1
linux/fs
Frank Mayhar f06febc96b timers: fix itimer/many thread hang
Overview

This patch reworks the handling of POSIX CPU timers, including the
ITIMER_PROF, ITIMER_VIRT timers and rlimit handling.  It was put together
with the help of Roland McGrath, the owner and original writer of this code.

The problem we ran into, and the reason for this rework, has to do with using
a profiling timer in a process with a large number of threads.  It appears
that the performance of the old implementation of run_posix_cpu_timers() was
at least O(n*3) (where "n" is the number of threads in a process) or worse.
Everything is fine with an increasing number of threads until the time taken
for that routine to run becomes the same as or greater than the tick time, at
which point things degrade rather quickly.

This patch fixes bug 9906, "Weird hang with NPTL and SIGPROF."

Code Changes

This rework corrects the implementation of run_posix_cpu_timers() to make it
run in constant time for a particular machine.  (Performance may vary between
one machine and another depending upon whether the kernel is built as single-
or multiprocessor and, in the latter case, depending upon the number of
running processors.)  To do this, at each tick we now update fields in
signal_struct as well as task_struct.  The run_posix_cpu_timers() function
uses those fields to make its decisions.

We define a new structure, "task_cputime," to contain user, system and
scheduler times and use these in appropriate places:

struct task_cputime {
	cputime_t utime;
	cputime_t stime;
	unsigned long long sum_exec_runtime;
};

This is included in the structure "thread_group_cputime," which is a new
substructure of signal_struct and which varies for uniprocessor versus
multiprocessor kernels.  For uniprocessor kernels, it uses "task_cputime" as
a simple substructure, while for multiprocessor kernels it is a pointer:

struct thread_group_cputime {
	struct task_cputime totals;
};

struct thread_group_cputime {
	struct task_cputime *totals;
};

We also add a new task_cputime substructure directly to signal_struct, to
cache the earliest expiration of process-wide timers, and task_cputime also
replaces the it_*_expires fields of task_struct (used for earliest expiration
of thread timers).  The "thread_group_cputime" structure contains process-wide
timers that are updated via account_user_time() and friends.  In the non-SMP
case the structure is a simple aggregator; unfortunately in the SMP case that
simplicity was not achievable due to cache-line contention between CPUs (in
one measured case performance was actually _worse_ on a 16-cpu system than
the same test on a 4-cpu system, due to this contention).  For SMP, the
thread_group_cputime counters are maintained as a per-cpu structure allocated
using alloc_percpu().  The timer functions update only the timer field in
the structure corresponding to the running CPU, obtained using per_cpu_ptr().

We define a set of inline functions in sched.h that we use to maintain the
thread_group_cputime structure and hide the differences between UP and SMP
implementations from the rest of the kernel.  The thread_group_cputime_init()
function initializes the thread_group_cputime structure for the given task.
The thread_group_cputime_alloc() is a no-op for UP; for SMP it calls the
out-of-line function thread_group_cputime_alloc_smp() to allocate and fill
in the per-cpu structures and fields.  The thread_group_cputime_free()
function, also a no-op for UP, in SMP frees the per-cpu structures.  The
thread_group_cputime_clone_thread() function (also a UP no-op) for SMP calls
thread_group_cputime_alloc() if the per-cpu structures haven't yet been
allocated.  The thread_group_cputime() function fills the task_cputime
structure it is passed with the contents of the thread_group_cputime fields;
in UP it's that simple but in SMP it must also safely check that tsk->signal
is non-NULL (if it is it just uses the appropriate fields of task_struct) and,
if so, sums the per-cpu values for each online CPU.  Finally, the three
functions account_group_user_time(), account_group_system_time() and
account_group_exec_runtime() are used by timer functions to update the
respective fields of the thread_group_cputime structure.

Non-SMP operation is trivial and will not be mentioned further.

The per-cpu structure is always allocated when a task creates its first new
thread, via a call to thread_group_cputime_clone_thread() from copy_signal().
It is freed at process exit via a call to thread_group_cputime_free() from
cleanup_signal().

All functions that formerly summed utime/stime/sum_sched_runtime values from
from all threads in the thread group now use thread_group_cputime() to
snapshot the values in the thread_group_cputime structure or the values in
the task structure itself if the per-cpu structure hasn't been allocated.

