aboutsummaryrefslogtreecommitdiffstats
path: root/arch/ia64/kernel
diff options
context:
space:
mode:
authorTony Luck <tony.luck@intel.com>2005-06-28 11:24:49 -0400
committerTony Luck <tony.luck@intel.com>2005-06-28 11:24:49 -0400
commit54522b6613a03807f057fd567794a31267ef85cb (patch)
treeab0d9d47779769dc5d75810aaff21972958a681f /arch/ia64/kernel
parent99f95e5286df2f69edab8a04c7080d986ee4233b (diff)
parentf2cbb4f01936a3e4225692e03b084b78c56d386d (diff)
Auto merge with /home/aegl/GIT/ia64-test
Diffstat (limited to 'arch/ia64/kernel')
-rw-r--r--arch/ia64/kernel/entry.S110
-rw-r--r--arch/ia64/kernel/fsys.S147
-rw-r--r--arch/ia64/kernel/gate.S62
-rw-r--r--arch/ia64/kernel/ia64_ksyms.c3
-rw-r--r--arch/ia64/kernel/ivt.S179
-rw-r--r--arch/ia64/kernel/ptrace.c22
6 files changed, 299 insertions, 224 deletions
diff --git a/arch/ia64/kernel/entry.S b/arch/ia64/kernel/entry.S
index 785a51b0ad8e..69f88d561d62 100644
--- a/arch/ia64/kernel/entry.S
+++ b/arch/ia64/kernel/entry.S
@@ -470,18 +470,6 @@ ENTRY(load_switch_stack)
470 br.cond.sptk.many b7 470 br.cond.sptk.many b7
471END(load_switch_stack) 471END(load_switch_stack)
472 472
473GLOBAL_ENTRY(__ia64_syscall)
474 .regstk 6,0,0,0
475 mov r15=in5 // put syscall number in place
476 break __BREAK_SYSCALL
477 movl r2=errno
478 cmp.eq p6,p7=-1,r10
479 ;;
480(p6) st4 [r2]=r8
481(p6) mov r8=-1
482 br.ret.sptk.many rp
483END(__ia64_syscall)
484
485GLOBAL_ENTRY(execve) 473GLOBAL_ENTRY(execve)
486 mov r15=__NR_execve // put syscall number in place 474 mov r15=__NR_execve // put syscall number in place
487 break __BREAK_SYSCALL 475 break __BREAK_SYSCALL
@@ -637,7 +625,7 @@ END(ia64_ret_from_syscall)
637 * r8-r11: restored (syscall return value(s)) 625 * r8-r11: restored (syscall return value(s))
638 * r12: restored (user-level stack pointer) 626 * r12: restored (user-level stack pointer)
639 * r13: restored (user-level thread pointer) 627 * r13: restored (user-level thread pointer)
640 * r14: cleared 628 * r14: set to __kernel_syscall_via_epc
641 * r15: restored (syscall #) 629 * r15: restored (syscall #)
642 * r16-r17: cleared 630 * r16-r17: cleared
643 * r18: user-level b6 631 * r18: user-level b6
@@ -658,7 +646,7 @@ END(ia64_ret_from_syscall)
658 * pr: restored (user-level pr) 646 * pr: restored (user-level pr)
659 * b0: restored (user-level rp) 647 * b0: restored (user-level rp)
660 * b6: restored 648 * b6: restored
661 * b7: cleared 649 * b7: set to __kernel_syscall_via_epc
662 * ar.unat: restored (user-level ar.unat) 650 * ar.unat: restored (user-level ar.unat)
663 * ar.pfs: restored (user-level ar.pfs) 651 * ar.pfs: restored (user-level ar.pfs)
664 * ar.rsc: restored (user-level ar.rsc) 652 * ar.rsc: restored (user-level ar.rsc)
@@ -704,72 +692,79 @@ ENTRY(ia64_leave_syscall)
704 ;; 692 ;;
705(p6) ld4 r31=[r18] // load current_thread_info()->flags 693(p6) ld4 r31=[r18] // load current_thread_info()->flags
706 ld8 r19=[r2],PT(B6)-PT(LOADRS) // load ar.rsc value for "loadrs" 694 ld8 r19=[r2],PT(B6)-PT(LOADRS) // load ar.rsc value for "loadrs"
707 mov b7=r0 // clear b7 695 nop.i 0
708 ;; 696 ;;
709 ld8 r23=[r3],PT(R11)-PT(AR_BSPSTORE) // load ar.bspstore (may be garbage) 697 mov r16=ar.bsp // M2 get existing backing store pointer
710 ld8 r18=[r2],PT(R9)-PT(B6) // load b6 698 ld8 r18=[r2],PT(R9)-PT(B6) // load b6
711(p6) and r15=TIF_WORK_MASK,r31 // any work other than TIF_SYSCALL_TRACE? 699(p6) and r15=TIF_WORK_MASK,r31 // any work other than TIF_SYSCALL_TRACE?
712 ;; 700 ;;
713 mov r16=ar.bsp // M2 get existing backing store pointer 701 ld8 r23=[r3],PT(R11)-PT(AR_BSPSTORE) // load ar.bspstore (may be garbage)
714(p6) cmp4.ne.unc p6,p0=r15, r0 // any special work pending? 702(p6) cmp4.ne.unc p6,p0=r15, r0 // any special work pending?
715(p6) br.cond.spnt .work_pending_syscall 703(p6) br.cond.spnt .work_pending_syscall
716 ;; 704 ;;
717 // start restoring the state saved on the kernel stack (struct pt_regs): 705 // start restoring the state saved on the kernel stack (struct pt_regs):
718 ld8 r9=[r2],PT(CR_IPSR)-PT(R9) 706 ld8 r9=[r2],PT(CR_IPSR)-PT(R9)
719 ld8 r11=[r3],PT(CR_IIP)-PT(R11) 707 ld8 r11=[r3],PT(CR_IIP)-PT(R11)
720 mov f6=f0 // clear f6 708(pNonSys) break 0 // bug check: we shouldn't be here if pNonSys is TRUE!
