summaryrefslogtreecommitdiff
path: root/arch/ia64
diff options
context:
space:
mode:
authorTony Luck <tony.luck@intel.com>2009-10-07 17:54:19 (GMT)
committerTony Luck <tony.luck@intel.com>2009-10-07 17:54:19 (GMT)
commit9d40ee200a527ce08ab8c793ba8ae3e242edbb0e (patch)
tree189b7dc9cde0ceb791bb3bdd1a1d9a22aee00336 /arch/ia64
parent0eca52a92735f43462165efe00a7e394345fb38e (diff)
downloadlinux-9d40ee200a527ce08ab8c793ba8ae3e242edbb0e.tar.xz
[IA64] Squeeze ticket locks back into 4 bytes.
Linus pointed out that other people have spent large amounts of time and effort to optimize the layout of frequently used structures. Often these have embedded locks, and the assumption is that a lock takes 4 bytes. Linus also pointed out how to work with the limited options for atomic instructions on Itanium. Signed-off-by: Tony Luck <tony.luck@intel.com>
Diffstat (limited to 'arch/ia64')
-rw-r--r--arch/ia64/include/asm/spinlock.h45
-rw-r--r--arch/ia64/include/asm/spinlock_types.h2
2 files changed, 27 insertions, 20 deletions
diff --git a/arch/ia64/include/asm/spinlock.h b/arch/ia64/include/asm/spinlock.h
index 30bb930..4fa5027 100644
--- a/arch/ia64/include/asm/spinlock.h
+++ b/arch/ia64/include/asm/spinlock.h
@@ -25,61 +25,68 @@
* by atomically noting the tail and incrementing it by one (thus adding
* ourself to the queue and noting our position), then waiting until the head
* becomes equal to the the initial value of the tail.
+ * The pad bits in the middle are used to prevent the next_ticket number
+ * overflowing into the now_serving number.
*
- * 63 32 31 0
+ * 31 17 16 15 14 0
* +----------------------------------------------------+
- * | next_ticket_number | now_serving |
+ * | now_serving | padding | next_ticket |
* +----------------------------------------------------+
*/
-#define TICKET_SHIFT 32
+#define TICKET_SHIFT 17
+#define TICKET_BITS 15
+#define TICKET_MASK ((1 << TICKET_BITS) - 1)
static __always_inline void __ticket_spin_lock(raw_spinlock_t *lock)
{
- int *p = (int *)&lock->lock, turn, now_serving;
+ int *p = (int *)&lock->lock, ticket, serve;
- now_serving = *p;
- turn = ia64_fetchadd(1, p+1, acq);
+ ticket = ia64_fetchadd(1, p, acq);
- if (turn == now_serving)
+ if (!(((ticket >> TICKET_SHIFT) ^ ticket) & TICKET_MASK))
return;
- do {
+ ia64_invala();
+
+ for (;;) {
+ asm volatile ("ld4.c.nc %0=[%1]" : "=r"(serve) : "r"(p) : "memory");
+
+ if (!(((serve >> TICKET_SHIFT) ^ ticket) & TICKET_MASK))
+ return;
cpu_relax();
- } while (ACCESS_ONCE(*p) != turn);
+ }
}
static __always_inline int __ticket_spin_trylock(raw_spinlock_t *lock)
{
- long tmp = ACCESS_ONCE(lock->lock), try;
+ int tmp = ACCESS_ONCE(lock->lock);
- if (!(((tmp >> TICKET_SHIFT) ^ tmp) & ((1L << TICKET_SHIFT) - 1))) {
- try = tmp + (1L << TICKET_SHIFT);
-
- return ia64_cmpxchg(acq, &lock->lock, tmp, try, sizeof (tmp)) == tmp;
- }
+ if (!(((tmp >> TICKET_SHIFT) ^ tmp) & TICKET_MASK))
+ return ia64_cmpxchg(acq, &lock->lock, tmp, tmp + 1, sizeof (tmp)) == tmp;
return 0;
}
static __always_inline void __ticket_spin_unlock(raw_spinlock_t *lock)
{
- int *p = (int *)&lock->lock;
+ unsigned short *p = (unsigned short *)&lock->lock + 1, tmp;
- (void)ia64_fetchadd(1, p, rel);
+ asm volatile ("ld2.bias %0=[%1]" : "=r"(tmp) : "r"(p));
+ ACCESS_ONCE(*p) = (tmp + 2) & ~1;
}
static inline int __ticket_spin_is_locked(raw_spinlock_t *lock)
{
long tmp = ACCESS_ONCE(lock->lock);
- return !!(((tmp >> TICKET_SHIFT) ^ tmp) & ((1L << TICKET_SHIFT) - 1));
+ return !!(((tmp >> TICKET_SHIFT) ^ tmp) & TICKET_MASK);
}
static inline int __ticket_spin_is_contended(raw_spinlock_t *lock)
{
long tmp = ACCESS_ONCE(lock->lock);
- return (((tmp >> TICKET_SHIFT) - tmp) & ((1L << TICKET_SHIFT) - 1)) > 1;
+ return ((tmp - (tmp >> TICKET_SHIFT)) & TICKET_MASK) > 1;
}
static inline int __raw_spin_is_locked(raw_spinlock_t *lock)
diff --git a/arch/ia64/include/asm/spinlock_types.h b/arch/ia64/include/asm/spinlock_types.h
index b61d136..474e46f 100644
--- a/arch/ia64/include/asm/spinlock_types.h
+++ b/arch/ia64/include/asm/spinlock_types.h
@@ -6,7 +6,7 @@
#endif
typedef struct {
- volatile unsigned long lock;
+ volatile unsigned int lock;
} raw_spinlock_t;
#define __RAW_SPIN_LOCK_UNLOCKED { 0 }