Finally, the code in kernel/posix-cpu-timers.c has changed quite a bit.
The run_posix_cpu_timers() function has been split into a fast path and a
slow path; the former safely checks whether there are any expired thread
timers and, if not, just returns, while the slow path does the heavy lifting.
With the dedicated thread group fields, timers are no longer "rebalanced" and
the process_timer_rebalance() function and related code has gone away.  All
summing loops are gone and all code that used them now uses the
thread_group_cputime() inline.  When process-wide timers are set, the new
task_cputime structure in signal_struct is used to cache the earliest
expiration; this is checked in the fast path.

Performance

The fix appears not to add significant overhead to existing operations.  It
generally performs the same as the current code except in two cases, one in
which it performs slightly worse (Case 5 below) and one in which it performs
very significantly better (Case 2 below).  Overall it's a wash except in those
two cases.

I've since done somewhat more involved testing on a dual-core Opteron system.

Case 1: With no itimer running, for a test with 100,000 threads, the fixed
	kernel took 1428.5 seconds, 513 seconds more than the unfixed system,
	all of which was spent in the system.  There were twice as many
	voluntary context switches with the fix as without it.

Case 2: With an itimer running at .01 second ticks and 4000 threads (the most
	an unmodified kernel can handle), the fixed kernel ran the test in
	eight percent of the time (5.8 seconds as opposed to 70 seconds) and
	had better tick accuracy (.012 seconds per tick as opposed to .023
	seconds per tick).

Case 3: A 4000-thread test with an initial timer tick of .01 second and an
	interval of 10,000 seconds (i.e. a timer that ticks only once) had
	very nearly the same performance in both cases:  6.3 seconds elapsed
	for the fixed kernel versus 5.5 seconds for the unfixed kernel.

With fewer threads (eight in these tests), the Case 1 test ran in essentially
the same time on both the modified and unmodified kernels (5.2 seconds versus
5.8 seconds).  The Case 2 test ran in about the same time as well, 5.9 seconds
versus 5.4 seconds but again with much better tick accuracy, .013 seconds per
tick versus .025 seconds per tick for the unmodified kernel.

Since the fix affected the rlimit code, I also tested soft and hard CPU limits.

Case 4: With a hard CPU limit of 20 seconds and eight threads (and an itimer
	running), the modified kernel was very slightly favored in that while
	it killed the process in 19.997 seconds of CPU time (5.002 seconds of
	wall time), only .003 seconds of that was system time, the rest was
	user time.  The unmodified kernel killed the process in 20.001 seconds
	of CPU (5.014 seconds of wall time) of which .016 seconds was system
	time.  Really, though, the results were too close to call.  The results
	were essentially the same with no itimer running.

Case 5: With a soft limit of 20 seconds and a hard limit of 2000 seconds
	(where the hard limit would never be reached) and an itimer running,
	the modified kernel exhibited worse tick accuracy than the unmodified
	kernel: .050 seconds/tick versus .028 seconds/tick.  Otherwise,
	performance was almost indistinguishable.  With no itimer running this
	test exhibited virtually identical behavior and times in both cases.

In times past I did some limited performance testing.  those results are below.

On a four-cpu Opteron system without this fix, a sixteen-thread test executed
in 3569.991 seconds, of which user was 3568.435s and system was 1.556s.  On
the same system with the fix, user and elapsed time were about the same, but
system time dropped to 0.007 seconds.  Performance with eight, four and one
thread were comparable.  Interestingly, the timer ticks with the fix seemed
more accurate:  The sixteen-thread test with the fix received 149543 ticks
for 0.024 seconds per tick, while the same test without the fix received 58720
for 0.061 seconds per tick.  Both cases were configured for an interval of
0.01 seconds.  Again, the other tests were comparable.  Each thread in this
test computed the primes up to 25,000,000.

I also did a test with a large number of threads, 100,000 threads, which is
impossible without the fix.  In this case each thread computed the primes only
up to 10,000 (to make the runtime manageable).  System time dominated, at
1546.968 seconds out of a total 2176.906 seconds (giving a user time of
629.938s).  It received 147651 ticks for 0.015 seconds per tick, still quite
accurate.  There is obviously no comparable test without the fix.

