gcc/libgcc/config/i386/linux-unwind.h

197 lines
6.8 KiB
C
Raw Normal View History

tm.texi (MD_UNWIND_SUPPORT): Document. * doc/tm.texi (MD_UNWIND_SUPPORT): Document. (MD_FALLBACK_FRAME_STATE_FOR): Update. * unwind-dw2.c (MD_UNWIND_SUPPORT): #include if defined. (uw_frame_state_for): Adjust MD_FALLBACK_FRAME_STATE_FOR invocation. (MD_FROB_UPDATE_CONTEXT): Remove default. (uw_update_context_1): Instead #ifdef invocation. * config/ia64/unwind-ia64.c (MD_UNWIND_SUPPORT): #include if defined. (uw_frame_state_for): Adjust MD_FALLBACK_FRAME_STATE_FOR invocation. * config/alpha/gnu.h (MD_FALLBACK_FRAME_STATE_FOR): Don't undef. (MD_UNWIND_SUPPORT): Undefine this instead. * config/i386/gnu.h: Likewise. * config/alpha/linux-unwind.h: New file, macro converted to function, extracted from.. * config/alpha/linux.h (MD_FALLBACK_FRAME_STATE_FOR): ..this. (MD_UNWIND_SUPPORT): Define. * config/alpha/vms-unwind.h, config/alpha/vms.h: Likewise. * config/i386/linux-unwind.h, config/i386/linux.h, config/i386/linux64.h: Likewise. * config/ia64/linux-unwind.h, config/ia64/linux.h: Likewise. MD_HANDLE_UNWABI too. * config/mips/linux-unwind.h, config/mips/linux.h: Likewise. * config/pa/linux-unwind.h, config/pa/pa32-linux.h: Likewise. * config/rs6000/darwin-unwind.h, config/rs6000/darwin.h: Likewise. * config/s390/linux-unwind.h, config/s390/linux.h: Likewise. * config/sparc/linux-unwind.h, config/sparc/linux.h, config/sparc/linux64.h: Likewise. * config/sh/linux-unwind.h, config/sh/linux.h: Likewise, but merge SH_FALLBACK_FRAME_FLOAT_STATE into sh_fallback_frame_state. * config/rs6000/linux-unwind.h, config/rs6000/linux.h, config/rs6000/linux64.h: Likewise. Split out get_sigcontext function. Use ARG_POINTER_REGNUM for 32-bit temp reg too. From-SVN: r87167
2004-09-08 00:17:19 +00:00
/* DWARF2 EH unwinding support for AMD x86-64 and x86.
2025-01-02 11:59:57 +01:00
Copyright (C) 2004-2025 Free Software Foundation, Inc.
tm.texi (MD_UNWIND_SUPPORT): Document. * doc/tm.texi (MD_UNWIND_SUPPORT): Document. (MD_FALLBACK_FRAME_STATE_FOR): Update. * unwind-dw2.c (MD_UNWIND_SUPPORT): #include if defined. (uw_frame_state_for): Adjust MD_FALLBACK_FRAME_STATE_FOR invocation. (MD_FROB_UPDATE_CONTEXT): Remove default. (uw_update_context_1): Instead #ifdef invocation. * config/ia64/unwind-ia64.c (MD_UNWIND_SUPPORT): #include if defined. (uw_frame_state_for): Adjust MD_FALLBACK_FRAME_STATE_FOR invocation. * config/alpha/gnu.h (MD_FALLBACK_FRAME_STATE_FOR): Don't undef. (MD_UNWIND_SUPPORT): Undefine this instead. * config/i386/gnu.h: Likewise. * config/alpha/linux-unwind.h: New file, macro converted to function, extracted from.. * config/alpha/linux.h (MD_FALLBACK_FRAME_STATE_FOR): ..this. (MD_UNWIND_SUPPORT): Define. * config/alpha/vms-unwind.h, config/alpha/vms.h: Likewise. * config/i386/linux-unwind.h, config/i386/linux.h, config/i386/linux64.h: Likewise. * config/ia64/linux-unwind.h, config/ia64/linux.h: Likewise. MD_HANDLE_UNWABI too. * config/mips/linux-unwind.h, config/mips/linux.h: Likewise. * config/pa/linux-unwind.h, config/pa/pa32-linux.h: Likewise. * config/rs6000/darwin-unwind.h, config/rs6000/darwin.h: Likewise. * config/s390/linux-unwind.h, config/s390/linux.h: Likewise. * config/sparc/linux-unwind.h, config/sparc/linux.h, config/sparc/linux64.h: Likewise. * config/sh/linux-unwind.h, config/sh/linux.h: Likewise, but merge SH_FALLBACK_FRAME_FLOAT_STATE into sh_fallback_frame_state. * config/rs6000/linux-unwind.h, config/rs6000/linux.h, config/rs6000/linux64.h: Likewise. Split out get_sigcontext function. Use ARG_POINTER_REGNUM for 32-bit temp reg too. From-SVN: r87167
2004-09-08 00:17:19 +00:00
This file is part of GCC.
GCC 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 3, or (at your option)
tm.texi (MD_UNWIND_SUPPORT): Document. * doc/tm.texi (MD_UNWIND_SUPPORT): Document. (MD_FALLBACK_FRAME_STATE_FOR): Update. * unwind-dw2.c (MD_UNWIND_SUPPORT): #include if defined. (uw_frame_state_for): Adjust MD_FALLBACK_FRAME_STATE_FOR invocation. (MD_FROB_UPDATE_CONTEXT): Remove default. (uw_update_context_1): Instead #ifdef invocation. * config/ia64/unwind-ia64.c (MD_UNWIND_SUPPORT): #include if defined. (uw_frame_state_for): Adjust MD_FALLBACK_FRAME_STATE_FOR invocation. * config/alpha/gnu.h (MD_FALLBACK_FRAME_STATE_FOR): Don't undef. (MD_UNWIND_SUPPORT): Undefine this instead. * config/i386/gnu.h: Likewise. * config/alpha/linux-unwind.h: New file, macro converted to function, extracted from.. * config/alpha/linux.h (MD_FALLBACK_FRAME_STATE_FOR): ..this. (MD_UNWIND_SUPPORT): Define. * config/alpha/vms-unwind.h, config/alpha/vms.h: Likewise. * config/i386/linux-unwind.h, config/i386/linux.h, config/i386/linux64.h: Likewise. * config/ia64/linux-unwind.h, config/ia64/linux.h: Likewise. MD_HANDLE_UNWABI too. * config/mips/linux-unwind.h, config/mips/linux.h: Likewise. * config/pa/linux-unwind.h, config/pa/pa32-linux.h: Likewise. * config/rs6000/darwin-unwind.h, config/rs6000/darwin.h: Likewise. * config/s390/linux-unwind.h, config/s390/linux.h: Likewise. * config/sparc/linux-unwind.h, config/sparc/linux.h, config/sparc/linux64.h: Likewise. * config/sh/linux-unwind.h, config/sh/linux.h: Likewise, but merge SH_FALLBACK_FRAME_FLOAT_STATE into sh_fallback_frame_state. * config/rs6000/linux-unwind.h, config/rs6000/linux.h, config/rs6000/linux64.h: Likewise. Split out get_sigcontext function. Use ARG_POINTER_REGNUM for 32-bit temp reg too. From-SVN: r87167
2004-09-08 00:17:19 +00:00
any later version.
GCC 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.
Under Section 7 of GPL version 3, you are granted additional
permissions described in the GCC Runtime Library Exception, version
3.1, as published by the Free Software Foundation.
You should have received a copy of the GNU General Public License and
a copy of the GCC Runtime Library Exception along with this program;
see the files COPYING3 and COPYING.RUNTIME respectively. If not, see
<http://www.gnu.org/licenses/>. */
tm.texi (MD_UNWIND_SUPPORT): Document. * doc/tm.texi (MD_UNWIND_SUPPORT): Document. (MD_FALLBACK_FRAME_STATE_FOR): Update. * unwind-dw2.c (MD_UNWIND_SUPPORT): #include if defined. (uw_frame_state_for): Adjust MD_FALLBACK_FRAME_STATE_FOR invocation. (MD_FROB_UPDATE_CONTEXT): Remove default. (uw_update_context_1): Instead #ifdef invocation. * config/ia64/unwind-ia64.c (MD_UNWIND_SUPPORT): #include if defined. (uw_frame_state_for): Adjust MD_FALLBACK_FRAME_STATE_FOR invocation. * config/alpha/gnu.h (MD_FALLBACK_FRAME_STATE_FOR): Don't undef. (MD_UNWIND_SUPPORT): Undefine this instead. * config/i386/gnu.h: Likewise. * config/alpha/linux-unwind.h: New file, macro converted to function, extracted from.. * config/alpha/linux.h (MD_FALLBACK_FRAME_STATE_FOR): ..this. (MD_UNWIND_SUPPORT): Define. * config/alpha/vms-unwind.h, config/alpha/vms.h: Likewise. * config/i386/linux-unwind.h, config/i386/linux.h, config/i386/linux64.h: Likewise. * config/ia64/linux-unwind.h, config/ia64/linux.h: Likewise. MD_HANDLE_UNWABI too. * config/mips/linux-unwind.h, config/mips/linux.h: Likewise. * config/pa/linux-unwind.h, config/pa/pa32-linux.h: Likewise. * config/rs6000/darwin-unwind.h, config/rs6000/darwin.h: Likewise. * config/s390/linux-unwind.h, config/s390/linux.h: Likewise. * config/sparc/linux-unwind.h, config/sparc/linux.h, config/sparc/linux64.h: Likewise. * config/sh/linux-unwind.h, config/sh/linux.h: Likewise, but merge SH_FALLBACK_FRAME_FLOAT_STATE into sh_fallback_frame_state. * config/rs6000/linux-unwind.h, config/rs6000/linux.h, config/rs6000/linux64.h: Likewise. Split out get_sigcontext function. Use ARG_POINTER_REGNUM for 32-bit temp reg too. From-SVN: r87167
2004-09-08 00:17:19 +00:00
Add Intel CET support for EH in libgcc. Control-flow Enforcement Technology (CET), published by Intel, introduces the Shadow Stack feature, which ensures a return from a function is done to exactly the same location from where the function was called. When EH is present the control-flow transfer may skip some stack frames and the shadow stack has to be adjusted not to signal a violation of a control-flow transfer. It's done by counting a number of skiping frames and adjasting shadow stack pointer by this number. Having new semantic of the 'ret' instruction if CET is supported in HW the 'ret' instruction cannot be generated in ix86_expand_epilogue when we are returning after EH is processed. Added a code in ix86_expand_epilogue to adjust Shadow Stack pointer and to generate an indirect jump instead of 'ret'. As sp register is used during this adjustment thus the argument in pro_epilogue_adjust_stack is changed to update cfa_reg based on whether control-flow instrumentation is set. Without updating the cfa_reg field there is an assert later in dwarf2 pass related to mismatch the stack register and cfa_reg value. gcc/ * config/i386/i386.c (ix86_expand_epilogue): Change simple return to indirect jump for EH return if control-flow protection is enabled. Change explicit 'false' argument in pro_epilogue_adjust_stack with a value of flag_cf_protection. * config/i386/i386.md (simple_return_indirect_internal): Remove SImode restriction to support 64-bit. libgcc/ * config/i386/linux-unwind.h: Include config/i386/shadow-stack-unwind.h. * config/i386/shadow-stack-unwind.h: New file. * unwind-dw2.c: (uw_install_context): Add a frame parameter and pass it to _Unwind_Frames_Extra. * unwind-generic.h (_Unwind_Frames_Extra): New. * unwind.inc (_Unwind_RaiseException_Phase2): Add frames_p parameter. Add local variable frames to count number of frames. (_Unwind_ForcedUnwind_Phase2): Likewise. (_Unwind_RaiseException): Add local variable frames to count number of frames, pass it to _Unwind_RaiseException_Phase2 and uw_install_context. (_Unwind_ForcedUnwind): Likewise. (_Unwind_Resume): Likewise. (_Unwind_Resume_or_Rethrow): Likewise. From-SVN: r254876
2017-11-17 16:21:23 +01:00
/* Unwind shadow stack for -fcf-protection -mshstk. */
#if defined __SHSTK__ && defined __CET__ && (__CET__ & 2) != 0
Add Intel CET support for EH in libgcc. Control-flow Enforcement Technology (CET), published by Intel, introduces the Shadow Stack feature, which ensures a return from a function is done to exactly the same location from where the function was called. When EH is present the control-flow transfer may skip some stack frames and the shadow stack has to be adjusted not to signal a violation of a control-flow transfer. It's done by counting a number of skiping frames and adjasting shadow stack pointer by this number. Having new semantic of the 'ret' instruction if CET is supported in HW the 'ret' instruction cannot be generated in ix86_expand_epilogue when we are returning after EH is processed. Added a code in ix86_expand_epilogue to adjust Shadow Stack pointer and to generate an indirect jump instead of 'ret'. As sp register is used during this adjustment thus the argument in pro_epilogue_adjust_stack is changed to update cfa_reg based on whether control-flow instrumentation is set. Without updating the cfa_reg field there is an assert later in dwarf2 pass related to mismatch the stack register and cfa_reg value. gcc/ * config/i386/i386.c (ix86_expand_epilogue): Change simple return to indirect jump for EH return if control-flow protection is enabled. Change explicit 'false' argument in pro_epilogue_adjust_stack with a value of flag_cf_protection. * config/i386/i386.md (simple_return_indirect_internal): Remove SImode restriction to support 64-bit. libgcc/ * config/i386/linux-unwind.h: Include config/i386/shadow-stack-unwind.h. * config/i386/shadow-stack-unwind.h: New file. * unwind-dw2.c: (uw_install_context): Add a frame parameter and pass it to _Unwind_Frames_Extra. * unwind-generic.h (_Unwind_Frames_Extra): New. * unwind.inc (_Unwind_RaiseException_Phase2): Add frames_p parameter. Add local variable frames to count number of frames. (_Unwind_ForcedUnwind_Phase2): Likewise. (_Unwind_RaiseException): Add local variable frames to count number of frames, pass it to _Unwind_RaiseException_Phase2 and uw_install_context. (_Unwind_ForcedUnwind): Likewise. (_Unwind_Resume): Likewise. (_Unwind_Resume_or_Rethrow): Likewise. From-SVN: r254876
2017-11-17 16:21:23 +01:00
# include "config/i386/shadow-stack-unwind.h"
#endif
tm.texi (MD_UNWIND_SUPPORT): Document. * doc/tm.texi (MD_UNWIND_SUPPORT): Document. (MD_FALLBACK_FRAME_STATE_FOR): Update. * unwind-dw2.c (MD_UNWIND_SUPPORT): #include if defined. (uw_frame_state_for): Adjust MD_FALLBACK_FRAME_STATE_FOR invocation. (MD_FROB_UPDATE_CONTEXT): Remove default. (uw_update_context_1): Instead #ifdef invocation. * config/ia64/unwind-ia64.c (MD_UNWIND_SUPPORT): #include if defined. (uw_frame_state_for): Adjust MD_FALLBACK_FRAME_STATE_FOR invocation. * config/alpha/gnu.h (MD_FALLBACK_FRAME_STATE_FOR): Don't undef. (MD_UNWIND_SUPPORT): Undefine this instead. * config/i386/gnu.h: Likewise. * config/alpha/linux-unwind.h: New file, macro converted to function, extracted from.. * config/alpha/linux.h (MD_FALLBACK_FRAME_STATE_FOR): ..this. (MD_UNWIND_SUPPORT): Define. * config/alpha/vms-unwind.h, config/alpha/vms.h: Likewise. * config/i386/linux-unwind.h, config/i386/linux.h, config/i386/linux64.h: Likewise. * config/ia64/linux-unwind.h, config/ia64/linux.h: Likewise. MD_HANDLE_UNWABI too. * config/mips/linux-unwind.h, config/mips/linux.h: Likewise. * config/pa/linux-unwind.h, config/pa/pa32-linux.h: Likewise. * config/rs6000/darwin-unwind.h, config/rs6000/darwin.h: Likewise. * config/s390/linux-unwind.h, config/s390/linux.h: Likewise. * config/sparc/linux-unwind.h, config/sparc/linux.h, config/sparc/linux64.h: Likewise. * config/sh/linux-unwind.h, config/sh/linux.h: Likewise, but merge SH_FALLBACK_FRAME_FLOAT_STATE into sh_fallback_frame_state. * config/rs6000/linux-unwind.h, config/rs6000/linux.h, config/rs6000/linux64.h: Likewise. Split out get_sigcontext function. Use ARG_POINTER_REGNUM for 32-bit temp reg too. From-SVN: r87167
2004-09-08 00:17:19 +00:00
/* Do code reading to identify a signal frame, and set the frame
state data appropriately. See unwind-dw2.c for the structs.
