/*--------------------------------------------------------------------*/ /*--- The core dispatch loop, for jumping to a code address. ---*/ /*--- dispatch-ppc64-linux.S ---*/ /*--------------------------------------------------------------------*/ /* This file is part of Valgrind, a dynamic binary instrumentation framework. Copyright (C) 2005-2009 Cerion Armour-Brown This program is free software; you can redistribute it and/or modify it under the terms of the GNU General Public License as published by the Free Software Foundation; either version 2 of the License, or (at your option) any later version. This program is distributed in the hope that it will be useful, but WITHOUT ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License for more details. You should have received a copy of the GNU General Public License along with this program; if not, write to the Free Software Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA 02111-1307, USA. The GNU General Public License is contained in the file COPYING. */ #include "pub_core_basics_asm.h" #include "pub_core_dispatch_asm.h" #include "pub_core_transtab_asm.h" #include "libvex_guest_offsets.h" /* for OFFSET_ppc64_CIA */ /* References to globals via the TOC */ /* .globl vgPlain_tt_fast .lcomm vgPlain_tt_fast,4,4 .type vgPlain_tt_fast, @object */ .section ".toc","aw" .tocent__vgPlain_tt_fast: .tc vgPlain_tt_fast[TC],vgPlain_tt_fast .tocent__vgPlain_tt_fastN: .tc vgPlain_tt_fastN[TC],vgPlain_tt_fastN .tocent__vgPlain_dispatch_ctr: .tc vgPlain_dispatch_ctr[TC],vgPlain_dispatch_ctr .tocent__vgPlain_machine_ppc64_has_VMX: .tc vgPlain_machine_ppc64_has_VMX[TC],vgPlain_machine_ppc64_has_VMX /*------------------------------------------------------------*/ /*--- ---*/ /*--- The dispatch loop. VG_(run_innerloop) is used to ---*/ /*--- run all translations except no-redir ones. ---*/ /*--- ---*/ /*------------------------------------------------------------*/ /*----------------------------------------------------*/ /*--- Preamble (set everything up) ---*/ /*----------------------------------------------------*/ /* signature: UWord VG_(run_innerloop) ( void* guest_state, UWord do_profiling ); */ .section ".text" .align 2 .globl VG_(run_innerloop) .section ".opd","aw" .align 3 VG_(run_innerloop): .quad .VG_(run_innerloop),.TOC.@tocbase,0 .previous .type .VG_(run_innerloop),@function .globl .VG_(run_innerloop) .VG_(run_innerloop): /* r3 holds guest_state */ /* r4 holds do_profiling */ /* ----- entry point to VG_(run_innerloop) ----- */ /* PPC64 ABI saves LR->16(prt_sp), CR->8(prt_sp)) */ /* Save lr, cr */ mflr 0 std 0,16(1) mfcr 0 std 0,8(1) /* New stack frame */ stdu 1,-624(1) /* sp should maintain 16-byte alignment */ /* Save callee-saved registers... */ /* Floating-point reg save area : 144 bytes */ stfd 31,616(1) stfd 30,608(1) stfd 29,600(1) stfd 28,592(1) stfd 27,584(1) stfd 26,576(1) stfd 25,568(1) stfd 24,560(1) stfd 23,552(1) stfd 22,544(1) stfd 21,536(1) stfd 20,528(1) stfd 19,520(1) stfd 18,512(1) stfd 17,504(1) stfd 16,496(1) stfd 15,488(1) stfd 