Signed-off-by: Frank Mayhar <fmayhar@google.com>
Cc: Roland McGrath <roland@redhat.com>
Cc: Alexey Dobriyan <adobriyan@gmail.com>
Cc: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
2008-09-14 16:25:35 +02:00
..
9p [PATCH] fix ->llseek() for a bunch of directories 2008-08-25 01:18:09 -04:00
adfs [PATCH] fix ->llseek() for a bunch of directories 2008-08-25 01:18:09 -04:00
affs [PATCH] fix ->llseek() for a bunch of directories 2008-08-25 01:18:09 -04:00
afs mm: rename page trylock 2008-08-04 21:31:34 -07:00
autofs
autofs4 [PATCH] fix ->llseek() for a bunch of directories 2008-08-25 01:18:09 -04:00
befs [PATCH] fix ->llseek() for a bunch of directories 2008-08-25 01:18:09 -04:00
bfs bfs: fix Lockdep warning 2008-09-13 14:41:51 -07:00
cifs [CIFS] Turn off Unicode during session establishment for plaintext authentication 2008-08-28 15:32:22 +00:00
coda [PATCH] sanitize __user_walk_fd() et.al. 2008-07-26 20:53:34 -04:00
configfs [PATCH] configfs: Consolidate locking around configfs_detach_prep() in configfs_rmdir() 2008-08-22 11:09:02 -07:00
cramfs cramfs: fix named-pipe handling 2008-08-20 15:40:32 -07:00
debugfs debugfs: Implement debugfs_remove_recursive() 2008-07-21 21:54:59 -07:00
devpts [PATCH] devpts: switch to IDA 2008-08-01 11:25:29 -04:00
dlm dlm: rename structs 2008-08-13 12:47:36 -05:00
ecryptfs eCryptfs: use page_alloc not kmalloc to get a page of memory 2008-07-28 16:30:21 -07:00
efs [PATCH] fix efs_lookup() 2008-08-25 01:18:04 -04:00
exportfs fs: replace remaining __FUNCTION__ occurrences 2008-04-30 08:29:54 -07:00
ext2 vfs: pagecache usage optimization for pagesize!=blocksize 2008-07-28 16:30:21 -07:00
ext3 [PATCH] fix races and leaks in vfs_quota_on() users 2008-08-01 11:25:25 -04:00
ext4 ext4: Initialize writeback_index to 0 when allocating a new inode 2008-08-19 21:14:52 -04:00
fat vfat: fix 'sync' mount deadlock due to BKL->lock_super conversion 2008-08-20 08:31:19 -07:00
freevxfs fs/freevxfs/: proper externs 2008-04-29 08:06:00 -07:00
fuse [PATCH] fix MAY_CHDIR/MAY_ACCESS/LOOKUP_ACCESS mess 2008-07-26 20:53:21 -04:00
gfs2 [PATCH] don't pass nameidata to gfs2_lookupi() 2008-07-26 20:53:36 -04:00
hfs [PATCH] f_count may wrap around 2008-07-26 20:53:40 -04:00
hfsplus [PATCH] f_count may wrap around 2008-07-26 20:53:40 -04:00
hostfs [PATCH] sanitize ->permission() prototype 2008-07-26 20:53:14 -04:00
hpfs [patch 05/14] hpfs: dont call permission() 2008-07-26 20:53:13 -04:00
hppfs [patch] hppfs: remove hppfs_permission 2008-07-26 20:53:07 -04:00
hugetlbfs SL*B: drop kmem cache argument from constructor 2008-07-26 12:00:07 -07:00
isofs SL*B: drop kmem cache argument from constructor 2008-07-26 12:00:07 -07:00
jbd Merge branch 'core/locking' into core/urgent 2008-08-12 00:11:49 +02:00
jbd2 Merge branch 'core/locking' into core/urgent 2008-08-12 00:11:49 +02:00
jffs2 removed unused #include <linux/version.h>'s 2008-08-23 12:14:12 -07:00
jfs [PATCH] sanitize ->permission() prototype 2008-07-26 20:53:14 -04:00
lockd Merge branch 'for-2.6.27' of git://linux-nfs.org/~bfields/linux 2008-08-12 16:39:22 -07:00
minix SL*B: drop kmem cache argument from constructor 2008-07-26 12:00:07 -07:00
msdos fatfs: add UTC timestamp option 2008-07-25 10:53:34 -07:00
ncpfs [PATCH] don't pass nameidata to __ncp_lookup_validate() 2008-07-26 20:53:37 -04:00
