提交 3c79c8b1 编写于 作者: D David Mosberger-Tang 提交者: Tony Luck

[IA64] Schedule fp-clearing insns at least 6 cycles after reading ar.bsp.

Decreases syscall overhead by approximately 6 cycles.
Signed-off-by: NDavid Mosberger-Tang <davidm@hpl.hp.com>
Signed-off-by: NTony Luck <tony.luck@intel.com>
上级 9ec1a7ad
...@@ -705,15 +705,15 @@ ENTRY(ia64_leave_syscall) ...@@ -705,15 +705,15 @@ ENTRY(ia64_leave_syscall)
// start restoring the state saved on the kernel stack (struct pt_regs): // start restoring the state saved on the kernel stack (struct pt_regs):
ld8 r9=[r2],PT(CR_IPSR)-PT(R9) ld8 r9=[r2],PT(CR_IPSR)-PT(R9)
ld8 r11=[r3],PT(CR_IIP)-PT(R11) ld8 r11=[r3],PT(CR_IIP)-PT(R11)
mov f6=f0 // clear f6 nop.i 0
;; ;;
invala // M0|1 invalidate ALAT invala // M0|1 invalidate ALAT
rsm psr.i | psr.ic // M2 initiate turning off of interrupt and interruption collection rsm psr.i | psr.ic // M2 initiate turning off of interrupt and interruption collection
mov f9=f0 // clear f9 nop.i 0
ld8 r29=[r2],16 // load cr.ipsr ld8 r29=[r2],16 // load cr.ipsr
ld8 r28=[r3],16 // load cr.iip ld8 r28=[r3],16 // load cr.iip
mov f8=f0 // clear f8 mov r22=r0 // clear r22
;; ;;
ld8 r30=[r2],16 // M0|1 load cr.ifs ld8 r30=[r2],16 // M0|1 load cr.ifs
ld8 r25=[r3],16 // M0|1 load ar.unat ld8 r25=[r3],16 // M0|1 load ar.unat
...@@ -721,15 +721,15 @@ ENTRY(ia64_leave_syscall) ...@@ -721,15 +721,15 @@ ENTRY(ia64_leave_syscall)
;; ;;
ld8 r26=[r2],PT(B0)-PT(AR_PFS) // M0|1 load ar.pfs ld8 r26=[r2],PT(B0)-PT(AR_PFS) // M0|1 load ar.pfs
(pKStk) mov r22=psr // M2 read PSR now that interrupts are disabled (pKStk) mov r22=psr // M2 read PSR now that interrupts are disabled
mov f10=f0 // clear f10 (pUStk) add r14=IA64_TASK_THREAD_ON_USTACK_OFFSET,r13
;; ;;
ld8 r21=[r2],PT(AR_RNAT)-PT(B0) // load b0 ld8 r21=[r2],PT(AR_RNAT)-PT(B0) // load b0
ld8 r27=[r3],PT(PR)-PT(AR_RSC) // load ar.rsc ld8 r27=[r3],PT(PR)-PT(AR_RSC) // load ar.rsc
mov f11=f0 // clear f11 mov f6=f0 // clear f6
;; ;;
ld8 r24=[r2],PT(AR_FPSR)-PT(AR_RNAT) // load ar.rnat (may be garbage) ld8 r24=[r2],PT(AR_FPSR)-PT(AR_RNAT) // load ar.rnat (may be garbage)
ld8 r31=[r3],PT(R1)-PT(PR) // load predicates ld8 r31=[r3],PT(R1)-PT(PR) // load predicates
(pUStk) add r14=IA64_TASK_THREAD_ON_USTACK_OFFSET,r13 mov f7=f0 // clear f7
;; ;;
ld8 r20=[r2],PT(R12)-PT(AR_FPSR) // load ar.fpsr ld8 r20=[r2],PT(R12)-PT(AR_FPSR) // load ar.fpsr
ld8.fill r1=[r3],16 // load r1 ld8.fill r1=[r3],16 // load r1
...@@ -737,24 +737,29 @@ ENTRY(ia64_leave_syscall) ...@@ -737,24 +737,29 @@ ENTRY(ia64_leave_syscall)
;; ;;
srlz.d // M0 ensure interruption collection is off srlz.d // M0 ensure interruption collection is off
ld8.fill r13=[r3],16 ld8.fill r13=[r3],16
mov f7=f0 // clear f7 mov f8=f0 // clear f8
;; ;;
ld8.fill r12=[r2] // restore r12 (sp) ld8.fill r12=[r2] // restore r12 (sp)
mov.m ar.ssd=r0 // M2 clear ar.ssd mov.m ar.ssd=r0 // M2 clear ar.ssd
mov r22=r0 // clear r22 mov b6=r18 // I0 restore b6
nop.m 0
mov f9=f0 // clear f9
shr.u r18=r19,16 // I0|1 get byte size of existing "dirty" partition
;;
ld8.fill r15=[r3] // restore r15 ld8.fill r15=[r3] // restore r15
(pUStk) st1 [r14]=r17 (pUStk) st1 [r14]=r17
addl r3=THIS_CPU(ia64_phys_stacked_size_p8),r0 addl r3=THIS_CPU(ia64_phys_stacked_size_p8),r0
;; ;;
(pUStk) ld4 r17=[r3] // r17 = cpu_data->phys_stacked_size_p8 (pUStk) ld4 r17=[r3] // r17 = cpu_data->phys_stacked_size_p8
mov.m ar.csd=r0 // M2 clear ar.csd mov.m ar.csd=r0 // M2 clear ar.csd
mov b6=r18 // I0 restore b6 mov f10=f0 // clear f10
;; ;;
mov r14=r0 // clear r14 mov r14=r0 // clear r14
shr.u r18=r19,16 // I0|1 get byte size of existing "dirty" partition mov f11=f0 // clear f11
(pKStk) br.cond.dpnt.many skip_rbs_switch (pKStk) br.cond.dpnt.many skip_rbs_switch
mov.m ar.ccv=r0 // clear ar.ccv mov.m ar.ccv=r0 // clear ar.ccv
(pNonSys) br.cond.dpnt.many dont_preserve_current_frame (pNonSys) br.cond.dpnt.many dont_preserve_current_frame
br.cond.sptk.many rbs_switch br.cond.sptk.many rbs_switch
......
Markdown is supported
0% .
You are about to add 0 people to the discussion. Proceed with caution.
先完成此消息的编辑!
想要评论请 注册