721 ;; 709 ;;
722 invala // M0|1 invalidate ALAT 710 invala // M0|1 invalidate ALAT
723 rsm psr.i | psr.ic // M2 initiate turning off of interrupt and interruption collection 711 rsm psr.i | psr.ic // M2 turn off interrupts and interruption collection
724 mov f9=f0 // clear f9 712 cmp.eq p9,p0=r0,r0 // A set p9 to indicate that we should restore cr.ifs
725 713
726 ld8 r29=[r2],16 // load cr.ipsr 714 ld8 r29=[r2],16 // M0|1 load cr.ipsr
727 ld8 r28=[r3],16 // load cr.iip 715 ld8 r28=[r3],16 // M0|1 load cr.iip
728 mov f8=f0 // clear f8 716 mov r22=r0 // A clear r22
729 ;; 717 ;;
730 ld8 r30=[r2],16 // M0|1 load cr.ifs 718 ld8 r30=[r2],16 // M0|1 load cr.ifs
731 ld8 r25=[r3],16 // M0|1 load ar.unat 719 ld8 r25=[r3],16 // M0|1 load ar.unat
732 cmp.eq p9,p0=r0,r0 // set p9 to indicate that we should restore cr.ifs 720(pUStk) add r14=IA64_TASK_THREAD_ON_USTACK_OFFSET,r13
733 ;; 721 ;;
734 ld8 r26=[r2],PT(B0)-PT(AR_PFS) // M0|1 load ar.pfs 722 ld8 r26=[r2],PT(B0)-PT(AR_PFS) // M0|1 load ar.pfs
735(pKStk) mov r22=psr // M2 read PSR now that interrupts are disabled 723(pKStk) mov r22=psr // M2 read PSR now that interrupts are disabled
736 mov f10=f0 // clear f10 724 nop 0
737 ;; 725 ;;
738 ld8 r21=[r2],PT(AR_RNAT)-PT(B0) // load b0 726 ld8 r21=[r2],PT(AR_RNAT)-PT(B0) // M0|1 load b0
739 ld8 r27=[r3],PT(PR)-PT(AR_RSC) // load ar.rsc 727 ld8 r27=[r3],PT(PR)-PT(AR_RSC) // M0|1 load ar.rsc
740 mov f11=f0 // clear f11 728 mov f6=f0 // F clear f6
741 ;; 729 ;;
742 ld8 r24=[r2],PT(AR_FPSR)-PT(AR_RNAT) // load ar.rnat (may be garbage) 730 ld8 r24=[r2],PT(AR_FPSR)-PT(AR_RNAT) // M0|1 load ar.rnat (may be garbage)
743 ld8 r31=[r3],PT(R1)-PT(PR) // load predicates 731 ld8 r31=[r3],PT(R1)-PT(PR) // M0|1 load predicates
744(pUStk) add r14=IA64_TASK_THREAD_ON_USTACK_OFFSET,r13 732 mov f7=f0 // F clear f7
745 ;; 733 ;;
746 ld8 r20=[r2],PT(R12)-PT(AR_FPSR) // load ar.fpsr 734 ld8 r20=[r2],PT(R12)-PT(AR_FPSR) // M0|1 load ar.fpsr
747 ld8.fill r1=[r3],16 // load r1 735 ld8.fill r1=[r3],16 // M0|1 load r1
748(pUStk) mov r17=1 736(pUStk) mov r17=1 // A
749 ;; 737 ;;
750 srlz.d // M0 ensure interruption collection is off 738(pUStk) st1 [r14]=r17 // M2|3
751 ld8.fill r13=[r3],16 739 ld8.fill r13=[r3],16 // M0|1
752 mov f7=f0 // clear f7 740 mov f8=f0 // F clear f8
753 ;; 741 ;;
754 ld8.fill r12=[r2] // restore r12 (sp) 742 ld8.fill r12=[r2] // M0|1 restore r12 (sp)
755 mov.m ar.ssd=r0 // M2 clear ar.ssd 743 ld8.fill r15=[r3] // M0|1 restore r15
756 mov r22=r0 // clear r22 744 mov b6=r18 // I0 restore b6
757 745
758 ld8.fill r15=[r3] // restore r15 746 addl r17=THIS_CPU(ia64_phys_stacked_size_p8),r0 // A
759(pUStk) st1 [r14]=r17 747 mov f9=f0 // F clear f9
760 addl r3=THIS_CPU(ia64_phys_stacked_size_p8),r0 748(pKStk) br.cond.dpnt.many skip_rbs_switch // B
749
750 srlz.d // M0 ensure interruption collection is off (for cover)
751 shr.u r18=r19,16 // I0|1 get byte size of existing "dirty" partition
752 cover // B add current frame into dirty partition & set cr.ifs
761 ;; 753 ;;
762(pUStk) ld4 r17=[r3] // r17 = cpu_data->phys_stacked_size_p8 754(pUStk) ld4 r17=[r17] // M0|1 r17 = cpu_data->phys_stacked_size_p8
763 mov.m ar.csd=r0 // M2 clear ar.csd 755 mov r19=ar.bsp // M2 get new backing store pointer
764 mov b6=r18 // I0 restore b6 756 mov f10=f0 // F clear f10
757
758 nop.m 0
759 movl r14=__kernel_syscall_via_epc // X
765 ;; 760 ;;
766 mov r14=r0 // clear r14 761 mov.m ar.csd=r0 // M2 clear ar.csd
767 shr.u r18=r19,16 // I0|1 get byte size of existing "dirty" partition 762 mov.m ar.ccv=r0 // M2 clear ar.ccv
768(pKStk) br.cond.dpnt.many skip_rbs_switch 763 mov b7=r14 // I0 clear b7 (hint with __kernel_syscall_via_epc)
769 764
770 mov.m ar.ccv=r0 // clear ar.ccv 765 mov.m ar.ssd=r0 // M2 clear ar.ssd
771(pNonSys) br.cond.dpnt.many dont_preserve_current_frame 766 mov f11=f0 // F clear f11
772 br.cond.sptk.many rbs_switch 767 br.cond.sptk.many rbs_switch // B
773END(ia64_leave_syscall) 768END(ia64_leave_syscall)
774 769
775#ifdef CONFIG_IA32_SUPPORT 770#ifdef CONFIG_IA32_SUPPORT
@@ -885,7 +880,7 @@ GLOBAL_ENTRY(ia64_leave_kernel)
885 ldf.fill f7=[r2],PT(F11)-PT(F7) 880 ldf.fill f7=[r2],PT(F11)-PT(F7)
886 ldf.fill f8=[r3],32 881 ldf.fill f8=[r3],32
887 ;; 882 ;;
888 srlz.i // ensure interruption collection is off 883 srlz.d // ensure that inter. collection is off (VHPT is don't care, since text is pinned)
889 mov ar.ccv=r15 884 mov ar.ccv=r15
890 ;; 885 ;;
891 ldf.fill f11=[r2] 886 ldf.fill f11=[r2]
@@ -945,11 +940,10 @@ GLOBAL_ENTRY(ia64_leave_kernel)
945 * NOTE: alloc, loadrs, and cover can't be predicated. 940 * NOTE: alloc, loadrs, and cover can't be predicated.
