This a complete rewrite of the oom killer's badness() heuristic which is
used to determine which task to kill in oom conditions. The goal is to
make it as simple and predictable as possible so the results are better
understood and we end up killing the task which will lead to the most
memory freeing while still respecting the fine-tuning from userspace.
Instead of basing the heuristic on mm->total_vm for each task, the task's
rss and swap space is used instead. This is a better indication of the
amount of memory that will be freeable if the oom killed task is chosen
and subsequently exits. This helps specifically in cases where KDE or
GNOME is chosen for oom kill on desktop systems instead of a memory
hogging task.
The baseline for the heuristic is a proportion of memory that each task is
currently using in memory plus swap compared to the amount of "allowable"
memory. "Allowable," in this sense, means the system-wide resources for
unconstrained oom conditions, the set of mempolicy nodes, the mems
attached to current's cpuset, or a memory controller's limit. The
proportion is given on a scale of 0 (never kill) to 1000 (always kill),
roughly meaning that if a task has a badness() score of 500 that the task
consumes approximately 50% of allowable memory resident in RAM or in swap
space.
The proportion is always relative to the amount of "allowable" memory and
not the total amount of RAM systemwide so that mempolicies and cpusets may
operate in isolation; they shall not need to know the true size of the
machine on which they are running if they are bound to a specific set of
nodes or mems, respectively.
Root tasks are given 3% extra memory just like __vm_enough_memory()
provides in LSMs. In the event of two tasks consuming similar amounts of
memory, it is generally better to save root's task.
Because of the change in the badness() heuristic's baseline, it is also
necessary to introduce a new user interface to tune it. It's not possible
to redefine the meaning of /proc/pid/oom_adj with a new scale since the
ABI cannot be changed for backward compatability. Instead, a new tunable,
/proc/pid/oom_score_adj, is added that ranges from -1000 to +1000. It may
be used to polarize the heuristic such that certain tasks are never
considered for oom kill while others may always be considered. The value
is added directly into the badness() score so a value of -500, for
example, means to discount 50% of its memory consumption in comparison to
other tasks either on the system, bound to the mempolicy, in the cpuset,
or sharing the same memory controller.
/proc/pid/oom_adj is changed so that its meaning is rescaled into the
units used by /proc/pid/oom_score_adj, and vice versa. Changing one of
these per-task tunables will rescale the value of the other to an
equivalent meaning. Although /proc/pid/oom_adj was originally defined as
a bitshift on the badness score, it now shares the same linear growth as
/proc/pid/oom_score_adj but with different granularity. This is required
so the ABI is not broken with userspace applications and allows oom_adj to
be deprecated for future removal.
Signed-off-by: David Rientjes <rientjes@google.com>
Cc: Nick Piggin <npiggin@suse.de>
Cc: KAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com>
Cc: KOSAKI Motohiro <kosaki.motohiro@jp.fujitsu.com>
Cc: Oleg Nesterov <oleg@redhat.com>
Cc: Balbir Singh <balbir@in.ibm.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
Concurrency managed workqueue needs to know when workers are going to
sleep and waking up. Using these two hooks, cmwq keeps track of the
current concurrency level and throttles execution of new works if it's
too high and wakes up another worker from the sleep hook if it becomes
too low.
This patch introduces PF_WQ_WORKER to identify workqueue workers and
adds the following two hooks.
* wq_worker_waking_up(): called when a worker is woken up.
* wq_worker_sleeping(): called when a worker is going to sleep and may
return a pointer to a local task which should be woken up. The
returned task is woken up using try_to_wake_up_local() which is
simplified ttwu which is called under rq lock and can only wake up
local tasks.
Both hooks are currently defined as noop in kernel/workqueue_sched.h.
Later cmwq implementation will replace them with proper
implementation.
These hooks are hard coded as they'll always be enabled.
Signed-off-by: Tejun Heo <tj@kernel.org>
Acked-by: Peter Zijlstra <peterz@infradead.org>
Cc: Mike Galbraith <efault@gmx.de>
Cc: Ingo Molnar <mingo@elte.hu>
This reverts commit 0ac0c0d0f8, which
caused cross-architecture build problems for all the wrong reasons.
IA64 already added its own version of __node_random(), but the fact is,
there is nothing architectural about the function, and the original
commit was just badly done. Revert it, since no fix is forthcoming.
Requested-by: Stephen Rothwell <sfr@canb.auug.org.au>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
copy_process(pid => &init_struct_pid) doesn't do attach_pid/etc.
It shouldn't, but this means that the idle threads run with the wrong
pids copied from the caller's task_struct. In x86 case the caller is
either kernel_init() thread or keventd.
In particular, this means that after the series of cpu_up/cpu_down an
idle thread (which never exits) can run with .pid pointing to nowhere.
Change fork_idle() to initialize idle->pids[] correctly. We only set
.pid = &init_struct_pid but do not add .node to list, INIT_TASK() does
the same for the boot-cpu idle thread (swapper).
Signed-off-by: Oleg Nesterov <oleg@redhat.com>
Cc: Cedric Le Goater <clg@fr.ibm.com>
Cc: Dave Hansen <haveblue@us.ibm.com>
Cc: Eric Biederman <ebiederm@xmission.com>
Cc: Herbert Poetzl <herbert@13thfloor.at>
Cc: Mathias Krause <Mathias.Krause@secunet.com>
Acked-by: Roland McGrath <roland@redhat.com>
Acked-by: Serge Hallyn <serue@us.ibm.com>
Cc: Sukadev Bhattiprolu <sukadev@us.ibm.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
No functional changes, just s/atomic_t count/int nr_threads/.
With the recent changes this counter has a single user, get_nr_threads()
And, none of its callers need the really accurate number of threads, not
to mention each caller obviously races with fork/exit. It is only used to
report this value to the user-space, except first_tid() uses it to avoid
the unnecessary while_each_thread() loop in the unlikely case.
It is a bit sad we need a word in struct signal_struct for this, perhaps
we can change get_nr_threads() to approximate the number of threads using
signal->live and kill ->nr_threads later.
[akpm@linux-foundation.org: coding-style fixes]
Signed-off-by: Oleg Nesterov <oleg@redhat.com>
Cc: Alexey Dobriyan <adobriyan@gmail.com>
Cc: "Eric W. Biederman" <ebiederm@xmission.com>
Acked-by: Roland McGrath <roland@redhat.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
check_unshare_flags(CLONE_SIGHAND) adds CLONE_THREAD to *flags_ptr if the
task is multithreaded to ensure unshare_thread() will fail.
Not only this is a bit strange way to return the error, this is absolutely
meaningless. If signal->count > 1 then sighand->count must be also > 1,
and unshare_sighand() will fail anyway.
In fact, all CLONE_THREAD/SIGHAND/VM checks inside sys_unshare() do not
look right. Fortunately this code doesn't really work anyway.
