Intel reported a performance regression caused by the following commit:
commit 848c4dd515
Author: Zach Brown <zach.brown@oracle.com>
Date: Mon Aug 20 17:12:01 2007 -0700
dio: zero struct dio with kzalloc instead of manually
This patch uses kzalloc to zero all of struct dio rather than
manually trying to track which fields we rely on being zero. It
passed aio+dio stress testing and some bug regression testing on
ext3.
This patch was introduced by Linus in the conversation that lead up
to Badari's minimal fix to manually zero .map_bh.b_state in commit:
6a648fa721
It makes the code a bit smaller. Maybe a couple fewer cachelines to
load, if we're lucky:
text data bss dec hex filename
3285925 568506 1304616 5159047 4eb887 vmlinux
3285797 568506 1304616 5158919 4eb807 vmlinux.patched
I was unable to measure a stable difference in the number of cpu
cycles spent in blockdev_direct_IO() when pushing aio+dio 256K reads
at ~340MB/s.
So the resulting intent of the patch isn't a performance gain but to
avoid exposing ourselves to the risk of finding another field like
.map_bh.b_state where we rely on zeroing but don't enforce it in the
code.
Zach surmised that zeroing out the page array was what caused most of
the problem, and suggested the approach taken in the attached patch for
resolving the issue. Intel re-tested with this patch and saw a 0.6%
performance gain (the original regression was 0.5%).
[akpm@linux-foundation.org: add comment]
Signed-off-by: Jeff Moyer <jmoyer@redhat.com>
Acked-by: Zach Brown <zach.brown@oracle.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
Don't know the reason, but it appears ki_wait field of iocb never gets used.
Signed-off-by: Shaohua Li <shaohua.li@intel.com>
Cc: Jeff Moyer <jmoyer@redhat.com>
Cc: Benjamin LaHaise <bcrl@kvack.org>
Cc: Zach Brown <zach.brown@oracle.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
Andrew Morton's compiler sees the following warning in FS-Cache:
fs/fscache/object-list.c: In function 'fscache_objlist_lookup':
fs/fscache/object-list.c:94: warning: 'obj' may be used uninitialized in this function
which my compiler doesn't. This is a false positive as obj can only be
used in the comparison against minobj if minobj has been set to something
other than NULL, but for that to happen, obj has to be first set to
something.
Deal with this by preclearing obj too.
Reported-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: David Howells <dhowells@redhat.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
Implement shrinking the reserved memory for crash kernel, if it is more
than enough.
For example, if you have already reserved 128M, now you just want 100M,
you can do:
# echo $((100*1024*1024)) > /sys/kernel/kexec_crash_size
Note, you can only do this before loading the crash kernel.
Signed-off-by: WANG Cong <amwang@redhat.com>
Cc: Neil Horman <nhorman@redhat.com>
Acked-by: Eric W. Biederman <ebiederm@xmission.com>
Cc: Andi Kleen <andi@firstfloor.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
dma_mask is, when interpreted as address, the last valid byte, and hence
comparison msut also be done using the last valid of the buffer in
question.
Also fix the open-coded instances in lib/swiotlb.c.
Signed-off-by: Jan Beulich <jbeulich@novell.com>
Cc: FUJITA Tomonori <fujita.tomonori@lab.ntt.co.jp>
Cc: Becky Bruce <beckyb@kernel.crashing.org>
Cc: "Luck, Tony" <tony.luck@intel.com>
Cc: Benjamin Herrenschmidt <benh@kernel.crashing.org>
Cc: Ingo Molnar <mingo@elte.hu>
Cc: Thomas Gleixner <tglx@linutronix.de>
Cc: "H. Peter Anvin" <hpa@zytor.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
Add support for 6 ranks per channel to the i5100 chipset. I have tested
the patch as far as possible with correctible errors and things appear
good. The DIMM mapping is correct for our board, but boards may differ.