14,480(1) /* General reg save area : 144 bytes */ std 31,472(1) std 30,464(1) std 29,456(1) std 28,448(1) std 27,440(1) std 26,432(1) std 25,424(1) std 24,416(1) std 23,408(1) std 22,400(1) std 21,392(1) std 20,384(1) std 19,376(1) std 18,368(1) std 17,360(1) std 16,352(1) std 15,344(1) std 14,336(1) /* Probably not necessary to save r13 (thread-specific ptr), as VEX stays clear of it... but what the hey. */ std 13,328(1) /* It's necessary to save/restore VRSAVE in the AIX / Darwin ABI. The Linux kernel might not actually use VRSAVE for its intended purpose, but it should be harmless to preserve anyway. */ /* r3, r4 are live here, so use r5 */ ld 5,.tocent__vgPlain_machine_ppc64_has_VMX@toc(2) ld 5,0(5) cmpldi 5,0 beq .LafterVMX1 /* VRSAVE save word : 32 bytes */ mfspr 5,256 /* vrsave reg is spr number 256 */ stw 5,324(1) /* Alignment padding : 4 bytes */ /* Vector reg save area (quadword aligned) : 192 bytes */ li 5,304 stvx 31,5,1 li 5,288 stvx 30,5,1 li 5,272 stvx 29,5,1 li 5,256 stvx 28,5,1 li 5,240 stvx 27,5,1 li 5,224 stvx 26,5,1 li 5,208 stvx 25,5,1 li 5,192 stvx 24,5,1 li 5,176 stvx 23,5,1 li 5,160 stvx 22,5,1 li 5,144 stvx 21,5,1 li 5,128 stvx 20,5,1 .LafterVMX1: /* Local variable space... */ /* r3 holds guest_state */ /* r4 holds do_profiling */ mr 31,3 std 3,104(1) /* spill orig guest_state ptr */ /* 96(sp) used later to check FPSCR[RM] */ /* 88(sp) used later to load fpscr with zero */ /* 48:87(sp) free */ /* Linkage Area (reserved) 40(sp) : TOC 32(sp) : link editor doubleword 24(sp) : compiler doubleword 16(sp) : LR 8(sp) : CR 0(sp) : back-chain */ // CAB TODO: Use a caller-saved reg for orig guest_state ptr // - rem to set non-allocateable in isel.c /* hold dispatch_ctr (=32bit value) in r29 */ ld 29,.tocent__vgPlain_dispatch_ctr@toc(2) lwz 29,0(29) /* 32-bit zero-extending load */ /* set host FPU control word to the default mode expected by VEX-generated code. See comments in libvex.h for more info. */ /* => get zero into f3 (tedious) fsub 3,3,3 is not a reliable way to do this, since if f3 holds a NaN or similar then we don't necessarily wind up with zero. */ li 5,0 stw 5,88(1) lfs 3,88(1) mtfsf 0xFF,3 /* fpscr = lo32 of f3 */ /* set host AltiVec control word to the default mode expected by VEX-generated code. */ ld 5,.tocent__vgPlain_machine_ppc64_has_VMX@toc(2) ld 5,0(5) cmpldi 5,0 beq .LafterVMX2 vspltisw 3,0x0 /* generate zero */ mtvscr 3 .LafterVMX2: /* make a stack frame for the code we are calling */ stdu 1,-48(1) /* fetch %CIA into r3 */ ld 3,OFFSET_ppc64_CIA(31) /* fall into main loop (the right one) */ /* r4 = do_profiling. It's probably trashed after here, but that's OK: we don't need it after here. */ cmplwi 4,0 beq .VG_(run_innerloop__dispatch_unprofiled) b .VG_(run_innerloop__dispatch_profiled) /*NOTREACHED*/ /*----------------------------------------------------*/ /*--- NO-PROFILING (standard) dispatcher ---*/ /*----------------------------------------------------*/ .section ".text" .align 2 .globl VG_(run_innerloop__dispatch_unprofiled) .section ".opd","aw" .align 3 VG_(run_innerloop__dispatch_unprofiled): .quad .VG_(run_innerloop__dispatch_unprofiled),.TOC.