946 */ 941 */
947(pNonSys) br.cond.dpnt dont_preserve_current_frame 942(pNonSys) br.cond.dpnt dont_preserve_current_frame
948
949rbs_switch:
950 cover // add current frame into dirty partition and set cr.ifs 943 cover // add current frame into dirty partition and set cr.ifs
951 ;; 944 ;;
952 mov r19=ar.bsp // get new backing store pointer 945 mov r19=ar.bsp // get new backing store pointer
946rbs_switch:
953 sub r16=r16,r18 // krbs = old bsp - size of dirty partition 947 sub r16=r16,r18 // krbs = old bsp - size of dirty partition
954 cmp.ne p9,p0=r0,r0 // clear p9 to skip restore of cr.ifs 948 cmp.ne p9,p0=r0,r0 // clear p9 to skip restore of cr.ifs
955 ;; 949 ;;
@@ -1024,14 +1018,14 @@ rse_clear_invalid:
1024 mov loc5=0 1018 mov loc5=0
1025 mov loc6=0 1019 mov loc6=0
1026 mov loc7=0 1020 mov loc7=0
1027(pRecurse) br.call.sptk.few b0=rse_clear_invalid 1021(pRecurse) br.call.dptk.few b0=rse_clear_invalid
1028 ;; 1022 ;;
1029 mov loc8=0 1023 mov loc8=0
1030 mov loc9=0 1024 mov loc9=0
1031 cmp.ne pReturn,p0=r0,in1 // if recursion count != 0, we need to do a br.ret 1025 cmp.ne pReturn,p0=r0,in1 // if recursion count != 0, we need to do a br.ret
1032 mov loc10=0 1026 mov loc10=0
1033 mov loc11=0 1027 mov loc11=0
1034(pReturn) br.ret.sptk.many b0 1028(pReturn) br.ret.dptk.many b0
1035#endif /* !CONFIG_ITANIUM */ 1029#endif /* !CONFIG_ITANIUM */
1036# undef pRecurse 1030# undef pRecurse
1037# undef pReturn 1031# undef pReturn
diff --git a/arch/ia64/kernel/fsys.S b/arch/ia64/kernel/fsys.S
index 962b6c4e32b5..7d7684a369d3 100644
--- a/arch/ia64/kernel/fsys.S
+++ b/arch/ia64/kernel/fsys.S
@@ -531,93 +531,114 @@ GLOBAL_ENTRY(fsys_bubble_down)
531 .altrp b6 531 .altrp b6
532 .body 532 .body
533 /* 533 /*
534 * We get here for syscalls that don't have a lightweight handler. For those, we 534 * We get here for syscalls that don't have a lightweight
535 * need to bubble down into the kernel and that requires setting up a minimal 535 * handler. For those, we need to bubble down into the kernel
536 * pt_regs structure, and initializing the CPU state more or less as if an 536 * and that requires setting up a minimal pt_regs structure,
537 * interruption had occurred. To make syscall-restarts work, we setup pt_regs 537 * and initializing the CPU state more or less as if an
538 * such that cr_iip points to the second instruction in syscall_via_break. 538 * interruption had occurred. To make syscall-restarts work,
539 * Decrementing the IP hence will restart the syscall via break and not 539 * we setup pt_regs such that cr_iip points to the second
540 * decrementing IP will return us to the caller, as usual. Note that we preserve 540 * instruction in syscall_via_break. Decrementing the IP
541 * the value of psr.pp rather than initializing it from dcr.pp. This makes it 541 * hence will restart the syscall via break and not
542 * possible to distinguish fsyscall execution from other privileged execution. 542 * decrementing IP will return us to the caller, as usual.
543 * Note that we preserve the value of psr.pp rather than
544 * initializing it from dcr.pp. This makes it possible to
545 * distinguish fsyscall execution from other privileged
546 * execution.
543 * 547 *
544 * On entry: 548 * On entry:
545 * - normal fsyscall handler register usage, except that we also have: 549 * - normal fsyscall handler register usage, except
550 * that we also have:
546 * - r18: address of syscall entry point 551 * - r18: address of syscall entry point
547 * - r21: ar.fpsr 552 * - r21: ar.fpsr
548 * - r26: ar.pfs 553 * - r26: ar.pfs
549 * - r27: ar.rsc 554 * - r27: ar.rsc
550 * - r29: psr 555 * - r29: psr
556 *
557 * We used to clear some PSR bits here but that requires slow
558 * serialization. Fortuntely, that isn't really necessary.
559 * The rationale is as follows: we used to clear bits
560 * ~PSR_PRESERVED_BITS in PSR.L. Since
561 * PSR_PRESERVED_BITS==PSR.{UP,MFL,MFH,PK,DT,PP,SP,RT,IC}, we
562 * ended up clearing PSR.{BE,AC,I,DFL,DFH,DI,DB,SI,TB}.