Don't use this at all if inhibit_libc is used. */
#ifndef inhibit_libc
#include <signal.h>
#include <sys/ucontext.h>
#ifdef __x86_64__
tm.texi (MD_UNWIND_SUPPORT): Document. * doc/tm.texi (MD_UNWIND_SUPPORT): Document. (MD_FALLBACK_FRAME_STATE_FOR): Update. * unwind-dw2.c (MD_UNWIND_SUPPORT): #include if defined. (uw_frame_state_for): Adjust MD_FALLBACK_FRAME_STATE_FOR invocation. (MD_FROB_UPDATE_CONTEXT): Remove default. (uw_update_context_1): Instead #ifdef invocation. * config/ia64/unwind-ia64.c (MD_UNWIND_SUPPORT): #include if defined. (uw_frame_state_for): Adjust MD_FALLBACK_FRAME_STATE_FOR invocation. * config/alpha/gnu.h (MD_FALLBACK_FRAME_STATE_FOR): Don't undef. (MD_UNWIND_SUPPORT): Undefine this instead. * config/i386/gnu.h: Likewise. * config/alpha/linux-unwind.h: New file, macro converted to function, extracted from.. * config/alpha/linux.h (MD_FALLBACK_FRAME_STATE_FOR): ..this. (MD_UNWIND_SUPPORT): Define. * config/alpha/vms-unwind.h, config/alpha/vms.h: Likewise. * config/i386/linux-unwind.h, config/i386/linux.h, config/i386/linux64.h: Likewise. * config/ia64/linux-unwind.h, config/ia64/linux.h: Likewise. MD_HANDLE_UNWABI too. * config/mips/linux-unwind.h, config/mips/linux.h: Likewise. * config/pa/linux-unwind.h, config/pa/pa32-linux.h: Likewise. * config/rs6000/darwin-unwind.h, config/rs6000/darwin.h: Likewise. * config/s390/linux-unwind.h, config/s390/linux.h: Likewise. * config/sparc/linux-unwind.h, config/sparc/linux.h, config/sparc/linux64.h: Likewise. * config/sh/linux-unwind.h, config/sh/linux.h: Likewise, but merge SH_FALLBACK_FRAME_FLOAT_STATE into sh_fallback_frame_state. * config/rs6000/linux-unwind.h, config/rs6000/linux.h, config/rs6000/linux64.h: Likewise. Split out get_sigcontext function. Use ARG_POINTER_REGNUM for 32-bit temp reg too. From-SVN: r87167
2004-09-08 00:17:19 +00:00
#define MD_FALLBACK_FRAME_STATE_FOR x86_64_fallback_frame_state
static _Unwind_Reason_Code
x86_64_fallback_frame_state (struct _Unwind_Context *context,
_Unwind_FrameState *fs)
{
unsigned char *pc = context->ra;
struct sigcontext *sc;
long new_cfa;
/* movq $__NR_rt_sigreturn, %rax ; syscall. */
#ifdef __LP64__
#define RT_SIGRETURN_SYSCALL 0x050f0000000fc0c7ULL
#else
#define RT_SIGRETURN_SYSCALL 0x050f40000201c0c7ULL
#endif
tm.texi (MD_UNWIND_SUPPORT): Document. * doc/tm.texi (MD_UNWIND_SUPPORT): Document. (MD_FALLBACK_FRAME_STATE_FOR): Update. * unwind-dw2.c (MD_UNWIND_SUPPORT): #include if defined. (uw_frame_state_for): Adjust MD_FALLBACK_FRAME_STATE_FOR invocation. (MD_FROB_UPDATE_CONTEXT): Remove default. (uw_update_context_1): Instead #ifdef invocation. * config/ia64/unwind-ia64.c (MD_UNWIND_SUPPORT): #include if defined. (uw_frame_state_for): Adjust MD_FALLBACK_FRAME_STATE_FOR invocation. * config/alpha/gnu.h (MD_FALLBACK_FRAME_STATE_FOR): Don't undef. (MD_UNWIND_SUPPORT): Undefine this instead. * config/i386/gnu.h: Likewise. * config/alpha/linux-unwind.h: New file, macro converted to function, extracted from.. * config/alpha/linux.h (MD_FALLBACK_FRAME_STATE_FOR): ..this. (MD_UNWIND_SUPPORT): Define. * config/alpha/vms-unwind.h, config/alpha/vms.h: Likewise. * config/i386/linux-unwind.h, config/i386/linux.h, config/i386/linux64.h: Likewise. * config/ia64/linux-unwind.h, config/ia64/linux.h: Likewise. MD_HANDLE_UNWABI too. * config/mips/linux-unwind.h, config/mips/linux.h: Likewise. * config/pa/linux-unwind.h, config/pa/pa32-linux.h: Likewise. * config/rs6000/darwin-unwind.h, config/rs6000/darwin.h: Likewise. * config/s390/linux-unwind.h, config/s390/linux.h: Likewise. * config/sparc/linux-unwind.h, config/sparc/linux.h, config/sparc/linux64.h: Likewise. * config/sh/linux-unwind.h, config/sh/linux.h: Likewise, but merge SH_FALLBACK_FRAME_FLOAT_STATE into sh_fallback_frame_state. * config/rs6000/linux-unwind.h, config/rs6000/linux.h, config/rs6000/linux64.h: Likewise. Split out get_sigcontext function. Use ARG_POINTER_REGNUM for 32-bit temp reg too. From-SVN: r87167
2004-09-08 00:17:19 +00:00
if (*(unsigned char *)(pc+0) == 0x48
&& *(unsigned long long *)(pc+1) == RT_SIGRETURN_SYSCALL)
tm.texi (MD_UNWIND_SUPPORT): Document. * doc/tm.texi (MD_UNWIND_SUPPORT): Document. (MD_FALLBACK_FRAME_STATE_FOR): Update. * unwind-dw2.c (MD_UNWIND_SUPPORT): #include if defined. (uw_frame_state_for): Adjust MD_FALLBACK_FRAME_STATE_FOR invocation. (MD_FROB_UPDATE_CONTEXT): Remove default. (uw_update_context_1): Instead #ifdef invocation. * config/ia64/unwind-ia64.c (MD_UNWIND_SUPPORT): #include if defined. (uw_frame_state_for): Adjust MD_FALLBACK_FRAME_STATE_FOR invocation. * config/alpha/gnu.h (MD_FALLBACK_FRAME_STATE_FOR): Don't undef. (MD_UNWIND_SUPPORT): Undefine this instead. * config/i386/gnu.h: Likewise. * config/alpha/linux-unwind.h: New file, macro converted to function, extracted from.. * config/alpha/linux.h (MD_FALLBACK_FRAME_STATE_FOR): ..this. (MD_UNWIND_SUPPORT): Define. * config/alpha/vms-unwind.h, config/alpha/vms.h: Likewise. * config/i386/linux-unwind.h, config/i386/linux.h, config/i386/linux64.h: Likewise. * config/ia64/linux-unwind.h, config/ia64/linux.h: Likewise. MD_HANDLE_UNWABI too. * config/mips/linux-unwind.h, config/mips/linux.h: Likewise. * config/pa/linux-unwind.h, config/pa/pa32-linux.h: Likewise. * config/rs6000/darwin-unwind.h, config/rs6000/darwin.h: Likewise. * config/s390/linux-unwind.h, config/s390/linux.h: Likewise. * config/sparc/linux-unwind.h, config/sparc/linux.h, config/sparc/linux64.h: Likewise. * config/sh/linux-unwind.h, config/sh/linux.h: Likewise, but merge SH_FALLBACK_FRAME_FLOAT_STATE into sh_fallback_frame_state. * config/rs6000/linux-unwind.h, config/rs6000/linux.h, config/rs6000/linux64.h: Likewise. Split out get_sigcontext function. Use ARG_POINTER_REGNUM for 32-bit temp reg too. From-SVN: r87167
2004-09-08 00:17:19 +00:00
{
ucontext_t *uc_ = context->cfa;
/* The void * cast is necessary to avoid an aliasing warning.
The aliasing warning is correct, but should not be a problem
because it does not alias anything. */
sc = (struct sigcontext *) (void *) &uc_->uc_mcontext;
tm.texi (MD_UNWIND_SUPPORT): Document. * doc/tm.texi (MD_UNWIND_SUPPORT): Document. (MD_FALLBACK_FRAME_STATE_FOR): Update. * unwind-dw2.c (MD_UNWIND_SUPPORT): #include if defined. (uw_frame_state_for): Adjust MD_FALLBACK_FRAME_STATE_FOR invocation. (MD_FROB_UPDATE_CONTEXT): Remove default. (uw_update_context_1): Instead #ifdef invocation. * config/ia64/unwind-ia64.c (MD_UNWIND_SUPPORT): #include if defined. (uw_frame_state_for): Adjust MD_FALLBACK_FRAME_STATE_FOR invocation. * config/alpha/gnu.h (MD_FALLBACK_FRAME_STATE_FOR): Don't undef. (MD_UNWIND_SUPPORT): Undefine this instead. * config/i386/gnu.h: Likewise. * config/alpha/linux-unwind.h: New file, macro converted to function, extracted from.. * config/alpha/linux.h (MD_FALLBACK_FRAME_STATE_FOR): ..this. (MD_UNWIND_SUPPORT): Define. * config/alpha/vms-unwind.h, config/alpha/vms.h: Likewise. * config/i386/linux-unwind.h, config/i386/linux.h, config/i386/linux64.h: Likewise. * config/ia64/linux-unwind.h, config/ia64/linux.h: Likewise. MD_HANDLE_UNWABI too. * config/mips/linux-unwind.h, config/mips/linux.h: Likewise. * config/pa/linux-unwind.h, config/pa/pa32-linux.h: Likewise. * config/rs6000/darwin-unwind.h, config/rs6000/darwin.h: Likewise. * config/s390/linux-unwind.h, config/s390/linux.h: Likewise. * config/sparc/linux-unwind.h, config/sparc/linux.h, config/sparc/linux64.h: Likewise. * config/sh/linux-unwind.h, config/sh/linux.h: Likewise, but merge SH_FALLBACK_FRAME_FLOAT_STATE into sh_fallback_frame_state. * config/rs6000/linux-unwind.h, config/rs6000/linux.h, config/rs6000/linux64.h: Likewise. Split out get_sigcontext function. Use ARG_POINTER_REGNUM for 32-bit temp reg too. From-SVN: r87167
2004-09-08 00:17:19 +00:00
}
else
return _URC_END_OF_STACK;
new_cfa = sc->rsp;
fs->regs.cfa_how = CFA_REG_OFFSET;
tm.texi (MD_UNWIND_SUPPORT): Document. * doc/tm.texi (MD_UNWIND_SUPPORT): Document. (MD_FALLBACK_FRAME_STATE_FOR): Update. * unwind-dw2.c (MD_UNWIND_SUPPORT): #include if defined. (uw_frame_state_for): Adjust MD_FALLBACK_FRAME_STATE_FOR invocation. (MD_FROB_UPDATE_CONTEXT): Remove default. (uw_update_context_1): Instead #ifdef invocation. * config/ia64/unwind-ia64.c (MD_UNWIND_SUPPORT): #include if defined. (uw_frame_state_for): Adjust MD_FALLBACK_FRAME_STATE_FOR invocation. * config/alpha/gnu.h (MD_FALLBACK_FRAME_STATE_FOR): Don't undef. (MD_UNWIND_SUPPORT): Undefine this instead. * config/i386/gnu.h: Likewise. * config/alpha/linux-unwind.h: New file, macro converted to function, extracted from.. * config/alpha/linux.h (MD_FALLBACK_FRAME_STATE_FOR): ..this. (MD_UNWIND_SUPPORT): Define. * config/alpha/vms-unwind.h, config/alpha/vms.h: Likewise. * config/i386/linux-unwind.h, config/i386/linux.h, config/i386/linux64.h: Likewise. * config/ia64/linux-unwind.h, config/ia64/linux.h: Likewise. MD_HANDLE_UNWABI too. * config/mips/linux-unwind.h, config/mips/linux.h: Likewise. * config/pa/linux-unwind.h, config/pa/pa32-linux.h: Likewise. * config/rs6000/darwin-unwind.h, config/rs6000/darwin.h: Likewise. * config/s390/linux-unwind.h, config/s390/linux.h: Likewise. * config/sparc/linux-unwind.h, config/sparc/linux.h, config/sparc/linux64.h: Likewise. * config/sh/linux-unwind.h, config/sh/linux.h: Likewise, but merge SH_FALLBACK_FRAME_FLOAT_STATE into sh_fallback_frame_state. * config/rs6000/linux-unwind.h, config/rs6000/linux.h, config/rs6000/linux64.h: Likewise. Split out get_sigcontext function. Use ARG_POINTER_REGNUM for 32-bit temp reg too. From-SVN: r87167
2004-09-08 00:17:19 +00:00
/* Register 7 is rsp */
fs->regs.cfa_reg = 7;
fs->regs.cfa_offset = new_cfa - (long) context->cfa;
tm.texi (MD_UNWIND_SUPPORT): Document. * doc/tm.texi (MD_UNWIND_SUPPORT): Document. (MD_FALLBACK_FRAME_STATE_FOR): Update. * unwind-dw2.c (MD_UNWIND_SUPPORT): #include if defined. (uw_frame_state_for): Adjust MD_FALLBACK_FRAME_STATE_FOR invocation. (MD_FROB_UPDATE_CONTEXT): Remove default. (uw_update_context_1): Instead #ifdef invocation. * config/ia64/unwind-ia64.c (MD_UNWIND_SUPPORT): #include if defined. (uw_frame_state_for): Adjust MD_FALLBACK_FRAME_STATE_FOR invocation. * config/alpha/gnu.h (MD_FALLBACK_FRAME_STATE_FOR): Don't undef. (MD_UNWIND_SUPPORT): Undefine this instead. * config/i386/gnu.h: Likewise. * config/alpha/linux-unwind.h: New file, macro converted to function, extracted from.. * config/alpha/linux.h (MD_FALLBACK_FRAME_STATE_FOR): ..this. (MD_UNWIND_SUPPORT): Define. * config/alpha/vms-unwind.h, config/alpha/vms.h: Likewise. * config/i386/linux-unwind.h, config/i386/linux.h, config/i386/linux64.h: Likewise. * config/ia64/linux-unwind.h, config/ia64/linux.h: Likewise. MD_HANDLE_UNWABI too. * config/mips/linux-unwind.h, config/mips/linux.h: Likewise. * config/pa/linux-unwind.h, config/pa/pa32-linux.h: Likewise. * config/rs6000/darwin-unwind.h, config/rs6000/darwin.h: Likewise. * config/s390/linux-unwind.h, config/s390/linux.h: Likewise. * config/sparc/linux-unwind.h, config/sparc/linux.h, config/sparc/linux64.h: Likewise. * config/sh/linux-unwind.h, config/sh/linux.h: Likewise, but merge SH_FALLBACK_FRAME_FLOAT_STATE into sh_fallback_frame_state. * config/rs6000/linux-unwind.h, config/rs6000/linux.h, config/rs6000/linux64.h: Likewise. Split out get_sigcontext function. Use ARG_POINTER_REGNUM for 32-bit temp reg too. From-SVN: r87167
2004-09-08 00:17:19 +00:00
/* The SVR4 register numbering macros aren't usable in libgcc. */
libgcc: Decrease size of _Unwind_FrameState and even more size of cleared area in uw_frame_state_for The following patch implements something that has Florian found as low hanging fruit in our unwinder and has been discussed in the https://gcc.gnu.org/wiki/cauldron2022#cauldron2022talks.inprocess_unwinding_bof talk. _Unwind_FrameState type seems to be (unlike the pre-GCC 3 frame_state which has been part of ABI) private to unwind-dw2.c + unwind.inc it includes, it is always defined on the stack of some entrypoints, initialized by static uw_frame_state_for and the address of it is also passed to other static functions or the static inlines handling machine dependent unwinding, but it isn't fortunately passed to any callbacks or public functions, so I think we can safely change it any time we want. Florian mentioned that the structure is large even on x86_64, 384 bytes there, starts with 328 bytes long element with frame_state_reg_info type which then starts with an array with __LIBGCC_DWARF_FRAME_REGISTERS__ + 1 elements, each of them is 16 bytes long, on x86_64 __LIBGCC_DWARF_FRAME_REGISTERS__ is just 17 but even that is big, on say riscv __LIBGCC_DWARF_FRAME_REGISTERS__ is I think 128, on powerpc 111, on sh 153 etc. And, we memset to zero the whole fs variable with the _Unwind_FrameState type at the start of the unwinding. The reason why each element is 16 byte (on 64-bit arches) is that it contains some pointer or pointer sized integer and then an enum (with just 7 different enumerators) + padding. The following patch decreases it by moving the enum into a separate array and using just one byte for each register in that second array. We could compress it even more, say 4 bits per register, but I don't want to uglify the code for it too much and make the accesses slower. Furthermore, the clearing of the object can clear only thos how array and members after it, because REG_UNSAVED enumerator (0) doesn't actually need any pointer or pointer sized integer, it is just the other kinds that need to have there something. By doing this, on x86_64 the above numbers change to _Unwind_FrameState type being now 264 bytes long, frame_state_reg_info 208 bytes and we don't clear the first 144 bytes of the object, so the memset is 120 bytes, so ~ 31% of the old clearing size. On riscv 64-bit assuming it has same structure layout rules for the few types used there that would be ~ 2160 bytes of _Unwind_FrameState type before and ~ 1264 bytes after, with the memset previously ~ 2160 bytes and after ~ 232 bytes after. We've also talked about possibly adding a number of initially initialized regs and initializing the rest lazily, but at least for x86_64 with 18 elements in the array that doesn't seem to be worth it anymore, especially because return address column is 16 there and that is usually the first thing to be touched. It might theory help with lots of registers if they are usually untouched, but would uglify and complicate any stores to how by having to check there for the not initialized yet cases and lazy initialization, and similarly for all reads of how to do there if below last initialized one, use how, otherwise imply REG_UNSAVED. The disadvantage of the patch is that touching reg[x].loc and how[x] now means 2 cachelines rather than one as before, and I admit beyond bootstrap/regtest I haven't benchmarked it in any way. 2022-10-06 Jakub Jelinek <jakub@redhat.com> * unwind-dw2.h (REG_UNSAVED, REG_SAVED_OFFSET, REG_SAVED_REG, REG_SAVED_EXP, REG_SAVED_VAL_OFFSET, REG_SAVED_VAL_EXP, REG_UNDEFINED): New anonymous enum, moved from inside of struct frame_state_reg_info. (struct frame_state_reg_info): Remove reg[].how element and the anonymous enum there. Add how element. * unwind-dw2.c: Include stddef.h. (uw_frame_state_for): Don't clear first offsetof (_Unwind_FrameState, regs.how[0]) bytes of *fs. (execute_cfa_program, __frame_state_for, uw_update_context_1, uw_update_context): Use fs->regs.how[X] instead of fs->regs.reg[X].how or fs.regs.how[X] instead of fs.regs.reg[X].how. * config/sh/linux-unwind.h (sh_fallback_frame_state): Likewise. * config/bfin/linux-unwind.h (bfin_fallback_frame_state): Likewise. * config/pa/linux-unwind.h (pa32_fallback_frame_state): Likewise. * config/pa/hpux-unwind.h (UPDATE_FS_FOR_SAR, UPDATE_FS_FOR_GR, UPDATE_FS_FOR_FR, UPDATE_FS_FOR_PC, pa_fallback_frame_state): Likewise. * config/alpha/vms-unwind.h (alpha_vms_fallback_frame_state): Likewise. * config/alpha/linux-unwind.h (alpha_fallback_frame_state): Likewise. * config/arc/linux-unwind.h (arc_fallback_frame_state, arc_frob_update_context): Likewise. * config/riscv/linux-unwind.h (riscv_fallback_frame_state): Likewise. * config/nios2/linux-unwind.h (NIOS2_REG): Likewise. * config/nds32/linux-unwind.h (NDS32_PUT_FS_REG): Likewise. * config/s390/tpf-unwind.h (s390_fallback_frame_state): Likewise. * config/s390/linux-unwind.h (s390_fallback_frame_state): Likewise. * config/sparc/sol2-unwind.h (sparc64_frob_update_context, MD_FALLBACK_FRAME_STATE_FOR): Likewise. * config/sparc/linux-unwind.h (sparc64_fallback_frame_state, sparc64_frob_update_context, sparc_fallback_frame_state): Likewise. * config/i386/sol2-unwind.h (x86_64_fallback_frame_state, x86_fallback_frame_state): Likewise. * config/i386/w32-unwind.h (i386_w32_fallback_frame_state): Likewise. * config/i386/linux-unwind.h (x86_64_fallback_frame_state, x86_fallback_frame_state): Likewise. * config/i386/freebsd-unwind.h (x86_64_freebsd_fallback_frame_state): Likewise. * config/i386/dragonfly-unwind.h (x86_64_dragonfly_fallback_frame_state): Likewise. * config/i386/gnu-unwind.h (x86_gnu_fallback_frame_state): Likewise. * config/csky/linux-unwind.h (csky_fallback_frame_state): Likewise. * config/aarch64/linux-unwind.h (aarch64_fallback_frame_state): Likewise. * config/aarch64/freebsd-unwind.h (aarch64_freebsd_fallback_frame_state): Likewise. * config/aarch64/aarch64-unwind.h (aarch64_frob_update_context): Likewise. * config/or1k/linux-unwind.h (or1k_fallback_frame_state): Likewise. * config/mips/linux-unwind.h (mips_fallback_frame_state): Likewise. * config/loongarch/linux-unwind.h (loongarch_fallback_frame_state): Likewise. * config/m68k/linux-unwind.h (m68k_fallback_frame_state): Likewise. * config/xtensa/linux-unwind.h (xtensa_fallback_frame_state): Likewise. * config/rs6000/darwin-fallback.c (set_offset): Likewise. * config/rs6000/aix-unwind.h (MD_FROB_UPDATE_CONTEXT): Likewise. * config/rs6000/linux-unwind.h (ppc_fallback_frame_state): Likewise. * config/rs6000/freebsd-unwind.h (frob_update_context): Likewise.