Signed-off-by: Nils Carlson <nils.carlson@ludd.ltu.se>
Acked-by: Arthur Jones <ajones@riverbed.com>
Signed-off-by: Doug Thompson <dougthompson@xmission.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
Addscrubbing to the i5100 chipset. The i5100 chipset only supports one
scrubbing rate, which is not constant but dependent on memory load. The
rate returned by this driver is an estimate based on some experimentation,
but is substantially closer to the truth than the speed supplied in the
documentation.
Also, scrubbing is done once, and then a done-bit is set. This means that
to accomplish continuous scrubbing a re-enabling mechanism must be used.
I have created the simplest possible such mechanism in the form of a
work-queue which will check every five minutes. This interval is quite
arbitrary but should be sufficient for all sizes of system memory.
Signed-off-by: Nils Carlson <nils.carlson@ludd.ltu.se>
Signed-off-by: Doug Thompson <dougthompson@xmission.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
The i5100 driver uses the word controller instead of channel in a lot of
places, this is simply a cleanup of the patch.
Signed-off-by: Nils Carlson <nils.carlson@ludd.ltu.se>
Signed-off-by: Doug Thompson <dougthompson@xmission.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
It decreases code size by 16 bytes on my gcc 4.4.1 on Core 2:
text data bss dec hex filename
4314 2216 8 6538 198a kernel/pid.o-BEFORE
4298 2216 8 6522 197a kernel/pid.o-AFTER
Signed-off-by: André Goddard Rosa <andre.goddard@gmail.com>
Cc: Pekka Enberg <penberg@cs.helsinki.fi>
Cc: Oleg Nesterov <oleg@redhat.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
Avoid calling kfree() under pidmap spinlock, calling it afterwards.
Normally kfree() is fast, but sometimes it can be slow, so avoid
calling it under the spinlock if we can do it.
Signed-off-by: André Goddard Rosa <andre.goddard@gmail.com>
Cc: Pekka Enberg <penberg@cs.helsinki.fi>
Cc: Oleg Nesterov <oleg@redhat.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
Currently all architectures but microblaze unconditionally define
USE_ELF_CORE_DUMP. The microblaze omission seems like an error to me, so
let's kill this ifdef and make sure we are the same everywhere.
Signed-off-by: Christoph Hellwig <hch@lst.de>
Acked-by: Hugh Dickins <hugh.dickins@tiscali.co.uk>
Cc: <linux-arch@vger.kernel.org>
Cc: Michal Simek <michal.simek@petalogix.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
KCS_IDLE and KCS_IDLE state have the same value, but in this function the
constants ending in _STATE are compared to the state variable.
Signed-off-by: Julia Lawall <julia@diku.dk>
Acked-by: Core Minyard <cminyard@mvista.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
We have HARD_MSGMAX lower on 64bit than on 32bit, since usually 64bit
machines have more memory than 32bit machines.
Making it higher on 64bit seems reasonable, and keep the original number
on 32bit.
Acked-by: Serge E. Hallyn <serue@us.ibm.com>
Cc: Cedric Le Goater <clg@fr.ibm.com>
Signed-off-by: WANG Cong <amwang@redhat.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
This line is unreachable, remove it.
[akpm@linux-foundation.org: remove unneeded initialisation of `err']
Signed-off-by: WANG Cong <amwang@redhat.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
If multiple simple decrements on the same semaphore are pending, then the
current code scans all decrement operations, even if the semaphore value
is already 0.
The patch optimizes that: if the semaphore value is 0, then there is no
need to scan the q->alter entries.
Note that this is a common case: It happens if 100 decrements by one are
pending and now an increment by one increases the semaphore value from 0
to 1. Without this patch, all 100 entries are scanned. With the patch,
only one entry is scanned, then woken up. Then the new rule triggers and
the scanning is aborted, without looking at the remaining 99 tasks.
With this patch, single sop increment/decrement by 1 are now O(1).