Signed-off-by: Oleg Nesterov <oleg@redhat.com>
Cc: Balbir Singh <balbir@linux.vnet.ibm.com>
Acked-by: Roland McGrath <roland@redhat.com>
Cc: Veaceslav Falico <vfalico@redhat.com>
Cc: Stanislaw Gruszka <sgruszka@redhat.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
Move taskstats_tgid_free() from __exit_signal() to free_signal_struct().
This way signal->stats never points to nowhere and we can read ->stats
lockless.
Signed-off-by: Oleg Nesterov <oleg@redhat.com>
Cc: Balbir Singh <balbir@linux.vnet.ibm.com>
Cc: Roland McGrath <roland@redhat.com>
Cc: Veaceslav Falico <vfalico@redhat.com>
Cc: Stanislaw Gruszka <sgruszka@redhat.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
Kill the empty thread_group_cputime_free() helper. It was needed to free
the per-cpu data which we no longer have.
Signed-off-by: Oleg Nesterov <oleg@redhat.com>
Cc: Balbir Singh <balbir@linux.vnet.ibm.com>
Cc: Roland McGrath <roland@redhat.com>
Cc: Veaceslav Falico <vfalico@redhat.com>
Cc: Stanislaw Gruszka <sgruszka@redhat.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
We have a lot of problems with accessing task_struct->signal, it can
"disappear" at any moment. Even current can't use its ->signal safely
after exit_notify(). ->siglock helps, but it is not convenient, not
always possible, and sometimes it makes sense to use task->signal even
after this task has already dead.
This patch adds the reference counter, sigcnt, into signal_struct. This
reference is owned by task_struct and it is dropped in
__put_task_struct(). Perhaps it makes sense to export
get/put_signal_struct() later, but currently I don't see the immediate
reason.
Rename __cleanup_signal() to free_signal_struct() and unexport it. With
the previous changes it does nothing except kmem_cache_free().
Change __exit_signal() to not clear/free ->signal, it will be freed when
the last reference to any thread in the thread group goes away.
Note:
- when the last thead exits signal->tty can point to nowhere, see
the next patch.
- with or without this patch signal_struct->count should go away,
or at least it should be "int nr_threads" for fs/proc. This will
be addressed later.
Signed-off-by: Oleg Nesterov <oleg@redhat.com>
Cc: Alan Cox <alan@linux.intel.com>
Cc: Ingo Molnar <mingo@elte.hu>
Cc: Peter Zijlstra <peterz@infradead.org>
Acked-by: Roland McGrath <roland@redhat.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
tty_kref_put() has two callsites in copy_process() paths,
1. if copy_process() suceeds it is called before we copy
signal->tty from parent
2. otherwise it is called from __cleanup_signal() under
bad_fork_cleanup_signal: label
In both cases tty_kref_put() is not right and unneeded because we don't
have the balancing tty_kref_get(). Fortunately, this is harmless because
this can only happen without CLONE_THREAD, and in this case signal->tty
must be NULL.
Remove tty_kref_put() from copy_process() and __cleanup_signal(), and
change another caller of __cleanup_signal(), __exit_signal(), to call
tty_kref_put() by hand.
I hope this change makes sense by itself, but it is also needed to make
->signal refcountable.
Signed-off-by: Oleg Nesterov <oleg@redhat.com>
Acked-by: Alan Cox <alan@linux.intel.com>
Acked-by: Roland McGrath <roland@redhat.com>
Cc: Greg KH <greg@kroah.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
Some workloads that create a large number of small files tend to assign
too many pages to node 0 (multi-node systems). Part of the reason is that
the rotor (in cpuset_mem_spread_node()) used to assign nodes starts at
node 0 for newly created tasks.
This patch changes the rotor to be initialized to a random node number of
the cpuset.
[akpm@linux-foundation.org: fix layout]
[Lee.Schermerhorn@hp.com: Define stub numa_random() for !NUMA configuration]
Signed-off-by: Jack Steiner <steiner@sgi.com>
Signed-off-by: Lee Schermerhorn <lee.schermerhorn@hp.com>
Cc: Christoph Lameter <cl@linux-foundation.org>
Cc: Pekka Enberg <penberg@cs.helsinki.fi>
Cc: Paul Menage <menage@google.com>
Cc: Jack Steiner <steiner@sgi.com>
Cc: Robin Holt <holt@sgi.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
* 'perf-core-for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/tip/linux-2.6-tip: (311 commits)
perf tools: Add mode to build without newt support
perf symbols: symbol inconsistency message should be done only at verbose=1
perf tui: Add explicit -lslang option
perf options: Type check all the remaining OPT_ variants
perf options: Type check OPT_BOOLEAN and fix the offenders
perf options: Check v type in OPT_U?INTEGER
perf options: Introduce OPT_UINTEGER
perf tui: Add workaround for slang < 2.1.4
perf record: Fix bug mismatch with -c option definition
perf options: Introduce OPT_U64
perf tui: Add help window to show key associations
perf tui: Make <- exit menus too
perf newt: Add single key shortcuts for zoom into DSO and threads
perf newt: Exit browser unconditionally when CTRL+C, q or Q is pressed
perf newt: Fix the 'A'/'a' shortcut for annotate
perf newt: Make <- exit the ui_browser
x86, perf: P4 PMU - fix counters management logic
perf newt: Make <- zoom out filters
perf report: Report number of events, not samples
perf hist: Clarify events_stats fields usage
...
Fix up trivial conflicts in kernel/fork.c and tools/perf/builtin-record.c
Originally, commit d899bf7b ("procfs: provide stack information for
threads") attempted to introduce a new feature for showing where the
threadstack was located and how many pages are being utilized by the
stack.
Commit c44972f1 ("procfs: disable per-task stack usage on NOMMU") was
applied to fix the NO_MMU case.
Commit 89240ba0 ("x86, fs: Fix x86 procfs stack information for threads on
64-bit") was applied to fix a bug in ia32 executables being loaded.
Commit 9ebd4eba7 ("procfs: fix /proc/<pid>/stat stack pointer for kernel
threads") was applied to fix a bug which had kernel threads printing a
userland stack address.
Commit 1306d603f ('proc: partially revert "procfs: provide stack
information for threads"') was then applied to revert the stack pages
being used to solve a significant performance regression.
This patch nearly undoes the effect of all these patches.
The reason for reverting these is it provides an unusable value in
field 28. For x86_64, a fork will result in the task->stack_start
value being updated to the current user top of stack and not the stack
start address. This unpredictability of the stack_start value makes
it worthless. That includes the intended use of showing how much stack
space a thread has.
Other architectures will get different values. As an example, ia64
gets 0. The do_fork() and copy_process() functions appear to treat the
stack_start and stack_size parameters as architecture specific.
I only partially reverted c44972f1 ("procfs: disable per-task stack usage
on NOMMU") . If I had completely reverted it, I would have had to change
mm/Makefile only build pagewalk.o when CONFIG_PROC_PAGE_MONITOR is
configured. Since I could not test the builds without significant effort,
I decided to not change mm/Makefile.