2022-10-06 11:04:52 +02:00
fs->regs.how[0] = REG_SAVED_OFFSET;
tm.texi (MD_UNWIND_SUPPORT): Document. * doc/tm.texi (MD_UNWIND_SUPPORT): Document. (MD_FALLBACK_FRAME_STATE_FOR): Update. * unwind-dw2.c (MD_UNWIND_SUPPORT): #include if defined. (uw_frame_state_for): Adjust MD_FALLBACK_FRAME_STATE_FOR invocation. (MD_FROB_UPDATE_CONTEXT): Remove default. (uw_update_context_1): Instead #ifdef invocation. * config/ia64/unwind-ia64.c (MD_UNWIND_SUPPORT): #include if defined. (uw_frame_state_for): Adjust MD_FALLBACK_FRAME_STATE_FOR invocation. * config/alpha/gnu.h (MD_FALLBACK_FRAME_STATE_FOR): Don't undef. (MD_UNWIND_SUPPORT): Undefine this instead. * config/i386/gnu.h: Likewise. * config/alpha/linux-unwind.h: New file, macro converted to function, extracted from.. * config/alpha/linux.h (MD_FALLBACK_FRAME_STATE_FOR): ..this. (MD_UNWIND_SUPPORT): Define. * config/alpha/vms-unwind.h, config/alpha/vms.h: Likewise. * config/i386/linux-unwind.h, config/i386/linux.h, config/i386/linux64.h: Likewise. * config/ia64/linux-unwind.h, config/ia64/linux.h: Likewise. MD_HANDLE_UNWABI too. * config/mips/linux-unwind.h, config/mips/linux.h: Likewise. * config/pa/linux-unwind.h, config/pa/pa32-linux.h: Likewise. * config/rs6000/darwin-unwind.h, config/rs6000/darwin.h: Likewise. * config/s390/linux-unwind.h, config/s390/linux.h: Likewise. * config/sparc/linux-unwind.h, config/sparc/linux.h, config/sparc/linux64.h: Likewise. * config/sh/linux-unwind.h, config/sh/linux.h: Likewise, but merge SH_FALLBACK_FRAME_FLOAT_STATE into sh_fallback_frame_state. * config/rs6000/linux-unwind.h, config/rs6000/linux.h, config/rs6000/linux64.h: Likewise. Split out get_sigcontext function. Use ARG_POINTER_REGNUM for 32-bit temp reg too. From-SVN: r87167
2004-09-08 00:17:19 +00:00
fs->regs.reg[0].loc.offset = (long)&sc->rax - new_cfa;
libgcc: Decrease size of _Unwind_FrameState and even more size of cleared area in uw_frame_state_for The following patch implements something that has Florian found as low hanging fruit in our unwinder and has been discussed in the https://gcc.gnu.org/wiki/cauldron2022#cauldron2022talks.inprocess_unwinding_bof talk. _Unwind_FrameState type seems to be (unlike the pre-GCC 3 frame_state which has been part of ABI) private to unwind-dw2.c + unwind.inc it includes, it is always defined on the stack of some entrypoints, initialized by static uw_frame_state_for and the address of it is also passed to other static functions or the static inlines handling machine dependent unwinding, but it isn't fortunately passed to any callbacks or public functions, so I think we can safely change it any time we want. Florian mentioned that the structure is large even on x86_64, 384 bytes there, starts with 328 bytes long element with frame_state_reg_info type which then starts with an array with __LIBGCC_DWARF_FRAME_REGISTERS__ + 1 elements, each of them is 16 bytes long, on x86_64 __LIBGCC_DWARF_FRAME_REGISTERS__ is just 17 but even that is big, on say riscv __LIBGCC_DWARF_FRAME_REGISTERS__ is I think 128, on powerpc 111, on sh 153 etc. And, we memset to zero the whole fs variable with the _Unwind_FrameState type at the start of the unwinding. The reason why each element is 16 byte (on 64-bit arches) is that it contains some pointer or pointer sized integer and then an enum (with just 7 different enumerators) + padding. The following patch decreases it by moving the enum into a separate array and using just one byte for each register in that second array. We could compress it even more, say 4 bits per register, but I don't want to uglify the code for it too much and make the accesses slower. Furthermore, the clearing of the object can clear only thos how array and members after it, because REG_UNSAVED enumerator (0) doesn't actually need any pointer or pointer sized integer, it is just the other kinds that need to have there something. By doing this, on x86_64 the above numbers change to _Unwind_FrameState type being now 264 bytes long, frame_state_reg_info 208 bytes and we don't clear the first 144 bytes of the object, so the memset is 120 bytes, so ~ 31% of the old clearing size. On riscv 64-bit assuming it has same structure layout rules for the few types used there that would be ~ 2160 bytes of _Unwind_FrameState type before and ~ 1264 bytes after, with the memset previously ~ 2160 bytes and after ~ 232 bytes after. We've also talked about possibly adding a number of initially initialized regs and initializing the rest lazily, but at least for x86_64 with 18 elements in the array that doesn't seem to be worth it anymore, especially because return address column is 16 there and that is usually the first thing to be touched. It might theory help with lots of registers if they are usually untouched, but would uglify and complicate any stores to how by having to check there for the not initialized yet cases and lazy initialization, and similarly for all reads of how to do there if below last initialized one, use how, otherwise imply REG_UNSAVED. The disadvantage of the patch is that touching reg[x].loc and how[x] now means 2 cachelines rather than one as before, and I admit beyond bootstrap/regtest I haven't benchmarked it in any way. 2022-10-06 Jakub Jelinek <jakub@redhat.com> * unwind-dw2.h (REG_UNSAVED, REG_SAVED_OFFSET, REG_SAVED_REG, REG_SAVED_EXP, REG_SAVED_VAL_OFFSET, REG_SAVED_VAL_EXP, REG_UNDEFINED): New anonymous enum, moved from inside of struct frame_state_reg_info. (struct frame_state_reg_info): Remove reg[].how element and the anonymous enum there. Add how element. * unwind-dw2.c: Include stddef.h. (uw_frame_state_for): Don't clear first offsetof (_Unwind_FrameState, regs.how[0]) bytes of *fs. (execute_cfa_program, __frame_state_for, uw_update_context_1, uw_update_context): Use fs->regs.how[X] instead of fs->regs.reg[X].how or fs.regs.how[X] instead of fs.regs.reg[X].how. * config/sh/linux-unwind.h (sh_fallback_frame_state): Likewise. * config/bfin/linux-unwind.h (bfin_fallback_frame_state): Likewise. * config/pa/linux-unwind.h (pa32_fallback_frame_state): Likewise. * config/pa/hpux-unwind.h (UPDATE_FS_FOR_SAR, UPDATE_FS_FOR_GR, UPDATE_FS_FOR_FR, UPDATE_FS_FOR_PC, pa_fallback_frame_state): Likewise. * config/alpha/vms-unwind.h (alpha_vms_fallback_frame_state): Likewise. * config/alpha/linux-unwind.h (alpha_fallback_frame_state): Likewise. * config/arc/linux-unwind.h (arc_fallback_frame_state, arc_frob_update_context): Likewise. * config/riscv/linux-unwind.h (riscv_fallback_frame_state): Likewise. * config/nios2/linux-unwind.h (NIOS2_REG): Likewise. * config/nds32/linux-unwind.h (NDS32_PUT_FS_REG): Likewise. * config/s390/tpf-unwind.h (s390_fallback_frame_state): Likewise. * config/s390/linux-unwind.h (s390_fallback_frame_state): Likewise. * config/sparc/sol2-unwind.h (sparc64_frob_update_context, MD_FALLBACK_FRAME_STATE_FOR): Likewise. * config/sparc/linux-unwind.h (sparc64_fallback_frame_state, sparc64_frob_update_context, sparc_fallback_frame_state): Likewise. * config/i386/sol2-unwind.h (x86_64_fallback_frame_state, x86_fallback_frame_state): Likewise. * config/i386/w32-unwind.h (i386_w32_fallback_frame_state): Likewise. * config/i386/linux-unwind.h (x86_64_fallback_frame_state, x86_fallback_frame_state): Likewise. * config/i386/freebsd-unwind.h (x86_64_freebsd_fallback_frame_state): Likewise. * config/i386/dragonfly-unwind.h (x86_64_dragonfly_fallback_frame_state): Likewise. * config/i386/gnu-unwind.h (x86_gnu_fallback_frame_state): Likewise. * config/csky/linux-unwind.h (csky_fallback_frame_state): Likewise. * config/aarch64/linux-unwind.h (aarch64_fallback_frame_state): Likewise. * config/aarch64/freebsd-unwind.h (aarch64_freebsd_fallback_frame_state): Likewise. * config/aarch64/aarch64-unwind.h (aarch64_frob_update_context): Likewise. * config/or1k/linux-unwind.h (or1k_fallback_frame_state): Likewise. * config/mips/linux-unwind.h (mips_fallback_frame_state): Likewise. * config/loongarch/linux-unwind.h (loongarch_fallback_frame_state): Likewise. * config/m68k/linux-unwind.h (m68k_fallback_frame_state): Likewise. * config/xtensa/linux-unwind.h (xtensa_fallback_frame_state): Likewise. * config/rs6000/darwin-fallback.c (set_offset): Likewise. * config/rs6000/aix-unwind.h (MD_FROB_UPDATE_CONTEXT): Likewise. * config/rs6000/linux-unwind.h (ppc_fallback_frame_state): Likewise. * config/rs6000/freebsd-unwind.h (frob_update_context): Likewise.
2022-10-06 11:04:52 +02:00
fs->regs.how[1] = REG_SAVED_OFFSET;
tm.texi (MD_UNWIND_SUPPORT): Document. * doc/tm.texi (MD_UNWIND_SUPPORT): Document. (MD_FALLBACK_FRAME_STATE_FOR): Update. * unwind-dw2.c (MD_UNWIND_SUPPORT): #include if defined. (uw_frame_state_for): Adjust MD_FALLBACK_FRAME_STATE_FOR invocation. (MD_FROB_UPDATE_CONTEXT): Remove default. (uw_update_context_1): Instead #ifdef invocation. * config/ia64/unwind-ia64.c (MD_UNWIND_SUPPORT): #include if defined. (uw_frame_state_for): Adjust MD_FALLBACK_FRAME_STATE_FOR invocation. * config/alpha/gnu.h (MD_FALLBACK_FRAME_STATE_FOR): Don't undef. (MD_UNWIND_SUPPORT): Undefine this instead. * config/i386/gnu.h: Likewise. * config/alpha/linux-unwind.h: New file, macro converted to function, extracted from.. * config/alpha/linux.h (MD_FALLBACK_FRAME_STATE_FOR): ..this. (MD_UNWIND_SUPPORT): Define. * config/alpha/vms-unwind.h, config/alpha/vms.h: Likewise. * config/i386/linux-unwind.h, config/i386/linux.h, config/i386/linux64.h: Likewise. * config/ia64/linux-unwind.h, config/ia64/linux.h: Likewise. MD_HANDLE_UNWABI too. * config/mips/linux-unwind.h, config/mips/linux.h: Likewise. * config/pa/linux-unwind.h, config/pa/pa32-linux.h: Likewise. * config/rs6000/darwin-unwind.h, config/rs6000/darwin.h: Likewise. * config/s390/linux-unwind.h, config/s390/linux.h: Likewise. * config/sparc/linux-unwind.h, config/sparc/linux.h, config/sparc/linux64.h: Likewise. * config/sh/linux-unwind.h, config/sh/linux.h: Likewise, but merge SH_FALLBACK_FRAME_FLOAT_STATE into sh_fallback_frame_state. * config/rs6000/linux-unwind.h, config/rs6000/linux.h, config/rs6000/linux64.h: Likewise. Split out get_sigcontext function. Use ARG_POINTER_REGNUM for 32-bit temp reg too. From-SVN: r87167
2004-09-08 00:17:19 +00:00
fs->regs.reg[1].loc.offset = (long)&sc->rdx - new_cfa;
libgcc: Decrease size of _Unwind_FrameState and even more size of cleared area in uw_frame_state_for The following patch implements something that has Florian found as low hanging fruit in our unwinder and has been discussed in the https://gcc.gnu.org/wiki/cauldron2022#cauldron2022talks.inprocess_unwinding_bof talk. _Unwind_FrameState type seems to be (unlike the pre-GCC 3 frame_state which has been part of ABI) private to unwind-dw2.c + unwind.inc it includes, it is always defined on the stack of some entrypoints, initialized by static uw_frame_state_for and the address of it is also passed to other static functions or the static inlines handling machine dependent unwinding, but it isn't fortunately passed to any callbacks or public functions, so I think we can safely change it any time we want. Florian mentioned that the structure is large even on x86_64, 384 bytes there, starts with 328 bytes long element with frame_state_reg_info type which then starts with an array with __LIBGCC_DWARF_FRAME_REGISTERS__ + 1 elements, each of them is 16 bytes long, on x86_64 __LIBGCC_DWARF_FRAME_REGISTERS__ is just 17 but even that is big, on say riscv __LIBGCC_DWARF_FRAME_REGISTERS__ is I think 128, on powerpc 111, on sh 153 etc. And, we memset to zero the whole fs variable with the _Unwind_FrameState type at the start of the unwinding. The reason why each element is 16 byte (on 64-bit arches) is that it contains some pointer or pointer sized integer and then an enum (with just 7 different enumerators) + padding. The following patch decreases it by moving the enum into a separate array and using just one byte for each register in that second array. We could compress it even more, say 4 bits per register, but I don't want to uglify the code for it too much and make the accesses slower. Furthermore, the clearing of the object can clear only thos how array and members after it, because REG_UNSAVED enumerator (0) doesn't actually need any pointer or pointer sized integer, it is just the other kinds that need to have there something. By doing this, on x86_64 the above numbers change to _Unwind_FrameState type being now 264 bytes long, frame_state_reg_info 208 bytes and we don't clear the first 144 bytes of the object, so the memset is 120 bytes, so ~ 31% of the old clearing size. On riscv 64-bit assuming it has same structure layout rules for the few types used there that would be ~ 2160 bytes of _Unwind_FrameState type before and ~ 1264 bytes after, with the memset previously ~ 2160 bytes and after ~ 232 bytes after. We've also talked about possibly adding a number of initially initialized regs and initializing the rest lazily, but at least for x86_64 with 18 elements in the array that doesn't seem to be worth it anymore, especially because return address column is 16 there and that is usually the first thing to be touched. It might theory help with lots of registers if they are usually untouched, but would uglify and complicate any stores to how by having to check there for the not initialized yet cases and lazy initialization, and similarly for all reads of how to do there if below last initialized one, use how, otherwise imply REG_UNSAVED. The disadvantage of the patch is that touching reg[x].loc and how[x] now means 2 cachelines rather than one as before, and I admit beyond bootstrap/regtest I haven't benchmarked it in any way. 2022-10-06 Jakub Jelinek <jakub@redhat.com> * unwind-dw2.h (REG_UNSAVED, REG_SAVED_OFFSET, REG_SAVED_REG, REG_SAVED_EXP, REG_SAVED_VAL_OFFSET, REG_SAVED_VAL_EXP, REG_UNDEFINED): New anonymous enum, moved from inside of struct frame_state_reg_info. (struct frame_state_reg_info): Remove reg[].how element and the anonymous enum there. Add how element. * unwind-dw2.c: Include stddef.h. (uw_frame_state_for): Don't clear first offsetof (_Unwind_FrameState, regs.how[0]) bytes of *fs. (execute_cfa_program, __frame_state_for, uw_update_context_1, uw_update_context): Use fs->regs.how[X] instead of fs->regs.reg[X].how or fs.regs.how[X] instead of fs.regs.reg[X].how. * config/sh/linux-unwind.h (sh_fallback_frame_state): Likewise. * config/bfin/linux-unwind.h (bfin_fallback_frame_state): Likewise. * config/pa/linux-unwind.h (pa32_fallback_frame_state): Likewise. * config/pa/hpux-unwind.h (UPDATE_FS_FOR_SAR, UPDATE_FS_FOR_GR, UPDATE_FS_FOR_FR, UPDATE_FS_FOR_PC, pa_fallback_frame_state): Likewise. * config/alpha/vms-unwind.h (alpha_vms_fallback_frame_state): Likewise. * config/alpha/linux-unwind.h (alpha_fallback_frame_state): Likewise. * config/arc/linux-unwind.h (arc_fallback_frame_state, arc_frob_update_context): Likewise. * config/riscv/linux-unwind.h (riscv_fallback_frame_state): Likewise. * config/nios2/linux-unwind.h (NIOS2_REG): Likewise. * config/nds32/linux-unwind.h (NDS32_PUT_FS_REG): Likewise. * config/s390/tpf-unwind.h (s390_fallback_frame_state): Likewise. * config/s390/linux-unwind.h (s390_fallback_frame_state): Likewise. * config/sparc/sol2-unwind.h (sparc64_frob_update_context, MD_FALLBACK_FRAME_STATE_FOR): Likewise. * config/sparc/linux-unwind.h (sparc64_fallback_frame_state, sparc64_frob_update_context, sparc_fallback_frame_state): Likewise. * config/i386/sol2-unwind.h (x86_64_fallback_frame_state, x86_fallback_frame_state): Likewise. * config/i386/w32-unwind.h (i386_w32_fallback_frame_state): Likewise. * config/i386/linux-unwind.h (x86_64_fallback_frame_state, x86_fallback_frame_state): Likewise. * config/i386/freebsd-unwind.h (x86_64_freebsd_fallback_frame_state): Likewise. * config/i386/dragonfly-unwind.h (x86_64_dragonfly_fallback_frame_state): Likewise. * config/i386/gnu-unwind.h (x86_gnu_fallback_frame_state): Likewise. * config/csky/linux-unwind.h (csky_fallback_frame_state): Likewise. * config/aarch64/linux-unwind.h (aarch64_fallback_frame_state): Likewise. * config/aarch64/freebsd-unwind.h (aarch64_freebsd_fallback_frame_state): Likewise. * config/aarch64/aarch64-unwind.h (aarch64_frob_update_context): Likewise. * config/or1k/linux-unwind.h (or1k_fallback_frame_state): Likewise. * config/mips/linux-unwind.h (mips_fallback_frame_state): Likewise. * config/loongarch/linux-unwind.h (loongarch_fallback_frame_state): Likewise. * config/m68k/linux-unwind.h (m68k_fallback_frame_state): Likewise. * config/xtensa/linux-unwind.h (xtensa_fallback_frame_state): Likewise. * config/rs6000/darwin-fallback.c (set_offset): Likewise. * config/rs6000/aix-unwind.h (MD_FROB_UPDATE_CONTEXT): Likewise. * config/rs6000/linux-unwind.h (ppc_fallback_frame_state): Likewise. * config/rs6000/freebsd-unwind.h (frob_update_context): Likewise.