(same as with Nick's patch)
Signed-off-by: Manfred Spraul <manfred@colorfullife.com>
Cc: Nick Piggin <npiggin@suse.de>
Cc: Pierre Peiffer <peifferp@gmail.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
sysv sem has the concept of semaphore arrays that consist out of multiple
semaphores. Atomic operations that affect multiple semaphores are
supported.
The patch optimizes single semaphore operation calls that affect only one
semaphore: It's not necessary to scan all pending operations, it is
sufficient to scan the per-semaphore list.
The idea is from Nick Piggin version of an ipc sem improvement, the
implementation is different: The code tries to keep as much common code as
possible.
As the result, the patch is simpler, but optimizes fewer cases.
Signed-off-by: Manfred Spraul <manfred@colorfullife.com>
Cc: Nick Piggin <npiggin@suse.de>
Cc: Pierre Peiffer <peifferp@gmail.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
Based on Nick's findings:
sysv sem has the concept of semaphore arrays that consist out of multiple
semaphores. Atomic operations that affect multiple semaphores are
supported.
The patch is the first step for optimizing simple, single semaphore
operations: In addition to the global list of all pending operations, a
2nd, per-semaphore list with the simple operations is added.
Note: this patch does not make sense by itself, the new list is used
nowhere.
Signed-off-by: Manfred Spraul <manfred@colorfullife.com>
Cc: Nick Piggin <npiggin@suse.de>
Cc: Pierre Peiffer <peifferp@gmail.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
Reduce the amount of scanning of the list of pending semaphore operations:
If try_atomic_semop failed, then no changes were applied. Thus no need to
restart.
Additionally, this patch correct an incorrect comment: It's possible to
wait for arbitrary semaphore values (do a dec by <x>, wait-for-zero, inc
by <x> in one atomic operation)
Both changes are from Nick Piggin, the patch is the result of a different
split of the individual changes.
Signed-off-by: Manfred Spraul <manfred@colorfullife.com>
Cc: Nick Piggin <npiggin@suse.de>
Cc: Pierre Peiffer <peifferp@gmail.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
The strange sysv semaphore wakeup scheme has a kind of busy-wait lock
involved, which could deadlock if preemption is enabled during the "lock".
It is an implementation detail (due to a spinlock being held) that this is
actually the case. However if "spinlocks" are made preemptible, or if the
sem lock is changed to a sleeping lock for example, then the wakeup would
become buggy. So this might be a bugfix for -rt kernels.
Imagine waker being preempted by wakee and never clearing IN_WAKEUP -- if
wakee has higher RT priority then there is a priority inversion deadlock.
Even if there is not a priority inversion to cause a deadlock, then there
is still time wasted spinning.
Signed-off-by: Nick Piggin <npiggin@suse.de>
Signed-off-by: Manfred Spraul <manfred@colorfullife.com>
Cc: Pierre Peiffer <peifferp@gmail.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
Replace the handcoded list operations in update_queue() with the standard
list_for_each_entry macros.
list_for_each_entry_safe() must be used, because list entries can
disappear immediately uppon the wakeup event.
Signed-off-by: Nick Piggin <npiggin@suse.de>
Signed-off-by: Manfred Spraul <manfred@colorfullife.com>
Cc: Pierre Peiffer <peifferp@gmail.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
Around a month ago, there was some discussion about an improvement of the
sysv sem algorithm: Most (at least: some important) users only use simple
semaphore operations, therefore it's worthwile to optimize this use case.
This patch:
Move last looked up sem_undo struct to the head of the task's undo list.
Attempt to move common entries to the front of the list so search time is
reduced. This reduces lookup_undo on oprofile of problematic SAP workload
by 30% (see patch 4 for a description of SAP workload).
Signed-off-by: Nick Piggin <npiggin@suse.de>
Signed-off-by: Manfred Spraul <manfred@colorfullife.com>
Cc: Pierre Peiffer <peifferp@gmail.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
We have apparently had a memory leak since
7ca7e564e0 "ipc: store ipcs into IDRs" in
2007. The idr of which 3 exist for each ipc namespace is never freed.
This patch simply frees them when the ipcns is freed. I don't believe any
idr_remove() are done from rcu (and could therefore be delayed until after
this idr_destroy()), so the patch should be safe. Some quick testing
showed no harm, and the memory leak fixed.
