kevin/linux
1
Fork 0
Code
e63340ae6b
linux/mm/slub.c

3521 lines
81 KiB
C
Raw Normal View History

SLUB core This is a new slab allocator which was motivated by the complexity of the existing code in mm/slab.c. It attempts to address a variety of concerns with the existing implementation. A. Management of object queues A particular concern was the complex management of the numerous object queues in SLAB. SLUB has no such queues. Instead we dedicate a slab for each allocating CPU and use objects from a slab directly instead of queueing them up. B. Storage overhead of object queues SLAB Object queues exist per node, per CPU. The alien cache queue even has a queue array that contain a queue for each processor on each node. For very large systems the number of queues and the number of objects that may be caught in those queues grows exponentially. On our systems with 1k nodes / processors we have several gigabytes just tied up for storing references to objects for those queues This does not include the objects that could be on those queues. One fears that the whole memory of the machine could one day be consumed by those queues. C. SLAB meta data overhead SLAB has overhead at the beginning of each slab. This means that data cannot be naturally aligned at the beginning of a slab block. SLUB keeps all meta data in the corresponding page_struct. Objects can be naturally aligned in the slab. F.e. a 128 byte object will be aligned at 128 byte boundaries and can fit tightly into a 4k page with no bytes left over. SLAB cannot do this. D. SLAB has a complex cache reaper SLUB does not need a cache reaper for UP systems. On SMP systems the per CPU slab may be pushed back into partial list but that operation is simple and does not require an iteration over a list of objects. SLAB expires per CPU, shared and alien object queues during cache reaping which may cause strange hold offs. E. SLAB has complex NUMA policy layer support SLUB pushes NUMA policy handling into the page allocator. This means that allocation is coarser (SLUB does interleave on a page level) but that situation was also present before 2.6.13. SLABs application of policies to individual slab objects allocated in SLAB is certainly a performance concern due to the frequent references to memory policies which may lead a sequence of objects to come from one node after another. SLUB will get a slab full of objects from one node and then will switch to the next. F. Reduction of the size of partial slab lists SLAB has per node partial lists. This means that over time a large number of partial slabs may accumulate on those lists. These can only be reused if allocator occur on specific nodes. SLUB has a global pool of partial slabs and will consume slabs from that pool to decrease fragmentation. G. Tunables SLAB has sophisticated tuning abilities for each slab cache. One can manipulate the queue sizes in detail. However, filling the queues still requires the uses of the spin lock to check out slabs. SLUB has a global parameter (min_slab_order) for tuning. Increasing the minimum slab order can decrease the locking overhead. The bigger the slab order the less motions of pages between per CPU and partial lists occur and the better SLUB will be scaling. G. Slab merging We often have slab caches with similar parameters. SLUB detects those on boot up and merges them into the corresponding general caches. This leads to more effective memory use. About 50% of all caches can be eliminated through slab merging. This will also decrease slab fragmentation because partial allocated slabs can be filled up again. Slab merging can be switched off by specifying slub_nomerge on boot up. Note that merging can expose heretofore unknown bugs in the kernel because corrupted objects may now be placed differently and corrupt differing neighboring objects. Enable sanity checks to find those. H. Diagnostics The current slab diagnostics are difficult to use and require a recompilation of the kernel. SLUB contains debugging code that is always available (but is kept out of the hot code paths). SLUB diagnostics can be enabled via the "slab_debug" option. Parameters can be specified to select a single or a group of slab caches for diagnostics. This means that the system is running with the usual performance and it is much more likely that race conditions can be reproduced. I. Resiliency If basic sanity checks are on then SLUB is capable of detecting common error conditions and recover as best as possible to allow the system to continue. J. Tracing Tracing can be enabled via the slab_debug=T,<slabcache> option during boot. SLUB will then protocol all actions on that slabcache and dump the object contents on free. K. On demand DMA cache creation. Generally DMA caches are not needed. If a kmalloc is used with __GFP_DMA then just create this single slabcache that is needed. For systems that have no ZONE_DMA requirement the support is completely eliminated. L. Performance increase Some benchmarks have shown speed improvements on kernbench in the range of 5-10%. The locking overhead of slub is based on the underlying base allocation size. If we can reliably allocate larger order pages then it is possible to increase slub performance much further. The anti-fragmentation patches may enable further performance increases. Tested on: i386 UP + SMP, x86_64 UP + SMP + NUMA emulation, IA64 NUMA + Simulator SLUB Boot options slub_nomerge Disable merging of slabs slub_min_order=x Require a minimum order for slab caches. This increases the managed chunk size and therefore reduces meta data and locking overhead. slub_min_objects=x Mininum objects per slab. Default is 8. slub_max_order=x Avoid generating slabs larger than order specified. slub_debug Enable all diagnostics for all caches slub_debug=<options> Enable selective options for all caches slub_debug=<o>,<cache> Enable selective options for a certain set of caches Available Debug options F Double Free checking, sanity and resiliency R Red zoning P Object / padding poisoning U Track last free / alloc T Trace all allocs / frees (only use for individual slabs). To use SLUB: Apply this patch and then select SLUB as the default slab allocator. [hugh@veritas.com: fix an oops-causing locking error] [akpm@linux-foundation.org: various stupid cleanups and small fixes] Signed-off-by: Christoph Lameter <clameter@sgi.com> Signed-off-by: Hugh Dickins <hugh@veritas.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2007-05-06 14:49:36 -07:00
/*
* SLUB: A slab allocator that limits cache line use instead of queuing
* objects in per cpu and per node lists.
*
* The allocator synchronizes using per slab locks and only
* uses a centralized lock to manage a pool of partial slabs.
*
* (C) 2007 SGI, Christoph Lameter <clameter@sgi.com>
*/
#include <linux/mm.h>
#include <linux/module.h>
#include <linux/bit_spinlock.h>
#include <linux/interrupt.h>
#include <linux/bitops.h>
#include <linux/slab.h>
#include <linux/seq_file.h>
#include <linux/cpu.h>
#include <linux/cpuset.h>
#include <linux/mempolicy.h>
#include <linux/ctype.h>
#include <linux/kallsyms.h>
/*
* Lock order:
* 1. slab_lock(page)
* 2. slab->list_lock
*
* The slab_lock protects operations on the object of a particular
* slab and its metadata in the page struct. If the slab lock
* has been taken then no allocations nor frees can be performed
* on the objects in the slab nor can the slab be added or removed
* from the partial or full lists since this would mean modifying
* the page_struct of the slab.
*
* The list_lock protects the partial and full list on each node and
* the partial slab counter. If taken then no new slabs may be added or
* removed from the lists nor make the number of partial slabs be modified.
* (Note that the total number of slabs is an atomic value that may be
* modified without taking the list lock).
*
* The list_lock is a centralized lock and thus we avoid taking it as
* much as possible. As long as SLUB does not have to handle partial
* slabs, operations can continue without any centralized lock. F.e.
* allocating a long series of objects that fill up slabs does not require
* the list lock.
*
* The lock order is sometimes inverted when we are trying to get a slab
* off a list. We take the list_lock and then look for a page on the list
* to use. While we do that objects in the slabs may be freed. We can
* only operate on the slab if we have also taken the slab_lock. So we use
* a slab_trylock() on the slab. If trylock was successful then no frees
* can occur anymore and we can use the slab for allocations etc. If the
* slab_trylock() does not succeed then frees are in progress in the slab and
* we must stay away from it for a while since we may cause a bouncing
* cacheline if we try to acquire the lock. So go onto the next slab.
* If all pages are busy then we may allocate a new slab instead of reusing
* a partial slab. A new slab has noone operating on it and thus there is
* no danger of cacheline contention.
*
* Interrupts are disabled during allocation and deallocation in order to
* make the slab allocator safe to use in the context of an irq. In addition
* interrupts are disabled to ensure that the processor does not change
* while handling per_cpu slabs, due to kernel preemption.
*
* SLUB assigns one slab for allocation to each processor.
* Allocations only occur from these slabs called cpu slabs.
*
* Slabs with free elements are kept on a partial list.
* There is no list for full slabs. If an object in a full slab is
* freed then the slab will show up again on the partial lists.
* Otherwise there is no need to track full slabs unless we have to
* track full slabs for debugging purposes.
*
* Slabs are freed when they become empty. Teardown and setup is
* minimal so we rely on the page allocators per cpu caches for
* fast frees and allocs.
*
* Overloading of page flags that are otherwise used for LRU management.
*
* PageActive The slab is used as a cpu cache. Allocations
* may be performed from the slab. The slab is not
* on any slab list and cannot be moved onto one.
*
* PageError Slab requires special handling due to debug
* options set. This moves slab handling out of
* the fast path.
*/
/*
* Issues still to be resolved:
*
* - The per cpu array is updated for each new slab and and is a remote
* cacheline for most nodes. This could become a bouncing cacheline given
* enough frequent updates. There are 16 pointers in a cacheline.so at
* max 16 cpus could compete. Likely okay.
*
* - Support PAGE_ALLOC_DEBUG. Should be easy to do.
*
* - Variable sizing of the per node arrays
*/
/* Enable to test recovery from slab corruption on boot */
#undef SLUB_RESILIENCY_TEST
#if PAGE_SHIFT <= 12
/*
* Small page size. Make sure that we do not fragment memory
*/
#define DEFAULT_MAX_ORDER 1
#define DEFAULT_MIN_OBJECTS 4
#else
/*
* Large page machines are customarily able to handle larger
* page orders.
*/
#define DEFAULT_MAX_ORDER 2
#define DEFAULT_MIN_OBJECTS 8
#endif
SLUB: Free slabs and sort partial slab lists in kmem_cache_shrink At kmem_cache_shrink check if we have any empty slabs on the partial if so then remove them. Also--as an anti-fragmentation measure--sort the partial slabs so that the most fully allocated ones come first and the least allocated last. The next allocations may fill up the nearly full slabs. Having the least allocated slabs last gives them the maximum chance that their remaining objects may be freed. Thus we can hopefully minimize the partial slabs. I think this is the best one can do in terms antifragmentation measures. Real defragmentation (meaning moving objects out of slabs with the least free objects to those that are almost full) can be implemted by reverse scanning through the list produced here but that would mean that we need to provide a callback at slab cache creation that allows the deletion or moving of an object. This will involve slab API changes, so defer for now. Cc: Mel Gorman <mel@skynet.ie> Signed-off-by: Christoph Lameter <clameter@sgi.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2007-05-06 14:49:46 -07:00
/*
* Mininum number of partial slabs. These will be left on the partial
* lists even if they are empty. kmem_cache_shrink may reclaim them.
*/
SLUB: Add MIN_PARTIAL We leave a mininum of partial slabs on nodes when we search for partial slabs on other node. Define a constant for that value. Then modify slub to keep MIN_PARTIAL slabs around. This avoids bad situations where a function frees the last object in a slab (which results in the page being returned to the page allocator) only to then allocate one again (which requires getting a page back from the page allocator if the partial list was empty). Keeping a couple of slabs on the partial list reduces overhead. Empty slabs are added to the end of the partial list to insure that partially allocated slabs are consumed first (defragmentation). Signed-off-by: Christoph Lameter <clameter@sgi.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2007-05-06 14:49:44 -07:00
#define MIN_PARTIAL 2
SLUB: Free slabs and sort partial slab lists in kmem_cache_shrink At kmem_cache_shrink check if we have any empty slabs on the partial if so then remove them. Also--as an anti-fragmentation measure--sort the partial slabs so that the most fully allocated ones come first and the least allocated last. The next allocations may fill up the nearly full slabs. Having the least allocated slabs last gives them the maximum chance that their remaining objects may be freed. Thus we can hopefully minimize the partial slabs. I think this is the best one can do in terms antifragmentation measures. Real defragmentation (meaning moving objects out of slabs with the least free objects to those that are almost full) can be implemted by reverse scanning through the list produced here but that would mean that we need to provide a callback at slab cache creation that allows the deletion or moving of an object. This will involve slab API changes, so defer for now. Cc: Mel Gorman <mel@skynet.ie> Signed-off-by: Christoph Lameter <clameter@sgi.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2007-05-06 14:49:46 -07:00
/*
* Maximum number of desirable partial slabs.
* The existence of more partial slabs makes kmem_cache_shrink
* sort the partial list by the number of objects in the.
*/
#define MAX_PARTIAL 10
SLUB core This is a new slab allocator which was motivated by the complexity of the existing code in mm/slab.c. It attempts to address a variety of concerns with the existing implementation. A. Management of object queues A particular concern was the complex management of the numerous object queues in SLAB. SLUB has no such queues. Instead we dedicate a slab for each allocating CPU and use objects from a slab directly instead of queueing them up. B. Storage overhead of object queues SLAB Object queues exist per node, per CPU. The alien cache queue even has a queue array that contain a queue for each processor on each node. For very large systems the number of queues and the number of objects that may be caught in those queues grows exponentially. On our systems with 1k nodes / processors we have several gigabytes just tied up for storing references to objects for those queues This does not include the objects that could be on those queues. One fears that the whole memory of the machine could one day be consumed by those queues. C. SLAB meta data overhead SLAB has overhead at the beginning of each slab. This means that data cannot be naturally aligned at the beginning of a slab block. SLUB keeps all meta data in the corresponding page_struct. Objects can be naturally aligned in the slab. F.e. a 128 byte object will be aligned at 128 byte boundaries and can fit tightly into a 4k page with no bytes left over. SLAB cannot do this. D. SLAB has a complex cache reaper SLUB does not need a cache reaper for UP systems. On SMP systems the per CPU slab may be pushed back into partial list but that operation is simple and does not require an iteration over a list of objects. SLAB expires per CPU, shared and alien object queues during cache reaping which may cause strange hold offs. E. SLAB has complex NUMA policy layer support SLUB pushes NUMA policy handling into the page allocator. This means that allocation is coarser (SLUB does interleave on a page level) but that situation was also present before 2.6.13. SLABs application of policies to individual slab objects allocated in SLAB is certainly a performance concern due to the frequent references to memory policies which may lead a sequence of objects to come from one node after another. SLUB will get a slab full of objects from one node and then will switch to the next. F. Reduction of the size of partial slab lists SLAB has per node partial lists. This means that over time a large number of partial slabs may accumulate on those lists. These can only be reused if allocator occur on specific nodes. SLUB has a global pool of partial slabs and will consume slabs from that pool to decrease fragmentation. G. Tunables SLAB has sophisticated tuning abilities for each slab cache. One can manipulate the queue sizes in detail. However, filling the queues still requires the uses of the spin lock to check out slabs. SLUB has a global parameter (min_slab_order) for tuning. Increasing the minimum slab order can decrease the locking overhead. The bigger the slab order the less motions of pages between per CPU and partial lists occur and the better SLUB will be scaling. G. Slab merging We often have slab caches with similar parameters. SLUB detects those on boot up and merges them into the corresponding general caches. This leads to more effective memory use. About 50% of all caches can be eliminated through slab merging. This will also decrease slab fragmentation because partial allocated slabs can be filled up again. Slab merging can be switched off by specifying slub_nomerge on boot up. Note that merging can expose heretofore unknown bugs in the kernel because corrupted objects may now be placed differently and corrupt differing neighboring objects. Enable sanity checks to find those. H. Diagnostics The current slab diagnostics are difficult to use and require a recompilation of the kernel. SLUB contains debugging code that is always available (but is kept out of the hot code paths). SLUB diagnostics can be enabled via the "slab_debug" option. Parameters can be specified to select a single or a group of slab caches for diagnostics. This means that the system is running with the usual performance and it is much more likely that race conditions can be reproduced. I. Resiliency If basic sanity checks are on then SLUB is capable of detecting common error conditions and recover as best as possible to allow the system to continue. J. Tracing Tracing can be enabled via the slab_debug=T,<slabcache> option during boot. SLUB will then protocol all actions on that slabcache and dump the object contents on free. K. On demand DMA cache creation. Generally DMA caches are not needed. If a kmalloc is used with __GFP_DMA then just create this single slabcache that is needed. For systems that have no ZONE_DMA requirement the support is completely eliminated. L. Performance increase Some benchmarks have shown speed improvements on kernbench in the range of 5-10%. The locking overhead of slub is based on the underlying base allocation size. If we can reliably allocate larger order pages then it is possible to increase slub performance much further. The anti-fragmentation patches may enable further performance increases. Tested on: i386 UP + SMP, x86_64 UP + SMP + NUMA emulation, IA64 NUMA + Simulator SLUB Boot options slub_nomerge Disable merging of slabs slub_min_order=x Require a minimum order for slab caches. This increases the managed chunk size and therefore reduces meta data and locking overhead. slub_min_objects=x Mininum objects per slab. Default is 8. slub_max_order=x Avoid generating slabs larger than order specified. slub_debug Enable all diagnostics for all caches slub_debug=<options> Enable selective options for all caches slub_debug=<o>,<cache> Enable selective options for a certain set of caches Available Debug options F Double Free checking, sanity and resiliency R Red zoning P Object / padding poisoning U Track last free / alloc T Trace all allocs / frees (only use for individual slabs). To use SLUB: Apply this patch and then select SLUB as the default slab allocator. [hugh@veritas.com: fix an oops-causing locking error] [akpm@linux-foundation.org: various stupid cleanups and small fixes] Signed-off-by: Christoph Lameter <clameter@sgi.com> Signed-off-by: Hugh Dickins <hugh@veritas.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2007-05-06 14:49:36 -07:00
#define DEBUG_DEFAULT_FLAGS (SLAB_DEBUG_FREE | SLAB_RED_ZONE | \
SLAB_POISON | SLAB_STORE_USER)
/*
* Set of flags that will prevent slab merging
*/
#define SLUB_NEVER_MERGE (SLAB_RED_ZONE | SLAB_POISON | SLAB_STORE_USER | \
SLAB_TRACE | SLAB_DESTROY_BY_RCU)
#define SLUB_MERGE_SAME (SLAB_DEBUG_FREE | SLAB_RECLAIM_ACCOUNT | \
SLAB_CACHE_DMA)
#ifndef ARCH_KMALLOC_MINALIGN
SLUB: change default alignments Structures may contain u64 items on 32 bit platforms that are only able to address 64 bit items on 64 bit boundaries. Change the mininum alignment of slabs to conform to those expectations. ARCH_KMALLOC_MINALIGN must be changed for good since a variety of structure are mixed in the general slabs. ARCH_SLAB_MINALIGN is changed because currently there is no consistent specification of object alignment. We may have that in the future when the KMEM_CACHE and related macros are used to generate slabs. These pass the alignment of the structure generated by the compiler to the slab. With KMEM_CACHE etc we could align structures that do not contain 64 bit values to 32 bit boundaries potentially saving some memory. Signed-off-by: Christoph Lameter <clameter@sgi.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2007-05-06 14:49:37 -07:00
#define ARCH_KMALLOC_MINALIGN __alignof__(unsigned long long)
SLUB core This is a new slab allocator which was motivated by the complexity of the existing code in mm/slab.c. It attempts to address a variety of concerns with the existing implementation. A. Management of object queues A particular concern was the complex management of the numerous object queues in SLAB. SLUB has no such queues. Instead we dedicate a slab for each allocating CPU and use objects from a slab directly instead of queueing them up. B. Storage overhead of object queues SLAB Object queues exist per node, per CPU. The alien cache queue even has a queue array that contain a queue for each processor on each node. For very large systems the number of queues and the number of objects that may be caught in those queues grows exponentially. On our systems with 1k nodes / processors we have several gigabytes just tied up for storing references to objects for those queues This does not include the objects that could be on those queues. One fears that the whole memory of the machine could one day be consumed by those queues. C. SLAB meta data overhead SLAB has overhead at the beginning of each slab. This means that data cannot be naturally aligned at the beginning of a slab block. SLUB keeps all meta data in the corresponding page_struct. Objects can be naturally aligned in the slab. F.e. a 128 byte object will be aligned at 128 byte boundaries and can fit tightly into a 4k page with no bytes left over. SLAB cannot do this. D. SLAB has a complex cache reaper SLUB does not need a cache reaper for UP systems. On SMP systems the per CPU slab may be pushed back into partial list but that operation is simple and does not require an iteration over a list of objects. SLAB expires per CPU, shared and alien object queues during cache reaping which may cause strange hold offs. E. SLAB has complex NUMA policy layer support SLUB pushes NUMA policy handling into the page allocator. This means that allocation is coarser (SLUB does interleave on a page level) but that situation was also present before 2.6.13. SLABs application of policies to individual slab objects allocated in SLAB is certainly a performance concern due to the frequent references to memory policies which may lead a sequence of objects to come from one node after another. SLUB will get a slab full of objects from one node and then will switch to the next. F. Reduction of the size of partial slab lists SLAB has per node partial lists. This means that over time a large number of partial slabs may accumulate on those lists. These can only be reused if allocator occur on specific nodes. SLUB has a global pool of partial slabs and will consume slabs from that pool to decrease fragmentation. G. Tunables SLAB has sophisticated tuning abilities for each slab cache. One can manipulate the queue sizes in detail. However, filling the queues still requires the uses of the spin lock to check out slabs. SLUB has a global parameter (min_slab_order) for tuning. Increasing the minimum slab order can decrease the locking overhead. The bigger the slab order the less motions of pages between per CPU and partial lists occur and the better SLUB will be scaling. G. Slab merging We often have slab caches with similar parameters. SLUB detects those on boot up and merges them into the corresponding general caches. This leads to more effective memory use. About 50% of all caches can be eliminated through slab merging. This will also decrease slab fragmentation because partial allocated slabs can be filled up again. Slab merging can be switched off by specifying slub_nomerge on boot up. Note that merging can expose heretofore unknown bugs in the kernel because corrupted objects may now be placed differently and corrupt differing neighboring objects. Enable sanity checks to find those. H. Diagnostics The current slab diagnostics are difficult to use and require a recompilation of the kernel. SLUB contains debugging code that is always available (but is kept out of the hot code paths). SLUB diagnostics can be enabled via the "slab_debug" option. Parameters can be specified to select a single or a group of slab caches for diagnostics. This means that the system is running with the usual performance and it is much more likely that race conditions can be reproduced. I. Resiliency If basic sanity checks are on then SLUB is capable of detecting common error conditions and recover as best as possible to allow the system to continue. J. Tracing Tracing can be enabled via the slab_debug=T,<slabcache> option during boot. SLUB will then protocol all actions on that slabcache and dump the object contents on free. K. On demand DMA cache creation. Generally DMA caches are not needed. If a kmalloc is used with __GFP_DMA then just create this single slabcache that is needed. For systems that have no ZONE_DMA requirement the support is completely eliminated. L. Performance increase Some benchmarks have shown speed improvements on kernbench in the range of 5-10%. The locking overhead of slub is based on the underlying base allocation size. If we can reliably allocate larger order pages then it is possible to increase slub performance much further. The anti-fragmentation patches may enable further performance increases. Tested on: i386 UP + SMP, x86_64 UP + SMP + NUMA emulation, IA64 NUMA + Simulator SLUB Boot options slub_nomerge Disable merging of slabs slub_min_order=x Require a minimum order for slab caches. This increases the managed chunk size and therefore reduces meta data and locking overhead. slub_min_objects=x Mininum objects per slab. Default is 8. slub_max_order=x Avoid generating slabs larger than order specified. slub_debug Enable all diagnostics for all caches slub_debug=<options> Enable selective options for all caches slub_debug=<o>,<cache> Enable selective options for a certain set of caches Available Debug options F Double Free checking, sanity and resiliency R Red zoning P Object / padding poisoning U Track last free / alloc T Trace all allocs / frees (only use for individual slabs). To use SLUB: Apply this patch and then select SLUB as the default slab allocator. [hugh@veritas.com: fix an oops-causing locking error] [akpm@linux-foundation.org: various stupid cleanups and small fixes] Signed-off-by: Christoph Lameter <clameter@sgi.com> Signed-off-by: Hugh Dickins <hugh@veritas.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2007-05-06 14:49:36 -07:00
#endif
#ifndef ARCH_SLAB_MINALIGN
SLUB: change default alignments Structures may contain u64 items on 32 bit platforms that are only able to address 64 bit items on 64 bit boundaries. Change the mininum alignment of slabs to conform to those expectations. ARCH_KMALLOC_MINALIGN must be changed for good since a variety of structure are mixed in the general slabs. ARCH_SLAB_MINALIGN is changed because currently there is no consistent specification of object alignment. We may have that in the future when the KMEM_CACHE and related macros are used to generate slabs. These pass the alignment of the structure generated by the compiler to the slab. With KMEM_CACHE etc we could align structures that do not contain 64 bit values to 32 bit boundaries potentially saving some memory. Signed-off-by: Christoph Lameter <clameter@sgi.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2007-05-06 14:49:37 -07:00
#define ARCH_SLAB_MINALIGN __alignof__(unsigned long long)
SLUB core This is a new slab allocator which was motivated by the complexity of the existing code in mm/slab.c. It attempts to address a variety of concerns with the existing implementation. A. Management of object queues A particular concern was the complex management of the numerous object queues in SLAB. SLUB has no such queues. Instead we dedicate a slab for each allocating CPU and use objects from a slab directly instead of queueing them up. B. Storage overhead of object queues SLAB Object queues exist per node, per CPU. The alien cache queue even has a queue array that contain a queue for each processor on each node. For very large systems the number of queues and the number of objects that may be caught in those queues grows exponentially. On our systems with 1k nodes / processors we have several gigabytes just tied up for storing references to objects for those queues This does not include the objects that could be on those queues. One fears that the whole memory of the machine could one day be consumed by those queues. C. SLAB meta data overhead SLAB has overhead at the beginning of each slab. This means that data cannot be naturally aligned at the beginning of a slab block. SLUB keeps all meta data in the corresponding page_struct. Objects can be naturally aligned in the slab. F.e. a 128 byte object will be aligned at 128 byte boundaries and can fit tightly into a 4k page with no bytes left over. SLAB cannot do this. D. SLAB has a complex cache reaper SLUB does not need a cache reaper for UP systems. On SMP systems the per CPU slab may be pushed back into partial list but that operation is simple and does not require an iteration over a list of objects. SLAB expires per CPU, shared and alien object queues during cache reaping which may cause strange hold offs. E. SLAB has complex NUMA policy layer support SLUB pushes NUMA policy handling into the page allocator. This means that allocation is coarser (SLUB does interleave on a page level) but that situation was also present before 2.6.13. SLABs application of policies to individual slab objects allocated in SLAB is certainly a performance concern due to the frequent references to memory policies which may lead a sequence of objects to come from one node after another. SLUB will get a slab full of objects from one node and then will switch to the next. F. Reduction of the size of partial slab lists SLAB has per node partial lists. This means that over time a large number of partial slabs may accumulate on those lists. These can only be reused if allocator occur on specific nodes. SLUB has a global pool of partial slabs and will consume slabs from that pool to decrease fragmentation. G. Tunables SLAB has sophisticated tuning abilities for each slab cache. One can manipulate the queue sizes in detail. However, filling the queues still requires the uses of the spin lock to check out slabs. SLUB has a global parameter (min_slab_order) for tuning. Increasing the minimum slab order can decrease the locking overhead. The bigger the slab order the less motions of pages between per CPU and partial lists occur and the better SLUB will be scaling. G. Slab merging We often have slab caches with similar parameters. SLUB detects those on boot up and merges them into the corresponding general caches. This leads to more effective memory use. About 50% of all caches can be eliminated through slab merging. This will also decrease slab fragmentation because partial allocated slabs can be filled up again. Slab merging can be switched off by specifying slub_nomerge on boot up. Note that merging can expose heretofore unknown bugs in the kernel because corrupted objects may now be placed differently and corrupt differing neighboring objects. Enable sanity checks to find those. H. Diagnostics The current slab diagnostics are difficult to use and require a recompilation of the kernel. SLUB contains debugging code that is always available (but is kept out of the hot code paths). SLUB diagnostics can be enabled via the "slab_debug" option. Parameters can be specified to select a single or a group of slab caches for diagnostics. This means that the system is running with the usual performance and it is much more likely that race conditions can be reproduced. I. Resiliency If basic sanity checks are on then SLUB is capable of detecting common error conditions and recover as best as possible to allow the system to continue. J. Tracing Tracing can be enabled via the slab_debug=T,<slabcache> option during boot. SLUB will then protocol all actions on that slabcache and dump the object contents on free. K. On demand DMA cache creation. Generally DMA caches are not needed. If a kmalloc is used with __GFP_DMA then just create this single slabcache that is needed. For systems that have no ZONE_DMA requirement the support is completely eliminated. L. Performance increase Some benchmarks have shown speed improvements on kernbench in the range of 5-10%. The locking overhead of slub is based on the underlying base allocation size. If we can reliably allocate larger order pages then it is possible to increase slub performance much further. The anti-fragmentation patches may enable further performance increases. Tested on: i386 UP + SMP, x86_64 UP + SMP + NUMA emulation, IA64 NUMA + Simulator SLUB Boot options slub_nomerge Disable merging of slabs slub_min_order=x Require a minimum order for slab caches. This increases the managed chunk size and therefore reduces meta data and locking overhead. slub_min_objects=x Mininum objects per slab. Default is 8. slub_max_order=x Avoid generating slabs larger than order specified. slub_debug Enable all diagnostics for all caches slub_debug=<options> Enable selective options for all caches slub_debug=<o>,<cache> Enable selective options for a certain set of caches Available Debug options F Double Free checking, sanity and resiliency R Red zoning P Object / padding poisoning U Track last free / alloc T Trace all allocs / frees (only use for individual slabs). To use SLUB: Apply this patch and then select SLUB as the default slab allocator. [hugh@veritas.com: fix an oops-causing locking error] [akpm@linux-foundation.org: various stupid cleanups and small fixes] Signed-off-by: Christoph Lameter <clameter@sgi.com> Signed-off-by: Hugh Dickins <hugh@veritas.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2007-05-06 14:49:36 -07:00
#endif
/* Internal SLUB flags */
#define __OBJECT_POISON 0x80000000 /* Poison object */
static int kmem_size = sizeof(struct kmem_cache);
#ifdef CONFIG_SMP
static struct notifier_block slab_notifier;
#endif
static enum {
DOWN, /* No slab functionality available */
PARTIAL, /* kmem_cache_open() works but kmalloc does not */
UP, /* Everything works */
SYSFS /* Sysfs up */
} slab_state = DOWN;
/* A list of all slab caches on the system */
static DECLARE_RWSEM(slub_lock);
LIST_HEAD(slab_caches);
#ifdef CONFIG_SYSFS
static int sysfs_slab_add(struct kmem_cache *);
static int sysfs_slab_alias(struct kmem_cache *, const char *);
static void sysfs_slab_remove(struct kmem_cache *);
#else
static int sysfs_slab_add(struct kmem_cache *s) { return 0; }
static int sysfs_slab_alias(struct kmem_cache *s, const char *p) { return 0; }
static void sysfs_slab_remove(struct kmem_cache *s) {}
#endif
/********************************************************************
* Core slab cache functions
*******************************************************************/
int slab_is_available(void)
{
return slab_state >= UP;
}
static inline struct kmem_cache_node *get_node(struct kmem_cache *s, int node)
{
#ifdef CONFIG_NUMA
return s->node[node];
#else
return &s->local_node;
#endif
}
/*
* Object debugging
*/
static void print_section(char *text, u8 *addr, unsigned int length)
{
int i, offset;
int newline = 1;
char ascii[17];
ascii[16] = 0;
for (i = 0; i < length; i++) {
if (newline) {
printk(KERN_ERR "%10s 0x%p: ", text, addr + i);
newline = 0;
}
printk(" %02x", addr[i]);
offset = i % 16;
ascii[offset] = isgraph(addr[i]) ? addr[i] : '.';
if (offset == 15) {
printk(" %s\n",ascii);
newline = 1;
}
}
if (!newline) {
i %= 16;
while (i < 16) {
printk(" ");
ascii[i] = ' ';
i++;
}
printk(" %s\n", ascii);
}
}
/*
* Slow version of get and set free pointer.
*
* This requires touching the cache lines of kmem_cache.
* The offset can also be obtained from the page. In that
* case it is in the cacheline that we already need to touch.
*/
static void *get_freepointer(struct kmem_cache *s, void *object)
{
return *(void **)(object + s->offset);
}
static void set_freepointer(struct kmem_cache *s, void *object, void *fp)
{
*(void **)(object + s->offset) = fp;
}
/*
* Tracking user of a slab.
*/
struct track {
void *addr; /* Called from address */
int cpu; /* Was running on cpu */
int pid; /* Pid context */
unsigned long when; /* When did the operation occur */
};
enum track_item { TRACK_ALLOC, TRACK_FREE };
static struct track *get_track(struct kmem_cache *s, void *object,
enum track_item alloc)
{
struct track *p;
if (s->offset)
p = object + s->offset + sizeof(void *);
else
p = object + s->inuse;
return p + alloc;
}
static void set_track(struct kmem_cache *s, void *object,
enum track_item alloc, void *addr)
{
struct track *p;
if (s->offset)
p = object + s->offset + sizeof(void *);
else
p = object + s->inuse;
p += alloc;
if (addr) {
p->addr = addr;
p->cpu = smp_processor_id();
p->pid = current ? current->pid : -1;
p->when = jiffies;
} else
memset(p, 0, sizeof(struct track));
}
static void init_tracking(struct kmem_cache *s, void *object)
{
if (s->flags & SLAB_STORE_USER) {
set_track(s, object, TRACK_FREE, NULL);
set_track(s, object, TRACK_ALLOC, NULL);
}
}
static void print_track(const char *s, struct track *t)
{
if (!t->addr)
return;
printk(KERN_ERR "%s: ", s);
__print_symbol("%s", (unsigned long)t->addr);
printk(" jiffies_ago=%lu cpu=%u pid=%d\n", jiffies - t->when, t->cpu, t->pid);
}
static void print_trailer(struct kmem_cache *s, u8 *p)
{
unsigned int off; /* Offset of last byte */
if (s->flags & SLAB_RED_ZONE)
print_section("Redzone", p + s->objsize,
s->inuse - s->objsize);
printk(KERN_ERR "FreePointer 0x%p -> 0x%p\n",
p + s->offset,
get_freepointer(s, p));
if (s->offset)
off = s->offset + sizeof(void *);
else
off = s->inuse;
if (s->flags & SLAB_STORE_USER) {
print_track("Last alloc", get_track(s, p, TRACK_ALLOC));
print_track("Last free ", get_track(s, p, TRACK_FREE));
off += 2 * sizeof(struct track);
}
if (off != s->size)
/* Beginning of the filler is the free pointer */
print_section("Filler", p + off, s->size - off);
}
static void object_err(struct kmem_cache *s, struct page *page,
u8 *object, char *reason)
{
u8 *addr = page_address(page);
printk(KERN_ERR "*** SLUB %s: %s@0x%p slab 0x%p\n",
s->name, reason, object, page);
printk(KERN_ERR " offset=%tu flags=0x%04lx inuse=%u freelist=0x%p\n",
object - addr, page->flags, page->inuse, page->freelist);
if (object > addr + 16)
print_section("Bytes b4", object - 16, 16);
print_section("Object", object, min(s->objsize, 128));
print_trailer(s, object);
dump_stack();
}
static void slab_err(struct kmem_cache *s, struct page *page, char *reason, ...)
{
va_list args;
char buf[100];
va_start(args, reason);
vsnprintf(buf, sizeof(buf), reason, args);
va_end(args);
printk(KERN_ERR "*** SLUB %s: %s in slab @0x%p\n", s->name, buf,
page);
dump_stack();
}
static void init_object(struct kmem_cache *s, void *object, int active)
{
u8 *p = object;
if (s->flags & __OBJECT_POISON) {
memset(p, POISON_FREE, s->objsize - 1);
p[s->objsize -1] = POISON_END;
}
if (s->flags & SLAB_RED_ZONE)
memset(p + s->objsize,
active ? SLUB_RED_ACTIVE : SLUB_RED_INACTIVE,
s->inuse - s->objsize);
}
static int check_bytes(u8 *start, unsigned int value, unsigned int bytes)
{
while (bytes) {
if (*start != (u8)value)
return 0;
start++;
bytes--;
}
return 1;
}
static int check_valid_pointer(struct kmem_cache *s, struct page *page,
void *object)
{
void *base;
if (!object)
return 1;
base = page_address(page);
if (object < base || object >= base + s->objects * s->size ||
(object - base) % s->size) {
return 0;
}
return 1;
}
/*
* Object layout:
*
* object address
* Bytes of the object to be managed.
* If the freepointer may overlay the object then the free
* pointer is the first word of the object.
* Poisoning uses 0x6b (POISON_FREE) and the last byte is
* 0xa5 (POISON_END)
*
* object + s->objsize
* Padding to reach word boundary. This is also used for Redzoning.
* Padding is extended to word size if Redzoning is enabled
* and objsize == inuse.
* We fill with 0xbb (RED_INACTIVE) for inactive objects and with
* 0xcc (RED_ACTIVE) for objects in use.
*
* object + s->inuse
* A. Free pointer (if we cannot overwrite object on free)
* B. Tracking data for SLAB_STORE_USER
* C. Padding to reach required alignment boundary
* Padding is done using 0x5a (POISON_INUSE)
*
* object + s->size
*
* If slabcaches are merged then the objsize and inuse boundaries are to
* be ignored. And therefore no slab options that rely on these boundaries
* may be used with merged slabcaches.
*/
static void restore_bytes(struct kmem_cache *s, char *message, u8 data,
void *from, void *to)
{
slub: remove object activities out of checking functions Make sure that the check function really only check things and do not perform activities. Extract the tracing and object seeding out of the two check functions and place them into slab_alloc and slab_free Signed-off-by: Christoph Lameter <clameter@sgi.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2007-05-06 14:49:47 -07:00
printk(KERN_ERR "@@@ SLUB %s: Restoring %s (0x%x) from 0x%p-0x%p\n",
SLUB core This is a new slab allocator which was motivated by the complexity of the existing code in mm/slab.c. It attempts to address a variety of concerns with the existing implementation. A. Management of object queues A particular concern was the complex management of the numerous object queues in SLAB. SLUB has no such queues. Instead we dedicate a slab for each allocating CPU and use objects from a slab directly instead of queueing them up. B. Storage overhead of object queues SLAB Object queues exist per node, per CPU. The alien cache queue even has a queue array that contain a queue for each processor on each node. For very large systems the number of queues and the number of objects that may be caught in those queues grows exponentially. On our systems with 1k nodes / processors we have several gigabytes just tied up for storing references to objects for those queues This does not include the objects that could be on those queues. One fears that the whole memory of the machine could one day be consumed by those queues. C. SLAB meta data overhead SLAB has overhead at the beginning of each slab. This means that data cannot be naturally aligned at the beginning of a slab block. SLUB keeps all meta data in the corresponding page_struct. Objects can be naturally aligned in the slab. F.e. a 128 byte object will be aligned at 128 byte boundaries and can fit tightly into a 4k page with no bytes left over. SLAB cannot do this. D. SLAB has a complex cache reaper SLUB does not need a cache reaper for UP systems. On SMP systems the per CPU slab may be pushed back into partial list but that operation is simple and does not require an iteration over a list of objects. SLAB expires per CPU, shared and alien object queues during cache reaping which may cause strange hold offs. E. SLAB has complex NUMA policy layer support SLUB pushes NUMA policy handling into the page allocator. This means that allocation is coarser (SLUB does interleave on a page level) but that situation was also present before 2.6.13. SLABs application of policies to individual slab objects allocated in SLAB is certainly a performance concern due to the frequent references to memory policies which may lead a sequence of objects to come from one node after another. SLUB will get a slab full of objects from one node and then will switch to the next. F. Reduction of the size of partial slab lists SLAB has per node partial lists. This means that over time a large number of partial slabs may accumulate on those lists. These can only be reused if allocator occur on specific nodes. SLUB has a global pool of partial slabs and will consume slabs from that pool to decrease fragmentation. G. Tunables SLAB has sophisticated tuning abilities for each slab cache. One can manipulate the queue sizes in detail. However, filling the queues still requires the uses of the spin lock to check out slabs. SLUB has a global parameter (min_slab_order) for tuning. Increasing the minimum slab order can decrease the locking overhead. The bigger the slab order the less motions of pages between per CPU and partial lists occur and the better SLUB will be scaling. G. Slab merging We often have slab caches with similar parameters. SLUB detects those on boot up and merges them into the corresponding general caches. This leads to more effective memory use. About 50% of all caches can be eliminated through slab merging. This will also decrease slab fragmentation because partial allocated slabs can be filled up again. Slab merging can be switched off by specifying slub_nomerge on boot up. Note that merging can expose heretofore unknown bugs in the kernel because corrupted objects may now be placed differently and corrupt differing neighboring objects. Enable sanity checks to find those. H. Diagnostics The current slab diagnostics are difficult to use and require a recompilation of the kernel. SLUB contains debugging code that is always available (but is kept out of the hot code paths). SLUB diagnostics can be enabled via the "slab_debug" option. Parameters can be specified to select a single or a group of slab caches for diagnostics. This means that the system is running with the usual performance and it is much more likely that race conditions can be reproduced. I. Resiliency If basic sanity checks are on then SLUB is capable of detecting common error conditions and recover as best as possible to allow the system to continue. J. Tracing Tracing can be enabled via the slab_debug=T,<slabcache> option during boot. SLUB will then protocol all actions on that slabcache and dump the object contents on free. K. On demand DMA cache creation. Generally DMA caches are not needed. If a kmalloc is used with __GFP_DMA then just create this single slabcache that is needed. For systems that have no ZONE_DMA requirement the support is completely eliminated. L. Performance increase Some benchmarks have shown speed improvements on kernbench in the range of 5-10%. The locking overhead of slub is based on the underlying base allocation size. If we can reliably allocate larger order pages then it is possible to increase slub performance much further. The anti-fragmentation patches may enable further performance increases. Tested on: i386 UP + SMP, x86_64 UP + SMP + NUMA emulation, IA64 NUMA + Simulator SLUB Boot options slub_nomerge Disable merging of slabs slub_min_order=x Require a minimum order for slab caches. This increases the managed chunk size and therefore reduces meta data and locking overhead. slub_min_objects=x Mininum objects per slab. Default is 8. slub_max_order=x Avoid generating slabs larger than order specified. slub_debug Enable all diagnostics for all caches slub_debug=<options> Enable selective options for all caches slub_debug=<o>,<cache> Enable selective options for a certain set of caches Available Debug options F Double Free checking, sanity and resiliency R Red zoning P Object / padding poisoning U Track last free / alloc T Trace all allocs / frees (only use for individual slabs). To use SLUB: Apply this patch and then select SLUB as the default slab allocator. [hugh@veritas.com: fix an oops-causing locking error] [akpm@linux-foundation.org: various stupid cleanups and small fixes] Signed-off-by: Christoph Lameter <clameter@sgi.com> Signed-off-by: Hugh Dickins <hugh@veritas.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2007-05-06 14:49:36 -07:00
s->name, message, data, from, to - 1);
memset(from, data, to - from);
}
static int check_pad_bytes(struct kmem_cache *s, struct page *page, u8 *p)
{
unsigned long off = s->inuse; /* The end of info */
if (s->offset)
/* Freepointer is placed after the object. */
off += sizeof(void *);
if (s->flags & SLAB_STORE_USER)
/* We also have user information there */
off += 2 * sizeof(struct track);
if (s->size == off)
return 1;
if (check_bytes(p + off, POISON_INUSE, s->size - off))
return 1;
object_err(s, page, p, "Object padding check fails");
/*
* Restore padding
*/
restore_bytes(s, "object padding", POISON_INUSE, p + off, p + s->size);
return 0;
}
static int slab_pad_check(struct kmem_cache *s, struct page *page)
{
u8 *p;
int length, remainder;
if (!(s->flags & SLAB_POISON))
return 1;
p = page_address(page);
length = s->objects * s->size;
remainder = (PAGE_SIZE << s->order) - length;
if (!remainder)
return 1;
if (!check_bytes(p + length, POISON_INUSE, remainder)) {
slub: remove object activities out of checking functions Make sure that the check function really only check things and do not perform activities. Extract the tracing and object seeding out of the two check functions and place them into slab_alloc and slab_free Signed-off-by: Christoph Lameter <clameter@sgi.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2007-05-06 14:49:47 -07:00
slab_err(s, page, "Padding check failed");
SLUB core This is a new slab allocator which was motivated by the complexity of the existing code in mm/slab.c. It attempts to address a variety of concerns with the existing implementation. A. Management of object queues A particular concern was the complex management of the numerous object queues in SLAB. SLUB has no such queues. Instead we dedicate a slab for each allocating CPU and use objects from a slab directly instead of queueing them up. B. Storage overhead of object queues SLAB Object queues exist per node, per CPU. The alien cache queue even has a queue array that contain a queue for each processor on each node. For very large systems the number of queues and the number of objects that may be caught in those queues grows exponentially. On our systems with 1k nodes / processors we have several gigabytes just tied up for storing references to objects for those queues This does not include the objects that could be on those queues. One fears that the whole memory of the machine could one day be consumed by those queues. C. SLAB meta data overhead SLAB has overhead at the beginning of each slab. This means that data cannot be naturally aligned at the beginning of a slab block. SLUB keeps all meta data in the corresponding page_struct. Objects can be naturally aligned in the slab. F.e. a 128 byte object will be aligned at 128 byte boundaries and can fit tightly into a 4k page with no bytes left over. SLAB cannot do this. D. SLAB has a complex cache reaper SLUB does not need a cache reaper for UP systems. On SMP systems the per CPU slab may be pushed back into partial list but that operation is simple and does not require an iteration over a list of objects. SLAB expires per CPU, shared and alien object queues during cache reaping which may cause strange hold offs. E. SLAB has complex NUMA policy layer support SLUB pushes NUMA policy handling into the page allocator. This means that allocation is coarser (SLUB does interleave on a page level) but that situation was also present before 2.6.13. SLABs application of policies to individual slab objects allocated in SLAB is certainly a performance concern due to the frequent references to memory policies which may lead a sequence of objects to come from one node after another. SLUB will get a slab full of objects from one node and then will switch to the next. F. Reduction of the size of partial slab lists SLAB has per node partial lists. This means that over time a large number of partial slabs may accumulate on those lists. These can only be reused if allocator occur on specific nodes. SLUB has a global pool of partial slabs and will consume slabs from that pool to decrease fragmentation. G. Tunables SLAB has sophisticated tuning abilities for each slab cache. One can manipulate the queue sizes in detail. However, filling the queues still requires the uses of the spin lock to check out slabs. SLUB has a global parameter (min_slab_order) for tuning. Increasing the minimum slab order can decrease the locking overhead. The bigger the slab order the less motions of pages between per CPU and partial lists occur and the better SLUB will be scaling. G. Slab merging We often have slab caches with similar parameters. SLUB detects those on boot up and merges them into the corresponding general caches. This leads to more effective memory use. About 50% of all caches can be eliminated through slab merging. This will also decrease slab fragmentation because partial allocated slabs can be filled up again. Slab merging can be switched off by specifying slub_nomerge on boot up. Note that merging can expose heretofore unknown bugs in the kernel because corrupted objects may now be placed differently and corrupt differing neighboring objects. Enable sanity checks to find those. H. Diagnostics The current slab diagnostics are difficult to use and require a recompilation of the kernel. SLUB contains debugging code that is always available (but is kept out of the hot code paths). SLUB diagnostics can be enabled via the "slab_debug" option. Parameters can be specified to select a single or a group of slab caches for diagnostics. This means that the system is running with the usual performance and it is much more likely that race conditions can be reproduced. I. Resiliency If basic sanity checks are on then SLUB is capable of detecting common error conditions and recover as best as possible to allow the system to continue. J. Tracing Tracing can be enabled via the slab_debug=T,<slabcache> option during boot. SLUB will then protocol all actions on that slabcache and dump the object contents on free. K. On demand DMA cache creation. Generally DMA caches are not needed. If a kmalloc is used with __GFP_DMA then just create this single slabcache that is needed. For systems that have no ZONE_DMA requirement the support is completely eliminated. L. Performance increase Some benchmarks have shown speed improvements on kernbench in the range of 5-10%. The locking overhead of slub is based on the underlying base allocation size. If we can reliably allocate larger order pages then it is possible to increase slub performance much further. The anti-fragmentation patches may enable further performance increases. Tested on: i386 UP + SMP, x86_64 UP + SMP + NUMA emulation, IA64 NUMA + Simulator SLUB Boot options slub_nomerge Disable merging of slabs slub_min_order=x Require a minimum order for slab caches. This increases the managed chunk size and therefore reduces meta data and locking overhead. slub_min_objects=x Mininum objects per slab. Default is 8. slub_max_order=x Avoid generating slabs larger than order specified. slub_debug Enable all diagnostics for all caches slub_debug=<options> Enable selective options for all caches slub_debug=<o>,<cache> Enable selective options for a certain set of caches Available Debug options F Double Free checking, sanity and resiliency R Red zoning P Object / padding poisoning U Track last free / alloc T Trace all allocs / frees (only use for individual slabs). To use SLUB: Apply this patch and then select SLUB as the default slab allocator. [hugh@veritas.com: fix an oops-causing locking error] [akpm@linux-foundation.org: various stupid cleanups and small fixes] Signed-off-by: Christoph Lameter <clameter@sgi.com> Signed-off-by: Hugh Dickins <hugh@veritas.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2007-05-06 14:49:36 -07:00
restore_bytes(s, "slab padding", POISON_INUSE, p + length,
p + length + remainder);
return 0;
}
return 1;
}
static int check_object(struct kmem_cache *s, struct page *page,
void *object, int active)
{
u8 *p = object;
u8 *endobject = object + s->objsize;
if (s->flags & SLAB_RED_ZONE) {
unsigned int red =
active ? SLUB_RED_ACTIVE : SLUB_RED_INACTIVE;
if (!check_bytes(endobject, red, s->inuse - s->objsize)) {
object_err(s, page, object,
active ? "Redzone Active" : "Redzone Inactive");
restore_bytes(s, "redzone", red,
endobject, object + s->inuse);
return 0;
}
} else {
if ((s->flags & SLAB_POISON) && s->objsize < s->inuse &&
!check_bytes(endobject, POISON_INUSE,
s->inuse - s->objsize)) {
object_err(s, page, p, "Alignment padding check fails");
/*
* Fix it so that there will not be another report.
*
* Hmmm... We may be corrupting an object that now expects
* to be longer than allowed.
*/
restore_bytes(s, "alignment padding", POISON_INUSE,
endobject, object + s->inuse);
}
}
if (s->flags & SLAB_POISON) {
if (!active && (s->flags & __OBJECT_POISON) &&
(!check_bytes(p, POISON_FREE, s->objsize - 1) ||
p[s->objsize - 1] != POISON_END)) {
object_err(s, page, p, "Poison check failed");
restore_bytes(s, "Poison", POISON_FREE,
p, p + s->objsize -1);
restore_bytes(s, "Poison", POISON_END,
p + s->objsize - 1, p + s->objsize);
return 0;
}
/*
* check_pad_bytes cleans up on its own.
*/
check_pad_bytes(s, page, p);
}
if (!s->offset && active)
/*
* Object and freepointer overlap. Cannot check
* freepointer while object is allocated.
*/
return 1;
/* Check free pointer validity */
if (!check_valid_pointer(s, page, get_freepointer(s, p))) {
object_err(s, page, p, "Freepointer corrupt");
/*
* No choice but to zap it and thus loose the remainder
* of the free objects in this slab. May cause
* another error because the object count maybe
* wrong now.