2022-10-06 11:04:52 +02:00
fs->regs.how[2] = REG_SAVED_OFFSET;
tm.texi (MD_UNWIND_SUPPORT): Document. * doc/tm.texi (MD_UNWIND_SUPPORT): Document. (MD_FALLBACK_FRAME_STATE_FOR): Update. * unwind-dw2.c (MD_UNWIND_SUPPORT): #include if defined. (uw_frame_state_for): Adjust MD_FALLBACK_FRAME_STATE_FOR invocation. (MD_FROB_UPDATE_CONTEXT): Remove default. (uw_update_context_1): Instead #ifdef invocation. * config/ia64/unwind-ia64.c (MD_UNWIND_SUPPORT): #include if defined. (uw_frame_state_for): Adjust MD_FALLBACK_FRAME_STATE_FOR invocation. * config/alpha/gnu.h (MD_FALLBACK_FRAME_STATE_FOR): Don't undef. (MD_UNWIND_SUPPORT): Undefine this instead. * config/i386/gnu.h: Likewise. * config/alpha/linux-unwind.h: New file, macro converted to function, extracted from.. * config/alpha/linux.h (MD_FALLBACK_FRAME_STATE_FOR): ..this. (MD_UNWIND_SUPPORT): Define. * config/alpha/vms-unwind.h, config/alpha/vms.h: Likewise. * config/i386/linux-unwind.h, config/i386/linux.h, config/i386/linux64.h: Likewise. * config/ia64/linux-unwind.h, config/ia64/linux.h: Likewise. MD_HANDLE_UNWABI too. * config/mips/linux-unwind.h, config/mips/linux.h: Likewise. * config/pa/linux-unwind.h, config/pa/pa32-linux.h: Likewise. * config/rs6000/darwin-unwind.h, config/rs6000/darwin.h: Likewise. * config/s390/linux-unwind.h, config/s390/linux.h: Likewise. * config/sparc/linux-unwind.h, config/sparc/linux.h, config/sparc/linux64.h: Likewise. * config/sh/linux-unwind.h, config/sh/linux.h: Likewise, but merge SH_FALLBACK_FRAME_FLOAT_STATE into sh_fallback_frame_state. * config/rs6000/linux-unwind.h, config/rs6000/linux.h, config/rs6000/linux64.h: Likewise. Split out get_sigcontext function. Use ARG_POINTER_REGNUM for 32-bit temp reg too. From-SVN: r87167
2004-09-08 00:17:19 +00:00
fs->regs.reg[2].loc.offset = (long)&sc->rcx - new_cfa;
libgcc: Decrease size of _Unwind_FrameState and even more size of cleared area in uw_frame_state_for The following patch implements something that has Florian found as low hanging fruit in our unwinder and has been discussed in the https://gcc.gnu.org/wiki/cauldron2022#cauldron2022talks.inprocess_unwinding_bof talk. _Unwind_FrameState type seems to be (unlike the pre-GCC 3 frame_state which has been part of ABI) private to unwind-dw2.c + unwind.inc it includes, it is always defined on the stack of some entrypoints, initialized by static uw_frame_state_for and the address of it is also passed to other static functions or the static inlines handling machine dependent unwinding, but it isn't fortunately passed to any callbacks or public functions, so I think we can safely change it any time we want. Florian mentioned that the structure is large even on x86_64, 384 bytes there, starts with 328 bytes long element with frame_state_reg_info type which then starts with an array with __LIBGCC_DWARF_FRAME_REGISTERS__ + 1 elements, each of them is 16 bytes long, on x86_64 __LIBGCC_DWARF_FRAME_REGISTERS__ is just 17 but even that is big, on say riscv __LIBGCC_DWARF_FRAME_REGISTERS__ is I think 128, on powerpc 111, on sh 153 etc. And, we memset to zero the whole fs variable with the _Unwind_FrameState type at the start of the unwinding. The reason why each element is 16 byte (on 64-bit arches) is that it contains some pointer or pointer sized integer and then an enum (with just 7 different enumerators) + padding. The following patch decreases it by moving the enum into a separate array and using just one byte for each register in that second array. We could compress it even more, say 4 bits per register, but I don't want to uglify the code for it too much and make the accesses slower. Furthermore, the clearing of the object can clear only thos how array and members after it, because REG_UNSAVED enumerator (0) doesn't actually need any pointer or pointer sized integer, it is just the other kinds that need to have there something. By doing this, on x86_64 the above numbers change to _Unwind_FrameState type being now 264 bytes long, frame_state_reg_info 208 bytes and we don't clear the first 144 bytes of the object, so the memset is 120 bytes, so ~ 31% of the old clearing size. On riscv 64-bit assuming it has same structure layout rules for the few types used there that would be ~ 2160 bytes of _Unwind_FrameState type before and ~ 1264 bytes after, with the memset previously ~ 2160 bytes and after ~ 232 bytes after. We've also talked about possibly adding a number of initially initialized regs and initializing the rest lazily, but at least for x86_64 with 18 elements in the array that doesn't seem to be worth it anymore, especially because return address column is 16 there and that is usually the first thing to be touched. It might theory help with lots of registers if they are usually untouched, but would uglify and complicate any stores to how by having to check there for the not initialized yet cases and lazy initialization, and similarly for all reads of how to do there if below last initialized one, use how, otherwise imply REG_UNSAVED. The disadvantage of the patch is that touching reg[x].loc and how[x] now means 2 cachelines rather than one as before, and I admit beyond bootstrap/regtest I haven't benchmarked it in any way. 2022-10-06 Jakub Jelinek <jakub@redhat.com> * unwind-dw2.h (REG_UNSAVED, REG_SAVED_OFFSET, REG_SAVED_REG, REG_SAVED_EXP, REG_SAVED_VAL_OFFSET, REG_SAVED_VAL_EXP, REG_UNDEFINED): New anonymous enum, moved from inside of struct frame_state_reg_info. (struct frame_state_reg_info): Remove reg[].how element and the anonymous enum there. Add how element. * unwind-dw2.c: Include stddef.h. (uw_frame_state_for): Don't clear first offsetof (_Unwind_FrameState, regs.how[0]) bytes of *fs. (execute_cfa_program, __frame_state_for, uw_update_context_1, uw_update_context): Use fs->regs.how[X] instead of fs->regs.reg[X].how or fs.regs.how[X] instead of fs.regs.reg[X].how. * config/sh/linux-unwind.h (sh_fallback_frame_state): Likewise. * config/bfin/linux-unwind.h (bfin_fallback_frame_state): Likewise. * config/pa/linux-unwind.h (pa32_fallback_frame_state): Likewise. * config/pa/hpux-unwind.h (UPDATE_FS_FOR_SAR, UPDATE_FS_FOR_GR, UPDATE_FS_FOR_FR, UPDATE_FS_FOR_PC, pa_fallback_frame_state): Likewise. * config/alpha/vms-unwind.h (alpha_vms_fallback_frame_state): Likewise. * config/alpha/linux-unwind.h (alpha_fallback_frame_state): Likewise. * config/arc/linux-unwind.h (arc_fallback_frame_state, arc_frob_update_context): Likewise. * config/riscv/linux-unwind.h (riscv_fallback_frame_state): Likewise. * config/nios2/linux-unwind.h (NIOS2_REG): Likewise. * config/nds32/linux-unwind.h (NDS32_PUT_FS_REG): Likewise. * config/s390/tpf-unwind.h (s390_fallback_frame_state): Likewise. * config/s390/linux-unwind.h (s390_fallback_frame_state): Likewise. * config/sparc/sol2-unwind.h (sparc64_frob_update_context, MD_FALLBACK_FRAME_STATE_FOR): Likewise. * config/sparc/linux-unwind.h (sparc64_fallback_frame_state, sparc64_frob_update_context, sparc_fallback_frame_state): Likewise. * config/i386/sol2-unwind.h (x86_64_fallback_frame_state, x86_fallback_frame_state): Likewise. * config/i386/w32-unwind.h (i386_w32_fallback_frame_state): Likewise. * config/i386/linux-unwind.h (x86_64_fallback_frame_state, x86_fallback_frame_state): Likewise. * config/i386/freebsd-unwind.h (x86_64_freebsd_fallback_frame_state): Likewise. * config/i386/dragonfly-unwind.h (x86_64_dragonfly_fallback_frame_state): Likewise. * config/i386/gnu-unwind.h (x86_gnu_fallback_frame_state): Likewise. * config/csky/linux-unwind.h (csky_fallback_frame_state): Likewise. * config/aarch64/linux-unwind.h (aarch64_fallback_frame_state): Likewise. * config/aarch64/freebsd-unwind.h (aarch64_freebsd_fallback_frame_state): Likewise. * config/aarch64/aarch64-unwind.h (aarch64_frob_update_context): Likewise. * config/or1k/linux-unwind.h (or1k_fallback_frame_state): Likewise. * config/mips/linux-unwind.h (mips_fallback_frame_state): Likewise. * config/loongarch/linux-unwind.h (loongarch_fallback_frame_state): Likewise. * config/m68k/linux-unwind.h (m68k_fallback_frame_state): Likewise. * config/xtensa/linux-unwind.h (xtensa_fallback_frame_state): Likewise. * config/rs6000/darwin-fallback.c (set_offset): Likewise. * config/rs6000/aix-unwind.h (MD_FROB_UPDATE_CONTEXT): Likewise. * config/rs6000/linux-unwind.h (ppc_fallback_frame_state): Likewise. * config/rs6000/freebsd-unwind.h (frob_update_context): Likewise.
2022-10-06 11:04:52 +02:00
fs->regs.how[3] = REG_SAVED_OFFSET;
tm.texi (MD_UNWIND_SUPPORT): Document. * doc/tm.texi (MD_UNWIND_SUPPORT): Document. (MD_FALLBACK_FRAME_STATE_FOR): Update. * unwind-dw2.c (MD_UNWIND_SUPPORT): #include if defined. (uw_frame_state_for): Adjust MD_FALLBACK_FRAME_STATE_FOR invocation. (MD_FROB_UPDATE_CONTEXT): Remove default. (uw_update_context_1): Instead #ifdef invocation. * config/ia64/unwind-ia64.c (MD_UNWIND_SUPPORT): #include if defined. (uw_frame_state_for): Adjust MD_FALLBACK_FRAME_STATE_FOR invocation. * config/alpha/gnu.h (MD_FALLBACK_FRAME_STATE_FOR): Don't undef. (MD_UNWIND_SUPPORT): Undefine this instead. * config/i386/gnu.h: Likewise. * config/alpha/linux-unwind.h: New file, macro converted to function, extracted from.. * config/alpha/linux.h (MD_FALLBACK_FRAME_STATE_FOR): ..this. (MD_UNWIND_SUPPORT): Define. * config/alpha/vms-unwind.h, config/alpha/vms.h: Likewise. * config/i386/linux-unwind.h, config/i386/linux.h, config/i386/linux64.h: Likewise. * config/ia64/linux-unwind.h, config/ia64/linux.h: Likewise. MD_HANDLE_UNWABI too. * config/mips/linux-unwind.h, config/mips/linux.h: Likewise. * config/pa/linux-unwind.h, config/pa/pa32-linux.h: Likewise. * config/rs6000/darwin-unwind.h, config/rs6000/darwin.h: Likewise. * config/s390/linux-unwind.h, config/s390/linux.h: Likewise. * config/sparc/linux-unwind.h, config/sparc/linux.h, config/sparc/linux64.h: Likewise. * config/sh/linux-unwind.h, config/sh/linux.h: Likewise, but merge SH_FALLBACK_FRAME_FLOAT_STATE into sh_fallback_frame_state. * config/rs6000/linux-unwind.h, config/rs6000/linux.h, config/rs6000/linux64.h: Likewise. Split out get_sigcontext function. Use ARG_POINTER_REGNUM for 32-bit temp reg too. From-SVN: r87167
2004-09-08 00:17:19 +00:00
fs->regs.reg[3].loc.offset = (long)&sc->rbx - new_cfa;
libgcc: Decrease size of _Unwind_FrameState and even more size of cleared area in uw_frame_state_for The following patch implements something that has Florian found as low hanging fruit in our unwinder and has been discussed in the https://gcc.gnu.org/wiki/cauldron2022#cauldron2022talks.inprocess_unwinding_bof talk. _Unwind_FrameState type seems to be (unlike the pre-GCC 3 frame_state which has been part of ABI) private to unwind-dw2.c + unwind.inc it includes, it is always defined on the stack of some entrypoints, initialized by static uw_frame_state_for and the address of it is also passed to other static functions or the static inlines handling machine dependent unwinding, but it isn't fortunately passed to any callbacks or public functions, so I think we can safely change it any time we want. Florian mentioned that the structure is large even on x86_64, 384 bytes there, starts with 328 bytes long element with frame_state_reg_info type which then starts with an array with __LIBGCC_DWARF_FRAME_REGISTERS__ + 1 elements, each of them is 16 bytes long, on x86_64 __LIBGCC_DWARF_FRAME_REGISTERS__ is just 17 but even that is big, on say riscv __LIBGCC_DWARF_FRAME_REGISTERS__ is I think 128, on powerpc 111, on sh 153 etc. And, we memset to zero the whole fs variable with the _Unwind_FrameState type at the start of the unwinding. The reason why each element is 16 byte (on 64-bit arches) is that it contains some pointer or pointer sized integer and then an enum (with just 7 different enumerators) + padding. The following patch decreases it by moving the enum into a separate array and using just one byte for each register in that second array. We could compress it even more, say 4 bits per register, but I don't want to uglify the code for it too much and make the accesses slower. Furthermore, the clearing of the object can clear only thos how array and members after it, because REG_UNSAVED enumerator (0) doesn't actually need any pointer or pointer sized integer, it is just the other kinds that need to have there something. By doing this, on x86_64 the above numbers change to _Unwind_FrameState type being now 264 bytes long, frame_state_reg_info 208 bytes and we don't clear the first 144 bytes of the object, so the memset is 120 bytes, so ~ 31% of the old clearing size. On riscv 64-bit assuming it has same structure layout rules for the few types used there that would be ~ 2160 bytes of _Unwind_FrameState type before and ~ 1264 bytes after, with the memset previously ~ 2160 bytes and after ~ 232 bytes after. We've also talked about possibly adding a number of initially initialized regs and initializing the rest lazily, but at least for x86_64 with 18 elements in the array that doesn't seem to be worth it anymore, especially because return address column is 16 there and that is usually the first thing to be touched. It might theory help with lots of registers if they are usually untouched, but would uglify and complicate any stores to how by having to check there for the not initialized yet cases and lazy initialization, and similarly for all reads of how to do there if below last initialized one, use how, otherwise imply REG_UNSAVED. The disadvantage of the patch is that touching reg[x].loc and how[x] now means 2 cachelines rather than one as before, and I admit beyond bootstrap/regtest I haven't benchmarked it in any way. 2022-10-06 Jakub Jelinek <jakub@redhat.com> * unwind-dw2.h (REG_UNSAVED, REG_SAVED_OFFSET, REG_SAVED_REG, REG_SAVED_EXP, REG_SAVED_VAL_OFFSET, REG_SAVED_VAL_EXP, REG_UNDEFINED): New anonymous enum, moved from inside of struct frame_state_reg_info. (struct frame_state_reg_info): Remove reg[].how element and the anonymous enum there. Add how element. * unwind-dw2.c: Include stddef.h. (uw_frame_state_for): Don't clear first offsetof (_Unwind_FrameState, regs.how[0]) bytes of *fs. (execute_cfa_program, __frame_state_for, uw_update_context_1, uw_update_context): Use fs->regs.how[X] instead of fs->regs.reg[X].how or fs.regs.how[X] instead of fs.regs.reg[X].how. * config/sh/linux-unwind.h (sh_fallback_frame_state): Likewise. * config/bfin/linux-unwind.h (bfin_fallback_frame_state): Likewise. * config/pa/linux-unwind.h (pa32_fallback_frame_state): Likewise. * config/pa/hpux-unwind.h (UPDATE_FS_FOR_SAR, UPDATE_FS_FOR_GR, UPDATE_FS_FOR_FR, UPDATE_FS_FOR_PC, pa_fallback_frame_state): Likewise. * config/alpha/vms-unwind.h (alpha_vms_fallback_frame_state): Likewise. * config/alpha/linux-unwind.h (alpha_fallback_frame_state): Likewise. * config/arc/linux-unwind.h (arc_fallback_frame_state, arc_frob_update_context): Likewise. * config/riscv/linux-unwind.h (riscv_fallback_frame_state): Likewise. * config/nios2/linux-unwind.h (NIOS2_REG): Likewise. * config/nds32/linux-unwind.h (NDS32_PUT_FS_REG): Likewise. * config/s390/tpf-unwind.h (s390_fallback_frame_state): Likewise. * config/s390/linux-unwind.h (s390_fallback_frame_state): Likewise. * config/sparc/sol2-unwind.h (sparc64_frob_update_context, MD_FALLBACK_FRAME_STATE_FOR): Likewise. * config/sparc/linux-unwind.h (sparc64_fallback_frame_state, sparc64_frob_update_context, sparc_fallback_frame_state): Likewise. * config/i386/sol2-unwind.h (x86_64_fallback_frame_state, x86_fallback_frame_state): Likewise. * config/i386/w32-unwind.h (i386_w32_fallback_frame_state): Likewise. * config/i386/linux-unwind.h (x86_64_fallback_frame_state, x86_fallback_frame_state): Likewise. * config/i386/freebsd-unwind.h (x86_64_freebsd_fallback_frame_state): Likewise. * config/i386/dragonfly-unwind.h (x86_64_dragonfly_fallback_frame_state): Likewise. * config/i386/gnu-unwind.h (x86_gnu_fallback_frame_state): Likewise. * config/csky/linux-unwind.h (csky_fallback_frame_state): Likewise. * config/aarch64/linux-unwind.h (aarch64_fallback_frame_state): Likewise. * config/aarch64/freebsd-unwind.h (aarch64_freebsd_fallback_frame_state): Likewise. * config/aarch64/aarch64-unwind.h (aarch64_frob_update_context): Likewise. * config/or1k/linux-unwind.h (or1k_fallback_frame_state): Likewise. * config/mips/linux-unwind.h (mips_fallback_frame_state): Likewise. * config/loongarch/linux-unwind.h (loongarch_fallback_frame_state): Likewise. * config/m68k/linux-unwind.h (m68k_fallback_frame_state): Likewise. * config/xtensa/linux-unwind.h (xtensa_fallback_frame_state): Likewise. * config/rs6000/darwin-fallback.c (set_offset): Likewise. * config/rs6000/aix-unwind.h (MD_FROB_UPDATE_CONTEXT): Likewise. * config/rs6000/linux-unwind.h (ppc_fallback_frame_state): Likewise. * config/rs6000/freebsd-unwind.h (frob_update_context): Likewise.