Caught by kmemleak.
Signed-off-by: Serge E. Hallyn <serue@us.ibm.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
Move the call to do_signal_stop() down, after tracehook call. This makes
->group_stop_count condition visible to tracers before do_signal_stop()
will participate in this group-stop.
Currently the patch has no effect, tracehook_get_signal() always returns 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>
Trivial, s/0/SI_USER/ in collect_signal() for grep.
This is a bit confusing, we don't know the source of this signal.
But we don't care, and "info->si_code = 0" is imho worse.
Signed-off-by: Oleg Nesterov <oleg@redhat.com>
Cc: Roland McGrath <roland@redhat.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>
Change send_signal() to use si_fromuser(). From now SEND_SIG_NOINFO
triggers the "from_ancestor_ns" check.
This fixes reparent_thread()->group_send_sig_info(pdeath_signal)
behaviour, before this patch send_signal() does not detect the
cross-namespace case when the child of the dying parent belongs to the
sub-namespace.
This patch can affect the behaviour of send_sig(), kill_pgrp() and
kill_pid() when the caller sends the signal to the sub-namespace with
"priv == 0" but surprisingly all callers seem to use them correctly,
including disassociate_ctty(on_exit).
Except: drivers/staging/comedi/drivers/addi-data/*.c incorrectly use
send_sig(priv => 0). But his is minor and should be fixed anyway.
Reported-by: Daniel Lezcano <dlezcano@fr.ibm.com>
Signed-off-by: Oleg Nesterov <oleg@redhat.com>
Cc: Roland McGrath <roland@redhat.com>
Reviewed-by: 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 changes in compiled code. The patch adds the new helper, si_fromuser()
and changes check_kill_permission() to use this helper.
The real effect of this patch is that from now we "officially" consider
SEND_SIG_NOINFO signal as "from user-space" signals. This is already true
if we look at the code which uses SEND_SIG_NOINFO, except __send_signal()
has another opinion - see the next patch.
The naming of these special SEND_SIG_XXX siginfo's is really bad
imho. From __send_signal()'s pov they mean
SEND_SIG_NOINFO from user
SEND_SIG_PRIV from kernel
SEND_SIG_FORCED no info
Signed-off-by: Oleg Nesterov <oleg@redhat.com>
Cc: Roland McGrath <roland@redhat.com>
Reviewed-by: Sukadev Bhattiprolu <sukadev@us.ibm.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
Suggested by Roland.
Unlike powepc, x86 always calls tracehook_report_syscall_exit(step) with
step = 0, and sends the trap by hand.
This results in unnecessary SIGTRAP when PTRACE_SINGLESTEP follows the
syscall-exit stop.
Change syscall_trace_leave() to pass the correct "step" argument to
tracehook and remove the send_sigtrap() logic.
Signed-off-by: Oleg Nesterov <oleg@redhat.com>
Acked-by: Roland McGrath <roland@redhat.com>
Cc: <linux-arch@vger.kernel.org>
Cc: Ingo Molnar <mingo@elte.hu>
Cc: Thomas Gleixner <tglx@linutronix.de>
Cc: "H. Peter Anvin" <hpa@zytor.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
Suggested by Roland.
Implement user_single_step_siginfo() for x86. Extract this code from
send_sigtrap().
Since x86 calls tracehook_report_syscall_exit(step => 0) the new helper is
not used yet.
Signed-off-by: Oleg Nesterov <oleg@redhat.com>
Acked-by: Roland McGrath <roland@redhat.com>
Cc: <linux-arch@vger.kernel.org>
Cc: Ingo Molnar <mingo@elte.hu>
Cc: Thomas Gleixner <tglx@linutronix.de>
Cc: "H. Peter Anvin" <hpa@zytor.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
Suggested by Roland.