*/
set_freepointer(s, p, NULL);
return 0;
}
return 1;
}
static int check_slab(struct kmem_cache *s, struct page *page)
{
VM_BUG_ON(!irqs_disabled());
if (!PageSlab(page)) {
slub: remove object activities out of checking functions Make sure that the check function really only check things and do not perform activities. Extract the tracing and object seeding out of the two check functions and place them into slab_alloc and slab_free Signed-off-by: Christoph Lameter <clameter@sgi.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2007-05-06 14:49:47 -07:00
slab_err(s, page, "Not a valid slab page flags=%lx "
"mapping=0x%p count=%d", page->flags, page->mapping,
SLUB core This is a new slab allocator which was motivated by the complexity of the existing code in mm/slab.c. It attempts to address a variety of concerns with the existing implementation. A. Management of object queues A particular concern was the complex management of the numerous object queues in SLAB. SLUB has no such queues. Instead we dedicate a slab for each allocating CPU and use objects from a slab directly instead of queueing them up. B. Storage overhead of object queues SLAB Object queues exist per node, per CPU. The alien cache queue even has a queue array that contain a queue for each processor on each node. For very large systems the number of queues and the number of objects that may be caught in those queues grows exponentially. On our systems with 1k nodes / processors we have several gigabytes just tied up for storing references to objects for those queues This does not include the objects that could be on those queues. One fears that the whole memory of the machine could one day be consumed by those queues. C. SLAB meta data overhead SLAB has overhead at the beginning of each slab. This means that data cannot be naturally aligned at the beginning of a slab block. SLUB keeps all meta data in the corresponding page_struct. Objects can be naturally aligned in the slab. F.e. a 128 byte object will be aligned at 128 byte boundaries and can fit tightly into a 4k page with no bytes left over. SLAB cannot do this. D. SLAB has a complex cache reaper SLUB does not need a cache reaper for UP systems. On SMP systems the per CPU slab may be pushed back into partial list but that operation is simple and does not require an iteration over a list of objects. SLAB expires per CPU, shared and alien object queues during cache reaping which may cause strange hold offs. E. SLAB has complex NUMA policy layer support SLUB pushes NUMA policy handling into the page allocator. This means that allocation is coarser (SLUB does interleave on a page level) but that situation was also present before 2.6.13. SLABs application of policies to individual slab objects allocated in SLAB is certainly a performance concern due to the frequent references to memory policies which may lead a sequence of objects to come from one node after another. SLUB will get a slab full of objects from one node and then will switch to the next. F. Reduction of the size of partial slab lists SLAB has per node partial lists. This means that over time a large number of partial slabs may accumulate on those lists. These can only be reused if allocator occur on specific nodes. SLUB has a global pool of partial slabs and will consume slabs from that pool to decrease fragmentation. G. Tunables SLAB has sophisticated tuning abilities for each slab cache. One can manipulate the queue sizes in detail. However, filling the queues still requires the uses of the spin lock to check out slabs. SLUB has a global parameter (min_slab_order) for tuning. Increasing the minimum slab order can decrease the locking overhead. The bigger the slab order the less motions of pages between per CPU and partial lists occur and the better SLUB will be scaling. G. Slab merging We often have slab caches with similar parameters. SLUB detects those on boot up and merges them into the corresponding general caches. This leads to more effective memory use. About 50% of all caches can be eliminated through slab merging. This will also decrease slab fragmentation because partial allocated slabs can be filled up again. Slab merging can be switched off by specifying slub_nomerge on boot up. Note that merging can expose heretofore unknown bugs in the kernel because corrupted objects may now be placed differently and corrupt differing neighboring objects. Enable sanity checks to find those. H. Diagnostics The current slab diagnostics are difficult to use and require a recompilation of the kernel. SLUB contains debugging code that is always available (but is kept out of the hot code paths). SLUB diagnostics can be enabled via the "slab_debug" option. Parameters can be specified to select a single or a group of slab caches for diagnostics. This means that the system is running with the usual performance and it is much more likely that race conditions can be reproduced. I. Resiliency If basic sanity checks are on then SLUB is capable of detecting common error conditions and recover as best as possible to allow the system to continue. J. Tracing Tracing can be enabled via the slab_debug=T,<slabcache> option during boot. SLUB will then protocol all actions on that slabcache and dump the object contents on free. K. On demand DMA cache creation. Generally DMA caches are not needed. If a kmalloc is used with __GFP_DMA then just create this single slabcache that is needed. For systems that have no ZONE_DMA requirement the support is completely eliminated. L. Performance increase Some benchmarks have shown speed improvements on kernbench in the range of 5-10%. The locking overhead of slub is based on the underlying base allocation size. If we can reliably allocate larger order pages then it is possible to increase slub performance much further. The anti-fragmentation patches may enable further performance increases. Tested on: i386 UP + SMP, x86_64 UP + SMP + NUMA emulation, IA64 NUMA + Simulator SLUB Boot options slub_nomerge Disable merging of slabs slub_min_order=x Require a minimum order for slab caches. This increases the managed chunk size and therefore reduces meta data and locking overhead. slub_min_objects=x Mininum objects per slab. Default is 8. slub_max_order=x Avoid generating slabs larger than order specified. slub_debug Enable all diagnostics for all caches slub_debug=<options> Enable selective options for all caches slub_debug=<o>,<cache> Enable selective options for a certain set of caches Available Debug options F Double Free checking, sanity and resiliency R Red zoning P Object / padding poisoning U Track last free / alloc T Trace all allocs / frees (only use for individual slabs). To use SLUB: Apply this patch and then select SLUB as the default slab allocator. [hugh@veritas.com: fix an oops-causing locking error] [akpm@linux-foundation.org: various stupid cleanups and small fixes] Signed-off-by: Christoph Lameter <clameter@sgi.com> Signed-off-by: Hugh Dickins <hugh@veritas.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2007-05-06 14:49:36 -07:00
page_count(page));
return 0;
}
if (page->offset * sizeof(void *) != s->offset) {
slub: remove object activities out of checking functions Make sure that the check function really only check things and do not perform activities. Extract the tracing and object seeding out of the two check functions and place them into slab_alloc and slab_free Signed-off-by: Christoph Lameter <clameter@sgi.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2007-05-06 14:49:47 -07:00
slab_err(s, page, "Corrupted offset %lu flags=0x%lx "
"mapping=0x%p count=%d",
SLUB core This is a new slab allocator which was motivated by the complexity of the existing code in mm/slab.c. It attempts to address a variety of concerns with the existing implementation. A. Management of object queues A particular concern was the complex management of the numerous object queues in SLAB. SLUB has no such queues. Instead we dedicate a slab for each allocating CPU and use objects from a slab directly instead of queueing them up. B. Storage overhead of object queues SLAB Object queues exist per node, per CPU. The alien cache queue even has a queue array that contain a queue for each processor on each node. For very large systems the number of queues and the number of objects that may be caught in those queues grows exponentially. On our systems with 1k nodes / processors we have several gigabytes just tied up for storing references to objects for those queues This does not include the objects that could be on those queues. One fears that the whole memory of the machine could one day be consumed by those queues. C. SLAB meta data overhead SLAB has overhead at the beginning of each slab. This means that data cannot be naturally aligned at the beginning of a slab block. SLUB keeps all meta data in the corresponding page_struct. Objects can be naturally aligned in the slab. F.e. a 128 byte object will be aligned at 128 byte boundaries and can fit tightly into a 4k page with no bytes left over. SLAB cannot do this. D. SLAB has a complex cache reaper SLUB does not need a cache reaper for UP systems. On SMP systems the per CPU slab may be pushed back into partial list but that operation is simple and does not require an iteration over a list of objects. SLAB expires per CPU, shared and alien object queues during cache reaping which may cause strange hold offs. E. SLAB has complex NUMA policy layer support SLUB pushes NUMA policy handling into the page allocator. This means that allocation is coarser (SLUB does interleave on a page level) but that situation was also present before 2.6.13. SLABs application of policies to individual slab objects allocated in SLAB is certainly a performance concern due to the frequent references to memory policies which may lead a sequence of objects to come from one node after another. SLUB will get a slab full of objects from one node and then will switch to the next. F. Reduction of the size of partial slab lists SLAB has per node partial lists. This means that over time a large number of partial slabs may accumulate on those lists. These can only be reused if allocator occur on specific nodes. SLUB has a global pool of partial slabs and will consume slabs from that pool to decrease fragmentation. G. Tunables SLAB has sophisticated tuning abilities for each slab cache. One can manipulate the queue sizes in detail. However, filling the queues still requires the uses of the spin lock to check out slabs. SLUB has a global parameter (min_slab_order) for tuning. Increasing the minimum slab order can decrease the locking overhead. The bigger the slab order the less motions of pages between per CPU and partial lists occur and the better SLUB will be scaling. G. Slab merging We often have slab caches with similar parameters. SLUB detects those on boot up and merges them into the corresponding general caches. This leads to more effective memory use. About 50% of all caches can be eliminated through slab merging. This will also decrease slab fragmentation because partial allocated slabs can be filled up again. Slab merging can be switched off by specifying slub_nomerge on boot up. Note that merging can expose heretofore unknown bugs in the kernel because corrupted objects may now be placed differently and corrupt differing neighboring objects. Enable sanity checks to find those. H. Diagnostics The current slab diagnostics are difficult to use and require a recompilation of the kernel. SLUB contains debugging code that is always available (but is kept out of the hot code paths). SLUB diagnostics can be enabled via the "slab_debug" option. Parameters can be specified to select a single or a group of slab caches for diagnostics. This means that the system is running with the usual performance and it is much more likely that race conditions can be reproduced. I. Resiliency If basic sanity checks are on then SLUB is capable of detecting common error conditions and recover as best as possible to allow the system to continue. J. Tracing Tracing can be enabled via the slab_debug=T,<slabcache> option during boot. SLUB will then protocol all actions on that slabcache and dump the object contents on free. K. On demand DMA cache creation. Generally DMA caches are not needed. If a kmalloc is used with __GFP_DMA then just create this single slabcache that is needed. For systems that have no ZONE_DMA requirement the support is completely eliminated. L. Performance increase Some benchmarks have shown speed improvements on kernbench in the range of 5-10%. The locking overhead of slub is based on the underlying base allocation size. If we can reliably allocate larger order pages then it is possible to increase slub performance much further. The anti-fragmentation patches may enable further performance increases. Tested on: i386 UP + SMP, x86_64 UP + SMP + NUMA emulation, IA64 NUMA + Simulator SLUB Boot options slub_nomerge Disable merging of slabs slub_min_order=x Require a minimum order for slab caches. This increases the managed chunk size and therefore reduces meta data and locking overhead. slub_min_objects=x Mininum objects per slab. Default is 8. slub_max_order=x Avoid generating slabs larger than order specified. slub_debug Enable all diagnostics for all caches slub_debug=<options> Enable selective options for all caches slub_debug=<o>,<cache> Enable selective options for a certain set of caches Available Debug options F Double Free checking, sanity and resiliency R Red zoning P Object / padding poisoning U Track last free / alloc T Trace all allocs / frees (only use for individual slabs). To use SLUB: Apply this patch and then select SLUB as the default slab allocator. [hugh@veritas.com: fix an oops-causing locking error] [akpm@linux-foundation.org: various stupid cleanups and small fixes] Signed-off-by: Christoph Lameter <clameter@sgi.com> Signed-off-by: Hugh Dickins <hugh@veritas.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2007-05-06 14:49:36 -07:00
(unsigned long)(page->offset * sizeof(void *)),
page->flags,
page->mapping,
page_count(page));
return 0;
}
if (page->inuse > s->objects) {
slub: remove object activities out of checking functions Make sure that the check function really only check things and do not perform activities. Extract the tracing and object seeding out of the two check functions and place them into slab_alloc and slab_free Signed-off-by: Christoph Lameter <clameter@sgi.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2007-05-06 14:49:47 -07:00
slab_err(s, page, "inuse %u > max %u @0x%p flags=%lx "
"mapping=0x%p count=%d",
s->name, page->inuse, s->objects, page->flags,
SLUB core This is a new slab allocator which was motivated by the complexity of the existing code in mm/slab.c. It attempts to address a variety of concerns with the existing implementation. A. Management of object queues A particular concern was the complex management of the numerous object queues in SLAB. SLUB has no such queues. Instead we dedicate a slab for each allocating CPU and use objects from a slab directly instead of queueing them up. B. Storage overhead of object queues SLAB Object queues exist per node, per CPU. The alien cache queue even has a queue array that contain a queue for each processor on each node. For very large systems the number of queues and the number of objects that may be caught in those queues grows exponentially. On our systems with 1k nodes / processors we have several gigabytes just tied up for storing references to objects for those queues This does not include the objects that could be on those queues. One fears that the whole memory of the machine could one day be consumed by those queues. C. SLAB meta data overhead SLAB has overhead at the beginning of each slab. This means that data cannot be naturally aligned at the beginning of a slab block. SLUB keeps all meta data in the corresponding page_struct. Objects can be naturally aligned in the slab. F.e. a 128 byte object will be aligned at 128 byte boundaries and can fit tightly into a 4k page with no bytes left over. SLAB cannot do this. D. SLAB has a complex cache reaper SLUB does not need a cache reaper for UP systems. On SMP systems the per CPU slab may be pushed back into partial list but that operation is simple and does not require an iteration over a list of objects. SLAB expires per CPU, shared and alien object queues during cache reaping which may cause strange hold offs. E. SLAB has complex NUMA policy layer support SLUB pushes NUMA policy handling into the page allocator. This means that allocation is coarser (SLUB does interleave on a page level) but that situation was also present before 2.6.13. SLABs application of policies to individual slab objects allocated in SLAB is certainly a performance concern due to the frequent references to memory policies which may lead a sequence of objects to come from one node after another. SLUB will get a slab full of objects from one node and then will switch to the next. F. Reduction of the size of partial slab lists SLAB has per node partial lists. This means that over time a large number of partial slabs may accumulate on those lists. These can only be reused if allocator occur on specific nodes. SLUB has a global pool of partial slabs and will consume slabs from that pool to decrease fragmentation. G. Tunables SLAB has sophisticated tuning abilities for each slab cache. One can manipulate the queue sizes in detail. However, filling the queues still requires the uses of the spin lock to check out slabs. SLUB has a global parameter (min_slab_order) for tuning. Increasing the minimum slab order can decrease the locking overhead. The bigger the slab order the less motions of pages between per CPU and partial lists occur and the better SLUB will be scaling. G. Slab merging We often have slab caches with similar parameters. SLUB detects those on boot up and merges them into the corresponding general caches. This leads to more effective memory use. About 50% of all caches can be eliminated through slab merging. This will also decrease slab fragmentation because partial allocated slabs can be filled up again. Slab merging can be switched off by specifying slub_nomerge on boot up. Note that merging can expose heretofore unknown bugs in the kernel because corrupted objects may now be placed differently and corrupt differing neighboring objects. Enable sanity checks to find those. H. Diagnostics The current slab diagnostics are difficult to use and require a recompilation of the kernel. SLUB contains debugging code that is always available (but is kept out of the hot code paths). SLUB diagnostics can be enabled via the "slab_debug" option. Parameters can be specified to select a single or a group of slab caches for diagnostics. This means that the system is running with the usual performance and it is much more likely that race conditions can be reproduced. I. Resiliency If basic sanity checks are on then SLUB is capable of detecting common error conditions and recover as best as possible to allow the system to continue. J. Tracing Tracing can be enabled via the slab_debug=T,<slabcache> option during boot. SLUB will then protocol all actions on that slabcache and dump the object contents on free. K. On demand DMA cache creation. Generally DMA caches are not needed. If a kmalloc is used with __GFP_DMA then just create this single slabcache that is needed. For systems that have no ZONE_DMA requirement the support is completely eliminated. L. Performance increase Some benchmarks have shown speed improvements on kernbench in the range of 5-10%. The locking overhead of slub is based on the underlying base allocation size. If we can reliably allocate larger order pages then it is possible to increase slub performance much further. The anti-fragmentation patches may enable further performance increases. Tested on: i386 UP + SMP, x86_64 UP + SMP + NUMA emulation, IA64 NUMA + Simulator SLUB Boot options slub_nomerge Disable merging of slabs slub_min_order=x Require a minimum order for slab caches. This increases the managed chunk size and therefore reduces meta data and locking overhead. slub_min_objects=x Mininum objects per slab. Default is 8. slub_max_order=x Avoid generating slabs larger than order specified. slub_debug Enable all diagnostics for all caches slub_debug=<options> Enable selective options for all caches slub_debug=<o>,<cache> Enable selective options for a certain set of caches Available Debug options F Double Free checking, sanity and resiliency R Red zoning P Object / padding poisoning U Track last free / alloc T Trace all allocs / frees (only use for individual slabs). To use SLUB: Apply this patch and then select SLUB as the default slab allocator. [hugh@veritas.com: fix an oops-causing locking error] [akpm@linux-foundation.org: various stupid cleanups and small fixes] Signed-off-by: Christoph Lameter <clameter@sgi.com> Signed-off-by: Hugh Dickins <hugh@veritas.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2007-05-06 14:49:36 -07:00
page->mapping, page_count(page));
return 0;
}
/* Slab_pad_check fixes things up after itself */
slab_pad_check(s, page);
return 1;
}
/*
* Determine if a certain object on a page is on the freelist and
* therefore free. Must hold the slab lock for cpu slabs to
* guarantee that the chains are consistent.
*/
static int on_freelist(struct kmem_cache *s, struct page *page, void *search)
{
int nr = 0;
void *fp = page->freelist;
void *object = NULL;
while (fp && nr <= s->objects) {
if (fp == search)
return 1;
if (!check_valid_pointer(s, page, fp)) {
if (object) {
object_err(s, page, object,
"Freechain corrupt");
set_freepointer(s, object, NULL);
break;
} else {
slub: remove object activities out of checking functions Make sure that the check function really only check things and do not perform activities. Extract the tracing and object seeding out of the two check functions and place them into slab_alloc and slab_free Signed-off-by: Christoph Lameter <clameter@sgi.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2007-05-06 14:49:47 -07:00
slab_err(s, page, "Freepointer 0x%p corrupt",
fp);
SLUB core This is a new slab allocator which was motivated by the complexity of the existing code in mm/slab.c. It attempts to address a variety of concerns with the existing implementation. A. Management of object queues A particular concern was the complex management of the numerous object queues in SLAB. SLUB has no such queues. Instead we dedicate a slab for each allocating CPU and use objects from a slab directly instead of queueing them up. B. Storage overhead of object queues SLAB Object queues exist per node, per CPU. The alien cache queue even has a queue array that contain a queue for each processor on each node. For very large systems the number of queues and the number of objects that may be caught in those queues grows exponentially. On our systems with 1k nodes / processors we have several gigabytes just tied up for storing references to objects for those queues This does not include the objects that could be on those queues. One fears that the whole memory of the machine could one day be consumed by those queues. C. SLAB meta data overhead SLAB has overhead at the beginning of each slab. This means that data cannot be naturally aligned at the beginning of a slab block. SLUB keeps all meta data in the corresponding page_struct. Objects can be naturally aligned in the slab. F.e. a 128 byte object will be aligned at 128 byte boundaries and can fit tightly into a 4k page with no bytes left over. SLAB cannot do this. D. SLAB has a complex cache reaper SLUB does not need a cache reaper for UP systems. On SMP systems the per CPU slab may be pushed back into partial list but that operation is simple and does not require an iteration over a list of objects. SLAB expires per CPU, shared and alien object queues during cache reaping which may cause strange hold offs. E. SLAB has complex NUMA policy layer support SLUB pushes NUMA policy handling into the page allocator. This means that allocation is coarser (SLUB does interleave on a page level) but that situation was also present before 2.6.13. SLABs application of policies to individual slab objects allocated in SLAB is certainly a performance concern due to the frequent references to memory policies which may lead a sequence of objects to come from one node after another. SLUB will get a slab full of objects from one node and then will switch to the next. F. Reduction of the size of partial slab lists SLAB has per node partial lists. This means that over time a large number of partial slabs may accumulate on those lists. These can only be reused if allocator occur on specific nodes. SLUB has a global pool of partial slabs and will consume slabs from that pool to decrease fragmentation. G. Tunables SLAB has sophisticated tuning abilities for each slab cache. One can manipulate the queue sizes in detail. However, filling the queues still requires the uses of the spin lock to check out slabs. SLUB has a global parameter (min_slab_order) for tuning. Increasing the minimum slab order can decrease the locking overhead. The bigger the slab order the less motions of pages between per CPU and partial lists occur and the better SLUB will be scaling. G. Slab merging We often have slab caches with similar parameters. SLUB detects those on boot up and merges them into the corresponding general caches. This leads to more effective memory use. About 50% of all caches can be eliminated through slab merging. This will also decrease slab fragmentation because partial allocated slabs can be filled up again. Slab merging can be switched off by specifying slub_nomerge on boot up. Note that merging can expose heretofore unknown bugs in the kernel because corrupted objects may now be placed differently and corrupt differing neighboring objects. Enable sanity checks to find those. H. Diagnostics The current slab diagnostics are difficult to use and require a recompilation of the kernel. SLUB contains debugging code that is always available (but is kept out of the hot code paths). SLUB diagnostics can be enabled via the "slab_debug" option. Parameters can be specified to select a single or a group of slab caches for diagnostics. This means that the system is running with the usual performance and it is much more likely that race conditions can be reproduced. I. Resiliency If basic sanity checks are on then SLUB is capable of detecting common error conditions and recover as best as possible to allow the system to continue. J. Tracing Tracing can be enabled via the slab_debug=T,<slabcache> option during boot. SLUB will then protocol all actions on that slabcache and dump the object contents on free. K. On demand DMA cache creation. Generally DMA caches are not needed. If a kmalloc is used with __GFP_DMA then just create this single slabcache that is needed. For systems that have no ZONE_DMA requirement the support is completely eliminated. L. Performance increase Some benchmarks have shown speed improvements on kernbench in the range of 5-10%. The locking overhead of slub is based on the underlying base allocation size. If we can reliably allocate larger order pages then it is possible to increase slub performance much further. The anti-fragmentation patches may enable further performance increases. Tested on: i386 UP + SMP, x86_64 UP + SMP + NUMA emulation, IA64 NUMA + Simulator SLUB Boot options slub_nomerge Disable merging of slabs slub_min_order=x Require a minimum order for slab caches. This increases the managed chunk size and therefore reduces meta data and locking overhead. slub_min_objects=x Mininum objects per slab. Default is 8. slub_max_order=x Avoid generating slabs larger than order specified. slub_debug Enable all diagnostics for all caches slub_debug=<options> Enable selective options for all caches slub_debug=<o>,<cache> Enable selective options for a certain set of caches Available Debug options F Double Free checking, sanity and resiliency R Red zoning P Object / padding poisoning U Track last free / alloc T Trace all allocs / frees (only use for individual slabs). To use SLUB: Apply this patch and then select SLUB as the default slab allocator. [hugh@veritas.com: fix an oops-causing locking error] [akpm@linux-foundation.org: various stupid cleanups and small fixes] Signed-off-by: Christoph Lameter <clameter@sgi.com> Signed-off-by: Hugh Dickins <hugh@veritas.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2007-05-06 14:49:36 -07:00
page->freelist = NULL;
page->inuse = s->objects;
slub: remove object activities out of checking functions Make sure that the check function really only check things and do not perform activities. Extract the tracing and object seeding out of the two check functions and place them into slab_alloc and slab_free Signed-off-by: Christoph Lameter <clameter@sgi.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2007-05-06 14:49:47 -07:00
printk(KERN_ERR "@@@ SLUB %s: Freelist "
"cleared. Slab 0x%p\n",
s->name, page);
SLUB core This is a new slab allocator which was motivated by the complexity of the existing code in mm/slab.c. It attempts to address a variety of concerns with the existing implementation. A. Management of object queues A particular concern was the complex management of the numerous object queues in SLAB. SLUB has no such queues. Instead we dedicate a slab for each allocating CPU and use objects from a slab directly instead of queueing them up. B. Storage overhead of object queues SLAB Object queues exist per node, per CPU. The alien cache queue even has a queue array that contain a queue for each processor on each node. For very large systems the number of queues and the number of objects that may be caught in those queues grows exponentially. On our systems with 1k nodes / processors we have several gigabytes just tied up for storing references to objects for those queues This does not include the objects that could be on those queues. One fears that the whole memory of the machine could one day be consumed by those queues. C. SLAB meta data overhead SLAB has overhead at the beginning of each slab. This means that data cannot be naturally aligned at the beginning of a slab block. SLUB keeps all meta data in the corresponding page_struct. Objects can be naturally aligned in the slab. F.e. a 128 byte object will be aligned at 128 byte boundaries and can fit tightly into a 4k page with no bytes left over. SLAB cannot do this. D. SLAB has a complex cache reaper SLUB does not need a cache reaper for UP systems. On SMP systems the per CPU slab may be pushed back into partial list but that operation is simple and does not require an iteration over a list of objects. SLAB expires per CPU, shared and alien object queues during cache reaping which may cause strange hold offs. E. SLAB has complex NUMA policy layer support SLUB pushes NUMA policy handling into the page allocator. This means that allocation is coarser (SLUB does interleave on a page level) but that situation was also present before 2.6.13. SLABs application of policies to individual slab objects allocated in SLAB is certainly a performance concern due to the frequent references to memory policies which may lead a sequence of objects to come from one node after another. SLUB will get a slab full of objects from one node and then will switch to the next. F. Reduction of the size of partial slab lists SLAB has per node partial lists. This means that over time a large number of partial slabs may accumulate on those lists. These can only be reused if allocator occur on specific nodes. SLUB has a global pool of partial slabs and will consume slabs from that pool to decrease fragmentation. G. Tunables SLAB has sophisticated tuning abilities for each slab cache. One can manipulate the queue sizes in detail. However, filling the queues still requires the uses of the spin lock to check out slabs. SLUB has a global parameter (min_slab_order) for tuning. Increasing the minimum slab order can decrease the locking overhead. The bigger the slab order the less motions of pages between per CPU and partial lists occur and the better SLUB will be scaling. G. Slab merging We often have slab caches with similar parameters. SLUB detects those on boot up and merges them into the corresponding general caches. This leads to more effective memory use. About 50% of all caches can be eliminated through slab merging. This will also decrease slab fragmentation because partial allocated slabs can be filled up again. Slab merging can be switched off by specifying slub_nomerge on boot up. Note that merging can expose heretofore unknown bugs in the kernel because corrupted objects may now be placed differently and corrupt differing neighboring objects. Enable sanity checks to find those. H. Diagnostics The current slab diagnostics are difficult to use and require a recompilation of the kernel. SLUB contains debugging code that is always available (but is kept out of the hot code paths). SLUB diagnostics can be enabled via the "slab_debug" option. Parameters can be specified to select a single or a group of slab caches for diagnostics. This means that the system is running with the usual performance and it is much more likely that race conditions can be reproduced. I. Resiliency If basic sanity checks are on then SLUB is capable of detecting common error conditions and recover as best as possible to allow the system to continue. J. Tracing Tracing can be enabled via the slab_debug=T,<slabcache> option during boot. SLUB will then protocol all actions on that slabcache and dump the object contents on free. K. On demand DMA cache creation. Generally DMA caches are not needed. If a kmalloc is used with __GFP_DMA then just create this single slabcache that is needed. For systems that have no ZONE_DMA requirement the support is completely eliminated. L. Performance increase Some benchmarks have shown speed improvements on kernbench in the range of 5-10%. The locking overhead of slub is based on the underlying base allocation size. If we can reliably allocate larger order pages then it is possible to increase slub performance much further. The anti-fragmentation patches may enable further performance increases. Tested on: i386 UP + SMP, x86_64 UP + SMP + NUMA emulation, IA64 NUMA + Simulator SLUB Boot options slub_nomerge Disable merging of slabs slub_min_order=x Require a minimum order for slab caches. This increases the managed chunk size and therefore reduces meta data and locking overhead. slub_min_objects=x Mininum objects per slab. Default is 8. slub_max_order=x Avoid generating slabs larger than order specified. slub_debug Enable all diagnostics for all caches slub_debug=<options> Enable selective options for all caches slub_debug=<o>,<cache> Enable selective options for a certain set of caches Available Debug options F Double Free checking, sanity and resiliency R Red zoning P Object / padding poisoning U Track last free / alloc T Trace all allocs / frees (only use for individual slabs). To use SLUB: Apply this patch and then select SLUB as the default slab allocator. [hugh@veritas.com: fix an oops-causing locking error] [akpm@linux-foundation.org: various stupid cleanups and small fixes] Signed-off-by: Christoph Lameter <clameter@sgi.com> Signed-off-by: Hugh Dickins <hugh@veritas.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2007-05-06 14:49:36 -07:00
return 0;
}
break;
}
object = fp;
fp = get_freepointer(s, object);
nr++;
}
if (page->inuse != s->objects - nr) {
slub: remove object activities out of checking functions Make sure that the check function really only check things and do not perform activities. Extract the tracing and object seeding out of the two check functions and place them into slab_alloc and slab_free Signed-off-by: Christoph Lameter <clameter@sgi.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2007-05-06 14:49:47 -07:00
slab_err(s, page, "Wrong object count. Counter is %d but "
"counted were %d", s, page, page->inuse,
s->objects - nr);
SLUB core This is a new slab allocator which was motivated by the complexity of the existing code in mm/slab.c. It attempts to address a variety of concerns with the existing implementation. A. Management of object queues A particular concern was the complex management of the numerous object queues in SLAB. SLUB has no such queues. Instead we dedicate a slab for each allocating CPU and use objects from a slab directly instead of queueing them up. B. Storage overhead of object queues SLAB Object queues exist per node, per CPU. The alien cache queue even has a queue array that contain a queue for each processor on each node. For very large systems the number of queues and the number of objects that may be caught in those queues grows exponentially. On our systems with 1k nodes / processors we have several gigabytes just tied up for storing references to objects for those queues This does not include the objects that could be on those queues. One fears that the whole memory of the machine could one day be consumed by those queues. C. SLAB meta data overhead SLAB has overhead at the beginning of each slab. This means that data cannot be naturally aligned at the beginning of a slab block. SLUB keeps all meta data in the corresponding page_struct. Objects can be naturally aligned in the slab. F.e. a 128 byte object will be aligned at 128 byte boundaries and can fit tightly into a 4k page with no bytes left over. SLAB cannot do this. D. SLAB has a complex cache reaper SLUB does not need a cache reaper for UP systems. On SMP systems the per CPU slab may be pushed back into partial list but that operation is simple and does not require an iteration over a list of objects. SLAB expires per CPU, shared and alien object queues during cache reaping which may cause strange hold offs. E. SLAB has complex NUMA policy layer support SLUB pushes NUMA policy handling into the page allocator. This means that allocation is coarser (SLUB does interleave on a page level) but that situation was also present before 2.6.13. SLABs application of policies to individual slab objects allocated in SLAB is certainly a performance concern due to the frequent references to memory policies which may lead a sequence of objects to come from one node after another. SLUB will get a slab full of objects from one node and then will switch to the next. F. Reduction of the size of partial slab lists SLAB has per node partial lists. This means that over time a large number of partial slabs may accumulate on those lists. These can only be reused if allocator occur on specific nodes. SLUB has a global pool of partial slabs and will consume slabs from that pool to decrease fragmentation. G. Tunables SLAB has sophisticated tuning abilities for each slab cache. One can manipulate the queue sizes in detail. However, filling the queues still requires the uses of the spin lock to check out slabs. SLUB has a global parameter (min_slab_order) for tuning. Increasing the minimum slab order can decrease the locking overhead. The bigger the slab order the less motions of pages between per CPU and partial lists occur and the better SLUB will be scaling. G. Slab merging We often have slab caches with similar parameters. SLUB detects those on boot up and merges them into the corresponding general caches. This leads to more effective memory use. About 50% of all caches can be eliminated through slab merging. This will also decrease slab fragmentation because partial allocated slabs can be filled up again. Slab merging can be switched off by specifying slub_nomerge on boot up. Note that merging can expose heretofore unknown bugs in the kernel because corrupted objects may now be placed differently and corrupt differing neighboring objects. Enable sanity checks to find those. H. Diagnostics The current slab diagnostics are difficult to use and require a recompilation of the kernel. SLUB contains debugging code that is always available (but is kept out of the hot code paths). SLUB diagnostics can be enabled via the "slab_debug" option. Parameters can be specified to select a single or a group of slab caches for diagnostics. This means that the system is running with the usual performance and it is much more likely that race conditions can be reproduced. I. Resiliency If basic sanity checks are on then SLUB is capable of detecting common error conditions and recover as best as possible to allow the system to continue. J. Tracing Tracing can be enabled via the slab_debug=T,<slabcache> option during boot. SLUB will then protocol all actions on that slabcache and dump the object contents on free. K. On demand DMA cache creation. Generally DMA caches are not needed. If a kmalloc is used with __GFP_DMA then just create this single slabcache that is needed. For systems that have no ZONE_DMA requirement the support is completely eliminated. L. Performance increase Some benchmarks have shown speed improvements on kernbench in the range of 5-10%. The locking overhead of slub is based on the underlying base allocation size. If we can reliably allocate larger order pages then it is possible to increase slub performance much further. The anti-fragmentation patches may enable further performance increases. Tested on: i386 UP + SMP, x86_64 UP + SMP + NUMA emulation, IA64 NUMA + Simulator SLUB Boot options slub_nomerge Disable merging of slabs slub_min_order=x Require a minimum order for slab caches. This increases the managed chunk size and therefore reduces meta data and locking overhead. slub_min_objects=x Mininum objects per slab. Default is 8. slub_max_order=x Avoid generating slabs larger than order specified. slub_debug Enable all diagnostics for all caches slub_debug=<options> Enable selective options for all caches slub_debug=<o>,<cache> Enable selective options for a certain set of caches Available Debug options F Double Free checking, sanity and resiliency R Red zoning P Object / padding poisoning U Track last free / alloc T Trace all allocs / frees (only use for individual slabs). To use SLUB: Apply this patch and then select SLUB as the default slab allocator. [hugh@veritas.com: fix an oops-causing locking error] [akpm@linux-foundation.org: various stupid cleanups and small fixes] Signed-off-by: Christoph Lameter <clameter@sgi.com> Signed-off-by: Hugh Dickins <hugh@veritas.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2007-05-06 14:49:36 -07:00
page->inuse = s->objects - nr;
slub: remove object activities out of checking functions Make sure that the check function really only check things and do not perform activities. Extract the tracing and object seeding out of the two check functions and place them into slab_alloc and slab_free Signed-off-by: Christoph Lameter <clameter@sgi.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2007-05-06 14:49:47 -07:00
printk(KERN_ERR "@@@ SLUB %s: Object count adjusted. "
"Slab @0x%p\n", s->name, page);
SLUB core This is a new slab allocator which was motivated by the complexity of the existing code in mm/slab.c. It attempts to address a variety of concerns with the existing implementation. A. Management of object queues A particular concern was the complex management of the numerous object queues in SLAB. SLUB has no such queues. Instead we dedicate a slab for each allocating CPU and use objects from a slab directly instead of queueing them up. B. Storage overhead of object queues SLAB Object queues exist per node, per CPU. The alien cache queue even has a queue array that contain a queue for each processor on each node. For very large systems the number of queues and the number of objects that may be caught in those queues grows exponentially. On our systems with 1k nodes / processors we have several gigabytes just tied up for storing references to objects for those queues This does not include the objects that could be on those queues. One fears that the whole memory of the machine could one day be consumed by those queues. C. SLAB meta data overhead SLAB has overhead at the beginning of each slab. This means that data cannot be naturally aligned at the beginning of a slab block. SLUB keeps all meta data in the corresponding page_struct. Objects can be naturally aligned in the slab. F.e. a 128 byte object will be aligned at 128 byte boundaries and can fit tightly into a 4k page with no bytes left over. SLAB cannot do this. D. SLAB has a complex cache reaper SLUB does not need a cache reaper for UP systems. On SMP systems the per CPU slab may be pushed back into partial list but that operation is simple and does not require an iteration over a list of objects. SLAB expires per CPU, shared and alien object queues during cache reaping which may cause strange hold offs. E. SLAB has complex NUMA policy layer support SLUB pushes NUMA policy handling into the page allocator. This means that allocation is coarser (SLUB does interleave on a page level) but that situation was also present before 2.6.13. SLABs application of policies to individual slab objects allocated in SLAB is certainly a performance concern due to the frequent references to memory policies which may lead a sequence of objects to come from one node after another. SLUB will get a slab full of objects from one node and then will switch to the next. F. Reduction of the size of partial slab lists SLAB has per node partial lists. This means that over time a large number of partial slabs may accumulate on those lists. These can only be reused if allocator occur on specific nodes. SLUB has a global pool of partial slabs and will consume slabs from that pool to decrease fragmentation. G. Tunables SLAB has sophisticated tuning abilities for each slab cache. One can manipulate the queue sizes in detail. However, filling the queues still requires the uses of the spin lock to check out slabs. SLUB has a global parameter (min_slab_order) for tuning. Increasing the minimum slab order can decrease the locking overhead. The bigger the slab order the less motions of pages between per CPU and partial lists occur and the better SLUB will be scaling. G. Slab merging We often have slab caches with similar parameters. SLUB detects those on boot up and merges them into the corresponding general caches. This leads to more effective memory use. About 50% of all caches can be eliminated through slab merging. This will also decrease slab fragmentation because partial allocated slabs can be filled up again. Slab merging can be switched off by specifying slub_nomerge on boot up. Note that merging can expose heretofore unknown bugs in the kernel because corrupted objects may now be placed differently and corrupt differing neighboring objects. Enable sanity checks to find those. H. Diagnostics The current slab diagnostics are difficult to use and require a recompilation of the kernel. SLUB contains debugging code that is always available (but is kept out of the hot code paths). SLUB diagnostics can be enabled via the "slab_debug" option. Parameters can be specified to select a single or a group of slab caches for diagnostics. This means that the system is running with the usual performance and it is much more likely that race conditions can be reproduced. I. Resiliency If basic sanity checks are on then SLUB is capable of detecting common error conditions and recover as best as possible to allow the system to continue. J. Tracing Tracing can be enabled via the slab_debug=T,<slabcache> option during boot. SLUB will then protocol all actions on that slabcache and dump the object contents on free. K. On demand DMA cache creation. Generally DMA caches are not needed. If a kmalloc is used with __GFP_DMA then just create this single slabcache that is needed. For systems that have no ZONE_DMA requirement the support is completely eliminated. L. Performance increase Some benchmarks have shown speed improvements on kernbench in the range of 5-10%. The locking overhead of slub is based on the underlying base allocation size. If we can reliably allocate larger order pages then it is possible to increase slub performance much further. The anti-fragmentation patches may enable further performance increases. Tested on: i386 UP + SMP, x86_64 UP + SMP + NUMA emulation, IA64 NUMA + Simulator SLUB Boot options slub_nomerge Disable merging of slabs slub_min_order=x Require a minimum order for slab caches. This increases the managed chunk size and therefore reduces meta data and locking overhead. slub_min_objects=x Mininum objects per slab. Default is 8. slub_max_order=x Avoid generating slabs larger than order specified. slub_debug Enable all diagnostics for all caches slub_debug=<options> Enable selective options for all caches slub_debug=<o>,<cache> Enable selective options for a certain set of caches Available Debug options F Double Free checking, sanity and resiliency R Red zoning P Object / padding poisoning U Track last free / alloc T Trace all allocs / frees (only use for individual slabs). To use SLUB: Apply this patch and then select SLUB as the default slab allocator. [hugh@veritas.com: fix an oops-causing locking error] [akpm@linux-foundation.org: various stupid cleanups and small fixes] Signed-off-by: Christoph Lameter <clameter@sgi.com> Signed-off-by: Hugh Dickins <hugh@veritas.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2007-05-06 14:49:36 -07:00
}
return search == NULL;
}
slub: enable tracking of full slabs If slab tracking is on then build a list of full slabs so that we can verify the integrity of all slabs and are also able to built list of alloc/free callers. Signed-off-by: Christoph Lameter <clameter@sgi.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2007-05-06 14:49:42 -07:00
/*
* Tracking of fully allocated slabs for debugging
*/
SLUB: Add MIN_PARTIAL We leave a mininum of partial slabs on nodes when we search for partial slabs on other node. Define a constant for that value. Then modify slub to keep MIN_PARTIAL slabs around. This avoids bad situations where a function frees the last object in a slab (which results in the page being returned to the page allocator) only to then allocate one again (which requires getting a page back from the page allocator if the partial list was empty). Keeping a couple of slabs on the partial list reduces overhead. Empty slabs are added to the end of the partial list to insure that partially allocated slabs are consumed first (defragmentation). Signed-off-by: Christoph Lameter <clameter@sgi.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2007-05-06 14:49:44 -07:00
static void add_full(struct kmem_cache_node *n, struct page *page)
slub: enable tracking of full slabs If slab tracking is on then build a list of full slabs so that we can verify the integrity of all slabs and are also able to built list of alloc/free callers. Signed-off-by: Christoph Lameter <clameter@sgi.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2007-05-06 14:49:42 -07:00
{
spin_lock(&n->list_lock);
list_add(&page->lru, &n->full);
spin_unlock(&n->list_lock);
}
static void remove_full(struct kmem_cache *s, struct page *page)
{
struct kmem_cache_node *n;
if (!(s->flags & SLAB_STORE_USER))
return;
n = get_node(s, page_to_nid(page));
spin_lock(&n->list_lock);
list_del(&page->lru);
spin_unlock(&n->list_lock);
}
SLUB core This is a new slab allocator which was motivated by the complexity of the existing code in mm/slab.c. It attempts to address a variety of concerns with the existing implementation. A. Management of object queues A particular concern was the complex management of the numerous object queues in SLAB. SLUB has no such queues. Instead we dedicate a slab for each allocating CPU and use objects from a slab directly instead of queueing them up. B. Storage overhead of object queues SLAB Object queues exist per node, per CPU. The alien cache queue even has a queue array that contain a queue for each processor on each node. For very large systems the number of queues and the number of objects that may be caught in those queues grows exponentially. On our systems with 1k nodes / processors we have several gigabytes just tied up for storing references to objects for those queues This does not include the objects that could be on those queues. One fears that the whole memory of the machine could one day be consumed by those queues. C. SLAB meta data overhead SLAB has overhead at the beginning of each slab. This means that data cannot be naturally aligned at the beginning of a slab block. SLUB keeps all meta data in the corresponding page_struct. Objects can be naturally aligned in the slab. F.e. a 128 byte object will be aligned at 128 byte boundaries and can fit tightly into a 4k page with no bytes left over. SLAB cannot do this. D. SLAB has a complex cache reaper SLUB does not need a cache reaper for UP systems. On SMP systems the per CPU slab may be pushed back into partial list but that operation is simple and does not require an iteration over a list of objects. SLAB expires per CPU, shared and alien object queues during cache reaping which may cause strange hold offs. E. SLAB has complex NUMA policy layer support SLUB pushes NUMA policy handling into the page allocator. This means that allocation is coarser (SLUB does interleave on a page level) but that situation was also present before 2.6.13. SLABs application of policies to individual slab objects allocated in SLAB is certainly a performance concern due to the frequent references to memory policies which may lead a sequence of objects to come from one node after another. SLUB will get a slab full of objects from one node and then will switch to the next. F. Reduction of the size of partial slab lists SLAB has per node partial lists. This means that over time a large number of partial slabs may accumulate on those lists. These can only be reused if allocator occur on specific nodes. SLUB has a global pool of partial slabs and will consume slabs from that pool to decrease fragmentation. G. Tunables SLAB has sophisticated tuning abilities for each slab cache. One can manipulate the queue sizes in detail. However, filling the queues still requires the uses of the spin lock to check out slabs. SLUB has a global parameter (min_slab_order) for tuning. Increasing the minimum slab order can decrease the locking overhead. The bigger the slab order the less motions of pages between per CPU and partial lists occur and the better SLUB will be scaling. G. Slab merging We often have slab caches with similar parameters. SLUB detects those on boot up and merges them into the corresponding general caches. This leads to more effective memory use. About 50% of all caches can be eliminated through slab merging. This will also decrease slab fragmentation because partial allocated slabs can be filled up again. Slab merging can be switched off by specifying slub_nomerge on boot up. Note that merging can expose heretofore unknown bugs in the kernel because corrupted objects may now be placed differently and corrupt differing neighboring objects. Enable sanity checks to find those. H. Diagnostics The current slab diagnostics are difficult to use and require a recompilation of the kernel. SLUB contains debugging code that is always available (but is kept out of the hot code paths). SLUB diagnostics can be enabled via the "slab_debug" option. Parameters can be specified to select a single or a group of slab caches for diagnostics. This means that the system is running with the usual performance and it is much more likely that race conditions can be reproduced. I. Resiliency If basic sanity checks are on then SLUB is capable of detecting common error conditions and recover as best as possible to allow the system to continue. J. Tracing Tracing can be enabled via the slab_debug=T,<slabcache> option during boot. SLUB will then protocol all actions on that slabcache and dump the object contents on free. K. On demand DMA cache creation. Generally DMA caches are not needed. If a kmalloc is used with __GFP_DMA then just create this single slabcache that is needed. For systems that have no ZONE_DMA requirement the support is completely eliminated. L. Performance increase Some benchmarks have shown speed improvements on kernbench in the range of 5-10%. The locking overhead of slub is based on the underlying base allocation size. If we can reliably allocate larger order pages then it is possible to increase slub performance much further. The anti-fragmentation patches may enable further performance increases. Tested on: i386 UP + SMP, x86_64 UP + SMP + NUMA emulation, IA64 NUMA + Simulator SLUB Boot options slub_nomerge Disable merging of slabs slub_min_order=x Require a minimum order for slab caches. This increases the managed chunk size and therefore reduces meta data and locking overhead. slub_min_objects=x Mininum objects per slab. Default is 8. slub_max_order=x Avoid generating slabs larger than order specified. slub_debug Enable all diagnostics for all caches slub_debug=<options> Enable selective options for all caches slub_debug=<o>,<cache> Enable selective options for a certain set of caches Available Debug options F Double Free checking, sanity and resiliency R Red zoning P Object / padding poisoning U Track last free / alloc T Trace all allocs / frees (only use for individual slabs). To use SLUB: Apply this patch and then select SLUB as the default slab allocator. [hugh@veritas.com: fix an oops-causing locking error] [akpm@linux-foundation.org: various stupid cleanups and small fixes] Signed-off-by: Christoph Lameter <clameter@sgi.com> Signed-off-by: Hugh Dickins <hugh@veritas.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2007-05-06 14:49:36 -07:00
static int alloc_object_checks(struct kmem_cache *s, struct page *page,
void *object)
{
if (!check_slab(s, page))
goto bad;
if (object && !on_freelist(s, page, object)) {
slub: remove object activities out of checking functions Make sure that the check function really only check things and do not perform activities. Extract the tracing and object seeding out of the two check functions and place them into slab_alloc and slab_free Signed-off-by: Christoph Lameter <clameter@sgi.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2007-05-06 14:49:47 -07:00
slab_err(s, page, "Object 0x%p already allocated", object);
goto bad;
SLUB core This is a new slab allocator which was motivated by the complexity of the existing code in mm/slab.c. It attempts to address a variety of concerns with the existing implementation. A. Management of object queues A particular concern was the complex management of the numerous object queues in SLAB. SLUB has no such queues. Instead we dedicate a slab for each allocating CPU and use objects from a slab directly instead of queueing them up. B. Storage overhead of object queues SLAB Object queues exist per node, per CPU. The alien cache queue even has a queue array that contain a queue for each processor on each node. For very large systems the number of queues and the number of objects that may be caught in those queues grows exponentially. On our systems with 1k nodes / processors we have several gigabytes just tied up for storing references to objects for those queues This does not include the objects that could be on those queues. One fears that the whole memory of the machine could one day be consumed by those queues. C. SLAB meta data overhead SLAB has overhead at the beginning of each slab. This means that data cannot be naturally aligned at the beginning of a slab block. SLUB keeps all meta data in the corresponding page_struct. Objects can be naturally aligned in the slab. F.e. a 128 byte object will be aligned at 128 byte boundaries and can fit tightly into a 4k page with no bytes left over. SLAB cannot do this. D. SLAB has a complex cache reaper SLUB does not need a cache reaper for UP systems. On SMP systems the per CPU slab may be pushed back into partial list but that operation is simple and does not require an iteration over a list of objects. SLAB expires per CPU, shared and alien object queues during cache reaping which may cause strange hold offs. E. SLAB has complex NUMA policy layer support SLUB pushes NUMA policy handling into the page allocator. This means that allocation is coarser (SLUB does interleave on a page level) but that situation was also present before 2.6.13. SLABs application of policies to individual slab objects allocated in SLAB is certainly a performance concern due to the frequent references to memory policies which may lead a sequence of objects to come from one node after another. SLUB will get a slab full of objects from one node and then will switch to the next. F. Reduction of the size of partial slab lists SLAB has per node partial lists. This means that over time a large number of partial slabs may accumulate on those lists. These can only be reused if allocator occur on specific nodes. SLUB has a global pool of partial slabs and will consume slabs from that pool to decrease fragmentation. G. Tunables SLAB has sophisticated tuning abilities for each slab cache. One can manipulate the queue sizes in detail. However, filling the queues still requires the uses of the spin lock to check out slabs. SLUB has a global parameter (min_slab_order) for tuning. Increasing the minimum slab order can decrease the locking overhead. The bigger the slab order the less motions of pages between per CPU and partial lists occur and the better SLUB will be scaling. G. Slab merging We often have slab caches with similar parameters. SLUB detects those on boot up and merges them into the corresponding general caches. This leads to more effective memory use. About 50% of all caches can be eliminated through slab merging. This will also decrease slab fragmentation because partial allocated slabs can be filled up again. Slab merging can be switched off by specifying slub_nomerge on boot up. Note that merging can expose heretofore unknown bugs in the kernel because corrupted objects may now be placed differently and corrupt differing neighboring objects. Enable sanity checks to find those. H. Diagnostics The current slab diagnostics are difficult to use and require a recompilation of the kernel. SLUB contains debugging code that is always available (but is kept out of the hot code paths). SLUB diagnostics can be enabled via the "slab_debug" option. Parameters can be specified to select a single or a group of slab caches for diagnostics. This means that the system is running with the usual performance and it is much more likely that race conditions can be reproduced. I. Resiliency If basic sanity checks are on then SLUB is capable of detecting common error conditions and recover as best as possible to allow the system to continue. J. Tracing Tracing can be enabled via the slab_debug=T,<slabcache> option during boot. SLUB will then protocol all actions on that slabcache and dump the object contents on free. K. On demand DMA cache creation. Generally DMA caches are not needed. If a kmalloc is used with __GFP_DMA then just create this single slabcache that is needed. For systems that have no ZONE_DMA requirement the support is completely eliminated. L. Performance increase Some benchmarks have shown speed improvements on kernbench in the range of 5-10%. The locking overhead of slub is based on the underlying base allocation size. If we can reliably allocate larger order pages then it is possible to increase slub performance much further. The anti-fragmentation patches may enable further performance increases. Tested on: i386 UP + SMP, x86_64 UP + SMP + NUMA emulation, IA64 NUMA + Simulator SLUB Boot options slub_nomerge Disable merging of slabs slub_min_order=x Require a minimum order for slab caches. This increases the managed chunk size and therefore reduces meta data and locking overhead. slub_min_objects=x Mininum objects per slab. Default is 8. slub_max_order=x Avoid generating slabs larger than order specified. slub_debug Enable all diagnostics for all caches slub_debug=<options> Enable selective options for all caches slub_debug=<o>,<cache> Enable selective options for a certain set of caches Available Debug options F Double Free checking, sanity and resiliency R Red zoning P Object / padding poisoning U Track last free / alloc T Trace all allocs / frees (only use for individual slabs). To use SLUB: Apply this patch and then select SLUB as the default slab allocator. [hugh@veritas.com: fix an oops-causing locking error] [akpm@linux-foundation.org: various stupid cleanups and small fixes] Signed-off-by: Christoph Lameter <clameter@sgi.com> Signed-off-by: Hugh Dickins <hugh@veritas.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2007-05-06 14:49:36 -07:00
}
if (!check_valid_pointer(s, page, object)) {
object_err(s, page, object, "Freelist Pointer check fails");
slub: remove object activities out of checking functions Make sure that the check function really only check things and do not perform activities. Extract the tracing and object seeding out of the two check functions and place them into slab_alloc and slab_free Signed-off-by: Christoph Lameter <clameter@sgi.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2007-05-06 14:49:47 -07:00
goto bad;
SLUB core This is a new slab allocator which was motivated by the complexity of the existing code in mm/slab.c. It attempts to address a variety of concerns with the existing implementation. A. Management of object queues A particular concern was the complex management of the numerous object queues in SLAB. SLUB has no such queues. Instead we dedicate a slab for each allocating CPU and use objects from a slab directly instead of queueing them up. B. Storage overhead of object queues SLAB Object queues exist per node, per CPU. The alien cache queue even has a queue array that contain a queue for each processor on each node. For very large systems the number of queues and the number of objects that may be caught in those queues grows exponentially. On our systems with 1k nodes / processors we have several gigabytes just tied up for storing references to objects for those queues This does not include the objects that could be on those queues. One fears that the whole memory of the machine could one day be consumed by those queues. C. SLAB meta data overhead SLAB has overhead at the beginning of each slab. This means that data cannot be naturally aligned at the beginning of a slab block. SLUB keeps all meta data in the corresponding page_struct. Objects can be naturally aligned in the slab. F.e. a 128 byte object will be aligned at 128 byte boundaries and can fit tightly into a 4k page with no bytes left over. SLAB cannot do this. D. SLAB has a complex cache reaper SLUB does not need a cache reaper for UP systems. On SMP systems the per CPU slab may be pushed back into partial list but that operation is simple and does not require an iteration over a list of objects. SLAB expires per CPU, shared and alien object queues during cache reaping which may cause strange hold offs. E. SLAB has complex NUMA policy layer support SLUB pushes NUMA policy handling into the page allocator. This means that allocation is coarser (SLUB does interleave on a page level) but that situation was also present before 2.6.13. SLABs application of policies to individual slab objects allocated in SLAB is certainly a performance concern due to the frequent references to memory policies which may lead a sequence of objects to come from one node after another. SLUB will get a slab full of objects from one node and then will switch to the next. F. Reduction of the size of partial slab lists SLAB has per node partial lists. This means that over time a large number of partial slabs may accumulate on those lists. These can only be reused if allocator occur on specific nodes. SLUB has a global pool of partial slabs and will consume slabs from that pool to decrease fragmentation. G. Tunables SLAB has sophisticated tuning abilities for each slab cache. One can manipulate the queue sizes in detail. However, filling the queues still requires the uses of the spin lock to check out slabs. SLUB has a global parameter (min_slab_order) for tuning. Increasing the minimum slab order can decrease the locking overhead. The bigger the slab order the less motions of pages between per CPU and partial lists occur and the better SLUB will be scaling. G. Slab merging We often have slab caches with similar parameters. SLUB detects those on boot up and merges them into the corresponding general caches. This leads to more effective memory use. About 50% of all caches can be eliminated through slab merging. This will also decrease slab fragmentation because partial allocated slabs can be filled up again. Slab merging can be switched off by specifying slub_nomerge on boot up. Note that merging can expose heretofore unknown bugs in the kernel because corrupted objects may now be placed differently and corrupt differing neighboring objects. Enable sanity checks to find those. H. Diagnostics The current slab diagnostics are difficult to use and require a recompilation of the kernel. SLUB contains debugging code that is always available (but is kept out of the hot code paths). SLUB diagnostics can be enabled via the "slab_debug" option. Parameters can be specified to select a single or a group of slab caches for diagnostics. This means that the system is running with the usual performance and it is much more likely that race conditions can be reproduced. I. Resiliency If basic sanity checks are on then SLUB is capable of detecting common error conditions and recover as best as possible to allow the system to continue. J. Tracing Tracing can be enabled via the slab_debug=T,<slabcache> option during boot. SLUB will then protocol all actions on that slabcache and dump the object contents on free. K. On demand DMA cache creation. Generally DMA caches are not needed. If a kmalloc is used with __GFP_DMA then just create this single slabcache that is needed. For systems that have no ZONE_DMA requirement the support is completely eliminated. L. Performance increase Some benchmarks have shown speed improvements on kernbench in the range of 5-10%. The locking overhead of slub is based on the underlying base allocation size. If we can reliably allocate larger order pages then it is possible to increase slub performance much further. The anti-fragmentation patches may enable further performance increases. Tested on: i386 UP + SMP, x86_64 UP + SMP + NUMA emulation, IA64 NUMA + Simulator SLUB Boot options slub_nomerge Disable merging of slabs slub_min_order=x Require a minimum order for slab caches. This increases the managed chunk size and therefore reduces meta data and locking overhead. slub_min_objects=x Mininum objects per slab. Default is 8. slub_max_order=x Avoid generating slabs larger than order specified. slub_debug Enable all diagnostics for all caches slub_debug=<options> Enable selective options for all caches slub_debug=<o>,<cache> Enable selective options for a certain set of caches Available Debug options F Double Free checking, sanity and resiliency R Red zoning P Object / padding poisoning U Track last free / alloc T Trace all allocs / frees (only use for individual slabs). To use SLUB: Apply this patch and then select SLUB as the default slab allocator. [hugh@veritas.com: fix an oops-causing locking error] [akpm@linux-foundation.org: various stupid cleanups and small fixes] Signed-off-by: Christoph Lameter <clameter@sgi.com> Signed-off-by: Hugh Dickins <hugh@veritas.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2007-05-06 14:49:36 -07:00
}
if (!object)
return 1;
if (!check_object(s, page, object, 0))
goto bad;
return 1;
bad:
if (PageSlab(page)) {
/*
* If this is a slab page then lets do the best we can
* to avoid issues in the future. Marking all objects
* as used avoids touching the remainder.