I only partially reverted 89240ba0 ("x86, fs: Fix x86 procfs stack
information for threads on 64-bit") . I left the KSTK_ESP() change in
place as that seemed worthwhile.
Signed-off-by: Robin Holt <holt@sgi.com>
Cc: Stefani Seibold <stefani@seibold.net>
Cc: KOSAKI Motohiro <kosaki.motohiro@jp.fujitsu.com>
Cc: Michal Simek <monstr@monstr.eu>
Cc: Ingo Molnar <mingo@elte.hu>
Cc: <stable@kernel.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
- We weren't zeroing p->rss_stat[] at fork()
- Consequently sync_mm_rss() was dereferencing tsk->mm for kernel
threads and was oopsing.
- Make __sync_task_rss_stat() static, too.
Addresses https://bugzilla.kernel.org/show_bug.cgi?id=15648
[akpm@linux-foundation.org: remove the BUG_ON(!mm->rss)]
Reported-by: Troels Liebe Bentsen <tlb@rapanden.dk>
Signed-off-by: KAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com>
"Michael S. Tsirkin" <mst@redhat.com>
Cc: Andrea Arcangeli <aarcange@redhat.com>
Cc: Rik van Riel <riel@redhat.com>
Cc: Minchan Kim <minchan.kim@gmail.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
Support for the PMU's BTS features has been upstreamed in
v2.6.32, but we still have the old and disabled ptrace-BTS,
as Linus noticed it not so long ago.
It's buggy: TIF_DEBUGCTLMSR is trampling all over that MSR without
regard for other uses (perf) and doesn't provide the flexibility
needed for perf either.
Its users are ptrace-block-step and ptrace-bts, since ptrace-bts
was never used and ptrace-block-step can be implemented using a
much simpler approach.
So axe all 3000 lines of it. That includes the *locked_memory*()
APIs in mm/mlock.c as well.
Reported-by: Linus Torvalds <torvalds@linux-foundation.org>
Signed-off-by: Peter Zijlstra <a.p.zijlstra@chello.nl>
Cc: Roland McGrath <roland@redhat.com>
Cc: Oleg Nesterov <oleg@redhat.com>
Cc: Markus Metzger <markus.t.metzger@intel.com>
Cc: Steven Rostedt <rostedt@goodmis.org>
Cc: Andrew Morton <akpm@linux-foundation.org>
LKML-Reference: <20100325135413.938004390@chello.nl>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
* 'core-fixes-for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/tip/linux-2.6-tip:
locking: Make sparse work with inline spinlocks and rwlocks
x86/mce: Fix RCU lockdep splats
rcu: Increase RCU CPU stall timeouts if PROVE_RCU
ftrace: Replace read_barrier_depends() with rcu_dereference_raw()
rcu: Suppress RCU lockdep warnings during early boot
rcu, ftrace: Fix RCU lockdep splat in ftrace_perf_buf_prepare()
rcu: Suppress __mpol_dup() false positive from RCU lockdep
rcu: Make rcu_read_lock_sched_held() handle !PREEMPT
rcu: Add control variables to lockdep_rcu_dereference() diagnostics
rcu, cgroup: Relax the check in task_subsys_state() as early boot is now handled by lockdep-RCU
rcu: Use wrapper function instead of exporting tasklist_lock
sched, rcu: Fix rcu_dereference() for RCU-lockdep
rcu: Make task_subsys_state() RCU-lockdep checks handle boot-time use
rcu: Fix holdoff for accelerated GPs for last non-dynticked CPU
x86/gart: Unexport gart_iommu_aperture
Fix trivial conflicts in kernel/trace/ftrace.c
Remove unneeded initializations in thread_group_cputime_init() and in
posix_cpu_timers_init_group(). They are useless after kmem_cache_zalloc()
was used in copy_signal().
Signed-off-by: Veaceslav Falico <vfalico@redhat.com>
Acked-by: Oleg Nesterov <oleg@redhat.com>
Cc: Roland McGrath <roland@redhat.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
Use kmem_cache_zalloc() on signal creation and remove unneeded
initialization lines in copy_signal().
Signed-off-by: Veaceslav Falico <vfalico@redhat.com>
Acked-by: Oleg Nesterov <oleg@redhat.com>
Cc: Roland McGrath <roland@redhat.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
Make sure compiler won't do weird things with limits. E.g. fetching them
twice may return 2 different values after writable limits are implemented.
I.e. either use rlimit helpers added in commit 3e10e716ab ("resource:
add helpers for fetching rlimits") or ACCESS_ONCE if not applicable.
Signed-off-by: Jiri Slaby <jslaby@suse.cz>
Cc: Ingo Molnar <mingo@elte.hu>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Thomas Gleixner <tglx@linutronix.de>
Cc: john stultz <johnstul@us.ibm.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
The old anon_vma code can lead to scalability issues with heavily forking
workloads. Specifically, each anon_vma will be shared between the parent
process and all its child processes.
In a workload with 1000 child processes and a VMA with 1000 anonymous
pages per process that get COWed, this leads to a system with a million
anonymous pages in the same anon_vma, each of which is mapped in just one
of the 1000 processes. However, the current rmap code needs to walk them
all, leading to O(N) scanning complexity for each page.
This can result in systems where one CPU is walking the page tables of
1000 processes in page_referenced_one, while all other CPUs are stuck on
the anon_vma lock. This leads to catastrophic failure for a benchmark
like AIM7, where the total number of processes can reach in the tens of
thousands. Real workloads are still a factor 10 less process intensive
than AIM7, but they are catching up.
This patch changes the way anon_vmas and VMAs are linked, which allows us
to associate multiple anon_vmas with a VMA. At fork time, each child
process gets its own anon_vmas, in which its COWed pages will be
instantiated. The parents' anon_vma is also linked to the VMA, because
non-COWed pages could be present in any of the children.
This reduces rmap scanning complexity to O(1) for the pages of the 1000
child processes, with O(N) complexity for at most 1/N pages in the system.
This reduces the average scanning cost in heavily forking workloads from
O(N) to 2.
The only real complexity in this patch stems from the fact that linking a
VMA to anon_vmas now involves memory allocations. This means vma_adjust
can fail, if it needs to attach a VMA to anon_vma structures. This in
turn means error handling needs to be added to the calling functions.
A second source of complexity is that, because there can be multiple
anon_vmas, the anon_vma linking in vma_adjust can no longer be done under
"the" anon_vma lock. To prevent the rmap code from walking up an
incomplete VMA, this patch introduces the VM_LOCK_RMAP VMA flag. This bit
flag uses the same slot as the NOMMU VM_MAPPED_COPY, with an ifdef in mm.h
to make sure it is impossible to compile a kernel that needs both symbolic
values for the same bitflag.
