[PATCH 1/2] ipc/sem.c: Avoid using spin_unlock_wait()

From: Manfred Spraul <manfred@colorfullife.com>
To: Peter Zijlstra <peterz@infradead.org>,
	Andrew Morton <akpm@linux-foundation.org>,
	Davidlohr Bueso <dave@stgolabs.net>
Cc: LKML <linux-kernel@vger.kernel.org>,
	Thomas Gleixner <tglx@linutronix.de>, Ingo Molnar <mingo@elte.hu>,
	"H. Peter Anvin" <hpa@zytor.com>,
	1vier1@web.de, felixh@informatik.uni-bremen.de,
	Manfred Spraul <manfred@colorfullife.com>
Subject: [PATCH 1/2] ipc/sem.c: Avoid using spin_unlock_wait()
Date: Sat,  1 Oct 2016 20:40:16 +0200	[thread overview]
Message-ID: <1475347217-2143-2-git-send-email-manfred@colorfullife.com> (raw)
In-Reply-To: <1475347217-2143-1-git-send-email-manfred@colorfullife.com>

a) The ACQUIRE in spin_lock() applies to the read, not to the store,
at least for powerpc. This forces to add a smp_mb() into the fast
path.

b) The memory barrier provided by spin_unlock_wait() is right now
arch dependent.

Therefore: Use spin_lock()/spin_unlock() instead of spin_unlock_wait().

Advantage: faster single op semop calls(), observed +8.9% on
x86. (the other solution would be arch dependencies in ipc/sem).

Disadvantage: slower complex op semop calls, if (and only if)
there are no sleeping operations.

The next patch adds hysteresis, this further reduces the
probability that the slow path is used.

Signed-off-by: Manfred Spraul <manfred@colorfullife.com>
---
 ipc/sem.c | 25 +++----------------------
 1 file changed, 3 insertions(+), 22 deletions(-)

diff --git a/ipc/sem.c b/ipc/sem.c
index 5e318c5..d5f2710 100644
--- a/ipc/sem.c
+++ b/ipc/sem.c
@@ -280,24 +280,13 @@ static void complexmode_enter(struct sem_array *sma)
 		return;
 	}
 
-	/* We need a full barrier after seting complex_mode:
-	 * The write to complex_mode must be visible
-	 * before we read the first sem->lock spinlock state.
-	 */
-	smp_store_mb(sma->complex_mode, true);
+	sma->complex_mode = true;
 
 	for (i = 0; i < sma->sem_nsems; i++) {
 		sem = sma->sem_base + i;
-		spin_unlock_wait(&sem->lock);
+		spin_lock(&sem->lock);
+		spin_unlock(&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 +352,6 @@ 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;
-- 
2.7.4