*/
printk(KERN_ERR "@@@ SLUB: %s slab 0x%p. Marking all objects used.\n",
s->name, page);
page->inuse = s->objects;
page->freelist = NULL;
/* Fix up fields that may be corrupted */
page->offset = s->offset / sizeof(void *);
}
return 0;
}
static int free_object_checks(struct kmem_cache *s, struct page *page,
void *object)
{
if (!check_slab(s, page))
goto fail;
if (!check_valid_pointer(s, page, object)) {
slub: remove object activities out of checking functions Make sure that the check function really only check things and do not perform activities. Extract the tracing and object seeding out of the two check functions and place them into slab_alloc and slab_free Signed-off-by: Christoph Lameter <clameter@sgi.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2007-05-06 14:49:47 -07:00
slab_err(s, page, "Invalid object pointer 0x%p", object);
SLUB core This is a new slab allocator which was motivated by the complexity of the existing code in mm/slab.c. It attempts to address a variety of concerns with the existing implementation. A. Management of object queues A particular concern was the complex management of the numerous object queues in SLAB. SLUB has no such queues. Instead we dedicate a slab for each allocating CPU and use objects from a slab directly instead of queueing them up. B. Storage overhead of object queues SLAB Object queues exist per node, per CPU. The alien cache queue even has a queue array that contain a queue for each processor on each node. For very large systems the number of queues and the number of objects that may be caught in those queues grows exponentially. On our systems with 1k nodes / processors we have several gigabytes just tied up for storing references to objects for those queues This does not include the objects that could be on those queues. One fears that the whole memory of the machine could one day be consumed by those queues. C. SLAB meta data overhead SLAB has overhead at the beginning of each slab. This means that data cannot be naturally aligned at the beginning of a slab block. SLUB keeps all meta data in the corresponding page_struct. Objects can be naturally aligned in the slab. F.e. a 128 byte object will be aligned at 128 byte boundaries and can fit tightly into a 4k page with no bytes left over. SLAB cannot do this. D. SLAB has a complex cache reaper SLUB does not need a cache reaper for UP systems. On SMP systems the per CPU slab may be pushed back into partial list but that operation is simple and does not require an iteration over a list of objects. SLAB expires per CPU, shared and alien object queues during cache reaping which may cause strange hold offs. E. SLAB has complex NUMA policy layer support SLUB pushes NUMA policy handling into the page allocator. This means that allocation is coarser (SLUB does interleave on a page level) but that situation was also present before 2.6.13. SLABs application of policies to individual slab objects allocated in SLAB is certainly a performance concern due to the frequent references to memory policies which may lead a sequence of objects to come from one node after another. SLUB will get a slab full of objects from one node and then will switch to the next. F. Reduction of the size of partial slab lists SLAB has per node partial lists. This means that over time a large number of partial slabs may accumulate on those lists. These can only be reused if allocator occur on specific nodes. SLUB has a global pool of partial slabs and will consume slabs from that pool to decrease fragmentation. G. Tunables SLAB has sophisticated tuning abilities for each slab cache. One can manipulate the queue sizes in detail. However, filling the queues still requires the uses of the spin lock to check out slabs. SLUB has a global parameter (min_slab_order) for tuning. Increasing the minimum slab order can decrease the locking overhead. The bigger the slab order the less motions of pages between per CPU and partial lists occur and the better SLUB will be scaling. G. Slab merging We often have slab caches with similar parameters. SLUB detects those on boot up and merges them into the corresponding general caches. This leads to more effective memory use. About 50% of all caches can be eliminated through slab merging. This will also decrease slab fragmentation because partial allocated slabs can be filled up again. Slab merging can be switched off by specifying slub_nomerge on boot up. Note that merging can expose heretofore unknown bugs in the kernel because corrupted objects may now be placed differently and corrupt differing neighboring objects. Enable sanity checks to find those. H. Diagnostics The current slab diagnostics are difficult to use and require a recompilation of the kernel. SLUB contains debugging code that is always available (but is kept out of the hot code paths). SLUB diagnostics can be enabled via the "slab_debug" option. Parameters can be specified to select a single or a group of slab caches for diagnostics. This means that the system is running with the usual performance and it is much more likely that race conditions can be reproduced. I. Resiliency If basic sanity checks are on then SLUB is capable of detecting common error conditions and recover as best as possible to allow the system to continue. J. Tracing Tracing can be enabled via the slab_debug=T,<slabcache> option during boot. SLUB will then protocol all actions on that slabcache and dump the object contents on free. K. On demand DMA cache creation. Generally DMA caches are not needed. If a kmalloc is used with __GFP_DMA then just create this single slabcache that is needed. For systems that have no ZONE_DMA requirement the support is completely eliminated. L. Performance increase Some benchmarks have shown speed improvements on kernbench in the range of 5-10%. The locking overhead of slub is based on the underlying base allocation size. If we can reliably allocate larger order pages then it is possible to increase slub performance much further. The anti-fragmentation patches may enable further performance increases. Tested on: i386 UP + SMP, x86_64 UP + SMP + NUMA emulation, IA64 NUMA + Simulator SLUB Boot options slub_nomerge Disable merging of slabs slub_min_order=x Require a minimum order for slab caches. This increases the managed chunk size and therefore reduces meta data and locking overhead. slub_min_objects=x Mininum objects per slab. Default is 8. slub_max_order=x Avoid generating slabs larger than order specified. slub_debug Enable all diagnostics for all caches slub_debug=<options> Enable selective options for all caches slub_debug=<o>,<cache> Enable selective options for a certain set of caches Available Debug options F Double Free checking, sanity and resiliency R Red zoning P Object / padding poisoning U Track last free / alloc T Trace all allocs / frees (only use for individual slabs). To use SLUB: Apply this patch and then select SLUB as the default slab allocator. [hugh@veritas.com: fix an oops-causing locking error] [akpm@linux-foundation.org: various stupid cleanups and small fixes] Signed-off-by: Christoph Lameter <clameter@sgi.com> Signed-off-by: Hugh Dickins <hugh@veritas.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2007-05-06 14:49:36 -07:00
goto fail;
}
if (on_freelist(s, page, object)) {
slub: remove object activities out of checking functions Make sure that the check function really only check things and do not perform activities. Extract the tracing and object seeding out of the two check functions and place them into slab_alloc and slab_free Signed-off-by: Christoph Lameter <clameter@sgi.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2007-05-06 14:49:47 -07:00
slab_err(s, page, "Object 0x%p already free", object);
SLUB core This is a new slab allocator which was motivated by the complexity of the existing code in mm/slab.c. It attempts to address a variety of concerns with the existing implementation. A. Management of object queues A particular concern was the complex management of the numerous object queues in SLAB. SLUB has no such queues. Instead we dedicate a slab for each allocating CPU and use objects from a slab directly instead of queueing them up. B. Storage overhead of object queues SLAB Object queues exist per node, per CPU. The alien cache queue even has a queue array that contain a queue for each processor on each node. For very large systems the number of queues and the number of objects that may be caught in those queues grows exponentially. On our systems with 1k nodes / processors we have several gigabytes just tied up for storing references to objects for those queues This does not include the objects that could be on those queues. One fears that the whole memory of the machine could one day be consumed by those queues. C. SLAB meta data overhead SLAB has overhead at the beginning of each slab. This means that data cannot be naturally aligned at the beginning of a slab block. SLUB keeps all meta data in the corresponding page_struct. Objects can be naturally aligned in the slab. F.e. a 128 byte object will be aligned at 128 byte boundaries and can fit tightly into a 4k page with no bytes left over. SLAB cannot do this. D. SLAB has a complex cache reaper SLUB does not need a cache reaper for UP systems. On SMP systems the per CPU slab may be pushed back into partial list but that operation is simple and does not require an iteration over a list of objects. SLAB expires per CPU, shared and alien object queues during cache reaping which may cause strange hold offs. E. SLAB has complex NUMA policy layer support SLUB pushes NUMA policy handling into the page allocator. This means that allocation is coarser (SLUB does interleave on a page level) but that situation was also present before 2.6.13. SLABs application of policies to individual slab objects allocated in SLAB is certainly a performance concern due to the frequent references to memory policies which may lead a sequence of objects to come from one node after another. SLUB will get a slab full of objects from one node and then will switch to the next. F. Reduction of the size of partial slab lists SLAB has per node partial lists. This means that over time a large number of partial slabs may accumulate on those lists. These can only be reused if allocator occur on specific nodes. SLUB has a global pool of partial slabs and will consume slabs from that pool to decrease fragmentation. G. Tunables SLAB has sophisticated tuning abilities for each slab cache. One can manipulate the queue sizes in detail. However, filling the queues still requires the uses of the spin lock to check out slabs. SLUB has a global parameter (min_slab_order) for tuning. Increasing the minimum slab order can decrease the locking overhead. The bigger the slab order the less motions of pages between per CPU and partial lists occur and the better SLUB will be scaling. G. Slab merging We often have slab caches with similar parameters. SLUB detects those on boot up and merges them into the corresponding general caches. This leads to more effective memory use. About 50% of all caches can be eliminated through slab merging. This will also decrease slab fragmentation because partial allocated slabs can be filled up again. Slab merging can be switched off by specifying slub_nomerge on boot up. Note that merging can expose heretofore unknown bugs in the kernel because corrupted objects may now be placed differently and corrupt differing neighboring objects. Enable sanity checks to find those. H. Diagnostics The current slab diagnostics are difficult to use and require a recompilation of the kernel. SLUB contains debugging code that is always available (but is kept out of the hot code paths). SLUB diagnostics can be enabled via the "slab_debug" option. Parameters can be specified to select a single or a group of slab caches for diagnostics. This means that the system is running with the usual performance and it is much more likely that race conditions can be reproduced. I. Resiliency If basic sanity checks are on then SLUB is capable of detecting common error conditions and recover as best as possible to allow the system to continue. J. Tracing Tracing can be enabled via the slab_debug=T,<slabcache> option during boot. SLUB will then protocol all actions on that slabcache and dump the object contents on free. K. On demand DMA cache creation. Generally DMA caches are not needed. If a kmalloc is used with __GFP_DMA then just create this single slabcache that is needed. For systems that have no ZONE_DMA requirement the support is completely eliminated. L. Performance increase Some benchmarks have shown speed improvements on kernbench in the range of 5-10%. The locking overhead of slub is based on the underlying base allocation size. If we can reliably allocate larger order pages then it is possible to increase slub performance much further. The anti-fragmentation patches may enable further performance increases. Tested on: i386 UP + SMP, x86_64 UP + SMP + NUMA emulation, IA64 NUMA + Simulator SLUB Boot options slub_nomerge Disable merging of slabs slub_min_order=x Require a minimum order for slab caches. This increases the managed chunk size and therefore reduces meta data and locking overhead. slub_min_objects=x Mininum objects per slab. Default is 8. slub_max_order=x Avoid generating slabs larger than order specified. slub_debug Enable all diagnostics for all caches slub_debug=<options> Enable selective options for all caches slub_debug=<o>,<cache> Enable selective options for a certain set of caches Available Debug options F Double Free checking, sanity and resiliency R Red zoning P Object / padding poisoning U Track last free / alloc T Trace all allocs / frees (only use for individual slabs). To use SLUB: Apply this patch and then select SLUB as the default slab allocator. [hugh@veritas.com: fix an oops-causing locking error] [akpm@linux-foundation.org: various stupid cleanups and small fixes] Signed-off-by: Christoph Lameter <clameter@sgi.com> Signed-off-by: Hugh Dickins <hugh@veritas.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2007-05-06 14:49:36 -07:00
goto fail;
}
if (!check_object(s, page, object, 1))
return 0;
if (unlikely(s != page->slab)) {
if (!PageSlab(page))
slub: remove object activities out of checking functions Make sure that the check function really only check things and do not perform activities. Extract the tracing and object seeding out of the two check functions and place them into slab_alloc and slab_free Signed-off-by: Christoph Lameter <clameter@sgi.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2007-05-06 14:49:47 -07:00
slab_err(s, page, "Attempt to free object(0x%p) "
"outside of slab", object);
SLUB core This is a new slab allocator which was motivated by the complexity of the existing code in mm/slab.c. It attempts to address a variety of concerns with the existing implementation. A. Management of object queues A particular concern was the complex management of the numerous object queues in SLAB. SLUB has no such queues. Instead we dedicate a slab for each allocating CPU and use objects from a slab directly instead of queueing them up. B. Storage overhead of object queues SLAB Object queues exist per node, per CPU. The alien cache queue even has a queue array that contain a queue for each processor on each node. For very large systems the number of queues and the number of objects that may be caught in those queues grows exponentially. On our systems with 1k nodes / processors we have several gigabytes just tied up for storing references to objects for those queues This does not include the objects that could be on those queues. One fears that the whole memory of the machine could one day be consumed by those queues. C. SLAB meta data overhead SLAB has overhead at the beginning of each slab. This means that data cannot be naturally aligned at the beginning of a slab block. SLUB keeps all meta data in the corresponding page_struct. Objects can be naturally aligned in the slab. F.e. a 128 byte object will be aligned at 128 byte boundaries and can fit tightly into a 4k page with no bytes left over. SLAB cannot do this. D. SLAB has a complex cache reaper SLUB does not need a cache reaper for UP systems. On SMP systems the per CPU slab may be pushed back into partial list but that operation is simple and does not require an iteration over a list of objects. SLAB expires per CPU, shared and alien object queues during cache reaping which may cause strange hold offs. E. SLAB has complex NUMA policy layer support SLUB pushes NUMA policy handling into the page allocator. This means that allocation is coarser (SLUB does interleave on a page level) but that situation was also present before 2.6.13. SLABs application of policies to individual slab objects allocated in SLAB is certainly a performance concern due to the frequent references to memory policies which may lead a sequence of objects to come from one node after another. SLUB will get a slab full of objects from one node and then will switch to the next. F. Reduction of the size of partial slab lists SLAB has per node partial lists. This means that over time a large number of partial slabs may accumulate on those lists. These can only be reused if allocator occur on specific nodes. SLUB has a global pool of partial slabs and will consume slabs from that pool to decrease fragmentation. G. Tunables SLAB has sophisticated tuning abilities for each slab cache. One can manipulate the queue sizes in detail. However, filling the queues still requires the uses of the spin lock to check out slabs. SLUB has a global parameter (min_slab_order) for tuning. Increasing the minimum slab order can decrease the locking overhead. The bigger the slab order the less motions of pages between per CPU and partial lists occur and the better SLUB will be scaling. G. Slab merging We often have slab caches with similar parameters. SLUB detects those on boot up and merges them into the corresponding general caches. This leads to more effective memory use. About 50% of all caches can be eliminated through slab merging. This will also decrease slab fragmentation because partial allocated slabs can be filled up again. Slab merging can be switched off by specifying slub_nomerge on boot up. Note that merging can expose heretofore unknown bugs in the kernel because corrupted objects may now be placed differently and corrupt differing neighboring objects. Enable sanity checks to find those. H. Diagnostics The current slab diagnostics are difficult to use and require a recompilation of the kernel. SLUB contains debugging code that is always available (but is kept out of the hot code paths). SLUB diagnostics can be enabled via the "slab_debug" option. Parameters can be specified to select a single or a group of slab caches for diagnostics. This means that the system is running with the usual performance and it is much more likely that race conditions can be reproduced. I. Resiliency If basic sanity checks are on then SLUB is capable of detecting common error conditions and recover as best as possible to allow the system to continue. J. Tracing Tracing can be enabled via the slab_debug=T,<slabcache> option during boot. SLUB will then protocol all actions on that slabcache and dump the object contents on free. K. On demand DMA cache creation. Generally DMA caches are not needed. If a kmalloc is used with __GFP_DMA then just create this single slabcache that is needed. For systems that have no ZONE_DMA requirement the support is completely eliminated. L. Performance increase Some benchmarks have shown speed improvements on kernbench in the range of 5-10%. The locking overhead of slub is based on the underlying base allocation size. If we can reliably allocate larger order pages then it is possible to increase slub performance much further. The anti-fragmentation patches may enable further performance increases. Tested on: i386 UP + SMP, x86_64 UP + SMP + NUMA emulation, IA64 NUMA + Simulator SLUB Boot options slub_nomerge Disable merging of slabs slub_min_order=x Require a minimum order for slab caches. This increases the managed chunk size and therefore reduces meta data and locking overhead. slub_min_objects=x Mininum objects per slab. Default is 8. slub_max_order=x Avoid generating slabs larger than order specified. slub_debug Enable all diagnostics for all caches slub_debug=<options> Enable selective options for all caches slub_debug=<o>,<cache> Enable selective options for a certain set of caches Available Debug options F Double Free checking, sanity and resiliency R Red zoning P Object / padding poisoning U Track last free / alloc T Trace all allocs / frees (only use for individual slabs). To use SLUB: Apply this patch and then select SLUB as the default slab allocator. [hugh@veritas.com: fix an oops-causing locking error] [akpm@linux-foundation.org: various stupid cleanups and small fixes] Signed-off-by: Christoph Lameter <clameter@sgi.com> Signed-off-by: Hugh Dickins <hugh@veritas.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2007-05-06 14:49:36 -07:00
else
slub: remove object activities out of checking functions Make sure that the check function really only check things and do not perform activities. Extract the tracing and object seeding out of the two check functions and place them into slab_alloc and slab_free Signed-off-by: Christoph Lameter <clameter@sgi.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2007-05-06 14:49:47 -07:00
if (!page->slab) {
SLUB core This is a new slab allocator which was motivated by the complexity of the existing code in mm/slab.c. It attempts to address a variety of concerns with the existing implementation. A. Management of object queues A particular concern was the complex management of the numerous object queues in SLAB. SLUB has no such queues. Instead we dedicate a slab for each allocating CPU and use objects from a slab directly instead of queueing them up. B. Storage overhead of object queues SLAB Object queues exist per node, per CPU. The alien cache queue even has a queue array that contain a queue for each processor on each node. For very large systems the number of queues and the number of objects that may be caught in those queues grows exponentially. On our systems with 1k nodes / processors we have several gigabytes just tied up for storing references to objects for those queues This does not include the objects that could be on those queues. One fears that the whole memory of the machine could one day be consumed by those queues. C. SLAB meta data overhead SLAB has overhead at the beginning of each slab. This means that data cannot be naturally aligned at the beginning of a slab block. SLUB keeps all meta data in the corresponding page_struct. Objects can be naturally aligned in the slab. F.e. a 128 byte object will be aligned at 128 byte boundaries and can fit tightly into a 4k page with no bytes left over. SLAB cannot do this. D. SLAB has a complex cache reaper SLUB does not need a cache reaper for UP systems. On SMP systems the per CPU slab may be pushed back into partial list but that operation is simple and does not require an iteration over a list of objects. SLAB expires per CPU, shared and alien object queues during cache reaping which may cause strange hold offs. E. SLAB has complex NUMA policy layer support SLUB pushes NUMA policy handling into the page allocator. This means that allocation is coarser (SLUB does interleave on a page level) but that situation was also present before 2.6.13. SLABs application of policies to individual slab objects allocated in SLAB is certainly a performance concern due to the frequent references to memory policies which may lead a sequence of objects to come from one node after another. SLUB will get a slab full of objects from one node and then will switch to the next. F. Reduction of the size of partial slab lists SLAB has per node partial lists. This means that over time a large number of partial slabs may accumulate on those lists. These can only be reused if allocator occur on specific nodes. SLUB has a global pool of partial slabs and will consume slabs from that pool to decrease fragmentation. G. Tunables SLAB has sophisticated tuning abilities for each slab cache. One can manipulate the queue sizes in detail. However, filling the queues still requires the uses of the spin lock to check out slabs. SLUB has a global parameter (min_slab_order) for tuning. Increasing the minimum slab order can decrease the locking overhead. The bigger the slab order the less motions of pages between per CPU and partial lists occur and the better SLUB will be scaling. G. Slab merging We often have slab caches with similar parameters. SLUB detects those on boot up and merges them into the corresponding general caches. This leads to more effective memory use. About 50% of all caches can be eliminated through slab merging. This will also decrease slab fragmentation because partial allocated slabs can be filled up again. Slab merging can be switched off by specifying slub_nomerge on boot up. Note that merging can expose heretofore unknown bugs in the kernel because corrupted objects may now be placed differently and corrupt differing neighboring objects. Enable sanity checks to find those. H. Diagnostics The current slab diagnostics are difficult to use and require a recompilation of the kernel. SLUB contains debugging code that is always available (but is kept out of the hot code paths). SLUB diagnostics can be enabled via the "slab_debug" option. Parameters can be specified to select a single or a group of slab caches for diagnostics. This means that the system is running with the usual performance and it is much more likely that race conditions can be reproduced. I. Resiliency If basic sanity checks are on then SLUB is capable of detecting common error conditions and recover as best as possible to allow the system to continue. J. Tracing Tracing can be enabled via the slab_debug=T,<slabcache> option during boot. SLUB will then protocol all actions on that slabcache and dump the object contents on free. K. On demand DMA cache creation. Generally DMA caches are not needed. If a kmalloc is used with __GFP_DMA then just create this single slabcache that is needed. For systems that have no ZONE_DMA requirement the support is completely eliminated. L. Performance increase Some benchmarks have shown speed improvements on kernbench in the range of 5-10%. The locking overhead of slub is based on the underlying base allocation size. If we can reliably allocate larger order pages then it is possible to increase slub performance much further. The anti-fragmentation patches may enable further performance increases. Tested on: i386 UP + SMP, x86_64 UP + SMP + NUMA emulation, IA64 NUMA + Simulator SLUB Boot options slub_nomerge Disable merging of slabs slub_min_order=x Require a minimum order for slab caches. This increases the managed chunk size and therefore reduces meta data and locking overhead. slub_min_objects=x Mininum objects per slab. Default is 8. slub_max_order=x Avoid generating slabs larger than order specified. slub_debug Enable all diagnostics for all caches slub_debug=<options> Enable selective options for all caches slub_debug=<o>,<cache> Enable selective options for a certain set of caches Available Debug options F Double Free checking, sanity and resiliency R Red zoning P Object / padding poisoning U Track last free / alloc T Trace all allocs / frees (only use for individual slabs). To use SLUB: Apply this patch and then select SLUB as the default slab allocator. [hugh@veritas.com: fix an oops-causing locking error] [akpm@linux-foundation.org: various stupid cleanups and small fixes] Signed-off-by: Christoph Lameter <clameter@sgi.com> Signed-off-by: Hugh Dickins <hugh@veritas.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2007-05-06 14:49:36 -07:00
printk(KERN_ERR
slub: remove object activities out of checking functions Make sure that the check function really only check things and do not perform activities. Extract the tracing and object seeding out of the two check functions and place them into slab_alloc and slab_free Signed-off-by: Christoph Lameter <clameter@sgi.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2007-05-06 14:49:47 -07:00
"SLUB <none>: no slab for object 0x%p.\n",
SLUB core This is a new slab allocator which was motivated by the complexity of the existing code in mm/slab.c. It attempts to address a variety of concerns with the existing implementation. A. Management of object queues A particular concern was the complex management of the numerous object queues in SLAB. SLUB has no such queues. Instead we dedicate a slab for each allocating CPU and use objects from a slab directly instead of queueing them up. B. Storage overhead of object queues SLAB Object queues exist per node, per CPU. The alien cache queue even has a queue array that contain a queue for each processor on each node. For very large systems the number of queues and the number of objects that may be caught in those queues grows exponentially. On our systems with 1k nodes / processors we have several gigabytes just tied up for storing references to objects for those queues This does not include the objects that could be on those queues. One fears that the whole memory of the machine could one day be consumed by those queues. C. SLAB meta data overhead SLAB has overhead at the beginning of each slab. This means that data cannot be naturally aligned at the beginning of a slab block. SLUB keeps all meta data in the corresponding page_struct. Objects can be naturally aligned in the slab. F.e. a 128 byte object will be aligned at 128 byte boundaries and can fit tightly into a 4k page with no bytes left over. SLAB cannot do this. D. SLAB has a complex cache reaper SLUB does not need a cache reaper for UP systems. On SMP systems the per CPU slab may be pushed back into partial list but that operation is simple and does not require an iteration over a list of objects. SLAB expires per CPU, shared and alien object queues during cache reaping which may cause strange hold offs. E. SLAB has complex NUMA policy layer support SLUB pushes NUMA policy handling into the page allocator. This means that allocation is coarser (SLUB does interleave on a page level) but that situation was also present before 2.6.13. SLABs application of policies to individual slab objects allocated in SLAB is certainly a performance concern due to the frequent references to memory policies which may lead a sequence of objects to come from one node after another. SLUB will get a slab full of objects from one node and then will switch to the next. F. Reduction of the size of partial slab lists SLAB has per node partial lists. This means that over time a large number of partial slabs may accumulate on those lists. These can only be reused if allocator occur on specific nodes. SLUB has a global pool of partial slabs and will consume slabs from that pool to decrease fragmentation. G. Tunables SLAB has sophisticated tuning abilities for each slab cache. One can manipulate the queue sizes in detail. However, filling the queues still requires the uses of the spin lock to check out slabs. SLUB has a global parameter (min_slab_order) for tuning. Increasing the minimum slab order can decrease the locking overhead. The bigger the slab order the less motions of pages between per CPU and partial lists occur and the better SLUB will be scaling. G. Slab merging We often have slab caches with similar parameters. SLUB detects those on boot up and merges them into the corresponding general caches. This leads to more effective memory use. About 50% of all caches can be eliminated through slab merging. This will also decrease slab fragmentation because partial allocated slabs can be filled up again. Slab merging can be switched off by specifying slub_nomerge on boot up. Note that merging can expose heretofore unknown bugs in the kernel because corrupted objects may now be placed differently and corrupt differing neighboring objects. Enable sanity checks to find those. H. Diagnostics The current slab diagnostics are difficult to use and require a recompilation of the kernel. SLUB contains debugging code that is always available (but is kept out of the hot code paths). SLUB diagnostics can be enabled via the "slab_debug" option. Parameters can be specified to select a single or a group of slab caches for diagnostics. This means that the system is running with the usual performance and it is much more likely that race conditions can be reproduced. I. Resiliency If basic sanity checks are on then SLUB is capable of detecting common error conditions and recover as best as possible to allow the system to continue. J. Tracing Tracing can be enabled via the slab_debug=T,<slabcache> option during boot. SLUB will then protocol all actions on that slabcache and dump the object contents on free. K. On demand DMA cache creation. Generally DMA caches are not needed. If a kmalloc is used with __GFP_DMA then just create this single slabcache that is needed. For systems that have no ZONE_DMA requirement the support is completely eliminated. L. Performance increase Some benchmarks have shown speed improvements on kernbench in the range of 5-10%. The locking overhead of slub is based on the underlying base allocation size. If we can reliably allocate larger order pages then it is possible to increase slub performance much further. The anti-fragmentation patches may enable further performance increases. Tested on: i386 UP + SMP, x86_64 UP + SMP + NUMA emulation, IA64 NUMA + Simulator SLUB Boot options slub_nomerge Disable merging of slabs slub_min_order=x Require a minimum order for slab caches. This increases the managed chunk size and therefore reduces meta data and locking overhead. slub_min_objects=x Mininum objects per slab. Default is 8. slub_max_order=x Avoid generating slabs larger than order specified. slub_debug Enable all diagnostics for all caches slub_debug=<options> Enable selective options for all caches slub_debug=<o>,<cache> Enable selective options for a certain set of caches Available Debug options F Double Free checking, sanity and resiliency R Red zoning P Object / padding poisoning U Track last free / alloc T Trace all allocs / frees (only use for individual slabs). To use SLUB: Apply this patch and then select SLUB as the default slab allocator. [hugh@veritas.com: fix an oops-causing locking error] [akpm@linux-foundation.org: various stupid cleanups and small fixes] Signed-off-by: Christoph Lameter <clameter@sgi.com> Signed-off-by: Hugh Dickins <hugh@veritas.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2007-05-06 14:49:36 -07:00
object);
slub: remove object activities out of checking functions Make sure that the check function really only check things and do not perform activities. Extract the tracing and object seeding out of the two check functions and place them into slab_alloc and slab_free Signed-off-by: Christoph Lameter <clameter@sgi.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2007-05-06 14:49:47 -07:00
dump_stack();
}
SLUB core This is a new slab allocator which was motivated by the complexity of the existing code in mm/slab.c. It attempts to address a variety of concerns with the existing implementation. A. Management of object queues A particular concern was the complex management of the numerous object queues in SLAB. SLUB has no such queues. Instead we dedicate a slab for each allocating CPU and use objects from a slab directly instead of queueing them up. B. Storage overhead of object queues SLAB Object queues exist per node, per CPU. The alien cache queue even has a queue array that contain a queue for each processor on each node. For very large systems the number of queues and the number of objects that may be caught in those queues grows exponentially. On our systems with 1k nodes / processors we have several gigabytes just tied up for storing references to objects for those queues This does not include the objects that could be on those queues. One fears that the whole memory of the machine could one day be consumed by those queues. C. SLAB meta data overhead SLAB has overhead at the beginning of each slab. This means that data cannot be naturally aligned at the beginning of a slab block. SLUB keeps all meta data in the corresponding page_struct. Objects can be naturally aligned in the slab. F.e. a 128 byte object will be aligned at 128 byte boundaries and can fit tightly into a 4k page with no bytes left over. SLAB cannot do this. D. SLAB has a complex cache reaper SLUB does not need a cache reaper for UP systems. On SMP systems the per CPU slab may be pushed back into partial list but that operation is simple and does not require an iteration over a list of objects. SLAB expires per CPU, shared and alien object queues during cache reaping which may cause strange hold offs. E. SLAB has complex NUMA policy layer support SLUB pushes NUMA policy handling into the page allocator. This means that allocation is coarser (SLUB does interleave on a page level) but that situation was also present before 2.6.13. SLABs application of policies to individual slab objects allocated in SLAB is certainly a performance concern due to the frequent references to memory policies which may lead a sequence of objects to come from one node after another. SLUB will get a slab full of objects from one node and then will switch to the next. F. Reduction of the size of partial slab lists SLAB has per node partial lists. This means that over time a large number of partial slabs may accumulate on those lists. These can only be reused if allocator occur on specific nodes. SLUB has a global pool of partial slabs and will consume slabs from that pool to decrease fragmentation. G. Tunables SLAB has sophisticated tuning abilities for each slab cache. One can manipulate the queue sizes in detail. However, filling the queues still requires the uses of the spin lock to check out slabs. SLUB has a global parameter (min_slab_order) for tuning. Increasing the minimum slab order can decrease the locking overhead. The bigger the slab order the less motions of pages between per CPU and partial lists occur and the better SLUB will be scaling. G. Slab merging We often have slab caches with similar parameters. SLUB detects those on boot up and merges them into the corresponding general caches. This leads to more effective memory use. About 50% of all caches can be eliminated through slab merging. This will also decrease slab fragmentation because partial allocated slabs can be filled up again. Slab merging can be switched off by specifying slub_nomerge on boot up. Note that merging can expose heretofore unknown bugs in the kernel because corrupted objects may now be placed differently and corrupt differing neighboring objects. Enable sanity checks to find those. H. Diagnostics The current slab diagnostics are difficult to use and require a recompilation of the kernel. SLUB contains debugging code that is always available (but is kept out of the hot code paths). SLUB diagnostics can be enabled via the "slab_debug" option. Parameters can be specified to select a single or a group of slab caches for diagnostics. This means that the system is running with the usual performance and it is much more likely that race conditions can be reproduced. I. Resiliency If basic sanity checks are on then SLUB is capable of detecting common error conditions and recover as best as possible to allow the system to continue. J. Tracing Tracing can be enabled via the slab_debug=T,<slabcache> option during boot. SLUB will then protocol all actions on that slabcache and dump the object contents on free. K. On demand DMA cache creation. Generally DMA caches are not needed. If a kmalloc is used with __GFP_DMA then just create this single slabcache that is needed. For systems that have no ZONE_DMA requirement the support is completely eliminated. L. Performance increase Some benchmarks have shown speed improvements on kernbench in the range of 5-10%. The locking overhead of slub is based on the underlying base allocation size. If we can reliably allocate larger order pages then it is possible to increase slub performance much further. The anti-fragmentation patches may enable further performance increases. Tested on: i386 UP + SMP, x86_64 UP + SMP + NUMA emulation, IA64 NUMA + Simulator SLUB Boot options slub_nomerge Disable merging of slabs slub_min_order=x Require a minimum order for slab caches. This increases the managed chunk size and therefore reduces meta data and locking overhead. slub_min_objects=x Mininum objects per slab. Default is 8. slub_max_order=x Avoid generating slabs larger than order specified. slub_debug Enable all diagnostics for all caches slub_debug=<options> Enable selective options for all caches slub_debug=<o>,<cache> Enable selective options for a certain set of caches Available Debug options F Double Free checking, sanity and resiliency R Red zoning P Object / padding poisoning U Track last free / alloc T Trace all allocs / frees (only use for individual slabs). To use SLUB: Apply this patch and then select SLUB as the default slab allocator. [hugh@veritas.com: fix an oops-causing locking error] [akpm@linux-foundation.org: various stupid cleanups and small fixes] Signed-off-by: Christoph Lameter <clameter@sgi.com> Signed-off-by: Hugh Dickins <hugh@veritas.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2007-05-06 14:49:36 -07:00
else
slub: remove object activities out of checking functions Make sure that the check function really only check things and do not perform activities. Extract the tracing and object seeding out of the two check functions and place them into slab_alloc and slab_free Signed-off-by: Christoph Lameter <clameter@sgi.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2007-05-06 14:49:47 -07:00
slab_err(s, page, "object at 0x%p belongs "
"to slab %s", object, page->slab->name);
SLUB core This is a new slab allocator which was motivated by the complexity of the existing code in mm/slab.c. It attempts to address a variety of concerns with the existing implementation. A. Management of object queues A particular concern was the complex management of the numerous object queues in SLAB. SLUB has no such queues. Instead we dedicate a slab for each allocating CPU and use objects from a slab directly instead of queueing them up. B. Storage overhead of object queues SLAB Object queues exist per node, per CPU. The alien cache queue even has a queue array that contain a queue for each processor on each node. For very large systems the number of queues and the number of objects that may be caught in those queues grows exponentially. On our systems with 1k nodes / processors we have several gigabytes just tied up for storing references to objects for those queues This does not include the objects that could be on those queues. One fears that the whole memory of the machine could one day be consumed by those queues. C. SLAB meta data overhead SLAB has overhead at the beginning of each slab. This means that data cannot be naturally aligned at the beginning of a slab block. SLUB keeps all meta data in the corresponding page_struct. Objects can be naturally aligned in the slab. F.e. a 128 byte object will be aligned at 128 byte boundaries and can fit tightly into a 4k page with no bytes left over. SLAB cannot do this. D. SLAB has a complex cache reaper SLUB does not need a cache reaper for UP systems. On SMP systems the per CPU slab may be pushed back into partial list but that operation is simple and does not require an iteration over a list of objects. SLAB expires per CPU, shared and alien object queues during cache reaping which may cause strange hold offs. E. SLAB has complex NUMA policy layer support SLUB pushes NUMA policy handling into the page allocator. This means that allocation is coarser (SLUB does interleave on a page level) but that situation was also present before 2.6.13. SLABs application of policies to individual slab objects allocated in SLAB is certainly a performance concern due to the frequent references to memory policies which may lead a sequence of objects to come from one node after another. SLUB will get a slab full of objects from one node and then will switch to the next. F. Reduction of the size of partial slab lists SLAB has per node partial lists. This means that over time a large number of partial slabs may accumulate on those lists. These can only be reused if allocator occur on specific nodes. SLUB has a global pool of partial slabs and will consume slabs from that pool to decrease fragmentation. G. Tunables SLAB has sophisticated tuning abilities for each slab cache. One can manipulate the queue sizes in detail. However, filling the queues still requires the uses of the spin lock to check out slabs. SLUB has a global parameter (min_slab_order) for tuning. Increasing the minimum slab order can decrease the locking overhead. The bigger the slab order the less motions of pages between per CPU and partial lists occur and the better SLUB will be scaling. G. Slab merging We often have slab caches with similar parameters. SLUB detects those on boot up and merges them into the corresponding general caches. This leads to more effective memory use. About 50% of all caches can be eliminated through slab merging. This will also decrease slab fragmentation because partial allocated slabs can be filled up again. Slab merging can be switched off by specifying slub_nomerge on boot up. Note that merging can expose heretofore unknown bugs in the kernel because corrupted objects may now be placed differently and corrupt differing neighboring objects. Enable sanity checks to find those. H. Diagnostics The current slab diagnostics are difficult to use and require a recompilation of the kernel. SLUB contains debugging code that is always available (but is kept out of the hot code paths). SLUB diagnostics can be enabled via the "slab_debug" option. Parameters can be specified to select a single or a group of slab caches for diagnostics. This means that the system is running with the usual performance and it is much more likely that race conditions can be reproduced. I. Resiliency If basic sanity checks are on then SLUB is capable of detecting common error conditions and recover as best as possible to allow the system to continue. J. Tracing Tracing can be enabled via the slab_debug=T,<slabcache> option during boot. SLUB will then protocol all actions on that slabcache and dump the object contents on free. K. On demand DMA cache creation. Generally DMA caches are not needed. If a kmalloc is used with __GFP_DMA then just create this single slabcache that is needed. For systems that have no ZONE_DMA requirement the support is completely eliminated. L. Performance increase Some benchmarks have shown speed improvements on kernbench in the range of 5-10%. The locking overhead of slub is based on the underlying base allocation size. If we can reliably allocate larger order pages then it is possible to increase slub performance much further. The anti-fragmentation patches may enable further performance increases. Tested on: i386 UP + SMP, x86_64 UP + SMP + NUMA emulation, IA64 NUMA + Simulator SLUB Boot options slub_nomerge Disable merging of slabs slub_min_order=x Require a minimum order for slab caches. This increases the managed chunk size and therefore reduces meta data and locking overhead. slub_min_objects=x Mininum objects per slab. Default is 8. slub_max_order=x Avoid generating slabs larger than order specified. slub_debug Enable all diagnostics for all caches slub_debug=<options> Enable selective options for all caches slub_debug=<o>,<cache> Enable selective options for a certain set of caches Available Debug options F Double Free checking, sanity and resiliency R Red zoning P Object / padding poisoning U Track last free / alloc T Trace all allocs / frees (only use for individual slabs). To use SLUB: Apply this patch and then select SLUB as the default slab allocator. [hugh@veritas.com: fix an oops-causing locking error] [akpm@linux-foundation.org: various stupid cleanups and small fixes] Signed-off-by: Christoph Lameter <clameter@sgi.com> Signed-off-by: Hugh Dickins <hugh@veritas.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2007-05-06 14:49:36 -07:00
goto fail;
}
return 1;
fail:
printk(KERN_ERR "@@@ SLUB: %s slab 0x%p object at 0x%p not freed.\n",
s->name, page, object);
return 0;
}
/*
* Slab allocation and freeing
*/
static struct page *allocate_slab(struct kmem_cache *s, gfp_t flags, int node)
{
struct page * page;
int pages = 1 << s->order;
if (s->order)
flags |= __GFP_COMP;
if (s->flags & SLAB_CACHE_DMA)
flags |= SLUB_DMA;
if (node == -1)
page = alloc_pages(flags, s->order);
else
page = alloc_pages_node(node, flags, s->order);
if (!page)
return NULL;
mod_zone_page_state(page_zone(page),
(s->flags & SLAB_RECLAIM_ACCOUNT) ?
NR_SLAB_RECLAIMABLE : NR_SLAB_UNRECLAIMABLE,
pages);
return page;
}
static void setup_object(struct kmem_cache *s, struct page *page,
void *object)
{
if (PageError(page)) {
init_object(s, object, 0);
init_tracking(s, object);
}
slab allocators: Remove SLAB_CTOR_ATOMIC SLAB_CTOR atomic is never used which is no surprise since I cannot imagine that one would want to do something serious in a constructor or destructor. In particular given that the slab allocators run with interrupts disabled. Actions in constructors and destructors are by their nature very limited and usually do not go beyond initializing variables and list operations. (The i386 pgd ctor and dtors do take a spinlock in constructor and destructor..... I think that is the furthest we go at this point.) There is no flag passed to the destructor so removing SLAB_CTOR_ATOMIC also establishes a certain symmetry. Signed-off-by: Christoph Lameter <clameter@sgi.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2007-05-06 14:50:17 -07:00
if (unlikely(s->ctor))
s->ctor(object, s, SLAB_CTOR_CONSTRUCTOR);
SLUB core This is a new slab allocator which was motivated by the complexity of the existing code in mm/slab.c. It attempts to address a variety of concerns with the existing implementation. A. Management of object queues A particular concern was the complex management of the numerous object queues in SLAB. SLUB has no such queues. Instead we dedicate a slab for each allocating CPU and use objects from a slab directly instead of queueing them up. B. Storage overhead of object queues SLAB Object queues exist per node, per CPU. The alien cache queue even has a queue array that contain a queue for each processor on each node. For very large systems the number of queues and the number of objects that may be caught in those queues grows exponentially. On our systems with 1k nodes / processors we have several gigabytes just tied up for storing references to objects for those queues This does not include the objects that could be on those queues. One fears that the whole memory of the machine could one day be consumed by those queues. C. SLAB meta data overhead SLAB has overhead at the beginning of each slab. This means that data cannot be naturally aligned at the beginning of a slab block. SLUB keeps all meta data in the corresponding page_struct. Objects can be naturally aligned in the slab. F.e. a 128 byte object will be aligned at 128 byte boundaries and can fit tightly into a 4k page with no bytes left over. SLAB cannot do this. D. SLAB has a complex cache reaper SLUB does not need a cache reaper for UP systems. On SMP systems the per CPU slab may be pushed back into partial list but that operation is simple and does not require an iteration over a list of objects. SLAB expires per CPU, shared and alien object queues during cache reaping which may cause strange hold offs. E. SLAB has complex NUMA policy layer support SLUB pushes NUMA policy handling into the page allocator. This means that allocation is coarser (SLUB does interleave on a page level) but that situation was also present before 2.6.13. SLABs application of policies to individual slab objects allocated in SLAB is certainly a performance concern due to the frequent references to memory policies which may lead a sequence of objects to come from one node after another. SLUB will get a slab full of objects from one node and then will switch to the next. F. Reduction of the size of partial slab lists SLAB has per node partial lists. This means that over time a large number of partial slabs may accumulate on those lists. These can only be reused if allocator occur on specific nodes. SLUB has a global pool of partial slabs and will consume slabs from that pool to decrease fragmentation. G. Tunables SLAB has sophisticated tuning abilities for each slab cache. One can manipulate the queue sizes in detail. However, filling the queues still requires the uses of the spin lock to check out slabs. SLUB has a global parameter (min_slab_order) for tuning. Increasing the minimum slab order can decrease the locking overhead. The bigger the slab order the less motions of pages between per CPU and partial lists occur and the better SLUB will be scaling. G. Slab merging We often have slab caches with similar parameters. SLUB detects those on boot up and merges them into the corresponding general caches. This leads to more effective memory use. About 50% of all caches can be eliminated through slab merging. This will also decrease slab fragmentation because partial allocated slabs can be filled up again. Slab merging can be switched off by specifying slub_nomerge on boot up. Note that merging can expose heretofore unknown bugs in the kernel because corrupted objects may now be placed differently and corrupt differing neighboring objects. Enable sanity checks to find those. H. Diagnostics The current slab diagnostics are difficult to use and require a recompilation of the kernel. SLUB contains debugging code that is always available (but is kept out of the hot code paths). SLUB diagnostics can be enabled via the "slab_debug" option. Parameters can be specified to select a single or a group of slab caches for diagnostics. This means that the system is running with the usual performance and it is much more likely that race conditions can be reproduced. I. Resiliency If basic sanity checks are on then SLUB is capable of detecting common error conditions and recover as best as possible to allow the system to continue. J. Tracing Tracing can be enabled via the slab_debug=T,<slabcache> option during boot. SLUB will then protocol all actions on that slabcache and dump the object contents on free. K. On demand DMA cache creation. Generally DMA caches are not needed. If a kmalloc is used with __GFP_DMA then just create this single slabcache that is needed. For systems that have no ZONE_DMA requirement the support is completely eliminated. L. Performance increase Some benchmarks have shown speed improvements on kernbench in the range of 5-10%. The locking overhead of slub is based on the underlying base allocation size. If we can reliably allocate larger order pages then it is possible to increase slub performance much further. The anti-fragmentation patches may enable further performance increases. Tested on: i386 UP + SMP, x86_64 UP + SMP + NUMA emulation, IA64 NUMA + Simulator SLUB Boot options slub_nomerge Disable merging of slabs slub_min_order=x Require a minimum order for slab caches. This increases the managed chunk size and therefore reduces meta data and locking overhead. slub_min_objects=x Mininum objects per slab. Default is 8. slub_max_order=x Avoid generating slabs larger than order specified. slub_debug Enable all diagnostics for all caches slub_debug=<options> Enable selective options for all caches slub_debug=<o>,<cache> Enable selective options for a certain set of caches Available Debug options F Double Free checking, sanity and resiliency R Red zoning P Object / padding poisoning U Track last free / alloc T Trace all allocs / frees (only use for individual slabs). To use SLUB: Apply this patch and then select SLUB as the default slab allocator. [hugh@veritas.com: fix an oops-causing locking error] [akpm@linux-foundation.org: various stupid cleanups and small fixes] Signed-off-by: Christoph Lameter <clameter@sgi.com> Signed-off-by: Hugh Dickins <hugh@veritas.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2007-05-06 14:49:36 -07:00
}
static struct page *new_slab(struct kmem_cache *s, gfp_t flags, int node)
{
struct page *page;
struct kmem_cache_node *n;
void *start;
void *end;
void *last;
void *p;
BUG_ON(flags & ~(GFP_DMA | GFP_LEVEL_MASK));
if (flags & __GFP_WAIT)
local_irq_enable();
page = allocate_slab(s, flags & GFP_LEVEL_MASK, node);
if (!page)
goto out;
n = get_node(s, page_to_nid(page));
if (n)
atomic_long_inc(&n->nr_slabs);
page->offset = s->offset / sizeof(void *);
page->slab = s;
page->flags |= 1 << PG_slab;
if (s->flags & (SLAB_DEBUG_FREE | SLAB_RED_ZONE | SLAB_POISON |
SLAB_STORE_USER | SLAB_TRACE))
page->flags |= 1 << PG_error;
start = page_address(page);
end = start + s->objects * s->size;
if (unlikely(s->flags & SLAB_POISON))
memset(start, POISON_INUSE, PAGE_SIZE << s->order);
last = start;
for (p = start + s->size; p < end; p += s->size) {
setup_object(s, page, last);
set_freepointer(s, last, p);
last = p;
}
setup_object(s, page, last);
set_freepointer(s, last, NULL);
page->freelist = start;
page->inuse = 0;
out:
if (flags & __GFP_WAIT)
local_irq_disable();
return page;
}
static void __free_slab(struct kmem_cache *s, struct page *page)
{
int pages = 1 << s->order;
if (unlikely(PageError(page) || s->dtor)) {
void *start = page_address(page);
void *end = start + (pages << PAGE_SHIFT);
void *p;
slab_pad_check(s, page);
for (p = start; p <= end - s->size; p += s->size) {
if (s->dtor)
s->dtor(p, s, 0);
check_object(s, page, p, 0);
}
}
mod_zone_page_state(page_zone(page),
(s->flags & SLAB_RECLAIM_ACCOUNT) ?
NR_SLAB_RECLAIMABLE : NR_SLAB_UNRECLAIMABLE,
- pages);
page->mapping = NULL;
__free_pages(page, s->order);
}
static void rcu_free_slab(struct rcu_head *h)
{
struct page *page;
page = container_of((struct list_head *)h, struct page, lru);
__free_slab(page->slab, page);
}
static void free_slab(struct kmem_cache *s, struct page *page)
{
if (unlikely(s->flags & SLAB_DESTROY_BY_RCU)) {
/*
* RCU free overloads the RCU head over the LRU
*/
struct rcu_head *head = (void *)&page->lru;
call_rcu(head, rcu_free_slab);
} else
__free_slab(s, page);
}
static void discard_slab(struct kmem_cache *s, struct page *page)
{
struct kmem_cache_node *n = get_node(s, page_to_nid(page));
atomic_long_dec(&n->nr_slabs);
reset_page_mapcount(page);
page->flags &= ~(1 << PG_slab | 1 << PG_error);
free_slab(s, page);
}
/*
* Per slab locking using the pagelock
*/
static __always_inline void slab_lock(struct page *page)
{
bit_spin_lock(PG_locked, &page->flags);
}
static __always_inline void slab_unlock(struct page *page)
{
bit_spin_unlock(PG_locked, &page->flags);
}
static __always_inline int slab_trylock(struct page *page)
{
int rc = 1;
rc = bit_spin_trylock(PG_locked, &page->flags);
return rc;
}
/*
* Management of partially allocated slabs
*/
SLUB: Add MIN_PARTIAL We leave a mininum of partial slabs on nodes when we search for partial slabs on other node. Define a constant for that value. Then modify slub to keep MIN_PARTIAL slabs around. This avoids bad situations where a function frees the last object in a slab (which results in the page being returned to the page allocator) only to then allocate one again (which requires getting a page back from the page allocator if the partial list was empty). Keeping a couple of slabs on the partial list reduces overhead. Empty slabs are added to the end of the partial list to insure that partially allocated slabs are consumed first (defragmentation). Signed-off-by: Christoph Lameter <clameter@sgi.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2007-05-06 14:49:44 -07:00
static void add_partial_tail(struct kmem_cache_node *n, struct page *page)
SLUB core This is a new slab allocator which was motivated by the complexity of the existing code in mm/slab.c. It attempts to address a variety of concerns with the existing implementation. A. Management of object queues A particular concern was the complex management of the numerous object queues in SLAB. SLUB has no such queues. Instead we dedicate a slab for each allocating CPU and use objects from a slab directly instead of queueing them up. B. Storage overhead of object queues SLAB Object queues exist per node, per CPU. The alien cache queue even has a queue array that contain a queue for each processor on each node. For very large systems the number of queues and the number of objects that may be caught in those queues grows exponentially. On our systems with 1k nodes / processors we have several gigabytes just tied up for storing references to objects for those queues This does not include the objects that could be on those queues. One fears that the whole memory of the machine could one day be consumed by those queues. C. SLAB meta data overhead SLAB has overhead at the beginning of each slab. This means that data cannot be naturally aligned at the beginning of a slab block. SLUB keeps all meta data in the corresponding page_struct. Objects can be naturally aligned in the slab. F.e. a 128 byte object will be aligned at 128 byte boundaries and can fit tightly into a 4k page with no bytes left over. SLAB cannot do this. D. SLAB has a complex cache reaper SLUB does not need a cache reaper for UP systems. On SMP systems the per CPU slab may be pushed back into partial list but that operation is simple and does not require an iteration over a list of objects. SLAB expires per CPU, shared and alien object queues during cache reaping which may cause strange hold offs. E. SLAB has complex NUMA policy layer support SLUB pushes NUMA policy handling into the page allocator. This means that allocation is coarser (SLUB does interleave on a page level) but that situation was also present before 2.6.13. SLABs application of policies to individual slab objects allocated in SLAB is certainly a performance concern due to the frequent references to memory policies which may lead a sequence of objects to come from one node after another. SLUB will get a slab full of objects from one node and then will switch to the next. F. Reduction of the size of partial slab lists SLAB has per node partial lists. This means that over time a large number of partial slabs may accumulate on those lists. These can only be reused if allocator occur on specific nodes. SLUB has a global pool of partial slabs and will consume slabs from that pool to decrease fragmentation. G. Tunables SLAB has sophisticated tuning abilities for each slab cache. One can manipulate the queue sizes in detail. However, filling the queues still requires the uses of the spin lock to check out slabs. SLUB has a global parameter (min_slab_order) for tuning. Increasing the minimum slab order can decrease the locking overhead. The bigger the slab order the less motions of pages between per CPU and partial lists occur and the better SLUB will be scaling. G. Slab merging We often have slab caches with similar parameters. SLUB detects those on boot up and merges them into the corresponding general caches. This leads to more effective memory use. About 50% of all caches can be eliminated through slab merging. This will also decrease slab fragmentation because partial allocated slabs can be filled up again. Slab merging can be switched off by specifying slub_nomerge on boot up. Note that merging can expose heretofore unknown bugs in the kernel because corrupted objects may now be placed differently and corrupt differing neighboring objects. Enable sanity checks to find those. H. Diagnostics The current slab diagnostics are difficult to use and require a recompilation of the kernel. SLUB contains debugging code that is always available (but is kept out of the hot code paths). SLUB diagnostics can be enabled via the "slab_debug" option. Parameters can be specified to select a single or a group of slab caches for diagnostics. This means that the system is running with the usual performance and it is much more likely that race conditions can be reproduced. I. Resiliency If basic sanity checks are on then SLUB is capable of detecting common error conditions and recover as best as possible to allow the system to continue. J. Tracing Tracing can be enabled via the slab_debug=T,<slabcache> option during boot. SLUB will then protocol all actions on that slabcache and dump the object contents on free. K. On demand DMA cache creation. Generally DMA caches are not needed. If a kmalloc is used with __GFP_DMA then just create this single slabcache that is needed. For systems that have no ZONE_DMA requirement the support is completely eliminated. L. Performance increase Some benchmarks have shown speed improvements on kernbench in the range of 5-10%. The locking overhead of slub is based on the underlying base allocation size. If we can reliably allocate larger order pages then it is possible to increase slub performance much further. The anti-fragmentation patches may enable further performance increases. Tested on: i386 UP + SMP, x86_64 UP + SMP + NUMA emulation, IA64 NUMA + Simulator SLUB Boot options slub_nomerge Disable merging of slabs slub_min_order=x Require a minimum order for slab caches. This increases the managed chunk size and therefore reduces meta data and locking overhead. slub_min_objects=x Mininum objects per slab. Default is 8. slub_max_order=x Avoid generating slabs larger than order specified. slub_debug Enable all diagnostics for all caches slub_debug=<options> Enable selective options for all caches slub_debug=<o>,<cache> Enable selective options for a certain set of caches Available Debug options F Double Free checking, sanity and resiliency R Red zoning P Object / padding poisoning U Track last free / alloc T Trace all allocs / frees (only use for individual slabs). To use SLUB: Apply this patch and then select SLUB as the default slab allocator. [hugh@veritas.com: fix an oops-causing locking error] [akpm@linux-foundation.org: various stupid cleanups and small fixes] Signed-off-by: Christoph Lameter <clameter@sgi.com> Signed-off-by: Hugh Dickins <hugh@veritas.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2007-05-06 14:49:36 -07:00
{
SLUB: Add MIN_PARTIAL We leave a mininum of partial slabs on nodes when we search for partial slabs on other node. Define a constant for that value. Then modify slub to keep MIN_PARTIAL slabs around. This avoids bad situations where a function frees the last object in a slab (which results in the page being returned to the page allocator) only to then allocate one again (which requires getting a page back from the page allocator if the partial list was empty). Keeping a couple of slabs on the partial list reduces overhead. Empty slabs are added to the end of the partial list to insure that partially allocated slabs are consumed first (defragmentation). Signed-off-by: Christoph Lameter <clameter@sgi.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2007-05-06 14:49:44 -07:00
spin_lock(&n->list_lock);
n->nr_partial++;
list_add_tail(&page->lru, &n->partial);
spin_unlock(&n->list_lock);
}
SLUB core This is a new slab allocator which was motivated by the complexity of the existing code in mm/slab.c. It attempts to address a variety of concerns with the existing implementation. A. Management of object queues A particular concern was the complex management of the numerous object queues in SLAB. SLUB has no such queues. Instead we dedicate a slab for each allocating CPU and use objects from a slab directly instead of queueing them up. B. Storage overhead of object queues SLAB Object queues exist per node, per CPU. The alien cache queue even has a queue array that contain a queue for each processor on each node. For very large systems the number of queues and the number of objects that may be caught in those queues grows exponentially. On our systems with 1k nodes / processors we have several gigabytes just tied up for storing references to objects for those queues This does not include the objects that could be on those queues. One fears that the whole memory of the machine could one day be consumed by those queues. C. SLAB meta data overhead SLAB has overhead at the beginning of each slab. This means that data cannot be naturally aligned at the beginning of a slab block. SLUB keeps all meta data in the corresponding page_struct. Objects can be naturally aligned in the slab. F.e. a 128 byte object will be aligned at 128 byte boundaries and can fit tightly into a 4k page with no bytes left over. SLAB cannot do this. D. SLAB has a complex cache reaper SLUB does not need a cache reaper for UP systems. On SMP systems the per CPU slab may be pushed back into partial list but that operation is simple and does not require an iteration over a list of objects. SLAB expires per CPU, shared and alien object queues during cache reaping which may cause strange hold offs. E. SLAB has complex NUMA policy layer support SLUB pushes NUMA policy handling into the page allocator. This means that allocation is coarser (SLUB does interleave on a page level) but that situation was also present before 2.6.13. SLABs application of policies to individual slab objects allocated in SLAB is certainly a performance concern due to the frequent references to memory policies which may lead a sequence of objects to come from one node after another. SLUB will get a slab full of objects from one node and then will switch to the next. F. Reduction of the size of partial slab lists SLAB has per node partial lists. This means that over time a large number of partial slabs may accumulate on those lists. These can only be reused if allocator occur on specific nodes. SLUB has a global pool of partial slabs and will consume slabs from that pool to decrease fragmentation. G. Tunables SLAB has sophisticated tuning abilities for each slab cache. One can manipulate the queue sizes in detail. However, filling the queues still requires the uses of the spin lock to check out slabs. SLUB has a global parameter (min_slab_order) for tuning. Increasing the minimum slab order can decrease the locking overhead. The bigger the slab order the less motions of pages between per CPU and partial lists occur and the better SLUB will be scaling. G. Slab merging We often have slab caches with similar parameters. SLUB detects those on boot up and merges them into the corresponding general caches. This leads to more effective memory use. About 50% of all caches can be eliminated through slab merging. This will also decrease slab fragmentation because partial allocated slabs can be filled up again. Slab merging can be switched off by specifying slub_nomerge on boot up. Note that merging can expose heretofore unknown bugs in the kernel because corrupted objects may now be placed differently and corrupt differing neighboring objects. Enable sanity checks to find those. H. Diagnostics The current slab diagnostics are difficult to use and require a recompilation of the kernel. SLUB contains debugging code that is always available (but is kept out of the hot code paths). SLUB diagnostics can be enabled via the "slab_debug" option. Parameters can be specified to select a single or a group of slab caches for diagnostics. This means that the system is running with the usual performance and it is much more likely that race conditions can be reproduced. I. Resiliency If basic sanity checks are on then SLUB is capable of detecting common error conditions and recover as best as possible to allow the system to continue. J. Tracing Tracing can be enabled via the slab_debug=T,<slabcache> option during boot. SLUB will then protocol all actions on that slabcache and dump the object contents on free. K. On demand DMA cache creation. Generally DMA caches are not needed. If a kmalloc is used with __GFP_DMA then just create this single slabcache that is needed. For systems that have no ZONE_DMA requirement the support is completely eliminated. L. Performance increase Some benchmarks have shown speed improvements on kernbench in the range of 5-10%. The locking overhead of slub is based on the underlying base allocation size. If we can reliably allocate larger order pages then it is possible to increase slub performance much further. The anti-fragmentation patches may enable further performance increases. Tested on: i386 UP + SMP, x86_64 UP + SMP + NUMA emulation, IA64 NUMA + Simulator SLUB Boot options slub_nomerge Disable merging of slabs slub_min_order=x Require a minimum order for slab caches. This increases the managed chunk size and therefore reduces meta data and locking overhead. slub_min_objects=x Mininum objects per slab. Default is 8. slub_max_order=x Avoid generating slabs larger than order specified. slub_debug Enable all diagnostics for all caches slub_debug=<options> Enable selective options for all caches slub_debug=<o>,<cache> Enable selective options for a certain set of caches Available Debug options F Double Free checking, sanity and resiliency R Red zoning P Object / padding poisoning U Track last free / alloc T Trace all allocs / frees (only use for individual slabs). To use SLUB: Apply this patch and then select SLUB as the default slab allocator. [hugh@veritas.com: fix an oops-causing locking error] [akpm@linux-foundation.org: various stupid cleanups and small fixes] Signed-off-by: Christoph Lameter <clameter@sgi.com> Signed-off-by: Hugh Dickins <hugh@veritas.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2007-05-06 14:49:36 -07:00
SLUB: Add MIN_PARTIAL We leave a mininum of partial slabs on nodes when we search for partial slabs on other node. Define a constant for that value. Then modify slub to keep MIN_PARTIAL slabs around. This avoids bad situations where a function frees the last object in a slab (which results in the page being returned to the page allocator) only to then allocate one again (which requires getting a page back from the page allocator if the partial list was empty). Keeping a couple of slabs on the partial list reduces overhead. Empty slabs are added to the end of the partial list to insure that partially allocated slabs are consumed first (defragmentation). Signed-off-by: Christoph Lameter <clameter@sgi.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2007-05-06 14:49:44 -07:00
static void add_partial(struct kmem_cache_node *n, struct page *page)
{
SLUB core This is a new slab allocator which was motivated by the complexity of the existing code in mm/slab.c. It attempts to address a variety of concerns with the existing implementation. A. Management of object queues A particular concern was the complex management of the numerous object queues in SLAB. SLUB has no such queues. Instead we dedicate a slab for each allocating CPU and use objects from a slab directly instead of queueing them up. B. Storage overhead of object queues SLAB Object queues exist per node, per CPU. The alien cache queue even has a queue array that contain a queue for each processor on each node. For very large systems the number of queues and the number of objects that may be caught in those queues grows exponentially. On our systems with 1k nodes / processors we have several gigabytes just tied up for storing references to objects for those queues This does not include the objects that could be on those queues. One fears that the whole memory of the machine could one day be consumed by those queues. C. SLAB meta data overhead SLAB has overhead at the beginning of each slab. This means that data cannot be naturally aligned at the beginning of a slab block. SLUB keeps all meta data in the corresponding page_struct. Objects can be naturally aligned in the slab. F.e. a 128 byte object will be aligned at 128 byte boundaries and can fit tightly into a 4k page with no bytes left over. SLAB cannot do this. D. SLAB has a complex cache reaper SLUB does not need a cache reaper for UP systems. On SMP systems the per CPU slab may be pushed back into partial list but that operation is simple and does not require an iteration over a list of objects. SLAB expires per CPU, shared and alien object queues during cache reaping which may cause strange hold offs. E. SLAB has complex NUMA policy layer support SLUB pushes NUMA policy handling into the page allocator. This means that allocation is coarser (SLUB does interleave on a page level) but that situation was also present before 2.6.13. SLABs application of policies to individual slab objects allocated in SLAB is certainly a performance concern due to the frequent references to memory policies which may lead a sequence of objects to come from one node after another. SLUB will get a slab full of objects from one node and then will switch to the next. F. Reduction of the size of partial slab lists SLAB has per node partial lists. This means that over time a large number of partial slabs may accumulate on those lists. These can only be reused if allocator occur on specific nodes. SLUB has a global pool of partial slabs and will consume slabs from that pool to decrease fragmentation. G. Tunables SLAB has sophisticated tuning abilities for each slab cache. One can manipulate the queue sizes in detail. However, filling the queues still requires the uses of the spin lock to check out slabs. SLUB has a global parameter (min_slab_order) for tuning. Increasing the minimum slab order can decrease the locking overhead. The bigger the slab order the less motions of pages between per CPU and partial lists occur and the better SLUB will be scaling. G. Slab merging We often have slab caches with similar parameters. SLUB detects those on boot up and merges them into the corresponding general caches. This leads to more effective memory use. About 50% of all caches can be eliminated through slab merging. This will also decrease slab fragmentation because partial allocated slabs can be filled up again. Slab merging can be switched off by specifying slub_nomerge on boot up. Note that merging can expose heretofore unknown bugs in the kernel because corrupted objects may now be placed differently and corrupt differing neighboring objects. Enable sanity checks to find those. H. Diagnostics The current slab diagnostics are difficult to use and require a recompilation of the kernel. SLUB contains debugging code that is always available (but is kept out of the hot code paths). SLUB diagnostics can be enabled via the "slab_debug" option. Parameters can be specified to select a single or a group of slab caches for diagnostics. This means that the system is running with the usual performance and it is much more likely that race conditions can be reproduced. I. Resiliency If basic sanity checks are on then SLUB is capable of detecting common error conditions and recover as best as possible to allow the system to continue. J. Tracing Tracing can be enabled via the slab_debug=T,<slabcache> option during boot. SLUB will then protocol all actions on that slabcache and dump the object contents on free. K. On demand DMA cache creation. Generally DMA caches are not needed. If a kmalloc is used with __GFP_DMA then just create this single slabcache that is needed. For systems that have no ZONE_DMA requirement the support is completely eliminated. L. Performance increase Some benchmarks have shown speed improvements on kernbench in the range of 5-10%. The locking overhead of slub is based on the underlying base allocation size. If we can reliably allocate larger order pages then it is possible to increase slub performance much further. The anti-fragmentation patches may enable further performance increases. Tested on: i386 UP + SMP, x86_64 UP + SMP + NUMA emulation, IA64 NUMA + Simulator SLUB Boot options slub_nomerge Disable merging of slabs slub_min_order=x Require a minimum order for slab caches. This increases the managed chunk size and therefore reduces meta data and locking overhead. slub_min_objects=x Mininum objects per slab. Default is 8. slub_max_order=x Avoid generating slabs larger than order specified. slub_debug Enable all diagnostics for all caches slub_debug=<options> Enable selective options for all caches slub_debug=<o>,<cache> Enable selective options for a certain set of caches Available Debug options F Double Free checking, sanity and resiliency R Red zoning P Object / padding poisoning U Track last free / alloc T Trace all allocs / frees (only use for individual slabs). To use SLUB: Apply this patch and then select SLUB as the default slab allocator. [hugh@veritas.com: fix an oops-causing locking error] [akpm@linux-foundation.org: various stupid cleanups and small fixes] Signed-off-by: Christoph Lameter <clameter@sgi.com> Signed-off-by: Hugh Dickins <hugh@veritas.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2007-05-06 14:49:36 -07:00
spin_lock(&n->list_lock);
n->nr_partial++;
list_add(&page->lru, &n->partial);
spin_unlock(&n->list_lock);
}
static void remove_partial(struct kmem_cache *s,
struct page *page)
{
struct kmem_cache_node *n = get_node(s, page_to_nid(page));
spin_lock(&n->list_lock);
list_del(&page->lru);
n->nr_partial--;
spin_unlock(&n->list_lock);
}
/*
* Lock page and remove it from the partial list
*
* Must hold list_lock
*/
static int lock_and_del_slab(struct kmem_cache_node *n, struct page *page)
{
if (slab_trylock(page)) {
list_del(&page->lru);
n->nr_partial--;
return 1;
}
return 0;
}
/*
* Try to get a partial slab from a specific node
*/
static struct page *get_partial_node(struct kmem_cache_node *n)
{
struct page *page;
/*
* Racy check. If we mistakenly see no partial slabs then we
* just allocate an empty slab. If we mistakenly try to get a
* partial slab then get_partials() will return NULL.