Change tracehook_report_syscall_exit() to look at step flag and send the
trap signal if needed.
This change affects ia64, microblaze, parisc, powerpc, sh. They pass
nonzero "step" argument to tracehook but since it was ignored the tracee
reports via ptrace_notify(), this is not right and not consistent.
- PTRACE_SETSIGINFO doesn't work
- if the tracer resumes the tracee with signr != 0 the new signal
is generated rather than delivering it
- If PT_TRACESYSGOOD is set the tracee reports the wrong exit_code
I don't have a powerpc machine, but I think this test-case should see the
difference:
#include <unistd.h>
#include <sys/ptrace.h>
#include <sys/wait.h>
#include <assert.h>
#include <stdio.h>
int main(void)
{
int pid, status;
if (!(pid = fork())) {
assert(ptrace(PTRACE_TRACEME) == 0);
kill(getpid(), SIGSTOP);
getppid();
return 0;
}
assert(pid == wait(&status));
assert(ptrace(PTRACE_SETOPTIONS, pid, 0, PTRACE_O_TRACESYSGOOD) == 0);
assert(ptrace(PTRACE_SYSCALL, pid, 0,0) == 0);
assert(pid == wait(&status));
assert(ptrace(PTRACE_SINGLESTEP, pid, 0,0) == 0);
assert(pid == wait(&status));
if (status == 0x57F)
return 0;
printf("kernel bug: status=%X shouldn't have 0x80\n", status);
return 1;
}
Signed-off-by: Oleg Nesterov <oleg@redhat.com>
Acked-by: Roland McGrath <roland@redhat.com>
Cc: <linux-arch@vger.kernel.org>
Cc: Ingo Molnar <mingo@elte.hu>
Cc: Thomas Gleixner <tglx@linutronix.de>
Cc: "H. Peter Anvin" <hpa@zytor.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
Suggested by Roland.
Currently there is no way to synthesize a single-stepping trap in the
arch-independent manner. This patch adds the default helper which fills
siginfo_t, arch/ can can override it.
Architetures which implement user_enable_single_step() should add
user_single_step_siginfo() also.
Signed-off-by: Oleg Nesterov <oleg@redhat.com>
Acked-by: Roland McGrath <roland@redhat.com>
Cc: <linux-arch@vger.kernel.org>
Cc: Ingo Molnar <mingo@elte.hu>
Cc: Thomas Gleixner <tglx@linutronix.de>
Cc: "H. Peter Anvin" <hpa@zytor.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>
No functional changes.
ptrace_init_task() looks confusing, as if we always auto-attach when "bool
ptrace" argument is true, while in fact we attach only if current is
traced.
Make the code more explicit and kill now unused ptrace_link().
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>
Variable `progress' isn't used in mem_cgroup_resize_limit() any more.
Remove it.
[akpm@linux-foundation.org: cleanup]
Signed-off-by: Bob Liu <lliubbo@gmail.com>
Cc: Daisuke Nishimura <nishimura@mxp.nes.nec.co.jp>
Reviewed-by: Balbir Singh <balbir@linux.vnet.ibm.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
memcg_tasklist was introduced at commit 7f4d454d(memcg: avoid deadlock
caused by race between oom and cpuset_attach) instead of cgroup_mutex to
fix a deadlock problem. The cgroup_mutex, which was removed by the
commit, in mem_cgroup_out_of_memory() was originally introduced at commit
c7ba5c9e (Memory controller: OOM handling).
IIUC, the intention of this cgroup_mutex was to prevent task move during
select_bad_process() so that situations like below can be avoided.
Assume cgroup "foo" has exceeded its limit and is about to trigger oom.
1. Process A, which has been in cgroup "baa" and uses large memory, is just
moved to cgroup "foo". Process A can be the candidates for being killed.
