[IA64] Optimize ticket spinlocks in fsys_rt_sigprocmask

Tony's fix (f574c84319) has a small bug,
it incorrectly uses "r3" as a scratch register in the first of the two
unlock paths ... it is also inefficient.  Optimize the fast path again.

Signed-off-by: Petr Tesarik <ptesarik@suse.cz>
Signed-off-by: Tony Luck <tony.luck@intel.com>
This commit is contained in:
Petr Tesarik 2010-09-15 15:35:48 -07:00 коммит произвёл Tony Luck
Родитель bfa88ea7ee
Коммит 2d2b690164
1 изменённых файлов: 13 добавлений и 33 удалений

Просмотреть файл

@ -420,34 +420,31 @@ EX(.fail_efault, ld8 r14=[r33]) // r14 <- *set
;;
RSM_PSR_I(p0, r18, r19) // mask interrupt delivery
mov ar.ccv=0
andcm r14=r14,r17 // filter out SIGKILL & SIGSTOP
mov r8=EINVAL // default to EINVAL
#ifdef CONFIG_SMP
// __ticket_spin_trylock(r31)
ld4 r17=[r31]
mov r8=EINVAL // default to EINVAL
;;
extr r9=r17,17,15
;;
xor r18=r17,r9
adds r19=1,r17
;;
extr.u r18=r18,0,15
;;
cmp.eq p0,p7=0,r18
(p7) br.cond.spnt.many .lock_contention
mov.m ar.ccv=r17
extr.u r9=r17,17,15
adds r19=1,r17
extr.u r18=r17,0,15
;;
cmpxchg4.acq r9=[r31],r19,ar.ccv
cmp.eq p6,p7=r9,r18
;;
(p6) cmpxchg4.acq r9=[r31],r19,ar.ccv
(p6) dep.z r20=r19,1,15 // next serving ticket for unlock
(p7) br.cond.spnt.many .lock_contention
;;
cmp4.eq p0,p7=r9,r17
adds r31=2,r31
(p7) br.cond.spnt.many .lock_contention
ld8 r3=[r2] // re-read current->blocked now that we hold the lock
;;
#else
ld8 r3=[r2] // re-read current->blocked now that we hold the lock
mov r8=EINVAL // default to EINVAL
#endif
add r18=IA64_TASK_PENDING_OFFSET+IA64_SIGPENDING_SIGNAL_OFFSET,r16
add r19=IA64_TASK_SIGNAL_OFFSET,r16
@ -503,16 +500,8 @@ EX(.fail_efault, ld8 r14=[r33]) // r14 <- *set
#ifdef CONFIG_SMP
// __ticket_spin_unlock(r31)
adds r31=2,r31
;;
ld2.bias r2=[r31]
mov r3=65534
;;
adds r2=2,r2
;;
and r3=r3,r2
;;
st2.rel [r31]=r3
st2.rel [r31]=r20
mov r20=0 // i must not leak kernel bits...
#endif
SSM_PSR_I(p0, p9, r31)
;;
@ -535,16 +524,7 @@ EX(.fail_efault, (p15) st8 [r34]=r3)
.sig_pending:
#ifdef CONFIG_SMP
// __ticket_spin_unlock(r31)
adds r31=2,r31
;;
ld2.bias r2=[r31]
mov r3=65534
;;
adds r2=2,r2
;;
and r3=r3,r2
;;
st2.rel [r31]=r3
st2.rel [r31]=r20 // release the lock
#endif
SSM_PSR_I(p0, p9, r17)
;;