2022-10-06 11:04:52 +02:00
fs->regs.how[4] = REG_SAVED_OFFSET;
tm.texi (MD_UNWIND_SUPPORT): Document. * doc/tm.texi (MD_UNWIND_SUPPORT): Document. (MD_FALLBACK_FRAME_STATE_FOR): Update. * unwind-dw2.c (MD_UNWIND_SUPPORT): #include if defined. (uw_frame_state_for): Adjust MD_FALLBACK_FRAME_STATE_FOR invocation. (MD_FROB_UPDATE_CONTEXT): Remove default. (uw_update_context_1): Instead #ifdef invocation. * config/ia64/unwind-ia64.c (MD_UNWIND_SUPPORT): #include if defined. (uw_frame_state_for): Adjust MD_FALLBACK_FRAME_STATE_FOR invocation. * config/alpha/gnu.h (MD_FALLBACK_FRAME_STATE_FOR): Don't undef. (MD_UNWIND_SUPPORT): Undefine this instead. * config/i386/gnu.h: Likewise. * config/alpha/linux-unwind.h: New file, macro converted to function, extracted from.. * config/alpha/linux.h (MD_FALLBACK_FRAME_STATE_FOR): ..this. (MD_UNWIND_SUPPORT): Define. * config/alpha/vms-unwind.h, config/alpha/vms.h: Likewise. * config/i386/linux-unwind.h, config/i386/linux.h, config/i386/linux64.h: Likewise. * config/ia64/linux-unwind.h, config/ia64/linux.h: Likewise. MD_HANDLE_UNWABI too. * config/mips/linux-unwind.h, config/mips/linux.h: Likewise. * config/pa/linux-unwind.h, config/pa/pa32-linux.h: Likewise. * config/rs6000/darwin-unwind.h, config/rs6000/darwin.h: Likewise. * config/s390/linux-unwind.h, config/s390/linux.h: Likewise. * config/sparc/linux-unwind.h, config/sparc/linux.h, config/sparc/linux64.h: Likewise. * config/sh/linux-unwind.h, config/sh/linux.h: Likewise, but merge SH_FALLBACK_FRAME_FLOAT_STATE into sh_fallback_frame_state. * config/rs6000/linux-unwind.h, config/rs6000/linux.h, config/rs6000/linux64.h: Likewise. Split out get_sigcontext function. Use ARG_POINTER_REGNUM for 32-bit temp reg too. From-SVN: r87167
2004-09-08 00:17:19 +00:00
fs->regs.reg[4].loc.offset = (long)&sc->rsi - new_cfa;
libgcc: Decrease size of _Unwind_FrameState and even more size of cleared area in uw_frame_state_for The following patch implements something that has Florian found as low hanging fruit in our unwinder and has been discussed in the https://gcc.gnu.org/wiki/cauldron2022#cauldron2022talks.inprocess_unwinding_bof talk. _Unwind_FrameState type seems to be (unlike the pre-GCC 3 frame_state which has been part of ABI) private to unwind-dw2.c + unwind.inc it includes, it is always defined on the stack of some entrypoints, initialized by static uw_frame_state_for and the address of it is also passed to other static functions or the static inlines handling machine dependent unwinding, but it isn't fortunately passed to any callbacks or public functions, so I think we can safely change it any time we want. Florian mentioned that the structure is large even on x86_64, 384 bytes there, starts with 328 bytes long element with frame_state_reg_info type which then starts with an array with __LIBGCC_DWARF_FRAME_REGISTERS__ + 1 elements, each of them is 16 bytes long, on x86_64 __LIBGCC_DWARF_FRAME_REGISTERS__ is just 17 but even that is big, on say riscv __LIBGCC_DWARF_FRAME_REGISTERS__ is I think 128, on powerpc 111, on sh 153 etc. And, we memset to zero the whole fs variable with the _Unwind_FrameState type at the start of the unwinding. The reason why each element is 16 byte (on 64-bit arches) is that it contains some pointer or pointer sized integer and then an enum (with just 7 different enumerators) + padding. The following patch decreases it by moving the enum into a separate array and using just one byte for each register in that second array. We could compress it even more, say 4 bits per register, but I don't want to uglify the code for it too much and make the accesses slower. Furthermore, the clearing of the object can clear only thos how array and members after it, because REG_UNSAVED enumerator (0) doesn't actually need any pointer or pointer sized integer, it is just the other kinds that need to have there something. By doing this, on x86_64 the above numbers change to _Unwind_FrameState type being now 264 bytes long, frame_state_reg_info 208 bytes and we don't clear the first 144 bytes of the object, so the memset is 120 bytes, so ~ 31% of the old clearing size. On riscv 64-bit assuming it has same structure layout rules for the few types used there that would be ~ 2160 bytes of _Unwind_FrameState type before and ~ 1264 bytes after, with the memset previously ~ 2160 bytes and after ~ 232 bytes after. We've also talked about possibly adding a number of initially initialized regs and initializing the rest lazily, but at least for x86_64 with 18 elements in the array that doesn't seem to be worth it anymore, especially because return address column is 16 there and that is usually the first thing to be touched. It might theory help with lots of registers if they are usually untouched, but would uglify and complicate any stores to how by having to check there for the not initialized yet cases and lazy initialization, and similarly for all reads of how to do there if below last initialized one, use how, otherwise imply REG_UNSAVED. The disadvantage of the patch is that touching reg[x].loc and how[x] now means 2 cachelines rather than one as before, and I admit beyond bootstrap/regtest I haven't benchmarked it in any way. 2022-10-06 Jakub Jelinek <jakub@redhat.com> * unwind-dw2.h (REG_UNSAVED, REG_SAVED_OFFSET, REG_SAVED_REG, REG_SAVED_EXP, REG_SAVED_VAL_OFFSET, REG_SAVED_VAL_EXP, REG_UNDEFINED): New anonymous enum, moved from inside of struct frame_state_reg_info. (struct frame_state_reg_info): Remove reg[].how element and the anonymous enum there. Add how element. * unwind-dw2.c: Include stddef.h. (uw_frame_state_for): Don't clear first offsetof (_Unwind_FrameState, regs.how[0]) bytes of *fs. (execute_cfa_program, __frame_state_for, uw_update_context_1, uw_update_context): Use fs->regs.how[X] instead of fs->regs.reg[X].how or fs.regs.how[X] instead of fs.regs.reg[X].how. * config/sh/linux-unwind.h (sh_fallback_frame_state): Likewise. * config/bfin/linux-unwind.h (bfin_fallback_frame_state): Likewise. * config/pa/linux-unwind.h (pa32_fallback_frame_state): Likewise. * config/pa/hpux-unwind.h (UPDATE_FS_FOR_SAR, UPDATE_FS_FOR_GR, UPDATE_FS_FOR_FR, UPDATE_FS_FOR_PC, pa_fallback_frame_state): Likewise. * config/alpha/vms-unwind.h (alpha_vms_fallback_frame_state): Likewise. * config/alpha/linux-unwind.h (alpha_fallback_frame_state): Likewise. * config/arc/linux-unwind.h (arc_fallback_frame_state, arc_frob_update_context): Likewise. * config/riscv/linux-unwind.h (riscv_fallback_frame_state): Likewise. * config/nios2/linux-unwind.h (NIOS2_REG): Likewise. * config/nds32/linux-unwind.h (NDS32_PUT_FS_REG): Likewise. * config/s390/tpf-unwind.h (s390_fallback_frame_state): Likewise. * config/s390/linux-unwind.h (s390_fallback_frame_state): Likewise. * config/sparc/sol2-unwind.h (sparc64_frob_update_context, MD_FALLBACK_FRAME_STATE_FOR): Likewise. * config/sparc/linux-unwind.h (sparc64_fallback_frame_state, sparc64_frob_update_context, sparc_fallback_frame_state): Likewise. * config/i386/sol2-unwind.h (x86_64_fallback_frame_state, x86_fallback_frame_state): Likewise. * config/i386/w32-unwind.h (i386_w32_fallback_frame_state): Likewise. * config/i386/linux-unwind.h (x86_64_fallback_frame_state, x86_fallback_frame_state): Likewise. * config/i386/freebsd-unwind.h (x86_64_freebsd_fallback_frame_state): Likewise. * config/i386/dragonfly-unwind.h (x86_64_dragonfly_fallback_frame_state): Likewise. * config/i386/gnu-unwind.h (x86_gnu_fallback_frame_state): Likewise. * config/csky/linux-unwind.h (csky_fallback_frame_state): Likewise. * config/aarch64/linux-unwind.h (aarch64_fallback_frame_state): Likewise. * config/aarch64/freebsd-unwind.h (aarch64_freebsd_fallback_frame_state): Likewise. * config/aarch64/aarch64-unwind.h (aarch64_frob_update_context): Likewise. * config/or1k/linux-unwind.h (or1k_fallback_frame_state): Likewise. * config/mips/linux-unwind.h (mips_fallback_frame_state): Likewise. * config/loongarch/linux-unwind.h (loongarch_fallback_frame_state): Likewise. * config/m68k/linux-unwind.h (m68k_fallback_frame_state): Likewise. * config/xtensa/linux-unwind.h (xtensa_fallback_frame_state): Likewise. * config/rs6000/darwin-fallback.c (set_offset): Likewise. * config/rs6000/aix-unwind.h (MD_FROB_UPDATE_CONTEXT): Likewise. * config/rs6000/linux-unwind.h (ppc_fallback_frame_state): Likewise. * config/rs6000/freebsd-unwind.h (frob_update_context): Likewise.
2022-10-06 11:04:52 +02:00
fs->regs.how[5] = REG_SAVED_OFFSET;
tm.texi (MD_UNWIND_SUPPORT): Document. * doc/tm.texi (MD_UNWIND_SUPPORT): Document. (MD_FALLBACK_FRAME_STATE_FOR): Update. * unwind-dw2.c (MD_UNWIND_SUPPORT): #include if defined. (uw_frame_state_for): Adjust MD_FALLBACK_FRAME_STATE_FOR invocation. (MD_FROB_UPDATE_CONTEXT): Remove default. (uw_update_context_1): Instead #ifdef invocation. * config/ia64/unwind-ia64.c (MD_UNWIND_SUPPORT): #include if defined. (uw_frame_state_for): Adjust MD_FALLBACK_FRAME_STATE_FOR invocation. * config/alpha/gnu.h (MD_FALLBACK_FRAME_STATE_FOR): Don't undef. (MD_UNWIND_SUPPORT): Undefine this instead. * config/i386/gnu.h: Likewise. * config/alpha/linux-unwind.h: New file, macro converted to function, extracted from.. * config/alpha/linux.h (MD_FALLBACK_FRAME_STATE_FOR): ..this. (MD_UNWIND_SUPPORT): Define. * config/alpha/vms-unwind.h, config/alpha/vms.h: Likewise. * config/i386/linux-unwind.h, config/i386/linux.h, config/i386/linux64.h: Likewise. * config/ia64/linux-unwind.h, config/ia64/linux.h: Likewise. MD_HANDLE_UNWABI too. * config/mips/linux-unwind.h, config/mips/linux.h: Likewise. * config/pa/linux-unwind.h, config/pa/pa32-linux.h: Likewise. * config/rs6000/darwin-unwind.h, config/rs6000/darwin.h: Likewise. * config/s390/linux-unwind.h, config/s390/linux.h: Likewise. * config/sparc/linux-unwind.h, config/sparc/linux.h, config/sparc/linux64.h: Likewise. * config/sh/linux-unwind.h, config/sh/linux.h: Likewise, but merge SH_FALLBACK_FRAME_FLOAT_STATE into sh_fallback_frame_state. * config/rs6000/linux-unwind.h, config/rs6000/linux.h, config/rs6000/linux64.h: Likewise. Split out get_sigcontext function. Use ARG_POINTER_REGNUM for 32-bit temp reg too. From-SVN: r87167
2004-09-08 00:17:19 +00:00
fs->regs.reg[5].loc.offset = (long)&sc->rdi - new_cfa;
libgcc: Decrease size of _Unwind_FrameState and even more size of cleared area in uw_frame_state_for The following patch implements something that has Florian found as low hanging fruit in our unwinder and has been discussed in the https://gcc.gnu.org/wiki/cauldron2022#cauldron2022talks.inprocess_unwinding_bof talk. _Unwind_FrameState type seems to be (unlike the pre-GCC 3 frame_state which has been part of ABI) private to unwind-dw2.c + unwind.inc it includes, it is always defined on the stack of some entrypoints, initialized by static uw_frame_state_for and the address of it is also passed to other static functions or the static inlines handling machine dependent unwinding, but it isn't fortunately passed to any callbacks or public functions, so I think we can safely change it any time we want. Florian mentioned that the structure is large even on x86_64, 384 bytes there, starts with 328 bytes long element with frame_state_reg_info type which then starts with an array with __LIBGCC_DWARF_FRAME_REGISTERS__ + 1 elements, each of them is 16 bytes long, on x86_64 __LIBGCC_DWARF_FRAME_REGISTERS__ is just 17 but even that is big, on say riscv __LIBGCC_DWARF_FRAME_REGISTERS__ is I think 128, on powerpc 111, on sh 153 etc. And, we memset to zero the whole fs variable with the _Unwind_FrameState type at the start of the unwinding. The reason why each element is 16 byte (on 64-bit arches) is that it contains some pointer or pointer sized integer and then an enum (with just 7 different enumerators) + padding. The following patch decreases it by moving the enum into a separate array and using just one byte for each register in that second array. We could compress it even more, say 4 bits per register, but I don't want to uglify the code for it too much and make the accesses slower. Furthermore, the clearing of the object can clear only thos how array and members after it, because REG_UNSAVED enumerator (0) doesn't actually need any pointer or pointer sized integer, it is just the other kinds that need to have there something. By doing this, on x86_64 the above numbers change to _Unwind_FrameState type being now 264 bytes long, frame_state_reg_info 208 bytes and we don't clear the first 144 bytes of the object, so the memset is 120 bytes, so ~ 31% of the old clearing size. On riscv 64-bit assuming it has same structure layout rules for the few types used there that would be ~ 2160 bytes of _Unwind_FrameState type before and ~ 1264 bytes after, with the memset previously ~ 2160 bytes and after ~ 232 bytes after. We've also talked about possibly adding a number of initially initialized regs and initializing the rest lazily, but at least for x86_64 with 18 elements in the array that doesn't seem to be worth it anymore, especially because return address column is 16 there and that is usually the first thing to be touched. It might theory help with lots of registers if they are usually untouched, but would uglify and complicate any stores to how by having to check there for the not initialized yet cases and lazy initialization, and similarly for all reads of how to do there if below last initialized one, use how, otherwise imply REG_UNSAVED. The disadvantage of the patch is that touching reg[x].loc and how[x] now means 2 cachelines rather than one as before, and I admit beyond bootstrap/regtest I haven't benchmarked it in any way. 2022-10-06 Jakub Jelinek <jakub@redhat.com> * unwind-dw2.h (REG_UNSAVED, REG_SAVED_OFFSET, REG_SAVED_REG, REG_SAVED_EXP, REG_SAVED_VAL_OFFSET, REG_SAVED_VAL_EXP, REG_UNDEFINED): New anonymous enum, moved from inside of struct frame_state_reg_info. (struct frame_state_reg_info): Remove reg[].how element and the anonymous enum there. Add how element. * unwind-dw2.c: Include stddef.h. (uw_frame_state_for): Don't clear first offsetof (_Unwind_FrameState, regs.how[0]) bytes of *fs. (execute_cfa_program, __frame_state_for, uw_update_context_1, uw_update_context): Use fs->regs.how[X] instead of fs->regs.reg[X].how or fs.regs.how[X] instead of fs.regs.reg[X].how. * config/sh/linux-unwind.h (sh_fallback_frame_state): Likewise. * config/bfin/linux-unwind.h (bfin_fallback_frame_state): Likewise. * config/pa/linux-unwind.h (pa32_fallback_frame_state): Likewise. * config/pa/hpux-unwind.h (UPDATE_FS_FOR_SAR, UPDATE_FS_FOR_GR, UPDATE_FS_FOR_FR, UPDATE_FS_FOR_PC, pa_fallback_frame_state): Likewise. * config/alpha/vms-unwind.h (alpha_vms_fallback_frame_state): Likewise. * config/alpha/linux-unwind.h (alpha_fallback_frame_state): Likewise. * config/arc/linux-unwind.h (arc_fallback_frame_state, arc_frob_update_context): Likewise. * config/riscv/linux-unwind.h (riscv_fallback_frame_state): Likewise. * config/nios2/linux-unwind.h (NIOS2_REG): Likewise. * config/nds32/linux-unwind.h (NDS32_PUT_FS_REG): Likewise. * config/s390/tpf-unwind.h (s390_fallback_frame_state): Likewise. * config/s390/linux-unwind.h (s390_fallback_frame_state): Likewise. * config/sparc/sol2-unwind.h (sparc64_frob_update_context, MD_FALLBACK_FRAME_STATE_FOR): Likewise. * config/sparc/linux-unwind.h (sparc64_fallback_frame_state, sparc64_frob_update_context, sparc_fallback_frame_state): Likewise. * config/i386/sol2-unwind.h (x86_64_fallback_frame_state, x86_fallback_frame_state): Likewise. * config/i386/w32-unwind.h (i386_w32_fallback_frame_state): Likewise. * config/i386/linux-unwind.h (x86_64_fallback_frame_state, x86_fallback_frame_state): Likewise. * config/i386/freebsd-unwind.h (x86_64_freebsd_fallback_frame_state): Likewise. * config/i386/dragonfly-unwind.h (x86_64_dragonfly_fallback_frame_state): Likewise. * config/i386/gnu-unwind.h (x86_gnu_fallback_frame_state): Likewise. * config/csky/linux-unwind.h (csky_fallback_frame_state): Likewise. * config/aarch64/linux-unwind.h (aarch64_fallback_frame_state): Likewise. * config/aarch64/freebsd-unwind.h (aarch64_freebsd_fallback_frame_state): Likewise. * config/aarch64/aarch64-unwind.h (aarch64_frob_update_context): Likewise. * config/or1k/linux-unwind.h (or1k_fallback_frame_state): Likewise. * config/mips/linux-unwind.h (mips_fallback_frame_state): Likewise. * config/loongarch/linux-unwind.h (loongarch_fallback_frame_state): Likewise. * config/m68k/linux-unwind.h (m68k_fallback_frame_state): Likewise. * config/xtensa/linux-unwind.h (xtensa_fallback_frame_state): Likewise. * config/rs6000/darwin-fallback.c (set_offset): Likewise. * config/rs6000/aix-unwind.h (MD_FROB_UPDATE_CONTEXT): Likewise. * config/rs6000/linux-unwind.h (ppc_fallback_frame_state): Likewise. * config/rs6000/freebsd-unwind.h (frob_update_context): Likewise.