2. Process B, which has been in cgroup "foo" and uses large memory, is just
moved from cgroup "foo". Process B can be excluded from the candidates for
being killed.
But these race window exists anyway even if we hold a lock, because
__mem_cgroup_try_charge() decides wether it should trigger oom or not
outside of the lock. So the original cgroup_mutex in
mem_cgroup_out_of_memory and thus current memcg_tasklist has no use. And
IMHO, those races are not so critical for users.
This patch removes it and make codes simpler.
Signed-off-by: Daisuke Nishimura <nishimura@mxp.nes.nec.co.jp>
Cc: Balbir Singh <balbir@linux.vnet.ibm.com>
Acked-by: KAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
task_in_mem_cgroup(), which is called by select_bad_process() to check
whether a task can be a candidate for being oom-killed from memcg's limit,
checks "curr->use_hierarchy"("curr" is the mem_cgroup the task belongs
to).
But this check return true(it's false positive) when:
<some path>/aa use_hierarchy == 0 <- hitting limit
<some path>/aa/00 use_hierarchy == 1 <- the task belongs to
This leads to killing an innocent task in aa/00. This patch is a fix for
this bug. And this patch also fixes the arg for
mem_cgroup_print_oom_info(). We should print information of mem_cgroup
which the task being killed, not current, belongs to.
Signed-off-by: Daisuke Nishimura <nishimura@mxp.nes.nec.co.jp>
Acked-by: Balbir Singh <balbir@linux.vnet.ibm.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
mem_cgroup_move_parent() calls try_charge first and cancel_charge on
failure. IMHO, charge/uncharge(especially charge) is high cost operation,
so we should avoid it as far as possible.
This patch tries to delay try_charge in mem_cgroup_move_parent() by
re-ordering checks it does.
And this patch renames mem_cgroup_move_account() to
__mem_cgroup_move_account(), changes the return value of
__mem_cgroup_move_account() from int to void, and adds a new
wrapper(mem_cgroup_move_account()), which checks whether a @pc is valid
for moving account and calls __mem_cgroup_move_account().
This patch removes the last caller of trylock_page_cgroup(), so removes
its definition too.
Signed-off-by: Daisuke Nishimura <nishimura@mxp.nes.nec.co.jp>
Acked-by: KAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com>
Cc: Balbir Singh <balbir@linux.vnet.ibm.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
There are some places calling both res_counter_uncharge() and css_put() to
cancel the charge and the refcnt we have got by mem_cgroup_tyr_charge().
This patch introduces mem_cgroup_cancel_charge() and call it in those
places.
Signed-off-by: KAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com>
Signed-off-by: Daisuke Nishimura <nishimura@mxp.nes.nec.co.jp>
Reviewed-by: Balbir Singh <balbir@linux.vnet.ibm.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
In global VM, FILE_MAPPED is used but memcg uses MAPPED_FILE. This makes
grep difficult. Replace memcg's MAPPED_FILE with FILE_MAPPED
And in global VM, mapped shared memory is accounted into FILE_MAPPED.
But memcg doesn't. fix it.
Note:
page_is_file_cache() just checks SwapBacked or not.
So, we need to check PageAnon.
Cc: Balbir Singh <balbir@in.ibm.com>
Reviewed-by: Daisuke Nishimura <nishimura@mxp.nes.nec.co.jp>
Signed-off-by: KAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
This is a patch for coalescing access to res_counter at charging by percpu
caching. At charge, memcg charges 64pages and remember it in percpu
cache. Because it's cache, drain/flush if necessary.
This version uses public percpu area.
2 benefits for using public percpu area.
1. Sum of stocked charge in the system is limited to # of cpus
not to the number of memcg. This shows better synchonization.
2. drain code for flush/cpuhotplug is very easy (and quick)
The most important point of this patch is that we never touch res_counter
in fast path. The res_counter is system-wide shared counter which is modified
very frequently. We shouldn't touch it as far as we can for avoiding
false sharing.