*/
if (!n || !n->nr_partial)
return NULL;
spin_lock(&n->list_lock);
list_for_each_entry(page, &n->partial, lru)
if (lock_and_del_slab(n, page))
goto out;
page = NULL;
out:
spin_unlock(&n->list_lock);
return page;
}
/*
* Get a page from somewhere. Search in increasing NUMA
* distances.
*/
static struct page *get_any_partial(struct kmem_cache *s, gfp_t flags)
{
#ifdef CONFIG_NUMA
struct zonelist *zonelist;
struct zone **z;
struct page *page;
/*
* The defrag ratio allows to configure the tradeoffs between
* inter node defragmentation and node local allocations.
* A lower defrag_ratio increases the tendency to do local
* allocations instead of scanning throught the partial
* lists on other nodes.
*
* If defrag_ratio is set to 0 then kmalloc() always
* returns node local objects. If its higher then kmalloc()
* may return off node objects in order to avoid fragmentation.
*
* A higher ratio means slabs may be taken from other nodes
* thus reducing the number of partial slabs on those nodes.
*
* If /sys/slab/xx/defrag_ratio is set to 100 (which makes
* defrag_ratio = 1000) then every (well almost) allocation
* will first attempt to defrag slab caches on other nodes. This
* means scanning over all nodes to look for partial slabs which
* may be a bit expensive to do on every slab allocation.
*/
if (!s->defrag_ratio || get_cycles() % 1024 > s->defrag_ratio)
return NULL;
zonelist = &NODE_DATA(slab_node(current->mempolicy))
->node_zonelists[gfp_zone(flags)];
for (z = zonelist->zones; *z; z++) {
struct kmem_cache_node *n;
n = get_node(s, zone_to_nid(*z));
if (n && cpuset_zone_allowed_hardwall(*z, flags) &&
SLUB: Add MIN_PARTIAL We leave a mininum of partial slabs on nodes when we search for partial slabs on other node. Define a constant for that value. Then modify slub to keep MIN_PARTIAL slabs around. This avoids bad situations where a function frees the last object in a slab (which results in the page being returned to the page allocator) only to then allocate one again (which requires getting a page back from the page allocator if the partial list was empty). Keeping a couple of slabs on the partial list reduces overhead. Empty slabs are added to the end of the partial list to insure that partially allocated slabs are consumed first (defragmentation). Signed-off-by: Christoph Lameter <clameter@sgi.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2007-05-06 14:49:44 -07:00
n->nr_partial > MIN_PARTIAL) {
SLUB core This is a new slab allocator which was motivated by the complexity of the existing code in mm/slab.c. It attempts to address a variety of concerns with the existing implementation. A. Management of object queues A particular concern was the complex management of the numerous object queues in SLAB. SLUB has no such queues. Instead we dedicate a slab for each allocating CPU and use objects from a slab directly instead of queueing them up. B. Storage overhead of object queues SLAB Object queues exist per node, per CPU. The alien cache queue even has a queue array that contain a queue for each processor on each node. For very large systems the number of queues and the number of objects that may be caught in those queues grows exponentially. On our systems with 1k nodes / processors we have several gigabytes just tied up for storing references to objects for those queues This does not include the objects that could be on those queues. One fears that the whole memory of the machine could one day be consumed by those queues. C. SLAB meta data overhead SLAB has overhead at the beginning of each slab. This means that data cannot be naturally aligned at the beginning of a slab block. SLUB keeps all meta data in the corresponding page_struct. Objects can be naturally aligned in the slab. F.e. a 128 byte object will be aligned at 128 byte boundaries and can fit tightly into a 4k page with no bytes left over. SLAB cannot do this. D. SLAB has a complex cache reaper SLUB does not need a cache reaper for UP systems. On SMP systems the per CPU slab may be pushed back into partial list but that operation is simple and does not require an iteration over a list of objects. SLAB expires per CPU, shared and alien object queues during cache reaping which may cause strange hold offs. E. SLAB has complex NUMA policy layer support SLUB pushes NUMA policy handling into the page allocator. This means that allocation is coarser (SLUB does interleave on a page level) but that situation was also present before 2.6.13. SLABs application of policies to individual slab objects allocated in SLAB is certainly a performance concern due to the frequent references to memory policies which may lead a sequence of objects to come from one node after another. SLUB will get a slab full of objects from one node and then will switch to the next. F. Reduction of the size of partial slab lists SLAB has per node partial lists. This means that over time a large number of partial slabs may accumulate on those lists. These can only be reused if allocator occur on specific nodes. SLUB has a global pool of partial slabs and will consume slabs from that pool to decrease fragmentation. G. Tunables SLAB has sophisticated tuning abilities for each slab cache. One can manipulate the queue sizes in detail. However, filling the queues still requires the uses of the spin lock to check out slabs. SLUB has a global parameter (min_slab_order) for tuning. Increasing the minimum slab order can decrease the locking overhead. The bigger the slab order the less motions of pages between per CPU and partial lists occur and the better SLUB will be scaling. G. Slab merging We often have slab caches with similar parameters. SLUB detects those on boot up and merges them into the corresponding general caches. This leads to more effective memory use. About 50% of all caches can be eliminated through slab merging. This will also decrease slab fragmentation because partial allocated slabs can be filled up again. Slab merging can be switched off by specifying slub_nomerge on boot up. Note that merging can expose heretofore unknown bugs in the kernel because corrupted objects may now be placed differently and corrupt differing neighboring objects. Enable sanity checks to find those. H. Diagnostics The current slab diagnostics are difficult to use and require a recompilation of the kernel. SLUB contains debugging code that is always available (but is kept out of the hot code paths). SLUB diagnostics can be enabled via the "slab_debug" option. Parameters can be specified to select a single or a group of slab caches for diagnostics. This means that the system is running with the usual performance and it is much more likely that race conditions can be reproduced. I. Resiliency If basic sanity checks are on then SLUB is capable of detecting common error conditions and recover as best as possible to allow the system to continue. J. Tracing Tracing can be enabled via the slab_debug=T,<slabcache> option during boot. SLUB will then protocol all actions on that slabcache and dump the object contents on free. K. On demand DMA cache creation. Generally DMA caches are not needed. If a kmalloc is used with __GFP_DMA then just create this single slabcache that is needed. For systems that have no ZONE_DMA requirement the support is completely eliminated. L. Performance increase Some benchmarks have shown speed improvements on kernbench in the range of 5-10%. The locking overhead of slub is based on the underlying base allocation size. If we can reliably allocate larger order pages then it is possible to increase slub performance much further. The anti-fragmentation patches may enable further performance increases. Tested on: i386 UP + SMP, x86_64 UP + SMP + NUMA emulation, IA64 NUMA + Simulator SLUB Boot options slub_nomerge Disable merging of slabs slub_min_order=x Require a minimum order for slab caches. This increases the managed chunk size and therefore reduces meta data and locking overhead. slub_min_objects=x Mininum objects per slab. Default is 8. slub_max_order=x Avoid generating slabs larger than order specified. slub_debug Enable all diagnostics for all caches slub_debug=<options> Enable selective options for all caches slub_debug=<o>,<cache> Enable selective options for a certain set of caches Available Debug options F Double Free checking, sanity and resiliency R Red zoning P Object / padding poisoning U Track last free / alloc T Trace all allocs / frees (only use for individual slabs). To use SLUB: Apply this patch and then select SLUB as the default slab allocator. [hugh@veritas.com: fix an oops-causing locking error] [akpm@linux-foundation.org: various stupid cleanups and small fixes] Signed-off-by: Christoph Lameter <clameter@sgi.com> Signed-off-by: Hugh Dickins <hugh@veritas.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2007-05-06 14:49:36 -07:00
page = get_partial_node(n);
if (page)
return page;
}
}
#endif
return NULL;
}
/*
* Get a partial page, lock it and return it.
*/
static struct page *get_partial(struct kmem_cache *s, gfp_t flags, int node)
{
struct page *page;
int searchnode = (node == -1) ? numa_node_id() : node;
page = get_partial_node(get_node(s, searchnode));
if (page || (flags & __GFP_THISNODE))
return page;
return get_any_partial(s, flags);
}
/*
* Move a page back to the lists.
*
* Must be called with the slab lock held.
*
* On exit the slab lock will have been dropped.
*/
static void putback_slab(struct kmem_cache *s, struct page *page)
{
SLUB: Add MIN_PARTIAL We leave a mininum of partial slabs on nodes when we search for partial slabs on other node. Define a constant for that value. Then modify slub to keep MIN_PARTIAL slabs around. This avoids bad situations where a function frees the last object in a slab (which results in the page being returned to the page allocator) only to then allocate one again (which requires getting a page back from the page allocator if the partial list was empty). Keeping a couple of slabs on the partial list reduces overhead. Empty slabs are added to the end of the partial list to insure that partially allocated slabs are consumed first (defragmentation). Signed-off-by: Christoph Lameter <clameter@sgi.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2007-05-06 14:49:44 -07:00
struct kmem_cache_node *n = get_node(s, page_to_nid(page));
SLUB core This is a new slab allocator which was motivated by the complexity of the existing code in mm/slab.c. It attempts to address a variety of concerns with the existing implementation. A. Management of object queues A particular concern was the complex management of the numerous object queues in SLAB. SLUB has no such queues. Instead we dedicate a slab for each allocating CPU and use objects from a slab directly instead of queueing them up. B. Storage overhead of object queues SLAB Object queues exist per node, per CPU. The alien cache queue even has a queue array that contain a queue for each processor on each node. For very large systems the number of queues and the number of objects that may be caught in those queues grows exponentially. On our systems with 1k nodes / processors we have several gigabytes just tied up for storing references to objects for those queues This does not include the objects that could be on those queues. One fears that the whole memory of the machine could one day be consumed by those queues. C. SLAB meta data overhead SLAB has overhead at the beginning of each slab. This means that data cannot be naturally aligned at the beginning of a slab block. SLUB keeps all meta data in the corresponding page_struct. Objects can be naturally aligned in the slab. F.e. a 128 byte object will be aligned at 128 byte boundaries and can fit tightly into a 4k page with no bytes left over. SLAB cannot do this. D. SLAB has a complex cache reaper SLUB does not need a cache reaper for UP systems. On SMP systems the per CPU slab may be pushed back into partial list but that operation is simple and does not require an iteration over a list of objects. SLAB expires per CPU, shared and alien object queues during cache reaping which may cause strange hold offs. E. SLAB has complex NUMA policy layer support SLUB pushes NUMA policy handling into the page allocator. This means that allocation is coarser (SLUB does interleave on a page level) but that situation was also present before 2.6.13. SLABs application of policies to individual slab objects allocated in SLAB is certainly a performance concern due to the frequent references to memory policies which may lead a sequence of objects to come from one node after another. SLUB will get a slab full of objects from one node and then will switch to the next. F. Reduction of the size of partial slab lists SLAB has per node partial lists. This means that over time a large number of partial slabs may accumulate on those lists. These can only be reused if allocator occur on specific nodes. SLUB has a global pool of partial slabs and will consume slabs from that pool to decrease fragmentation. G. Tunables SLAB has sophisticated tuning abilities for each slab cache. One can manipulate the queue sizes in detail. However, filling the queues still requires the uses of the spin lock to check out slabs. SLUB has a global parameter (min_slab_order) for tuning. Increasing the minimum slab order can decrease the locking overhead. The bigger the slab order the less motions of pages between per CPU and partial lists occur and the better SLUB will be scaling. G. Slab merging We often have slab caches with similar parameters. SLUB detects those on boot up and merges them into the corresponding general caches. This leads to more effective memory use. About 50% of all caches can be eliminated through slab merging. This will also decrease slab fragmentation because partial allocated slabs can be filled up again. Slab merging can be switched off by specifying slub_nomerge on boot up. Note that merging can expose heretofore unknown bugs in the kernel because corrupted objects may now be placed differently and corrupt differing neighboring objects. Enable sanity checks to find those. H. Diagnostics The current slab diagnostics are difficult to use and require a recompilation of the kernel. SLUB contains debugging code that is always available (but is kept out of the hot code paths). SLUB diagnostics can be enabled via the "slab_debug" option. Parameters can be specified to select a single or a group of slab caches for diagnostics. This means that the system is running with the usual performance and it is much more likely that race conditions can be reproduced. I. Resiliency If basic sanity checks are on then SLUB is capable of detecting common error conditions and recover as best as possible to allow the system to continue. J. Tracing Tracing can be enabled via the slab_debug=T,<slabcache> option during boot. SLUB will then protocol all actions on that slabcache and dump the object contents on free. K. On demand DMA cache creation. Generally DMA caches are not needed. If a kmalloc is used with __GFP_DMA then just create this single slabcache that is needed. For systems that have no ZONE_DMA requirement the support is completely eliminated. L. Performance increase Some benchmarks have shown speed improvements on kernbench in the range of 5-10%. The locking overhead of slub is based on the underlying base allocation size. If we can reliably allocate larger order pages then it is possible to increase slub performance much further. The anti-fragmentation patches may enable further performance increases. Tested on: i386 UP + SMP, x86_64 UP + SMP + NUMA emulation, IA64 NUMA + Simulator SLUB Boot options slub_nomerge Disable merging of slabs slub_min_order=x Require a minimum order for slab caches. This increases the managed chunk size and therefore reduces meta data and locking overhead. slub_min_objects=x Mininum objects per slab. Default is 8. slub_max_order=x Avoid generating slabs larger than order specified. slub_debug Enable all diagnostics for all caches slub_debug=<options> Enable selective options for all caches slub_debug=<o>,<cache> Enable selective options for a certain set of caches Available Debug options F Double Free checking, sanity and resiliency R Red zoning P Object / padding poisoning U Track last free / alloc T Trace all allocs / frees (only use for individual slabs). To use SLUB: Apply this patch and then select SLUB as the default slab allocator. [hugh@veritas.com: fix an oops-causing locking error] [akpm@linux-foundation.org: various stupid cleanups and small fixes] Signed-off-by: Christoph Lameter <clameter@sgi.com> Signed-off-by: Hugh Dickins <hugh@veritas.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2007-05-06 14:49:36 -07:00
if (page->inuse) {
SLUB: Add MIN_PARTIAL We leave a mininum of partial slabs on nodes when we search for partial slabs on other node. Define a constant for that value. Then modify slub to keep MIN_PARTIAL slabs around. This avoids bad situations where a function frees the last object in a slab (which results in the page being returned to the page allocator) only to then allocate one again (which requires getting a page back from the page allocator if the partial list was empty). Keeping a couple of slabs on the partial list reduces overhead. Empty slabs are added to the end of the partial list to insure that partially allocated slabs are consumed first (defragmentation). Signed-off-by: Christoph Lameter <clameter@sgi.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2007-05-06 14:49:44 -07:00
SLUB core This is a new slab allocator which was motivated by the complexity of the existing code in mm/slab.c. It attempts to address a variety of concerns with the existing implementation. A. Management of object queues A particular concern was the complex management of the numerous object queues in SLAB. SLUB has no such queues. Instead we dedicate a slab for each allocating CPU and use objects from a slab directly instead of queueing them up. B. Storage overhead of object queues SLAB Object queues exist per node, per CPU. The alien cache queue even has a queue array that contain a queue for each processor on each node. For very large systems the number of queues and the number of objects that may be caught in those queues grows exponentially. On our systems with 1k nodes / processors we have several gigabytes just tied up for storing references to objects for those queues This does not include the objects that could be on those queues. One fears that the whole memory of the machine could one day be consumed by those queues. C. SLAB meta data overhead SLAB has overhead at the beginning of each slab. This means that data cannot be naturally aligned at the beginning of a slab block. SLUB keeps all meta data in the corresponding page_struct. Objects can be naturally aligned in the slab. F.e. a 128 byte object will be aligned at 128 byte boundaries and can fit tightly into a 4k page with no bytes left over. SLAB cannot do this. D. SLAB has a complex cache reaper SLUB does not need a cache reaper for UP systems. On SMP systems the per CPU slab may be pushed back into partial list but that operation is simple and does not require an iteration over a list of objects. SLAB expires per CPU, shared and alien object queues during cache reaping which may cause strange hold offs. E. SLAB has complex NUMA policy layer support SLUB pushes NUMA policy handling into the page allocator. This means that allocation is coarser (SLUB does interleave on a page level) but that situation was also present before 2.6.13. SLABs application of policies to individual slab objects allocated in SLAB is certainly a performance concern due to the frequent references to memory policies which may lead a sequence of objects to come from one node after another. SLUB will get a slab full of objects from one node and then will switch to the next. F. Reduction of the size of partial slab lists SLAB has per node partial lists. This means that over time a large number of partial slabs may accumulate on those lists. These can only be reused if allocator occur on specific nodes. SLUB has a global pool of partial slabs and will consume slabs from that pool to decrease fragmentation. G. Tunables SLAB has sophisticated tuning abilities for each slab cache. One can manipulate the queue sizes in detail. However, filling the queues still requires the uses of the spin lock to check out slabs. SLUB has a global parameter (min_slab_order) for tuning. Increasing the minimum slab order can decrease the locking overhead. The bigger the slab order the less motions of pages between per CPU and partial lists occur and the better SLUB will be scaling. G. Slab merging We often have slab caches with similar parameters. SLUB detects those on boot up and merges them into the corresponding general caches. This leads to more effective memory use. About 50% of all caches can be eliminated through slab merging. This will also decrease slab fragmentation because partial allocated slabs can be filled up again. Slab merging can be switched off by specifying slub_nomerge on boot up. Note that merging can expose heretofore unknown bugs in the kernel because corrupted objects may now be placed differently and corrupt differing neighboring objects. Enable sanity checks to find those. H. Diagnostics The current slab diagnostics are difficult to use and require a recompilation of the kernel. SLUB contains debugging code that is always available (but is kept out of the hot code paths). SLUB diagnostics can be enabled via the "slab_debug" option. Parameters can be specified to select a single or a group of slab caches for diagnostics. This means that the system is running with the usual performance and it is much more likely that race conditions can be reproduced. I. Resiliency If basic sanity checks are on then SLUB is capable of detecting common error conditions and recover as best as possible to allow the system to continue. J. Tracing Tracing can be enabled via the slab_debug=T,<slabcache> option during boot. SLUB will then protocol all actions on that slabcache and dump the object contents on free. K. On demand DMA cache creation. Generally DMA caches are not needed. If a kmalloc is used with __GFP_DMA then just create this single slabcache that is needed. For systems that have no ZONE_DMA requirement the support is completely eliminated. L. Performance increase Some benchmarks have shown speed improvements on kernbench in the range of 5-10%. The locking overhead of slub is based on the underlying base allocation size. If we can reliably allocate larger order pages then it is possible to increase slub performance much further. The anti-fragmentation patches may enable further performance increases. Tested on: i386 UP + SMP, x86_64 UP + SMP + NUMA emulation, IA64 NUMA + Simulator SLUB Boot options slub_nomerge Disable merging of slabs slub_min_order=x Require a minimum order for slab caches. This increases the managed chunk size and therefore reduces meta data and locking overhead. slub_min_objects=x Mininum objects per slab. Default is 8. slub_max_order=x Avoid generating slabs larger than order specified. slub_debug Enable all diagnostics for all caches slub_debug=<options> Enable selective options for all caches slub_debug=<o>,<cache> Enable selective options for a certain set of caches Available Debug options F Double Free checking, sanity and resiliency R Red zoning P Object / padding poisoning U Track last free / alloc T Trace all allocs / frees (only use for individual slabs). To use SLUB: Apply this patch and then select SLUB as the default slab allocator. [hugh@veritas.com: fix an oops-causing locking error] [akpm@linux-foundation.org: various stupid cleanups and small fixes] Signed-off-by: Christoph Lameter <clameter@sgi.com> Signed-off-by: Hugh Dickins <hugh@veritas.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2007-05-06 14:49:36 -07:00
if (page->freelist)
SLUB: Add MIN_PARTIAL We leave a mininum of partial slabs on nodes when we search for partial slabs on other node. Define a constant for that value. Then modify slub to keep MIN_PARTIAL slabs around. This avoids bad situations where a function frees the last object in a slab (which results in the page being returned to the page allocator) only to then allocate one again (which requires getting a page back from the page allocator if the partial list was empty). Keeping a couple of slabs on the partial list reduces overhead. Empty slabs are added to the end of the partial list to insure that partially allocated slabs are consumed first (defragmentation). Signed-off-by: Christoph Lameter <clameter@sgi.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2007-05-06 14:49:44 -07:00
add_partial(n, page);
else if (PageError(page) && (s->flags & SLAB_STORE_USER))
add_full(n, page);
SLUB core This is a new slab allocator which was motivated by the complexity of the existing code in mm/slab.c. It attempts to address a variety of concerns with the existing implementation. A. Management of object queues A particular concern was the complex management of the numerous object queues in SLAB. SLUB has no such queues. Instead we dedicate a slab for each allocating CPU and use objects from a slab directly instead of queueing them up. B. Storage overhead of object queues SLAB Object queues exist per node, per CPU. The alien cache queue even has a queue array that contain a queue for each processor on each node. For very large systems the number of queues and the number of objects that may be caught in those queues grows exponentially. On our systems with 1k nodes / processors we have several gigabytes just tied up for storing references to objects for those queues This does not include the objects that could be on those queues. One fears that the whole memory of the machine could one day be consumed by those queues. C. SLAB meta data overhead SLAB has overhead at the beginning of each slab. This means that data cannot be naturally aligned at the beginning of a slab block. SLUB keeps all meta data in the corresponding page_struct. Objects can be naturally aligned in the slab. F.e. a 128 byte object will be aligned at 128 byte boundaries and can fit tightly into a 4k page with no bytes left over. SLAB cannot do this. D. SLAB has a complex cache reaper SLUB does not need a cache reaper for UP systems. On SMP systems the per CPU slab may be pushed back into partial list but that operation is simple and does not require an iteration over a list of objects. SLAB expires per CPU, shared and alien object queues during cache reaping which may cause strange hold offs. E. SLAB has complex NUMA policy layer support SLUB pushes NUMA policy handling into the page allocator. This means that allocation is coarser (SLUB does interleave on a page level) but that situation was also present before 2.6.13. SLABs application of policies to individual slab objects allocated in SLAB is certainly a performance concern due to the frequent references to memory policies which may lead a sequence of objects to come from one node after another. SLUB will get a slab full of objects from one node and then will switch to the next. F. Reduction of the size of partial slab lists SLAB has per node partial lists. This means that over time a large number of partial slabs may accumulate on those lists. These can only be reused if allocator occur on specific nodes. SLUB has a global pool of partial slabs and will consume slabs from that pool to decrease fragmentation. G. Tunables SLAB has sophisticated tuning abilities for each slab cache. One can manipulate the queue sizes in detail. However, filling the queues still requires the uses of the spin lock to check out slabs. SLUB has a global parameter (min_slab_order) for tuning. Increasing the minimum slab order can decrease the locking overhead. The bigger the slab order the less motions of pages between per CPU and partial lists occur and the better SLUB will be scaling. G. Slab merging We often have slab caches with similar parameters. SLUB detects those on boot up and merges them into the corresponding general caches. This leads to more effective memory use. About 50% of all caches can be eliminated through slab merging. This will also decrease slab fragmentation because partial allocated slabs can be filled up again. Slab merging can be switched off by specifying slub_nomerge on boot up. Note that merging can expose heretofore unknown bugs in the kernel because corrupted objects may now be placed differently and corrupt differing neighboring objects. Enable sanity checks to find those. H. Diagnostics The current slab diagnostics are difficult to use and require a recompilation of the kernel. SLUB contains debugging code that is always available (but is kept out of the hot code paths). SLUB diagnostics can be enabled via the "slab_debug" option. Parameters can be specified to select a single or a group of slab caches for diagnostics. This means that the system is running with the usual performance and it is much more likely that race conditions can be reproduced. I. Resiliency If basic sanity checks are on then SLUB is capable of detecting common error conditions and recover as best as possible to allow the system to continue. J. Tracing Tracing can be enabled via the slab_debug=T,<slabcache> option during boot. SLUB will then protocol all actions on that slabcache and dump the object contents on free. K. On demand DMA cache creation. Generally DMA caches are not needed. If a kmalloc is used with __GFP_DMA then just create this single slabcache that is needed. For systems that have no ZONE_DMA requirement the support is completely eliminated. L. Performance increase Some benchmarks have shown speed improvements on kernbench in the range of 5-10%. The locking overhead of slub is based on the underlying base allocation size. If we can reliably allocate larger order pages then it is possible to increase slub performance much further. The anti-fragmentation patches may enable further performance increases. Tested on: i386 UP + SMP, x86_64 UP + SMP + NUMA emulation, IA64 NUMA + Simulator SLUB Boot options slub_nomerge Disable merging of slabs slub_min_order=x Require a minimum order for slab caches. This increases the managed chunk size and therefore reduces meta data and locking overhead. slub_min_objects=x Mininum objects per slab. Default is 8. slub_max_order=x Avoid generating slabs larger than order specified. slub_debug Enable all diagnostics for all caches slub_debug=<options> Enable selective options for all caches slub_debug=<o>,<cache> Enable selective options for a certain set of caches Available Debug options F Double Free checking, sanity and resiliency R Red zoning P Object / padding poisoning U Track last free / alloc T Trace all allocs / frees (only use for individual slabs). To use SLUB: Apply this patch and then select SLUB as the default slab allocator. [hugh@veritas.com: fix an oops-causing locking error] [akpm@linux-foundation.org: various stupid cleanups and small fixes] Signed-off-by: Christoph Lameter <clameter@sgi.com> Signed-off-by: Hugh Dickins <hugh@veritas.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2007-05-06 14:49:36 -07:00
slab_unlock(page);
SLUB: Add MIN_PARTIAL We leave a mininum of partial slabs on nodes when we search for partial slabs on other node. Define a constant for that value. Then modify slub to keep MIN_PARTIAL slabs around. This avoids bad situations where a function frees the last object in a slab (which results in the page being returned to the page allocator) only to then allocate one again (which requires getting a page back from the page allocator if the partial list was empty). Keeping a couple of slabs on the partial list reduces overhead. Empty slabs are added to the end of the partial list to insure that partially allocated slabs are consumed first (defragmentation). Signed-off-by: Christoph Lameter <clameter@sgi.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2007-05-06 14:49:44 -07:00
SLUB core This is a new slab allocator which was motivated by the complexity of the existing code in mm/slab.c. It attempts to address a variety of concerns with the existing implementation. A. Management of object queues A particular concern was the complex management of the numerous object queues in SLAB. SLUB has no such queues. Instead we dedicate a slab for each allocating CPU and use objects from a slab directly instead of queueing them up. B. Storage overhead of object queues SLAB Object queues exist per node, per CPU. The alien cache queue even has a queue array that contain a queue for each processor on each node. For very large systems the number of queues and the number of objects that may be caught in those queues grows exponentially. On our systems with 1k nodes / processors we have several gigabytes just tied up for storing references to objects for those queues This does not include the objects that could be on those queues. One fears that the whole memory of the machine could one day be consumed by those queues. C. SLAB meta data overhead SLAB has overhead at the beginning of each slab. This means that data cannot be naturally aligned at the beginning of a slab block. SLUB keeps all meta data in the corresponding page_struct. Objects can be naturally aligned in the slab. F.e. a 128 byte object will be aligned at 128 byte boundaries and can fit tightly into a 4k page with no bytes left over. SLAB cannot do this. D. SLAB has a complex cache reaper SLUB does not need a cache reaper for UP systems. On SMP systems the per CPU slab may be pushed back into partial list but that operation is simple and does not require an iteration over a list of objects. SLAB expires per CPU, shared and alien object queues during cache reaping which may cause strange hold offs. E. SLAB has complex NUMA policy layer support SLUB pushes NUMA policy handling into the page allocator. This means that allocation is coarser (SLUB does interleave on a page level) but that situation was also present before 2.6.13. SLABs application of policies to individual slab objects allocated in SLAB is certainly a performance concern due to the frequent references to memory policies which may lead a sequence of objects to come from one node after another. SLUB will get a slab full of objects from one node and then will switch to the next. F. Reduction of the size of partial slab lists SLAB has per node partial lists. This means that over time a large number of partial slabs may accumulate on those lists. These can only be reused if allocator occur on specific nodes. SLUB has a global pool of partial slabs and will consume slabs from that pool to decrease fragmentation. G. Tunables SLAB has sophisticated tuning abilities for each slab cache. One can manipulate the queue sizes in detail. However, filling the queues still requires the uses of the spin lock to check out slabs. SLUB has a global parameter (min_slab_order) for tuning. Increasing the minimum slab order can decrease the locking overhead. The bigger the slab order the less motions of pages between per CPU and partial lists occur and the better SLUB will be scaling. G. Slab merging We often have slab caches with similar parameters. SLUB detects those on boot up and merges them into the corresponding general caches. This leads to more effective memory use. About 50% of all caches can be eliminated through slab merging. This will also decrease slab fragmentation because partial allocated slabs can be filled up again. Slab merging can be switched off by specifying slub_nomerge on boot up. Note that merging can expose heretofore unknown bugs in the kernel because corrupted objects may now be placed differently and corrupt differing neighboring objects. Enable sanity checks to find those. H. Diagnostics The current slab diagnostics are difficult to use and require a recompilation of the kernel. SLUB contains debugging code that is always available (but is kept out of the hot code paths). SLUB diagnostics can be enabled via the "slab_debug" option. Parameters can be specified to select a single or a group of slab caches for diagnostics. This means that the system is running with the usual performance and it is much more likely that race conditions can be reproduced. I. Resiliency If basic sanity checks are on then SLUB is capable of detecting common error conditions and recover as best as possible to allow the system to continue. J. Tracing Tracing can be enabled via the slab_debug=T,<slabcache> option during boot. SLUB will then protocol all actions on that slabcache and dump the object contents on free. K. On demand DMA cache creation. Generally DMA caches are not needed. If a kmalloc is used with __GFP_DMA then just create this single slabcache that is needed. For systems that have no ZONE_DMA requirement the support is completely eliminated. L. Performance increase Some benchmarks have shown speed improvements on kernbench in the range of 5-10%. The locking overhead of slub is based on the underlying base allocation size. If we can reliably allocate larger order pages then it is possible to increase slub performance much further. The anti-fragmentation patches may enable further performance increases. Tested on: i386 UP + SMP, x86_64 UP + SMP + NUMA emulation, IA64 NUMA + Simulator SLUB Boot options slub_nomerge Disable merging of slabs slub_min_order=x Require a minimum order for slab caches. This increases the managed chunk size and therefore reduces meta data and locking overhead. slub_min_objects=x Mininum objects per slab. Default is 8. slub_max_order=x Avoid generating slabs larger than order specified. slub_debug Enable all diagnostics for all caches slub_debug=<options> Enable selective options for all caches slub_debug=<o>,<cache> Enable selective options for a certain set of caches Available Debug options F Double Free checking, sanity and resiliency R Red zoning P Object / padding poisoning U Track last free / alloc T Trace all allocs / frees (only use for individual slabs). To use SLUB: Apply this patch and then select SLUB as the default slab allocator. [hugh@veritas.com: fix an oops-causing locking error] [akpm@linux-foundation.org: various stupid cleanups and small fixes] Signed-off-by: Christoph Lameter <clameter@sgi.com> Signed-off-by: Hugh Dickins <hugh@veritas.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2007-05-06 14:49:36 -07:00
} else {
SLUB: Add MIN_PARTIAL We leave a mininum of partial slabs on nodes when we search for partial slabs on other node. Define a constant for that value. Then modify slub to keep MIN_PARTIAL slabs around. This avoids bad situations where a function frees the last object in a slab (which results in the page being returned to the page allocator) only to then allocate one again (which requires getting a page back from the page allocator if the partial list was empty). Keeping a couple of slabs on the partial list reduces overhead. Empty slabs are added to the end of the partial list to insure that partially allocated slabs are consumed first (defragmentation). Signed-off-by: Christoph Lameter <clameter@sgi.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2007-05-06 14:49:44 -07:00
if (n->nr_partial < MIN_PARTIAL) {
/*
* Adding an empty page to the partial slabs in order
* to avoid page allocator overhead. This page needs to
* come after all the others that are not fully empty
* in order to make sure that we do maximum
* defragmentation.
*/
add_partial_tail(n, page);
slab_unlock(page);
} else {
slab_unlock(page);
discard_slab(s, page);
}
SLUB core This is a new slab allocator which was motivated by the complexity of the existing code in mm/slab.c. It attempts to address a variety of concerns with the existing implementation. A. Management of object queues A particular concern was the complex management of the numerous object queues in SLAB. SLUB has no such queues. Instead we dedicate a slab for each allocating CPU and use objects from a slab directly instead of queueing them up. B. Storage overhead of object queues SLAB Object queues exist per node, per CPU. The alien cache queue even has a queue array that contain a queue for each processor on each node. For very large systems the number of queues and the number of objects that may be caught in those queues grows exponentially. On our systems with 1k nodes / processors we have several gigabytes just tied up for storing references to objects for those queues This does not include the objects that could be on those queues. One fears that the whole memory of the machine could one day be consumed by those queues. C. SLAB meta data overhead SLAB has overhead at the beginning of each slab. This means that data cannot be naturally aligned at the beginning of a slab block. SLUB keeps all meta data in the corresponding page_struct. Objects can be naturally aligned in the slab. F.e. a 128 byte object will be aligned at 128 byte boundaries and can fit tightly into a 4k page with no bytes left over. SLAB cannot do this. D. SLAB has a complex cache reaper SLUB does not need a cache reaper for UP systems. On SMP systems the per CPU slab may be pushed back into partial list but that operation is simple and does not require an iteration over a list of objects. SLAB expires per CPU, shared and alien object queues during cache reaping which may cause strange hold offs. E. SLAB has complex NUMA policy layer support SLUB pushes NUMA policy handling into the page allocator. This means that allocation is coarser (SLUB does interleave on a page level) but that situation was also present before 2.6.13. SLABs application of policies to individual slab objects allocated in SLAB is certainly a performance concern due to the frequent references to memory policies which may lead a sequence of objects to come from one node after another. SLUB will get a slab full of objects from one node and then will switch to the next. F. Reduction of the size of partial slab lists SLAB has per node partial lists. This means that over time a large number of partial slabs may accumulate on those lists. These can only be reused if allocator occur on specific nodes. SLUB has a global pool of partial slabs and will consume slabs from that pool to decrease fragmentation. G. Tunables SLAB has sophisticated tuning abilities for each slab cache. One can manipulate the queue sizes in detail. However, filling the queues still requires the uses of the spin lock to check out slabs. SLUB has a global parameter (min_slab_order) for tuning. Increasing the minimum slab order can decrease the locking overhead. The bigger the slab order the less motions of pages between per CPU and partial lists occur and the better SLUB will be scaling. G. Slab merging We often have slab caches with similar parameters. SLUB detects those on boot up and merges them into the corresponding general caches. This leads to more effective memory use. About 50% of all caches can be eliminated through slab merging. This will also decrease slab fragmentation because partial allocated slabs can be filled up again. Slab merging can be switched off by specifying slub_nomerge on boot up. Note that merging can expose heretofore unknown bugs in the kernel because corrupted objects may now be placed differently and corrupt differing neighboring objects. Enable sanity checks to find those. H. Diagnostics The current slab diagnostics are difficult to use and require a recompilation of the kernel. SLUB contains debugging code that is always available (but is kept out of the hot code paths). SLUB diagnostics can be enabled via the "slab_debug" option. Parameters can be specified to select a single or a group of slab caches for diagnostics. This means that the system is running with the usual performance and it is much more likely that race conditions can be reproduced. I. Resiliency If basic sanity checks are on then SLUB is capable of detecting common error conditions and recover as best as possible to allow the system to continue. J. Tracing Tracing can be enabled via the slab_debug=T,<slabcache> option during boot. SLUB will then protocol all actions on that slabcache and dump the object contents on free. K. On demand DMA cache creation. Generally DMA caches are not needed. If a kmalloc is used with __GFP_DMA then just create this single slabcache that is needed. For systems that have no ZONE_DMA requirement the support is completely eliminated. L. Performance increase Some benchmarks have shown speed improvements on kernbench in the range of 5-10%. The locking overhead of slub is based on the underlying base allocation size. If we can reliably allocate larger order pages then it is possible to increase slub performance much further. The anti-fragmentation patches may enable further performance increases. Tested on: i386 UP + SMP, x86_64 UP + SMP + NUMA emulation, IA64 NUMA + Simulator SLUB Boot options slub_nomerge Disable merging of slabs slub_min_order=x Require a minimum order for slab caches. This increases the managed chunk size and therefore reduces meta data and locking overhead. slub_min_objects=x Mininum objects per slab. Default is 8. slub_max_order=x Avoid generating slabs larger than order specified. slub_debug Enable all diagnostics for all caches slub_debug=<options> Enable selective options for all caches slub_debug=<o>,<cache> Enable selective options for a certain set of caches Available Debug options F Double Free checking, sanity and resiliency R Red zoning P Object / padding poisoning U Track last free / alloc T Trace all allocs / frees (only use for individual slabs). To use SLUB: Apply this patch and then select SLUB as the default slab allocator. [hugh@veritas.com: fix an oops-causing locking error] [akpm@linux-foundation.org: various stupid cleanups and small fixes] Signed-off-by: Christoph Lameter <clameter@sgi.com> Signed-off-by: Hugh Dickins <hugh@veritas.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2007-05-06 14:49:36 -07:00
}
}
/*
* Remove the cpu slab
*/
static void deactivate_slab(struct kmem_cache *s, struct page *page, int cpu)
{
s->cpu_slab[cpu] = NULL;
ClearPageActive(page);
putback_slab(s, page);
}
static void flush_slab(struct kmem_cache *s, struct page *page, int cpu)
{
slab_lock(page);
deactivate_slab(s, page, cpu);
}
/*
* Flush cpu slab.
* Called from IPI handler with interrupts disabled.
*/
static void __flush_cpu_slab(struct kmem_cache *s, int cpu)
{
struct page *page = s->cpu_slab[cpu];
if (likely(page))
flush_slab(s, page, cpu);
}
static void flush_cpu_slab(void *d)
{
struct kmem_cache *s = d;
int cpu = smp_processor_id();
__flush_cpu_slab(s, cpu);
}
static void flush_all(struct kmem_cache *s)
{
#ifdef CONFIG_SMP
on_each_cpu(flush_cpu_slab, s, 1, 1);
#else
unsigned long flags;
local_irq_save(flags);
flush_cpu_slab(s);
local_irq_restore(flags);
#endif
}
/*
* slab_alloc is optimized to only modify two cachelines on the fast path
* (aside from the stack):
*
* 1. The page struct
* 2. The first cacheline of the object to be allocated.
*
* The only cache lines that are read (apart from code) is the
* per cpu array in the kmem_cache struct.
*
* Fastpath is not possible if we need to get a new slab or have
* debugging enabled (which means all slabs are marked with PageError)
*/
slub: fix object tracking Object tracking did not work the right way for several call chains. Fix this up by adding a new parameter to slub_alloc and slub_free that specifies the caller address explicitly. Signed-off-by: Christoph Lameter <clameter@sgi.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2007-05-06 14:49:42 -07:00
static void *slab_alloc(struct kmem_cache *s,
gfp_t gfpflags, int node, void *addr)
SLUB core This is a new slab allocator which was motivated by the complexity of the existing code in mm/slab.c. It attempts to address a variety of concerns with the existing implementation. A. Management of object queues A particular concern was the complex management of the numerous object queues in SLAB. SLUB has no such queues. Instead we dedicate a slab for each allocating CPU and use objects from a slab directly instead of queueing them up. B. Storage overhead of object queues SLAB Object queues exist per node, per CPU. The alien cache queue even has a queue array that contain a queue for each processor on each node. For very large systems the number of queues and the number of objects that may be caught in those queues grows exponentially. On our systems with 1k nodes / processors we have several gigabytes just tied up for storing references to objects for those queues This does not include the objects that could be on those queues. One fears that the whole memory of the machine could one day be consumed by those queues. C. SLAB meta data overhead SLAB has overhead at the beginning of each slab. This means that data cannot be naturally aligned at the beginning of a slab block. SLUB keeps all meta data in the corresponding page_struct. Objects can be naturally aligned in the slab. F.e. a 128 byte object will be aligned at 128 byte boundaries and can fit tightly into a 4k page with no bytes left over. SLAB cannot do this. D. SLAB has a complex cache reaper SLUB does not need a cache reaper for UP systems. On SMP systems the per CPU slab may be pushed back into partial list but that operation is simple and does not require an iteration over a list of objects. SLAB expires per CPU, shared and alien object queues during cache reaping which may cause strange hold offs. E. SLAB has complex NUMA policy layer support SLUB pushes NUMA policy handling into the page allocator. This means that allocation is coarser (SLUB does interleave on a page level) but that situation was also present before 2.6.13. SLABs application of policies to individual slab objects allocated in SLAB is certainly a performance concern due to the frequent references to memory policies which may lead a sequence of objects to come from one node after another. SLUB will get a slab full of objects from one node and then will switch to the next. F. Reduction of the size of partial slab lists SLAB has per node partial lists. This means that over time a large number of partial slabs may accumulate on those lists. These can only be reused if allocator occur on specific nodes. SLUB has a global pool of partial slabs and will consume slabs from that pool to decrease fragmentation. G. Tunables SLAB has sophisticated tuning abilities for each slab cache. One can manipulate the queue sizes in detail. However, filling the queues still requires the uses of the spin lock to check out slabs. SLUB has a global parameter (min_slab_order) for tuning. Increasing the minimum slab order can decrease the locking overhead. The bigger the slab order the less motions of pages between per CPU and partial lists occur and the better SLUB will be scaling. G. Slab merging We often have slab caches with similar parameters. SLUB detects those on boot up and merges them into the corresponding general caches. This leads to more effective memory use. About 50% of all caches can be eliminated through slab merging. This will also decrease slab fragmentation because partial allocated slabs can be filled up again. Slab merging can be switched off by specifying slub_nomerge on boot up. Note that merging can expose heretofore unknown bugs in the kernel because corrupted objects may now be placed differently and corrupt differing neighboring objects. Enable sanity checks to find those. H. Diagnostics The current slab diagnostics are difficult to use and require a recompilation of the kernel. SLUB contains debugging code that is always available (but is kept out of the hot code paths). SLUB diagnostics can be enabled via the "slab_debug" option. Parameters can be specified to select a single or a group of slab caches for diagnostics. This means that the system is running with the usual performance and it is much more likely that race conditions can be reproduced. I. Resiliency If basic sanity checks are on then SLUB is capable of detecting common error conditions and recover as best as possible to allow the system to continue. J. Tracing Tracing can be enabled via the slab_debug=T,<slabcache> option during boot. SLUB will then protocol all actions on that slabcache and dump the object contents on free. K. On demand DMA cache creation. Generally DMA caches are not needed. If a kmalloc is used with __GFP_DMA then just create this single slabcache that is needed. For systems that have no ZONE_DMA requirement the support is completely eliminated. L. Performance increase Some benchmarks have shown speed improvements on kernbench in the range of 5-10%. The locking overhead of slub is based on the underlying base allocation size. If we can reliably allocate larger order pages then it is possible to increase slub performance much further. The anti-fragmentation patches may enable further performance increases. Tested on: i386 UP + SMP, x86_64 UP + SMP + NUMA emulation, IA64 NUMA + Simulator SLUB Boot options slub_nomerge Disable merging of slabs slub_min_order=x Require a minimum order for slab caches. This increases the managed chunk size and therefore reduces meta data and locking overhead. slub_min_objects=x Mininum objects per slab. Default is 8. slub_max_order=x Avoid generating slabs larger than order specified. slub_debug Enable all diagnostics for all caches slub_debug=<options> Enable selective options for all caches slub_debug=<o>,<cache> Enable selective options for a certain set of caches Available Debug options F Double Free checking, sanity and resiliency R Red zoning P Object / padding poisoning U Track last free / alloc T Trace all allocs / frees (only use for individual slabs). To use SLUB: Apply this patch and then select SLUB as the default slab allocator. [hugh@veritas.com: fix an oops-causing locking error] [akpm@linux-foundation.org: various stupid cleanups and small fixes] Signed-off-by: Christoph Lameter <clameter@sgi.com> Signed-off-by: Hugh Dickins <hugh@veritas.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2007-05-06 14:49:36 -07:00
{
struct page *page;
void **object;
unsigned long flags;
int cpu;
local_irq_save(flags);
cpu = smp_processor_id();
page = s->cpu_slab[cpu];
if (!page)
goto new_slab;
slab_lock(page);
if (unlikely(node != -1 && page_to_nid(page) != node))
goto another_slab;
redo:
object = page->freelist;
if (unlikely(!object))
goto another_slab;
if (unlikely(PageError(page)))
goto debug;
have_object:
page->inuse++;
page->freelist = object[page->offset];
slab_unlock(page);
local_irq_restore(flags);
return object;
another_slab:
deactivate_slab(s, page, cpu);
new_slab:
page = get_partial(s, gfpflags, node);
if (likely(page)) {
have_slab:
s->cpu_slab[cpu] = page;
SetPageActive(page);
goto redo;
}
page = new_slab(s, gfpflags, node);
if (page) {
cpu = smp_processor_id();
if (s->cpu_slab[cpu]) {
/*
* Someone else populated the cpu_slab while we enabled
* interrupts, or we have got scheduled on another cpu.
* The page may not be on the requested node.
*/
if (node == -1 ||
page_to_nid(s->cpu_slab[cpu]) == node) {
/*
* Current cpuslab is acceptable and we
* want the current one since its cache hot
*/
discard_slab(s, page);
page = s->cpu_slab[cpu];
slab_lock(page);
goto redo;
}
/* Dump the current slab */
flush_slab(s, s->cpu_slab[cpu], cpu);
}
slab_lock(page);
goto have_slab;
}
local_irq_restore(flags);
return NULL;
debug:
if (!alloc_object_checks(s, page, object))
goto another_slab;
if (s->flags & SLAB_STORE_USER)
slub: fix object tracking Object tracking did not work the right way for several call chains. Fix this up by adding a new parameter to slub_alloc and slub_free that specifies the caller address explicitly. Signed-off-by: Christoph Lameter <clameter@sgi.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2007-05-06 14:49:42 -07:00
set_track(s, object, TRACK_ALLOC, addr);
slub: remove object activities out of checking functions Make sure that the check function really only check things and do not perform activities. Extract the tracing and object seeding out of the two check functions and place them into slab_alloc and slab_free Signed-off-by: Christoph Lameter <clameter@sgi.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2007-05-06 14:49:47 -07:00
if (s->flags & SLAB_TRACE) {
printk(KERN_INFO "TRACE %s alloc 0x%p inuse=%d fp=0x%p\n",
s->name, object, page->inuse,
page->freelist);
dump_stack();
}
init_object(s, object, 1);
SLUB core This is a new slab allocator which was motivated by the complexity of the existing code in mm/slab.c. It attempts to address a variety of concerns with the existing implementation. A. Management of object queues A particular concern was the complex management of the numerous object queues in SLAB. SLUB has no such queues. Instead we dedicate a slab for each allocating CPU and use objects from a slab directly instead of queueing them up. B. Storage overhead of object queues SLAB Object queues exist per node, per CPU. The alien cache queue even has a queue array that contain a queue for each processor on each node. For very large systems the number of queues and the number of objects that may be caught in those queues grows exponentially. On our systems with 1k nodes / processors we have several gigabytes just tied up for storing references to objects for those queues This does not include the objects that could be on those queues. One fears that the whole memory of the machine could one day be consumed by those queues. C. SLAB meta data overhead SLAB has overhead at the beginning of each slab. This means that data cannot be naturally aligned at the beginning of a slab block. SLUB keeps all meta data in the corresponding page_struct. Objects can be naturally aligned in the slab. F.e. a 128 byte object will be aligned at 128 byte boundaries and can fit tightly into a 4k page with no bytes left over. SLAB cannot do this. D. SLAB has a complex cache reaper SLUB does not need a cache reaper for UP systems. On SMP systems the per CPU slab may be pushed back into partial list but that operation is simple and does not require an iteration over a list of objects. SLAB expires per CPU, shared and alien object queues during cache reaping which may cause strange hold offs. E. SLAB has complex NUMA policy layer support SLUB pushes NUMA policy handling into the page allocator. This means that allocation is coarser (SLUB does interleave on a page level) but that situation was also present before 2.6.13. SLABs application of policies to individual slab objects allocated in SLAB is certainly a performance concern due to the frequent references to memory policies which may lead a sequence of objects to come from one node after another. SLUB will get a slab full of objects from one node and then will switch to the next. F. Reduction of the size of partial slab lists SLAB has per node partial lists. This means that over time a large number of partial slabs may accumulate on those lists. These can only be reused if allocator occur on specific nodes. SLUB has a global pool of partial slabs and will consume slabs from that pool to decrease fragmentation. G. Tunables SLAB has sophisticated tuning abilities for each slab cache. One can manipulate the queue sizes in detail. However, filling the queues still requires the uses of the spin lock to check out slabs. SLUB has a global parameter (min_slab_order) for tuning. Increasing the minimum slab order can decrease the locking overhead. The bigger the slab order the less motions of pages between per CPU and partial lists occur and the better SLUB will be scaling. G. Slab merging We often have slab caches with similar parameters. SLUB detects those on boot up and merges them into the corresponding general caches. This leads to more effective memory use. About 50% of all caches can be eliminated through slab merging. This will also decrease slab fragmentation because partial allocated slabs can be filled up again. Slab merging can be switched off by specifying slub_nomerge on boot up. Note that merging can expose heretofore unknown bugs in the kernel because corrupted objects may now be placed differently and corrupt differing neighboring objects. Enable sanity checks to find those. H. Diagnostics The current slab diagnostics are difficult to use and require a recompilation of the kernel. SLUB contains debugging code that is always available (but is kept out of the hot code paths). SLUB diagnostics can be enabled via the "slab_debug" option. Parameters can be specified to select a single or a group of slab caches for diagnostics. This means that the system is running with the usual performance and it is much more likely that race conditions can be reproduced. I. Resiliency If basic sanity checks are on then SLUB is capable of detecting common error conditions and recover as best as possible to allow the system to continue. J. Tracing Tracing can be enabled via the slab_debug=T,<slabcache> option during boot. SLUB will then protocol all actions on that slabcache and dump the object contents on free. K. On demand DMA cache creation. Generally DMA caches are not needed. If a kmalloc is used with __GFP_DMA then just create this single slabcache that is needed. For systems that have no ZONE_DMA requirement the support is completely eliminated. L. Performance increase Some benchmarks have shown speed improvements on kernbench in the range of 5-10%. The locking overhead of slub is based on the underlying base allocation size. If we can reliably allocate larger order pages then it is possible to increase slub performance much further. The anti-fragmentation patches may enable further performance increases. Tested on: i386 UP + SMP, x86_64 UP + SMP + NUMA emulation, IA64 NUMA + Simulator SLUB Boot options slub_nomerge Disable merging of slabs slub_min_order=x Require a minimum order for slab caches. This increases the managed chunk size and therefore reduces meta data and locking overhead. slub_min_objects=x Mininum objects per slab. Default is 8. slub_max_order=x Avoid generating slabs larger than order specified. slub_debug Enable all diagnostics for all caches slub_debug=<options> Enable selective options for all caches slub_debug=<o>,<cache> Enable selective options for a certain set of caches Available Debug options F Double Free checking, sanity and resiliency R Red zoning P Object / padding poisoning U Track last free / alloc T Trace all allocs / frees (only use for individual slabs). To use SLUB: Apply this patch and then select SLUB as the default slab allocator. [hugh@veritas.com: fix an oops-causing locking error] [akpm@linux-foundation.org: various stupid cleanups and small fixes] Signed-off-by: Christoph Lameter <clameter@sgi.com> Signed-off-by: Hugh Dickins <hugh@veritas.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2007-05-06 14:49:36 -07:00
goto have_object;
}
void *kmem_cache_alloc(struct kmem_cache *s, gfp_t gfpflags)
{
slub: fix object tracking Object tracking did not work the right way for several call chains. Fix this up by adding a new parameter to slub_alloc and slub_free that specifies the caller address explicitly. Signed-off-by: Christoph Lameter <clameter@sgi.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2007-05-06 14:49:42 -07:00
return slab_alloc(s, gfpflags, -1, __builtin_return_address(0));
SLUB core This is a new slab allocator which was motivated by the complexity of the existing code in mm/slab.c. It attempts to address a variety of concerns with the existing implementation. A. Management of object queues A particular concern was the complex management of the numerous object queues in SLAB. SLUB has no such queues. Instead we dedicate a slab for each allocating CPU and use objects from a slab directly instead of queueing them up. B. Storage overhead of object queues SLAB Object queues exist per node, per CPU. The alien cache queue even has a queue array that contain a queue for each processor on each node. For very large systems the number of queues and the number of objects that may be caught in those queues grows exponentially. On our systems with 1k nodes / processors we have several gigabytes just tied up for storing references to objects for those queues This does not include the objects that could be on those queues. One fears that the whole memory of the machine could one day be consumed by those queues. C. SLAB meta data overhead SLAB has overhead at the beginning of each slab. This means that data cannot be naturally aligned at the beginning of a slab block. SLUB keeps all meta data in the corresponding page_struct. Objects can be naturally aligned in the slab. F.e. a 128 byte object will be aligned at 128 byte boundaries and can fit tightly into a 4k page with no bytes left over. SLAB cannot do this. D. SLAB has a complex cache reaper SLUB does not need a cache reaper for UP systems. On SMP systems the per CPU slab may be pushed back into partial list but that operation is simple and does not require an iteration over a list of objects. SLAB expires per CPU, shared and alien object queues during cache reaping which may cause strange hold offs. E. SLAB has complex NUMA policy layer support SLUB pushes NUMA policy handling into the page allocator. This means that allocation is coarser (SLUB does interleave on a page level) but that situation was also present before 2.6.13. SLABs application of policies to individual slab objects allocated in SLAB is certainly a performance concern due to the frequent references to memory policies which may lead a sequence of objects to come from one node after another. SLUB will get a slab full of objects from one node and then will switch to the next. F. Reduction of the size of partial slab lists SLAB has per node partial lists. This means that over time a large number of partial slabs may accumulate on those lists. These can only be reused if allocator occur on specific nodes. SLUB has a global pool of partial slabs and will consume slabs from that pool to decrease fragmentation. G. Tunables SLAB has sophisticated tuning abilities for each slab cache. One can manipulate the queue sizes in detail. However, filling the queues still requires the uses of the spin lock to check out slabs. SLUB has a global parameter (min_slab_order) for tuning. Increasing the minimum slab order can decrease the locking overhead. The bigger the slab order the less motions of pages between per CPU and partial lists occur and the better SLUB will be scaling. G. Slab merging We often have slab caches with similar parameters. SLUB detects those on boot up and merges them into the corresponding general caches. This leads to more effective memory use. About 50% of all caches can be eliminated through slab merging. This will also decrease slab fragmentation because partial allocated slabs can be filled up again. Slab merging can be switched off by specifying slub_nomerge on boot up. Note that merging can expose heretofore unknown bugs in the kernel because corrupted objects may now be placed differently and corrupt differing neighboring objects. Enable sanity checks to find those. H. Diagnostics The current slab diagnostics are difficult to use and require a recompilation of the kernel. SLUB contains debugging code that is always available (but is kept out of the hot code paths). SLUB diagnostics can be enabled via the "slab_debug" option. Parameters can be specified to select a single or a group of slab caches for diagnostics. This means that the system is running with the usual performance and it is much more likely that race conditions can be reproduced. I. Resiliency If basic sanity checks are on then SLUB is capable of detecting common error conditions and recover as best as possible to allow the system to continue. J. Tracing Tracing can be enabled via the slab_debug=T,<slabcache> option during boot. SLUB will then protocol all actions on that slabcache and dump the object contents on free. K. On demand DMA cache creation. Generally DMA caches are not needed. If a kmalloc is used with __GFP_DMA then just create this single slabcache that is needed. For systems that have no ZONE_DMA requirement the support is completely eliminated. L. Performance increase Some benchmarks have shown speed improvements on kernbench in the range of 5-10%. The locking overhead of slub is based on the underlying base allocation size. If we can reliably allocate larger order pages then it is possible to increase slub performance much further. The anti-fragmentation patches may enable further performance increases. Tested on: i386 UP + SMP, x86_64 UP + SMP + NUMA emulation, IA64 NUMA + Simulator SLUB Boot options slub_nomerge Disable merging of slabs slub_min_order=x Require a minimum order for slab caches. This increases the managed chunk size and therefore reduces meta data and locking overhead. slub_min_objects=x Mininum objects per slab. Default is 8. slub_max_order=x Avoid generating slabs larger than order specified. slub_debug Enable all diagnostics for all caches slub_debug=<options> Enable selective options for all caches slub_debug=<o>,<cache> Enable selective options for a certain set of caches Available Debug options F Double Free checking, sanity and resiliency R Red zoning P Object / padding poisoning U Track last free / alloc T Trace all allocs / frees (only use for individual slabs). To use SLUB: Apply this patch and then select SLUB as the default slab allocator. [hugh@veritas.com: fix an oops-causing locking error] [akpm@linux-foundation.org: various stupid cleanups and small fixes] Signed-off-by: Christoph Lameter <clameter@sgi.com> Signed-off-by: Hugh Dickins <hugh@veritas.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2007-05-06 14:49:36 -07:00
}
EXPORT_SYMBOL(kmem_cache_alloc);
#ifdef CONFIG_NUMA
void *kmem_cache_alloc_node(struct kmem_cache *s, gfp_t gfpflags, int node)
{
slub: fix object tracking Object tracking did not work the right way for several call chains. Fix this up by adding a new parameter to slub_alloc and slub_free that specifies the caller address explicitly. Signed-off-by: Christoph Lameter <clameter@sgi.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2007-05-06 14:49:42 -07:00
return slab_alloc(s, gfpflags, node, __builtin_return_address(0));
SLUB core This is a new slab allocator which was motivated by the complexity of the existing code in mm/slab.c. It attempts to address a variety of concerns with the existing implementation. A. Management of object queues A particular concern was the complex management of the numerous object queues in SLAB. SLUB has no such queues. Instead we dedicate a slab for each allocating CPU and use objects from a slab directly instead of queueing them up. B. Storage overhead of object queues SLAB Object queues exist per node, per CPU. The alien cache queue even has a queue array that contain a queue for each processor on each node. For very large systems the number of queues and the number of objects that may be caught in those queues grows exponentially. On our systems with 1k nodes / processors we have several gigabytes just tied up for storing references to objects for those queues This does not include the objects that could be on those queues. One fears that the whole memory of the machine could one day be consumed by those queues. C. SLAB meta data overhead SLAB has overhead at the beginning of each slab. This means that data cannot be naturally aligned at the beginning of a slab block. SLUB keeps all meta data in the corresponding page_struct. Objects can be naturally aligned in the slab. F.e. a 128 byte object will be aligned at 128 byte boundaries and can fit tightly into a 4k page with no bytes left over. SLAB cannot do this. D. SLAB has a complex cache reaper SLUB does not need a cache reaper for UP systems. On SMP systems the per CPU slab may be pushed back into partial list but that operation is simple and does not require an iteration over a list of objects. SLAB expires per CPU, shared and alien object queues during cache reaping which may cause strange hold offs. E. SLAB has complex NUMA policy layer support SLUB pushes NUMA policy handling into the page allocator. This means that allocation is coarser (SLUB does interleave on a page level) but that situation was also present before 2.6.13. SLABs application of policies to individual slab objects allocated in SLAB is certainly a performance concern due to the frequent references to memory policies which may lead a sequence of objects to come from one node after another. SLUB will get a slab full of objects from one node and then will switch to the next. F. Reduction of the size of partial slab lists SLAB has per node partial lists. This means that over time a large number of partial slabs may accumulate on those lists. These can only be reused if allocator occur on specific nodes. SLUB has a global pool of partial slabs and will consume slabs from that pool to decrease fragmentation. G. Tunables SLAB has sophisticated tuning abilities for each slab cache. One can manipulate the queue sizes in detail. However, filling the queues still requires the uses of the spin lock to check out slabs. SLUB has a global parameter (min_slab_order) for tuning. Increasing the minimum slab order can decrease the locking overhead. The bigger the slab order the less motions of pages between per CPU and partial lists occur and the better SLUB will be scaling. G. Slab merging We often have slab caches with similar parameters. SLUB detects those on boot up and merges them into the corresponding general caches. This leads to more effective memory use. About 50% of all caches can be eliminated through slab merging. This will also decrease slab fragmentation because partial allocated slabs can be filled up again. Slab merging can be switched off by specifying slub_nomerge on boot up. Note that merging can expose heretofore unknown bugs in the kernel because corrupted objects may now be placed differently and corrupt differing neighboring objects. Enable sanity checks to find those. H. Diagnostics The current slab diagnostics are difficult to use and require a recompilation of the kernel. SLUB contains debugging code that is always available (but is kept out of the hot code paths). SLUB diagnostics can be enabled via the "slab_debug" option. Parameters can be specified to select a single or a group of slab caches for diagnostics. This means that the system is running with the usual performance and it is much more likely that race conditions can be reproduced. I. Resiliency If basic sanity checks are on then SLUB is capable of detecting common error conditions and recover as best as possible to allow the system to continue. J. Tracing Tracing can be enabled via the slab_debug=T,<slabcache> option during boot. SLUB will then protocol all actions on that slabcache and dump the object contents on free. K. On demand DMA cache creation. Generally DMA caches are not needed. If a kmalloc is used with __GFP_DMA then just create this single slabcache that is needed. For systems that have no ZONE_DMA requirement the support is completely eliminated. L. Performance increase Some benchmarks have shown speed improvements on kernbench in the range of 5-10%. The locking overhead of slub is based on the underlying base allocation size. If we can reliably allocate larger order pages then it is possible to increase slub performance much further. The anti-fragmentation patches may enable further performance increases. Tested on: i386 UP + SMP, x86_64 UP + SMP + NUMA emulation, IA64 NUMA + Simulator SLUB Boot options slub_nomerge Disable merging of slabs slub_min_order=x Require a minimum order for slab caches. This increases the managed chunk size and therefore reduces meta data and locking overhead. slub_min_objects=x Mininum objects per slab. Default is 8. slub_max_order=x Avoid generating slabs larger than order specified. slub_debug Enable all diagnostics for all caches slub_debug=<options> Enable selective options for all caches slub_debug=<o>,<cache> Enable selective options for a certain set of caches Available Debug options F Double Free checking, sanity and resiliency R Red zoning P Object / padding poisoning U Track last free / alloc T Trace all allocs / frees (only use for individual slabs). To use SLUB: Apply this patch and then select SLUB as the default slab allocator. [hugh@veritas.com: fix an oops-causing locking error] [akpm@linux-foundation.org: various stupid cleanups and small fixes] Signed-off-by: Christoph Lameter <clameter@sgi.com> Signed-off-by: Hugh Dickins <hugh@veritas.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2007-05-06 14:49:36 -07:00
}
EXPORT_SYMBOL(kmem_cache_alloc_node);
#endif
/*
* The fastpath only writes the cacheline of the page struct and the first
* cacheline of the object.