563 * However,
564 *
565 * PSR.BE : already is turned off in __kernel_syscall_via_epc()
566 * PSR.AC : don't care (kernel normally turns PSR.AC on)
567 * PSR.I : already turned off by the time fsys_bubble_down gets
568 * invoked
569 * PSR.DFL: always 0 (kernel never turns it on)
570 * PSR.DFH: don't care --- kernel never touches f32-f127 on its own
571 * initiative
572 * PSR.DI : always 0 (kernel never turns it on)
573 * PSR.SI : always 0 (kernel never turns it on)
574 * PSR.DB : don't care --- kernel never enables kernel-level
575 * breakpoints
576 * PSR.TB : must be 0 already; if it wasn't zero on entry to
577 * __kernel_syscall_via_epc, the branch to fsys_bubble_down
578 * will trigger a taken branch; the taken-trap-handler then
579 * converts the syscall into a break-based system-call.
551 */ 580 */
552# define PSR_PRESERVED_BITS (IA64_PSR_UP | IA64_PSR_MFL | IA64_PSR_MFH | IA64_PSR_PK \
553 | IA64_PSR_DT | IA64_PSR_PP | IA64_PSR_SP | IA64_PSR_RT \
554 | IA64_PSR_IC)
555 /* 581 /*
556 * Reading psr.l gives us only bits 0-31, psr.it, and psr.mc. The rest we have 582 * Reading psr.l gives us only bits 0-31, psr.it, and psr.mc.
557 * to synthesize. 583 * The rest we have to synthesize.
558 */ 584 */
559# define PSR_ONE_BITS ((3 << IA64_PSR_CPL0_BIT) | (0x1 << IA64_PSR_RI_BIT) \ 585# define PSR_ONE_BITS ((3 << IA64_PSR_CPL0_BIT) \
586 | (0x1 << IA64_PSR_RI_BIT) \
560 | IA64_PSR_BN | IA64_PSR_I) 587 | IA64_PSR_BN | IA64_PSR_I)
561 588
562 invala 589 invala // M0|1
563 movl r8=PSR_ONE_BITS 590 movl r14=ia64_ret_from_syscall // X
564 591
565 mov r25=ar.unat // save ar.unat (5 cyc) 592 nop.m 0
566 movl r9=PSR_PRESERVED_BITS 593 movl r28=__kernel_syscall_via_break // X create cr.iip
594 ;;
567 595
568 mov ar.rsc=0 // set enforced lazy mode, pl 0, little-endian, loadrs=0 596 mov r2=r16 // A get task addr to addl-addressable register
569 movl r28=__kernel_syscall_via_break 597 adds r16=IA64_TASK_THREAD_ON_USTACK_OFFSET,r16 // A
598 mov r31=pr // I0 save pr (2 cyc)
570 ;; 599 ;;
571 mov r23=ar.bspstore // save ar.bspstore (12 cyc) 600 st1 [r16]=r0 // M2|3 clear current->thread.on_ustack flag
572 mov r31=pr // save pr (2 cyc) 601 addl r22=IA64_RBS_OFFSET,r2 // A compute base of RBS
573 mov r20=r1 // save caller's gp in r20 602 add r3=TI_FLAGS+IA64_TASK_SIZE,r2 // A
574 ;; 603 ;;
575 mov r2=r16 // copy current task addr to addl-addressable register 604 ld4 r3=[r3] // M0|1 r3 = current_thread_info()->flags
576 and r9=r9,r29 605 lfetch.fault.excl.nt1 [r22] // M0|1 prefetch register backing-store
577 mov r19=b6 // save b6 (2 cyc) 606 nop.i 0
578 ;; 607 ;;
579 mov psr.l=r9 // slam the door (17 cyc to srlz.i) 608 mov ar.rsc=0 // M2 set enforced lazy mode, pl 0, LE, loadrs=0
580 or r29=r8,r29 // construct cr.ipsr value to save 609 nop.m 0
581 addl r22=IA64_RBS_OFFSET,r2 // compute base of RBS 610 nop.i 0
582 ;; 611 ;;
583 // GAS reports a spurious RAW hazard on the read of ar.rnat because it thinks 612 mov r23=ar.bspstore // M2 (12 cyc) save ar.bspstore
584 // we may be reading ar.itc after writing to psr.l. Avoid that message with 613 mov.m r24=ar.rnat // M2 (5 cyc) read ar.rnat (dual-issues!)
585 // this directive: 614 nop.i 0
586 dv_serialize_data
587 mov.m r24=ar.rnat // read ar.rnat (5 cyc lat)
588 lfetch.fault.excl.nt1 [r22]
589 adds r16=IA64_TASK_THREAD_ON_USTACK_OFFSET,r2
590
591 // ensure previous insn group is issued before we stall for srlz.i:
592 ;; 615 ;;
593 srlz.i // ensure new psr.l has been established 616 mov ar.bspstore=r22 // M2 (6 cyc) switch to kernel RBS
594 ///////////////////////////////////////////////////////////////////////////// 617 movl r8=PSR_ONE_BITS // X
595 ////////// from this point on, execution is not interruptible anymore
596 /////////////////////////////////////////////////////////////////////////////
597 addl r1=IA64_STK_OFFSET-IA64_PT_REGS_SIZE,r2 // compute base of memory stack
598 cmp.ne pKStk,pUStk=r0,r0 // set pKStk <- 0, pUStk <- 1
599 ;; 618 ;;
600 st1 [r16]=r0 // clear current->thread.on_ustack flag 619 mov r25=ar.unat // M2 (5 cyc) save ar.unat
601 mov ar.bspstore=r22 // switch to kernel RBS 620 mov r19=b6 // I0 save b6 (2 cyc)
602 mov b6=r18 // copy syscall entry-point to b6 (7 cyc) 621 mov r20=r1 // A save caller's gp in r20
