* [PATCH] signals: Avoid unnecessary taking of sighand->siglock
@ 2016-09-22 18:25 Waiman Long
2016-09-23 9:32 ` Oleg Nesterov
0 siblings, 1 reply; 3+ messages in thread
From: Waiman Long @ 2016-09-22 18:25 UTC (permalink / raw)
To: Andrew Morton, Ingo Molnar, Oleg Nesterov, Thomas Gleixner,
Stas Sergeev
Cc: linux-kernel, Scott J Norton, Douglas Hatch, Waiman Long
When running certain database workload on a high-end system with many
CPUs, it was found that spinlock contention in the sigprocmask syscalls
became a significant portion of the overall CPU cycles as shown below.
9.30% 9.30% 905387 dataserver /proc/kcore 0x7fff8163f4d2
[k] _raw_spin_lock_irq
|
---_raw_spin_lock_irq
|
|--99.34%-- __set_current_blocked
| sigprocmask
| sys_rt_sigprocmask
| system_call_fastpath
| |
| |--50.63%-- __swapcontext
| | |
| | |--99.91%-- upsleepgeneric
| |
| |--49.36%-- __setcontext
| | ktskRun
Looking further into the swapcontext function in glibc, it was found
that the function always call sigprocmask() without checking if there
are changes in the signal mask.
This patch adds a check in the __set_current_blocked() function to
avoid taking the sighand->siglock spinlock if there is no change in
the signal mask. This will prevent unneeded spinlock contention when
many threads are trying to call sigprocmask().
With this patch applied, the spinlock contention in sigprocmask() was
gone.
This patch is currently only active for 64-bit architectures.
Signed-off-by: Waiman Long <Waiman.Long@hpe.com>
---
kernel/signal.c | 10 ++++++++++
1 files changed, 10 insertions(+), 0 deletions(-)
diff --git a/kernel/signal.c b/kernel/signal.c
index af21afc..5850b11 100644
--- a/kernel/signal.c
+++ b/kernel/signal.c
@@ -2485,6 +2485,16 @@ void __set_current_blocked(const sigset_t *newset)
{
struct task_struct *tsk = current;
+ /*
+ * In case the signal mask hasn't changed, we won't need to take
+ * the lock. As the current blocked mask can be modified by other
+ * CPUs, we need to do an atomic read without lock. In other words,
+ * this check will only be done on 64-bit systems.
+ */
+#if _NSIG_WORDS == 1
+ if (READ_ONCE(tsk->blocked.sig[0]) == newset->sig[0])
+ return;
+#endif
spin_lock_irq(&tsk->sighand->siglock);
__set_task_blocked(tsk, newset);
spin_unlock_irq(&tsk->sighand->siglock);
--
1.7.1
^ permalink raw reply related [flat|nested] 3+ messages in thread* Re: [PATCH] signals: Avoid unnecessary taking of sighand->siglock
2016-09-22 18:25 [PATCH] signals: Avoid unnecessary taking of sighand->siglock Waiman Long
@ 2016-09-23 9:32 ` Oleg Nesterov
2016-09-26 21:21 ` Waiman Long
0 siblings, 1 reply; 3+ messages in thread
From: Oleg Nesterov @ 2016-09-23 9:32 UTC (permalink / raw)
To: Waiman Long
Cc: Andrew Morton, Ingo Molnar, Thomas Gleixner, Stas Sergeev,
linux-kernel, Scott J Norton, Douglas Hatch
On 09/22, Waiman Long wrote:
>
> This patch is currently only active for 64-bit architectures.
Why?
> --- a/kernel/signal.c
> +++ b/kernel/signal.c
> @@ -2485,6 +2485,16 @@ void __set_current_blocked(const sigset_t *newset)
> {
> struct task_struct *tsk = current;
>
> + /*
> + * In case the signal mask hasn't changed, we won't need to take
> + * the lock. As the current blocked mask can be modified by other
> + * CPUs,
No, nobody else should modify current->blocked.
Yes, we need to cleanup the usage of force_sig_info(), and probably remove
the "struct task_struct *t" argument.
> we need to do an atomic read without lock. In other words,
> + * this check will only be done on 64-bit systems.
> + */
> +#if _NSIG_WORDS == 1
> + if (READ_ONCE(tsk->blocked.sig[0]) == newset->sig[0])
> + return;
> +#endif
OK, agreed, but this should not depend on _NSIG_WORDS == 1 and
READ_ONCE() looks confusing. It seems you need to add the new helper
into include/linux/signal.h.
Oleg.
^ permalink raw reply [flat|nested] 3+ messages in thread* Re: [PATCH] signals: Avoid unnecessary taking of sighand->siglock
2016-09-23 9:32 ` Oleg Nesterov
@ 2016-09-26 21:21 ` Waiman Long
0 siblings, 0 replies; 3+ messages in thread
From: Waiman Long @ 2016-09-26 21:21 UTC (permalink / raw)
To: Oleg Nesterov
Cc: Andrew Morton, Ingo Molnar, Thomas Gleixner, Stas Sergeev,
linux-kernel, Scott J Norton, Douglas Hatch
On 09/23/2016 05:32 AM, Oleg Nesterov wrote:
> On 09/22, Waiman Long wrote:
>> This patch is currently only active for 64-bit architectures.
> Why?
>
>> --- a/kernel/signal.c
>> +++ b/kernel/signal.c
>> @@ -2485,6 +2485,16 @@ void __set_current_blocked(const sigset_t *newset)
>> {
>> struct task_struct *tsk = current;
>>
>> + /*
>> + * In case the signal mask hasn't changed, we won't need to take
>> + * the lock. As the current blocked mask can be modified by other
>> + * CPUs,
> No, nobody else should modify current->blocked.
Thanks for the clarification. I am just not 100% sure about that. I will
modify the comment.
>
> Yes, we need to cleanup the usage of force_sig_info(), and probably remove
> the "struct task_struct *t" argument.
>
>> we need to do an atomic read without lock. In other words,
>> + * this check will only be done on 64-bit systems.
>> + */
>> +#if _NSIG_WORDS == 1
>> + if (READ_ONCE(tsk->blocked.sig[0]) == newset->sig[0])
>> + return;
>> +#endif
> OK, agreed, but this should not depend on _NSIG_WORDS == 1 and
> READ_ONCE() looks confusing. It seems you need to add the new helper
> into include/linux/signal.h.
Yes, I will add a helper into signal.h and call it instead.
Cheers,
Longman
^ permalink raw reply [flat|nested] 3+ messages in thread
end of thread, other threads:[~2016-09-26 21:55 UTC | newest]
Thread overview: 3+ messages (download: mbox.gz follow: Atom feed
-- links below jump to the message on this page --
2016-09-22 18:25 [PATCH] signals: Avoid unnecessary taking of sighand->siglock Waiman Long
2016-09-23 9:32 ` Oleg Nesterov
2016-09-26 21:21 ` Waiman Long
This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox;
as well as URLs for NNTP newsgroup(s).