Some test results:
Without the anon_vma changes, when AIM7 hits around 9.7k users (on a test
box with 16GB RAM and not quite enough IO), the system ends up running
>99% in system time, with every CPU on the same anon_vma lock in the
pageout code.
With these changes, AIM7 hits the cross-over point around 29.7k users.
This happens with ~99% IO wait time, there never seems to be any spike in
system time. The anon_vma lock contention appears to be resolved.
[akpm@linux-foundation.org: cleanups]
Signed-off-by: Rik van Riel <riel@redhat.com>
Cc: KOSAKI Motohiro <kosaki.motohiro@jp.fujitsu.com>
Cc: Larry Woodman <lwoodman@redhat.com>
Cc: Lee Schermerhorn <Lee.Schermerhorn@hp.com>
Cc: Minchan Kim <minchan.kim@gmail.com>
Cc: Andrea Arcangeli <aarcange@redhat.com>
Cc: Hugh Dickins <hugh.dickins@tiscali.co.uk>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
Presently, per-mm statistics counter is defined by macro in sched.h
This patch modifies it to
- defined in mm.h as inlinf functions
- use array instead of macro's name creation.
This patch is for reducing patch size in future patch to modify
implementation of per-mm counter.
Signed-off-by: KAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com>
Reviewed-by: Minchan Kim <minchan.kim@gmail.com>
Cc: Christoph Lameter <cl@linux-foundation.org>
Cc: Lee Schermerhorn <lee.schermerhorn@hp.com>
Cc: David Rientjes <rientjes@google.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
There are a number of issues:
1) TASK_WAKING vs cgroup_clone (cpusets)
copy_process():
sched_fork()
child->state = TASK_WAKING; /* waiting for wake_up_new_task() */
if (current->nsproxy != p->nsproxy)
ns_cgroup_clone()
cgroup_clone()
mutex_lock(inode->i_mutex)
mutex_lock(cgroup_mutex)
cgroup_attach_task()
ss->can_attach()
ss->attach() [ -> cpuset_attach() ]
cpuset_attach_task()
set_cpus_allowed_ptr();
while (child->state == TASK_WAKING)
cpu_relax();
will deadlock the system.
2) cgroup_clone (cpusets) vs copy_process
So even if the above would work we still have:
copy_process():
if (current->nsproxy != p->nsproxy)
ns_cgroup_clone()
cgroup_clone()
mutex_lock(inode->i_mutex)
mutex_lock(cgroup_mutex)
cgroup_attach_task()
ss->can_attach()
ss->attach() [ -> cpuset_attach() ]
cpuset_attach_task()
set_cpus_allowed_ptr();
...
p->cpus_allowed = current->cpus_allowed
over-writing the modified cpus_allowed.
3) fork() vs hotplug
if we unplug the child's cpu after the sanity check when the child
gets attached to the task_list but before wake_up_new_task() shit
will meet with fan.
Solve all these issues by moving fork cpu selection into
wake_up_new_task().
Reported-by: Serge E. Hallyn <serue@us.ibm.com>
Tested-by: Serge E. Hallyn <serue@us.ibm.com>
Signed-off-by: Peter Zijlstra <a.p.zijlstra@chello.nl>
LKML-Reference: <1264106190.4283.1314.camel@laptop>
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
Thanks to Roland who pointed out de_thread() issues.
Currently we add sub-threads to ->real_parent->children list. This buys
nothing but slows down do_wait().
With this patch ->children contains only main threads (group leaders).
The only complication is that forget_original_parent() should iterate over
sub-threads by hand, and de_thread() needs another list_replace() when it
changes ->group_leader.
Henceforth do_wait_thread() can never see task_detached() && !EXIT_DEAD
tasks, we can remove this check (and we can unify do_wait_thread() and
ptrace_do_wait()).
This change can confuse the optimistic search in mm_update_next_owner(),
but this is fixable and minor.
Perhaps badness() and oom_kill_process() should be updated, but they
should be fixed in any case.
Signed-off-by: Oleg Nesterov <oleg@redhat.com>
Cc: Roland McGrath <roland@redhat.com>
Cc: Ingo Molnar <mingo@elte.hu>
Cc: Ratan Nalumasu <rnalumasu@gmail.com>
Cc: Vitaly Mayatskikh <vmayatsk@redhat.com>
Cc: David Rientjes <rientjes@google.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
If the tracee calls fork() after PTRACE_SINGLESTEP, the forked child
starts with TIF_SINGLESTEP/X86_EFLAGS_TF bits copied from ptraced parent.
This is not right, especially when the new child is not auto-attaced: in
this case it is killed by SIGTRAP.
Change copy_process() to call user_disable_single_step(). Tested on x86.
Test-case:
#include <stdio.h>
#include <unistd.h>
#include <signal.h>
#include <sys/ptrace.h>
#include <sys/wait.h>
#include <assert.h>
int main(void)
{
int pid, status;
if (!(pid = fork())) {
assert(ptrace(PTRACE_TRACEME) == 0);
kill(getpid(), SIGSTOP);
if (!fork()) {
/* kernel bug: this child will be killed by SIGTRAP */
printf("Hello world\n");
return 43;
}
wait(&status);
return WEXITSTATUS(status);
}
for (;;) {
assert(pid == wait(&status));
if (WIFEXITED(status))
break;
assert(ptrace(PTRACE_SINGLESTEP, pid, 0,0) == 0);
}
assert(WEXITSTATUS(status) == 43);
return 0;
}
Signed-off-by: Oleg Nesterov <oleg@redhat.com>
Acked-by: Roland McGrath <roland@redhat.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
In massive parallel enviroment, res_counter can be a performance
bottleneck. One strong techinque to reduce lock contention is reducing
calls by coalescing some amount of calls into one.
Considering charge/uncharge chatacteristic,
- charge is done one by one via demand-paging.
- uncharge is done by
- in chunk at munmap, truncate, exit, execve...
- one by one via vmscan/paging.
It seems we have a chance to coalesce uncharges for improving scalability
at unmap/truncation.
This patch is a for coalescing uncharge. For avoiding scattering memcg's
structure to functions under /mm, this patch adds memcg batch uncharge
information to the task. A reason for per-task batching is for making use
of caller's context information. We do batched uncharge (deleyed
uncharge) when truncation/unmap occurs but do direct uncharge when
uncharge is called by memory reclaim (vmscan.c).
The degree of coalescing depends on callers
- at invalidate/trucate... pagevec size
- at unmap ....ZAP_BLOCK_SIZE
(memory itself will be freed in this degree.)
Then, we'll not coalescing too much.
On x86-64 8cpu server, I tested overheads of memcg at page fault by
running a program which does map/fault/unmap in a loop. Running
a task per a cpu by taskset and see sum of the number of page faults
in 60secs.