*
* No special cachelines need to be read
*/
slub: fix object tracking Object tracking did not work the right way for several call chains. Fix this up by adding a new parameter to slub_alloc and slub_free that specifies the caller address explicitly. Signed-off-by: Christoph Lameter <clameter@sgi.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2007-05-06 14:49:42 -07:00
static void slab_free(struct kmem_cache *s, struct page *page,
void *x, void *addr)
SLUB core This is a new slab allocator which was motivated by the complexity of the existing code in mm/slab.c. It attempts to address a variety of concerns with the existing implementation. A. Management of object queues A particular concern was the complex management of the numerous object queues in SLAB. SLUB has no such queues. Instead we dedicate a slab for each allocating CPU and use objects from a slab directly instead of queueing them up. B. Storage overhead of object queues SLAB Object queues exist per node, per CPU. The alien cache queue even has a queue array that contain a queue for each processor on each node. For very large systems the number of queues and the number of objects that may be caught in those queues grows exponentially. On our systems with 1k nodes / processors we have several gigabytes just tied up for storing references to objects for those queues This does not include the objects that could be on those queues. One fears that the whole memory of the machine could one day be consumed by those queues. C. SLAB meta data overhead SLAB has overhead at the beginning of each slab. This means that data cannot be naturally aligned at the beginning of a slab block. SLUB keeps all meta data in the corresponding page_struct. Objects can be naturally aligned in the slab. F.e. a 128 byte object will be aligned at 128 byte boundaries and can fit tightly into a 4k page with no bytes left over. SLAB cannot do this. D. SLAB has a complex cache reaper SLUB does not need a cache reaper for UP systems. On SMP systems the per CPU slab may be pushed back into partial list but that operation is simple and does not require an iteration over a list of objects. SLAB expires per CPU, shared and alien object queues during cache reaping which may cause strange hold offs. E. SLAB has complex NUMA policy layer support SLUB pushes NUMA policy handling into the page allocator. This means that allocation is coarser (SLUB does interleave on a page level) but that situation was also present before 2.6.13. SLABs application of policies to individual slab objects allocated in SLAB is certainly a performance concern due to the frequent references to memory policies which may lead a sequence of objects to come from one node after another. SLUB will get a slab full of objects from one node and then will switch to the next. F. Reduction of the size of partial slab lists SLAB has per node partial lists. This means that over time a large number of partial slabs may accumulate on those lists. These can only be reused if allocator occur on specific nodes. SLUB has a global pool of partial slabs and will consume slabs from that pool to decrease fragmentation. G. Tunables SLAB has sophisticated tuning abilities for each slab cache. One can manipulate the queue sizes in detail. However, filling the queues still requires the uses of the spin lock to check out slabs. SLUB has a global parameter (min_slab_order) for tuning. Increasing the minimum slab order can decrease the locking overhead. The bigger the slab order the less motions of pages between per CPU and partial lists occur and the better SLUB will be scaling. G. Slab merging We often have slab caches with similar parameters. SLUB detects those on boot up and merges them into the corresponding general caches. This leads to more effective memory use. About 50% of all caches can be eliminated through slab merging. This will also decrease slab fragmentation because partial allocated slabs can be filled up again. Slab merging can be switched off by specifying slub_nomerge on boot up. Note that merging can expose heretofore unknown bugs in the kernel because corrupted objects may now be placed differently and corrupt differing neighboring objects. Enable sanity checks to find those. H. Diagnostics The current slab diagnostics are difficult to use and require a recompilation of the kernel. SLUB contains debugging code that is always available (but is kept out of the hot code paths). SLUB diagnostics can be enabled via the "slab_debug" option. Parameters can be specified to select a single or a group of slab caches for diagnostics. This means that the system is running with the usual performance and it is much more likely that race conditions can be reproduced. I. Resiliency If basic sanity checks are on then SLUB is capable of detecting common error conditions and recover as best as possible to allow the system to continue. J. Tracing Tracing can be enabled via the slab_debug=T,<slabcache> option during boot. SLUB will then protocol all actions on that slabcache and dump the object contents on free. K. On demand DMA cache creation. Generally DMA caches are not needed. If a kmalloc is used with __GFP_DMA then just create this single slabcache that is needed. For systems that have no ZONE_DMA requirement the support is completely eliminated. L. Performance increase Some benchmarks have shown speed improvements on kernbench in the range of 5-10%. The locking overhead of slub is based on the underlying base allocation size. If we can reliably allocate larger order pages then it is possible to increase slub performance much further. The anti-fragmentation patches may enable further performance increases. Tested on: i386 UP + SMP, x86_64 UP + SMP + NUMA emulation, IA64 NUMA + Simulator SLUB Boot options slub_nomerge Disable merging of slabs slub_min_order=x Require a minimum order for slab caches. This increases the managed chunk size and therefore reduces meta data and locking overhead. slub_min_objects=x Mininum objects per slab. Default is 8. slub_max_order=x Avoid generating slabs larger than order specified. slub_debug Enable all diagnostics for all caches slub_debug=<options> Enable selective options for all caches slub_debug=<o>,<cache> Enable selective options for a certain set of caches Available Debug options F Double Free checking, sanity and resiliency R Red zoning P Object / padding poisoning U Track last free / alloc T Trace all allocs / frees (only use for individual slabs). To use SLUB: Apply this patch and then select SLUB as the default slab allocator. [hugh@veritas.com: fix an oops-causing locking error] [akpm@linux-foundation.org: various stupid cleanups and small fixes] Signed-off-by: Christoph Lameter <clameter@sgi.com> Signed-off-by: Hugh Dickins <hugh@veritas.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2007-05-06 14:49:36 -07:00
{
void *prior;
void **object = (void *)x;
unsigned long flags;
local_irq_save(flags);
slab_lock(page);
if (unlikely(PageError(page)))
goto debug;
checks_ok:
prior = object[page->offset] = page->freelist;
page->freelist = object;
page->inuse--;
if (unlikely(PageActive(page)))
/*
* Cpu slabs are never on partial lists and are
* never freed.
*/
goto out_unlock;
if (unlikely(!page->inuse))
goto slab_empty;
/*
* Objects left in the slab. If it
* was not on the partial list before
* then add it.
*/
if (unlikely(!prior))
SLUB: Add MIN_PARTIAL We leave a mininum of partial slabs on nodes when we search for partial slabs on other node. Define a constant for that value. Then modify slub to keep MIN_PARTIAL slabs around. This avoids bad situations where a function frees the last object in a slab (which results in the page being returned to the page allocator) only to then allocate one again (which requires getting a page back from the page allocator if the partial list was empty). Keeping a couple of slabs on the partial list reduces overhead. Empty slabs are added to the end of the partial list to insure that partially allocated slabs are consumed first (defragmentation). Signed-off-by: Christoph Lameter <clameter@sgi.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2007-05-06 14:49:44 -07:00
add_partial(get_node(s, page_to_nid(page)), page);
SLUB core This is a new slab allocator which was motivated by the complexity of the existing code in mm/slab.c. It attempts to address a variety of concerns with the existing implementation. A. Management of object queues A particular concern was the complex management of the numerous object queues in SLAB. SLUB has no such queues. Instead we dedicate a slab for each allocating CPU and use objects from a slab directly instead of queueing them up. B. Storage overhead of object queues SLAB Object queues exist per node, per CPU. The alien cache queue even has a queue array that contain a queue for each processor on each node. For very large systems the number of queues and the number of objects that may be caught in those queues grows exponentially. On our systems with 1k nodes / processors we have several gigabytes just tied up for storing references to objects for those queues This does not include the objects that could be on those queues. One fears that the whole memory of the machine could one day be consumed by those queues. C. SLAB meta data overhead SLAB has overhead at the beginning of each slab. This means that data cannot be naturally aligned at the beginning of a slab block. SLUB keeps all meta data in the corresponding page_struct. Objects can be naturally aligned in the slab. F.e. a 128 byte object will be aligned at 128 byte boundaries and can fit tightly into a 4k page with no bytes left over. SLAB cannot do this. D. SLAB has a complex cache reaper SLUB does not need a cache reaper for UP systems. On SMP systems the per CPU slab may be pushed back into partial list but that operation is simple and does not require an iteration over a list of objects. SLAB expires per CPU, shared and alien object queues during cache reaping which may cause strange hold offs. E. SLAB has complex NUMA policy layer support SLUB pushes NUMA policy handling into the page allocator. This means that allocation is coarser (SLUB does interleave on a page level) but that situation was also present before 2.6.13. SLABs application of policies to individual slab objects allocated in SLAB is certainly a performance concern due to the frequent references to memory policies which may lead a sequence of objects to come from one node after another. SLUB will get a slab full of objects from one node and then will switch to the next. F. Reduction of the size of partial slab lists SLAB has per node partial lists. This means that over time a large number of partial slabs may accumulate on those lists. These can only be reused if allocator occur on specific nodes. SLUB has a global pool of partial slabs and will consume slabs from that pool to decrease fragmentation. G. Tunables SLAB has sophisticated tuning abilities for each slab cache. One can manipulate the queue sizes in detail. However, filling the queues still requires the uses of the spin lock to check out slabs. SLUB has a global parameter (min_slab_order) for tuning. Increasing the minimum slab order can decrease the locking overhead. The bigger the slab order the less motions of pages between per CPU and partial lists occur and the better SLUB will be scaling. G. Slab merging We often have slab caches with similar parameters. SLUB detects those on boot up and merges them into the corresponding general caches. This leads to more effective memory use. About 50% of all caches can be eliminated through slab merging. This will also decrease slab fragmentation because partial allocated slabs can be filled up again. Slab merging can be switched off by specifying slub_nomerge on boot up. Note that merging can expose heretofore unknown bugs in the kernel because corrupted objects may now be placed differently and corrupt differing neighboring objects. Enable sanity checks to find those. H. Diagnostics The current slab diagnostics are difficult to use and require a recompilation of the kernel. SLUB contains debugging code that is always available (but is kept out of the hot code paths). SLUB diagnostics can be enabled via the "slab_debug" option. Parameters can be specified to select a single or a group of slab caches for diagnostics. This means that the system is running with the usual performance and it is much more likely that race conditions can be reproduced. I. Resiliency If basic sanity checks are on then SLUB is capable of detecting common error conditions and recover as best as possible to allow the system to continue. J. Tracing Tracing can be enabled via the slab_debug=T,<slabcache> option during boot. SLUB will then protocol all actions on that slabcache and dump the object contents on free. K. On demand DMA cache creation. Generally DMA caches are not needed. If a kmalloc is used with __GFP_DMA then just create this single slabcache that is needed. For systems that have no ZONE_DMA requirement the support is completely eliminated. L. Performance increase Some benchmarks have shown speed improvements on kernbench in the range of 5-10%. The locking overhead of slub is based on the underlying base allocation size. If we can reliably allocate larger order pages then it is possible to increase slub performance much further. The anti-fragmentation patches may enable further performance increases. Tested on: i386 UP + SMP, x86_64 UP + SMP + NUMA emulation, IA64 NUMA + Simulator SLUB Boot options slub_nomerge Disable merging of slabs slub_min_order=x Require a minimum order for slab caches. This increases the managed chunk size and therefore reduces meta data and locking overhead. slub_min_objects=x Mininum objects per slab. Default is 8. slub_max_order=x Avoid generating slabs larger than order specified. slub_debug Enable all diagnostics for all caches slub_debug=<options> Enable selective options for all caches slub_debug=<o>,<cache> Enable selective options for a certain set of caches Available Debug options F Double Free checking, sanity and resiliency R Red zoning P Object / padding poisoning U Track last free / alloc T Trace all allocs / frees (only use for individual slabs). To use SLUB: Apply this patch and then select SLUB as the default slab allocator. [hugh@veritas.com: fix an oops-causing locking error] [akpm@linux-foundation.org: various stupid cleanups and small fixes] Signed-off-by: Christoph Lameter <clameter@sgi.com> Signed-off-by: Hugh Dickins <hugh@veritas.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2007-05-06 14:49:36 -07:00
out_unlock:
slab_unlock(page);
local_irq_restore(flags);
return;
slab_empty:
if (prior)
/*
slub: enable tracking of full slabs If slab tracking is on then build a list of full slabs so that we can verify the integrity of all slabs and are also able to built list of alloc/free callers. Signed-off-by: Christoph Lameter <clameter@sgi.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2007-05-06 14:49:42 -07:00
* Slab on the partial list.
SLUB core This is a new slab allocator which was motivated by the complexity of the existing code in mm/slab.c. It attempts to address a variety of concerns with the existing implementation. A. Management of object queues A particular concern was the complex management of the numerous object queues in SLAB. SLUB has no such queues. Instead we dedicate a slab for each allocating CPU and use objects from a slab directly instead of queueing them up. B. Storage overhead of object queues SLAB Object queues exist per node, per CPU. The alien cache queue even has a queue array that contain a queue for each processor on each node. For very large systems the number of queues and the number of objects that may be caught in those queues grows exponentially. On our systems with 1k nodes / processors we have several gigabytes just tied up for storing references to objects for those queues This does not include the objects that could be on those queues. One fears that the whole memory of the machine could one day be consumed by those queues. C. SLAB meta data overhead SLAB has overhead at the beginning of each slab. This means that data cannot be naturally aligned at the beginning of a slab block. SLUB keeps all meta data in the corresponding page_struct. Objects can be naturally aligned in the slab. F.e. a 128 byte object will be aligned at 128 byte boundaries and can fit tightly into a 4k page with no bytes left over. SLAB cannot do this. D. SLAB has a complex cache reaper SLUB does not need a cache reaper for UP systems. On SMP systems the per CPU slab may be pushed back into partial list but that operation is simple and does not require an iteration over a list of objects. SLAB expires per CPU, shared and alien object queues during cache reaping which may cause strange hold offs. E. SLAB has complex NUMA policy layer support SLUB pushes NUMA policy handling into the page allocator. This means that allocation is coarser (SLUB does interleave on a page level) but that situation was also present before 2.6.13. SLABs application of policies to individual slab objects allocated in SLAB is certainly a performance concern due to the frequent references to memory policies which may lead a sequence of objects to come from one node after another. SLUB will get a slab full of objects from one node and then will switch to the next. F. Reduction of the size of partial slab lists SLAB has per node partial lists. This means that over time a large number of partial slabs may accumulate on those lists. These can only be reused if allocator occur on specific nodes. SLUB has a global pool of partial slabs and will consume slabs from that pool to decrease fragmentation. G. Tunables SLAB has sophisticated tuning abilities for each slab cache. One can manipulate the queue sizes in detail. However, filling the queues still requires the uses of the spin lock to check out slabs. SLUB has a global parameter (min_slab_order) for tuning. Increasing the minimum slab order can decrease the locking overhead. The bigger the slab order the less motions of pages between per CPU and partial lists occur and the better SLUB will be scaling. G. Slab merging We often have slab caches with similar parameters. SLUB detects those on boot up and merges them into the corresponding general caches. This leads to more effective memory use. About 50% of all caches can be eliminated through slab merging. This will also decrease slab fragmentation because partial allocated slabs can be filled up again. Slab merging can be switched off by specifying slub_nomerge on boot up. Note that merging can expose heretofore unknown bugs in the kernel because corrupted objects may now be placed differently and corrupt differing neighboring objects. Enable sanity checks to find those. H. Diagnostics The current slab diagnostics are difficult to use and require a recompilation of the kernel. SLUB contains debugging code that is always available (but is kept out of the hot code paths). SLUB diagnostics can be enabled via the "slab_debug" option. Parameters can be specified to select a single or a group of slab caches for diagnostics. This means that the system is running with the usual performance and it is much more likely that race conditions can be reproduced. I. Resiliency If basic sanity checks are on then SLUB is capable of detecting common error conditions and recover as best as possible to allow the system to continue. J. Tracing Tracing can be enabled via the slab_debug=T,<slabcache> option during boot. SLUB will then protocol all actions on that slabcache and dump the object contents on free. K. On demand DMA cache creation. Generally DMA caches are not needed. If a kmalloc is used with __GFP_DMA then just create this single slabcache that is needed. For systems that have no ZONE_DMA requirement the support is completely eliminated. L. Performance increase Some benchmarks have shown speed improvements on kernbench in the range of 5-10%. The locking overhead of slub is based on the underlying base allocation size. If we can reliably allocate larger order pages then it is possible to increase slub performance much further. The anti-fragmentation patches may enable further performance increases. Tested on: i386 UP + SMP, x86_64 UP + SMP + NUMA emulation, IA64 NUMA + Simulator SLUB Boot options slub_nomerge Disable merging of slabs slub_min_order=x Require a minimum order for slab caches. This increases the managed chunk size and therefore reduces meta data and locking overhead. slub_min_objects=x Mininum objects per slab. Default is 8. slub_max_order=x Avoid generating slabs larger than order specified. slub_debug Enable all diagnostics for all caches slub_debug=<options> Enable selective options for all caches slub_debug=<o>,<cache> Enable selective options for a certain set of caches Available Debug options F Double Free checking, sanity and resiliency R Red zoning P Object / padding poisoning U Track last free / alloc T Trace all allocs / frees (only use for individual slabs). To use SLUB: Apply this patch and then select SLUB as the default slab allocator. [hugh@veritas.com: fix an oops-causing locking error] [akpm@linux-foundation.org: various stupid cleanups and small fixes] Signed-off-by: Christoph Lameter <clameter@sgi.com> Signed-off-by: Hugh Dickins <hugh@veritas.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2007-05-06 14:49:36 -07:00
*/
remove_partial(s, page);
slab_unlock(page);
discard_slab(s, page);
local_irq_restore(flags);
return;
debug:
slub: fix object tracking Object tracking did not work the right way for several call chains. Fix this up by adding a new parameter to slub_alloc and slub_free that specifies the caller address explicitly. Signed-off-by: Christoph Lameter <clameter@sgi.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2007-05-06 14:49:42 -07:00
if (!free_object_checks(s, page, x))
goto out_unlock;
slub: enable tracking of full slabs If slab tracking is on then build a list of full slabs so that we can verify the integrity of all slabs and are also able to built list of alloc/free callers. Signed-off-by: Christoph Lameter <clameter@sgi.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2007-05-06 14:49:42 -07:00
if (!PageActive(page) && !page->freelist)
remove_full(s, page);
slub: fix object tracking Object tracking did not work the right way for several call chains. Fix this up by adding a new parameter to slub_alloc and slub_free that specifies the caller address explicitly. Signed-off-by: Christoph Lameter <clameter@sgi.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2007-05-06 14:49:42 -07:00
if (s->flags & SLAB_STORE_USER)
set_track(s, x, TRACK_FREE, addr);
slub: remove object activities out of checking functions Make sure that the check function really only check things and do not perform activities. Extract the tracing and object seeding out of the two check functions and place them into slab_alloc and slab_free Signed-off-by: Christoph Lameter <clameter@sgi.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2007-05-06 14:49:47 -07:00
if (s->flags & SLAB_TRACE) {
printk(KERN_INFO "TRACE %s free 0x%p inuse=%d fp=0x%p\n",
s->name, object, page->inuse,
page->freelist);
print_section("Object", (void *)object, s->objsize);
dump_stack();
}
init_object(s, object, 0);
slub: fix object tracking Object tracking did not work the right way for several call chains. Fix this up by adding a new parameter to slub_alloc and slub_free that specifies the caller address explicitly. Signed-off-by: Christoph Lameter <clameter@sgi.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2007-05-06 14:49:42 -07:00
goto checks_ok;
SLUB core This is a new slab allocator which was motivated by the complexity of the existing code in mm/slab.c. It attempts to address a variety of concerns with the existing implementation. A. Management of object queues A particular concern was the complex management of the numerous object queues in SLAB. SLUB has no such queues. Instead we dedicate a slab for each allocating CPU and use objects from a slab directly instead of queueing them up. B. Storage overhead of object queues SLAB Object queues exist per node, per CPU. The alien cache queue even has a queue array that contain a queue for each processor on each node. For very large systems the number of queues and the number of objects that may be caught in those queues grows exponentially. On our systems with 1k nodes / processors we have several gigabytes just tied up for storing references to objects for those queues This does not include the objects that could be on those queues. One fears that the whole memory of the machine could one day be consumed by those queues. C. SLAB meta data overhead SLAB has overhead at the beginning of each slab. This means that data cannot be naturally aligned at the beginning of a slab block. SLUB keeps all meta data in the corresponding page_struct. Objects can be naturally aligned in the slab. F.e. a 128 byte object will be aligned at 128 byte boundaries and can fit tightly into a 4k page with no bytes left over. SLAB cannot do this. D. SLAB has a complex cache reaper SLUB does not need a cache reaper for UP systems. On SMP systems the per CPU slab may be pushed back into partial list but that operation is simple and does not require an iteration over a list of objects. SLAB expires per CPU, shared and alien object queues during cache reaping which may cause strange hold offs. E. SLAB has complex NUMA policy layer support SLUB pushes NUMA policy handling into the page allocator. This means that allocation is coarser (SLUB does interleave on a page level) but that situation was also present before 2.6.13. SLABs application of policies to individual slab objects allocated in SLAB is certainly a performance concern due to the frequent references to memory policies which may lead a sequence of objects to come from one node after another. SLUB will get a slab full of objects from one node and then will switch to the next. F. Reduction of the size of partial slab lists SLAB has per node partial lists. This means that over time a large number of partial slabs may accumulate on those lists. These can only be reused if allocator occur on specific nodes. SLUB has a global pool of partial slabs and will consume slabs from that pool to decrease fragmentation. G. Tunables SLAB has sophisticated tuning abilities for each slab cache. One can manipulate the queue sizes in detail. However, filling the queues still requires the uses of the spin lock to check out slabs. SLUB has a global parameter (min_slab_order) for tuning. Increasing the minimum slab order can decrease the locking overhead. The bigger the slab order the less motions of pages between per CPU and partial lists occur and the better SLUB will be scaling. G. Slab merging We often have slab caches with similar parameters. SLUB detects those on boot up and merges them into the corresponding general caches. This leads to more effective memory use. About 50% of all caches can be eliminated through slab merging. This will also decrease slab fragmentation because partial allocated slabs can be filled up again. Slab merging can be switched off by specifying slub_nomerge on boot up. Note that merging can expose heretofore unknown bugs in the kernel because corrupted objects may now be placed differently and corrupt differing neighboring objects. Enable sanity checks to find those. H. Diagnostics The current slab diagnostics are difficult to use and require a recompilation of the kernel. SLUB contains debugging code that is always available (but is kept out of the hot code paths). SLUB diagnostics can be enabled via the "slab_debug" option. Parameters can be specified to select a single or a group of slab caches for diagnostics. This means that the system is running with the usual performance and it is much more likely that race conditions can be reproduced. I. Resiliency If basic sanity checks are on then SLUB is capable of detecting common error conditions and recover as best as possible to allow the system to continue. J. Tracing Tracing can be enabled via the slab_debug=T,<slabcache> option during boot. SLUB will then protocol all actions on that slabcache and dump the object contents on free. K. On demand DMA cache creation. Generally DMA caches are not needed. If a kmalloc is used with __GFP_DMA then just create this single slabcache that is needed. For systems that have no ZONE_DMA requirement the support is completely eliminated. L. Performance increase Some benchmarks have shown speed improvements on kernbench in the range of 5-10%. The locking overhead of slub is based on the underlying base allocation size. If we can reliably allocate larger order pages then it is possible to increase slub performance much further. The anti-fragmentation patches may enable further performance increases. Tested on: i386 UP + SMP, x86_64 UP + SMP + NUMA emulation, IA64 NUMA + Simulator SLUB Boot options slub_nomerge Disable merging of slabs slub_min_order=x Require a minimum order for slab caches. This increases the managed chunk size and therefore reduces meta data and locking overhead. slub_min_objects=x Mininum objects per slab. Default is 8. slub_max_order=x Avoid generating slabs larger than order specified. slub_debug Enable all diagnostics for all caches slub_debug=<options> Enable selective options for all caches slub_debug=<o>,<cache> Enable selective options for a certain set of caches Available Debug options F Double Free checking, sanity and resiliency R Red zoning P Object / padding poisoning U Track last free / alloc T Trace all allocs / frees (only use for individual slabs). To use SLUB: Apply this patch and then select SLUB as the default slab allocator. [hugh@veritas.com: fix an oops-causing locking error] [akpm@linux-foundation.org: various stupid cleanups and small fixes] Signed-off-by: Christoph Lameter <clameter@sgi.com> Signed-off-by: Hugh Dickins <hugh@veritas.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2007-05-06 14:49:36 -07:00
}
void kmem_cache_free(struct kmem_cache *s, void *x)
{
slub: fix object tracking Object tracking did not work the right way for several call chains. Fix this up by adding a new parameter to slub_alloc and slub_free that specifies the caller address explicitly. Signed-off-by: Christoph Lameter <clameter@sgi.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2007-05-06 14:49:42 -07:00
struct page *page;
SLUB core This is a new slab allocator which was motivated by the complexity of the existing code in mm/slab.c. It attempts to address a variety of concerns with the existing implementation. A. Management of object queues A particular concern was the complex management of the numerous object queues in SLAB. SLUB has no such queues. Instead we dedicate a slab for each allocating CPU and use objects from a slab directly instead of queueing them up. B. Storage overhead of object queues SLAB Object queues exist per node, per CPU. The alien cache queue even has a queue array that contain a queue for each processor on each node. For very large systems the number of queues and the number of objects that may be caught in those queues grows exponentially. On our systems with 1k nodes / processors we have several gigabytes just tied up for storing references to objects for those queues This does not include the objects that could be on those queues. One fears that the whole memory of the machine could one day be consumed by those queues. C. SLAB meta data overhead SLAB has overhead at the beginning of each slab. This means that data cannot be naturally aligned at the beginning of a slab block. SLUB keeps all meta data in the corresponding page_struct. Objects can be naturally aligned in the slab. F.e. a 128 byte object will be aligned at 128 byte boundaries and can fit tightly into a 4k page with no bytes left over. SLAB cannot do this. D. SLAB has a complex cache reaper SLUB does not need a cache reaper for UP systems. On SMP systems the per CPU slab may be pushed back into partial list but that operation is simple and does not require an iteration over a list of objects. SLAB expires per CPU, shared and alien object queues during cache reaping which may cause strange hold offs. E. SLAB has complex NUMA policy layer support SLUB pushes NUMA policy handling into the page allocator. This means that allocation is coarser (SLUB does interleave on a page level) but that situation was also present before 2.6.13. SLABs application of policies to individual slab objects allocated in SLAB is certainly a performance concern due to the frequent references to memory policies which may lead a sequence of objects to come from one node after another. SLUB will get a slab full of objects from one node and then will switch to the next. F. Reduction of the size of partial slab lists SLAB has per node partial lists. This means that over time a large number of partial slabs may accumulate on those lists. These can only be reused if allocator occur on specific nodes. SLUB has a global pool of partial slabs and will consume slabs from that pool to decrease fragmentation. G. Tunables SLAB has sophisticated tuning abilities for each slab cache. One can manipulate the queue sizes in detail. However, filling the queues still requires the uses of the spin lock to check out slabs. SLUB has a global parameter (min_slab_order) for tuning. Increasing the minimum slab order can decrease the locking overhead. The bigger the slab order the less motions of pages between per CPU and partial lists occur and the better SLUB will be scaling. G. Slab merging We often have slab caches with similar parameters. SLUB detects those on boot up and merges them into the corresponding general caches. This leads to more effective memory use. About 50% of all caches can be eliminated through slab merging. This will also decrease slab fragmentation because partial allocated slabs can be filled up again. Slab merging can be switched off by specifying slub_nomerge on boot up. Note that merging can expose heretofore unknown bugs in the kernel because corrupted objects may now be placed differently and corrupt differing neighboring objects. Enable sanity checks to find those. H. Diagnostics The current slab diagnostics are difficult to use and require a recompilation of the kernel. SLUB contains debugging code that is always available (but is kept out of the hot code paths). SLUB diagnostics can be enabled via the "slab_debug" option. Parameters can be specified to select a single or a group of slab caches for diagnostics. This means that the system is running with the usual performance and it is much more likely that race conditions can be reproduced. I. Resiliency If basic sanity checks are on then SLUB is capable of detecting common error conditions and recover as best as possible to allow the system to continue. J. Tracing Tracing can be enabled via the slab_debug=T,<slabcache> option during boot. SLUB will then protocol all actions on that slabcache and dump the object contents on free. K. On demand DMA cache creation. Generally DMA caches are not needed. If a kmalloc is used with __GFP_DMA then just create this single slabcache that is needed. For systems that have no ZONE_DMA requirement the support is completely eliminated. L. Performance increase Some benchmarks have shown speed improvements on kernbench in the range of 5-10%. The locking overhead of slub is based on the underlying base allocation size. If we can reliably allocate larger order pages then it is possible to increase slub performance much further. The anti-fragmentation patches may enable further performance increases. Tested on: i386 UP + SMP, x86_64 UP + SMP + NUMA emulation, IA64 NUMA + Simulator SLUB Boot options slub_nomerge Disable merging of slabs slub_min_order=x Require a minimum order for slab caches. This increases the managed chunk size and therefore reduces meta data and locking overhead. slub_min_objects=x Mininum objects per slab. Default is 8. slub_max_order=x Avoid generating slabs larger than order specified. slub_debug Enable all diagnostics for all caches slub_debug=<options> Enable selective options for all caches slub_debug=<o>,<cache> Enable selective options for a certain set of caches Available Debug options F Double Free checking, sanity and resiliency R Red zoning P Object / padding poisoning U Track last free / alloc T Trace all allocs / frees (only use for individual slabs). To use SLUB: Apply this patch and then select SLUB as the default slab allocator. [hugh@veritas.com: fix an oops-causing locking error] [akpm@linux-foundation.org: various stupid cleanups and small fixes] Signed-off-by: Christoph Lameter <clameter@sgi.com> Signed-off-by: Hugh Dickins <hugh@veritas.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2007-05-06 14:49:36 -07:00
Add virt_to_head_page and consolidate code in slab and slub Signed-off-by: Christoph Lameter <clameter@sgi.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2007-05-06 14:49:41 -07:00
page = virt_to_head_page(x);
SLUB core This is a new slab allocator which was motivated by the complexity of the existing code in mm/slab.c. It attempts to address a variety of concerns with the existing implementation. A. Management of object queues A particular concern was the complex management of the numerous object queues in SLAB. SLUB has no such queues. Instead we dedicate a slab for each allocating CPU and use objects from a slab directly instead of queueing them up. B. Storage overhead of object queues SLAB Object queues exist per node, per CPU. The alien cache queue even has a queue array that contain a queue for each processor on each node. For very large systems the number of queues and the number of objects that may be caught in those queues grows exponentially. On our systems with 1k nodes / processors we have several gigabytes just tied up for storing references to objects for those queues This does not include the objects that could be on those queues. One fears that the whole memory of the machine could one day be consumed by those queues. C. SLAB meta data overhead SLAB has overhead at the beginning of each slab. This means that data cannot be naturally aligned at the beginning of a slab block. SLUB keeps all meta data in the corresponding page_struct. Objects can be naturally aligned in the slab. F.e. a 128 byte object will be aligned at 128 byte boundaries and can fit tightly into a 4k page with no bytes left over. SLAB cannot do this. D. SLAB has a complex cache reaper SLUB does not need a cache reaper for UP systems. On SMP systems the per CPU slab may be pushed back into partial list but that operation is simple and does not require an iteration over a list of objects. SLAB expires per CPU, shared and alien object queues during cache reaping which may cause strange hold offs. E. SLAB has complex NUMA policy layer support SLUB pushes NUMA policy handling into the page allocator. This means that allocation is coarser (SLUB does interleave on a page level) but that situation was also present before 2.6.13. SLABs application of policies to individual slab objects allocated in SLAB is certainly a performance concern due to the frequent references to memory policies which may lead a sequence of objects to come from one node after another. SLUB will get a slab full of objects from one node and then will switch to the next. F. Reduction of the size of partial slab lists SLAB has per node partial lists. This means that over time a large number of partial slabs may accumulate on those lists. These can only be reused if allocator occur on specific nodes. SLUB has a global pool of partial slabs and will consume slabs from that pool to decrease fragmentation. G. Tunables SLAB has sophisticated tuning abilities for each slab cache. One can manipulate the queue sizes in detail. However, filling the queues still requires the uses of the spin lock to check out slabs. SLUB has a global parameter (min_slab_order) for tuning. Increasing the minimum slab order can decrease the locking overhead. The bigger the slab order the less motions of pages between per CPU and partial lists occur and the better SLUB will be scaling. G. Slab merging We often have slab caches with similar parameters. SLUB detects those on boot up and merges them into the corresponding general caches. This leads to more effective memory use. About 50% of all caches can be eliminated through slab merging. This will also decrease slab fragmentation because partial allocated slabs can be filled up again. Slab merging can be switched off by specifying slub_nomerge on boot up. Note that merging can expose heretofore unknown bugs in the kernel because corrupted objects may now be placed differently and corrupt differing neighboring objects. Enable sanity checks to find those. H. Diagnostics The current slab diagnostics are difficult to use and require a recompilation of the kernel. SLUB contains debugging code that is always available (but is kept out of the hot code paths). SLUB diagnostics can be enabled via the "slab_debug" option. Parameters can be specified to select a single or a group of slab caches for diagnostics. This means that the system is running with the usual performance and it is much more likely that race conditions can be reproduced. I. Resiliency If basic sanity checks are on then SLUB is capable of detecting common error conditions and recover as best as possible to allow the system to continue. J. Tracing Tracing can be enabled via the slab_debug=T,<slabcache> option during boot. SLUB will then protocol all actions on that slabcache and dump the object contents on free. K. On demand DMA cache creation. Generally DMA caches are not needed. If a kmalloc is used with __GFP_DMA then just create this single slabcache that is needed. For systems that have no ZONE_DMA requirement the support is completely eliminated. L. Performance increase Some benchmarks have shown speed improvements on kernbench in the range of 5-10%. The locking overhead of slub is based on the underlying base allocation size. If we can reliably allocate larger order pages then it is possible to increase slub performance much further. The anti-fragmentation patches may enable further performance increases. Tested on: i386 UP + SMP, x86_64 UP + SMP + NUMA emulation, IA64 NUMA + Simulator SLUB Boot options slub_nomerge Disable merging of slabs slub_min_order=x Require a minimum order for slab caches. This increases the managed chunk size and therefore reduces meta data and locking overhead. slub_min_objects=x Mininum objects per slab. Default is 8. slub_max_order=x Avoid generating slabs larger than order specified. slub_debug Enable all diagnostics for all caches slub_debug=<options> Enable selective options for all caches slub_debug=<o>,<cache> Enable selective options for a certain set of caches Available Debug options F Double Free checking, sanity and resiliency R Red zoning P Object / padding poisoning U Track last free / alloc T Trace all allocs / frees (only use for individual slabs). To use SLUB: Apply this patch and then select SLUB as the default slab allocator. [hugh@veritas.com: fix an oops-causing locking error] [akpm@linux-foundation.org: various stupid cleanups and small fixes] Signed-off-by: Christoph Lameter <clameter@sgi.com> Signed-off-by: Hugh Dickins <hugh@veritas.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2007-05-06 14:49:36 -07:00
slub: fix object tracking Object tracking did not work the right way for several call chains. Fix this up by adding a new parameter to slub_alloc and slub_free that specifies the caller address explicitly. Signed-off-by: Christoph Lameter <clameter@sgi.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2007-05-06 14:49:42 -07:00
slab_free(s, page, x, __builtin_return_address(0));
SLUB core This is a new slab allocator which was motivated by the complexity of the existing code in mm/slab.c. It attempts to address a variety of concerns with the existing implementation. A. Management of object queues A particular concern was the complex management of the numerous object queues in SLAB. SLUB has no such queues. Instead we dedicate a slab for each allocating CPU and use objects from a slab directly instead of queueing them up. B. Storage overhead of object queues SLAB Object queues exist per node, per CPU. The alien cache queue even has a queue array that contain a queue for each processor on each node. For very large systems the number of queues and the number of objects that may be caught in those queues grows exponentially. On our systems with 1k nodes / processors we have several gigabytes just tied up for storing references to objects for those queues This does not include the objects that could be on those queues. One fears that the whole memory of the machine could one day be consumed by those queues. C. SLAB meta data overhead SLAB has overhead at the beginning of each slab. This means that data cannot be naturally aligned at the beginning of a slab block. SLUB keeps all meta data in the corresponding page_struct. Objects can be naturally aligned in the slab. F.e. a 128 byte object will be aligned at 128 byte boundaries and can fit tightly into a 4k page with no bytes left over. SLAB cannot do this. D. SLAB has a complex cache reaper SLUB does not need a cache reaper for UP systems. On SMP systems the per CPU slab may be pushed back into partial list but that operation is simple and does not require an iteration over a list of objects. SLAB expires per CPU, shared and alien object queues during cache reaping which may cause strange hold offs. E. SLAB has complex NUMA policy layer support SLUB pushes NUMA policy handling into the page allocator. This means that allocation is coarser (SLUB does interleave on a page level) but that situation was also present before 2.6.13. SLABs application of policies to individual slab objects allocated in SLAB is certainly a performance concern due to the frequent references to memory policies which may lead a sequence of objects to come from one node after another. SLUB will get a slab full of objects from one node and then will switch to the next. F. Reduction of the size of partial slab lists SLAB has per node partial lists. This means that over time a large number of partial slabs may accumulate on those lists. These can only be reused if allocator occur on specific nodes. SLUB has a global pool of partial slabs and will consume slabs from that pool to decrease fragmentation. G. Tunables SLAB has sophisticated tuning abilities for each slab cache. One can manipulate the queue sizes in detail. However, filling the queues still requires the uses of the spin lock to check out slabs. SLUB has a global parameter (min_slab_order) for tuning. Increasing the minimum slab order can decrease the locking overhead. The bigger the slab order the less motions of pages between per CPU and partial lists occur and the better SLUB will be scaling. G. Slab merging We often have slab caches with similar parameters. SLUB detects those on boot up and merges them into the corresponding general caches. This leads to more effective memory use. About 50% of all caches can be eliminated through slab merging. This will also decrease slab fragmentation because partial allocated slabs can be filled up again. Slab merging can be switched off by specifying slub_nomerge on boot up. Note that merging can expose heretofore unknown bugs in the kernel because corrupted objects may now be placed differently and corrupt differing neighboring objects. Enable sanity checks to find those. H. Diagnostics The current slab diagnostics are difficult to use and require a recompilation of the kernel. SLUB contains debugging code that is always available (but is kept out of the hot code paths). SLUB diagnostics can be enabled via the "slab_debug" option. Parameters can be specified to select a single or a group of slab caches for diagnostics. This means that the system is running with the usual performance and it is much more likely that race conditions can be reproduced. I. Resiliency If basic sanity checks are on then SLUB is capable of detecting common error conditions and recover as best as possible to allow the system to continue. J. Tracing Tracing can be enabled via the slab_debug=T,<slabcache> option during boot. SLUB will then protocol all actions on that slabcache and dump the object contents on free. K. On demand DMA cache creation. Generally DMA caches are not needed. If a kmalloc is used with __GFP_DMA then just create this single slabcache that is needed. For systems that have no ZONE_DMA requirement the support is completely eliminated. L. Performance increase Some benchmarks have shown speed improvements on kernbench in the range of 5-10%. The locking overhead of slub is based on the underlying base allocation size. If we can reliably allocate larger order pages then it is possible to increase slub performance much further. The anti-fragmentation patches may enable further performance increases. Tested on: i386 UP + SMP, x86_64 UP + SMP + NUMA emulation, IA64 NUMA + Simulator SLUB Boot options slub_nomerge Disable merging of slabs slub_min_order=x Require a minimum order for slab caches. This increases the managed chunk size and therefore reduces meta data and locking overhead. slub_min_objects=x Mininum objects per slab. Default is 8. slub_max_order=x Avoid generating slabs larger than order specified. slub_debug Enable all diagnostics for all caches slub_debug=<options> Enable selective options for all caches slub_debug=<o>,<cache> Enable selective options for a certain set of caches Available Debug options F Double Free checking, sanity and resiliency R Red zoning P Object / padding poisoning U Track last free / alloc T Trace all allocs / frees (only use for individual slabs). To use SLUB: Apply this patch and then select SLUB as the default slab allocator. [hugh@veritas.com: fix an oops-causing locking error] [akpm@linux-foundation.org: various stupid cleanups and small fixes] Signed-off-by: Christoph Lameter <clameter@sgi.com> Signed-off-by: Hugh Dickins <hugh@veritas.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2007-05-06 14:49:36 -07:00
}
EXPORT_SYMBOL(kmem_cache_free);
/* Figure out on which slab object the object resides */
static struct page *get_object_page(const void *x)
{
Add virt_to_head_page and consolidate code in slab and slub Signed-off-by: Christoph Lameter <clameter@sgi.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2007-05-06 14:49:41 -07:00
struct page *page = virt_to_head_page(x);
SLUB core This is a new slab allocator which was motivated by the complexity of the existing code in mm/slab.c. It attempts to address a variety of concerns with the existing implementation. A. Management of object queues A particular concern was the complex management of the numerous object queues in SLAB. SLUB has no such queues. Instead we dedicate a slab for each allocating CPU and use objects from a slab directly instead of queueing them up. B. Storage overhead of object queues SLAB Object queues exist per node, per CPU. The alien cache queue even has a queue array that contain a queue for each processor on each node. For very large systems the number of queues and the number of objects that may be caught in those queues grows exponentially. On our systems with 1k nodes / processors we have several gigabytes just tied up for storing references to objects for those queues This does not include the objects that could be on those queues. One fears that the whole memory of the machine could one day be consumed by those queues. C. SLAB meta data overhead SLAB has overhead at the beginning of each slab. This means that data cannot be naturally aligned at the beginning of a slab block. SLUB keeps all meta data in the corresponding page_struct. Objects can be naturally aligned in the slab. F.e. a 128 byte object will be aligned at 128 byte boundaries and can fit tightly into a 4k page with no bytes left over. SLAB cannot do this. D. SLAB has a complex cache reaper SLUB does not need a cache reaper for UP systems. On SMP systems the per CPU slab may be pushed back into partial list but that operation is simple and does not require an iteration over a list of objects. SLAB expires per CPU, shared and alien object queues during cache reaping which may cause strange hold offs. E. SLAB has complex NUMA policy layer support SLUB pushes NUMA policy handling into the page allocator. This means that allocation is coarser (SLUB does interleave on a page level) but that situation was also present before 2.6.13. SLABs application of policies to individual slab objects allocated in SLAB is certainly a performance concern due to the frequent references to memory policies which may lead a sequence of objects to come from one node after another. SLUB will get a slab full of objects from one node and then will switch to the next. F. Reduction of the size of partial slab lists SLAB has per node partial lists. This means that over time a large number of partial slabs may accumulate on those lists. These can only be reused if allocator occur on specific nodes. SLUB has a global pool of partial slabs and will consume slabs from that pool to decrease fragmentation. G. Tunables SLAB has sophisticated tuning abilities for each slab cache. One can manipulate the queue sizes in detail. However, filling the queues still requires the uses of the spin lock to check out slabs. SLUB has a global parameter (min_slab_order) for tuning. Increasing the minimum slab order can decrease the locking overhead. The bigger the slab order the less motions of pages between per CPU and partial lists occur and the better SLUB will be scaling. G. Slab merging We often have slab caches with similar parameters. SLUB detects those on boot up and merges them into the corresponding general caches. This leads to more effective memory use. About 50% of all caches can be eliminated through slab merging. This will also decrease slab fragmentation because partial allocated slabs can be filled up again. Slab merging can be switched off by specifying slub_nomerge on boot up. Note that merging can expose heretofore unknown bugs in the kernel because corrupted objects may now be placed differently and corrupt differing neighboring objects. Enable sanity checks to find those. H. Diagnostics The current slab diagnostics are difficult to use and require a recompilation of the kernel. SLUB contains debugging code that is always available (but is kept out of the hot code paths). SLUB diagnostics can be enabled via the "slab_debug" option. Parameters can be specified to select a single or a group of slab caches for diagnostics. This means that the system is running with the usual performance and it is much more likely that race conditions can be reproduced. I. Resiliency If basic sanity checks are on then SLUB is capable of detecting common error conditions and recover as best as possible to allow the system to continue. J. Tracing Tracing can be enabled via the slab_debug=T,<slabcache> option during boot. SLUB will then protocol all actions on that slabcache and dump the object contents on free. K. On demand DMA cache creation. Generally DMA caches are not needed. If a kmalloc is used with __GFP_DMA then just create this single slabcache that is needed. For systems that have no ZONE_DMA requirement the support is completely eliminated. L. Performance increase Some benchmarks have shown speed improvements on kernbench in the range of 5-10%. The locking overhead of slub is based on the underlying base allocation size. If we can reliably allocate larger order pages then it is possible to increase slub performance much further. The anti-fragmentation patches may enable further performance increases. Tested on: i386 UP + SMP, x86_64 UP + SMP + NUMA emulation, IA64 NUMA + Simulator SLUB Boot options slub_nomerge Disable merging of slabs slub_min_order=x Require a minimum order for slab caches. This increases the managed chunk size and therefore reduces meta data and locking overhead. slub_min_objects=x Mininum objects per slab. Default is 8. slub_max_order=x Avoid generating slabs larger than order specified. slub_debug Enable all diagnostics for all caches slub_debug=<options> Enable selective options for all caches slub_debug=<o>,<cache> Enable selective options for a certain set of caches Available Debug options F Double Free checking, sanity and resiliency R Red zoning P Object / padding poisoning U Track last free / alloc T Trace all allocs / frees (only use for individual slabs). To use SLUB: Apply this patch and then select SLUB as the default slab allocator. [hugh@veritas.com: fix an oops-causing locking error] [akpm@linux-foundation.org: various stupid cleanups and small fixes] Signed-off-by: Christoph Lameter <clameter@sgi.com> Signed-off-by: Hugh Dickins <hugh@veritas.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2007-05-06 14:49:36 -07:00
if (!PageSlab(page))
return NULL;
return page;
}
/*
* kmem_cache_open produces objects aligned at "size" and the first object
* is placed at offset 0 in the slab (We have no metainformation on the
* slab, all slabs are in essence "off slab").
*
* In order to get the desired alignment one just needs to align the
* size.
*
* Notice that the allocation order determines the sizes of the per cpu
* caches. Each processor has always one slab available for allocations.
* Increasing the allocation order reduces the number of times that slabs
* must be moved on and off the partial lists and therefore may influence
* locking overhead.
*
* The offset is used to relocate the free list link in each object. It is
* therefore possible to move the free list link behind the object. This
* is necessary for RCU to work properly and also useful for debugging.
*/
/*
* Mininum / Maximum order of slab pages. This influences locking overhead
* and slab fragmentation. A higher order reduces the number of partial slabs
* and increases the number of allocations possible without having to
* take the list_lock.
*/
static int slub_min_order;
static int slub_max_order = DEFAULT_MAX_ORDER;
/*
* Minimum number of objects per slab. This is necessary in order to
* reduce locking overhead. Similar to the queue size in SLAB.
*/
static int slub_min_objects = DEFAULT_MIN_OBJECTS;
/*
* Merge control. If this is set then no merging of slab caches will occur.
*/
static int slub_nomerge;
/*
* Debug settings:
*/
static int slub_debug;
static char *slub_debug_slabs;
/*
* Calculate the order of allocation given an slab object size.
*
* The order of allocation has significant impact on other elements
* of the system. Generally order 0 allocations should be preferred
* since they do not cause fragmentation in the page allocator. Larger
* objects may have problems with order 0 because there may be too much
* space left unused in a slab. We go to a higher order if more than 1/8th
* of the slab would be wasted.
*
* In order to reach satisfactory performance we must ensure that
* a minimum number of objects is in one slab. Otherwise we may
* generate too much activity on the partial lists. This is less a
* concern for large slabs though. slub_max_order specifies the order
* where we begin to stop considering the number of objects in a slab.
*
* Higher order allocations also allow the placement of more objects
* in a slab and thereby reduce object handling overhead. If the user
* has requested a higher mininum order then we start with that one
* instead of zero.
*/
static int calculate_order(int size)
{
int order;
int rem;
for (order = max(slub_min_order, fls(size - 1) - PAGE_SHIFT);
order < MAX_ORDER; order++) {
unsigned long slab_size = PAGE_SIZE << order;
if (slub_max_order > order &&
slab_size < slub_min_objects * size)
continue;
if (slab_size < size)
continue;
rem = slab_size % size;
if (rem <= (PAGE_SIZE << order) / 8)
break;
}
if (order >= MAX_ORDER)
return -E2BIG;
return order;
}
/*
* Function to figure out which alignment to use from the
* various ways of specifying it.
*/
static unsigned long calculate_alignment(unsigned long flags,
unsigned long align, unsigned long size)
{
/*
* If the user wants hardware cache aligned objects then
* follow that suggestion if the object is sufficiently
* large.
*
* The hardware cache alignment cannot override the
* specified alignment though. If that is greater
* then use it.
