Re: [lkp] [ipc/sem.c] 99ac0dfffc: aim9.shared_memory.ops_per_sec -8.9% regression

[Manfred Spraul]

Hi,

On 09/06/2016 08:42 AM, kernel test robot wrote:
> FYI, we noticed a -8.9% regression of aim9.shared_memory.ops_per_sec due to commit:
>
> commit 99ac0dfffcfb34326a880e90e06c30a2a882c692 ("ipc/sem.c: fix complex_count vs. simple op race")
> https://git.kernel.org/pub/scm/linux/kernel/git/next/linux-next.git master
>
> in testcase: aim9
> on test machine: 4 threads Intel(R) Core(TM) i3-3220 CPU @ 3.30GHz with 4G memory
> with following parameters:
>
>
> commit:
>    5ae37d6d64 ("config/android: Remove CONFIG_IPV6_PRIVACY")
>    99ac0dfffc ("ipc/sem.c: fix complex_count vs. simple op race")
>
> 5ae37d6d64c605ee 99ac0dfffcfb34326a880e90e0
> ---------------- --------------------------
>         fail:runs  %reproduction    fail:runs
>             |             |             |
>           %stddev     %change         %stddev
>               \          |                \
>     1398301 ±  1%      -8.9%    1274143 ±  0%  aim9.shared_memory.ops_per_sec
The shared memory test uses sem_op for token passing.

I think the root cause is the additional smp_mb() after spin_lock(), at 
least I got +8.9% when I removed the barrier.

#./sem_scalebench -t 30 -c 1 -p 1 -i 1 -h 4 -m 0 -d 1 -o 1 -f

(Note: Just removing the barrier is not an option)

--
    Manfred
diff --git a/ipc/sem.c b/ipc/sem.c
index 5e318c5..85cc5bf 100644
--- a/ipc/sem.c
+++ b/ipc/sem.c
@@ -363,14 +363,6 @@ static inline int sem_lock(struct sem_array *sma, struct sembuf *sops,
 		 */
 		spin_lock(&sem->lock);
 
-		/*
-		 * See 51d7d5205d33
-		 * ("powerpc: Add smp_mb() to arch_spin_is_locked()"):
-		 * A full barrier is required: the write of sem->lock
-		 * must be visible before the read is executed
-		 */
-		smp_mb();
-
 		if (!smp_load_acquire(&sma->complex_mode)) {
 			/* fast path successful! */
 			return sops->sem_num;