d/freebsd-skq
1
0
Fork 0
Code Issues Pull Requests Packages Projects Releases Wiki Activity

Fix some bugs related to adaptive spinning:

Browse Source 
In the lockmgr support:
- GIANT_RESTORE() is just called when the sleep finishes, so the current
  code can ends up into a giant unlock problem.  Fix it by appropriately
  call GIANT_RESTORE() when needed.  Note that this is not exactly ideal
  because for any interation of the adaptive spinning we drop and restore
  Giant, but the overhead should be not a factor.
- In the lock held in exclusive mode case, after the adaptive spinning is
  brought to completition, we should just retry to acquire the lock
  instead to fallthrough. Fix that.
- Fix a style nit

In the sx support:
- Call GIANT_SAVE() before than looping. This saves some overhead because
  in the current code GIANT_SAVE() is called several times.

Tested by:	Giovanni Trematerra <giovanni dot trematerra at gmail dot com>
This commit is contained in:
attilio 2009-09-02 17:33:51 +00:00
parent 7cb9716a59
commit ce3b4baf3c
2 changed files with 12 additions and 3 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

13
sys/kern/kern_lock.c
View File

@ -467,7 +467,10 @@ __lockmgr_args(struct lock *lk, u_int flags, struct lock_object *ilk,
/*
* If the owner is running on another CPU, spin until
* the owner stops running or the state of the lock
* changes.
* changes. We need a double-state handle here
* because for a failed acquisition the lock can be
* either held in exclusive mode or shared mode
* (for the writer starvation avoidance technique).
*/
if (LK_CAN_ADAPT(lk, flags) && (x & LK_SHARE) == 0 &&
LK_HOLDER(x) != LK_KERNPROC) {
@ -491,8 +494,10 @@ __lockmgr_args(struct lock *lk, u_int flags, struct lock_object *ilk,
while (LK_HOLDER(lk->lk_lock) ==
(uintptr_t)owner && TD_IS_RUNNING(owner))
cpu_spinwait();
GIANT_RESTORE();
continue;
} else if (LK_CAN_ADAPT(lk, flags) &&
(x & LK_SHARE) !=0 && LK_SHARERS(x) &&
(x & LK_SHARE) != 0 && LK_SHARERS(x) &&
spintries < alk_retries) {
if (flags & LK_INTERLOCK) {
class->lc_unlock(ilk);
@ -511,6 +516,7 @@ __lockmgr_args(struct lock *lk, u_int flags, struct lock_object *ilk,
break;
cpu_spinwait();
}
GIANT_RESTORE();
if (i != alk_loops)
continue;
}
@ -704,6 +710,8 @@ __lockmgr_args(struct lock *lk, u_int flags, struct lock_object *ilk,
while (LK_HOLDER(lk->lk_lock) ==
(uintptr_t)owner && TD_IS_RUNNING(owner))
cpu_spinwait();
GIANT_RESTORE();
continue;
} else if (LK_CAN_ADAPT(lk, flags) &&
(x & LK_SHARE) != 0 && LK_SHARERS(x) &&
spintries < alk_retries) {
@ -727,6 +735,7 @@ __lockmgr_args(struct lock *lk, u_int flags, struct lock_object *ilk,
break;
cpu_spinwait();
}
GIANT_RESTORE();
if (i != alk_loops)
continue;
}

2
sys/kern/kern_sx.c
View File

@ -531,13 +531,13 @@ _sx_xlock_hard(struct sx *sx, uintptr_t tid, int opts, const char *file,
continue;
}
} else if (SX_SHARERS(x) && spintries < asx_retries) {
GIANT_SAVE();
spintries++;
for (i = 0; i < asx_loops; i++) {
if (LOCK_LOG_TEST(&sx->lock_object, 0))
CTR4(KTR_LOCK,
"%s: shared spinning on %p with %u and %u",
__func__, sx, spintries, i);
GIANT_SAVE();
x = sx->sx_lock;
if ((x & SX_LOCK_SHARED) == 0 ||
SX_SHARERS(x) == 0)