*/
slab allocators: Remove obsolete SLAB_MUST_HWCACHE_ALIGN This patch was recently posted to lkml and acked by Pekka. The flag SLAB_MUST_HWCACHE_ALIGN is 1. Never checked by SLAB at all. 2. A duplicate of SLAB_HWCACHE_ALIGN for SLUB 3. Fulfills the role of SLAB_HWCACHE_ALIGN for SLOB. The only remaining use is in sparc64 and ppc64 and their use there reflects some earlier role that the slab flag once may have had. If its specified then SLAB_HWCACHE_ALIGN is also specified. The flag is confusing, inconsistent and has no purpose. Remove it. Acked-by: Pekka Enberg <penberg@cs.helsinki.fi> Signed-off-by: Christoph Lameter <clameter@sgi.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2007-05-06 14:49:56 -07:00
if ((flags & SLAB_HWCACHE_ALIGN) &&
SLUB core This is a new slab allocator which was motivated by the complexity of the existing code in mm/slab.c. It attempts to address a variety of concerns with the existing implementation. A. Management of object queues A particular concern was the complex management of the numerous object queues in SLAB. SLUB has no such queues. Instead we dedicate a slab for each allocating CPU and use objects from a slab directly instead of queueing them up. B. Storage overhead of object queues SLAB Object queues exist per node, per CPU. The alien cache queue even has a queue array that contain a queue for each processor on each node. For very large systems the number of queues and the number of objects that may be caught in those queues grows exponentially. On our systems with 1k nodes / processors we have several gigabytes just tied up for storing references to objects for those queues This does not include the objects that could be on those queues. One fears that the whole memory of the machine could one day be consumed by those queues. C. SLAB meta data overhead SLAB has overhead at the beginning of each slab. This means that data cannot be naturally aligned at the beginning of a slab block. SLUB keeps all meta data in the corresponding page_struct. Objects can be naturally aligned in the slab. F.e. a 128 byte object will be aligned at 128 byte boundaries and can fit tightly into a 4k page with no bytes left over. SLAB cannot do this. D. SLAB has a complex cache reaper SLUB does not need a cache reaper for UP systems. On SMP systems the per CPU slab may be pushed back into partial list but that operation is simple and does not require an iteration over a list of objects. SLAB expires per CPU, shared and alien object queues during cache reaping which may cause strange hold offs. E. SLAB has complex NUMA policy layer support SLUB pushes NUMA policy handling into the page allocator. This means that allocation is coarser (SLUB does interleave on a page level) but that situation was also present before 2.6.13. SLABs application of policies to individual slab objects allocated in SLAB is certainly a performance concern due to the frequent references to memory policies which may lead a sequence of objects to come from one node after another. SLUB will get a slab full of objects from one node and then will switch to the next. F. Reduction of the size of partial slab lists SLAB has per node partial lists. This means that over time a large number of partial slabs may accumulate on those lists. These can only be reused if allocator occur on specific nodes. SLUB has a global pool of partial slabs and will consume slabs from that pool to decrease fragmentation. G. Tunables SLAB has sophisticated tuning abilities for each slab cache. One can manipulate the queue sizes in detail. However, filling the queues still requires the uses of the spin lock to check out slabs. SLUB has a global parameter (min_slab_order) for tuning. Increasing the minimum slab order can decrease the locking overhead. The bigger the slab order the less motions of pages between per CPU and partial lists occur and the better SLUB will be scaling. G. Slab merging We often have slab caches with similar parameters. SLUB detects those on boot up and merges them into the corresponding general caches. This leads to more effective memory use. About 50% of all caches can be eliminated through slab merging. This will also decrease slab fragmentation because partial allocated slabs can be filled up again. Slab merging can be switched off by specifying slub_nomerge on boot up. Note that merging can expose heretofore unknown bugs in the kernel because corrupted objects may now be placed differently and corrupt differing neighboring objects. Enable sanity checks to find those. H. Diagnostics The current slab diagnostics are difficult to use and require a recompilation of the kernel. SLUB contains debugging code that is always available (but is kept out of the hot code paths). SLUB diagnostics can be enabled via the "slab_debug" option. Parameters can be specified to select a single or a group of slab caches for diagnostics. This means that the system is running with the usual performance and it is much more likely that race conditions can be reproduced. I. Resiliency If basic sanity checks are on then SLUB is capable of detecting common error conditions and recover as best as possible to allow the system to continue. J. Tracing Tracing can be enabled via the slab_debug=T,<slabcache> option during boot. SLUB will then protocol all actions on that slabcache and dump the object contents on free. K. On demand DMA cache creation. Generally DMA caches are not needed. If a kmalloc is used with __GFP_DMA then just create this single slabcache that is needed. For systems that have no ZONE_DMA requirement the support is completely eliminated. L. Performance increase Some benchmarks have shown speed improvements on kernbench in the range of 5-10%. The locking overhead of slub is based on the underlying base allocation size. If we can reliably allocate larger order pages then it is possible to increase slub performance much further. The anti-fragmentation patches may enable further performance increases. Tested on: i386 UP + SMP, x86_64 UP + SMP + NUMA emulation, IA64 NUMA + Simulator SLUB Boot options slub_nomerge Disable merging of slabs slub_min_order=x Require a minimum order for slab caches. This increases the managed chunk size and therefore reduces meta data and locking overhead. slub_min_objects=x Mininum objects per slab. Default is 8. slub_max_order=x Avoid generating slabs larger than order specified. slub_debug Enable all diagnostics for all caches slub_debug=<options> Enable selective options for all caches slub_debug=<o>,<cache> Enable selective options for a certain set of caches Available Debug options F Double Free checking, sanity and resiliency R Red zoning P Object / padding poisoning U Track last free / alloc T Trace all allocs / frees (only use for individual slabs). To use SLUB: Apply this patch and then select SLUB as the default slab allocator. [hugh@veritas.com: fix an oops-causing locking error] [akpm@linux-foundation.org: various stupid cleanups and small fixes] Signed-off-by: Christoph Lameter <clameter@sgi.com> Signed-off-by: Hugh Dickins <hugh@veritas.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2007-05-06 14:49:36 -07:00
size > L1_CACHE_BYTES / 2)
return max_t(unsigned long, align, L1_CACHE_BYTES);
if (align < ARCH_SLAB_MINALIGN)
return ARCH_SLAB_MINALIGN;
return ALIGN(align, sizeof(void *));
}
static void init_kmem_cache_node(struct kmem_cache_node *n)
{
n->nr_partial = 0;
atomic_long_set(&n->nr_slabs, 0);
spin_lock_init(&n->list_lock);
INIT_LIST_HEAD(&n->partial);
slub: enable tracking of full slabs If slab tracking is on then build a list of full slabs so that we can verify the integrity of all slabs and are also able to built list of alloc/free callers. Signed-off-by: Christoph Lameter <clameter@sgi.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2007-05-06 14:49:42 -07:00
INIT_LIST_HEAD(&n->full);
SLUB core This is a new slab allocator which was motivated by the complexity of the existing code in mm/slab.c. It attempts to address a variety of concerns with the existing implementation. A. Management of object queues A particular concern was the complex management of the numerous object queues in SLAB. SLUB has no such queues. Instead we dedicate a slab for each allocating CPU and use objects from a slab directly instead of queueing them up. B. Storage overhead of object queues SLAB Object queues exist per node, per CPU. The alien cache queue even has a queue array that contain a queue for each processor on each node. For very large systems the number of queues and the number of objects that may be caught in those queues grows exponentially. On our systems with 1k nodes / processors we have several gigabytes just tied up for storing references to objects for those queues This does not include the objects that could be on those queues. One fears that the whole memory of the machine could one day be consumed by those queues. C. SLAB meta data overhead SLAB has overhead at the beginning of each slab. This means that data cannot be naturally aligned at the beginning of a slab block. SLUB keeps all meta data in the corresponding page_struct. Objects can be naturally aligned in the slab. F.e. a 128 byte object will be aligned at 128 byte boundaries and can fit tightly into a 4k page with no bytes left over. SLAB cannot do this. D. SLAB has a complex cache reaper SLUB does not need a cache reaper for UP systems. On SMP systems the per CPU slab may be pushed back into partial list but that operation is simple and does not require an iteration over a list of objects. SLAB expires per CPU, shared and alien object queues during cache reaping which may cause strange hold offs. E. SLAB has complex NUMA policy layer support SLUB pushes NUMA policy handling into the page allocator. This means that allocation is coarser (SLUB does interleave on a page level) but that situation was also present before 2.6.13. SLABs application of policies to individual slab objects allocated in SLAB is certainly a performance concern due to the frequent references to memory policies which may lead a sequence of objects to come from one node after another. SLUB will get a slab full of objects from one node and then will switch to the next. F. Reduction of the size of partial slab lists SLAB has per node partial lists. This means that over time a large number of partial slabs may accumulate on those lists. These can only be reused if allocator occur on specific nodes. SLUB has a global pool of partial slabs and will consume slabs from that pool to decrease fragmentation. G. Tunables SLAB has sophisticated tuning abilities for each slab cache. One can manipulate the queue sizes in detail. However, filling the queues still requires the uses of the spin lock to check out slabs. SLUB has a global parameter (min_slab_order) for tuning. Increasing the minimum slab order can decrease the locking overhead. The bigger the slab order the less motions of pages between per CPU and partial lists occur and the better SLUB will be scaling. G. Slab merging We often have slab caches with similar parameters. SLUB detects those on boot up and merges them into the corresponding general caches. This leads to more effective memory use. About 50% of all caches can be eliminated through slab merging. This will also decrease slab fragmentation because partial allocated slabs can be filled up again. Slab merging can be switched off by specifying slub_nomerge on boot up. Note that merging can expose heretofore unknown bugs in the kernel because corrupted objects may now be placed differently and corrupt differing neighboring objects. Enable sanity checks to find those. H. Diagnostics The current slab diagnostics are difficult to use and require a recompilation of the kernel. SLUB contains debugging code that is always available (but is kept out of the hot code paths). SLUB diagnostics can be enabled via the "slab_debug" option. Parameters can be specified to select a single or a group of slab caches for diagnostics. This means that the system is running with the usual performance and it is much more likely that race conditions can be reproduced. I. Resiliency If basic sanity checks are on then SLUB is capable of detecting common error conditions and recover as best as possible to allow the system to continue. J. Tracing Tracing can be enabled via the slab_debug=T,<slabcache> option during boot. SLUB will then protocol all actions on that slabcache and dump the object contents on free. K. On demand DMA cache creation. Generally DMA caches are not needed. If a kmalloc is used with __GFP_DMA then just create this single slabcache that is needed. For systems that have no ZONE_DMA requirement the support is completely eliminated. L. Performance increase Some benchmarks have shown speed improvements on kernbench in the range of 5-10%. The locking overhead of slub is based on the underlying base allocation size. If we can reliably allocate larger order pages then it is possible to increase slub performance much further. The anti-fragmentation patches may enable further performance increases. Tested on: i386 UP + SMP, x86_64 UP + SMP + NUMA emulation, IA64 NUMA + Simulator SLUB Boot options slub_nomerge Disable merging of slabs slub_min_order=x Require a minimum order for slab caches. This increases the managed chunk size and therefore reduces meta data and locking overhead. slub_min_objects=x Mininum objects per slab. Default is 8. slub_max_order=x Avoid generating slabs larger than order specified. slub_debug Enable all diagnostics for all caches slub_debug=<options> Enable selective options for all caches slub_debug=<o>,<cache> Enable selective options for a certain set of caches Available Debug options F Double Free checking, sanity and resiliency R Red zoning P Object / padding poisoning U Track last free / alloc T Trace all allocs / frees (only use for individual slabs). To use SLUB: Apply this patch and then select SLUB as the default slab allocator. [hugh@veritas.com: fix an oops-causing locking error] [akpm@linux-foundation.org: various stupid cleanups and small fixes] Signed-off-by: Christoph Lameter <clameter@sgi.com> Signed-off-by: Hugh Dickins <hugh@veritas.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2007-05-06 14:49:36 -07:00
}
#ifdef CONFIG_NUMA
/*
* No kmalloc_node yet so do it by hand. We know that this is the first
* slab on the node for this slabcache. There are no concurrent accesses
* possible.
*
* Note that this function only works on the kmalloc_node_cache
* when allocating for the kmalloc_node_cache.
*/
static struct kmem_cache_node * __init early_kmem_cache_node_alloc(gfp_t gfpflags,
int node)
{
struct page *page;
struct kmem_cache_node *n;
BUG_ON(kmalloc_caches->size < sizeof(struct kmem_cache_node));
page = new_slab(kmalloc_caches, gfpflags | GFP_THISNODE, node);
/* new_slab() disables interupts */
local_irq_enable();
BUG_ON(!page);
n = page->freelist;
BUG_ON(!n);
page->freelist = get_freepointer(kmalloc_caches, n);
page->inuse++;
kmalloc_caches->node[node] = n;
init_object(kmalloc_caches, n, 1);
init_kmem_cache_node(n);
atomic_long_inc(&n->nr_slabs);
SLUB: Add MIN_PARTIAL We leave a mininum of partial slabs on nodes when we search for partial slabs on other node. Define a constant for that value. Then modify slub to keep MIN_PARTIAL slabs around. This avoids bad situations where a function frees the last object in a slab (which results in the page being returned to the page allocator) only to then allocate one again (which requires getting a page back from the page allocator if the partial list was empty). Keeping a couple of slabs on the partial list reduces overhead. Empty slabs are added to the end of the partial list to insure that partially allocated slabs are consumed first (defragmentation). Signed-off-by: Christoph Lameter <clameter@sgi.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2007-05-06 14:49:44 -07:00
add_partial(n, page);
SLUB core This is a new slab allocator which was motivated by the complexity of the existing code in mm/slab.c. It attempts to address a variety of concerns with the existing implementation. A. Management of object queues A particular concern was the complex management of the numerous object queues in SLAB. SLUB has no such queues. Instead we dedicate a slab for each allocating CPU and use objects from a slab directly instead of queueing them up. B. Storage overhead of object queues SLAB Object queues exist per node, per CPU. The alien cache queue even has a queue array that contain a queue for each processor on each node. For very large systems the number of queues and the number of objects that may be caught in those queues grows exponentially. On our systems with 1k nodes / processors we have several gigabytes just tied up for storing references to objects for those queues This does not include the objects that could be on those queues. One fears that the whole memory of the machine could one day be consumed by those queues. C. SLAB meta data overhead SLAB has overhead at the beginning of each slab. This means that data cannot be naturally aligned at the beginning of a slab block. SLUB keeps all meta data in the corresponding page_struct. Objects can be naturally aligned in the slab. F.e. a 128 byte object will be aligned at 128 byte boundaries and can fit tightly into a 4k page with no bytes left over. SLAB cannot do this. D. SLAB has a complex cache reaper SLUB does not need a cache reaper for UP systems. On SMP systems the per CPU slab may be pushed back into partial list but that operation is simple and does not require an iteration over a list of objects. SLAB expires per CPU, shared and alien object queues during cache reaping which may cause strange hold offs. E. SLAB has complex NUMA policy layer support SLUB pushes NUMA policy handling into the page allocator. This means that allocation is coarser (SLUB does interleave on a page level) but that situation was also present before 2.6.13. SLABs application of policies to individual slab objects allocated in SLAB is certainly a performance concern due to the frequent references to memory policies which may lead a sequence of objects to come from one node after another. SLUB will get a slab full of objects from one node and then will switch to the next. F. Reduction of the size of partial slab lists SLAB has per node partial lists. This means that over time a large number of partial slabs may accumulate on those lists. These can only be reused if allocator occur on specific nodes. SLUB has a global pool of partial slabs and will consume slabs from that pool to decrease fragmentation. G. Tunables SLAB has sophisticated tuning abilities for each slab cache. One can manipulate the queue sizes in detail. However, filling the queues still requires the uses of the spin lock to check out slabs. SLUB has a global parameter (min_slab_order) for tuning. Increasing the minimum slab order can decrease the locking overhead. The bigger the slab order the less motions of pages between per CPU and partial lists occur and the better SLUB will be scaling. G. Slab merging We often have slab caches with similar parameters. SLUB detects those on boot up and merges them into the corresponding general caches. This leads to more effective memory use. About 50% of all caches can be eliminated through slab merging. This will also decrease slab fragmentation because partial allocated slabs can be filled up again. Slab merging can be switched off by specifying slub_nomerge on boot up. Note that merging can expose heretofore unknown bugs in the kernel because corrupted objects may now be placed differently and corrupt differing neighboring objects. Enable sanity checks to find those. H. Diagnostics The current slab diagnostics are difficult to use and require a recompilation of the kernel. SLUB contains debugging code that is always available (but is kept out of the hot code paths). SLUB diagnostics can be enabled via the "slab_debug" option. Parameters can be specified to select a single or a group of slab caches for diagnostics. This means that the system is running with the usual performance and it is much more likely that race conditions can be reproduced. I. Resiliency If basic sanity checks are on then SLUB is capable of detecting common error conditions and recover as best as possible to allow the system to continue. J. Tracing Tracing can be enabled via the slab_debug=T,<slabcache> option during boot. SLUB will then protocol all actions on that slabcache and dump the object contents on free. K. On demand DMA cache creation. Generally DMA caches are not needed. If a kmalloc is used with __GFP_DMA then just create this single slabcache that is needed. For systems that have no ZONE_DMA requirement the support is completely eliminated. L. Performance increase Some benchmarks have shown speed improvements on kernbench in the range of 5-10%. The locking overhead of slub is based on the underlying base allocation size. If we can reliably allocate larger order pages then it is possible to increase slub performance much further. The anti-fragmentation patches may enable further performance increases. Tested on: i386 UP + SMP, x86_64 UP + SMP + NUMA emulation, IA64 NUMA + Simulator SLUB Boot options slub_nomerge Disable merging of slabs slub_min_order=x Require a minimum order for slab caches. This increases the managed chunk size and therefore reduces meta data and locking overhead. slub_min_objects=x Mininum objects per slab. Default is 8. slub_max_order=x Avoid generating slabs larger than order specified. slub_debug Enable all diagnostics for all caches slub_debug=<options> Enable selective options for all caches slub_debug=<o>,<cache> Enable selective options for a certain set of caches Available Debug options F Double Free checking, sanity and resiliency R Red zoning P Object / padding poisoning U Track last free / alloc T Trace all allocs / frees (only use for individual slabs). To use SLUB: Apply this patch and then select SLUB as the default slab allocator. [hugh@veritas.com: fix an oops-causing locking error] [akpm@linux-foundation.org: various stupid cleanups and small fixes] Signed-off-by: Christoph Lameter <clameter@sgi.com> Signed-off-by: Hugh Dickins <hugh@veritas.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2007-05-06 14:49:36 -07:00
return n;
}
static void free_kmem_cache_nodes(struct kmem_cache *s)
{
int node;
for_each_online_node(node) {
struct kmem_cache_node *n = s->node[node];
if (n && n != &s->local_node)
kmem_cache_free(kmalloc_caches, n);
s->node[node] = NULL;
}
}
static int init_kmem_cache_nodes(struct kmem_cache *s, gfp_t gfpflags)
{
int node;
int local_node;
if (slab_state >= UP)
local_node = page_to_nid(virt_to_page(s));
else
local_node = 0;
for_each_online_node(node) {
struct kmem_cache_node *n;
if (local_node == node)
n = &s->local_node;
else {
if (slab_state == DOWN) {
n = early_kmem_cache_node_alloc(gfpflags,
node);
continue;
}
n = kmem_cache_alloc_node(kmalloc_caches,
gfpflags, node);
if (!n) {
free_kmem_cache_nodes(s);
return 0;
}
}
s->node[node] = n;
init_kmem_cache_node(n);
}
return 1;
}
#else
static void free_kmem_cache_nodes(struct kmem_cache *s)
{
}
static int init_kmem_cache_nodes(struct kmem_cache *s, gfp_t gfpflags)
{
init_kmem_cache_node(&s->local_node);
return 1;
}
#endif
/*
* calculate_sizes() determines the order and the distribution of data within
* a slab object.
*/
static int calculate_sizes(struct kmem_cache *s)
{
unsigned long flags = s->flags;
unsigned long size = s->objsize;
unsigned long align = s->align;
/*
* Determine if we can poison the object itself. If the user of
* the slab may touch the object after free or before allocation
* then we should never poison the object itself.
*/
if ((flags & SLAB_POISON) && !(flags & SLAB_DESTROY_BY_RCU) &&
!s->ctor && !s->dtor)
s->flags |= __OBJECT_POISON;
else
s->flags &= ~__OBJECT_POISON;
/*
* Round up object size to the next word boundary. We can only
* place the free pointer at word boundaries and this determines
* the possible location of the free pointer.
*/
size = ALIGN(size, sizeof(void *));
/*
* If we are redzoning then check if there is some space between the
* end of the object and the free pointer. If not then add an
* additional word, so that we can establish a redzone between
* the object and the freepointer to be able to check for overwrites.
*/
if ((flags & SLAB_RED_ZONE) && size == s->objsize)
size += sizeof(void *);
/*
* With that we have determined how much of the slab is in actual
* use by the object. This is the potential offset to the free
* pointer.
*/
s->inuse = size;
if (((flags & (SLAB_DESTROY_BY_RCU | SLAB_POISON)) ||
s->ctor || s->dtor)) {
/*
* Relocate free pointer after the object if it is not
* permitted to overwrite the first word of the object on
* kmem_cache_free.
*
* This is the case if we do RCU, have a constructor or
* destructor or are poisoning the objects.
*/
s->offset = size;
size += sizeof(void *);
}
if (flags & SLAB_STORE_USER)
/*
* Need to store information about allocs and frees after
* the object.
*/
size += 2 * sizeof(struct track);
if (flags & DEBUG_DEFAULT_FLAGS)
/*
* Add some empty padding so that we can catch
* overwrites from earlier objects rather than let
* tracking information or the free pointer be
* corrupted if an user writes before the start
* of the object.
*/
size += sizeof(void *);
/*
* Determine the alignment based on various parameters that the
* user specified (this is unecessarily complex due to the attempt
* to be compatible with SLAB. Should be cleaned up some day).
*/
align = calculate_alignment(flags, align, s->objsize);
/*
* SLUB stores one object immediately after another beginning from
* offset 0. In order to align the objects we have to simply size
* each object to conform to the alignment.
*/
size = ALIGN(size, align);
s->size = size;
s->order = calculate_order(size);
if (s->order < 0)
return 0;
/*
* Determine the number of objects per slab
*/
s->objects = (PAGE_SIZE << s->order) / size;
/*
* Verify that the number of objects is within permitted limits.
* The page->inuse field is only 16 bit wide! So we cannot have
* more than 64k objects per slab.
*/
if (!s->objects || s->objects > 65535)
return 0;
return 1;
}
static int __init finish_bootstrap(void)
{
struct list_head *h;
int err;
slab_state = SYSFS;
list_for_each(h, &slab_caches) {
struct kmem_cache *s =
container_of(h, struct kmem_cache, list);
err = sysfs_slab_add(s);
BUG_ON(err);
}
return 0;
}
static int kmem_cache_open(struct kmem_cache *s, gfp_t gfpflags,
const char *name, size_t size,
size_t align, unsigned long flags,
void (*ctor)(void *, struct kmem_cache *, unsigned long),
void (*dtor)(void *, struct kmem_cache *, unsigned long))
{
memset(s, 0, kmem_size);
s->name = name;
s->ctor = ctor;
s->dtor = dtor;
s->objsize = size;
s->flags = flags;
s->align = align;
/*
* The page->offset field is only 16 bit wide. This is an offset
* in units of words from the beginning of an object. If the slab
* size is bigger then we cannot move the free pointer behind the
* object anymore.
*
* On 32 bit platforms the limit is 256k. On 64bit platforms
* the limit is 512k.
*
* Debugging or ctor/dtors may create a need to move the free
* pointer. Fail if this happens.
*/
if (s->size >= 65535 * sizeof(void *)) {
BUG_ON(flags & (SLAB_RED_ZONE | SLAB_POISON |
SLAB_STORE_USER | SLAB_DESTROY_BY_RCU));
BUG_ON(ctor || dtor);
}
else
/*
* Enable debugging if selected on the kernel commandline.
*/
if (slub_debug && (!slub_debug_slabs ||
strncmp(slub_debug_slabs, name,
strlen(slub_debug_slabs)) == 0))
s->flags |= slub_debug;
if (!calculate_sizes(s))
goto error;
s->refcount = 1;
#ifdef CONFIG_NUMA
s->defrag_ratio = 100;
#endif
if (init_kmem_cache_nodes(s, gfpflags & ~SLUB_DMA))
return 1;
error:
if (flags & SLAB_PANIC)
panic("Cannot create slab %s size=%lu realsize=%u "
"order=%u offset=%u flags=%lx\n",
s->name, (unsigned long)size, s->size, s->order,
s->offset, flags);
return 0;
}
EXPORT_SYMBOL(kmem_cache_open);
/*
* Check if a given pointer is valid
*/
int kmem_ptr_validate(struct kmem_cache *s, const void *object)
{
struct page * page;
void *addr;
page = get_object_page(object);
if (!page || s != page->slab)
/* No slab or wrong slab */
return 0;
addr = page_address(page);
if (object < addr || object >= addr + s->objects * s->size)
/* Out of bounds */
return 0;
if ((object - addr) % s->size)
/* Improperly aligned */
return 0;
/*
* We could also check if the object is on the slabs freelist.
* But this would be too expensive and it seems that the main
* purpose of kmem_ptr_valid is to check if the object belongs
* to a certain slab.
*/
return 1;
}
EXPORT_SYMBOL(kmem_ptr_validate);
/*
* Determine the size of a slab object
*/
unsigned int kmem_cache_size(struct kmem_cache *s)
{
return s->objsize;
}
EXPORT_SYMBOL(kmem_cache_size);
const char *kmem_cache_name(struct kmem_cache *s)
{
return s->name;
}
EXPORT_SYMBOL(kmem_cache_name);
/*
* Attempt to free all slabs on a node
*/
static int free_list(struct kmem_cache *s, struct kmem_cache_node *n,
struct list_head *list)
{
int slabs_inuse = 0;
unsigned long flags;
struct page *page, *h;
spin_lock_irqsave(&n->list_lock, flags);
list_for_each_entry_safe(page, h, list, lru)
if (!page->inuse) {
list_del(&page->lru);
discard_slab(s, page);
} else
slabs_inuse++;
spin_unlock_irqrestore(&n->list_lock, flags);
return slabs_inuse;
}
/*
* Release all resources used by slab cache
*/
static int kmem_cache_close(struct kmem_cache *s)
{
int node;
flush_all(s);
/* Attempt to free all objects */
for_each_online_node(node) {
struct kmem_cache_node *n = get_node(s, node);
SLUB: Free slabs and sort partial slab lists in kmem_cache_shrink At kmem_cache_shrink check if we have any empty slabs on the partial if so then remove them. Also--as an anti-fragmentation measure--sort the partial slabs so that the most fully allocated ones come first and the least allocated last. The next allocations may fill up the nearly full slabs. Having the least allocated slabs last gives them the maximum chance that their remaining objects may be freed. Thus we can hopefully minimize the partial slabs. I think this is the best one can do in terms antifragmentation measures. Real defragmentation (meaning moving objects out of slabs with the least free objects to those that are almost full) can be implemted by reverse scanning through the list produced here but that would mean that we need to provide a callback at slab cache creation that allows the deletion or moving of an object. This will involve slab API changes, so defer for now. Cc: Mel Gorman <mel@skynet.ie> Signed-off-by: Christoph Lameter <clameter@sgi.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2007-05-06 14:49:46 -07:00
n->nr_partial -= free_list(s, n, &n->partial);
SLUB core This is a new slab allocator which was motivated by the complexity of the existing code in mm/slab.c. It attempts to address a variety of concerns with the existing implementation. A. Management of object queues A particular concern was the complex management of the numerous object queues in SLAB. SLUB has no such queues. Instead we dedicate a slab for each allocating CPU and use objects from a slab directly instead of queueing them up. B. Storage overhead of object queues SLAB Object queues exist per node, per CPU. The alien cache queue even has a queue array that contain a queue for each processor on each node. For very large systems the number of queues and the number of objects that may be caught in those queues grows exponentially. On our systems with 1k nodes / processors we have several gigabytes just tied up for storing references to objects for those queues This does not include the objects that could be on those queues. One fears that the whole memory of the machine could one day be consumed by those queues. C. SLAB meta data overhead SLAB has overhead at the beginning of each slab. This means that data cannot be naturally aligned at the beginning of a slab block. SLUB keeps all meta data in the corresponding page_struct. Objects can be naturally aligned in the slab. F.e. a 128 byte object will be aligned at 128 byte boundaries and can fit tightly into a 4k page with no bytes left over. SLAB cannot do this. D. SLAB has a complex cache reaper SLUB does not need a cache reaper for UP systems. On SMP systems the per CPU slab may be pushed back into partial list but that operation is simple and does not require an iteration over a list of objects. SLAB expires per CPU, shared and alien object queues during cache reaping which may cause strange hold offs. E. SLAB has complex NUMA policy layer support SLUB pushes NUMA policy handling into the page allocator. This means that allocation is coarser (SLUB does interleave on a page level) but that situation was also present before 2.6.13. SLABs application of policies to individual slab objects allocated in SLAB is certainly a performance concern due to the frequent references to memory policies which may lead a sequence of objects to come from one node after another. SLUB will get a slab full of objects from one node and then will switch to the next. F. Reduction of the size of partial slab lists SLAB has per node partial lists. This means that over time a large number of partial slabs may accumulate on those lists. These can only be reused if allocator occur on specific nodes. SLUB has a global pool of partial slabs and will consume slabs from that pool to decrease fragmentation. G. Tunables SLAB has sophisticated tuning abilities for each slab cache. One can manipulate the queue sizes in detail. However, filling the queues still requires the uses of the spin lock to check out slabs. SLUB has a global parameter (min_slab_order) for tuning. Increasing the minimum slab order can decrease the locking overhead. The bigger the slab order the less motions of pages between per CPU and partial lists occur and the better SLUB will be scaling. G. Slab merging We often have slab caches with similar parameters. SLUB detects those on boot up and merges them into the corresponding general caches. This leads to more effective memory use. About 50% of all caches can be eliminated through slab merging. This will also decrease slab fragmentation because partial allocated slabs can be filled up again. Slab merging can be switched off by specifying slub_nomerge on boot up. Note that merging can expose heretofore unknown bugs in the kernel because corrupted objects may now be placed differently and corrupt differing neighboring objects. Enable sanity checks to find those. H. Diagnostics The current slab diagnostics are difficult to use and require a recompilation of the kernel. SLUB contains debugging code that is always available (but is kept out of the hot code paths). SLUB diagnostics can be enabled via the "slab_debug" option. Parameters can be specified to select a single or a group of slab caches for diagnostics. This means that the system is running with the usual performance and it is much more likely that race conditions can be reproduced. I. Resiliency If basic sanity checks are on then SLUB is capable of detecting common error conditions and recover as best as possible to allow the system to continue. J. Tracing Tracing can be enabled via the slab_debug=T,<slabcache> option during boot. SLUB will then protocol all actions on that slabcache and dump the object contents on free. K. On demand DMA cache creation. Generally DMA caches are not needed. If a kmalloc is used with __GFP_DMA then just create this single slabcache that is needed. For systems that have no ZONE_DMA requirement the support is completely eliminated. L. Performance increase Some benchmarks have shown speed improvements on kernbench in the range of 5-10%. The locking overhead of slub is based on the underlying base allocation size. If we can reliably allocate larger order pages then it is possible to increase slub performance much further. The anti-fragmentation patches may enable further performance increases. Tested on: i386 UP + SMP, x86_64 UP + SMP + NUMA emulation, IA64 NUMA + Simulator SLUB Boot options slub_nomerge Disable merging of slabs slub_min_order=x Require a minimum order for slab caches. This increases the managed chunk size and therefore reduces meta data and locking overhead. slub_min_objects=x Mininum objects per slab. Default is 8. slub_max_order=x Avoid generating slabs larger than order specified. slub_debug Enable all diagnostics for all caches slub_debug=<options> Enable selective options for all caches slub_debug=<o>,<cache> Enable selective options for a certain set of caches Available Debug options F Double Free checking, sanity and resiliency R Red zoning P Object / padding poisoning U Track last free / alloc T Trace all allocs / frees (only use for individual slabs). To use SLUB: Apply this patch and then select SLUB as the default slab allocator. [hugh@veritas.com: fix an oops-causing locking error] [akpm@linux-foundation.org: various stupid cleanups and small fixes] Signed-off-by: Christoph Lameter <clameter@sgi.com> Signed-off-by: Hugh Dickins <hugh@veritas.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2007-05-06 14:49:36 -07:00
if (atomic_long_read(&n->nr_slabs))
return 1;
}
free_kmem_cache_nodes(s);
return 0;
}
/*
* Close a cache and release the kmem_cache structure
* (must be used for caches created using kmem_cache_create)
*/
void kmem_cache_destroy(struct kmem_cache *s)
{
down_write(&slub_lock);
s->refcount--;
if (!s->refcount) {
list_del(&s->list);
if (kmem_cache_close(s))
WARN_ON(1);
sysfs_slab_remove(s);
kfree(s);
}
up_write(&slub_lock);
}
EXPORT_SYMBOL(kmem_cache_destroy);
/********************************************************************
* Kmalloc subsystem
*******************************************************************/
struct kmem_cache kmalloc_caches[KMALLOC_SHIFT_HIGH + 1] __cacheline_aligned;
EXPORT_SYMBOL(kmalloc_caches);
#ifdef CONFIG_ZONE_DMA
static struct kmem_cache *kmalloc_caches_dma[KMALLOC_SHIFT_HIGH + 1];
#endif
static int __init setup_slub_min_order(char *str)
{
get_option (&str, &slub_min_order);
return 1;
}
__setup("slub_min_order=", setup_slub_min_order);
static int __init setup_slub_max_order(char *str)
{
get_option (&str, &slub_max_order);
return 1;
}
__setup("slub_max_order=", setup_slub_max_order);
static int __init setup_slub_min_objects(char *str)
{
get_option (&str, &slub_min_objects);
return 1;
}
__setup("slub_min_objects=", setup_slub_min_objects);
static int __init setup_slub_nomerge(char *str)
{
slub_nomerge = 1;
return 1;
}
__setup("slub_nomerge", setup_slub_nomerge);
static int __init setup_slub_debug(char *str)
{
if (!str || *str != '=')
slub_debug = DEBUG_DEFAULT_FLAGS;
else {
str++;
if (*str == 0 || *str == ',')
slub_debug = DEBUG_DEFAULT_FLAGS;
else
for( ;*str && *str != ','; str++)
switch (*str) {
case 'f' : case 'F' :
slub_debug |= SLAB_DEBUG_FREE;
break;
case 'z' : case 'Z' :
slub_debug |= SLAB_RED_ZONE;
break;
case 'p' : case 'P' :
slub_debug |= SLAB_POISON;
break;
case 'u' : case 'U' :
slub_debug |= SLAB_STORE_USER;
break;
case 't' : case 'T' :
slub_debug |= SLAB_TRACE;
break;
default:
printk(KERN_ERR "slub_debug option '%c' "
"unknown. skipped\n",*str);
}
}
if (*str == ',')
slub_debug_slabs = str + 1;
return 1;
}
__setup("slub_debug", setup_slub_debug);
static struct kmem_cache *create_kmalloc_cache(struct kmem_cache *s,
const char *name, int size, gfp_t gfp_flags)
{
unsigned int flags = 0;
if (gfp_flags & SLUB_DMA)
flags = SLAB_CACHE_DMA;
down_write(&slub_lock);
if (!kmem_cache_open(s, gfp_flags, name, size, ARCH_KMALLOC_MINALIGN,
flags, NULL, NULL))
goto panic;
list_add(&s->list, &slab_caches);
up_write(&slub_lock);
if (sysfs_slab_add(s))
goto panic;
return s;
panic:
panic("Creation of kmalloc slab %s size=%d failed.\n", name, size);
}
static struct kmem_cache *get_slab(size_t size, gfp_t flags)
{
int index = kmalloc_index(size);
SLUB: allocate smallest object size if the user asks for 0 bytes Makes SLUB behave like SLAB in this area to avoid issues.... Throw a stack dump to alert people. At some point the behavior should be switched back. NULL is no memory as far as I can tell and if the use asked for 0 bytes then he need to get no memory. Signed-off-by: Christoph Lameter <clameter@sgi.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2007-05-06 14:49:38 -07:00
if (!index)
SLUB core This is a new slab allocator which was motivated by the complexity of the existing code in mm/slab.c. It attempts to address a variety of concerns with the existing implementation. A. Management of object queues A particular concern was the complex management of the numerous object queues in SLAB. SLUB has no such queues. Instead we dedicate a slab for each allocating CPU and use objects from a slab directly instead of queueing them up. B. Storage overhead of object queues SLAB Object queues exist per node, per CPU. The alien cache queue even has a queue array that contain a queue for each processor on each node. For very large systems the number of queues and the number of objects that may be caught in those queues grows exponentially. On our systems with 1k nodes / processors we have several gigabytes just tied up for storing references to objects for those queues This does not include the objects that could be on those queues. One fears that the whole memory of the machine could one day be consumed by those queues. C. SLAB meta data overhead SLAB has overhead at the beginning of each slab. This means that data cannot be naturally aligned at the beginning of a slab block. SLUB keeps all meta data in the corresponding page_struct. Objects can be naturally aligned in the slab. F.e. a 128 byte object will be aligned at 128 byte boundaries and can fit tightly into a 4k page with no bytes left over. SLAB cannot do this. D. SLAB has a complex cache reaper SLUB does not need a cache reaper for UP systems. On SMP systems the per CPU slab may be pushed back into partial list but that operation is simple and does not require an iteration over a list of objects. SLAB expires per CPU, shared and alien object queues during cache reaping which may cause strange hold offs. E. SLAB has complex NUMA policy layer support SLUB pushes NUMA policy handling into the page allocator. This means that allocation is coarser (SLUB does interleave on a page level) but that situation was also present before 2.6.13. SLABs application of policies to individual slab objects allocated in SLAB is certainly a performance concern due to the frequent references to memory policies which may lead a sequence of objects to come from one node after another. SLUB will get a slab full of objects from one node and then will switch to the next. F. Reduction of the size of partial slab lists SLAB has per node partial lists. This means that over time a large number of partial slabs may accumulate on those lists. These can only be reused if allocator occur on specific nodes. SLUB has a global pool of partial slabs and will consume slabs from that pool to decrease fragmentation. G. Tunables SLAB has sophisticated tuning abilities for each slab cache. One can manipulate the queue sizes in detail. However, filling the queues still requires the uses of the spin lock to check out slabs. SLUB has a global parameter (min_slab_order) for tuning. Increasing the minimum slab order can decrease the locking overhead. The bigger the slab order the less motions of pages between per CPU and partial lists occur and the better SLUB will be scaling. G. Slab merging We often have slab caches with similar parameters. SLUB detects those on boot up and merges them into the corresponding general caches. This leads to more effective memory use. About 50% of all caches can be eliminated through slab merging. This will also decrease slab fragmentation because partial allocated slabs can be filled up again. Slab merging can be switched off by specifying slub_nomerge on boot up. Note that merging can expose heretofore unknown bugs in the kernel because corrupted objects may now be placed differently and corrupt differing neighboring objects. Enable sanity checks to find those. H. Diagnostics The current slab diagnostics are difficult to use and require a recompilation of the kernel. SLUB contains debugging code that is always available (but is kept out of the hot code paths). SLUB diagnostics can be enabled via the "slab_debug" option. Parameters can be specified to select a single or a group of slab caches for diagnostics. This means that the system is running with the usual performance and it is much more likely that race conditions can be reproduced. I. Resiliency If basic sanity checks are on then SLUB is capable of detecting common error conditions and recover as best as possible to allow the system to continue. J. Tracing Tracing can be enabled via the slab_debug=T,<slabcache> option during boot. SLUB will then protocol all actions on that slabcache and dump the object contents on free. K. On demand DMA cache creation. Generally DMA caches are not needed. If a kmalloc is used with __GFP_DMA then just create this single slabcache that is needed. For systems that have no ZONE_DMA requirement the support is completely eliminated. L. Performance increase Some benchmarks have shown speed improvements on kernbench in the range of 5-10%. The locking overhead of slub is based on the underlying base allocation size. If we can reliably allocate larger order pages then it is possible to increase slub performance much further. The anti-fragmentation patches may enable further performance increases. Tested on: i386 UP + SMP, x86_64 UP + SMP + NUMA emulation, IA64 NUMA + Simulator SLUB Boot options slub_nomerge Disable merging of slabs slub_min_order=x Require a minimum order for slab caches. This increases the managed chunk size and therefore reduces meta data and locking overhead. slub_min_objects=x Mininum objects per slab. Default is 8. slub_max_order=x Avoid generating slabs larger than order specified. slub_debug Enable all diagnostics for all caches slub_debug=<options> Enable selective options for all caches slub_debug=<o>,<cache> Enable selective options for a certain set of caches Available Debug options F Double Free checking, sanity and resiliency R Red zoning P Object / padding poisoning U Track last free / alloc T Trace all allocs / frees (only use for individual slabs). To use SLUB: Apply this patch and then select SLUB as the default slab allocator. [hugh@veritas.com: fix an oops-causing locking error] [akpm@linux-foundation.org: various stupid cleanups and small fixes] Signed-off-by: Christoph Lameter <clameter@sgi.com> Signed-off-by: Hugh Dickins <hugh@veritas.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2007-05-06 14:49:36 -07:00
return NULL;
/* Allocation too large? */
BUG_ON(index < 0);
#ifdef CONFIG_ZONE_DMA
if ((flags & SLUB_DMA)) {
struct kmem_cache *s;
struct kmem_cache *x;
char *text;
size_t realsize;
s = kmalloc_caches_dma[index];
if (s)
return s;
/* Dynamically create dma cache */
x = kmalloc(kmem_size, flags & ~SLUB_DMA);
if (!x)
panic("Unable to allocate memory for dma cache\n");
if (index <= KMALLOC_SHIFT_HIGH)
realsize = 1 << index;
else {
if (index == 1)
realsize = 96;
else
realsize = 192;
}
text = kasprintf(flags & ~SLUB_DMA, "kmalloc_dma-%d",
(unsigned int)realsize);
s = create_kmalloc_cache(x, text, realsize, flags);
kmalloc_caches_dma[index] = s;
return s;
}
#endif
return &kmalloc_caches[index];
}
void *__kmalloc(size_t size, gfp_t flags)
{
struct kmem_cache *s = get_slab(size, flags);
if (s)
slub: fix object tracking Object tracking did not work the right way for several call chains. Fix this up by adding a new parameter to slub_alloc and slub_free that specifies the caller address explicitly. Signed-off-by: Christoph Lameter <clameter@sgi.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2007-05-06 14:49:42 -07:00
return slab_alloc(s, flags, -1, __builtin_return_address(0));
SLUB core This is a new slab allocator which was motivated by the complexity of the existing code in mm/slab.c. It attempts to address a variety of concerns with the existing implementation. A. Management of object queues A particular concern was the complex management of the numerous object queues in SLAB. SLUB has no such queues. Instead we dedicate a slab for each allocating CPU and use objects from a slab directly instead of queueing them up. B. Storage overhead of object queues SLAB Object queues exist per node, per CPU. The alien cache queue even has a queue array that contain a queue for each processor on each node. For very large systems the number of queues and the number of objects that may be caught in those queues grows exponentially. On our systems with 1k nodes / processors we have several gigabytes just tied up for storing references to objects for those queues This does not include the objects that could be on those queues. One fears that the whole memory of the machine could one day be consumed by those queues. C. SLAB meta data overhead SLAB has overhead at the beginning of each slab. This means that data cannot be naturally aligned at the beginning of a slab block. SLUB keeps all meta data in the corresponding page_struct. Objects can be naturally aligned in the slab. F.e. a 128 byte object will be aligned at 128 byte boundaries and can fit tightly into a 4k page with no bytes left over. SLAB cannot do this. D. SLAB has a complex cache reaper SLUB does not need a cache reaper for UP systems. On SMP systems the per CPU slab may be pushed back into partial list but that operation is simple and does not require an iteration over a list of objects. SLAB expires per CPU, shared and alien object queues during cache reaping which may cause strange hold offs. E. SLAB has complex NUMA policy layer support SLUB pushes NUMA policy handling into the page allocator. This means that allocation is coarser (SLUB does interleave on a page level) but that situation was also present before 2.6.13. SLABs application of policies to individual slab objects allocated in SLAB is certainly a performance concern due to the frequent references to memory policies which may lead a sequence of objects to come from one node after another. SLUB will get a slab full of objects from one node and then will switch to the next. F. Reduction of the size of partial slab lists SLAB has per node partial lists. This means that over time a large number of partial slabs may accumulate on those lists. These can only be reused if allocator occur on specific nodes. SLUB has a global pool of partial slabs and will consume slabs from that pool to decrease fragmentation. G. Tunables SLAB has sophisticated tuning abilities for each slab cache. One can manipulate the queue sizes in detail. However, filling the queues still requires the uses of the spin lock to check out slabs. SLUB has a global parameter (min_slab_order) for tuning. Increasing the minimum slab order can decrease the locking overhead. The bigger the slab order the less motions of pages between per CPU and partial lists occur and the better SLUB will be scaling. G. Slab merging We often have slab caches with similar parameters. SLUB detects those on boot up and merges them into the corresponding general caches. This leads to more effective memory use. About 50% of all caches can be eliminated through slab merging. This will also decrease slab fragmentation because partial allocated slabs can be filled up again. Slab merging can be switched off by specifying slub_nomerge on boot up. Note that merging can expose heretofore unknown bugs in the kernel because corrupted objects may now be placed differently and corrupt differing neighboring objects. Enable sanity checks to find those. H. Diagnostics The current slab diagnostics are difficult to use and require a recompilation of the kernel. SLUB contains debugging code that is always available (but is kept out of the hot code paths). SLUB diagnostics can be enabled via the "slab_debug" option. Parameters can be specified to select a single or a group of slab caches for diagnostics. This means that the system is running with the usual performance and it is much more likely that race conditions can be reproduced. I. Resiliency If basic sanity checks are on then SLUB is capable of detecting common error conditions and recover as best as possible to allow the system to continue. J. Tracing Tracing can be enabled via the slab_debug=T,<slabcache> option during boot. SLUB will then protocol all actions on that slabcache and dump the object contents on free. K. On demand DMA cache creation. Generally DMA caches are not needed. If a kmalloc is used with __GFP_DMA then just create this single slabcache that is needed. For systems that have no ZONE_DMA requirement the support is completely eliminated. L. Performance increase Some benchmarks have shown speed improvements on kernbench in the range of 5-10%. The locking overhead of slub is based on the underlying base allocation size. If we can reliably allocate larger order pages then it is possible to increase slub performance much further. The anti-fragmentation patches may enable further performance increases. Tested on: i386 UP + SMP, x86_64 UP + SMP + NUMA emulation, IA64 NUMA + Simulator SLUB Boot options slub_nomerge Disable merging of slabs slub_min_order=x Require a minimum order for slab caches. This increases the managed chunk size and therefore reduces meta data and locking overhead. slub_min_objects=x Mininum objects per slab. Default is 8. slub_max_order=x Avoid generating slabs larger than order specified. slub_debug Enable all diagnostics for all caches slub_debug=<options> Enable selective options for all caches slub_debug=<o>,<cache> Enable selective options for a certain set of caches Available Debug options F Double Free checking, sanity and resiliency R Red zoning P Object / padding poisoning U Track last free / alloc T Trace all allocs / frees (only use for individual slabs). To use SLUB: Apply this patch and then select SLUB as the default slab allocator. [hugh@veritas.com: fix an oops-causing locking error] [akpm@linux-foundation.org: various stupid cleanups and small fixes] Signed-off-by: Christoph Lameter <clameter@sgi.com> Signed-off-by: Hugh Dickins <hugh@veritas.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2007-05-06 14:49:36 -07:00
return NULL;
}
EXPORT_SYMBOL(__kmalloc);
#ifdef CONFIG_NUMA
void *__kmalloc_node(size_t size, gfp_t flags, int node)
{
struct kmem_cache *s = get_slab(size, flags);
if (s)
slub: fix object tracking Object tracking did not work the right way for several call chains. Fix this up by adding a new parameter to slub_alloc and slub_free that specifies the caller address explicitly. Signed-off-by: Christoph Lameter <clameter@sgi.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2007-05-06 14:49:42 -07:00
return slab_alloc(s, flags, node, __builtin_return_address(0));
SLUB core This is a new slab allocator which was motivated by the complexity of the existing code in mm/slab.c. It attempts to address a variety of concerns with the existing implementation. A. Management of object queues A particular concern was the complex management of the numerous object queues in SLAB. SLUB has no such queues. Instead we dedicate a slab for each allocating CPU and use objects from a slab directly instead of queueing them up. B. Storage overhead of object queues SLAB Object queues exist per node, per CPU. The alien cache queue even has a queue array that contain a queue for each processor on each node. For very large systems the number of queues and the number of objects that may be caught in those queues grows exponentially. On our systems with 1k nodes / processors we have several gigabytes just tied up for storing references to objects for those queues This does not include the objects that could be on those queues. One fears that the whole memory of the machine could one day be consumed by those queues. C. SLAB meta data overhead SLAB has overhead at the beginning of each slab. This means that data cannot be naturally aligned at the beginning of a slab block. SLUB keeps all meta data in the corresponding page_struct. Objects can be naturally aligned in the slab. F.e. a 128 byte object will be aligned at 128 byte boundaries and can fit tightly into a 4k page with no bytes left over. SLAB cannot do this. D. SLAB has a complex cache reaper SLUB does not need a cache reaper for UP systems. On SMP systems the per CPU slab may be pushed back into partial list but that operation is simple and does not require an iteration over a list of objects. SLAB expires per CPU, shared and alien object queues during cache reaping which may cause strange hold offs. E. SLAB has complex NUMA policy layer support SLUB pushes NUMA policy handling into the page allocator. This means that allocation is coarser (SLUB does interleave on a page level) but that situation was also present before 2.6.13. SLABs application of policies to individual slab objects allocated in SLAB is certainly a performance concern due to the frequent references to memory policies which may lead a sequence of objects to come from one node after another. SLUB will get a slab full of objects from one node and then will switch to the next. F. Reduction of the size of partial slab lists SLAB has per node partial lists. This means that over time a large number of partial slabs may accumulate on those lists. These can only be reused if allocator occur on specific nodes. SLUB has a global pool of partial slabs and will consume slabs from that pool to decrease fragmentation. G. Tunables SLAB has sophisticated tuning abilities for each slab cache. One can manipulate the queue sizes in detail. However, filling the queues still requires the uses of the spin lock to check out slabs. SLUB has a global parameter (min_slab_order) for tuning. Increasing the minimum slab order can decrease the locking overhead. The bigger the slab order the less motions of pages between per CPU and partial lists occur and the better SLUB will be scaling. G. Slab merging We often have slab caches with similar parameters. SLUB detects those on boot up and merges them into the corresponding general caches. This leads to more effective memory use. About 50% of all caches can be eliminated through slab merging. This will also decrease slab fragmentation because partial allocated slabs can be filled up again. Slab merging can be switched off by specifying slub_nomerge on boot up. Note that merging can expose heretofore unknown bugs in the kernel because corrupted objects may now be placed differently and corrupt differing neighboring objects. Enable sanity checks to find those. H. Diagnostics The current slab diagnostics are difficult to use and require a recompilation of the kernel. SLUB contains debugging code that is always available (but is kept out of the hot code paths). SLUB diagnostics can be enabled via the "slab_debug" option. Parameters can be specified to select a single or a group of slab caches for diagnostics. This means that the system is running with the usual performance and it is much more likely that race conditions can be reproduced. I. Resiliency If basic sanity checks are on then SLUB is capable of detecting common error conditions and recover as best as possible to allow the system to continue. J. Tracing Tracing can be enabled via the slab_debug=T,<slabcache> option during boot. SLUB will then protocol all actions on that slabcache and dump the object contents on free. K. On demand DMA cache creation. Generally DMA caches are not needed. If a kmalloc is used with __GFP_DMA then just create this single slabcache that is needed. For systems that have no ZONE_DMA requirement the support is completely eliminated. L. Performance increase Some benchmarks have shown speed improvements on kernbench in the range of 5-10%. The locking overhead of slub is based on the underlying base allocation size. If we can reliably allocate larger order pages then it is possible to increase slub performance much further. The anti-fragmentation patches may enable further performance increases. Tested on: i386 UP + SMP, x86_64 UP + SMP + NUMA emulation, IA64 NUMA + Simulator SLUB Boot options slub_nomerge Disable merging of slabs slub_min_order=x Require a minimum order for slab caches. This increases the managed chunk size and therefore reduces meta data and locking overhead. slub_min_objects=x Mininum objects per slab. Default is 8. slub_max_order=x Avoid generating slabs larger than order specified. slub_debug Enable all diagnostics for all caches slub_debug=<options> Enable selective options for all caches slub_debug=<o>,<cache> Enable selective options for a certain set of caches Available Debug options F Double Free checking, sanity and resiliency R Red zoning P Object / padding poisoning U Track last free / alloc T Trace all allocs / frees (only use for individual slabs). To use SLUB: Apply this patch and then select SLUB as the default slab allocator. [hugh@veritas.com: fix an oops-causing locking error] [akpm@linux-foundation.org: various stupid cleanups and small fixes] Signed-off-by: Christoph Lameter <clameter@sgi.com> Signed-off-by: Hugh Dickins <hugh@veritas.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2007-05-06 14:49:36 -07:00
return NULL;
}
EXPORT_SYMBOL(__kmalloc_node);
#endif
size_t ksize(const void *object)
{
struct page *page = get_object_page(object);
struct kmem_cache *s;
BUG_ON(!page);
s = page->slab;
BUG_ON(!s);
/*
* Debugging requires use of the padding between object
* and whatever may come after it.
*/
if (s->flags & (SLAB_RED_ZONE | SLAB_POISON))
return s->objsize;
/*
* If we have the need to store the freelist pointer
* back there or track user information then we can
* only use the space before that information.