603 add r3=TI_FLAGS+IA64_TASK_SIZE,r2
604 ;; 622 ;;
605 ld4 r3=[r3] // r2 = current_thread_info()->flags 623 or r29=r8,r29 // A construct cr.ipsr value to save
606 mov r18=ar.bsp // save (kernel) ar.bsp (12 cyc) 624 mov b6=r18 // I0 copy syscall entry-point to b6 (7 cyc)
607 mov ar.rsc=0x3 // set eager mode, pl 0, little-endian, loadrs=0 625 addl r1=IA64_STK_OFFSET-IA64_PT_REGS_SIZE,r2 // A compute base of memory stack
608 br.call.sptk.many b7=ia64_syscall_setup 626
609 ;; 627 mov r18=ar.bsp // M2 save (kernel) ar.bsp (12 cyc)
610 ssm psr.i 628 cmp.ne pKStk,pUStk=r0,r0 // A set pKStk <- 0, pUStk <- 1
611 movl r2=ia64_ret_from_syscall 629 br.call.sptk.many b7=ia64_syscall_setup // B
612 ;; 630 ;;
613 mov rp=r2 // set the real return addr 631 mov ar.rsc=0x3 // M2 set eager mode, pl 0, LE, loadrs=0
614 and r3=_TIF_SYSCALL_TRACEAUDIT,r3 632 mov rp=r14 // I0 set the real return addr
633 and r3=_TIF_SYSCALL_TRACEAUDIT,r3 // A
615 ;; 634 ;;
616 cmp.eq p8,p0=r3,r0 635 ssm psr.i // M2 we're on kernel stacks now, reenable irqs
636 cmp.eq p8,p0=r3,r0 // A
637(p10) br.cond.spnt.many ia64_ret_from_syscall // B return if bad call-frame or r15 is a NaT
617 638
618(p10) br.cond.spnt.many ia64_ret_from_syscall // p10==true means out registers are more than 8 639 nop.m 0
619(p8) br.call.sptk.many b6=b6 // ignore this return addr 640(p8) br.call.sptk.many b6=b6 // B (ignore return address)
620 br.cond.sptk ia64_trace_syscall 641 br.cond.spnt ia64_trace_syscall // B
621END(fsys_bubble_down) 642END(fsys_bubble_down)
622 643
623 .rodata 644 .rodata
diff --git a/arch/ia64/kernel/gate.S b/arch/ia64/kernel/gate.S
index facf75acdc85..86948ce63e43 100644
--- a/arch/ia64/kernel/gate.S
+++ b/arch/ia64/kernel/gate.S
@@ -72,38 +72,40 @@ GLOBAL_ENTRY(__kernel_syscall_via_epc)
72 * bundle get executed. The remaining code must be safe even if 72 * bundle get executed. The remaining code must be safe even if
73 * they do not get executed. 73 * they do not get executed.
74 */ 74 */
75 adds r17=-1024,r15 75 adds r17=-1024,r15 // A
76 mov r10=0 // default to successful syscall execution 76 mov r10=0 // A default to successful syscall execution
77 epc 77 epc // B causes split-issue
78} 78}
79 ;; 79 ;;
80 rsm psr.be // note: on McKinley "rsm psr.be/srlz.d" is slightly faster than "rum psr.be" 80 rsm psr.be | psr.i // M2 (5 cyc to srlz.d)
81 LOAD_FSYSCALL_TABLE(r14) 81 LOAD_FSYSCALL_TABLE(r14) // X
82
83 mov r16=IA64_KR(CURRENT) // 12 cycle read latency
84 tnat.nz p10,p9=r15
85 mov r19=NR_syscalls-1
86 ;; 82 ;;
87 shladd r18=r17,3,r14 83 mov r16=IA64_KR(CURRENT) // M2 (12 cyc)
88 84 shladd r18=r17,3,r14 // A
89 srlz.d 85 mov r19=NR_syscalls-1 // A
90 cmp.ne p8,p0=r0,r0 // p8 <- FALSE 86 ;;
91 /* Note: if r17 is a NaT, p6 will be set to zero. */ 87 lfetch [r18] // M0|1
92 cmp.geu p6,p7=r19,r17 // (syscall > 0 && syscall < 1024+NR_syscalls)? 88 mov r29=psr // M2 (12 cyc)
93 ;; 89 // If r17 is a NaT, p6 will be zero
94(p6) ld8 r18=[r18] 90 cmp.geu p6,p7=r19,r17 // A (sysnr > 0 && sysnr < 1024+NR_syscalls)?
95 mov r21=ar.fpsr 91 ;;
96 add r14=-8,r14 // r14 <- addr of fsys_bubble_down entry 92 mov r21=ar.fpsr // M2 (12 cyc)
97 ;; 93 tnat.nz p10,p9=r15 // I0
98(p6) mov b7=r18 94 mov.i r26=ar.pfs // I0 (would stall anyhow due to srlz.d...)
99(p6) tbit.z p8,p0=r18,0 95 ;;
100(p8) br.dptk.many b7 96 srlz.d // M0 (forces split-issue) ensure PSR.BE==0
101 97(p6) ld8 r18=[r18] // M0|1
102(p6) rsm psr.i 98 nop.i 0
103 mov r27=ar.rsc 99 ;;
104 mov r26=ar.pfs 100 nop.m 0
101(p6) tbit.z.unc p8,p0=r18,0 // I0 (dual-issues with "mov b7=r18"!)
102 nop.i 0
105 ;; 103 ;;
106 mov r29=psr // read psr (12 cyc load latency) 104(p8) ssm psr.i
105(p6) mov b7=r18 // I0
106(p8) br.dptk.many b7 // B
107
108 mov r27=ar.rsc // M2 (12 cyc)
107/* 109/*
108 * brl.cond doesn't work as intended because the linker would convert this branch 110 * brl.cond doesn't work as intended because the linker would convert this branch
109 * into a branch to a PLT. Perhaps there will be a way to avoid this with some 111 * into a branch to a PLT. Perhaps there will be a way to avoid this with some
@@ -111,6 +113,8 @@ GLOBAL_ENTRY(__kernel_syscall_via_epc)
111 * instead. 113 * instead.