@tocbase,0 .previous .type .VG_(run_innerloop__dispatch_unprofiled),@function .globl .VG_(run_innerloop__dispatch_unprofiled) .VG_(run_innerloop__dispatch_unprofiled): /* At entry: Live regs: r1 (=sp) r2 (toc pointer) r3 (=CIA = next guest address) r29 (=dispatch_ctr) r31 (=guest_state) Stack state: 152(r1) (=orig guest_state) 144(r1) (=var space for FPSCR[RM]) */ /* Has the guest state pointer been messed with? If yes, exit. Also set up & VG_(tt_fast) early in an attempt at better scheduling. */ ld 9,152(1) /* original guest_state ptr */ ld 5, .tocent__vgPlain_tt_fast@toc(2) /* &VG_(tt_fast) */ cmpd 9,31 bne .gsp_changed /* save the jump address in the guest state */ std 3,OFFSET_ppc64_CIA(31) /* Are we out of timeslice? If yes, defer to scheduler. */ subi 29,29,1 cmpldi 29,0 beq .counter_is_zero /* try a fast lookup in the translation cache */ /* r4 = VG_TT_FAST_HASH(addr) * sizeof(FastCacheEntry) = ((r3 >>u 2) & VG_TT_FAST_MASK) << 4 */ rldicl 4,3, 62, 64-VG_TT_FAST_BITS /* entry# */ sldi 4,4,4 /* entry# * sizeof(FastCacheEntry) */ add 5,5,4 /* & VG_(tt_fast)[entry#] */ ld 6,0(5) /* .guest */ ld 7,8(5) /* .host */ cmpd 3,6 bne .fast_lookup_failed /* Found a match. Call .host. */ mtctr 7 bctrl /* On return from guest code: r3 holds destination (original) address. r31 may be unchanged (guest_state), or may indicate further details of the control transfer requested to *r3. */ /* start over */ b .VG_(run_innerloop__dispatch_unprofiled) /*NOTREACHED*/ .size VG_(run_innerloop), .-VG_(run_innerloop) /*----------------------------------------------------*/ /*--- PROFILING dispatcher (can be much slower) ---*/ /*----------------------------------------------------*/ .section ".text" .align 2 .globl VG_(run_innerloop__dispatch_profiled) .section ".opd","aw" .align 3 VG_(run_innerloop__dispatch_profiled): .quad .VG_(run_innerloop__dispatch_profiled),.TOC.@tocbase,0 .previous .type .VG_(run_innerloop__dispatch_profiled),@function .globl .VG_(run_innerloop__dispatch_profiled) .VG_(run_innerloop__dispatch_profiled): /* At entry: Live regs: r1 (=sp) r2 (toc pointer) r3 (=CIA = next guest address) r29 (=dispatch_ctr) r31 (=guest_state) Stack state: 152(r1) (=orig guest_state) 144(r1) (=var space for FPSCR[RM]) */ /* Has the guest state pointer been messed with? If yes, exit. Also set up & VG_(tt_fast) early in an attempt at better scheduling. */ ld 9,152(1) /* original guest_state ptr */ ld 5, .tocent__vgPlain_tt_fast@toc(2) /* &VG_(tt_fast) */ cmpd 9,31 bne .gsp_changed /* save the jump address in the guest state */ std 3,OFFSET_ppc64_CIA(31) /* Are we out of timeslice? If yes, defer to scheduler. */ subi 29,29,1 cmpldi 29,0 beq .counter_is_zero /* try a fast lookup in the translation cache */ /* r4 = VG_TT_FAST_HASH(addr) * sizeof(FastCacheEntry) = ((r3 >>u 2) & VG_TT_FAST_MASK) << 4 */ rldicl 4,3, 62, 64-VG_TT_FAST_BITS /* entry# */ sldi 4,4,4 /* entry# * sizeof(FastCacheEntry) */ add 5,5,4 /* & VG_(tt_fast)[entry#] */ ld 6,0(5) /* .guest */ ld 7,8(5) /* .host */ cmpd 3,6 bne .fast_lookup_failed /* increment bb profile counter VG_(tt_fastN)[x] (=32bit val) */ ld 9, .tocent__vgPlain_tt_fastN@toc(2) srdi 4, 4,1 /* entry# * sizeof(UInt*) */ ldx 9, 9,4 /* r7 = VG_(tt_fastN)[VG_TT_HASH(addr)] */ lwz 6, 0(9) /* *(UInt*)r7 ++ */ addi 6, 6,1 stw 6, 0(9) /* Found a match. Call .host. */ mtctr 7 bctrl /* On return from guest code: r3 holds destination (original) address. r31 may be unchanged (guest_state), or may indicate further details of the control transfer requested to *r3. */ /* start over */ b .VG_(run_innerloop__dispatch_profiled) /*NOTREACHED*/ .size VG_(run_a_noredir_translation), .-VG_(run_a_noredir_translation) /*----------------------------------------------------*/ /*--- exit points ---*/ /*----------------------------------------------------*/ .gsp_changed: /* Someone messed with the gsp (in r31). Have to defer to scheduler to resolve this. dispatch ctr is not yet decremented, so no need to increment. */ /* %CIA is NOT up to date here. First, need to write %r3 back to %CIA, but without trashing %r31 since that holds the value we want to return to the scheduler. Hence use %r5 transiently for the guest state pointer. */ ld 5,152(1) /* original guest_state ptr */ std 3,OFFSET_ppc64_CIA(5) mr 3,31 /* r3 = new gsp value */ b .run_innerloop_exit /*NOTREACHED*/ .counter_is_zero: /* %CIA is up to date */ /* back out decrement of the dispatch counter */ addi 29,29,1 li 3,VG_TRC_INNER_COUNTERZERO b .run_innerloop_exit .fast_lookup_failed: /* %CIA is up to date */ /* back out decrement of the dispatch counter */ addi 29,29,1 li 3,VG_TRC_INNER_FASTMISS b .run_innerloop_exit /* All exits from the dispatcher go through here. r3 holds the return value. */ .run_innerloop_exit: /* We're leaving. Check that nobody messed with VSCR or FPSCR. */ /* Set fpscr back to a known state, since vex-generated code may have messed with fpscr[rm]. */ li 5,0 addi 1,1,-16 stw 5,0(1) lfs 3,0(1) addi 1,1,16 mtfsf 0xFF,3 /* fpscr = f3 */ /* Using r11 - value used again further on, so don't trash! */ ld 11,.tocent__vgPlain_machine_ppc64_has_VMX@toc(2) ld 11,0(11) cmpldi 11,0 beq .LafterVMX8 /* Check VSCR[NJ] == 1 */ /* first generate 4x 0x00010000 */ vspltisw 4,0x1 /* 4x 0x00000001 */ vspltisw 5,0x0 /* zero */ vsldoi 6,4,5,0x2 /* <<2*8 => 4x 0x00010000 */ /* retrieve VSCR and mask wanted bits */ mfvscr 7 vand 7,7,6 /* gives NJ flag */ vspltw 7,7,0x3 /* flags-word to all lanes */ vcmpequw. 8,6,7 /* CR[24] = 1 if v6 == v7 */ bt 24,.invariant_violation /* branch if all_equal */ .LafterVMX8: /* otherwise we're OK */ b .run_innerloop_exit_REALLY .invariant_violation: li 3,VG_TRC_INVARIANT_FAILED b .run_innerloop_exit_REALLY .run_innerloop_exit_REALLY: /* r3 holds VG_TRC_* value to return */ /* Return to parent stack */ addi 1,1,48 /* Write ctr to VG_(dispatch_ctr) (=32bit value) */ ld 5,.tocent__vgPlain_dispatch_ctr@toc(2) stw 29,0(5) /* Restore cr */ lwz 0,44(1) mtcr 0 /* Restore callee-saved registers... */ /* Floating-point