*/
if (s->flags & (SLAB_DESTROY_BY_RCU | SLAB_STORE_USER))
return s->inuse;
/*
* Else we can use all the padding etc for the allocation
*/
return s->size;
}
EXPORT_SYMBOL(ksize);
void kfree(const void *x)
{
struct kmem_cache *s;
struct page *page;
if (!x)
return;
Add virt_to_head_page and consolidate code in slab and slub Signed-off-by: Christoph Lameter <clameter@sgi.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2007-05-06 14:49:41 -07:00
page = virt_to_head_page(x);
SLUB core This is a new slab allocator which was motivated by the complexity of the existing code in mm/slab.c. It attempts to address a variety of concerns with the existing implementation. A. Management of object queues A particular concern was the complex management of the numerous object queues in SLAB. SLUB has no such queues. Instead we dedicate a slab for each allocating CPU and use objects from a slab directly instead of queueing them up. B. Storage overhead of object queues SLAB Object queues exist per node, per CPU. The alien cache queue even has a queue array that contain a queue for each processor on each node. For very large systems the number of queues and the number of objects that may be caught in those queues grows exponentially. On our systems with 1k nodes / processors we have several gigabytes just tied up for storing references to objects for those queues This does not include the objects that could be on those queues. One fears that the whole memory of the machine could one day be consumed by those queues. C. SLAB meta data overhead SLAB has overhead at the beginning of each slab. This means that data cannot be naturally aligned at the beginning of a slab block. SLUB keeps all meta data in the corresponding page_struct. Objects can be naturally aligned in the slab. F.e. a 128 byte object will be aligned at 128 byte boundaries and can fit tightly into a 4k page with no bytes left over. SLAB cannot do this. D. SLAB has a complex cache reaper SLUB does not need a cache reaper for UP systems. On SMP systems the per CPU slab may be pushed back into partial list but that operation is simple and does not require an iteration over a list of objects. SLAB expires per CPU, shared and alien object queues during cache reaping which may cause strange hold offs. E. SLAB has complex NUMA policy layer support SLUB pushes NUMA policy handling into the page allocator. This means that allocation is coarser (SLUB does interleave on a page level) but that situation was also present before 2.6.13. SLABs application of policies to individual slab objects allocated in SLAB is certainly a performance concern due to the frequent references to memory policies which may lead a sequence of objects to come from one node after another. SLUB will get a slab full of objects from one node and then will switch to the next. F. Reduction of the size of partial slab lists SLAB has per node partial lists. This means that over time a large number of partial slabs may accumulate on those lists. These can only be reused if allocator occur on specific nodes. SLUB has a global pool of partial slabs and will consume slabs from that pool to decrease fragmentation. G. Tunables SLAB has sophisticated tuning abilities for each slab cache. One can manipulate the queue sizes in detail. However, filling the queues still requires the uses of the spin lock to check out slabs. SLUB has a global parameter (min_slab_order) for tuning. Increasing the minimum slab order can decrease the locking overhead. The bigger the slab order the less motions of pages between per CPU and partial lists occur and the better SLUB will be scaling. G. Slab merging We often have slab caches with similar parameters. SLUB detects those on boot up and merges them into the corresponding general caches. This leads to more effective memory use. About 50% of all caches can be eliminated through slab merging. This will also decrease slab fragmentation because partial allocated slabs can be filled up again. Slab merging can be switched off by specifying slub_nomerge on boot up. Note that merging can expose heretofore unknown bugs in the kernel because corrupted objects may now be placed differently and corrupt differing neighboring objects. Enable sanity checks to find those. H. Diagnostics The current slab diagnostics are difficult to use and require a recompilation of the kernel. SLUB contains debugging code that is always available (but is kept out of the hot code paths). SLUB diagnostics can be enabled via the "slab_debug" option. Parameters can be specified to select a single or a group of slab caches for diagnostics. This means that the system is running with the usual performance and it is much more likely that race conditions can be reproduced. I. Resiliency If basic sanity checks are on then SLUB is capable of detecting common error conditions and recover as best as possible to allow the system to continue. J. Tracing Tracing can be enabled via the slab_debug=T,<slabcache> option during boot. SLUB will then protocol all actions on that slabcache and dump the object contents on free. K. On demand DMA cache creation. Generally DMA caches are not needed. If a kmalloc is used with __GFP_DMA then just create this single slabcache that is needed. For systems that have no ZONE_DMA requirement the support is completely eliminated. L. Performance increase Some benchmarks have shown speed improvements on kernbench in the range of 5-10%. The locking overhead of slub is based on the underlying base allocation size. If we can reliably allocate larger order pages then it is possible to increase slub performance much further. The anti-fragmentation patches may enable further performance increases. Tested on: i386 UP + SMP, x86_64 UP + SMP + NUMA emulation, IA64 NUMA + Simulator SLUB Boot options slub_nomerge Disable merging of slabs slub_min_order=x Require a minimum order for slab caches. This increases the managed chunk size and therefore reduces meta data and locking overhead. slub_min_objects=x Mininum objects per slab. Default is 8. slub_max_order=x Avoid generating slabs larger than order specified. slub_debug Enable all diagnostics for all caches slub_debug=<options> Enable selective options for all caches slub_debug=<o>,<cache> Enable selective options for a certain set of caches Available Debug options F Double Free checking, sanity and resiliency R Red zoning P Object / padding poisoning U Track last free / alloc T Trace all allocs / frees (only use for individual slabs). To use SLUB: Apply this patch and then select SLUB as the default slab allocator. [hugh@veritas.com: fix an oops-causing locking error] [akpm@linux-foundation.org: various stupid cleanups and small fixes] Signed-off-by: Christoph Lameter <clameter@sgi.com> Signed-off-by: Hugh Dickins <hugh@veritas.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2007-05-06 14:49:36 -07:00
s = page->slab;
slub: fix object tracking Object tracking did not work the right way for several call chains. Fix this up by adding a new parameter to slub_alloc and slub_free that specifies the caller address explicitly. Signed-off-by: Christoph Lameter <clameter@sgi.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2007-05-06 14:49:42 -07:00
slab_free(s, page, (void *)x, __builtin_return_address(0));
SLUB core This is a new slab allocator which was motivated by the complexity of the existing code in mm/slab.c. It attempts to address a variety of concerns with the existing implementation. A. Management of object queues A particular concern was the complex management of the numerous object queues in SLAB. SLUB has no such queues. Instead we dedicate a slab for each allocating CPU and use objects from a slab directly instead of queueing them up. B. Storage overhead of object queues SLAB Object queues exist per node, per CPU. The alien cache queue even has a queue array that contain a queue for each processor on each node. For very large systems the number of queues and the number of objects that may be caught in those queues grows exponentially. On our systems with 1k nodes / processors we have several gigabytes just tied up for storing references to objects for those queues This does not include the objects that could be on those queues. One fears that the whole memory of the machine could one day be consumed by those queues. C. SLAB meta data overhead SLAB has overhead at the beginning of each slab. This means that data cannot be naturally aligned at the beginning of a slab block. SLUB keeps all meta data in the corresponding page_struct. Objects can be naturally aligned in the slab. F.e. a 128 byte object will be aligned at 128 byte boundaries and can fit tightly into a 4k page with no bytes left over. SLAB cannot do this. D. SLAB has a complex cache reaper SLUB does not need a cache reaper for UP systems. On SMP systems the per CPU slab may be pushed back into partial list but that operation is simple and does not require an iteration over a list of objects. SLAB expires per CPU, shared and alien object queues during cache reaping which may cause strange hold offs. E. SLAB has complex NUMA policy layer support SLUB pushes NUMA policy handling into the page allocator. This means that allocation is coarser (SLUB does interleave on a page level) but that situation was also present before 2.6.13. SLABs application of policies to individual slab objects allocated in SLAB is certainly a performance concern due to the frequent references to memory policies which may lead a sequence of objects to come from one node after another. SLUB will get a slab full of objects from one node and then will switch to the next. F. Reduction of the size of partial slab lists SLAB has per node partial lists. This means that over time a large number of partial slabs may accumulate on those lists. These can only be reused if allocator occur on specific nodes. SLUB has a global pool of partial slabs and will consume slabs from that pool to decrease fragmentation. G. Tunables SLAB has sophisticated tuning abilities for each slab cache. One can manipulate the queue sizes in detail. However, filling the queues still requires the uses of the spin lock to check out slabs. SLUB has a global parameter (min_slab_order) for tuning. Increasing the minimum slab order can decrease the locking overhead. The bigger the slab order the less motions of pages between per CPU and partial lists occur and the better SLUB will be scaling. G. Slab merging We often have slab caches with similar parameters. SLUB detects those on boot up and merges them into the corresponding general caches. This leads to more effective memory use. About 50% of all caches can be eliminated through slab merging. This will also decrease slab fragmentation because partial allocated slabs can be filled up again. Slab merging can be switched off by specifying slub_nomerge on boot up. Note that merging can expose heretofore unknown bugs in the kernel because corrupted objects may now be placed differently and corrupt differing neighboring objects. Enable sanity checks to find those. H. Diagnostics The current slab diagnostics are difficult to use and require a recompilation of the kernel. SLUB contains debugging code that is always available (but is kept out of the hot code paths). SLUB diagnostics can be enabled via the "slab_debug" option. Parameters can be specified to select a single or a group of slab caches for diagnostics. This means that the system is running with the usual performance and it is much more likely that race conditions can be reproduced. I. Resiliency If basic sanity checks are on then SLUB is capable of detecting common error conditions and recover as best as possible to allow the system to continue. J. Tracing Tracing can be enabled via the slab_debug=T,<slabcache> option during boot. SLUB will then protocol all actions on that slabcache and dump the object contents on free. K. On demand DMA cache creation. Generally DMA caches are not needed. If a kmalloc is used with __GFP_DMA then just create this single slabcache that is needed. For systems that have no ZONE_DMA requirement the support is completely eliminated. L. Performance increase Some benchmarks have shown speed improvements on kernbench in the range of 5-10%. The locking overhead of slub is based on the underlying base allocation size. If we can reliably allocate larger order pages then it is possible to increase slub performance much further. The anti-fragmentation patches may enable further performance increases. Tested on: i386 UP + SMP, x86_64 UP + SMP + NUMA emulation, IA64 NUMA + Simulator SLUB Boot options slub_nomerge Disable merging of slabs slub_min_order=x Require a minimum order for slab caches. This increases the managed chunk size and therefore reduces meta data and locking overhead. slub_min_objects=x Mininum objects per slab. Default is 8. slub_max_order=x Avoid generating slabs larger than order specified. slub_debug Enable all diagnostics for all caches slub_debug=<options> Enable selective options for all caches slub_debug=<o>,<cache> Enable selective options for a certain set of caches Available Debug options F Double Free checking, sanity and resiliency R Red zoning P Object / padding poisoning U Track last free / alloc T Trace all allocs / frees (only use for individual slabs). To use SLUB: Apply this patch and then select SLUB as the default slab allocator. [hugh@veritas.com: fix an oops-causing locking error] [akpm@linux-foundation.org: various stupid cleanups and small fixes] Signed-off-by: Christoph Lameter <clameter@sgi.com> Signed-off-by: Hugh Dickins <hugh@veritas.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2007-05-06 14:49:36 -07:00
}
EXPORT_SYMBOL(kfree);
SLUB: Free slabs and sort partial slab lists in kmem_cache_shrink At kmem_cache_shrink check if we have any empty slabs on the partial if so then remove them. Also--as an anti-fragmentation measure--sort the partial slabs so that the most fully allocated ones come first and the least allocated last. The next allocations may fill up the nearly full slabs. Having the least allocated slabs last gives them the maximum chance that their remaining objects may be freed. Thus we can hopefully minimize the partial slabs. I think this is the best one can do in terms antifragmentation measures. Real defragmentation (meaning moving objects out of slabs with the least free objects to those that are almost full) can be implemted by reverse scanning through the list produced here but that would mean that we need to provide a callback at slab cache creation that allows the deletion or moving of an object. This will involve slab API changes, so defer for now. Cc: Mel Gorman <mel@skynet.ie> Signed-off-by: Christoph Lameter <clameter@sgi.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2007-05-06 14:49:46 -07:00
/*
* kmem_cache_shrink removes empty slabs from the partial lists
* and then sorts the partially allocated slabs by the number
* of items in use. The slabs with the most items in use
* come first. New allocations will remove these from the
* partial list because they are full. The slabs with the
* least items are placed last. If it happens that the objects
* are freed then the page can be returned to the page allocator.
*/
int kmem_cache_shrink(struct kmem_cache *s)
{
int node;
int i;
struct kmem_cache_node *n;
struct page *page;
struct page *t;
struct list_head *slabs_by_inuse =
kmalloc(sizeof(struct list_head) * s->objects, GFP_KERNEL);
unsigned long flags;
if (!slabs_by_inuse)
return -ENOMEM;
flush_all(s);
for_each_online_node(node) {
n = get_node(s, node);
if (!n->nr_partial)
continue;
for (i = 0; i < s->objects; i++)
INIT_LIST_HEAD(slabs_by_inuse + i);
spin_lock_irqsave(&n->list_lock, flags);
/*
* Build lists indexed by the items in use in
* each slab or free slabs if empty.
*
* Note that concurrent frees may occur while
* we hold the list_lock. page->inuse here is
* the upper limit.
*/
list_for_each_entry_safe(page, t, &n->partial, lru) {
if (!page->inuse && slab_trylock(page)) {
/*
* Must hold slab lock here because slab_free
* may have freed the last object and be
* waiting to release the slab.
*/
list_del(&page->lru);
n->nr_partial--;
slab_unlock(page);
discard_slab(s, page);
} else {
if (n->nr_partial > MAX_PARTIAL)
list_move(&page->lru,
slabs_by_inuse + page->inuse);
}
}
if (n->nr_partial <= MAX_PARTIAL)
goto out;
/*
* Rebuild the partial list with the slabs filled up
* most first and the least used slabs at the end.
*/
for (i = s->objects - 1; i >= 0; i--)
list_splice(slabs_by_inuse + i, n->partial.prev);
out:
spin_unlock_irqrestore(&n->list_lock, flags);
}
kfree(slabs_by_inuse);
return 0;
}
EXPORT_SYMBOL(kmem_cache_shrink);
SLUB core This is a new slab allocator which was motivated by the complexity of the existing code in mm/slab.c. It attempts to address a variety of concerns with the existing implementation. A. Management of object queues A particular concern was the complex management of the numerous object queues in SLAB. SLUB has no such queues. Instead we dedicate a slab for each allocating CPU and use objects from a slab directly instead of queueing them up. B. Storage overhead of object queues SLAB Object queues exist per node, per CPU. The alien cache queue even has a queue array that contain a queue for each processor on each node. For very large systems the number of queues and the number of objects that may be caught in those queues grows exponentially. On our systems with 1k nodes / processors we have several gigabytes just tied up for storing references to objects for those queues This does not include the objects that could be on those queues. One fears that the whole memory of the machine could one day be consumed by those queues. C. SLAB meta data overhead SLAB has overhead at the beginning of each slab. This means that data cannot be naturally aligned at the beginning of a slab block. SLUB keeps all meta data in the corresponding page_struct. Objects can be naturally aligned in the slab. F.e. a 128 byte object will be aligned at 128 byte boundaries and can fit tightly into a 4k page with no bytes left over. SLAB cannot do this. D. SLAB has a complex cache reaper SLUB does not need a cache reaper for UP systems. On SMP systems the per CPU slab may be pushed back into partial list but that operation is simple and does not require an iteration over a list of objects. SLAB expires per CPU, shared and alien object queues during cache reaping which may cause strange hold offs. E. SLAB has complex NUMA policy layer support SLUB pushes NUMA policy handling into the page allocator. This means that allocation is coarser (SLUB does interleave on a page level) but that situation was also present before 2.6.13. SLABs application of policies to individual slab objects allocated in SLAB is certainly a performance concern due to the frequent references to memory policies which may lead a sequence of objects to come from one node after another. SLUB will get a slab full of objects from one node and then will switch to the next. F. Reduction of the size of partial slab lists SLAB has per node partial lists. This means that over time a large number of partial slabs may accumulate on those lists. These can only be reused if allocator occur on specific nodes. SLUB has a global pool of partial slabs and will consume slabs from that pool to decrease fragmentation. G. Tunables SLAB has sophisticated tuning abilities for each slab cache. One can manipulate the queue sizes in detail. However, filling the queues still requires the uses of the spin lock to check out slabs. SLUB has a global parameter (min_slab_order) for tuning. Increasing the minimum slab order can decrease the locking overhead. The bigger the slab order the less motions of pages between per CPU and partial lists occur and the better SLUB will be scaling. G. Slab merging We often have slab caches with similar parameters. SLUB detects those on boot up and merges them into the corresponding general caches. This leads to more effective memory use. About 50% of all caches can be eliminated through slab merging. This will also decrease slab fragmentation because partial allocated slabs can be filled up again. Slab merging can be switched off by specifying slub_nomerge on boot up. Note that merging can expose heretofore unknown bugs in the kernel because corrupted objects may now be placed differently and corrupt differing neighboring objects. Enable sanity checks to find those. H. Diagnostics The current slab diagnostics are difficult to use and require a recompilation of the kernel. SLUB contains debugging code that is always available (but is kept out of the hot code paths). SLUB diagnostics can be enabled via the "slab_debug" option. Parameters can be specified to select a single or a group of slab caches for diagnostics. This means that the system is running with the usual performance and it is much more likely that race conditions can be reproduced. I. Resiliency If basic sanity checks are on then SLUB is capable of detecting common error conditions and recover as best as possible to allow the system to continue. J. Tracing Tracing can be enabled via the slab_debug=T,<slabcache> option during boot. SLUB will then protocol all actions on that slabcache and dump the object contents on free. K. On demand DMA cache creation. Generally DMA caches are not needed. If a kmalloc is used with __GFP_DMA then just create this single slabcache that is needed. For systems that have no ZONE_DMA requirement the support is completely eliminated. L. Performance increase Some benchmarks have shown speed improvements on kernbench in the range of 5-10%. The locking overhead of slub is based on the underlying base allocation size. If we can reliably allocate larger order pages then it is possible to increase slub performance much further. The anti-fragmentation patches may enable further performance increases. Tested on: i386 UP + SMP, x86_64 UP + SMP + NUMA emulation, IA64 NUMA + Simulator SLUB Boot options slub_nomerge Disable merging of slabs slub_min_order=x Require a minimum order for slab caches. This increases the managed chunk size and therefore reduces meta data and locking overhead. slub_min_objects=x Mininum objects per slab. Default is 8. slub_max_order=x Avoid generating slabs larger than order specified. slub_debug Enable all diagnostics for all caches slub_debug=<options> Enable selective options for all caches slub_debug=<o>,<cache> Enable selective options for a certain set of caches Available Debug options F Double Free checking, sanity and resiliency R Red zoning P Object / padding poisoning U Track last free / alloc T Trace all allocs / frees (only use for individual slabs). To use SLUB: Apply this patch and then select SLUB as the default slab allocator. [hugh@veritas.com: fix an oops-causing locking error] [akpm@linux-foundation.org: various stupid cleanups and small fixes] Signed-off-by: Christoph Lameter <clameter@sgi.com> Signed-off-by: Hugh Dickins <hugh@veritas.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2007-05-06 14:49:36 -07:00
/**
* krealloc - reallocate memory. The contents will remain unchanged.
*
* @p: object to reallocate memory for.
* @new_size: how many bytes of memory are required.
* @flags: the type of memory to allocate.
*
* The contents of the object pointed to are preserved up to the
* lesser of the new and old sizes. If @p is %NULL, krealloc()
* behaves exactly like kmalloc(). If @size is 0 and @p is not a
* %NULL pointer, the object pointed to is freed.
*/
void *krealloc(const void *p, size_t new_size, gfp_t flags)
{
struct kmem_cache *new_cache;
void *ret;
struct page *page;
if (unlikely(!p))
return kmalloc(new_size, flags);
if (unlikely(!new_size)) {
kfree(p);
return NULL;
}
Add virt_to_head_page and consolidate code in slab and slub Signed-off-by: Christoph Lameter <clameter@sgi.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2007-05-06 14:49:41 -07:00
page = virt_to_head_page(p);
SLUB core This is a new slab allocator which was motivated by the complexity of the existing code in mm/slab.c. It attempts to address a variety of concerns with the existing implementation. A. Management of object queues A particular concern was the complex management of the numerous object queues in SLAB. SLUB has no such queues. Instead we dedicate a slab for each allocating CPU and use objects from a slab directly instead of queueing them up. B. Storage overhead of object queues SLAB Object queues exist per node, per CPU. The alien cache queue even has a queue array that contain a queue for each processor on each node. For very large systems the number of queues and the number of objects that may be caught in those queues grows exponentially. On our systems with 1k nodes / processors we have several gigabytes just tied up for storing references to objects for those queues This does not include the objects that could be on those queues. One fears that the whole memory of the machine could one day be consumed by those queues. C. SLAB meta data overhead SLAB has overhead at the beginning of each slab. This means that data cannot be naturally aligned at the beginning of a slab block. SLUB keeps all meta data in the corresponding page_struct. Objects can be naturally aligned in the slab. F.e. a 128 byte object will be aligned at 128 byte boundaries and can fit tightly into a 4k page with no bytes left over. SLAB cannot do this. D. SLAB has a complex cache reaper SLUB does not need a cache reaper for UP systems. On SMP systems the per CPU slab may be pushed back into partial list but that operation is simple and does not require an iteration over a list of objects. SLAB expires per CPU, shared and alien object queues during cache reaping which may cause strange hold offs. E. SLAB has complex NUMA policy layer support SLUB pushes NUMA policy handling into the page allocator. This means that allocation is coarser (SLUB does interleave on a page level) but that situation was also present before 2.6.13. SLABs application of policies to individual slab objects allocated in SLAB is certainly a performance concern due to the frequent references to memory policies which may lead a sequence of objects to come from one node after another. SLUB will get a slab full of objects from one node and then will switch to the next. F. Reduction of the size of partial slab lists SLAB has per node partial lists. This means that over time a large number of partial slabs may accumulate on those lists. These can only be reused if allocator occur on specific nodes. SLUB has a global pool of partial slabs and will consume slabs from that pool to decrease fragmentation. G. Tunables SLAB has sophisticated tuning abilities for each slab cache. One can manipulate the queue sizes in detail. However, filling the queues still requires the uses of the spin lock to check out slabs. SLUB has a global parameter (min_slab_order) for tuning. Increasing the minimum slab order can decrease the locking overhead. The bigger the slab order the less motions of pages between per CPU and partial lists occur and the better SLUB will be scaling. G. Slab merging We often have slab caches with similar parameters. SLUB detects those on boot up and merges them into the corresponding general caches. This leads to more effective memory use. About 50% of all caches can be eliminated through slab merging. This will also decrease slab fragmentation because partial allocated slabs can be filled up again. Slab merging can be switched off by specifying slub_nomerge on boot up. Note that merging can expose heretofore unknown bugs in the kernel because corrupted objects may now be placed differently and corrupt differing neighboring objects. Enable sanity checks to find those. H. Diagnostics The current slab diagnostics are difficult to use and require a recompilation of the kernel. SLUB contains debugging code that is always available (but is kept out of the hot code paths). SLUB diagnostics can be enabled via the "slab_debug" option. Parameters can be specified to select a single or a group of slab caches for diagnostics. This means that the system is running with the usual performance and it is much more likely that race conditions can be reproduced. I. Resiliency If basic sanity checks are on then SLUB is capable of detecting common error conditions and recover as best as possible to allow the system to continue. J. Tracing Tracing can be enabled via the slab_debug=T,<slabcache> option during boot. SLUB will then protocol all actions on that slabcache and dump the object contents on free. K. On demand DMA cache creation. Generally DMA caches are not needed. If a kmalloc is used with __GFP_DMA then just create this single slabcache that is needed. For systems that have no ZONE_DMA requirement the support is completely eliminated. L. Performance increase Some benchmarks have shown speed improvements on kernbench in the range of 5-10%. The locking overhead of slub is based on the underlying base allocation size. If we can reliably allocate larger order pages then it is possible to increase slub performance much further. The anti-fragmentation patches may enable further performance increases. Tested on: i386 UP + SMP, x86_64 UP + SMP + NUMA emulation, IA64 NUMA + Simulator SLUB Boot options slub_nomerge Disable merging of slabs slub_min_order=x Require a minimum order for slab caches. This increases the managed chunk size and therefore reduces meta data and locking overhead. slub_min_objects=x Mininum objects per slab. Default is 8. slub_max_order=x Avoid generating slabs larger than order specified. slub_debug Enable all diagnostics for all caches slub_debug=<options> Enable selective options for all caches slub_debug=<o>,<cache> Enable selective options for a certain set of caches Available Debug options F Double Free checking, sanity and resiliency R Red zoning P Object / padding poisoning U Track last free / alloc T Trace all allocs / frees (only use for individual slabs). To use SLUB: Apply this patch and then select SLUB as the default slab allocator. [hugh@veritas.com: fix an oops-causing locking error] [akpm@linux-foundation.org: various stupid cleanups and small fixes] Signed-off-by: Christoph Lameter <clameter@sgi.com> Signed-off-by: Hugh Dickins <hugh@veritas.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2007-05-06 14:49:36 -07:00
new_cache = get_slab(new_size, flags);
/*
* If new size fits in the current cache, bail out.
*/
if (likely(page->slab == new_cache))
return (void *)p;
ret = kmalloc(new_size, flags);
if (ret) {
memcpy(ret, p, min(new_size, ksize(p)));
kfree(p);
}
return ret;
}
EXPORT_SYMBOL(krealloc);
/********************************************************************
* Basic setup of slabs
*******************************************************************/
void __init kmem_cache_init(void)
{
int i;
#ifdef CONFIG_NUMA
/*
* Must first have the slab cache available for the allocations of the
* struct kmalloc_cache_node's. There is special bootstrap code in
* kmem_cache_open for slab_state == DOWN.
*/
create_kmalloc_cache(&kmalloc_caches[0], "kmem_cache_node",
sizeof(struct kmem_cache_node), GFP_KERNEL);
#endif
/* Able to allocate the per node structures */
slab_state = PARTIAL;
/* Caches that are not of the two-to-the-power-of size */
create_kmalloc_cache(&kmalloc_caches[1],
"kmalloc-96", 96, GFP_KERNEL);
create_kmalloc_cache(&kmalloc_caches[2],
"kmalloc-192", 192, GFP_KERNEL);
for (i = KMALLOC_SHIFT_LOW; i <= KMALLOC_SHIFT_HIGH; i++)
create_kmalloc_cache(&kmalloc_caches[i],
"kmalloc", 1 << i, GFP_KERNEL);
slab_state = UP;
/* Provide the correct kmalloc names now that the caches are up */
for (i = KMALLOC_SHIFT_LOW; i <= KMALLOC_SHIFT_HIGH; i++)
kmalloc_caches[i]. name =
kasprintf(GFP_KERNEL, "kmalloc-%d", 1 << i);
#ifdef CONFIG_SMP
register_cpu_notifier(&slab_notifier);
#endif
if (nr_cpu_ids) /* Remove when nr_cpu_ids is fixed upstream ! */
kmem_size = offsetof(struct kmem_cache, cpu_slab)
+ nr_cpu_ids * sizeof(struct page *);
printk(KERN_INFO "SLUB: Genslabs=%d, HWalign=%d, Order=%d-%d, MinObjects=%d,"
" Processors=%d, Nodes=%d\n",
KMALLOC_SHIFT_HIGH, L1_CACHE_BYTES,
slub_min_order, slub_max_order, slub_min_objects,
nr_cpu_ids, nr_node_ids);
}
/*
* Find a mergeable slab cache
*/
static int slab_unmergeable(struct kmem_cache *s)
{
if (slub_nomerge || (s->flags & SLUB_NEVER_MERGE))
return 1;
if (s->ctor || s->dtor)
return 1;
return 0;
}
static struct kmem_cache *find_mergeable(size_t size,
size_t align, unsigned long flags,
void (*ctor)(void *, struct kmem_cache *, unsigned long),
void (*dtor)(void *, struct kmem_cache *, unsigned long))
{
struct list_head *h;
if (slub_nomerge || (flags & SLUB_NEVER_MERGE))
return NULL;
if (ctor || dtor)
return NULL;
size = ALIGN(size, sizeof(void *));
align = calculate_alignment(flags, align, size);
size = ALIGN(size, align);
list_for_each(h, &slab_caches) {
struct kmem_cache *s =
container_of(h, struct kmem_cache, list);
if (slab_unmergeable(s))
continue;
if (size > s->size)
continue;
if (((flags | slub_debug) & SLUB_MERGE_SAME) !=
(s->flags & SLUB_MERGE_SAME))
continue;
/*
* Check if alignment is compatible.
* Courtesy of Adrian Drzewiecki
*/
if ((s->size & ~(align -1)) != s->size)
continue;
if (s->size - size >= sizeof(void *))
continue;
return s;
}
return NULL;
}
struct kmem_cache *kmem_cache_create(const char *name, size_t size,
size_t align, unsigned long flags,
void (*ctor)(void *, struct kmem_cache *, unsigned long),
void (*dtor)(void *, struct kmem_cache *, unsigned long))
{
struct kmem_cache *s;
down_write(&slub_lock);
s = find_mergeable(size, align, flags, dtor, ctor);
if (s) {
s->refcount++;
/*
* Adjust the object sizes so that we clear
* the complete object on kzalloc.
*/
s->objsize = max(s->objsize, (int)size);
s->inuse = max_t(int, s->inuse, ALIGN(size, sizeof(void *)));
if (sysfs_slab_alias(s, name))
goto err;
} else {
s = kmalloc(kmem_size, GFP_KERNEL);
if (s && kmem_cache_open(s, GFP_KERNEL, name,
size, align, flags, ctor, dtor)) {
if (sysfs_slab_add(s)) {
kfree(s);
goto err;
}
list_add(&s->list, &slab_caches);
} else
kfree(s);
}
up_write(&slub_lock);
return s;
err:
up_write(&slub_lock);
if (flags & SLAB_PANIC)
panic("Cannot create slabcache %s\n", name);
else
s = NULL;
return s;
}
EXPORT_SYMBOL(kmem_cache_create);
void *kmem_cache_zalloc(struct kmem_cache *s, gfp_t flags)
{
void *x;
slub: fix object tracking Object tracking did not work the right way for several call chains. Fix this up by adding a new parameter to slub_alloc and slub_free that specifies the caller address explicitly. Signed-off-by: Christoph Lameter <clameter@sgi.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2007-05-06 14:49:42 -07:00
x = slab_alloc(s, flags, -1, __builtin_return_address(0));
SLUB core This is a new slab allocator which was motivated by the complexity of the existing code in mm/slab.c. It attempts to address a variety of concerns with the existing implementation. A. Management of object queues A particular concern was the complex management of the numerous object queues in SLAB. SLUB has no such queues. Instead we dedicate a slab for each allocating CPU and use objects from a slab directly instead of queueing them up. B. Storage overhead of object queues SLAB Object queues exist per node, per CPU. The alien cache queue even has a queue array that contain a queue for each processor on each node. For very large systems the number of queues and the number of objects that may be caught in those queues grows exponentially. On our systems with 1k nodes / processors we have several gigabytes just tied up for storing references to objects for those queues This does not include the objects that could be on those queues. One fears that the whole memory of the machine could one day be consumed by those queues. C. SLAB meta data overhead SLAB has overhead at the beginning of each slab. This means that data cannot be naturally aligned at the beginning of a slab block. SLUB keeps all meta data in the corresponding page_struct. Objects can be naturally aligned in the slab. F.e. a 128 byte object will be aligned at 128 byte boundaries and can fit tightly into a 4k page with no bytes left over. SLAB cannot do this. D. SLAB has a complex cache reaper SLUB does not need a cache reaper for UP systems. On SMP systems the per CPU slab may be pushed back into partial list but that operation is simple and does not require an iteration over a list of objects. SLAB expires per CPU, shared and alien object queues during cache reaping which may cause strange hold offs. E. SLAB has complex NUMA policy layer support SLUB pushes NUMA policy handling into the page allocator. This means that allocation is coarser (SLUB does interleave on a page level) but that situation was also present before 2.6.13. SLABs application of policies to individual slab objects allocated in SLAB is certainly a performance concern due to the frequent references to memory policies which may lead a sequence of objects to come from one node after another. SLUB will get a slab full of objects from one node and then will switch to the next. F. Reduction of the size of partial slab lists SLAB has per node partial lists. This means that over time a large number of partial slabs may accumulate on those lists. These can only be reused if allocator occur on specific nodes. SLUB has a global pool of partial slabs and will consume slabs from that pool to decrease fragmentation. G. Tunables SLAB has sophisticated tuning abilities for each slab cache. One can manipulate the queue sizes in detail. However, filling the queues still requires the uses of the spin lock to check out slabs. SLUB has a global parameter (min_slab_order) for tuning. Increasing the minimum slab order can decrease the locking overhead. The bigger the slab order the less motions of pages between per CPU and partial lists occur and the better SLUB will be scaling. G. Slab merging We often have slab caches with similar parameters. SLUB detects those on boot up and merges them into the corresponding general caches. This leads to more effective memory use. About 50% of all caches can be eliminated through slab merging. This will also decrease slab fragmentation because partial allocated slabs can be filled up again. Slab merging can be switched off by specifying slub_nomerge on boot up. Note that merging can expose heretofore unknown bugs in the kernel because corrupted objects may now be placed differently and corrupt differing neighboring objects. Enable sanity checks to find those. H. Diagnostics The current slab diagnostics are difficult to use and require a recompilation of the kernel. SLUB contains debugging code that is always available (but is kept out of the hot code paths). SLUB diagnostics can be enabled via the "slab_debug" option. Parameters can be specified to select a single or a group of slab caches for diagnostics. This means that the system is running with the usual performance and it is much more likely that race conditions can be reproduced. I. Resiliency If basic sanity checks are on then SLUB is capable of detecting common error conditions and recover as best as possible to allow the system to continue. J. Tracing Tracing can be enabled via the slab_debug=T,<slabcache> option during boot. SLUB will then protocol all actions on that slabcache and dump the object contents on free. K. On demand DMA cache creation. Generally DMA caches are not needed. If a kmalloc is used with __GFP_DMA then just create this single slabcache that is needed. For systems that have no ZONE_DMA requirement the support is completely eliminated. L. Performance increase Some benchmarks have shown speed improvements on kernbench in the range of 5-10%. The locking overhead of slub is based on the underlying base allocation size. If we can reliably allocate larger order pages then it is possible to increase slub performance much further. The anti-fragmentation patches may enable further performance increases. Tested on: i386 UP + SMP, x86_64 UP + SMP + NUMA emulation, IA64 NUMA + Simulator SLUB Boot options slub_nomerge Disable merging of slabs slub_min_order=x Require a minimum order for slab caches. This increases the managed chunk size and therefore reduces meta data and locking overhead. slub_min_objects=x Mininum objects per slab. Default is 8. slub_max_order=x Avoid generating slabs larger than order specified. slub_debug Enable all diagnostics for all caches slub_debug=<options> Enable selective options for all caches slub_debug=<o>,<cache> Enable selective options for a certain set of caches Available Debug options F Double Free checking, sanity and resiliency R Red zoning P Object / padding poisoning U Track last free / alloc T Trace all allocs / frees (only use for individual slabs). To use SLUB: Apply this patch and then select SLUB as the default slab allocator. [hugh@veritas.com: fix an oops-causing locking error] [akpm@linux-foundation.org: various stupid cleanups and small fixes] Signed-off-by: Christoph Lameter <clameter@sgi.com> Signed-off-by: Hugh Dickins <hugh@veritas.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2007-05-06 14:49:36 -07:00
if (x)
memset(x, 0, s->objsize);
return x;
}
EXPORT_SYMBOL(kmem_cache_zalloc);
#ifdef CONFIG_SMP
static void for_all_slabs(void (*func)(struct kmem_cache *, int), int cpu)
{
struct list_head *h;
down_read(&slub_lock);
list_for_each(h, &slab_caches) {
struct kmem_cache *s =
container_of(h, struct kmem_cache, list);
func(s, cpu);
}
up_read(&slub_lock);
}
/*
* Use the cpu notifier to insure that the slab are flushed
* when necessary.
*/
static int __cpuinit slab_cpuup_callback(struct notifier_block *nfb,
unsigned long action, void *hcpu)
{
long cpu = (long)hcpu;
switch (action) {
case CPU_UP_CANCELED:
case CPU_DEAD:
for_all_slabs(__flush_cpu_slab, cpu);
break;
default:
break;
}
return NOTIFY_OK;
}
static struct notifier_block __cpuinitdata slab_notifier =
{ &slab_cpuup_callback, NULL, 0 };
#endif
#ifdef CONFIG_NUMA
/*****************************************************************
* Generic reaper used to support the page allocator
* (the cpu slabs are reaped by a per slab workqueue).
*
* Maybe move this to the page allocator?
****************************************************************/
static DEFINE_PER_CPU(unsigned long, reap_node);
static void init_reap_node(int cpu)
{
int node;
node = next_node(cpu_to_node(cpu), node_online_map);
if (node == MAX_NUMNODES)
node = first_node(node_online_map);
__get_cpu_var(reap_node) = node;
}
static void next_reap_node(void)
{
int node = __get_cpu_var(reap_node);
/*
* Also drain per cpu pages on remote zones
*/
if (node != numa_node_id())
drain_node_pages(node);
node = next_node(node, node_online_map);
if (unlikely(node >= MAX_NUMNODES))
node = first_node(node_online_map);
__get_cpu_var(reap_node) = node;
}
#else
#define init_reap_node(cpu) do { } while (0)
#define next_reap_node(void) do { } while (0)
#endif
#define REAPTIMEOUT_CPUC (2*HZ)
#ifdef CONFIG_SMP
static DEFINE_PER_CPU(struct delayed_work, reap_work);
static void cache_reap(struct work_struct *unused)
{
next_reap_node();
refresh_cpu_vm_stats(smp_processor_id());
schedule_delayed_work(&__get_cpu_var(reap_work),
REAPTIMEOUT_CPUC);
}
static void __devinit start_cpu_timer(int cpu)
{
struct delayed_work *reap_work = &per_cpu(reap_work, cpu);
/*
* When this gets called from do_initcalls via cpucache_init(),
* init_workqueues() has already run, so keventd will be setup
* at that time.
*/
if (keventd_up() && reap_work->work.func == NULL) {
init_reap_node(cpu);
INIT_DELAYED_WORK(reap_work, cache_reap);
schedule_delayed_work_on(cpu, reap_work, HZ + 3 * cpu);
}
}
static int __init cpucache_init(void)
{
int cpu;
/*
* Register the timers that drain pcp pages and update vm statistics
*/
for_each_online_cpu(cpu)
start_cpu_timer(cpu);
return 0;
}
__initcall(cpucache_init);
#endif
#ifdef SLUB_RESILIENCY_TEST
static unsigned long validate_slab_cache(struct kmem_cache *s);
static void resiliency_test(void)
{
u8 *p;
printk(KERN_ERR "SLUB resiliency testing\n");
printk(KERN_ERR "-----------------------\n");
printk(KERN_ERR "A. Corruption after allocation\n");
p = kzalloc(16, GFP_KERNEL);
p[16] = 0x12;
printk(KERN_ERR "\n1. kmalloc-16: Clobber Redzone/next pointer"
" 0x12->0x%p\n\n", p + 16);
validate_slab_cache(kmalloc_caches + 4);
/* Hmmm... The next two are dangerous */
p = kzalloc(32, GFP_KERNEL);
p[32 + sizeof(void *)] = 0x34;
printk(KERN_ERR "\n2. kmalloc-32: Clobber next pointer/next slab"
" 0x34 -> -0x%p\n", p);
printk(KERN_ERR "If allocated object is overwritten then not detectable\n\n");
validate_slab_cache(kmalloc_caches + 5);
p = kzalloc(64, GFP_KERNEL);
p += 64 + (get_cycles() & 0xff) * sizeof(void *);
*p = 0x56;
printk(KERN_ERR "\n3. kmalloc-64: corrupting random byte 0x56->0x%p\n",
p);
printk(KERN_ERR "If allocated object is overwritten then not detectable\n\n");
validate_slab_cache(kmalloc_caches + 6);
printk(KERN_ERR "\nB. Corruption after free\n");
p = kzalloc(128, GFP_KERNEL);
kfree(p);
*p = 0x78;
printk(KERN_ERR "1. kmalloc-128: Clobber first word 0x78->0x%p\n\n", p);
validate_slab_cache(kmalloc_caches + 7);
p = kzalloc(256, GFP_KERNEL);
kfree(p);
p[50] = 0x9a;
printk(KERN_ERR "\n2. kmalloc-256: Clobber 50th byte 0x9a->0x%p\n\n", p);
validate_slab_cache(kmalloc_caches + 8);
p = kzalloc(512, GFP_KERNEL);
kfree(p);
p[512] = 0xab;
printk(KERN_ERR "\n3. kmalloc-512: Clobber redzone 0xab->0x%p\n\n", p);
validate_slab_cache(kmalloc_caches + 9);
}
#else
static void resiliency_test(void) {};
#endif
/*
* These are not as efficient as kmalloc for the non debug case.
* We do not have the page struct available so we have to touch one
* cacheline in struct kmem_cache to check slab flags.
*/
void *__kmalloc_track_caller(size_t size, gfp_t gfpflags, void *caller)
{
struct kmem_cache *s = get_slab(size, gfpflags);
if (!s)
return NULL;
slub: fix object tracking Object tracking did not work the right way for several call chains. Fix this up by adding a new parameter to slub_alloc and slub_free that specifies the caller address explicitly. Signed-off-by: Christoph Lameter <clameter@sgi.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2007-05-06 14:49:42 -07:00
return slab_alloc(s, gfpflags, -1, caller);
SLUB core This is a new slab allocator which was motivated by the complexity of the existing code in mm/slab.c. It attempts to address a variety of concerns with the existing implementation. A. Management of object queues A particular concern was the complex management of the numerous object queues in SLAB. SLUB has no such queues. Instead we dedicate a slab for each allocating CPU and use objects from a slab directly instead of queueing them up. B. Storage overhead of object queues SLAB Object queues exist per node, per CPU. The alien cache queue even has a queue array that contain a queue for each processor on each node. For very large systems the number of queues and the number of objects that may be caught in those queues grows exponentially. On our systems with 1k nodes / processors we have several gigabytes just tied up for storing references to objects for those queues This does not include the objects that could be on those queues. One fears that the whole memory of the machine could one day be consumed by those queues. C. SLAB meta data overhead SLAB has overhead at the beginning of each slab. This means that data cannot be naturally aligned at the beginning of a slab block. SLUB keeps all meta data in the corresponding page_struct. Objects can be naturally aligned in the slab. F.e. a 128 byte object will be aligned at 128 byte boundaries and can fit tightly into a 4k page with no bytes left over. SLAB cannot do this. D. SLAB has a complex cache reaper SLUB does not need a cache reaper for UP systems. On SMP systems the per CPU slab may be pushed back into partial list but that operation is simple and does not require an iteration over a list of objects. SLAB expires per CPU, shared and alien object queues during cache reaping which may cause strange hold offs. E. SLAB has complex NUMA policy layer support SLUB pushes NUMA policy handling into the page allocator. This means that allocation is coarser (SLUB does interleave on a page level) but that situation was also present before 2.6.13. SLABs application of policies to individual slab objects allocated in SLAB is certainly a performance concern due to the frequent references to memory policies which may lead a sequence of objects to come from one node after another. SLUB will get a slab full of objects from one node and then will switch to the next. F. Reduction of the size of partial slab lists SLAB has per node partial lists. This means that over time a large number of partial slabs may accumulate on those lists. These can only be reused if allocator occur on specific nodes. SLUB has a global pool of partial slabs and will consume slabs from that pool to decrease fragmentation. G. Tunables SLAB has sophisticated tuning abilities for each slab cache. One can manipulate the queue sizes in detail. However, filling the queues still requires the uses of the spin lock to check out slabs. SLUB has a global parameter (min_slab_order) for tuning. Increasing the minimum slab order can decrease the locking overhead. The bigger the slab order the less motions of pages between per CPU and partial lists occur and the better SLUB will be scaling. G. Slab merging We often have slab caches with similar parameters. SLUB detects those on boot up and merges them into the corresponding general caches. This leads to more effective memory use. About 50% of all caches can be eliminated through slab merging. This will also decrease slab fragmentation because partial allocated slabs can be filled up again. Slab merging can be switched off by specifying slub_nomerge on boot up. Note that merging can expose heretofore unknown bugs in the kernel because corrupted objects may now be placed differently and corrupt differing neighboring objects. Enable sanity checks to find those. H. Diagnostics The current slab diagnostics are difficult to use and require a recompilation of the kernel. SLUB contains debugging code that is always available (but is kept out of the hot code paths). SLUB diagnostics can be enabled via the "slab_debug" option. Parameters can be specified to select a single or a group of slab caches for diagnostics. This means that the system is running with the usual performance and it is much more likely that race conditions can be reproduced. I. Resiliency If basic sanity checks are on then SLUB is capable of detecting common error conditions and recover as best as possible to allow the system to continue. J. Tracing Tracing can be enabled via the slab_debug=T,<slabcache> option during boot. SLUB will then protocol all actions on that slabcache and dump the object contents on free. K. On demand DMA cache creation. Generally DMA caches are not needed. If a kmalloc is used with __GFP_DMA then just create this single slabcache that is needed. For systems that have no ZONE_DMA requirement the support is completely eliminated. L. Performance increase Some benchmarks have shown speed improvements on kernbench in the range of 5-10%. The locking overhead of slub is based on the underlying base allocation size. If we can reliably allocate larger order pages then it is possible to increase slub performance much further. The anti-fragmentation patches may enable further performance increases. Tested on: i386 UP + SMP, x86_64 UP + SMP + NUMA emulation, IA64 NUMA + Simulator SLUB Boot options slub_nomerge Disable merging of slabs slub_min_order=x Require a minimum order for slab caches. This increases the managed chunk size and therefore reduces meta data and locking overhead. slub_min_objects=x Mininum objects per slab. Default is 8. slub_max_order=x Avoid generating slabs larger than order specified. slub_debug Enable all diagnostics for all caches slub_debug=<options> Enable selective options for all caches slub_debug=<o>,<cache> Enable selective options for a certain set of caches Available Debug options F Double Free checking, sanity and resiliency R Red zoning P Object / padding poisoning U Track last free / alloc T Trace all allocs / frees (only use for individual slabs). To use SLUB: Apply this patch and then select SLUB as the default slab allocator. [hugh@veritas.com: fix an oops-causing locking error] [akpm@linux-foundation.org: various stupid cleanups and small fixes] Signed-off-by: Christoph Lameter <clameter@sgi.com> Signed-off-by: Hugh Dickins <hugh@veritas.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2007-05-06 14:49:36 -07:00
}
void *__kmalloc_node_track_caller(size_t size, gfp_t gfpflags,
int node, void *caller)
{
struct kmem_cache *s = get_slab(size, gfpflags);
if (!s)
return NULL;
slub: fix object tracking Object tracking did not work the right way for several call chains. Fix this up by adding a new parameter to slub_alloc and slub_free that specifies the caller address explicitly. Signed-off-by: Christoph Lameter <clameter@sgi.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2007-05-06 14:49:42 -07:00
return slab_alloc(s, gfpflags, node, caller);
SLUB core This is a new slab allocator which was motivated by the complexity of the existing code in mm/slab.c. It attempts to address a variety of concerns with the existing implementation. A. Management of object queues A particular concern was the complex management of the numerous object queues in SLAB. SLUB has no such queues. Instead we dedicate a slab for each allocating CPU and use objects from a slab directly instead of queueing them up. B. Storage overhead of object queues SLAB Object queues exist per node, per CPU. The alien cache queue even has a queue array that contain a queue for each processor on each node. For very large systems the number of queues and the number of objects that may be caught in those queues grows exponentially. On our systems with 1k nodes / processors we have several gigabytes just tied up for storing references to objects for those queues This does not include the objects that could be on those queues. One fears that the whole memory of the machine could one day be consumed by those queues. C. SLAB meta data overhead SLAB has overhead at the beginning of each slab. This means that data cannot be naturally aligned at the beginning of a slab block. SLUB keeps all meta data in the corresponding page_struct. Objects can be naturally aligned in the slab. F.e. a 128 byte object will be aligned at 128 byte boundaries and can fit tightly into a 4k page with no bytes left over. SLAB cannot do this. D. SLAB has a complex cache reaper SLUB does not need a cache reaper for UP systems. On SMP systems the per CPU slab may be pushed back into partial list but that operation is simple and does not require an iteration over a list of objects. SLAB expires per CPU, shared and alien object queues during cache reaping which may cause strange hold offs. E. SLAB has complex NUMA policy layer support SLUB pushes NUMA policy handling into the page allocator. This means that allocation is coarser (SLUB does interleave on a page level) but that situation was also present before 2.6.13. SLABs application of policies to individual slab objects allocated in SLAB is certainly a performance concern due to the frequent references to memory policies which may lead a sequence of objects to come from one node after another. SLUB will get a slab full of objects from one node and then will switch to the next. F. Reduction of the size of partial slab lists SLAB has per node partial lists. This means that over time a large number of partial slabs may accumulate on those lists. These can only be reused if allocator occur on specific nodes. SLUB has a global pool of partial slabs and will consume slabs from that pool to decrease fragmentation. G. Tunables SLAB has sophisticated tuning abilities for each slab cache. One can manipulate the queue sizes in detail. However, filling the queues still requires the uses of the spin lock to check out slabs. SLUB has a global parameter (min_slab_order) for tuning. Increasing the minimum slab order can decrease the locking overhead. The bigger the slab order the less motions of pages between per CPU and partial lists occur and the better SLUB will be scaling. G. Slab merging We often have slab caches with similar parameters. SLUB detects those on boot up and merges them into the corresponding general caches. This leads to more effective memory use. About 50% of all caches can be eliminated through slab merging. This will also decrease slab fragmentation because partial allocated slabs can be filled up again. Slab merging can be switched off by specifying slub_nomerge on boot up. Note that merging can expose heretofore unknown bugs in the kernel because corrupted objects may now be placed differently and corrupt differing neighboring objects. Enable sanity checks to find those. H. Diagnostics The current slab diagnostics are difficult to use and require a recompilation of the kernel. SLUB contains debugging code that is always available (but is kept out of the hot code paths). SLUB diagnostics can be enabled via the "slab_debug" option. Parameters can be specified to select a single or a group of slab caches for diagnostics. This means that the system is running with the usual performance and it is much more likely that race conditions can be reproduced. I. Resiliency If basic sanity checks are on then SLUB is capable of detecting common error conditions and recover as best as possible to allow the system to continue. J. Tracing Tracing can be enabled via the slab_debug=T,<slabcache> option during boot. SLUB will then protocol all actions on that slabcache and dump the object contents on free. K. On demand DMA cache creation. Generally DMA caches are not needed. If a kmalloc is used with __GFP_DMA then just create this single slabcache that is needed. For systems that have no ZONE_DMA requirement the support is completely eliminated. L. Performance increase Some benchmarks have shown speed improvements on kernbench in the range of 5-10%. The locking overhead of slub is based on the underlying base allocation size. If we can reliably allocate larger order pages then it is possible to increase slub performance much further. The anti-fragmentation patches may enable further performance increases. Tested on: i386 UP + SMP, x86_64 UP + SMP + NUMA emulation, IA64 NUMA + Simulator SLUB Boot options slub_nomerge Disable merging of slabs slub_min_order=x Require a minimum order for slab caches. This increases the managed chunk size and therefore reduces meta data and locking overhead. slub_min_objects=x Mininum objects per slab. Default is 8. slub_max_order=x Avoid generating slabs larger than order specified. slub_debug Enable all diagnostics for all caches slub_debug=<options> Enable selective options for all caches slub_debug=<o>,<cache> Enable selective options for a certain set of caches Available Debug options F Double Free checking, sanity and resiliency R Red zoning P Object / padding poisoning U Track last free / alloc T Trace all allocs / frees (only use for individual slabs). To use SLUB: Apply this patch and then select SLUB as the default slab allocator. [hugh@veritas.com: fix an oops-causing locking error] [akpm@linux-foundation.org: various stupid cleanups and small fixes] Signed-off-by: Christoph Lameter <clameter@sgi.com> Signed-off-by: Hugh Dickins <hugh@veritas.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2007-05-06 14:49:36 -07:00
}
#ifdef CONFIG_SYSFS
slub: validation of slabs (metadata and guard zones) This enables validation of slab. Validation means that all objects are checked to see if there are redzone violations, if padding has been overwritten or any pointers have been corrupted. Also checks the consistency of slab counters. Validation enables the detection of metadata corruption without the kernel having to execute code that actually uses (allocs/frees) and object. It allows one to make sure that the slab metainformation and the guard values around an object have not been compromised. A single slabcache can be checked by writing a 1 to the "validate" file. i.e. echo 1 >/sys/slab/kmalloc-128/validate or use the slabinfo tool to check all slabs slabinfo -v Error messages will show up in the syslog. Note that validation can only reach slabs that are on a list. This means that we are usually restricted to partial slabs and active slabs unless SLAB_STORE_USER is active which will build a full slab list and allows validation of slabs that are fully in use. Booting with "slub_debug" set will enable SLAB_STORE_USER and then full diagnostic are available. Note that we attempt to push cpu slabs back to the lists when we start the check. If the cpu slab is reactivated before we get to it (another processor grabs it before we get to it) then it cannot be checked. Signed-off-by: Christoph Lameter <clameter@sgi.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2007-05-06 14:49:43 -07:00
static int validate_slab(struct kmem_cache *s, struct page *page)
{
void *p;
void *addr = page_address(page);
unsigned long map[BITS_TO_LONGS(s->objects)];
if (!check_slab(s, page) ||
!on_freelist(s, page, NULL))
return 0;
/* Now we know that a valid freelist exists */
bitmap_zero(map, s->objects);
for(p = page->freelist; p; p = get_freepointer(s, p)) {
set_bit((p - addr) / s->size, map);
if (!check_object(s, page, p, 0))
return 0;
}
for(p = addr; p < addr + s->objects * s->size; p += s->size)
if (!test_bit((p - addr) / s->size, map))
if (!check_object(s, page, p, 1))
return 0;
return 1;
}
static void validate_slab_slab(struct kmem_cache *s, struct page *page)
{
if (slab_trylock(page)) {
validate_slab(s, page);
slab_unlock(page);
} else
printk(KERN_INFO "SLUB %s: Skipped busy slab 0x%p\n",
s->name, page);
if (s->flags & DEBUG_DEFAULT_FLAGS) {
if (!PageError(page))
printk(KERN_ERR "SLUB %s: PageError not set "
"on slab 0x%p\n", s->name, page);
} else {
if (PageError(page))
printk(KERN_ERR "SLUB %s: PageError set on "
"slab 0x%p\n", s->name, page);
}
}
static int validate_slab_node(struct kmem_cache *s, struct kmem_cache_node *n)
{
unsigned long count = 0;
struct page *page;
unsigned long flags;
spin_lock_irqsave(&n->list_lock, flags);
list_for_each_entry(page, &n->partial, lru) {
validate_slab_slab(s, page);
count++;
}
if (count != n->nr_partial)
printk(KERN_ERR "SLUB %s: %ld partial slabs counted but "
"counter=%ld\n", s->name, count, n->nr_partial);
if (!(s->flags & SLAB_STORE_USER))
goto out;
list_for_each_entry(page, &n->full, lru) {
validate_slab_slab(s, page);
count++;
}
if (count != atomic_long_read(&n->nr_slabs))
printk(KERN_ERR "SLUB: %s %ld slabs counted but "
"counter=%ld\n", s->name, count,
atomic_long_read(&n->nr_slabs));
out:
spin_unlock_irqrestore(&n->list_lock, flags);
return count;
}
static unsigned long validate_slab_cache(struct kmem_cache *s)
{
int node;
unsigned long count = 0;
flush_all(s);
for_each_online_node(node) {
struct kmem_cache_node *n = get_node(s, node);
count += validate_slab_node(s, n);
}
return count;
}
slub: add ability to list alloc / free callers per slab This patch enables listing the callers who allocated or freed objects in a cache. For example to list the allocators for kmalloc-128 do cat /sys/slab/kmalloc-128/alloc_calls 7 sn_io_slot_fixup+0x40/0x700 7 sn_io_slot_fixup+0x80/0x700 9 sn_bus_fixup+0xe0/0x380 6 param_sysfs_setup+0xf0/0x280 276 percpu_populate+0xf0/0x1a0 19 __register_chrdev_region+0x30/0x360 8 expand_files+0x2e0/0x6e0 1 sys_epoll_create+0x60/0x200 1 __mounts_open+0x140/0x2c0 65 kmem_alloc+0x110/0x280 3 alloc_disk_node+0xe0/0x200 33 as_get_io_context+0x90/0x280 74 kobject_kset_add_dir+0x40/0x140 12 pci_create_bus+0x2a0/0x5c0 1 acpi_ev_create_gpe_block+0x120/0x9e0 41 con_insert_unipair+0x100/0x1c0 1 uart_open+0x1c0/0xba0 1 dma_pool_create+0xe0/0x340 2 neigh_table_init_no_netlink+0x260/0x4c0 6 neigh_parms_alloc+0x30/0x200 1 netlink_kernel_create+0x130/0x320 5 fz_hash_alloc+0x50/0xe0 2 sn_common_hubdev_init+0xd0/0x6e0 28 kernel_param_sysfs_setup+0x30/0x180 72 process_zones+0x70/0x2e0 cat /sys/slab/kmalloc-128/free_calls 558 <not-available> 3 sn_io_slot_fixup+0x600/0x700 84 free_fdtable_rcu+0x120/0x260 2 seq_release+0x40/0x60 6 kmem_free+0x70/0xc0 24 free_as_io_context+0x20/0x200 1 acpi_get_object_info+0x3a0/0x3e0 1 acpi_add_single_object+0xcf0/0x1e40 2 con_release_unimap+0x80/0x140 1 free+0x20/0x40 SLAB_STORE_USER must be enabled for a slab cache by either booting with "slab_debug" or enabling user tracking specifically for the slab of interest. Signed-off-by: Christoph Lameter <clameter@sgi.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2007-05-06 14:49:45 -07:00
/*
* Generate lists of locations where slabcache objects are allocated
* and freed.
*/
struct location {
unsigned long count;
void *addr;
};
struct loc_track {
unsigned long max;
unsigned long count;
struct location *loc;
};
static void free_loc_track(struct loc_track *t)
{
if (t->max)
free_pages((unsigned long)t->loc,
get_order(sizeof(struct location) * t->max));
}
static int alloc_loc_track(struct loc_track *t, unsigned long max)
{
struct location *l;
int order;
if (!max)
max = PAGE_SIZE / sizeof(struct location);
order = get_order(sizeof(struct location) * max);
l = (void *)__get_free_pages(GFP_KERNEL, order);
if (!l)
return 0;
if (t->count) {
memcpy(l, t->loc, sizeof(struct location) * t->count);
free_loc_track(t);
}
t->max = max;
t->loc = l;
return 1;
}
static int add_location(struct loc_track *t, struct kmem_cache *s,
void *addr)
{
long start, end, pos;
struct location *l;
void *caddr;
start = -1;
end = t->count;
for ( ; ; ) {
pos = start + (end - start + 1) / 2;
/*
* There is nothing at "end". If we end up there
* we need to add something to before end.