nfs NFS: Restore missing hunk in NFS mount option parser 2008-09-08 15:35:19 -07:00
nfs_common
nfsd nfsd: fix buffer overrun decoding NFSv4 acl 2008-09-01 14:24:24 -04:00
nls
ntfs NTFS: update homepage 2008-09-02 19:21:37 -07:00
ocfs2 ocfs2: Fix a bug in direct IO read. 2008-09-10 01:44:08 -07:00
omfs omfs: fix oops when file metadata is corrupted 2008-08-15 08:35:44 -07:00
openpromfs SL*B: drop kmem cache argument from constructor 2008-07-26 12:00:07 -07:00
partitions rescan_partitions(): make device capacity errors non-fatal 2008-09-13 14:41:52 -07:00
proc timers: fix itimer/many thread hang 2008-09-14 16:25:35 +02:00
qnx4 SL*B: drop kmem cache argument from constructor 2008-07-26 12:00:07 -07:00
ramfs ramfs: enable splice write 2008-07-04 09:52:14 +02:00
reiserfs reiserfs: removed duplicated #include 2008-08-12 16:07:30 -07:00
romfs romfs_readpage: don't report errors for pages beyond i_size 2008-07-30 14:30:34 -07:00
smbfs [PATCH] sanitize ->permission() prototype 2008-07-26 20:53:14 -04:00
sysfs Use WARN() in fs/sysfs 2008-07-26 12:00:07 -07:00
sysv SL*B: drop kmem cache argument from constructor 2008-07-26 12:00:07 -07:00
ubifs UBIFS: make minimum fanout 3 2008-09-05 20:02:35 +03:00
udf udf: add llseek method 2008-09-08 20:31:04 +02:00
ufs Revert "UFS: add const to parser token table" 2008-08-04 16:50:38 -07:00
vfat fatfs: add UTC timestamp option 2008-07-25 10:53:34 -07:00
xfs [PATCH] fix ->llseek() for a bunch of directories 2008-08-25 01:18:09 -04:00
aio.c [PATCH] f_count may wrap around 2008-07-26 20:53:40 -04:00
anon_inodes.c flag parameters: NONBLOCK in anon_inode_getfd 2008-07-24 10:47:28 -07:00
attr.c [patch 4/4] vfs: immutable inode checking cleanup 2008-07-26 20:53:28 -04:00
bad_inode.c [PATCH] sanitize ->permission() prototype 2008-07-26 20:53:14 -04:00
binfmt_aout.c tracehook: exec 2008-07-26 12:00:08 -07:00
binfmt_elf_fdpic.c binfmt_elf_fdpic: Magical stack pointer index, for NEW_AUX_ENT compat. 2008-07-28 18:10:28 +09:00
binfmt_elf.c timers: fix itimer/many thread hang 2008-09-14 16:25:35 +02:00
binfmt_em86.c binfmt_misc.c: avoid potential kernel stack overflow 2008-04-29 08:06:04 -07:00
binfmt_flat.c binfmt_flat: Stub in a FLAT_PLAT_INIT(). 2008-08-11 20:17:55 +09:00
binfmt_misc.c binfmt_misc: fix false -ENOEXEC when coupled with other binary handlers 2008-08-20 15:40:31 -07:00
binfmt_script.c binfmt_misc.c: avoid potential kernel stack overflow 2008-04-29 08:06:04 -07:00
binfmt_som.c tracehook: exec 2008-07-26 12:00:08 -07:00
bio-integrity.c bio-integrity: remove EXPORT_SYMBOL for bio_integrity_init_slab() 2008-07-28 16:30:21 -07:00
bio.c bio: fix __bio_copy_iov() handling of bio->bv_len 2008-08-27 09:50:19 +02:00
block_dev.c [PATCH] switch mtd and dm-table to lookup_bdev() 2008-08-01 11:25:31 -04:00
buffer.c block: submit_bh() inadvertently discards barrier flag on a sync write 2008-08-27 09:50:19 +02:00
char_dev.c Remove the lock_kernel() call from chrdev_open() 2008-06-20 14:05:53 -06:00
compat_binfmt_elf.c
compat_ioctl.c remove unused #include <linux/dirent.h>'s 2008-07-25 10:53:34 -07:00
compat.c [PATCH] fix regular readdir() and friends 2008-08-25 01:18:08 -04:00
dcache.c [PATCH] change d_add_ci argument ordering 2008-08-25 01:18:05 -04:00
dcookies.c
direct-io.c dio: use get_user_pages_fast 2008-07-26 12:00:06 -07:00
dnotify.c [PATCH] split linux/file.h 2008-05-01 13:08:16 -04:00