112 */ 114 */
113#ifdef CONFIG_ITANIUM 115#ifdef CONFIG_ITANIUM
116(p6) add r14=-8,r14 // r14 <- addr of fsys_bubble_down entry
117 ;;
114(p6) ld8 r14=[r14] // r14 <- fsys_bubble_down 118(p6) ld8 r14=[r14] // r14 <- fsys_bubble_down
115 ;; 119 ;;
116(p6) mov b7=r14 120(p6) mov b7=r14
@@ -118,7 +122,7 @@ GLOBAL_ENTRY(__kernel_syscall_via_epc)
118#else 122#else
119 BRL_COND_FSYS_BUBBLE_DOWN(p6) 123 BRL_COND_FSYS_BUBBLE_DOWN(p6)
120#endif 124#endif
121 125 ssm psr.i
122 mov r10=-1 126 mov r10=-1
123(p10) mov r8=EINVAL 127(p10) mov r8=EINVAL
124(p9) mov r8=ENOSYS 128(p9) mov r8=ENOSYS
diff --git a/arch/ia64/kernel/ia64_ksyms.c b/arch/ia64/kernel/ia64_ksyms.c
index 7bbf019c9867..01572814abe4 100644
--- a/arch/ia64/kernel/ia64_ksyms.c
+++ b/arch/ia64/kernel/ia64_ksyms.c
@@ -58,9 +58,6 @@ EXPORT_SYMBOL(__strlen_user);
58EXPORT_SYMBOL(__strncpy_from_user); 58EXPORT_SYMBOL(__strncpy_from_user);
59EXPORT_SYMBOL(__strnlen_user); 59EXPORT_SYMBOL(__strnlen_user);
60 60
61#include <asm/unistd.h>
62EXPORT_SYMBOL(__ia64_syscall);
63
64/* from arch/ia64/lib */ 61/* from arch/ia64/lib */
65extern void __divsi3(void); 62extern void __divsi3(void);
66extern void __udivsi3(void); 63extern void __udivsi3(void);
diff --git a/arch/ia64/kernel/ivt.S b/arch/ia64/kernel/ivt.S
index 2bc085a73e30..b28d2212a779 100644
--- a/arch/ia64/kernel/ivt.S
+++ b/arch/ia64/kernel/ivt.S
@@ -1,7 +1,7 @@
1/* 1/*
2 * arch/ia64/kernel/ivt.S 2 * arch/ia64/kernel/ivt.S
3 * 3 *
4 * Copyright (C) 1998-2001, 2003 Hewlett-Packard Co 4 * Copyright (C) 1998-2001, 2003, 2005 Hewlett-Packard Co
5 * Stephane Eranian <eranian@hpl.hp.com> 5 * Stephane Eranian <eranian@hpl.hp.com>
6 * David Mosberger <davidm@hpl.hp.com> 6 * David Mosberger <davidm@hpl.hp.com>
7 * Copyright (C) 2000, 2002-2003 Intel Co 7 * Copyright (C) 2000, 2002-2003 Intel Co
@@ -692,82 +692,118 @@ ENTRY(break_fault)
692 * to prevent leaking bits from kernel to user level. 692 * to prevent leaking bits from kernel to user level.
693 */ 693 */
694 DBG_FAULT(11) 694 DBG_FAULT(11)
695 mov r16=IA64_KR(CURRENT) // r16 = current task; 12 cycle read lat. 695 mov.m r16=IA64_KR(CURRENT) // M2 r16 <- current task (12 cyc)
696 mov r17=cr.iim 696 mov r29=cr.ipsr // M2 (12 cyc)
697 mov r18=__IA64_BREAK_SYSCALL 697 mov r31=pr // I0 (2 cyc)
698 mov r21=ar.fpsr 698
699 mov r29=cr.ipsr 699 mov r17=cr.iim // M2 (2 cyc)
700 mov r19=b6 700 mov.m r27=ar.rsc // M2 (12 cyc)
701 mov r25=ar.unat 701 mov r18=__IA64_BREAK_SYSCALL // A
702 mov r27=ar.rsc 702
703 mov r26=ar.pfs 703 mov.m ar.rsc=0 // M2
704 mov r28=cr.iip 704 mov.m r21=ar.fpsr // M2 (12 cyc)
705 mov r31=pr // prepare to save predicates 705 mov r19=b6 // I0 (2 cyc)
706 mov r20=r1 706 ;;
707 ;; 707 mov.m r23=ar.bspstore // M2 (12 cyc)
708 mov.m r24=ar.rnat // M2 (5 cyc)
709 mov.i r26=ar.pfs // I0 (2 cyc)
710
711 invala // M0|1
712 nop.m 0 // M
713 mov r20=r1 // A save r1
714
715 nop.m 0
716 movl r30=sys_call_table // X
717
718 mov r28=cr.iip // M2 (2 cyc)
719 cmp.eq p0,p7=r18,r17 // I0 is this a system call?
720(p7) br.cond.spnt non_syscall // B no ->
721 //
722 // From this point on, we are definitely on the syscall-path
723 // and we can use (non-banked) scratch registers.
724 //
725///////////////////////////////////////////////////////////////////////
726 mov r1=r16 // A move task-pointer to "addl"-addressable reg
727 mov r2=r16 // A setup r2 for ia64_syscall_setup
728 add r9=TI_FLAGS+IA64_TASK_SIZE,r16 // A r9 = &current_thread_info()->flags
729
708 adds r16=IA64_TASK_THREAD_ON_USTACK_OFFSET,r16 730 adds r16=IA64_TASK_THREAD_ON_USTACK_OFFSET,r16
709 cmp.eq p0,p7=r18,r17 // is this a system call? (p7 <- false, if so) 731 adds r15=-1024,r15 // A subtract 1024 from syscall number
710(p7) br.cond.spnt non_syscall 732 mov r3=NR_syscalls - 1
711 ;; 733 ;;
712 ld1 r17=[r16] // load current->thread.on_ustack flag 734 ld1.bias r17=[r16] // M0|1 r17 = current->thread.on_ustack flag
713 st1 [r16]=r0 // clear current->thread.on_ustack flag 735 ld4 r9=[r9] // M0|1 r9 = current_thread_info()->flags
714 add r1=-IA64_TASK_THREAD_ON_USTACK_OFFSET,r16 // set r1 for MINSTATE_START_SAVE_MIN_VIRT 736 extr.u r8=r29,41,2 // I0 extract ei field from cr.ipsr
737
738 shladd r30=r15,3,r30 // A r30 = sys_call_table + 8*(syscall-1024)
739 addl r22=IA64_RBS_OFFSET,r1 // A compute base of RBS
740 cmp.leu p6,p7=r15,r3 // A syscall number in range?