*/
if (pos == end)
break;
caddr = t->loc[pos].addr;
if (addr == caddr) {
t->loc[pos].count++;
return 1;
}
if (addr < caddr)
end = pos;
else
start = pos;
}
/*
* Not found. Insert new tracking element
*/
if (t->count >= t->max && !alloc_loc_track(t, 2 * t->max))
return 0;
l = t->loc + pos;
if (pos < t->count)
memmove(l + 1, l,
(t->count - pos) * sizeof(struct location));
t->count++;
l->count = 1;
l->addr = addr;
return 1;
}
static void process_slab(struct loc_track *t, struct kmem_cache *s,
struct page *page, enum track_item alloc)
{
void *addr = page_address(page);
unsigned long map[BITS_TO_LONGS(s->objects)];
void *p;
bitmap_zero(map, s->objects);
for (p = page->freelist; p; p = get_freepointer(s, p))
set_bit((p - addr) / s->size, map);
for (p = addr; p < addr + s->objects * s->size; p += s->size)
if (!test_bit((p - addr) / s->size, map)) {
void *addr = get_track(s, p, alloc)->addr;
add_location(t, s, addr);
}
}
static int list_locations(struct kmem_cache *s, char *buf,
enum track_item alloc)
{
int n = 0;
unsigned long i;
struct loc_track t;
int node;
t.count = 0;
t.max = 0;
/* Push back cpu slabs */
flush_all(s);
for_each_online_node(node) {
struct kmem_cache_node *n = get_node(s, node);
unsigned long flags;
struct page *page;
if (!atomic_read(&n->nr_slabs))
continue;
spin_lock_irqsave(&n->list_lock, flags);
list_for_each_entry(page, &n->partial, lru)
process_slab(&t, s, page, alloc);
list_for_each_entry(page, &n->full, lru)
process_slab(&t, s, page, alloc);
spin_unlock_irqrestore(&n->list_lock, flags);
}
for (i = 0; i < t.count; i++) {
void *addr = t.loc[i].addr;
if (n > PAGE_SIZE - 100)
break;
n += sprintf(buf + n, "%7ld ", t.loc[i].count);
if (addr)
n += sprint_symbol(buf + n, (unsigned long)t.loc[i].addr);
else
n += sprintf(buf + n, "<not-available>");
n += sprintf(buf + n, "\n");
}
free_loc_track(&t);
if (!t.count)
n += sprintf(buf, "No data\n");
return n;
}
SLUB core This is a new slab allocator which was motivated by the complexity of the existing code in mm/slab.c. It attempts to address a variety of concerns with the existing implementation. A. Management of object queues A particular concern was the complex management of the numerous object queues in SLAB. SLUB has no such queues. Instead we dedicate a slab for each allocating CPU and use objects from a slab directly instead of queueing them up. B. Storage overhead of object queues SLAB Object queues exist per node, per CPU. The alien cache queue even has a queue array that contain a queue for each processor on each node. For very large systems the number of queues and the number of objects that may be caught in those queues grows exponentially. On our systems with 1k nodes / processors we have several gigabytes just tied up for storing references to objects for those queues This does not include the objects that could be on those queues. One fears that the whole memory of the machine could one day be consumed by those queues. C. SLAB meta data overhead SLAB has overhead at the beginning of each slab. This means that data cannot be naturally aligned at the beginning of a slab block. SLUB keeps all meta data in the corresponding page_struct. Objects can be naturally aligned in the slab. F.e. a 128 byte object will be aligned at 128 byte boundaries and can fit tightly into a 4k page with no bytes left over. SLAB cannot do this. D. SLAB has a complex cache reaper SLUB does not need a cache reaper for UP systems. On SMP systems the per CPU slab may be pushed back into partial list but that operation is simple and does not require an iteration over a list of objects. SLAB expires per CPU, shared and alien object queues during cache reaping which may cause strange hold offs. E. SLAB has complex NUMA policy layer support SLUB pushes NUMA policy handling into the page allocator. This means that allocation is coarser (SLUB does interleave on a page level) but that situation was also present before 2.6.13. SLABs application of policies to individual slab objects allocated in SLAB is certainly a performance concern due to the frequent references to memory policies which may lead a sequence of objects to come from one node after another. SLUB will get a slab full of objects from one node and then will switch to the next. F. Reduction of the size of partial slab lists SLAB has per node partial lists. This means that over time a large number of partial slabs may accumulate on those lists. These can only be reused if allocator occur on specific nodes. SLUB has a global pool of partial slabs and will consume slabs from that pool to decrease fragmentation. G. Tunables SLAB has sophisticated tuning abilities for each slab cache. One can manipulate the queue sizes in detail. However, filling the queues still requires the uses of the spin lock to check out slabs. SLUB has a global parameter (min_slab_order) for tuning. Increasing the minimum slab order can decrease the locking overhead. The bigger the slab order the less motions of pages between per CPU and partial lists occur and the better SLUB will be scaling. G. Slab merging We often have slab caches with similar parameters. SLUB detects those on boot up and merges them into the corresponding general caches. This leads to more effective memory use. About 50% of all caches can be eliminated through slab merging. This will also decrease slab fragmentation because partial allocated slabs can be filled up again. Slab merging can be switched off by specifying slub_nomerge on boot up. Note that merging can expose heretofore unknown bugs in the kernel because corrupted objects may now be placed differently and corrupt differing neighboring objects. Enable sanity checks to find those. H. Diagnostics The current slab diagnostics are difficult to use and require a recompilation of the kernel. SLUB contains debugging code that is always available (but is kept out of the hot code paths). SLUB diagnostics can be enabled via the "slab_debug" option. Parameters can be specified to select a single or a group of slab caches for diagnostics. This means that the system is running with the usual performance and it is much more likely that race conditions can be reproduced. I. Resiliency If basic sanity checks are on then SLUB is capable of detecting common error conditions and recover as best as possible to allow the system to continue. J. Tracing Tracing can be enabled via the slab_debug=T,<slabcache> option during boot. SLUB will then protocol all actions on that slabcache and dump the object contents on free. K. On demand DMA cache creation. Generally DMA caches are not needed. If a kmalloc is used with __GFP_DMA then just create this single slabcache that is needed. For systems that have no ZONE_DMA requirement the support is completely eliminated. L. Performance increase Some benchmarks have shown speed improvements on kernbench in the range of 5-10%. The locking overhead of slub is based on the underlying base allocation size. If we can reliably allocate larger order pages then it is possible to increase slub performance much further. The anti-fragmentation patches may enable further performance increases. Tested on: i386 UP + SMP, x86_64 UP + SMP + NUMA emulation, IA64 NUMA + Simulator SLUB Boot options slub_nomerge Disable merging of slabs slub_min_order=x Require a minimum order for slab caches. This increases the managed chunk size and therefore reduces meta data and locking overhead. slub_min_objects=x Mininum objects per slab. Default is 8. slub_max_order=x Avoid generating slabs larger than order specified. slub_debug Enable all diagnostics for all caches slub_debug=<options> Enable selective options for all caches slub_debug=<o>,<cache> Enable selective options for a certain set of caches Available Debug options F Double Free checking, sanity and resiliency R Red zoning P Object / padding poisoning U Track last free / alloc T Trace all allocs / frees (only use for individual slabs). To use SLUB: Apply this patch and then select SLUB as the default slab allocator. [hugh@veritas.com: fix an oops-causing locking error] [akpm@linux-foundation.org: various stupid cleanups and small fixes] Signed-off-by: Christoph Lameter <clameter@sgi.com> Signed-off-by: Hugh Dickins <hugh@veritas.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2007-05-06 14:49:36 -07:00
static unsigned long count_partial(struct kmem_cache_node *n)
{
unsigned long flags;
unsigned long x = 0;
struct page *page;
spin_lock_irqsave(&n->list_lock, flags);
list_for_each_entry(page, &n->partial, lru)
x += page->inuse;
spin_unlock_irqrestore(&n->list_lock, flags);
return x;
}
enum slab_stat_type {
SL_FULL,
SL_PARTIAL,
SL_CPU,
SL_OBJECTS
};
#define SO_FULL (1 << SL_FULL)
#define SO_PARTIAL (1 << SL_PARTIAL)
#define SO_CPU (1 << SL_CPU)
#define SO_OBJECTS (1 << SL_OBJECTS)
static unsigned long slab_objects(struct kmem_cache *s,
char *buf, unsigned long flags)
{
unsigned long total = 0;
int cpu;
int node;
int x;
unsigned long *nodes;
unsigned long *per_cpu;
nodes = kzalloc(2 * sizeof(unsigned long) * nr_node_ids, GFP_KERNEL);
per_cpu = nodes + nr_node_ids;
for_each_possible_cpu(cpu) {
struct page *page = s->cpu_slab[cpu];
int node;
if (page) {
node = page_to_nid(page);
if (flags & SO_CPU) {
int x = 0;
if (flags & SO_OBJECTS)
x = page->inuse;
else
x = 1;
total += x;
nodes[node] += x;
}
per_cpu[node]++;
}
}
for_each_online_node(node) {
struct kmem_cache_node *n = get_node(s, node);
if (flags & SO_PARTIAL) {
if (flags & SO_OBJECTS)
x = count_partial(n);
else
x = n->nr_partial;
total += x;
nodes[node] += x;
}
if (flags & SO_FULL) {
int full_slabs = atomic_read(&n->nr_slabs)
- per_cpu[node]
- n->nr_partial;
if (flags & SO_OBJECTS)
x = full_slabs * s->objects;
else
x = full_slabs;
total += x;
nodes[node] += x;
}
}
x = sprintf(buf, "%lu", total);
#ifdef CONFIG_NUMA
for_each_online_node(node)
if (nodes[node])
x += sprintf(buf + x, " N%d=%lu",
node, nodes[node]);
#endif
kfree(nodes);
return x + sprintf(buf + x, "\n");
}
static int any_slab_objects(struct kmem_cache *s)
{
int node;
int cpu;
for_each_possible_cpu(cpu)
if (s->cpu_slab[cpu])
return 1;
for_each_node(node) {
struct kmem_cache_node *n = get_node(s, node);
if (n->nr_partial || atomic_read(&n->nr_slabs))
return 1;
}
return 0;
}
#define to_slab_attr(n) container_of(n, struct slab_attribute, attr)
#define to_slab(n) container_of(n, struct kmem_cache, kobj);
struct slab_attribute {
struct attribute attr;
ssize_t (*show)(struct kmem_cache *s, char *buf);
ssize_t (*store)(struct kmem_cache *s, const char *x, size_t count);
};
#define SLAB_ATTR_RO(_name) \
static struct slab_attribute _name##_attr = __ATTR_RO(_name)
#define SLAB_ATTR(_name) \
static struct slab_attribute _name##_attr = \
__ATTR(_name, 0644, _name##_show, _name##_store)
static ssize_t slab_size_show(struct kmem_cache *s, char *buf)
{
return sprintf(buf, "%d\n", s->size);
}
SLAB_ATTR_RO(slab_size);
static ssize_t align_show(struct kmem_cache *s, char *buf)
{
return sprintf(buf, "%d\n", s->align);
}
SLAB_ATTR_RO(align);
static ssize_t object_size_show(struct kmem_cache *s, char *buf)
{
return sprintf(buf, "%d\n", s->objsize);
}
SLAB_ATTR_RO(object_size);
static ssize_t objs_per_slab_show(struct kmem_cache *s, char *buf)
{
return sprintf(buf, "%d\n", s->objects);
}
SLAB_ATTR_RO(objs_per_slab);
static ssize_t order_show(struct kmem_cache *s, char *buf)
{
return sprintf(buf, "%d\n", s->order);
}
SLAB_ATTR_RO(order);
static ssize_t ctor_show(struct kmem_cache *s, char *buf)
{
if (s->ctor) {
int n = sprint_symbol(buf, (unsigned long)s->ctor);
return n + sprintf(buf + n, "\n");
}
return 0;
}
SLAB_ATTR_RO(ctor);
static ssize_t dtor_show(struct kmem_cache *s, char *buf)
{
if (s->dtor) {
int n = sprint_symbol(buf, (unsigned long)s->dtor);
return n + sprintf(buf + n, "\n");
}
return 0;
}
SLAB_ATTR_RO(dtor);
static ssize_t aliases_show(struct kmem_cache *s, char *buf)
{
return sprintf(buf, "%d\n", s->refcount - 1);
}
SLAB_ATTR_RO(aliases);
static ssize_t slabs_show(struct kmem_cache *s, char *buf)
{
return slab_objects(s, buf, SO_FULL|SO_PARTIAL|SO_CPU);
}
SLAB_ATTR_RO(slabs);
static ssize_t partial_show(struct kmem_cache *s, char *buf)
{
return slab_objects(s, buf, SO_PARTIAL);
}
SLAB_ATTR_RO(partial);
static ssize_t cpu_slabs_show(struct kmem_cache *s, char *buf)
{
return slab_objects(s, buf, SO_CPU);
}
SLAB_ATTR_RO(cpu_slabs);
static ssize_t objects_show(struct kmem_cache *s, char *buf)
{
return slab_objects(s, buf, SO_FULL|SO_PARTIAL|SO_CPU|SO_OBJECTS);
}
SLAB_ATTR_RO(objects);
static ssize_t sanity_checks_show(struct kmem_cache *s, char *buf)
{
return sprintf(buf, "%d\n", !!(s->flags & SLAB_DEBUG_FREE));
}
static ssize_t sanity_checks_store(struct kmem_cache *s,
const char *buf, size_t length)
{
s->flags &= ~SLAB_DEBUG_FREE;
if (buf[0] == '1')
s->flags |= SLAB_DEBUG_FREE;
return length;
}
SLAB_ATTR(sanity_checks);
static ssize_t trace_show(struct kmem_cache *s, char *buf)
{
return sprintf(buf, "%d\n", !!(s->flags & SLAB_TRACE));
}
static ssize_t trace_store(struct kmem_cache *s, const char *buf,
size_t length)
{
s->flags &= ~SLAB_TRACE;
if (buf[0] == '1')
s->flags |= SLAB_TRACE;
return length;
}
SLAB_ATTR(trace);
static ssize_t reclaim_account_show(struct kmem_cache *s, char *buf)
{
return sprintf(buf, "%d\n", !!(s->flags & SLAB_RECLAIM_ACCOUNT));
}
static ssize_t reclaim_account_store(struct kmem_cache *s,
const char *buf, size_t length)
{
s->flags &= ~SLAB_RECLAIM_ACCOUNT;
if (buf[0] == '1')
s->flags |= SLAB_RECLAIM_ACCOUNT;
return length;
}
SLAB_ATTR(reclaim_account);
static ssize_t hwcache_align_show(struct kmem_cache *s, char *buf)
{
slab allocators: Remove obsolete SLAB_MUST_HWCACHE_ALIGN This patch was recently posted to lkml and acked by Pekka. The flag SLAB_MUST_HWCACHE_ALIGN is 1. Never checked by SLAB at all. 2. A duplicate of SLAB_HWCACHE_ALIGN for SLUB 3. Fulfills the role of SLAB_HWCACHE_ALIGN for SLOB. The only remaining use is in sparc64 and ppc64 and their use there reflects some earlier role that the slab flag once may have had. If its specified then SLAB_HWCACHE_ALIGN is also specified. The flag is confusing, inconsistent and has no purpose. Remove it. Acked-by: Pekka Enberg <penberg@cs.helsinki.fi> Signed-off-by: Christoph Lameter <clameter@sgi.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2007-05-06 14:49:56 -07:00
return sprintf(buf, "%d\n", !!(s->flags & SLAB_HWCACHE_ALIGN));
SLUB core This is a new slab allocator which was motivated by the complexity of the existing code in mm/slab.c. It attempts to address a variety of concerns with the existing implementation. A. Management of object queues A particular concern was the complex management of the numerous object queues in SLAB. SLUB has no such queues. Instead we dedicate a slab for each allocating CPU and use objects from a slab directly instead of queueing them up. B. Storage overhead of object queues SLAB Object queues exist per node, per CPU. The alien cache queue even has a queue array that contain a queue for each processor on each node. For very large systems the number of queues and the number of objects that may be caught in those queues grows exponentially. On our systems with 1k nodes / processors we have several gigabytes just tied up for storing references to objects for those queues This does not include the objects that could be on those queues. One fears that the whole memory of the machine could one day be consumed by those queues. C. SLAB meta data overhead SLAB has overhead at the beginning of each slab. This means that data cannot be naturally aligned at the beginning of a slab block. SLUB keeps all meta data in the corresponding page_struct. Objects can be naturally aligned in the slab. F.e. a 128 byte object will be aligned at 128 byte boundaries and can fit tightly into a 4k page with no bytes left over. SLAB cannot do this. D. SLAB has a complex cache reaper SLUB does not need a cache reaper for UP systems. On SMP systems the per CPU slab may be pushed back into partial list but that operation is simple and does not require an iteration over a list of objects. SLAB expires per CPU, shared and alien object queues during cache reaping which may cause strange hold offs. E. SLAB has complex NUMA policy layer support SLUB pushes NUMA policy handling into the page allocator. This means that allocation is coarser (SLUB does interleave on a page level) but that situation was also present before 2.6.13. SLABs application of policies to individual slab objects allocated in SLAB is certainly a performance concern due to the frequent references to memory policies which may lead a sequence of objects to come from one node after another. SLUB will get a slab full of objects from one node and then will switch to the next. F. Reduction of the size of partial slab lists SLAB has per node partial lists. This means that over time a large number of partial slabs may accumulate on those lists. These can only be reused if allocator occur on specific nodes. SLUB has a global pool of partial slabs and will consume slabs from that pool to decrease fragmentation. G. Tunables SLAB has sophisticated tuning abilities for each slab cache. One can manipulate the queue sizes in detail. However, filling the queues still requires the uses of the spin lock to check out slabs. SLUB has a global parameter (min_slab_order) for tuning. Increasing the minimum slab order can decrease the locking overhead. The bigger the slab order the less motions of pages between per CPU and partial lists occur and the better SLUB will be scaling. G. Slab merging We often have slab caches with similar parameters. SLUB detects those on boot up and merges them into the corresponding general caches. This leads to more effective memory use. About 50% of all caches can be eliminated through slab merging. This will also decrease slab fragmentation because partial allocated slabs can be filled up again. Slab merging can be switched off by specifying slub_nomerge on boot up. Note that merging can expose heretofore unknown bugs in the kernel because corrupted objects may now be placed differently and corrupt differing neighboring objects. Enable sanity checks to find those. H. Diagnostics The current slab diagnostics are difficult to use and require a recompilation of the kernel. SLUB contains debugging code that is always available (but is kept out of the hot code paths). SLUB diagnostics can be enabled via the "slab_debug" option. Parameters can be specified to select a single or a group of slab caches for diagnostics. This means that the system is running with the usual performance and it is much more likely that race conditions can be reproduced. I. Resiliency If basic sanity checks are on then SLUB is capable of detecting common error conditions and recover as best as possible to allow the system to continue. J. Tracing Tracing can be enabled via the slab_debug=T,<slabcache> option during boot. SLUB will then protocol all actions on that slabcache and dump the object contents on free. K. On demand DMA cache creation. Generally DMA caches are not needed. If a kmalloc is used with __GFP_DMA then just create this single slabcache that is needed. For systems that have no ZONE_DMA requirement the support is completely eliminated. L. Performance increase Some benchmarks have shown speed improvements on kernbench in the range of 5-10%. The locking overhead of slub is based on the underlying base allocation size. If we can reliably allocate larger order pages then it is possible to increase slub performance much further. The anti-fragmentation patches may enable further performance increases. Tested on: i386 UP + SMP, x86_64 UP + SMP + NUMA emulation, IA64 NUMA + Simulator SLUB Boot options slub_nomerge Disable merging of slabs slub_min_order=x Require a minimum order for slab caches. This increases the managed chunk size and therefore reduces meta data and locking overhead. slub_min_objects=x Mininum objects per slab. Default is 8. slub_max_order=x Avoid generating slabs larger than order specified. slub_debug Enable all diagnostics for all caches slub_debug=<options> Enable selective options for all caches slub_debug=<o>,<cache> Enable selective options for a certain set of caches Available Debug options F Double Free checking, sanity and resiliency R Red zoning P Object / padding poisoning U Track last free / alloc T Trace all allocs / frees (only use for individual slabs). To use SLUB: Apply this patch and then select SLUB as the default slab allocator. [hugh@veritas.com: fix an oops-causing locking error] [akpm@linux-foundation.org: various stupid cleanups and small fixes] Signed-off-by: Christoph Lameter <clameter@sgi.com> Signed-off-by: Hugh Dickins <hugh@veritas.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2007-05-06 14:49:36 -07:00
}
SLAB_ATTR_RO(hwcache_align);
#ifdef CONFIG_ZONE_DMA
static ssize_t cache_dma_show(struct kmem_cache *s, char *buf)
{
return sprintf(buf, "%d\n", !!(s->flags & SLAB_CACHE_DMA));
}
SLAB_ATTR_RO(cache_dma);
#endif
static ssize_t destroy_by_rcu_show(struct kmem_cache *s, char *buf)
{
return sprintf(buf, "%d\n", !!(s->flags & SLAB_DESTROY_BY_RCU));
}
SLAB_ATTR_RO(destroy_by_rcu);
static ssize_t red_zone_show(struct kmem_cache *s, char *buf)
{
return sprintf(buf, "%d\n", !!(s->flags & SLAB_RED_ZONE));
}
static ssize_t red_zone_store(struct kmem_cache *s,
const char *buf, size_t length)
{
if (any_slab_objects(s))
return -EBUSY;
s->flags &= ~SLAB_RED_ZONE;
if (buf[0] == '1')
s->flags |= SLAB_RED_ZONE;
calculate_sizes(s);
return length;
}
SLAB_ATTR(red_zone);
static ssize_t poison_show(struct kmem_cache *s, char *buf)
{
return sprintf(buf, "%d\n", !!(s->flags & SLAB_POISON));
}
static ssize_t poison_store(struct kmem_cache *s,
const char *buf, size_t length)
{
if (any_slab_objects(s))
return -EBUSY;
s->flags &= ~SLAB_POISON;
if (buf[0] == '1')
s->flags |= SLAB_POISON;
calculate_sizes(s);
return length;
}
SLAB_ATTR(poison);
static ssize_t store_user_show(struct kmem_cache *s, char *buf)
{
return sprintf(buf, "%d\n", !!(s->flags & SLAB_STORE_USER));
}
static ssize_t store_user_store(struct kmem_cache *s,
const char *buf, size_t length)
{
if (any_slab_objects(s))
return -EBUSY;
s->flags &= ~SLAB_STORE_USER;
if (buf[0] == '1')
s->flags |= SLAB_STORE_USER;
calculate_sizes(s);
return length;
}
SLAB_ATTR(store_user);
slub: validation of slabs (metadata and guard zones) This enables validation of slab. Validation means that all objects are checked to see if there are redzone violations, if padding has been overwritten or any pointers have been corrupted. Also checks the consistency of slab counters. Validation enables the detection of metadata corruption without the kernel having to execute code that actually uses (allocs/frees) and object. It allows one to make sure that the slab metainformation and the guard values around an object have not been compromised. A single slabcache can be checked by writing a 1 to the "validate" file. i.e. echo 1 >/sys/slab/kmalloc-128/validate or use the slabinfo tool to check all slabs slabinfo -v Error messages will show up in the syslog. Note that validation can only reach slabs that are on a list. This means that we are usually restricted to partial slabs and active slabs unless SLAB_STORE_USER is active which will build a full slab list and allows validation of slabs that are fully in use. Booting with "slub_debug" set will enable SLAB_STORE_USER and then full diagnostic are available. Note that we attempt to push cpu slabs back to the lists when we start the check. If the cpu slab is reactivated before we get to it (another processor grabs it before we get to it) then it cannot be checked. Signed-off-by: Christoph Lameter <clameter@sgi.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2007-05-06 14:49:43 -07:00
static ssize_t validate_show(struct kmem_cache *s, char *buf)
{
return 0;
}
static ssize_t validate_store(struct kmem_cache *s,
const char *buf, size_t length)
{
if (buf[0] == '1')
validate_slab_cache(s);
else
return -EINVAL;
return length;
}
SLAB_ATTR(validate);
SLUB: Free slabs and sort partial slab lists in kmem_cache_shrink At kmem_cache_shrink check if we have any empty slabs on the partial if so then remove them. Also--as an anti-fragmentation measure--sort the partial slabs so that the most fully allocated ones come first and the least allocated last. The next allocations may fill up the nearly full slabs. Having the least allocated slabs last gives them the maximum chance that their remaining objects may be freed. Thus we can hopefully minimize the partial slabs. I think this is the best one can do in terms antifragmentation measures. Real defragmentation (meaning moving objects out of slabs with the least free objects to those that are almost full) can be implemted by reverse scanning through the list produced here but that would mean that we need to provide a callback at slab cache creation that allows the deletion or moving of an object. This will involve slab API changes, so defer for now. Cc: Mel Gorman <mel@skynet.ie> Signed-off-by: Christoph Lameter <clameter@sgi.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2007-05-06 14:49:46 -07:00
static ssize_t shrink_show(struct kmem_cache *s, char *buf)
{
return 0;
}
static ssize_t shrink_store(struct kmem_cache *s,
const char *buf, size_t length)
{
if (buf[0] == '1') {
int rc = kmem_cache_shrink(s);
if (rc)
return rc;
} else
return -EINVAL;
return length;
}
SLAB_ATTR(shrink);
slub: add ability to list alloc / free callers per slab This patch enables listing the callers who allocated or freed objects in a cache. For example to list the allocators for kmalloc-128 do cat /sys/slab/kmalloc-128/alloc_calls 7 sn_io_slot_fixup+0x40/0x700 7 sn_io_slot_fixup+0x80/0x700 9 sn_bus_fixup+0xe0/0x380 6 param_sysfs_setup+0xf0/0x280 276 percpu_populate+0xf0/0x1a0 19 __register_chrdev_region+0x30/0x360 8 expand_files+0x2e0/0x6e0 1 sys_epoll_create+0x60/0x200 1 __mounts_open+0x140/0x2c0 65 kmem_alloc+0x110/0x280 3 alloc_disk_node+0xe0/0x200 33 as_get_io_context+0x90/0x280 74 kobject_kset_add_dir+0x40/0x140 12 pci_create_bus+0x2a0/0x5c0 1 acpi_ev_create_gpe_block+0x120/0x9e0 41 con_insert_unipair+0x100/0x1c0 1 uart_open+0x1c0/0xba0 1 dma_pool_create+0xe0/0x340 2 neigh_table_init_no_netlink+0x260/0x4c0 6 neigh_parms_alloc+0x30/0x200 1 netlink_kernel_create+0x130/0x320 5 fz_hash_alloc+0x50/0xe0 2 sn_common_hubdev_init+0xd0/0x6e0 28 kernel_param_sysfs_setup+0x30/0x180 72 process_zones+0x70/0x2e0 cat /sys/slab/kmalloc-128/free_calls 558 <not-available> 3 sn_io_slot_fixup+0x600/0x700 84 free_fdtable_rcu+0x120/0x260 2 seq_release+0x40/0x60 6 kmem_free+0x70/0xc0 24 free_as_io_context+0x20/0x200 1 acpi_get_object_info+0x3a0/0x3e0 1 acpi_add_single_object+0xcf0/0x1e40 2 con_release_unimap+0x80/0x140 1 free+0x20/0x40 SLAB_STORE_USER must be enabled for a slab cache by either booting with "slab_debug" or enabling user tracking specifically for the slab of interest. Signed-off-by: Christoph Lameter <clameter@sgi.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2007-05-06 14:49:45 -07:00
static ssize_t alloc_calls_show(struct kmem_cache *s, char *buf)
{
if (!(s->flags & SLAB_STORE_USER))
return -ENOSYS;
return list_locations(s, buf, TRACK_ALLOC);
}
SLAB_ATTR_RO(alloc_calls);
static ssize_t free_calls_show(struct kmem_cache *s, char *buf)
{
if (!(s->flags & SLAB_STORE_USER))
return -ENOSYS;
return list_locations(s, buf, TRACK_FREE);
}
SLAB_ATTR_RO(free_calls);
SLUB core This is a new slab allocator which was motivated by the complexity of the existing code in mm/slab.c. It attempts to address a variety of concerns with the existing implementation. A. Management of object queues A particular concern was the complex management of the numerous object queues in SLAB. SLUB has no such queues. Instead we dedicate a slab for each allocating CPU and use objects from a slab directly instead of queueing them up. B. Storage overhead of object queues SLAB Object queues exist per node, per CPU. The alien cache queue even has a queue array that contain a queue for each processor on each node. For very large systems the number of queues and the number of objects that may be caught in those queues grows exponentially. On our systems with 1k nodes / processors we have several gigabytes just tied up for storing references to objects for those queues This does not include the objects that could be on those queues. One fears that the whole memory of the machine could one day be consumed by those queues. C. SLAB meta data overhead SLAB has overhead at the beginning of each slab. This means that data cannot be naturally aligned at the beginning of a slab block. SLUB keeps all meta data in the corresponding page_struct. Objects can be naturally aligned in the slab. F.e. a 128 byte object will be aligned at 128 byte boundaries and can fit tightly into a 4k page with no bytes left over. SLAB cannot do this. D. SLAB has a complex cache reaper SLUB does not need a cache reaper for UP systems. On SMP systems the per CPU slab may be pushed back into partial list but that operation is simple and does not require an iteration over a list of objects. SLAB expires per CPU, shared and alien object queues during cache reaping which may cause strange hold offs. E. SLAB has complex NUMA policy layer support SLUB pushes NUMA policy handling into the page allocator. This means that allocation is coarser (SLUB does interleave on a page level) but that situation was also present before 2.6.13. SLABs application of policies to individual slab objects allocated in SLAB is certainly a performance concern due to the frequent references to memory policies which may lead a sequence of objects to come from one node after another. SLUB will get a slab full of objects from one node and then will switch to the next. F. Reduction of the size of partial slab lists SLAB has per node partial lists. This means that over time a large number of partial slabs may accumulate on those lists. These can only be reused if allocator occur on specific nodes. SLUB has a global pool of partial slabs and will consume slabs from that pool to decrease fragmentation. G. Tunables SLAB has sophisticated tuning abilities for each slab cache. One can manipulate the queue sizes in detail. However, filling the queues still requires the uses of the spin lock to check out slabs. SLUB has a global parameter (min_slab_order) for tuning. Increasing the minimum slab order can decrease the locking overhead. The bigger the slab order the less motions of pages between per CPU and partial lists occur and the better SLUB will be scaling. G. Slab merging We often have slab caches with similar parameters. SLUB detects those on boot up and merges them into the corresponding general caches. This leads to more effective memory use. About 50% of all caches can be eliminated through slab merging. This will also decrease slab fragmentation because partial allocated slabs can be filled up again. Slab merging can be switched off by specifying slub_nomerge on boot up. Note that merging can expose heretofore unknown bugs in the kernel because corrupted objects may now be placed differently and corrupt differing neighboring objects. Enable sanity checks to find those. H. Diagnostics The current slab diagnostics are difficult to use and require a recompilation of the kernel. SLUB contains debugging code that is always available (but is kept out of the hot code paths). SLUB diagnostics can be enabled via the "slab_debug" option. Parameters can be specified to select a single or a group of slab caches for diagnostics. This means that the system is running with the usual performance and it is much more likely that race conditions can be reproduced. I. Resiliency If basic sanity checks are on then SLUB is capable of detecting common error conditions and recover as best as possible to allow the system to continue. J. Tracing Tracing can be enabled via the slab_debug=T,<slabcache> option during boot. SLUB will then protocol all actions on that slabcache and dump the object contents on free. K. On demand DMA cache creation. Generally DMA caches are not needed. If a kmalloc is used with __GFP_DMA then just create this single slabcache that is needed. For systems that have no ZONE_DMA requirement the support is completely eliminated. L. Performance increase Some benchmarks have shown speed improvements on kernbench in the range of 5-10%. The locking overhead of slub is based on the underlying base allocation size. If we can reliably allocate larger order pages then it is possible to increase slub performance much further. The anti-fragmentation patches may enable further performance increases. Tested on: i386 UP + SMP, x86_64 UP + SMP + NUMA emulation, IA64 NUMA + Simulator SLUB Boot options slub_nomerge Disable merging of slabs slub_min_order=x Require a minimum order for slab caches. This increases the managed chunk size and therefore reduces meta data and locking overhead. slub_min_objects=x Mininum objects per slab. Default is 8. slub_max_order=x Avoid generating slabs larger than order specified. slub_debug Enable all diagnostics for all caches slub_debug=<options> Enable selective options for all caches slub_debug=<o>,<cache> Enable selective options for a certain set of caches Available Debug options F Double Free checking, sanity and resiliency R Red zoning P Object / padding poisoning U Track last free / alloc T Trace all allocs / frees (only use for individual slabs). To use SLUB: Apply this patch and then select SLUB as the default slab allocator. [hugh@veritas.com: fix an oops-causing locking error] [akpm@linux-foundation.org: various stupid cleanups and small fixes] Signed-off-by: Christoph Lameter <clameter@sgi.com> Signed-off-by: Hugh Dickins <hugh@veritas.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2007-05-06 14:49:36 -07:00
#ifdef CONFIG_NUMA
static ssize_t defrag_ratio_show(struct kmem_cache *s, char *buf)
{
return sprintf(buf, "%d\n", s->defrag_ratio / 10);
}
static ssize_t defrag_ratio_store(struct kmem_cache *s,
const char *buf, size_t length)
{
int n = simple_strtoul(buf, NULL, 10);
if (n < 100)
s->defrag_ratio = n * 10;
return length;
}
SLAB_ATTR(defrag_ratio);
#endif
static struct attribute * slab_attrs[] = {
&slab_size_attr.attr,
&object_size_attr.attr,
&objs_per_slab_attr.attr,
&order_attr.attr,
&objects_attr.attr,
&slabs_attr.attr,
&partial_attr.attr,
&cpu_slabs_attr.attr,
&ctor_attr.attr,
&dtor_attr.attr,
&aliases_attr.attr,
&align_attr.attr,
&sanity_checks_attr.attr,
&trace_attr.attr,
&hwcache_align_attr.attr,
&reclaim_account_attr.attr,
&destroy_by_rcu_attr.attr,
&red_zone_attr.attr,
&poison_attr.attr,
&store_user_attr.attr,
slub: validation of slabs (metadata and guard zones) This enables validation of slab. Validation means that all objects are checked to see if there are redzone violations, if padding has been overwritten or any pointers have been corrupted. Also checks the consistency of slab counters. Validation enables the detection of metadata corruption without the kernel having to execute code that actually uses (allocs/frees) and object. It allows one to make sure that the slab metainformation and the guard values around an object have not been compromised. A single slabcache can be checked by writing a 1 to the "validate" file. i.e. echo 1 >/sys/slab/kmalloc-128/validate or use the slabinfo tool to check all slabs slabinfo -v Error messages will show up in the syslog. Note that validation can only reach slabs that are on a list. This means that we are usually restricted to partial slabs and active slabs unless SLAB_STORE_USER is active which will build a full slab list and allows validation of slabs that are fully in use. Booting with "slub_debug" set will enable SLAB_STORE_USER and then full diagnostic are available. Note that we attempt to push cpu slabs back to the lists when we start the check. If the cpu slab is reactivated before we get to it (another processor grabs it before we get to it) then it cannot be checked. Signed-off-by: Christoph Lameter <clameter@sgi.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2007-05-06 14:49:43 -07:00
&validate_attr.attr,
SLUB: Free slabs and sort partial slab lists in kmem_cache_shrink At kmem_cache_shrink check if we have any empty slabs on the partial if so then remove them. Also--as an anti-fragmentation measure--sort the partial slabs so that the most fully allocated ones come first and the least allocated last. The next allocations may fill up the nearly full slabs. Having the least allocated slabs last gives them the maximum chance that their remaining objects may be freed. Thus we can hopefully minimize the partial slabs. I think this is the best one can do in terms antifragmentation measures. Real defragmentation (meaning moving objects out of slabs with the least free objects to those that are almost full) can be implemted by reverse scanning through the list produced here but that would mean that we need to provide a callback at slab cache creation that allows the deletion or moving of an object. This will involve slab API changes, so defer for now. Cc: Mel Gorman <mel@skynet.ie> Signed-off-by: Christoph Lameter <clameter@sgi.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2007-05-06 14:49:46 -07:00
&shrink_attr.attr,
slub: add ability to list alloc / free callers per slab This patch enables listing the callers who allocated or freed objects in a cache. For example to list the allocators for kmalloc-128 do cat /sys/slab/kmalloc-128/alloc_calls 7 sn_io_slot_fixup+0x40/0x700 7 sn_io_slot_fixup+0x80/0x700 9 sn_bus_fixup+0xe0/0x380 6 param_sysfs_setup+0xf0/0x280 276 percpu_populate+0xf0/0x1a0 19 __register_chrdev_region+0x30/0x360 8 expand_files+0x2e0/0x6e0 1 sys_epoll_create+0x60/0x200 1 __mounts_open+0x140/0x2c0 65 kmem_alloc+0x110/0x280 3 alloc_disk_node+0xe0/0x200 33 as_get_io_context+0x90/0x280 74 kobject_kset_add_dir+0x40/0x140 12 pci_create_bus+0x2a0/0x5c0 1 acpi_ev_create_gpe_block+0x120/0x9e0 41 con_insert_unipair+0x100/0x1c0 1 uart_open+0x1c0/0xba0 1 dma_pool_create+0xe0/0x340 2 neigh_table_init_no_netlink+0x260/0x4c0 6 neigh_parms_alloc+0x30/0x200 1 netlink_kernel_create+0x130/0x320 5 fz_hash_alloc+0x50/0xe0 2 sn_common_hubdev_init+0xd0/0x6e0 28 kernel_param_sysfs_setup+0x30/0x180 72 process_zones+0x70/0x2e0 cat /sys/slab/kmalloc-128/free_calls 558 <not-available> 3 sn_io_slot_fixup+0x600/0x700 84 free_fdtable_rcu+0x120/0x260 2 seq_release+0x40/0x60 6 kmem_free+0x70/0xc0 24 free_as_io_context+0x20/0x200 1 acpi_get_object_info+0x3a0/0x3e0 1 acpi_add_single_object+0xcf0/0x1e40 2 con_release_unimap+0x80/0x140 1 free+0x20/0x40 SLAB_STORE_USER must be enabled for a slab cache by either booting with "slab_debug" or enabling user tracking specifically for the slab of interest. Signed-off-by: Christoph Lameter <clameter@sgi.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2007-05-06 14:49:45 -07:00
&alloc_calls_attr.attr,
&free_calls_attr.attr,
SLUB core This is a new slab allocator which was motivated by the complexity of the existing code in mm/slab.c. It attempts to address a variety of concerns with the existing implementation. A. Management of object queues A particular concern was the complex management of the numerous object queues in SLAB. SLUB has no such queues. Instead we dedicate a slab for each allocating CPU and use objects from a slab directly instead of queueing them up. B. Storage overhead of object queues SLAB Object queues exist per node, per CPU. The alien cache queue even has a queue array that contain a queue for each processor on each node. For very large systems the number of queues and the number of objects that may be caught in those queues grows exponentially. On our systems with 1k nodes / processors we have several gigabytes just tied up for storing references to objects for those queues This does not include the objects that could be on those queues. One fears that the whole memory of the machine could one day be consumed by those queues. C. SLAB meta data overhead SLAB has overhead at the beginning of each slab. This means that data cannot be naturally aligned at the beginning of a slab block. SLUB keeps all meta data in the corresponding page_struct. Objects can be naturally aligned in the slab. F.e. a 128 byte object will be aligned at 128 byte boundaries and can fit tightly into a 4k page with no bytes left over. SLAB cannot do this. D. SLAB has a complex cache reaper SLUB does not need a cache reaper for UP systems. On SMP systems the per CPU slab may be pushed back into partial list but that operation is simple and does not require an iteration over a list of objects. SLAB expires per CPU, shared and alien object queues during cache reaping which may cause strange hold offs. E. SLAB has complex NUMA policy layer support SLUB pushes NUMA policy handling into the page allocator. This means that allocation is coarser (SLUB does interleave on a page level) but that situation was also present before 2.6.13. SLABs application of policies to individual slab objects allocated in SLAB is certainly a performance concern due to the frequent references to memory policies which may lead a sequence of objects to come from one node after another. SLUB will get a slab full of objects from one node and then will switch to the next. F. Reduction of the size of partial slab lists SLAB has per node partial lists. This means that over time a large number of partial slabs may accumulate on those lists. These can only be reused if allocator occur on specific nodes. SLUB has a global pool of partial slabs and will consume slabs from that pool to decrease fragmentation. G. Tunables SLAB has sophisticated tuning abilities for each slab cache. One can manipulate the queue sizes in detail. However, filling the queues still requires the uses of the spin lock to check out slabs. SLUB has a global parameter (min_slab_order) for tuning. Increasing the minimum slab order can decrease the locking overhead. The bigger the slab order the less motions of pages between per CPU and partial lists occur and the better SLUB will be scaling. G. Slab merging We often have slab caches with similar parameters. SLUB detects those on boot up and merges them into the corresponding general caches. This leads to more effective memory use. About 50% of all caches can be eliminated through slab merging. This will also decrease slab fragmentation because partial allocated slabs can be filled up again. Slab merging can be switched off by specifying slub_nomerge on boot up. Note that merging can expose heretofore unknown bugs in the kernel because corrupted objects may now be placed differently and corrupt differing neighboring objects. Enable sanity checks to find those. H. Diagnostics The current slab diagnostics are difficult to use and require a recompilation of the kernel. SLUB contains debugging code that is always available (but is kept out of the hot code paths). SLUB diagnostics can be enabled via the "slab_debug" option. Parameters can be specified to select a single or a group of slab caches for diagnostics. This means that the system is running with the usual performance and it is much more likely that race conditions can be reproduced. I. Resiliency If basic sanity checks are on then SLUB is capable of detecting common error conditions and recover as best as possible to allow the system to continue. J. Tracing Tracing can be enabled via the slab_debug=T,<slabcache> option during boot. SLUB will then protocol all actions on that slabcache and dump the object contents on free. K. On demand DMA cache creation. Generally DMA caches are not needed. If a kmalloc is used with __GFP_DMA then just create this single slabcache that is needed. For systems that have no ZONE_DMA requirement the support is completely eliminated. L. Performance increase Some benchmarks have shown speed improvements on kernbench in the range of 5-10%. The locking overhead of slub is based on the underlying base allocation size. If we can reliably allocate larger order pages then it is possible to increase slub performance much further. The anti-fragmentation patches may enable further performance increases. Tested on: i386 UP + SMP, x86_64 UP + SMP + NUMA emulation, IA64 NUMA + Simulator SLUB Boot options slub_nomerge Disable merging of slabs slub_min_order=x Require a minimum order for slab caches. This increases the managed chunk size and therefore reduces meta data and locking overhead. slub_min_objects=x Mininum objects per slab. Default is 8. slub_max_order=x Avoid generating slabs larger than order specified. slub_debug Enable all diagnostics for all caches slub_debug=<options> Enable selective options for all caches slub_debug=<o>,<cache> Enable selective options for a certain set of caches Available Debug options F Double Free checking, sanity and resiliency R Red zoning P Object / padding poisoning U Track last free / alloc T Trace all allocs / frees (only use for individual slabs). To use SLUB: Apply this patch and then select SLUB as the default slab allocator. [hugh@veritas.com: fix an oops-causing locking error] [akpm@linux-foundation.org: various stupid cleanups and small fixes] Signed-off-by: Christoph Lameter <clameter@sgi.com> Signed-off-by: Hugh Dickins <hugh@veritas.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2007-05-06 14:49:36 -07:00
#ifdef CONFIG_ZONE_DMA
&cache_dma_attr.attr,
#endif
#ifdef CONFIG_NUMA
&defrag_ratio_attr.attr,
#endif
NULL
};
static struct attribute_group slab_attr_group = {
.attrs = slab_attrs,
};
static ssize_t slab_attr_show(struct kobject *kobj,
struct attribute *attr,
char *buf)
{
struct slab_attribute *attribute;
struct kmem_cache *s;
int err;
attribute = to_slab_attr(attr);
s = to_slab(kobj);
if (!attribute->show)
return -EIO;
err = attribute->show(s, buf);
return err;
}
static ssize_t slab_attr_store(struct kobject *kobj,
struct attribute *attr,
const char *buf, size_t len)
{
struct slab_attribute *attribute;
struct kmem_cache *s;
int err;
attribute = to_slab_attr(attr);
s = to_slab(kobj);
if (!attribute->store)
return -EIO;
err = attribute->store(s, buf, len);
return err;
}
static struct sysfs_ops slab_sysfs_ops = {
.show = slab_attr_show,
.store = slab_attr_store,
};
static struct kobj_type slab_ktype = {
.sysfs_ops = &slab_sysfs_ops,
};
static int uevent_filter(struct kset *kset, struct kobject *kobj)
{
struct kobj_type *ktype = get_ktype(kobj);
if (ktype == &slab_ktype)
return 1;
return 0;
}
static struct kset_uevent_ops slab_uevent_ops = {
.filter = uevent_filter,
};
decl_subsys(slab, &slab_ktype, &slab_uevent_ops);
#define ID_STR_LENGTH 64
/* Create a unique string id for a slab cache:
* format
* :[flags-]size:[memory address of kmemcache]
*/
static char *create_unique_id(struct kmem_cache *s)
{
char *name = kmalloc(ID_STR_LENGTH, GFP_KERNEL);
char *p = name;
BUG_ON(!name);
*p++ = ':';
/*
* First flags affecting slabcache operations. We will only
* get here for aliasable slabs so we do not need to support
* too many flags. The flags here must cover all flags that
* are matched during merging to guarantee that the id is
* unique.
*/
if (s->flags & SLAB_CACHE_DMA)
*p++ = 'd';
if (s->flags & SLAB_RECLAIM_ACCOUNT)
*p++ = 'a';
if (s->flags & SLAB_DEBUG_FREE)
*p++ = 'F';
if (p != name + 1)
*p++ = '-';
p += sprintf(p, "%07d", s->size);
BUG_ON(p > name + ID_STR_LENGTH - 1);
return name;
}
static int sysfs_slab_add(struct kmem_cache *s)
{
int err;
const char *name;
int unmergeable;
if (slab_state < SYSFS)
/* Defer until later */
return 0;
unmergeable = slab_unmergeable(s);
if (unmergeable) {
/*
* Slabcache can never be merged so we can use the name proper.
* This is typically the case for debug situations. In that
* case we can catch duplicate names easily.
*/
Fix up SLUB compile The newly merged SLUB allocator patches had been generated before the removal of "struct subsystem", and ended up applying fine, but wouldn't build based on the current tree as a result. Fix up that merge error - not that SLUB is likely really ready for showtime yet, but at least I can fix the trivial stuff. Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2007-05-07 12:31:58 -07:00
sysfs_remove_link(&slab_subsys.kobj, s->name);
SLUB core This is a new slab allocator which was motivated by the complexity of the existing code in mm/slab.c. It attempts to address a variety of concerns with the existing implementation. A. Management of object queues A particular concern was the complex management of the numerous object queues in SLAB. SLUB has no such queues. Instead we dedicate a slab for each allocating CPU and use objects from a slab directly instead of queueing them up. B. Storage overhead of object queues SLAB Object queues exist per node, per CPU. The alien cache queue even has a queue array that contain a queue for each processor on each node. For very large systems the number of queues and the number of objects that may be caught in those queues grows exponentially. On our systems with 1k nodes / processors we have several gigabytes just tied up for storing references to objects for those queues This does not include the objects that could be on those queues. One fears that the whole memory of the machine could one day be consumed by those queues. C. SLAB meta data overhead SLAB has overhead at the beginning of each slab. This means that data cannot be naturally aligned at the beginning of a slab block. SLUB keeps all meta data in the corresponding page_struct. Objects can be naturally aligned in the slab. F.e. a 128 byte object will be aligned at 128 byte boundaries and can fit tightly into a 4k page with no bytes left over. SLAB cannot do this. D. SLAB has a complex cache reaper SLUB does not need a cache reaper for UP systems. On SMP systems the per CPU slab may be pushed back into partial list but that operation is simple and does not require an iteration over a list of objects. SLAB expires per CPU, shared and alien object queues during cache reaping which may cause strange hold offs. E. SLAB has complex NUMA policy layer support SLUB pushes NUMA policy handling into the page allocator. This means that allocation is coarser (SLUB does interleave on a page level) but that situation was also present before 2.6.13. SLABs application of policies to individual slab objects allocated in SLAB is certainly a performance concern due to the frequent references to memory policies which may lead a sequence of objects to come from one node after another. SLUB will get a slab full of objects from one node and then will switch to the next. F. Reduction of the size of partial slab lists SLAB has per node partial lists. This means that over time a large number of partial slabs may accumulate on those lists. These can only be reused if allocator occur on specific nodes. SLUB has a global pool of partial slabs and will consume slabs from that pool to decrease fragmentation. G. Tunables SLAB has sophisticated tuning abilities for each slab cache. One can manipulate the queue sizes in detail. However, filling the queues still requires the uses of the spin lock to check out slabs. SLUB has a global parameter (min_slab_order) for tuning. Increasing the minimum slab order can decrease the locking overhead. The bigger the slab order the less motions of pages between per CPU and partial lists occur and the better SLUB will be scaling. G. Slab merging We often have slab caches with similar parameters. SLUB detects those on boot up and merges them into the corresponding general caches. This leads to more effective memory use. About 50% of all caches can be eliminated through slab merging. This will also decrease slab fragmentation because partial allocated slabs can be filled up again. Slab merging can be switched off by specifying slub_nomerge on boot up. Note that merging can expose heretofore unknown bugs in the kernel because corrupted objects may now be placed differently and corrupt differing neighboring objects. Enable sanity checks to find those. H. Diagnostics The current slab diagnostics are difficult to use and require a recompilation of the kernel. SLUB contains debugging code that is always available (but is kept out of the hot code paths). SLUB diagnostics can be enabled via the "slab_debug" option. Parameters can be specified to select a single or a group of slab caches for diagnostics. This means that the system is running with the usual performance and it is much more likely that race conditions can be reproduced. I. Resiliency If basic sanity checks are on then SLUB is capable of detecting common error conditions and recover as best as possible to allow the system to continue. J. Tracing Tracing can be enabled via the slab_debug=T,<slabcache> option during boot. SLUB will then protocol all actions on that slabcache and dump the object contents on free. K. On demand DMA cache creation. Generally DMA caches are not needed. If a kmalloc is used with __GFP_DMA then just create this single slabcache that is needed. For systems that have no ZONE_DMA requirement the support is completely eliminated. L. Performance increase Some benchmarks have shown speed improvements on kernbench in the range of 5-10%. The locking overhead of slub is based on the underlying base allocation size. If we can reliably allocate larger order pages then it is possible to increase slub performance much further. The anti-fragmentation patches may enable further performance increases. Tested on: i386 UP + SMP, x86_64 UP + SMP + NUMA emulation, IA64 NUMA + Simulator SLUB Boot options slub_nomerge Disable merging of slabs slub_min_order=x Require a minimum order for slab caches. This increases the managed chunk size and therefore reduces meta data and locking overhead. slub_min_objects=x Mininum objects per slab. Default is 8. slub_max_order=x Avoid generating slabs larger than order specified. slub_debug Enable all diagnostics for all caches slub_debug=<options> Enable selective options for all caches slub_debug=<o>,<cache> Enable selective options for a certain set of caches Available Debug options F Double Free checking, sanity and resiliency R Red zoning P Object / padding poisoning U Track last free / alloc T Trace all allocs / frees (only use for individual slabs). To use SLUB: Apply this patch and then select SLUB as the default slab allocator. [hugh@veritas.com: fix an oops-causing locking error] [akpm@linux-foundation.org: various stupid cleanups and small fixes] Signed-off-by: Christoph Lameter <clameter@sgi.com> Signed-off-by: Hugh Dickins <hugh@veritas.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2007-05-06 14:49:36 -07:00
name = s->name;
} else {
/*
* Create a unique name for the slab as a target
* for the symlinks.
*/
name = create_unique_id(s);
}
kobj_set_kset_s(s, slab_subsys);
kobject_set_name(&s->kobj, name);
kobject_init(&s->kobj);
err = kobject_add(&s->kobj);
if (err)
return err;
err = sysfs_create_group(&s->kobj, &slab_attr_group);
if (err)
return err;
kobject_uevent(&s->kobj, KOBJ_ADD);
if (!unmergeable) {
/* Setup first alias */
sysfs_slab_alias(s, s->name);
kfree(name);
}
return 0;
}
static void sysfs_slab_remove(struct kmem_cache *s)
{
kobject_uevent(&s->kobj, KOBJ_REMOVE);
kobject_del(&s->kobj);
}
/*
* Need to buffer aliases during bootup until sysfs becomes
* available lest we loose that information.
*/
struct saved_alias {
struct kmem_cache *s;
const char *name;
struct saved_alias *next;
};
struct saved_alias *alias_list;
static int sysfs_slab_alias(struct kmem_cache *s, const char *name)
{
struct saved_alias *al;
if (slab_state == SYSFS) {
/*
* If we have a leftover link then remove it.
*/
Fix up SLUB compile The newly merged SLUB allocator patches had been generated before the removal of "struct subsystem", and ended up applying fine, but wouldn't build based on the current tree as a result. Fix up that merge error - not that SLUB is likely really ready for showtime yet, but at least I can fix the trivial stuff. Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2007-05-07 12:31:58 -07:00
sysfs_remove_link(&slab_subsys.kobj, name);
return sysfs_create_link(&slab_subsys.kobj,
SLUB core This is a new slab allocator which was motivated by the complexity of the existing code in mm/slab.c. It attempts to address a variety of concerns with the existing implementation. A. Management of object queues A particular concern was the complex management of the numerous object queues in SLAB. SLUB has no such queues. Instead we dedicate a slab for each allocating CPU and use objects from a slab directly instead of queueing them up. B. Storage overhead of object queues SLAB Object queues exist per node, per CPU. The alien cache queue even has a queue array that contain a queue for each processor on each node. For very large systems the number of queues and the number of objects that may be caught in those queues grows exponentially. On our systems with 1k nodes / processors we have several gigabytes just tied up for storing references to objects for those queues This does not include the objects that could be on those queues. One fears that the whole memory of the machine could one day be consumed by those queues. C. SLAB meta data overhead SLAB has overhead at the beginning of each slab. This means that data cannot be naturally aligned at the beginning of a slab block. SLUB keeps all meta data in the corresponding page_struct. Objects can be naturally aligned in the slab. F.e. a 128 byte object will be aligned at 128 byte boundaries and can fit tightly into a 4k page with no bytes left over. SLAB cannot do this. D. SLAB has a complex cache reaper SLUB does not need a cache reaper for UP systems. On SMP systems the per CPU slab may be pushed back into partial list but that operation is simple and does not require an iteration over a list of objects. SLAB expires per CPU, shared and alien object queues during cache reaping which may cause strange hold offs. E. SLAB has complex NUMA policy layer support SLUB pushes NUMA policy handling into the page allocator. This means that allocation is coarser (SLUB does interleave on a page level) but that situation was also present before 2.6.13. SLABs application of policies to individual slab objects allocated in SLAB is certainly a performance concern due to the frequent references to memory policies which may lead a sequence of objects to come from one node after another. SLUB will get a slab full of objects from one node and then will switch to the next. F. Reduction of the size of partial slab lists SLAB has per node partial lists. This means that over time a large number of partial slabs may accumulate on those lists. These can only be reused if allocator occur on specific nodes. SLUB has a global pool of partial slabs and will consume slabs from that pool to decrease fragmentation. G. Tunables SLAB has sophisticated tuning abilities for each slab cache. One can manipulate the queue sizes in detail. However, filling the queues still requires the uses of the spin lock to check out slabs. SLUB has a global parameter (min_slab_order) for tuning. Increasing the minimum slab order can decrease the locking overhead. The bigger the slab order the less motions of pages between per CPU and partial lists occur and the better SLUB will be scaling. G. Slab merging We often have slab caches with similar parameters. SLUB detects those on boot up and merges them into the corresponding general caches. This leads to more effective memory use. About 50% of all caches can be eliminated through slab merging. This will also decrease slab fragmentation because partial allocated slabs can be filled up again. Slab merging can be switched off by specifying slub_nomerge on boot up. Note that merging can expose heretofore unknown bugs in the kernel because corrupted objects may now be placed differently and corrupt differing neighboring objects. Enable sanity checks to find those. H. Diagnostics The current slab diagnostics are difficult to use and require a recompilation of the kernel. SLUB contains debugging code that is always available (but is kept out of the hot code paths). SLUB diagnostics can be enabled via the "slab_debug" option. Parameters can be specified to select a single or a group of slab caches for diagnostics. This means that the system is running with the usual performance and it is much more likely that race conditions can be reproduced. I. Resiliency If basic sanity checks are on then SLUB is capable of detecting common error conditions and recover as best as possible to allow the system to continue. J. Tracing Tracing can be enabled via the slab_debug=T,<slabcache> option during boot. SLUB will then protocol all actions on that slabcache and dump the object contents on free. K. On demand DMA cache creation. Generally DMA caches are not needed. If a kmalloc is used with __GFP_DMA then just create this single slabcache that is needed. For systems that have no ZONE_DMA requirement the support is completely eliminated. L. Performance increase Some benchmarks have shown speed improvements on kernbench in the range of 5-10%. The locking overhead of slub is based on the underlying base allocation size. If we can reliably allocate larger order pages then it is possible to increase slub performance much further. The anti-fragmentation patches may enable further performance increases. Tested on: i386 UP + SMP, x86_64 UP + SMP + NUMA emulation, IA64 NUMA + Simulator SLUB Boot options slub_nomerge Disable merging of slabs slub_min_order=x Require a minimum order for slab caches. This increases the managed chunk size and therefore reduces meta data and locking overhead. slub_min_objects=x Mininum objects per slab. Default is 8. slub_max_order=x Avoid generating slabs larger than order specified. slub_debug Enable all diagnostics for all caches slub_debug=<options> Enable selective options for all caches slub_debug=<o>,<cache> Enable selective options for a certain set of caches Available Debug options F Double Free checking, sanity and resiliency R Red zoning P Object / padding poisoning U Track last free / alloc T Trace all allocs / frees (only use for individual slabs). To use SLUB: Apply this patch and then select SLUB as the default slab allocator. [hugh@veritas.com: fix an oops-causing locking error] [akpm@linux-foundation.org: various stupid cleanups and small fixes] Signed-off-by: Christoph Lameter <clameter@sgi.com> Signed-off-by: Hugh Dickins <hugh@veritas.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2007-05-06 14:49:36 -07:00
&s->kobj, name);
}
al = kmalloc(sizeof(struct saved_alias), GFP_KERNEL);
if (!al)
return -ENOMEM;
al->s = s;
al->name = name;
al->next = alias_list;
alias_list = al;
return 0;
}
static int __init slab_sysfs_init(void)
{
int err;
err = subsystem_register(&slab_subsys);
if (err) {
printk(KERN_ERR "Cannot register slab subsystem.\n");
return -ENOSYS;
}
finish_bootstrap();
while (alias_list) {
struct saved_alias *al = alias_list;
alias_list = alias_list->next;
err = sysfs_slab_alias(al->s, al->name);
BUG_ON(err);
kfree(al);
}
resiliency_test();
return 0;
}
__initcall(slab_sysfs_init);
#else
__initcall(finish_bootstrap);
#endif
Copy Permalink