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 cache miss/faults
[in a child cgroup]
18444157 page-faults # 0.039 M/sec ( +- 0.133% )
69.96 cache miss/faults
[ + coalescing uncharge patch]
27133719 page-faults # 0.057 M/sec ( +- 0.155% )
47.16 cache miss/faults
[ + coalescing uncharge patch + this patch ]
34224709 page-faults # 0.072 M/sec ( +- 0.173% )
34.69 cache miss/faults
Changelog (since Oct/2):
- updated comments
- replaced get_cpu_var() with __get_cpu_var() if possible.
- removed mutex for system-wide drain. adds a counter instead of it.
- removed CONFIG_HOTPLUG_CPU
Changelog (old):
- rebased onto the latest mmotm
- moved charge size check before __GFP_WAIT check for avoiding unnecesary
- added asynchronous flush routine.
- fixed bugs pointed out by Nishimura-san.
[akpm@linux-foundation.org: tweak comments]
[nishimura@mxp.nes.nec.co.jp: don't do INIT_WORK() repeatedly against the same work_struct]
Signed-off-by: KAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com>
Cc: Balbir Singh <balbir@in.ibm.com>
Signed-off-by: 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>
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>
A memory cgroup has a memory.memsw.usage_in_bytes file. It shows the sum
of the usage of pages and swapents in the cgroup. Presently the root
cgroup's memsw.usage_in_bytes shows the wrong value - the number of
swapents are not added.
So take MEM_CGROUP_STAT_SWAPOUT into account.
Signed-off-by: Kirill A. Shutemov <kirill@shutemov.name>
Reviewed-by: Daisuke Nishimura <nishimura@mxp.nes.nec.co.jp>
Acked-by: KAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com>
Cc: <stable@kernel.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
Using create_proc_entry() + ->proc_fops assignment is racy because
->proc_fops will be NULL for some time, use proc_create() to avoid race.
Signed-off-by: Alexey Dobriyan <adobriyan@gmail.com>
Cc: Randy Dunlap <randy.dunlap@oracle.com>
Cc: Jonathan Corbet <corbet@lwn.net>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
Example is outdated, it still uses old ->read_proc interfaces and "fb"
example is plain racy. There are better examples all over the tree.
Docbook itself says almost nothing about /proc and contain quite a number
of simply wrong facts, e.g. device nodes support. What it does is
describing at great length interface which are going to be removed.
There are Documentation/filesystems/seq_file.txt in exchange.
Signed-off-by: Alexey Dobriyan <adobriyan@gmail.com>
Acked-by: Erik Mouw <mouw@nl.linux.org>
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>
If a patch adds ioctls, then Documentation/ioctl/ioctl-number.txt
should also be updated.
Remove reference to the OSDL PLM build farm.
Signed-off-by: Randy Dunlap <randy.dunlap@oracle.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
It is not necessary to write custom code for convert calendar time to
broken-down time. time_to_tm() is more generic to do that.
Signed-off-by: Zhao Lei <zhaolei@cn.fujitsu.com>
Cc: OGAWA Hirofumi <hirofumi@mail.parknet.co.jp>
Cc: Thomas Gleixner <tglx@linutronix.de>
Cc: Ingo Molnar <mingo@elte.hu>
Cc: Pavel Machek <pavel@ucw.cz>
Cc: Andi Kleen <andi@firstfloor.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
Use bitmap_weight instead of doing hweight32 for each 32bit in bitmap.
Signed-off-by: Akinobu Mita <akinobu.mita@gmail.com>
Cc: Mikulas Patocka <mikulas@artax.karlin.mff.cuni.cz>
Cc: Al Viro <viro@zeniv.linux.org.uk>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
Use hweight32 instead of counting for each bit
Signed-off-by: Akinobu Mita <akinobu.mita@gmail.com>
Cc: Mikulas Patocka <mikulas@artax.karlin.mff.cuni.cz>
Cc: Al Viro <viro@zeniv.linux.org.uk>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
Jeff Moyer