[without memcg config]
40156968 page-faults # 0.085 M/sec ( +- 0.046% )
27.67 cache-miss/faults
[root cgroup]
36659599 page-faults # 0.077 M/sec ( +- 0.247% )
31.58 miss/faults
[in a child cgroup]
18444157 page-faults # 0.039 M/sec ( +- 0.133% )
69.96 miss/faults
[child with this patch]
27133719 page-faults # 0.057 M/sec ( +- 0.155% )
47.16 miss/faults
We can see some amounts of improvement.
(root cgroup doesn't affected by this patch)
Another patch for "charge" will follow this and above will be improved more.
Changelog(since 2009/10/02):
- renamed filed of memcg_batch (as pages to bytes, memsw to memsw_bytes)
- some clean up and commentary/description updates.
- added initialize code to copy_process(). (possible bug fix)
Changelog(old):
- fixed !CONFIG_MEM_CGROUP case.
- rebased onto the latest mmotm + softlimit fix patches.
- unified patch for callers
- added commetns.
- make ->do_batch as bool.
- removed css_get() at el. We don't need it.
Signed-off-by: KAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com>
Cc: Balbir Singh <balbir@in.ibm.com>
Cc: Daisuke Nishimura <nishimura@mxp.nes.nec.co.jp>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
Convert locks which cannot be sleeping locks in preempt-rt to
raw_spinlocks.
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
Acked-by: Peter Zijlstra <peterz@infradead.org>
Acked-by: Ingo Molnar <mingo@elte.hu>
* 'for-2.6.33' of git://git.kernel.dk/linux-2.6-block: (113 commits)
cfq-iosched: Do not access cfqq after freeing it
block: include linux/err.h to use ERR_PTR
cfq-iosched: use call_rcu() instead of doing grace period stall on queue exit
blkio: Allow CFQ group IO scheduling even when CFQ is a module
blkio: Implement dynamic io controlling policy registration
blkio: Export some symbols from blkio as its user CFQ can be a module
block: Fix io_context leak after failure of clone with CLONE_IO
block: Fix io_context leak after clone with CLONE_IO
cfq-iosched: make nonrot check logic consistent
io controller: quick fix for blk-cgroup and modular CFQ
cfq-iosched: move IO controller declerations to a header file
cfq-iosched: fix compile problem with !CONFIG_CGROUP
blkio: Documentation
blkio: Wait on sync-noidle queue even if rq_noidle = 1
blkio: Implement group_isolation tunable
blkio: Determine async workload length based on total number of queues
blkio: Wait for cfq queue to get backlogged if group is empty
blkio: Propagate cgroup weight updation to cfq groups
blkio: Drop the reference to queue once the task changes cgroup
blkio: Provide some isolation between groups
...
* 'kvm-updates/2.6.33' of git://git.kernel.org/pub/scm/virt/kvm/kvm: (84 commits)
KVM: VMX: Fix comparison of guest efer with stale host value
KVM: s390: Fix prefix register checking in arch/s390/kvm/sigp.c
KVM: Drop user return notifier when disabling virtualization on a cpu
KVM: VMX: Disable unrestricted guest when EPT disabled
KVM: x86 emulator: limit instructions to 15 bytes
KVM: s390: Make psw available on all exits, not just a subset
KVM: x86: Add KVM_GET/SET_VCPU_EVENTS
KVM: VMX: Report unexpected simultaneous exceptions as internal errors
KVM: Allow internal errors reported to userspace to carry extra data
KVM: Reorder IOCTLs in main kvm.h
KVM: x86: Polish exception injection via KVM_SET_GUEST_DEBUG
KVM: only clear irq_source_id if irqchip is present
KVM: x86: disallow KVM_{SET,GET}_LAPIC without allocated in-kernel lapic
KVM: x86: disallow multiple KVM_CREATE_IRQCHIP
KVM: VMX: Remove vmx->msr_offset_efer
KVM: MMU: update invlpg handler comment
KVM: VMX: move CR3/PDPTR update to vmx_set_cr3
KVM: remove duplicated task_switch check
KVM: powerpc: Fix BUILD_BUG_ON condition
KVM: VMX: Use shared msr infrastructure
...
Trivial conflicts due to new Kconfig options in arch/Kconfig and kernel/Makefile
With CLONE_IO, parent's io_context->nr_tasks is incremented, but never
decremented whenever copy_process() fails afterwards, which prevents
exit_io_context() from calling IO schedulers exit functions.
Give a task_struct to exit_io_context(), and call exit_io_context() instead of
put_io_context() in copy_process() cleanup path.
Signed-off-by: Louis Rilling <louis.rilling@kerlabs.com>
Signed-off-by: Jens Axboe <jens.axboe@oracle.com>
This is a real fix for problem of utime/stime values decreasing
described in the thread:
http://lkml.org/lkml/2009/11/3/522
Now cputime is accounted in the following way:
- {u,s}time in task_struct are increased every time when the thread
is interrupted by a tick (timer interrupt).
- When a thread exits, its {u,s}time are added to signal->{u,s}time,
after adjusted by task_times().
- When all threads in a thread_group exits, accumulated {u,s}time
(and also c{u,s}time) in signal struct are added to c{u,s}time
in signal struct of the group's parent.
So {u,s}time in task struct are "raw" tick count, while
{u,s}time and c{u,s}time in signal struct are "adjusted" values.
And accounted values are used by:
- task_times(), to get cputime of a thread:
This function returns adjusted values that originates from raw
{u,s}time and scaled by sum_exec_runtime that accounted by CFS.
- thread_group_cputime(), to get cputime of a thread group:
This function returns sum of all {u,s}time of living threads in
the group, plus {u,s}time in the signal struct that is sum of
adjusted cputimes of all exited threads belonged to the group.
The problem is the return value of thread_group_cputime(),
because it is mixed sum of "raw" value and "adjusted" value:
group's {u,s}time = foreach(thread){{u,s}time} + exited({u,s}time)
This misbehavior can break {u,s}time monotonicity.
Assume that if there is a thread that have raw values greater
than adjusted values (e.g. interrupted by 1000Hz ticks 50 times
but only runs 45ms) and if it exits, cputime will decrease (e.g.
-5ms).
To fix this, we could do:
group's {u,s}time = foreach(t){task_times(t)} + exited({u,s}time)
But task_times() contains hard divisions, so applying it for
every thread should be avoided.
This patch fixes the above problem in the following way:
- Modify thread's exit (= __exit_signal()) not to use task_times().
It means {u,s}time in signal struct accumulates raw values instead
of adjusted values. As the result it makes thread_group_cputime()
to return pure sum of "raw" values.
- Introduce a new function thread_group_times(*task, *utime, *stime)
that converts "raw" values of thread_group_cputime() to "adjusted"
values, in same calculation procedure as task_times().
- Modify group's exit (= wait_task_zombie()) to use this introduced
thread_group_times(). It make c{u,s}time in signal struct to
have adjusted values like before this patch.