regs */ lfd 31,616(1) lfd 30,608(1) lfd 29,600(1) lfd 28,592(1) lfd 27,584(1) lfd 26,576(1) lfd 25,568(1) lfd 24,560(1) lfd 23,552(1) lfd 22,544(1) lfd 21,536(1) lfd 20,528(1) lfd 19,520(1) lfd 18,512(1) lfd 17,504(1) lfd 16,496(1) lfd 15,488(1) lfd 14,480(1) /* General regs */ ld 31,472(1) ld 30,464(1) ld 29,456(1) ld 28,448(1) ld 27,440(1) ld 26,432(1) ld 25,424(1) ld 24,416(1) ld 23,408(1) ld 22,400(1) ld 21,392(1) ld 20,384(1) ld 19,376(1) ld 18,368(1) ld 17,360(1) ld 16,352(1) ld 15,344(1) ld 14,336(1) ld 13,328(1) /* r11 already holds VG_(machine_ppc64_has_VMX) value */ cmpldi 11,0 beq .LafterVMX9 /* VRSAVE */ lwz 4,324(1) mfspr 4,256 /* VRSAVE reg is spr number 256 */ /* Vector regs */ li 4,304 lvx 31,4,1 li 4,288 lvx 30,4,1 li 4,272 lvx 29,4,1 li 4,256 lvx 28,4,1 li 4,240 lvx 27,4,1 li 4,224 lvx 26,4,1 li 4,208 lvx 25,4,1 li 4,192 lvx 24,4,1 li 4,176 lvx 23,4,1 li 4,160 lvx 22,4,1 li 4,144 lvx 21,4,1 li 4,128 lvx 20,4,1 .LafterVMX9: /* reset cr, lr, sp */ ld 0,632(1) /* stack_size + 8 */ mtcr 0 ld 0,640(1) /* stack_size + 16 */ mtlr 0 addi 1,1,624 /* stack_size */ blr /*------------------------------------------------------------*/ /*--- ---*/ /*--- A special dispatcher, for running no-redir ---*/ /*--- translations. Just runs the given translation once. ---*/ /*--- ---*/ /*------------------------------------------------------------*/ /* signature: void VG_(run_a_noredir_translation) ( UWord* argblock ); */ /* Run a no-redir translation. argblock points to 4 UWords, 2 to carry args and 2 to carry results: 0: input: ptr to translation 1: input: ptr to guest state 2: output: next guest PC 3: output: guest state pointer afterwards (== thread return code) */ .section ".text" .align 2 .globl VG_(run_a_noredir_translation) .section ".opd","aw" .align 3 VG_(run_a_noredir_translation): .quad .VG_(run_a_noredir_translation),.TOC.@tocbase,0 .previous .type .VG_(run_a_noredir_translation),@function .globl .VG_(run_a_noredir_translation) .VG_(run_a_noredir_translation): /* save callee-save int regs, & lr */ stdu 1,-512(1) std 14,256(1) std 15,264(1) std 16,272(1) std 17,280(1) std 18,288(1) std 19,296(1) std 20,304(1) std 21,312(1) std 22,320(1) std 23,328(1) std 24,336(1) std 25,344(1) std 26,352(1) std 27,360(1) std 28,368(1) std 29,376(1) std 30,384(1) std 31,392(1) mflr 31 std 31,400(1) std 2,408(1) /* also preserve R2, just in case .. */ std 3,416(1) ld 31,8(3) ld 30,0(3) mtlr 30 blrl ld 4,416(1) std 3, 16(4) std 31,24(4) ld 14,256(1) ld 15,264(1) ld 16,272(1) ld 17,280(1) ld 18,288(1) ld 19,296(1) ld 20,304(1) ld 21,312(1) ld 22,320(1) ld 23,328(1) ld 24,336(1) ld 25,344(1) ld 26,352(1) ld 27,360(1) ld 28,368(1) ld 29,376(1) ld 30,384(1) ld 31,400(1) mtlr 31 ld 31,392(1) ld 2,408(1) /* also preserve R2, just in case .. */ addi 1,1,512 blr /* Let the linker know we don't need an executable stack */ .section .note.GNU-stack,"",@progbits /*--------------------------------------------------------------------*/ /*--- end ---*/ /*--------------------------------------------------------------------*/