2022-10-06 11:04:52 +02:00
fs->regs.how[6] = REG_SAVED_OFFSET;
tm.texi (MD_UNWIND_SUPPORT): Document. * doc/tm.texi (MD_UNWIND_SUPPORT): Document. (MD_FALLBACK_FRAME_STATE_FOR): Update. * unwind-dw2.c (MD_UNWIND_SUPPORT): #include if defined. (uw_frame_state_for): Adjust MD_FALLBACK_FRAME_STATE_FOR invocation. (MD_FROB_UPDATE_CONTEXT): Remove default. (uw_update_context_1): Instead #ifdef invocation. * config/ia64/unwind-ia64.c (MD_UNWIND_SUPPORT): #include if defined. (uw_frame_state_for): Adjust MD_FALLBACK_FRAME_STATE_FOR invocation. * config/alpha/gnu.h (MD_FALLBACK_FRAME_STATE_FOR): Don't undef. (MD_UNWIND_SUPPORT): Undefine this instead. * config/i386/gnu.h: Likewise. * config/alpha/linux-unwind.h: New file, macro converted to function, extracted from.. * config/alpha/linux.h (MD_FALLBACK_FRAME_STATE_FOR): ..this. (MD_UNWIND_SUPPORT): Define. * config/alpha/vms-unwind.h, config/alpha/vms.h: Likewise. * config/i386/linux-unwind.h, config/i386/linux.h, config/i386/linux64.h: Likewise. * config/ia64/linux-unwind.h, config/ia64/linux.h: Likewise. MD_HANDLE_UNWABI too. * config/mips/linux-unwind.h, config/mips/linux.h: Likewise. * config/pa/linux-unwind.h, config/pa/pa32-linux.h: Likewise. * config/rs6000/darwin-unwind.h, config/rs6000/darwin.h: Likewise. * config/s390/linux-unwind.h, config/s390/linux.h: Likewise. * config/sparc/linux-unwind.h, config/sparc/linux.h, config/sparc/linux64.h: Likewise. * config/sh/linux-unwind.h, config/sh/linux.h: Likewise, but merge SH_FALLBACK_FRAME_FLOAT_STATE into sh_fallback_frame_state. * config/rs6000/linux-unwind.h, config/rs6000/linux.h, config/rs6000/linux64.h: Likewise. Split out get_sigcontext function. Use ARG_POINTER_REGNUM for 32-bit temp reg too. From-SVN: r87167
2004-09-08 00:17:19 +00:00
fs->regs.reg[6].loc.offset = (long)&sc->rbp - new_cfa;
libgcc: Decrease size of _Unwind_FrameState and even more size of cleared area in uw_frame_state_for The following patch implements something that has Florian found as low hanging fruit in our unwinder and has been discussed in the https://gcc.gnu.org/wiki/cauldron2022#cauldron2022talks.inprocess_unwinding_bof talk. _Unwind_FrameState type seems to be (unlike the pre-GCC 3 frame_state which has been part of ABI) private to unwind-dw2.c + unwind.inc it includes, it is always defined on the stack of some entrypoints, initialized by static uw_frame_state_for and the address of it is also passed to other static functions or the static inlines handling machine dependent unwinding, but it isn't fortunately passed to any callbacks or public functions, so I think we can safely change it any time we want. Florian mentioned that the structure is large even on x86_64, 384 bytes there, starts with 328 bytes long element with frame_state_reg_info type which then starts with an array with __LIBGCC_DWARF_FRAME_REGISTERS__ + 1 elements, each of them is 16 bytes long, on x86_64 __LIBGCC_DWARF_FRAME_REGISTERS__ is just 17 but even that is big, on say riscv __LIBGCC_DWARF_FRAME_REGISTERS__ is I think 128, on powerpc 111, on sh 153 etc. And, we memset to zero the whole fs variable with the _Unwind_FrameState type at the start of the unwinding. The reason why each element is 16 byte (on 64-bit arches) is that it contains some pointer or pointer sized integer and then an enum (with just 7 different enumerators) + padding. The following patch decreases it by moving the enum into a separate array and using just one byte for each register in that second array. We could compress it even more, say 4 bits per register, but I don't want to uglify the code for it too much and make the accesses slower. Furthermore, the clearing of the object can clear only thos how array and members after it, because REG_UNSAVED enumerator (0) doesn't actually need any pointer or pointer sized integer, it is just the other kinds that need to have there something. By doing this, on x86_64 the above numbers change to _Unwind_FrameState type being now 264 bytes long, frame_state_reg_info 208 bytes and we don't clear the first 144 bytes of the object, so the memset is 120 bytes, so ~ 31% of the old clearing size. On riscv 64-bit assuming it has same structure layout rules for the few types used there that would be ~ 2160 bytes of _Unwind_FrameState type before and ~ 1264 bytes after, with the memset previously ~ 2160 bytes and after ~ 232 bytes after. We've also talked about possibly adding a number of initially initialized regs and initializing the rest lazily, but at least for x86_64 with 18 elements in the array that doesn't seem to be worth it anymore, especially because return address column is 16 there and that is usually the first thing to be touched. It might theory help with lots of registers if they are usually untouched, but would uglify and complicate any stores to how by having to check there for the not initialized yet cases and lazy initialization, and similarly for all reads of how to do there if below last initialized one, use how, otherwise imply REG_UNSAVED. The disadvantage of the patch is that touching reg[x].loc and how[x] now means 2 cachelines rather than one as before, and I admit beyond bootstrap/regtest I haven't benchmarked it in any way. 2022-10-06 Jakub Jelinek <jakub@redhat.com> * unwind-dw2.h (REG_UNSAVED, REG_SAVED_OFFSET, REG_SAVED_REG, REG_SAVED_EXP, REG_SAVED_VAL_OFFSET, REG_SAVED_VAL_EXP, REG_UNDEFINED): New anonymous enum, moved from inside of struct frame_state_reg_info. (struct frame_state_reg_info): Remove reg[].how element and the anonymous enum there. Add how element. * unwind-dw2.c: Include stddef.h. (uw_frame_state_for): Don't clear first offsetof (_Unwind_FrameState, regs.how[0]) bytes of *fs. (execute_cfa_program, __frame_state_for, uw_update_context_1, uw_update_context): Use fs->regs.how[X] instead of fs->regs.reg[X].how or fs.regs.how[X] instead of fs.regs.reg[X].how. * config/sh/linux-unwind.h (sh_fallback_frame_state): Likewise. * config/bfin/linux-unwind.h (bfin_fallback_frame_state): Likewise. * config/pa/linux-unwind.h (pa32_fallback_frame_state): Likewise. * config/pa/hpux-unwind.h (UPDATE_FS_FOR_SAR, UPDATE_FS_FOR_GR, UPDATE_FS_FOR_FR, UPDATE_FS_FOR_PC, pa_fallback_frame_state): Likewise. * config/alpha/vms-unwind.h (alpha_vms_fallback_frame_state): Likewise. * config/alpha/linux-unwind.h (alpha_fallback_frame_state): Likewise. * config/arc/linux-unwind.h (arc_fallback_frame_state, arc_frob_update_context): Likewise. * config/riscv/linux-unwind.h (riscv_fallback_frame_state): Likewise. * config/nios2/linux-unwind.h (NIOS2_REG): Likewise. * config/nds32/linux-unwind.h (NDS32_PUT_FS_REG): Likewise. * config/s390/tpf-unwind.h (s390_fallback_frame_state): Likewise. * config/s390/linux-unwind.h (s390_fallback_frame_state): Likewise. * config/sparc/sol2-unwind.h (sparc64_frob_update_context, MD_FALLBACK_FRAME_STATE_FOR): Likewise. * config/sparc/linux-unwind.h (sparc64_fallback_frame_state, sparc64_frob_update_context, sparc_fallback_frame_state): Likewise. * config/i386/sol2-unwind.h (x86_64_fallback_frame_state, x86_fallback_frame_state): Likewise. * config/i386/w32-unwind.h (i386_w32_fallback_frame_state): Likewise. * config/i386/linux-unwind.h (x86_64_fallback_frame_state, x86_fallback_frame_state): Likewise. * config/i386/freebsd-unwind.h (x86_64_freebsd_fallback_frame_state): Likewise. * config/i386/dragonfly-unwind.h (x86_64_dragonfly_fallback_frame_state): Likewise. * config/i386/gnu-unwind.h (x86_gnu_fallback_frame_state): Likewise. * config/csky/linux-unwind.h (csky_fallback_frame_state): Likewise. * config/aarch64/linux-unwind.h (aarch64_fallback_frame_state): Likewise. * config/aarch64/freebsd-unwind.h (aarch64_freebsd_fallback_frame_state): Likewise. * config/aarch64/aarch64-unwind.h (aarch64_frob_update_context): Likewise. * config/or1k/linux-unwind.h (or1k_fallback_frame_state): Likewise. * config/mips/linux-unwind.h (mips_fallback_frame_state): Likewise. * config/loongarch/linux-unwind.h (loongarch_fallback_frame_state): Likewise. * config/m68k/linux-unwind.h (m68k_fallback_frame_state): Likewise. * config/xtensa/linux-unwind.h (xtensa_fallback_frame_state): Likewise. * config/rs6000/darwin-fallback.c (set_offset): Likewise. * config/rs6000/aix-unwind.h (MD_FROB_UPDATE_CONTEXT): Likewise. * config/rs6000/linux-unwind.h (ppc_fallback_frame_state): Likewise. * config/rs6000/freebsd-unwind.h (frob_update_context): Likewise.
2022-10-06 11:04:52 +02:00
fs->regs.how[8] = REG_SAVED_OFFSET;
tm.texi (MD_UNWIND_SUPPORT): Document. * doc/tm.texi (MD_UNWIND_SUPPORT): Document. (MD_FALLBACK_FRAME_STATE_FOR): Update. * unwind-dw2.c (MD_UNWIND_SUPPORT): #include if defined. (uw_frame_state_for): Adjust MD_FALLBACK_FRAME_STATE_FOR invocation. (MD_FROB_UPDATE_CONTEXT): Remove default. (uw_update_context_1): Instead #ifdef invocation. * config/ia64/unwind-ia64.c (MD_UNWIND_SUPPORT): #include if defined. (uw_frame_state_for): Adjust MD_FALLBACK_FRAME_STATE_FOR invocation. * config/alpha/gnu.h (MD_FALLBACK_FRAME_STATE_FOR): Don't undef. (MD_UNWIND_SUPPORT): Undefine this instead. * config/i386/gnu.h: Likewise. * config/alpha/linux-unwind.h: New file, macro converted to function, extracted from.. * config/alpha/linux.h (MD_FALLBACK_FRAME_STATE_FOR): ..this. (MD_UNWIND_SUPPORT): Define. * config/alpha/vms-unwind.h, config/alpha/vms.h: Likewise. * config/i386/linux-unwind.h, config/i386/linux.h, config/i386/linux64.h: Likewise. * config/ia64/linux-unwind.h, config/ia64/linux.h: Likewise. MD_HANDLE_UNWABI too. * config/mips/linux-unwind.h, config/mips/linux.h: Likewise. * config/pa/linux-unwind.h, config/pa/pa32-linux.h: Likewise. * config/rs6000/darwin-unwind.h, config/rs6000/darwin.h: Likewise. * config/s390/linux-unwind.h, config/s390/linux.h: Likewise. * config/sparc/linux-unwind.h, config/sparc/linux.h, config/sparc/linux64.h: Likewise. * config/sh/linux-unwind.h, config/sh/linux.h: Likewise, but merge SH_FALLBACK_FRAME_FLOAT_STATE into sh_fallback_frame_state. * config/rs6000/linux-unwind.h, config/rs6000/linux.h, config/rs6000/linux64.h: Likewise. Split out get_sigcontext function. Use ARG_POINTER_REGNUM for 32-bit temp reg too. From-SVN: r87167
2004-09-08 00:17:19 +00:00
fs->regs.reg[8].loc.offset = (long)&sc->r8 - new_cfa;
libgcc: Decrease size of _Unwind_FrameState and even more size of cleared area in uw_frame_state_for The following patch implements something that has Florian found as low hanging fruit in our unwinder and has been discussed in the https://gcc.gnu.org/wiki/cauldron2022#cauldron2022talks.inprocess_unwinding_bof talk. _Unwind_FrameState type seems to be (unlike the pre-GCC 3 frame_state which has been part of ABI) private to unwind-dw2.c + unwind.inc it includes, it is always defined on the stack of some entrypoints, initialized by static uw_frame_state_for and the address of it is also passed to other static functions or the static inlines handling machine dependent unwinding, but it isn't fortunately passed to any callbacks or public functions, so I think we can safely change it any time we want. Florian mentioned that the structure is large even on x86_64, 384 bytes there, starts with 328 bytes long element with frame_state_reg_info type which then starts with an array with __LIBGCC_DWARF_FRAME_REGISTERS__ + 1 elements, each of them is 16 bytes long, on x86_64 __LIBGCC_DWARF_FRAME_REGISTERS__ is just 17 but even that is big, on say riscv __LIBGCC_DWARF_FRAME_REGISTERS__ is I think 128, on powerpc 111, on sh 153 etc. And, we memset to zero the whole fs variable with the _Unwind_FrameState type at the start of the unwinding. The reason why each element is 16 byte (on 64-bit arches) is that it contains some pointer or pointer sized integer and then an enum (with just 7 different enumerators) + padding. The following patch decreases it by moving the enum into a separate array and using just one byte for each register in that second array. We could compress it even more, say 4 bits per register, but I don't want to uglify the code for it too much and make the accesses slower. Furthermore, the clearing of the object can clear only thos how array and members after it, because REG_UNSAVED enumerator (0) doesn't actually need any pointer or pointer sized integer, it is just the other kinds that need to have there something. By doing this, on x86_64 the above numbers change to _Unwind_FrameState type being now 264 bytes long, frame_state_reg_info 208 bytes and we don't clear the first 144 bytes of the object, so the memset is 120 bytes, so ~ 31% of the old clearing size. On riscv 64-bit assuming it has same structure layout rules for the few types used there that would be ~ 2160 bytes of _Unwind_FrameState type before and ~ 1264 bytes after, with the memset previously ~ 2160 bytes and after ~ 232 bytes after. We've also talked about possibly adding a number of initially initialized regs and initializing the rest lazily, but at least for x86_64 with 18 elements in the array that doesn't seem to be worth it anymore, especially because return address column is 16 there and that is usually the first thing to be touched. It might theory help with lots of registers if they are usually untouched, but would uglify and complicate any stores to how by having to check there for the not initialized yet cases and lazy initialization, and similarly for all reads of how to do there if below last initialized one, use how, otherwise imply REG_UNSAVED. The disadvantage of the patch is that touching reg[x].loc and how[x] now means 2 cachelines rather than one as before, and I admit beyond bootstrap/regtest I haven't benchmarked it in any way. 2022-10-06 Jakub Jelinek <jakub@redhat.com> * unwind-dw2.h (REG_UNSAVED, REG_SAVED_OFFSET, REG_SAVED_REG, REG_SAVED_EXP, REG_SAVED_VAL_OFFSET, REG_SAVED_VAL_EXP, REG_UNDEFINED): New anonymous enum, moved from inside of struct frame_state_reg_info. (struct frame_state_reg_info): Remove reg[].how element and the anonymous enum there. Add how element. * unwind-dw2.c: Include stddef.h. (uw_frame_state_for): Don't clear first offsetof (_Unwind_FrameState, regs.how[0]) bytes of *fs. (execute_cfa_program, __frame_state_for, uw_update_context_1, uw_update_context): Use fs->regs.how[X] instead of fs->regs.reg[X].how or fs.regs.how[X] instead of fs.regs.reg[X].how. * config/sh/linux-unwind.h (sh_fallback_frame_state): Likewise. * config/bfin/linux-unwind.h (bfin_fallback_frame_state): Likewise. * config/pa/linux-unwind.h (pa32_fallback_frame_state): Likewise. * config/pa/hpux-unwind.h (UPDATE_FS_FOR_SAR, UPDATE_FS_FOR_GR, UPDATE_FS_FOR_FR, UPDATE_FS_FOR_PC, pa_fallback_frame_state): Likewise. * config/alpha/vms-unwind.h (alpha_vms_fallback_frame_state): Likewise. * config/alpha/linux-unwind.h (alpha_fallback_frame_state): Likewise. * config/arc/linux-unwind.h (arc_fallback_frame_state, arc_frob_update_context): Likewise. * config/riscv/linux-unwind.h (riscv_fallback_frame_state): Likewise. * config/nios2/linux-unwind.h (NIOS2_REG): Likewise. * config/nds32/linux-unwind.h (NDS32_PUT_FS_REG): Likewise. * config/s390/tpf-unwind.h (s390_fallback_frame_state): Likewise. * config/s390/linux-unwind.h (s390_fallback_frame_state): Likewise. * config/sparc/sol2-unwind.h (sparc64_frob_update_context, MD_FALLBACK_FRAME_STATE_FOR): Likewise. * config/sparc/linux-unwind.h (sparc64_fallback_frame_state, sparc64_frob_update_context, sparc_fallback_frame_state): Likewise. * config/i386/sol2-unwind.h (x86_64_fallback_frame_state, x86_fallback_frame_state): Likewise. * config/i386/w32-unwind.h (i386_w32_fallback_frame_state): Likewise. * config/i386/linux-unwind.h (x86_64_fallback_frame_state, x86_fallback_frame_state): Likewise. * config/i386/freebsd-unwind.h (x86_64_freebsd_fallback_frame_state): Likewise. * config/i386/dragonfly-unwind.h (x86_64_dragonfly_fallback_frame_state): Likewise. * config/i386/gnu-unwind.h (x86_gnu_fallback_frame_state): Likewise. * config/csky/linux-unwind.h (csky_fallback_frame_state): Likewise. * config/aarch64/linux-unwind.h (aarch64_fallback_frame_state): Likewise. * config/aarch64/freebsd-unwind.h (aarch64_freebsd_fallback_frame_state): Likewise. * config/aarch64/aarch64-unwind.h (aarch64_frob_update_context): Likewise. * config/or1k/linux-unwind.h (or1k_fallback_frame_state): Likewise. * config/mips/linux-unwind.h (mips_fallback_frame_state): Likewise. * config/loongarch/linux-unwind.h (loongarch_fallback_frame_state): Likewise. * config/m68k/linux-unwind.h (m68k_fallback_frame_state): Likewise. * config/xtensa/linux-unwind.h (xtensa_fallback_frame_state): Likewise. * config/rs6000/darwin-fallback.c (set_offset): Likewise. * config/rs6000/aix-unwind.h (MD_FROB_UPDATE_CONTEXT): Likewise. * config/rs6000/linux-unwind.h (ppc_fallback_frame_state): Likewise. * config/rs6000/freebsd-unwind.h (frob_update_context): Likewise.