- Replace some thread_group_cputime() by thread_group_times().
This replacements are only applied where conveys the "adjusted"
cputime to users, and where already uses task_times() near by it.
(i.e. sys_times(), getrusage(), and /proc/<PID>/stat.)
This patch have a positive side effect:
- Before this patch, if a group contains many short-life threads
(e.g. runs 0.9ms and not interrupted by ticks), the group's
cputime could be invisible since thread's cputime was accumulated
after adjusted: imagine adjustment function as adj(ticks, runtime),
{adj(0, 0.9) + adj(0, 0.9) + ....} = {0 + 0 + ....} = 0.
After this patch it will not happen because the adjustment is
applied after accumulated.
v2:
- remove if()s, put new variables into signal_struct.
Signed-off-by: Hidetoshi Seto <seto.hidetoshi@jp.fujitsu.com>
Acked-by: Peter Zijlstra <peterz@infradead.org>
Cc: Spencer Candland <spencer@bluehost.com>
Cc: Americo Wang <xiyou.wangcong@gmail.com>
Cc: Oleg Nesterov <oleg@redhat.com>
Cc: Balbir Singh <balbir@in.ibm.com>
Cc: Stanislaw Gruszka <sgruszka@redhat.com>
LKML-Reference: <4B162517.8040909@jp.fujitsu.com>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
- Remove if({u,s}t)s because no one call it with NULL now.
- Use cputime_{add,sub}().
- Add ifndef-endif for prev_{u,s}time since they are used
only when !VIRT_CPU_ACCOUNTING.
Signed-off-by: Hidetoshi Seto <seto.hidetoshi@jp.fujitsu.com>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Spencer Candland <spencer@bluehost.com>
Cc: Americo Wang <xiyou.wangcong@gmail.com>
Cc: Oleg Nesterov <oleg@redhat.com>
Cc: Balbir Singh <balbir@in.ibm.com>
Cc: Stanislaw Gruszka <sgruszka@redhat.com>
LKML-Reference: <4B1624C7.7040302@jp.fujitsu.com>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
fork() clones all thread_info flags, including
TIF_USER_RETURN_NOTIFY; if the new task is first scheduled on a cpu
which doesn't have user return notifiers set, this causes user
return notifiers to trigger without any way of clearing itself.
This is easy to trigger with a forky workload on the host in
parallel with kvm, resulting in a cpu in an endless loop on the
verge of returning to userspace.
Fix by dropping the TIF_USER_RETURN_NOTIFY immediately after fork.
Signed-off-by: Avi Kivity <avi@redhat.com>
LKML-Reference: <1259505288-16559-1-git-send-email-avi@redhat.com>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
nr_processes() returns the sum of the per cpu counter process_counts for
all online CPUs. This counter is incremented for the current CPU on
fork() and decremented for the current CPU on exit(). Since a process
does not necessarily fork and exit on the same CPU the process_count for
an individual CPU can be either positive or negative and effectively has
no meaning in isolation.
Therefore calculating the sum of process_counts over only the online
CPUs omits the processes which were started or stopped on any CPU which
has since been unplugged. Only the sum of process_counts across all
possible CPUs has meaning.
The only caller of nr_processes() is proc_root_getattr() which
calculates the number of links to /proc as
stat->nlink = proc_root.nlink + nr_processes();
You don't have to be all that unlucky for the nr_processes() to return a
negative value leading to a negative number of links (or rather, an
apparently enormous number of links). If this happens then you can get
failures where things like "ls /proc" start to fail because they got an
-EOVERFLOW from some stat() call.
Example with some debugging inserted to show what goes on:
# ps haux|wc -l
nr_processes: CPU0: 90
nr_processes: CPU1: 1030
nr_processes: CPU2: -900
nr_processes: CPU3: -136
nr_processes: TOTAL: 84
proc_root_getattr. nlink 12 + nr_processes() 84 = 96
84
# echo 0 >/sys/devices/system/cpu/cpu1/online
# ps haux|wc -l
nr_processes: CPU0: 85
nr_processes: CPU2: -901
nr_processes: CPU3: -137
nr_processes: TOTAL: -953
proc_root_getattr. nlink 12 + nr_processes() -953 = -941
75
# stat /proc/
nr_processes: CPU0: 84
nr_processes: CPU2: -901
nr_processes: CPU3: -137
nr_processes: TOTAL: -954
proc_root_getattr. nlink 12 + nr_processes() -954 = -942
File: `/proc/'
Size: 0 Blocks: 0 IO Block: 1024 directory
Device: 3h/3d Inode: 1 Links: 4294966354
Access: (0555/dr-xr-xr-x) Uid: ( 0/ root) Gid: ( 0/ root)
Access: 2009-11-03 09:06:55.000000000 +0000
Modify: 2009-11-03 09:06:55.000000000 +0000
Change: 2009-11-03 09:06:55.000000000 +0000
I'm not 100% convinced that the per_cpu regions remain valid for offline
CPUs, although my testing suggests that they do. If not then I think the
correct solution would be to aggregate the process_count for a given CPU
into a global base value in cpu_down().
This bug appears to pre-date the transition to git and it looks like it
may even have been present in linux-2.6.0-test7-bk3 since it looks like
the code Rusty patched in http://lwn.net/Articles/64773/ was already
wrong.
Signed-off-by: Ian Campbell <ian.campbell@citrix.com>
Cc: Andrew Morton <akpm@linux-foundation.org>
Cc: Rusty Russell <rusty@rustcorp.com.au>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
* 'core-fixes-for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/tip/linux-2.6-tip:
futex: fix requeue_pi key imbalance
futex: Fix typo in FUTEX_WAIT/WAKE_BITSET_PRIVATE definitions
rcu: Place root rcu_node structure in separate lockdep class
rcu: Make hot-unplugged CPU relinquish its own RCU callbacks
rcu: Move rcu_barrier() to rcutree
futex: Move exit_pi_state() call to release_mm()
futex: Nullify robust lists after cleanup
futex: Fix locking imbalance
panic: Fix panic message visibility by calling bust_spinlocks(0) before dying
rcu: Replace the rcu_barrier enum with pointer to call_rcu*() function
rcu: Clean up code based on review feedback from Josh Triplett, part 4
rcu: Clean up code based on review feedback from Josh Triplett, part 3
rcu: Fix rcu_lock_map build failure on CONFIG_PROVE_LOCKING=y
rcu: Clean up code to address Ingo's checkpatch feedback
rcu: Clean up code based on review feedback from Josh Triplett, part 2
rcu: Clean up code based on review feedback from Josh Triplett
exit_pi_state() is called from do_exit() but not from do_execve().
Move it to release_mm() so it gets called from do_execve() as well.
