Re: [PATCH] ipc/sem.c: Fix complex_count vs. simple op race

[Andrew Morton]

On Thu, 21 Jul 2016 19:54:55 +0200 Manfred Spraul <manfred@xxxxxxxxxxxxxxxx> wrote:

> Next update:
> - switch to smp_store_mb() instead of WRITE_ONCE();smp_mb();
> - introduce SEM_GLOBAL_LOCK instead of magic -1.
> - do not use READ_ONCE() for the unlocked&unordered test:
>   READ_ONCE doesn't make sense for unlocked&unordered code.
> - document why smp_mb() is required after spin_lock().

I assume "ipc/sem.c: remove duplicated memory barriers" is still
relevant?


From: Manfred Spraul <manfred@xxxxxxxxxxxxxxxx>
Subject: ipc/sem.c: remove duplicated memory barriers

With 2c610022711 ("locking/qspinlock: Fix spin_unlock_wait() some more"),
memory barriers were added into spin_unlock_wait().  Thus another barrier
is not required.

And as explained in 055ce0fd1b8 ("locking/qspinlock: Add comments"),
spin_lock() provides a barrier so that reads within the critical section
cannot happen before the write for the lock is visible.  i.e.  spin_lock
provides an acquire barrier after the write of the lock variable, this
barrier pairs with the smp_mb() in complexmode_enter().

Link: http://lkml.kernel.org/r/1468386412-3608-3-git-send-email-manfred@xxxxxxxxxxxxxxxx
Signed-off-by: Manfred Spraul <manfred@xxxxxxxxxxxxxxxx>
Cc: "H. Peter Anvin" <hpa@xxxxxxxxx>
Cc: Peter Zijlstra <peterz@xxxxxxxxxxxxx>
Cc: Davidlohr Bueso <dave@xxxxxxxxxxxx>
Cc: Thomas Gleixner <tglx@xxxxxxxxxxxxx>
Cc: Ingo Molnar <mingo@xxxxxxx>
Cc: <1vier1@xxxxxx>
Cc: <felixh@xxxxxxxxxxxxxxxxxxxxxxxx>
Signed-off-by: Andrew Morton <akpm@xxxxxxxxxxxxxxxxxxxx>
---

 ipc/sem.c |   16 ----------------
 1 file changed, 16 deletions(-)

diff -puN ipc/sem.c~ipc-semc-remove-duplicated-memory-barriers ipc/sem.c
--- a/ipc/sem.c~ipc-semc-remove-duplicated-memory-barriers
+++ a/ipc/sem.c
@@ -290,14 +290,6 @@ static void complexmode_enter(struct sem
 		sem = sma->sem_base + i;
 		spin_unlock_wait(&sem->lock);
 	}
-	/*
-	 * spin_unlock_wait() is not a memory barriers, it is only a
-	 * control barrier. The code must pair with spin_unlock(&sem->lock),
-	 * thus just the control barrier is insufficient.
-	 *
-	 * smp_rmb() is sufficient, as writes cannot pass the control barrier.
-	 */
-	smp_rmb();
 }
 
 /*
@@ -363,14 +355,6 @@ static inline int sem_lock(struct sem_ar
 		 */
 		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;
_