715 ;; 741 ;;
716 invala
717 742
718 /* adjust return address so we skip over the break instruction: */ 743 lfetch.fault.excl.nt1 [r22] // M0|1 prefetch RBS
744(p6) ld8 r30=[r30] // M0|1 load address of syscall entry point
745 tnat.nz.or p7,p0=r15 // I0 is syscall nr a NaT?
719 746
720 extr.u r8=r29,41,2 // extract ei field from cr.ipsr 747 mov.m ar.bspstore=r22 // M2 switch to kernel RBS
721 ;; 748 cmp.eq p8,p9=2,r8 // A isr.ei==2?
722 cmp.eq p6,p7=2,r8 // isr.ei==2?
723 mov r2=r1 // setup r2 for ia64_syscall_setup
724 ;;
725(p6) mov r8=0 // clear ei to 0
726(p6) adds r28=16,r28 // switch cr.iip to next bundle cr.ipsr.ei wrapped
727(p7) adds r8=1,r8 // increment ei to next slot
728 ;;
729 cmp.eq pKStk,pUStk=r0,r17 // are we in kernel mode already?
730 dep r29=r8,r29,41,2 // insert new ei into cr.ipsr
731 ;; 749 ;;
732 750
733 // switch from user to kernel RBS: 751(p8) mov r8=0 // A clear ei to 0
734 MINSTATE_START_SAVE_MIN_VIRT 752(p7) movl r30=sys_ni_syscall // X
735 br.call.sptk.many b7=ia64_syscall_setup
736 ;;
737 MINSTATE_END_SAVE_MIN_VIRT // switch to bank 1
738 ssm psr.ic | PSR_DEFAULT_BITS
739 ;;
740 srlz.i // guarantee that interruption collection is on
741 mov r3=NR_syscalls - 1
742 ;;
743(p15) ssm psr.i // restore psr.i
744 // p10==true means out registers are more than 8 or r15's Nat is true
745(p10) br.cond.spnt.many ia64_ret_from_syscall
746 ;;
747 movl r16=sys_call_table
748 753
749 adds r15=-1024,r15 // r15 contains the syscall number---subtract 1024 754(p8) adds r28=16,r28 // A switch cr.iip to next bundle
750 movl r2=ia64_ret_from_syscall 755(p9) adds r8=1,r8 // A increment ei to next slot
751 ;; 756 nop.i 0
752 shladd r20=r15,3,r16 // r20 = sys_call_table + 8*(syscall-1024)
753 cmp.leu p6,p7=r15,r3 // (syscall > 0 && syscall < 1024 + NR_syscalls) ?
754 mov rp=r2 // set the real return addr
755 ;; 757 ;;
756(p6) ld8 r20=[r20] // load address of syscall entry point
757(p7) movl r20=sys_ni_syscall
758 758
759 add r2=TI_FLAGS+IA64_TASK_SIZE,r13 759 mov.m r25=ar.unat // M2 (5 cyc)
760 ;; 760 dep r29=r8,r29,41,2 // I0 insert new ei into cr.ipsr
761 ld4 r2=[r2] // r2 = current_thread_info()->flags 761 adds r15=1024,r15 // A restore original syscall number
762 ;; 762 //
763 and r2=_TIF_SYSCALL_TRACEAUDIT,r2 // mask trace or audit 763 // If any of the above loads miss in L1D, we'll stall here until
764 // the data arrives.
765 //
766///////////////////////////////////////////////////////////////////////
767 st1 [r16]=r0 // M2|3 clear current->thread.on_ustack flag
768 mov b6=r30 // I0 setup syscall handler branch reg early
769 cmp.eq pKStk,pUStk=r0,r17 // A were we on kernel stacks already?
770
771 and r9=_TIF_SYSCALL_TRACEAUDIT,r9 // A mask trace or audit
772 mov r18=ar.bsp // M2 (12 cyc)
773(pKStk) br.cond.spnt .break_fixup // B we're already in kernel-mode -- fix up RBS
774 ;;
775.back_from_break_fixup:
776(pUStk) addl r1=IA64_STK_OFFSET-IA64_PT_REGS_SIZE,r1 // A compute base of memory stack
777 cmp.eq p14,p0=r9,r0 // A are syscalls being traced/audited?
778 br.call.sptk.many b7=ia64_syscall_setup // B
7791:
780 mov ar.rsc=0x3 // M2 set eager mode, pl 0, LE, loadrs=0
781 nop 0
782 bsw.1 // B (6 cyc) regs are saved, switch to bank 1
764 ;; 783 ;;
765 cmp.eq p8,p0=r2,r0 784
766 mov b6=r20 785 ssm psr.ic | PSR_DEFAULT_BITS // M2 now it's safe to re-enable intr.-collection
786 movl r3=ia64_ret_from_syscall // X
767 ;; 787 ;;
768(p8) br.call.sptk.many b6=b6 // ignore this return addr 788
769 br.cond.sptk ia64_trace_syscall 789 srlz.i // M0 ensure interruption collection is on
790 mov rp=r3 // I0 set the real return addr
791(p10) br.cond.spnt.many ia64_ret_from_syscall // B return if bad call-frame or r15 is a NaT
792
793(p15) ssm psr.i // M2 restore psr.i
794(p14) br.call.sptk.many b6=b6 // B invoke syscall-handker (ignore return addr)
795 br.cond.spnt.many ia64_trace_syscall // B do syscall-tracing thingamagic
770 // NOT REACHED 796 // NOT REACHED
797///////////////////////////////////////////////////////////////////////
798 // On entry, we optimistically assumed that we're coming from user-space.