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
LKML-Reference: <new-submission>
Cc: stable@kernel.org
Cc: Anirban Sinha <ani@anirban.org>
Cc: Peter Zijlstra <peterz@infradead.org>
The robust list pointers of user space held futexes are kept intact
over an exec() call. When the exec'ed task exits exit_robust_list() is
called with the stale pointer. The risk of corruption is minimal, but
still it is incorrect to keep the pointers valid. Actually glibc
should uninstall the robust list before calling exec() but we have to
deal with it anyway.
Nullify the pointers after [compat_]exit_robust_list() has been
called.
Reported-by: Anirban Sinha <ani@anirban.org>
Signed-off-by: Peter Zijlstra <peterz@infradead.org>
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
LKML-Reference: <new-submission>
Cc: stable@kernel.org
Because the binfmt is not different between threads in the same process,
it can be moved from task_struct to mm_struct. And binfmt moudle is
handled per mm_struct instead of task_struct.
Signed-off-by: Hiroshi Shimamoto <h-shimamoto@ct.jp.nec.com>
Acked-by: Oleg Nesterov <oleg@redhat.com>
Cc: Rusty Russell <rusty@rustcorp.com.au>
Acked-by: Roland McGrath <roland@redhat.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
->ioctx_lock and ->ioctx_list are used only under CONFIG_AIO.
Signed-off-by: Alexey Dobriyan <adobriyan@gmail.com>
Cc: Zach Brown <zach.brown@oracle.com>
Cc: Benjamin LaHaise <bcrl@kvack.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
When global or container-init processes use CLONE_PARENT, they create a
multi-rooted process tree. Besides siblings of global init remain as
zombies on exit since they are not reaped by their parent (swapper). So
prevent global and container-inits from creating siblings.
Signed-off-by: Sukadev Bhattiprolu <sukadev@us.ibm.com>
Acked-by: Eric W. Biederman <ebiederm@xmission.com>
Acked-by: Roland McGrath <roland@redhat.com>
Cc: Oren Laadan <orenl@cs.columbia.edu>
Cc: Oleg Nesterov <oleg@redhat.com>
Cc: Serge Hallyn <serue@us.ibm.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
A patch to give a better overview of the userland application stack usage,
especially for embedded linux.
Currently you are only able to dump the main process/thread stack usage
which is showed in /proc/pid/status by the "VmStk" Value. But you get no
information about the consumed stack memory of the the threads.
There is an enhancement in the /proc/<pid>/{task/*,}/*maps and which marks
the vm mapping where the thread stack pointer reside with "[thread stack
xxxxxxxx]". xxxxxxxx is the maximum size of stack. This is a value
information, because libpthread doesn't set the start of the stack to the
top of the mapped area, depending of the pthread usage.
A sample output of /proc/<pid>/task/<tid>/maps looks like:
08048000-08049000 r-xp 00000000 03:00 8312 /opt/z
08049000-0804a000 rw-p 00001000 03:00 8312 /opt/z
0804a000-0806b000 rw-p 00000000 00:00 0 [heap]
a7d12000-a7d13000 ---p 00000000 00:00 0
a7d13000-a7f13000 rw-p 00000000 00:00 0 [thread stack: 001ff4b4]
a7f13000-a7f14000 ---p 00000000 00:00 0
a7f14000-a7f36000 rw-p 00000000 00:00 0
a7f36000-a8069000 r-xp 00000000 03:00 4222 /lib/libc.so.6
a8069000-a806b000 r--p 00133000 03:00 4222 /lib/libc.so.6
a806b000-a806c000 rw-p 00135000 03:00 4222 /lib/libc.so.6
a806c000-a806f000 rw-p 00000000 00:00 0
a806f000-a8083000 r-xp 00000000 03:00 14462 /lib/libpthread.so.0
a8083000-a8084000 r--p 00013000 03:00 14462 /lib/libpthread.so.0
a8084000-a8085000 rw-p 00014000 03:00 14462 /lib/libpthread.so.0
a8085000-a8088000 rw-p 00000000 00:00 0
a8088000-a80a4000 r-xp 00000000 03:00 8317 /lib/ld-linux.so.2
a80a4000-a80a5000 r--p 0001b000 03:00 8317 /lib/ld-linux.so.2
a80a5000-a80a6000 rw-p 0001c000 03:00 8317 /lib/ld-linux.so.2
afaf5000-afb0a000 rw-p 00000000 00:00 0 [stack]
ffffe000-fffff000 r-xp 00000000 00:00 0 [vdso]
Also there is a new entry "stack usage" in /proc/<pid>/{task/*,}/status
which will you give the current stack usage in kb.
A sample output of /proc/self/status looks like:
Name: cat
State: R (running)
Tgid: 507
Pid: 507
.
.
.
CapBnd: fffffffffffffeff
voluntary_ctxt_switches: 0
nonvoluntary_ctxt_switches: 0
Stack usage: 12 kB
I also fixed stack base address in /proc/<pid>/{task/*,}/stat to the base
address of the associated thread stack and not the one of the main
process. This makes more sense.
[akpm@linux-foundation.org: fs/proc/array.c now needs walk_page_range()]
Signed-off-by: Stefani Seibold <stefani@seibold.net>
Cc: Ingo Molnar <mingo@elte.hu>
Cc: Peter Zijlstra <a.p.zijlstra@chello.nl>
Cc: Alexey Dobriyan <adobriyan@gmail.com>
Cc: "Eric W. Biederman" <ebiederm@xmission.com>
Cc: Randy Dunlap <randy.dunlap@oracle.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
Currently, OOM logic callflow is here.
__out_of_memory()
select_bad_process() for each task
badness() calculate badness of one task
oom_kill_process() search child
oom_kill_task() kill target task and mm shared tasks with it
example, process-A have two thread, thread-A and thread-B and it have very
fat memory and each thread have following oom_adj and oom_score.
thread-A: oom_adj = OOM_DISABLE, oom_score = 0
thread-B: oom_adj = 0, oom_score = very-high
Then, select_bad_process() select thread-B, but oom_kill_task() refuse
kill the task because thread-A have OOM_DISABLE. Thus __out_of_memory()
call select_bad_process() again. but select_bad_process() select the same
task. It mean kernel fall in livelock.
The fact is, select_bad_process() must select killable task. otherwise
OOM logic go into livelock.
And root cause is, oom_adj shouldn't be per-thread value. it should be
per-process value because OOM-killer kill a process, not thread. Thus
This patch moves oomkilladj (now more appropriately named oom_adj) from
struct task_struct to struct signal_struct. it naturally prevent
select_bad_process() choose wrong task.