2022-10-06 11:04:52 +02:00
fs->regs.how[9] = REG_SAVED_OFFSET;
tm.texi (MD_UNWIND_SUPPORT): Document. * doc/tm.texi (MD_UNWIND_SUPPORT): Document. (MD_FALLBACK_FRAME_STATE_FOR): Update. * unwind-dw2.c (MD_UNWIND_SUPPORT): #include if defined. (uw_frame_state_for): Adjust MD_FALLBACK_FRAME_STATE_FOR invocation. (MD_FROB_UPDATE_CONTEXT): Remove default. (uw_update_context_1): Instead #ifdef invocation. * config/ia64/unwind-ia64.c (MD_UNWIND_SUPPORT): #include if defined. (uw_frame_state_for): Adjust MD_FALLBACK_FRAME_STATE_FOR invocation. * config/alpha/gnu.h (MD_FALLBACK_FRAME_STATE_FOR): Don't undef. (MD_UNWIND_SUPPORT): Undefine this instead. * config/i386/gnu.h: Likewise. * config/alpha/linux-unwind.h: New file, macro converted to function, extracted from.. * config/alpha/linux.h (MD_FALLBACK_FRAME_STATE_FOR): ..this. (MD_UNWIND_SUPPORT): Define. * config/alpha/vms-unwind.h, config/alpha/vms.h: Likewise. * config/i386/linux-unwind.h, config/i386/linux.h, config/i386/linux64.h: Likewise. * config/ia64/linux-unwind.h, config/ia64/linux.h: Likewise. MD_HANDLE_UNWABI too. * config/mips/linux-unwind.h, config/mips/linux.h: Likewise. * config/pa/linux-unwind.h, config/pa/pa32-linux.h: Likewise. * config/rs6000/darwin-unwind.h, config/rs6000/darwin.h: Likewise. * config/s390/linux-unwind.h, config/s390/linux.h: Likewise. * config/sparc/linux-unwind.h, config/sparc/linux.h, config/sparc/linux64.h: Likewise. * config/sh/linux-unwind.h, config/sh/linux.h: Likewise, but merge SH_FALLBACK_FRAME_FLOAT_STATE into sh_fallback_frame_state. * config/rs6000/linux-unwind.h, config/rs6000/linux.h, config/rs6000/linux64.h: Likewise. Split out get_sigcontext function. Use ARG_POINTER_REGNUM for 32-bit temp reg too. From-SVN: r87167
2004-09-08 00:17:19 +00:00
fs->regs.reg[9].loc.offset = (long)&sc->r9 - new_cfa;
libgcc: Decrease size of _Unwind_FrameState and even more size of cleared area in uw_frame_state_for The following patch implements something that has Florian found as low hanging fruit in our unwinder and has been discussed in the https://gcc.gnu.org/wiki/cauldron2022#cauldron2022talks.inprocess_unwinding_bof talk. _Unwind_FrameState type seems to be (unlike the pre-GCC 3 frame_state which has been part of ABI) private to unwind-dw2.c + unwind.inc it includes, it is always defined on the stack of some entrypoints, initialized by static uw_frame_state_for and the address of it is also passed to other static functions or the static inlines handling machine dependent unwinding, but it isn't fortunately passed to any callbacks or public functions, so I think we can safely change it any time we want. Florian mentioned that the structure is large even on x86_64, 384 bytes there, starts with 328 bytes long element with frame_state_reg_info type which then starts with an array with __LIBGCC_DWARF_FRAME_REGISTERS__ + 1 elements, each of them is 16 bytes long, on x86_64 __LIBGCC_DWARF_FRAME_REGISTERS__ is just 17 but even that is big, on say riscv __LIBGCC_DWARF_FRAME_REGISTERS__ is I think 128, on powerpc 111, on sh 153 etc. And, we memset to zero the whole fs variable with the _Unwind_FrameState type at the start of the unwinding. The reason why each element is 16 byte (on 64-bit arches) is that it contains some pointer or pointer sized integer and then an enum (with just 7 different enumerators) + padding. The following patch decreases it by moving the enum into a separate array and using just one byte for each register in that second array. We could compress it even more, say 4 bits per register, but I don't want to uglify the code for it too much and make the accesses slower. Furthermore, the clearing of the object can clear only thos how array and members after it, because REG_UNSAVED enumerator (0) doesn't actually need any pointer or pointer sized integer, it is just the other kinds that need to have there something. By doing this, on x86_64 the above numbers change to _Unwind_FrameState type being now 264 bytes long, frame_state_reg_info 208 bytes and we don't clear the first 144 bytes of the object, so the memset is 120 bytes, so ~ 31% of the old clearing size. On riscv 64-bit assuming it has same structure layout rules for the few types used there that would be ~ 2160 bytes of _Unwind_FrameState type before and ~ 1264 bytes after, with the memset previously ~ 2160 bytes and after ~ 232 bytes after. We've also talked about possibly adding a number of initially initialized regs and initializing the rest lazily, but at least for x86_64 with 18 elements in the array that doesn't seem to be worth it anymore, especially because return address column is 16 there and that is usually the first thing to be touched. It might theory help with lots of registers if they are usually untouched, but would uglify and complicate any stores to how by having to check there for the not initialized yet cases and lazy initialization, and similarly for all reads of how to do there if below last initialized one, use how, otherwise imply REG_UNSAVED. The disadvantage of the patch is that touching reg[x].loc and how[x] now means 2 cachelines rather than one as before, and I admit beyond bootstrap/regtest I haven't benchmarked it in any way. 2022-10-06 Jakub Jelinek <jakub@redhat.com> * unwind-dw2.h (REG_UNSAVED, REG_SAVED_OFFSET, REG_SAVED_REG, REG_SAVED_EXP, REG_SAVED_VAL_OFFSET, REG_SAVED_VAL_EXP, REG_UNDEFINED): New anonymous enum, moved from inside of struct frame_state_reg_info. (struct frame_state_reg_info): Remove reg[].how element and the anonymous enum there. Add how element. * unwind-dw2.c: Include stddef.h. (uw_frame_state_for): Don't clear first offsetof (_Unwind_FrameState, regs.how[0]) bytes of *fs. (execute_cfa_program, __frame_state_for, uw_update_context_1, uw_update_context): Use fs->regs.how[X] instead of fs->regs.reg[X].how or fs.regs.how[X] instead of fs.regs.reg[X].how. * config/sh/linux-unwind.h (sh_fallback_frame_state): Likewise. * config/bfin/linux-unwind.h (bfin_fallback_frame_state): Likewise. * config/pa/linux-unwind.h (pa32_fallback_frame_state): Likewise. * config/pa/hpux-unwind.h (UPDATE_FS_FOR_SAR, UPDATE_FS_FOR_GR, UPDATE_FS_FOR_FR, UPDATE_FS_FOR_PC, pa_fallback_frame_state): Likewise. * config/alpha/vms-unwind.h (alpha_vms_fallback_frame_state): Likewise. * config/alpha/linux-unwind.h (alpha_fallback_frame_state): Likewise. * config/arc/linux-unwind.h (arc_fallback_frame_state, arc_frob_update_context): Likewise. * config/riscv/linux-unwind.h (riscv_fallback_frame_state): Likewise. * config/nios2/linux-unwind.h (NIOS2_REG): Likewise. * config/nds32/linux-unwind.h (NDS32_PUT_FS_REG): Likewise. * config/s390/tpf-unwind.h (s390_fallback_frame_state): Likewise. * config/s390/linux-unwind.h (s390_fallback_frame_state): Likewise. * config/sparc/sol2-unwind.h (sparc64_frob_update_context, MD_FALLBACK_FRAME_STATE_FOR): Likewise. * config/sparc/linux-unwind.h (sparc64_fallback_frame_state, sparc64_frob_update_context, sparc_fallback_frame_state): Likewise. * config/i386/sol2-unwind.h (x86_64_fallback_frame_state, x86_fallback_frame_state): Likewise. * config/i386/w32-unwind.h (i386_w32_fallback_frame_state): Likewise. * config/i386/linux-unwind.h (x86_64_fallback_frame_state, x86_fallback_frame_state): Likewise. * config/i386/freebsd-unwind.h (x86_64_freebsd_fallback_frame_state): Likewise. * config/i386/dragonfly-unwind.h (x86_64_dragonfly_fallback_frame_state): Likewise. * config/i386/gnu-unwind.h (x86_gnu_fallback_frame_state): Likewise. * config/csky/linux-unwind.h (csky_fallback_frame_state): Likewise. * config/aarch64/linux-unwind.h (aarch64_fallback_frame_state): Likewise. * config/aarch64/freebsd-unwind.h (aarch64_freebsd_fallback_frame_state): Likewise. * config/aarch64/aarch64-unwind.h (aarch64_frob_update_context): Likewise. * config/or1k/linux-unwind.h (or1k_fallback_frame_state): Likewise. * config/mips/linux-unwind.h (mips_fallback_frame_state): Likewise. * config/loongarch/linux-unwind.h (loongarch_fallback_frame_state): Likewise. * config/m68k/linux-unwind.h (m68k_fallback_frame_state): Likewise. * config/xtensa/linux-unwind.h (xtensa_fallback_frame_state): Likewise. * config/rs6000/darwin-fallback.c (set_offset): Likewise. * config/rs6000/aix-unwind.h (MD_FROB_UPDATE_CONTEXT): Likewise. * config/rs6000/linux-unwind.h (ppc_fallback_frame_state): Likewise. * config/rs6000/freebsd-unwind.h (frob_update_context): Likewise.
2022-10-06 11:04:52 +02:00
fs->regs.how[10] = REG_SAVED_OFFSET;
tm.texi (MD_UNWIND_SUPPORT): Document. * doc/tm.texi (MD_UNWIND_SUPPORT): Document. (MD_FALLBACK_FRAME_STATE_FOR): Update. * unwind-dw2.c (MD_UNWIND_SUPPORT): #include if defined. (uw_frame_state_for): Adjust MD_FALLBACK_FRAME_STATE_FOR invocation. (MD_FROB_UPDATE_CONTEXT): Remove default. (uw_update_context_1): Instead #ifdef invocation. * config/ia64/unwind-ia64.c (MD_UNWIND_SUPPORT): #include if defined. (uw_frame_state_for): Adjust MD_FALLBACK_FRAME_STATE_FOR invocation. * config/alpha/gnu.h (MD_FALLBACK_FRAME_STATE_FOR): Don't undef. (MD_UNWIND_SUPPORT): Undefine this instead. * config/i386/gnu.h: Likewise. * config/alpha/linux-unwind.h: New file, macro converted to function, extracted from.. * config/alpha/linux.h (MD_FALLBACK_FRAME_STATE_FOR): ..this. (MD_UNWIND_SUPPORT): Define. * config/alpha/vms-unwind.h, config/alpha/vms.h: Likewise. * config/i386/linux-unwind.h, config/i386/linux.h, config/i386/linux64.h: Likewise. * config/ia64/linux-unwind.h, config/ia64/linux.h: Likewise. MD_HANDLE_UNWABI too. * config/mips/linux-unwind.h, config/mips/linux.h: Likewise. * config/pa/linux-unwind.h, config/pa/pa32-linux.h: Likewise. * config/rs6000/darwin-unwind.h, config/rs6000/darwin.h: Likewise. * config/s390/linux-unwind.h, config/s390/linux.h: Likewise. * config/sparc/linux-unwind.h, config/sparc/linux.h, config/sparc/linux64.h: Likewise. * config/sh/linux-unwind.h, config/sh/linux.h: Likewise, but merge SH_FALLBACK_FRAME_FLOAT_STATE into sh_fallback_frame_state. * config/rs6000/linux-unwind.h, config/rs6000/linux.h, config/rs6000/linux64.h: Likewise. Split out get_sigcontext function. Use ARG_POINTER_REGNUM for 32-bit temp reg too. From-SVN: r87167
2004-09-08 00:17:19 +00:00
fs->regs.reg[10].loc.offset = (long)&sc->r10 - new_cfa;
libgcc: Decrease size of _Unwind_FrameState and even more size of cleared area in uw_frame_state_for The following patch implements something that has Florian found as low hanging fruit in our unwinder and has been discussed in the https://gcc.gnu.org/wiki/cauldron2022#cauldron2022talks.inprocess_unwinding_bof talk. _Unwind_FrameState type seems to be (unlike the pre-GCC 3 frame_state which has been part of ABI) private to unwind-dw2.c + unwind.inc it includes, it is always defined on the stack of some entrypoints, initialized by static uw_frame_state_for and the address of it is also passed to other static functions or the static inlines handling machine dependent unwinding, but it isn't fortunately passed to any callbacks or public functions, so I think we can safely change it any time we want. Florian mentioned that the structure is large even on x86_64, 384 bytes there, starts with 328 bytes long element with frame_state_reg_info type which then starts with an array with __LIBGCC_DWARF_FRAME_REGISTERS__ + 1 elements, each of them is 16 bytes long, on x86_64 __LIBGCC_DWARF_FRAME_REGISTERS__ is just 17 but even that is big, on say riscv __LIBGCC_DWARF_FRAME_REGISTERS__ is I think 128, on powerpc 111, on sh 153 etc. And, we memset to zero the whole fs variable with the _Unwind_FrameState type at the start of the unwinding. The reason why each element is 16 byte (on 64-bit arches) is that it contains some pointer or pointer sized integer and then an enum (with just 7 different enumerators) + padding. The following patch decreases it by moving the enum into a separate array and using just one byte for each register in that second array. We could compress it even more, say 4 bits per register, but I don't want to uglify the code for it too much and make the accesses slower. Furthermore, the clearing of the object can clear only thos how array and members after it, because REG_UNSAVED enumerator (0) doesn't actually need any pointer or pointer sized integer, it is just the other kinds that need to have there something. By doing this, on x86_64 the above numbers change to _Unwind_FrameState type being now 264 bytes long, frame_state_reg_info 208 bytes and we don't clear the first 144 bytes of the object, so the memset is 120 bytes, so ~ 31% of the old clearing size. On riscv 64-bit assuming it has same structure layout rules for the few types used there that would be ~ 2160 bytes of _Unwind_FrameState type before and ~ 1264 bytes after, with the memset previously ~ 2160 bytes and after ~ 232 bytes after. We've also talked about possibly adding a number of initially initialized regs and initializing the rest lazily, but at least for x86_64 with 18 elements in the array that doesn't seem to be worth it anymore, especially because return address column is 16 there and that is usually the first thing to be touched. It might theory help with lots of registers if they are usually untouched, but would uglify and complicate any stores to how by having to check there for the not initialized yet cases and lazy initialization, and similarly for all reads of how to do there if below last initialized one, use how, otherwise imply REG_UNSAVED. The disadvantage of the patch is that touching reg[x].loc and how[x] now means 2 cachelines rather than one as before, and I admit beyond bootstrap/regtest I haven't benchmarked it in any way. 2022-10-06 Jakub Jelinek <jakub@redhat.com> * unwind-dw2.h (REG_UNSAVED, REG_SAVED_OFFSET, REG_SAVED_REG, REG_SAVED_EXP, REG_SAVED_VAL_OFFSET, REG_SAVED_VAL_EXP, REG_UNDEFINED): New anonymous enum, moved from inside of struct frame_state_reg_info. (struct frame_state_reg_info): Remove reg[].how element and the anonymous enum there. Add how element. * unwind-dw2.c: Include stddef.h. (uw_frame_state_for): Don't clear first offsetof (_Unwind_FrameState, regs.how[0]) bytes of *fs. (execute_cfa_program, __frame_state_for, uw_update_context_1, uw_update_context): Use fs->regs.how[X] instead of fs->regs.reg[X].how or fs.regs.how[X] instead of fs.regs.reg[X].how. * config/sh/linux-unwind.h (sh_fallback_frame_state): Likewise. * config/bfin/linux-unwind.h (bfin_fallback_frame_state): Likewise. * config/pa/linux-unwind.h (pa32_fallback_frame_state): Likewise. * config/pa/hpux-unwind.h (UPDATE_FS_FOR_SAR, UPDATE_FS_FOR_GR, UPDATE_FS_FOR_FR, UPDATE_FS_FOR_PC, pa_fallback_frame_state): Likewise. * config/alpha/vms-unwind.h (alpha_vms_fallback_frame_state): Likewise. * config/alpha/linux-unwind.h (alpha_fallback_frame_state): Likewise. * config/arc/linux-unwind.h (arc_fallback_frame_state, arc_frob_update_context): Likewise. * config/riscv/linux-unwind.h (riscv_fallback_frame_state): Likewise. * config/nios2/linux-unwind.h (NIOS2_REG): Likewise. * config/nds32/linux-unwind.h (NDS32_PUT_FS_REG): Likewise. * config/s390/tpf-unwind.h (s390_fallback_frame_state): Likewise. * config/s390/linux-unwind.h (s390_fallback_frame_state): Likewise. * config/sparc/sol2-unwind.h (sparc64_frob_update_context, MD_FALLBACK_FRAME_STATE_FOR): Likewise. * config/sparc/linux-unwind.h (sparc64_fallback_frame_state, sparc64_frob_update_context, sparc_fallback_frame_state): Likewise. * config/i386/sol2-unwind.h (x86_64_fallback_frame_state, x86_fallback_frame_state): Likewise. * config/i386/w32-unwind.h (i386_w32_fallback_frame_state): Likewise. * config/i386/linux-unwind.h (x86_64_fallback_frame_state, x86_fallback_frame_state): Likewise. * config/i386/freebsd-unwind.h (x86_64_freebsd_fallback_frame_state): Likewise. * config/i386/dragonfly-unwind.h (x86_64_dragonfly_fallback_frame_state): Likewise. * config/i386/gnu-unwind.h (x86_gnu_fallback_frame_state): Likewise. * config/csky/linux-unwind.h (csky_fallback_frame_state): Likewise. * config/aarch64/linux-unwind.h (aarch64_fallback_frame_state): Likewise. * config/aarch64/freebsd-unwind.h (aarch64_freebsd_fallback_frame_state): Likewise. * config/aarch64/aarch64-unwind.h (aarch64_frob_update_context): Likewise. * config/or1k/linux-unwind.h (or1k_fallback_frame_state): Likewise. * config/mips/linux-unwind.h (mips_fallback_frame_state): Likewise. * config/loongarch/linux-unwind.h (loongarch_fallback_frame_state): Likewise. * config/m68k/linux-unwind.h (m68k_fallback_frame_state): Likewise. * config/xtensa/linux-unwind.h (xtensa_fallback_frame_state): Likewise. * config/rs6000/darwin-fallback.c (set_offset): Likewise. * config/rs6000/aix-unwind.h (MD_FROB_UPDATE_CONTEXT): Likewise. * config/rs6000/linux-unwind.h (ppc_fallback_frame_state): Likewise. * config/rs6000/freebsd-unwind.h (frob_update_context): Likewise.
2022-10-06 11:04:52 +02:00
fs->regs.how[11] = REG_SAVED_OFFSET;
tm.texi (MD_UNWIND_SUPPORT): Document. * doc/tm.texi (MD_UNWIND_SUPPORT): Document. (MD_FALLBACK_FRAME_STATE_FOR): Update. * unwind-dw2.c (MD_UNWIND_SUPPORT): #include if defined. (uw_frame_state_for): Adjust MD_FALLBACK_FRAME_STATE_FOR invocation. (MD_FROB_UPDATE_CONTEXT): Remove default. (uw_update_context_1): Instead #ifdef invocation. * config/ia64/unwind-ia64.c (MD_UNWIND_SUPPORT): #include if defined. (uw_frame_state_for): Adjust MD_FALLBACK_FRAME_STATE_FOR invocation. * config/alpha/gnu.h (MD_FALLBACK_FRAME_STATE_FOR): Don't undef. (MD_UNWIND_SUPPORT): Undefine this instead. * config/i386/gnu.h: Likewise. * config/alpha/linux-unwind.h: New file, macro converted to function, extracted from.. * config/alpha/linux.h (MD_FALLBACK_FRAME_STATE_FOR): ..this. (MD_UNWIND_SUPPORT): Define. * config/alpha/vms-unwind.h, config/alpha/vms.h: Likewise. * config/i386/linux-unwind.h, config/i386/linux.h, config/i386/linux64.h: Likewise. * config/ia64/linux-unwind.h, config/ia64/linux.h: Likewise. MD_HANDLE_UNWABI too. * config/mips/linux-unwind.h, config/mips/linux.h: Likewise. * config/pa/linux-unwind.h, config/pa/pa32-linux.h: Likewise. * config/rs6000/darwin-unwind.h, config/rs6000/darwin.h: Likewise. * config/s390/linux-unwind.h, config/s390/linux.h: Likewise. * config/sparc/linux-unwind.h, config/sparc/linux.h, config/sparc/linux64.h: Likewise. * config/sh/linux-unwind.h, config/sh/linux.h: Likewise, but merge SH_FALLBACK_FRAME_FLOAT_STATE into sh_fallback_frame_state. * config/rs6000/linux-unwind.h, config/rs6000/linux.h, config/rs6000/linux64.h: Likewise. Split out get_sigcontext function. Use ARG_POINTER_REGNUM for 32-bit temp reg too. From-SVN: r87167
2004-09-08 00:17:19 +00:00
fs->regs.reg[11].loc.offset = (long)&sc->r11 - new_cfa;
libgcc: Decrease size of _Unwind_FrameState and even more size of cleared area in uw_frame_state_for The following patch implements something that has Florian found as low hanging fruit in our unwinder and has been discussed in the https://gcc.gnu.org/wiki/cauldron2022#cauldron2022talks.inprocess_unwinding_bof talk. _Unwind_FrameState type seems to be (unlike the pre-GCC 3 frame_state which has been part of ABI) private to unwind-dw2.c + unwind.inc it includes, it is always defined on the stack of some entrypoints, initialized by static uw_frame_state_for and the address of it is also passed to other static functions or the static inlines handling machine dependent unwinding, but it isn't fortunately passed to any callbacks or public functions, so I think we can safely change it any time we want. Florian mentioned that the structure is large even on x86_64, 384 bytes there, starts with 328 bytes long element with frame_state_reg_info type which then starts with an array with __LIBGCC_DWARF_FRAME_REGISTERS__ + 1 elements, each of them is 16 bytes long, on x86_64 __LIBGCC_DWARF_FRAME_REGISTERS__ is just 17 but even that is big, on say riscv __LIBGCC_DWARF_FRAME_REGISTERS__ is I think 128, on powerpc 111, on sh 153 etc. And, we memset to zero the whole fs variable with the _Unwind_FrameState type at the start of the unwinding. The reason why each element is 16 byte (on 64-bit arches) is that it contains some pointer or pointer sized integer and then an enum (with just 7 different enumerators) + padding. The following patch decreases it by moving the enum into a separate array and using just one byte for each register in that second array. We could compress it even more, say 4 bits per register, but I don't want to uglify the code for it too much and make the accesses slower. Furthermore, the clearing of the object can clear only thos how array and members after it, because REG_UNSAVED enumerator (0) doesn't actually need any pointer or pointer sized integer, it is just the other kinds that need to have there something. By doing this, on x86_64 the above numbers change to _Unwind_FrameState type being now 264 bytes long, frame_state_reg_info 208 bytes and we don't clear the first 144 bytes of the object, so the memset is 120 bytes, so ~ 31% of the old clearing size. On riscv 64-bit assuming it has same structure layout rules for the few types used there that would be ~ 2160 bytes of _Unwind_FrameState type before and ~ 1264 bytes after, with the memset previously ~ 2160 bytes and after ~ 232 bytes after. We've also talked about possibly adding a number of initially initialized regs and initializing the rest lazily, but at least for x86_64 with 18 elements in the array that doesn't seem to be worth it anymore, especially because return address column is 16 there and that is usually the first thing to be touched. It might theory help with lots of registers if they are usually untouched, but would uglify and complicate any stores to how by having to check there for the not initialized yet cases and lazy initialization, and similarly for all reads of how to do there if below last initialized one, use how, otherwise imply REG_UNSAVED. The disadvantage of the patch is that touching reg[x].loc and how[x] now means 2 cachelines rather than one as before, and I admit beyond bootstrap/regtest I haven't benchmarked it in any way. 2022-10-06 Jakub Jelinek <jakub@redhat.com> * unwind-dw2.h (REG_UNSAVED, REG_SAVED_OFFSET, REG_SAVED_REG, REG_SAVED_EXP, REG_SAVED_VAL_OFFSET, REG_SAVED_VAL_EXP, REG_UNDEFINED): New anonymous enum, moved from inside of struct frame_state_reg_info. (struct frame_state_reg_info): Remove reg[].how element and the anonymous enum there. Add how element. * unwind-dw2.c: Include stddef.h. (uw_frame_state_for): Don't clear first offsetof (_Unwind_FrameState, regs.how[0]) bytes of *fs. (execute_cfa_program, __frame_state_for, uw_update_context_1, uw_update_context): Use fs->regs.how[X] instead of fs->regs.reg[X].how or fs.regs.how[X] instead of fs.regs.reg[X].how. * config/sh/linux-unwind.h (sh_fallback_frame_state): Likewise. * config/bfin/linux-unwind.h (bfin_fallback_frame_state): Likewise. * config/pa/linux-unwind.h (pa32_fallback_frame_state): Likewise. * config/pa/hpux-unwind.h (UPDATE_FS_FOR_SAR, UPDATE_FS_FOR_GR, UPDATE_FS_FOR_FR, UPDATE_FS_FOR_PC, pa_fallback_frame_state): Likewise. * config/alpha/vms-unwind.h (alpha_vms_fallback_frame_state): Likewise. * config/alpha/linux-unwind.h (alpha_fallback_frame_state): Likewise. * config/arc/linux-unwind.h (arc_fallback_frame_state, arc_frob_update_context): Likewise. * config/riscv/linux-unwind.h (riscv_fallback_frame_state): Likewise. * config/nios2/linux-unwind.h (NIOS2_REG): Likewise. * config/nds32/linux-unwind.h (NDS32_PUT_FS_REG): Likewise. * config/s390/tpf-unwind.h (s390_fallback_frame_state): Likewise. * config/s390/linux-unwind.h (s390_fallback_frame_state): Likewise. * config/sparc/sol2-unwind.h (sparc64_frob_update_context, MD_FALLBACK_FRAME_STATE_FOR): Likewise. * config/sparc/linux-unwind.h (sparc64_fallback_frame_state, sparc64_frob_update_context, sparc_fallback_frame_state): Likewise. * config/i386/sol2-unwind.h (x86_64_fallback_frame_state, x86_fallback_frame_state): Likewise. * config/i386/w32-unwind.h (i386_w32_fallback_frame_state): Likewise. * config/i386/linux-unwind.h (x86_64_fallback_frame_state, x86_fallback_frame_state): Likewise. * config/i386/freebsd-unwind.h (x86_64_freebsd_fallback_frame_state): Likewise. * config/i386/dragonfly-unwind.h (x86_64_dragonfly_fallback_frame_state): Likewise. * config/i386/gnu-unwind.h (x86_gnu_fallback_frame_state): Likewise. * config/csky/linux-unwind.h (csky_fallback_frame_state): Likewise. * config/aarch64/linux-unwind.h (aarch64_fallback_frame_state): Likewise. * config/aarch64/freebsd-unwind.h (aarch64_freebsd_fallback_frame_state): Likewise. * config/aarch64/aarch64-unwind.h (aarch64_frob_update_context): Likewise. * config/or1k/linux-unwind.h (or1k_fallback_frame_state): Likewise. * config/mips/linux-unwind.h (mips_fallback_frame_state): Likewise. * config/loongarch/linux-unwind.h (loongarch_fallback_frame_state): Likewise. * config/m68k/linux-unwind.h (m68k_fallback_frame_state): Likewise. * config/xtensa/linux-unwind.h (xtensa_fallback_frame_state): Likewise. * config/rs6000/darwin-fallback.c (set_offset): Likewise. * config/rs6000/aix-unwind.h (MD_FROB_UPDATE_CONTEXT): Likewise. * config/rs6000/linux-unwind.h (ppc_fallback_frame_state): Likewise. * config/rs6000/freebsd-unwind.h (frob_update_context): Likewise.