dquot.c [PATCH] fix races and leaks in vfs_quota_on() users 2008-08-01 11:25:25 -04:00
drop_caches.c vfs: skip inodes without pages to free in drop_pagecache_sb() 2008-04-29 08:06:05 -07:00
eventfd.c flag parameters: check magic constants 2008-07-24 10:47:29 -07:00
eventpoll.c fs/eventpoll.c: fix sys_epoll_create1() comment 2008-08-12 16:07:30 -07:00
exec.c exec: include pagemap.h again to fix build 2008-07-28 16:30:20 -07:00
fcntl.c [PATCH] clean dup2() up a bit 2008-08-01 11:25:24 -04:00
fifo.c [PATCH] reuse xxx_fifo_fops for xxx_pipe_fops 2008-07-26 20:53:06 -04:00
file_table.c [PATCH] f_count may wrap around 2008-07-26 20:53:40 -04:00
file.c [PATCH] merge locate_fd() and get_unused_fd() 2008-08-01 11:25:23 -04:00
filesystems.c
fs-writeback.c VFS: export sync_sb_inodes 2008-07-14 19:10:52 +03:00
generic_acl.c
inode.c fs/inode.c: properly init address_space->writeback_index 2008-08-15 08:35:44 -07:00
inotify_user.c [PATCH] sanitize __user_walk_fd() et.al. 2008-07-26 20:53:34 -04:00
inotify.c
internal.h
ioctl.c make vfs_ioctl() static 2008-04-29 08:06:00 -07:00
ioprio.c fix setpriority(PRIO_PGRP) thread iterator breakage 2008-08-20 15:40:32 -07:00
Kconfig [CIFS] Reorder cifs config item for better clarity 2008-08-26 18:32:28 +00:00
Kconfig.binfmt sh: Initial ELF FDPIC support. 2008-07-28 18:10:28 +09:00
libfs.c VFS: increase pseudo-filesystem block size to PAGE_SIZE 2008-07-30 09:41:44 -07:00
locks.c SL*B: drop kmem cache argument from constructor 2008-07-26 12:00:07 -07:00
Makefile omfs: update kbuild to include OMFS 2008-07-26 12:00:05 -07:00
mbcache.c
mpage.c vfs: add hooks for ext4's delayed allocation support 2008-07-11 19:27:31 -04:00
namei.c [patch 3/4] vfs: remove unused nameidata argument of may_create() 2008-08-01 11:25:30 -04:00
namespace.c [PATCH] pass struct path * to do_add_mount() 2008-08-01 11:25:32 -04:00
nfsctl.c
no-block.c
open.c [PATCH] merge locate_fd() and get_unused_fd() 2008-08-01 11:25:23 -04:00
pipe.c [PATCH] reuse xxx_fifo_fops for xxx_pipe_fops 2008-07-26 20:53:06 -04:00
pnode.c
pnode.h
posix_acl.c
quota_v1.c quota: move function-macros from quota.h to quotaops.h 2008-07-25 10:53:35 -07:00
quota_v2.c quota: move function-macros from quota.h to quotaops.h 2008-07-25 10:53:35 -07:00
quota.c quota: cleanup loop in sync_dquots() 2008-07-25 10:53:35 -07:00
read_write.c Remove BKL from remote_llseek v2 2008-07-02 15:06:27 -06:00
read_write.h
readdir.c [PATCH] fix regular readdir() and friends 2008-08-25 01:18:08 -04:00
select.c Fix performance regression on lmbench select benchmark 2008-06-22 12:23:15 -07:00
seq_file.c [PATCH] deal with the first call of ->show() generating no output 2008-08-25 01:18:10 -04:00
signalfd.c flag parameters: check magic constants 2008-07-24 10:47:29 -07:00
splice.c mm: rename page trylock 2008-08-04 21:31:34 -07:00
stack.c
stat.c [PATCH] sanitize __user_walk_fd() et.al. 2008-07-26 20:53:34 -04:00
super.c fix soft lock up at NFS mount via per-SB LRU-list of unused dentries 2008-07-24 10:47:15 -07:00
sync.c SYNC_FILE_RANGE_WRITE may and will block. Document that. 2008-07-24 10:47:17 -07:00
timerfd.c flag parameters: check magic constants 2008-07-24 10:47:29 -07:00
utimes.c [PATCH] sanitize __user_walk_fd() et.al. 2008-07-26 20:53:34 -04:00
xattr_acl.c
xattr.c [PATCH] sanitize __user_walk_fd() et.al. 2008-07-26 20:53:34 -04:00