Signed-off-by: KOSAKI Motohiro <kosaki.motohiro@jp.fujitsu.com>
Cc: Paul Menage <menage@google.com>
Cc: David Rientjes <rientjes@google.com>
Cc: KAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com>
Cc: Rik van Riel <riel@redhat.com>
Cc: Oleg Nesterov <oleg@redhat.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
Rawhide users have reported hang at startup when cryptsetup is run: the
same problem can be simply reproduced by running a program int main() {
mlockall(MCL_CURRENT | MCL_FUTURE); return 0; }
The problem is that exit_mmap() applies munlock_vma_pages_all() to
clean up VM_LOCKED areas, and its current implementation (stupidly)
tries to fault in absent pages, for example where PROT_NONE prevented
them being faulted in when mlocking. Whereas the "ksm: fix oom
deadlock" patch, knowing there's a race by which KSM might try to fault
in pages after exit_mmap() had finally zapped the range, backs out of
such faults doing nothing when its ksm_test_exit() notices mm_users 0.
So revert that part of "ksm: fix oom deadlock" which moved the
ksm_exit() call from before exit_mmap() to the middle of exit_mmap();
and remove those ksm_test_exit() checks from the page fault paths, so
allowing the munlocking to proceed without interference.
ksm_exit, if there are rmap_items still chained on this mm slot, takes
mmap_sem write side: so preventing KSM from working on an mm while
exit_mmap runs. And KSM will bail out as soon as it notices that
mm_users is already zero, thanks to its internal ksm_test_exit checks.
So that when a task is killed by OOM killer or the user, KSM will not
indefinitely prevent it from running exit_mmap to release its memory.
This does break a part of what "ksm: fix oom deadlock" was trying to
achieve. When unmerging KSM (echo 2 >/sys/kernel/mm/ksm), and even
when ksmd itself has to cancel a KSM page, it is possible that the
first OOM-kill victim would be the KSM process being faulted: then its
memory won't be freed until a second victim has been selected (freeing
memory for the unmerging fault to complete).
But the OOM killer is already liable to kill a second victim once the
intended victim's p->mm goes to NULL: so there's not much point in
rejecting this KSM patch before fixing that OOM behaviour. It is very
much more important to allow KSM users to boot up, than to haggle over
an unlikely and poorly supported OOM case.
We also intend to fix munlocking to not fault pages: at which point
this patch _could_ be reverted; though that would be controversial, so
we hope to find a better solution.
Signed-off-by: Andrea Arcangeli <aarcange@redhat.com>
Acked-by: Justin M. Forbes <jforbes@redhat.com>
Acked-for-now-by: Hugh Dickins <hugh.dickins@tiscali.co.uk>
Cc: Izik Eidus <ieidus@redhat.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
There's a now-obvious deadlock in KSM's out-of-memory handling:
imagine ksmd or KSM_RUN_UNMERGE handling, holding ksm_thread_mutex,
trying to allocate a page to break KSM in an mm which becomes the
OOM victim (quite likely in the unmerge case): it's killed and goes
to exit, and hangs there waiting to acquire ksm_thread_mutex.
Clearly we must not require ksm_thread_mutex in __ksm_exit, simple
though that made everything else: perhaps use mmap_sem somehow?
And part of the answer lies in the comments on unmerge_ksm_pages:
__ksm_exit should also leave all the rmap_item removal to ksmd.
But there's a fundamental problem, that KSM relies upon mmap_sem to
guarantee the consistency of the mm it's dealing with, yet exit_mmap
tears down an mm without taking mmap_sem. And bumping mm_users won't
help at all, that just ensures that the pages the OOM killer assumes
are on their way to being freed will not be freed.
The best answer seems to be, to move the ksm_exit callout from just
before exit_mmap, to the middle of exit_mmap: after the mm's pages
have been freed (if the mmu_gather is flushed), but before its page
tables and vma structures have been freed; and down_write,up_write
mmap_sem there to serialize with KSM's own reliance on mmap_sem.
But KSM then needs to be careful, whenever it downs mmap_sem, to
check that the mm is not already exiting: there's a danger of using
find_vma on a layout that's being torn apart, or writing into page
tables which have been freed for reuse; and even do_anonymous_page
and __do_fault need to check they're not being called by break_ksm
to reinstate a pte after zap_pte_range has zapped that page table.
Though it might be clearer to add an exiting flag, set while holding
mmap_sem in __ksm_exit, that wouldn't cover the issue of reinstating
a zapped pte. All we need is to check whether mm_users is 0 - but
must remember that ksmd may detect that before __ksm_exit is reached.
So, ksm_test_exit(mm) added to comment such checks on mm->mm_users.
__ksm_exit now has to leave clearing up the rmap_items to ksmd,
that needs ksm_thread_mutex; but shift the exiting mm just after the
ksm_scan cursor so that it will soon be dealt with. __ksm_enter raise
mm_count to hold the mm_struct, ksmd's exit processing (exactly like
its processing when it finds all VM_MERGEABLEs unmapped) mmdrop it,
similar procedure for KSM_RUN_UNMERGE (which has stopped ksmd).
But also give __ksm_exit a fast path: when there's no complication
(no rmap_items attached to mm and it's not at the ksm_scan cursor),
it can safely do all the exiting work itself. This is not just an
optimization: when ksmd is not running, the raised mm_count would
otherwise leak mm_structs.
Signed-off-by: Hugh Dickins <hugh.dickins@tiscali.co.uk>
Acked-by: Izik Eidus <ieidus@redhat.com>
Cc: Andrea Arcangeli <aarcange@redhat.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
This patch presents the mm interface to a dummy version of ksm.c, for
better scrutiny of that interface: the real ksm.c follows later.
When CONFIG_KSM is not set, madvise(2) reject MADV_MERGEABLE and
MADV_UNMERGEABLE with EINVAL, since that seems more helpful than
pretending that they can be serviced. But when CONFIG_KSM=y, accept them
even if KSM is not currently running, and even on areas which KSM will not
touch (e.g. hugetlb or shared file or special driver mappings).
Like other madvices, report ENOMEM despite success if any area in the
range is unmapped, and use EAGAIN to report out of memory.
Define vma flag VM_MERGEABLE to identify an area on which KSM may try
merging pages: leave it to ksm_madvise() to decide whether to set it.
Define mm flag MMF_VM_MERGEABLE to identify an mm which might contain
VM_MERGEABLE areas, to minimize callouts when forking or exiting.
Based upon earlier patches by Chris Wright and Izik Eidus.
Signed-off-by: Hugh Dickins <hugh.dickins@tiscali.co.uk>
Signed-off-by: Chris Wright <chrisw@redhat.com>
Signed-off-by: Izik Eidus <ieidus@redhat.com>
Cc: Michael Kerrisk <mtk.manpages@gmail.com>
Cc: Andrea Arcangeli <aarcange@redhat.com>
Cc: Rik van Riel <riel@redhat.com>
Cc: Wu Fengguang <fengguang.wu@intel.com>
Cc: Balbir Singh <balbir@in.ibm.com>
Cc: Hugh Dickins <hugh.dickins@tiscali.co.uk>
Cc: KAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com>
Cc: Lee Schermerhorn <lee.schermerhorn@hp.com>
Cc: Avi Kivity <avi@redhat.com>
Cc: Nick Piggin <nickpiggin@yahoo.com.au>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
