lkml.org 
[lkml]   [2010]   [Nov]   [3]   [last100]   RSS Feed
Views: [wrap][no wrap]   [headers]  [forward] 
 
Messages in this thread
Patch in this message
/
From
Subject[PATCH 19/20] x86/ticketlocks: use overlapping read to eliminate mb()
Date
From: Jeremy Fitzhardinge <jeremy.fitzhardinge@citrix.com>

When reading the 'waiting' counter, use a longer-than-necessary read
which also overlaps 'head'. This read is guaranteed to be in-order
with respect to and unlock writes to 'head', thereby eliminating the
need for an explicit mb() to enforce the read-after-write ordering.

Signed-off-by: Jeremy Fitzhardinge <jeremy.fitzhardinge@citrix.com>
---
arch/x86/include/asm/spinlock.h | 14 +++++++-------
arch/x86/include/asm/spinlock_types.h | 24 ++++++++++++++++++++----
2 files changed, 27 insertions(+), 11 deletions(-)
diff --git a/arch/x86/include/asm/spinlock.h b/arch/x86/include/asm/spinlock.h
index c7455e1..a16b6e4 100644
--- a/arch/x86/include/asm/spinlock.h
+++ b/arch/x86/include/asm/spinlock.h
@@ -86,14 +86,14 @@ static inline void __ticket_sub_waiting(struct arch_spinlock *lock)

static __always_inline bool __ticket_lock_waiters(const struct arch_spinlock *lock)
{
- /*
- * Make sure everyone sees the unlock write before we check the
- * waiting count. The processor ordering doesn't guarantee this
- * because they're different memory locations.
+ /*
+ * lock->waiting_head is a union element which aliases both
+ * 'waiting' and 'head'. Since the memory access overlaps
+ * 'head', the read is forced to be in-order with respect to
+ * unlock writes to 'head', eliminating the need for an
+ * explicit mb(). (Intel memory ordering rules.)
*/
- mb();
-
- return ACCESS_ONCE(lock->waiting) != 0;
+ return ((__ticket_t)ACCESS_ONCE(lock->waiting_head)) != 0;
}
#endif /* CONFIG_PARAVIRT_SPINLOCKS */

diff --git a/arch/x86/include/asm/spinlock_types.h b/arch/x86/include/asm/spinlock_types.h
index def8010..307ef0b 100644
--- a/arch/x86/include/asm/spinlock_types.h
+++ b/arch/x86/include/asm/spinlock_types.h
@@ -18,6 +18,24 @@ typedef u32 __ticketpair_t;
#define TICKET_SHIFT (sizeof(__ticket_t) * 8)
#define TICKET_MASK ((1 << TICKET_SHIFT) - 1)

+#ifdef CONFIG_PARAVIRT_SPINLOCKS
+typedef struct arch_spinlock {
+ union {
+ struct {
+ __ticket_t waiting;
+ union {
+ __ticketpair_t ticketpair;
+ struct __raw_tickets {
+ __ticket_t head, tail;
+ } tickets;
+ };
+ };
+ __ticketpair_t waiting_head; /* aliases waiting and head */
+ };
+} arch_spinlock_t;
+
+#define __ARCH_SPIN_LOCK_UNLOCKED { { { 0, { 0 } } } }
+#else
typedef struct arch_spinlock {
union {
__ticketpair_t ticketpair;
@@ -25,12 +43,10 @@ typedef struct arch_spinlock {
__ticket_t head, tail;
} tickets;
};
-#ifdef CONFIG_PARAVIRT_SPINLOCKS
- __ticket_t waiting;
-#endif
} arch_spinlock_t;

-#define __ARCH_SPIN_LOCK_UNLOCKED { { .tickets = {0, 0} } }
+#define __ARCH_SPIN_LOCK_UNLOCKED { { 0 } }
+#endif

typedef struct {
unsigned int lock;
--
1.7.2.3


\
 
 \ /
  Last update: 2010-11-03 16:13    [from the cache]
©2003-2011 Jasper Spaans