799 // For the rare cases where a system-call is done from within the kernel,
800 // we fix things up at this point:
801.break_fixup:
802 add r1=-IA64_PT_REGS_SIZE,sp // A allocate space for pt_regs structure
803 mov ar.rnat=r24 // M2 restore kernel's AR.RNAT
804 ;;
805 mov ar.bspstore=r23 // M2 restore kernel's AR.BSPSTORE
806 br.cond.sptk .back_from_break_fixup
771END(break_fault) 807END(break_fault)
772 808
773 .org ia64_ivt+0x3000 809 .org ia64_ivt+0x3000
@@ -842,8 +878,6 @@ END(interrupt)
842 * - r31: saved pr 878 * - r31: saved pr
843 * - b0: original contents (to be saved) 879 * - b0: original contents (to be saved)
844 * On exit: 880 * On exit:
845 * - executing on bank 1 registers
846 * - psr.ic enabled, interrupts restored
847 * - p10: TRUE if syscall is invoked with more than 8 out 881 * - p10: TRUE if syscall is invoked with more than 8 out
848 * registers or r15's Nat is true 882 * registers or r15's Nat is true
849 * - r1: kernel's gp 883 * - r1: kernel's gp
@@ -851,8 +885,11 @@ END(interrupt)
851 * - r8: -EINVAL if p10 is true 885 * - r8: -EINVAL if p10 is true
852 * - r12: points to kernel stack 886 * - r12: points to kernel stack
853 * - r13: points to current task 887 * - r13: points to current task
888 * - r14: preserved (same as on entry)
889 * - p13: preserved
854 * - p15: TRUE if interrupts need to be re-enabled 890 * - p15: TRUE if interrupts need to be re-enabled
855 * - ar.fpsr: set to kernel settings 891 * - ar.fpsr: set to kernel settings
892 * - b6: preserved (same as on entry)
856 */ 893 */
857GLOBAL_ENTRY(ia64_syscall_setup) 894GLOBAL_ENTRY(ia64_syscall_setup)
858#if PT(B6) != 0 895#if PT(B6) != 0
@@ -920,10 +957,10 @@ GLOBAL_ENTRY(ia64_syscall_setup)
920(p13) mov in5=-1 957(p13) mov in5=-1
921 ;; 958 ;;
922 st8 [r16]=r21,PT(R8)-PT(AR_FPSR) // save ar.fpsr 959 st8 [r16]=r21,PT(R8)-PT(AR_FPSR) // save ar.fpsr
923 tnat.nz p14,p0=in6 960 tnat.nz p13,p0=in6
924 cmp.lt p10,p9=r11,r8 // frame size can't be more than local+8 961 cmp.lt p10,p9=r11,r8 // frame size can't be more than local+8
925 ;; 962 ;;
926 stf8 [r16]=f1 // ensure pt_regs.r8 != 0 (see handle_syscall_error) 963 mov r8=1
927(p9) tnat.nz p10,p0=r15 964(p9) tnat.nz p10,p0=r15
928 adds r12=-16,r1 // switch to kernel memory stack (with 16 bytes of scratch) 965 adds r12=-16,r1 // switch to kernel memory stack (with 16 bytes of scratch)
929 966
@@ -934,9 +971,9 @@ GLOBAL_ENTRY(ia64_syscall_setup)
934 mov r13=r2 // establish `current' 971 mov r13=r2 // establish `current'
935 movl r1=__gp // establish kernel global pointer 972 movl r1=__gp // establish kernel global pointer
936 ;; 973 ;;
937(p14) mov in6=-1 974 st8 [r16]=r8 // ensure pt_regs.r8 != 0 (see handle_syscall_error)
975(p13) mov in6=-1
938(p8) mov in7=-1 976(p8) mov in7=-1
939 nop.i 0
940 977
941 cmp.eq pSys,pNonSys=r0,r0 // set pSys=1, pNonSys=0 978 cmp.eq pSys,pNonSys=r0,r0 // set pSys=1, pNonSys=0
942 movl r17=FPSR_DEFAULT 979 movl r17=FPSR_DEFAULT
@@ -1007,6 +1044,8 @@ END(dispatch_illegal_op_fault)
1007 FAULT(17) 1044 FAULT(17)
1008 1045
1009ENTRY(non_syscall) 1046ENTRY(non_syscall)
1047 mov ar.rsc=r27 // restore ar.rsc before SAVE_MIN_WITH_COVER
1048 ;;
1010 SAVE_MIN_WITH_COVER 1049 SAVE_MIN_WITH_COVER
1011 1050
1012 // There is no particular reason for this code to be here, other than that 1051 // There is no particular reason for this code to be here, other than that
diff --git a/arch/ia64/kernel/ptrace.c b/arch/ia64/kernel/ptrace.c
index 6d57aebad485..bbb8bc7c0552 100644
--- a/arch/ia64/kernel/ptrace.c
+++ b/arch/ia64/kernel/ptrace.c
@@ -725,12 +725,32 @@ convert_to_non_syscall (struct task_struct *child, struct pt_regs *pt,
725 break; 725 break;
726 } 726 }
727 727
728 /*
729 * Note: at the time of this call, the target task is blocked
730 * in notify_resume_user() and by clearling PRED_LEAVE_SYSCALL
731 * (aka, "pLvSys") we redirect execution from
732 * .work_pending_syscall_end to .work_processed_kernel.
733 */
728 unw_get_pr(&prev_info, &pr); 734 unw_get_pr(&prev_info, &pr);
729 pr &= ~(1UL << PRED_SYSCALL); 735 pr &= ~((1UL << PRED_SYSCALL) | (1UL << PRED_LEAVE_SYSCALL));
730 pr |= (1UL << PRED_NON_SYSCALL); 736 pr |= (1UL << PRED_NON_SYSCALL);
731 unw_set_pr(&prev_info, pr); 737 unw_set_pr(&prev_info, pr);
732 738
733 pt->cr_ifs = (1UL << 63) | cfm; 739 pt->cr_ifs = (1UL << 63) | cfm;
740 /*
741 * Clear the memory that is NOT written on syscall-entry to
742 * ensure we do not leak kernel-state to user when execution
743 * resumes.
744 */
745 pt->r2 = 0;
746 pt->r3 = 0;
747 pt->r14 = 0;
748 memset(&pt->r16, 0, 16*8); /* clear r16-r31 */
749 memset(&pt->f6, 0, 6*16); /* clear f6-f11 */
750 pt->b7 = 0;
751 pt->ar_ccv = 0;
752 pt->ar_csd = 0;
753 pt->ar_ssd = 0;
734} 754}
735 755
736static int 756static int