2022-10-06 11:04:52 +02:00
fs->regs.how[12] = REG_SAVED_OFFSET;
tm.texi (MD_UNWIND_SUPPORT): Document. * doc/tm.texi (MD_UNWIND_SUPPORT): Document. (MD_FALLBACK_FRAME_STATE_FOR): Update. * unwind-dw2.c (MD_UNWIND_SUPPORT): #include if defined. (uw_frame_state_for): Adjust MD_FALLBACK_FRAME_STATE_FOR invocation. (MD_FROB_UPDATE_CONTEXT): Remove default. (uw_update_context_1): Instead #ifdef invocation. * config/ia64/unwind-ia64.c (MD_UNWIND_SUPPORT): #include if defined. (uw_frame_state_for): Adjust MD_FALLBACK_FRAME_STATE_FOR invocation. * config/alpha/gnu.h (MD_FALLBACK_FRAME_STATE_FOR): Don't undef. (MD_UNWIND_SUPPORT): Undefine this instead. * config/i386/gnu.h: Likewise. * config/alpha/linux-unwind.h: New file, macro converted to function, extracted from.. * config/alpha/linux.h (MD_FALLBACK_FRAME_STATE_FOR): ..this. (MD_UNWIND_SUPPORT): Define. * config/alpha/vms-unwind.h, config/alpha/vms.h: Likewise. * config/i386/linux-unwind.h, config/i386/linux.h, config/i386/linux64.h: Likewise. * config/ia64/linux-unwind.h, config/ia64/linux.h: Likewise. MD_HANDLE_UNWABI too. * config/mips/linux-unwind.h, config/mips/linux.h: Likewise. * config/pa/linux-unwind.h, config/pa/pa32-linux.h: Likewise. * config/rs6000/darwin-unwind.h, config/rs6000/darwin.h: Likewise. * config/s390/linux-unwind.h, config/s390/linux.h: Likewise. * config/sparc/linux-unwind.h, config/sparc/linux.h, config/sparc/linux64.h: Likewise. * config/sh/linux-unwind.h, config/sh/linux.h: Likewise, but merge SH_FALLBACK_FRAME_FLOAT_STATE into sh_fallback_frame_state. * config/rs6000/linux-unwind.h, config/rs6000/linux.h, config/rs6000/linux64.h: Likewise. Split out get_sigcontext function. Use ARG_POINTER_REGNUM for 32-bit temp reg too. From-SVN: r87167
2004-09-08 00:17:19 +00:00
fs->regs.reg[12].loc.offset = (long)&sc->r12 - new_cfa;
libgcc: Decrease size of _Unwind_FrameState and even more size of cleared area in uw_frame_state_for The following patch implements something that has Florian found as low hanging fruit in our unwinder and has been discussed in the https://gcc.gnu.org/wiki/cauldron2022#cauldron2022talks.inprocess_unwinding_bof talk. _Unwind_FrameState type seems to be (unlike the pre-GCC 3 frame_state which has been part of ABI) private to unwind-dw2.c + unwind.inc it includes, it is always defined on the stack of some entrypoints, initialized by static uw_frame_state_for and the address of it is also passed to other static functions or the static inlines handling machine dependent unwinding, but it isn't fortunately passed to any callbacks or public functions, so I think we can safely change it any time we want. Florian mentioned that the structure is large even on x86_64, 384 bytes there, starts with 328 bytes long element with frame_state_reg_info type which then starts with an array with __LIBGCC_DWARF_FRAME_REGISTERS__ + 1 elements, each of them is 16 bytes long, on x86_64 __LIBGCC_DWARF_FRAME_REGISTERS__ is just 17 but even that is big, on say riscv __LIBGCC_DWARF_FRAME_REGISTERS__ is I think 128, on powerpc 111, on sh 153 etc. And, we memset to zero the whole fs variable with the _Unwind_FrameState type at the start of the unwinding. The reason why each element is 16 byte (on 64-bit arches) is that it contains some pointer or pointer sized integer and then an enum (with just 7 different enumerators) + padding. The following patch decreases it by moving the enum into a separate array and using just one byte for each register in that second array. We could compress it even more, say 4 bits per register, but I don't want to uglify the code for it too much and make the accesses slower. Furthermore, the clearing of the object can clear only thos how array and members after it, because REG_UNSAVED enumerator (0) doesn't actually need any pointer or pointer sized integer, it is just the other kinds that need to have there something. By doing this, on x86_64 the above numbers change to _Unwind_FrameState type being now 264 bytes long, frame_state_reg_info 208 bytes and we don't clear the first 144 bytes of the object, so the memset is 120 bytes, so ~ 31% of the old clearing size. On riscv 64-bit assuming it has same structure layout rules for the few types used there that would be ~ 2160 bytes of _Unwind_FrameState type before and ~ 1264 bytes after, with the memset previously ~ 2160 bytes and after ~ 232 bytes after. We've also talked about possibly adding a number of initially initialized regs and initializing the rest lazily, but at least for x86_64 with 18 elements in the array that doesn't seem to be worth it anymore, especially because return address column is 16 there and that is usually the first thing to be touched. It might theory help with lots of registers if they are usually untouched, but would uglify and complicate any stores to how by having to check there for the not initialized yet cases and lazy initialization, and similarly for all reads of how to do there if below last initialized one, use how, otherwise imply REG_UNSAVED. The disadvantage of the patch is that touching reg[x].loc and how[x] now means 2 cachelines rather than one as before, and I admit beyond bootstrap/regtest I haven't benchmarked it in any way. 2022-10-06 Jakub Jelinek <jakub@redhat.com> * unwind-dw2.h (REG_UNSAVED, REG_SAVED_OFFSET, REG_SAVED_REG, REG_SAVED_EXP, REG_SAVED_VAL_OFFSET, REG_SAVED_VAL_EXP, REG_UNDEFINED): New anonymous enum, moved from inside of struct frame_state_reg_info. (struct frame_state_reg_info): Remove reg[].how element and the anonymous enum there. Add how element. * unwind-dw2.c: Include stddef.h. (uw_frame_state_for): Don't clear first offsetof (_Unwind_FrameState, regs.how[0]) bytes of *fs. (execute_cfa_program, __frame_state_for, uw_update_context_1, uw_update_context): Use fs->regs.how[X] instead of fs->regs.reg[X].how or fs.regs.how[X] instead of fs.regs.reg[X].how. * config/sh/linux-unwind.h (sh_fallback_frame_state): Likewise. * config/bfin/linux-unwind.h (bfin_fallback_frame_state): Likewise. * config/pa/linux-unwind.h (pa32_fallback_frame_state): Likewise. * config/pa/hpux-unwind.h (UPDATE_FS_FOR_SAR, UPDATE_FS_FOR_GR, UPDATE_FS_FOR_FR, UPDATE_FS_FOR_PC, pa_fallback_frame_state): Likewise. * config/alpha/vms-unwind.h (alpha_vms_fallback_frame_state): Likewise. * config/alpha/linux-unwind.h (alpha_fallback_frame_state): Likewise. * config/arc/linux-unwind.h (arc_fallback_frame_state, arc_frob_update_context): Likewise. * config/riscv/linux-unwind.h (riscv_fallback_frame_state): Likewise. * config/nios2/linux-unwind.h (NIOS2_REG): Likewise. * config/nds32/linux-unwind.h (NDS32_PUT_FS_REG): Likewise. * config/s390/tpf-unwind.h (s390_fallback_frame_state): Likewise. * config/s390/linux-unwind.h (s390_fallback_frame_state): Likewise. * config/sparc/sol2-unwind.h (sparc64_frob_update_context, MD_FALLBACK_FRAME_STATE_FOR): Likewise. * config/sparc/linux-unwind.h (sparc64_fallback_frame_state, sparc64_frob_update_context, sparc_fallback_frame_state): Likewise. * config/i386/sol2-unwind.h (x86_64_fallback_frame_state, x86_fallback_frame_state): Likewise. * config/i386/w32-unwind.h (i386_w32_fallback_frame_state): Likewise. * config/i386/linux-unwind.h (x86_64_fallback_frame_state, x86_fallback_frame_state): Likewise. * config/i386/freebsd-unwind.h (x86_64_freebsd_fallback_frame_state): Likewise. * config/i386/dragonfly-unwind.h (x86_64_dragonfly_fallback_frame_state): Likewise. * config/i386/gnu-unwind.h (x86_gnu_fallback_frame_state): Likewise. * config/csky/linux-unwind.h (csky_fallback_frame_state): Likewise. * config/aarch64/linux-unwind.h (aarch64_fallback_frame_state): Likewise. * config/aarch64/freebsd-unwind.h (aarch64_freebsd_fallback_frame_state): Likewise. * config/aarch64/aarch64-unwind.h (aarch64_frob_update_context): Likewise. * config/or1k/linux-unwind.h (or1k_fallback_frame_state): Likewise. * config/mips/linux-unwind.h (mips_fallback_frame_state): Likewise. * config/loongarch/linux-unwind.h (loongarch_fallback_frame_state): Likewise. * config/m68k/linux-unwind.h (m68k_fallback_frame_state): Likewise. * config/xtensa/linux-unwind.h (xtensa_fallback_frame_state): Likewise. * config/rs6000/darwin-fallback.c (set_offset): Likewise. * config/rs6000/aix-unwind.h (MD_FROB_UPDATE_CONTEXT): Likewise. * config/rs6000/linux-unwind.h (ppc_fallback_frame_state): Likewise. * config/rs6000/freebsd-unwind.h (frob_update_context): Likewise.
2022-10-06 11:04:52 +02:00
fs->regs.how[13] = REG_SAVED_OFFSET;
tm.texi (MD_UNWIND_SUPPORT): Document. * doc/tm.texi (MD_UNWIND_SUPPORT): Document. (MD_FALLBACK_FRAME_STATE_FOR): Update. * unwind-dw2.c (MD_UNWIND_SUPPORT): #include if defined. (uw_frame_state_for): Adjust MD_FALLBACK_FRAME_STATE_FOR invocation. (MD_FROB_UPDATE_CONTEXT): Remove default. (uw_update_context_1): Instead #ifdef invocation. * config/ia64/unwind-ia64.c (MD_UNWIND_SUPPORT): #include if defined. (uw_frame_state_for): Adjust MD_FALLBACK_FRAME_STATE_FOR invocation. * config/alpha/gnu.h (MD_FALLBACK_FRAME_STATE_FOR): Don't undef. (MD_UNWIND_SUPPORT): Undefine this instead. * config/i386/gnu.h: Likewise. * config/alpha/linux-unwind.h: New file, macro converted to function, extracted from.. * config/alpha/linux.h (MD_FALLBACK_FRAME_STATE_FOR): ..this. (MD_UNWIND_SUPPORT): Define. * config/alpha/vms-unwind.h, config/alpha/vms.h: Likewise. * config/i386/linux-unwind.h, config/i386/linux.h, config/i386/linux64.h: Likewise. * config/ia64/linux-unwind.h, config/ia64/linux.h: Likewise. MD_HANDLE_UNWABI too. * config/mips/linux-unwind.h, config/mips/linux.h: Likewise. * config/pa/linux-unwind.h, config/pa/pa32-linux.h: Likewise. * config/rs6000/darwin-unwind.h, config/rs6000/darwin.h: Likewise. * config/s390/linux-unwind.h, config/s390/linux.h: Likewise. * config/sparc/linux-unwind.h, config/sparc/linux.h, config/sparc/linux64.h: Likewise. * config/sh/linux-unwind.h, config/sh/linux.h: Likewise, but merge SH_FALLBACK_FRAME_FLOAT_STATE into sh_fallback_frame_state. * config/rs6000/linux-unwind.h, config/rs6000/linux.h, config/rs6000/linux64.h: Likewise. Split out get_sigcontext function. Use ARG_POINTER_REGNUM for 32-bit temp reg too. From-SVN: r87167
2004-09-08 00:17:19 +00:00
fs->regs.reg[13].loc.offset = (long)&sc->r13 - new_cfa;
libgcc: Decrease size of _Unwind_FrameState and even more size of cleared area in uw_frame_state_for The following patch implements something that has Florian found as low hanging fruit in our unwinder and has been discussed in the https://gcc.gnu.org/wiki/cauldron2022#cauldron2022talks.inprocess_unwinding_bof talk. _Unwind_FrameState type seems to be (unlike the pre-GCC 3 frame_state which has been part of ABI) private to unwind-dw2.c + unwind.inc it includes, it is always defined on the stack of some entrypoints, initialized by static uw_frame_state_for and the address of it is also passed to other static functions or the static inlines handling machine dependent unwinding, but it isn't fortunately passed to any callbacks or public functions, so I think we can safely change it any time we want. Florian mentioned that the structure is large even on x86_64, 384 bytes there, starts with 328 bytes long element with frame_state_reg_info type which then starts with an array with __LIBGCC_DWARF_FRAME_REGISTERS__ + 1 elements, each of them is 16 bytes long, on x86_64 __LIBGCC_DWARF_FRAME_REGISTERS__ is just 17 but even that is big, on say riscv __LIBGCC_DWARF_FRAME_REGISTERS__ is I think 128, on powerpc 111, on sh 153 etc. And, we memset to zero the whole fs variable with the _Unwind_FrameState type at the start of the unwinding. The reason why each element is 16 byte (on 64-bit arches) is that it contains some pointer or pointer sized integer and then an enum (with just 7 different enumerators) + padding. The following patch decreases it by moving the enum into a separate array and using just one byte for each register in that second array. We could compress it even more, say 4 bits per register, but I don't want to uglify the code for it too much and make the accesses slower. Furthermore, the clearing of the object can clear only thos how array and members after it, because REG_UNSAVED enumerator (0) doesn't actually need any pointer or pointer sized integer, it is just the other kinds that need to have there something. By doing this, on x86_64 the above numbers change to _Unwind_FrameState type being now 264 bytes long, frame_state_reg_info 208 bytes and we don't clear the first 144 bytes of the object, so the memset is 120 bytes, so ~ 31% of the old clearing size. On riscv 64-bit assuming it has same structure layout rules for the few types used there that would be ~ 2160 bytes of _Unwind_FrameState type before and ~ 1264 bytes after, with the memset previously ~ 2160 bytes and after ~ 232 bytes after. We've also talked about possibly adding a number of initially initialized regs and initializing the rest lazily, but at least for x86_64 with 18 elements in the array that doesn't seem to be worth it anymore, especially because return address column is 16 there and that is usually the first thing to be touched. It might theory help with lots of registers if they are usually untouched, but would uglify and complicate any stores to how by having to check there for the not initialized yet cases and lazy initialization, and similarly for all reads of how to do there if below last initialized one, use how, otherwise imply REG_UNSAVED. The disadvantage of the patch is that touching reg[x].loc and how[x] now means 2 cachelines rather than one as before, and I admit beyond bootstrap/regtest I haven't benchmarked it in any way. 2022-10-06 Jakub Jelinek <jakub@redhat.com> * unwind-dw2.h (REG_UNSAVED, REG_SAVED_OFFSET, REG_SAVED_REG, REG_SAVED_EXP, REG_SAVED_VAL_OFFSET, REG_SAVED_VAL_EXP, REG_UNDEFINED): New anonymous enum, moved from inside of struct frame_state_reg_info. (struct frame_state_reg_info): Remove reg[].how element and the anonymous enum there. Add how element. * unwind-dw2.c: Include stddef.h. (uw_frame_state_for): Don't clear first offsetof (_Unwind_FrameState, regs.how[0]) bytes of *fs. (execute_cfa_program, __frame_state_for, uw_update_context_1, uw_update_context): Use fs->regs.how[X] instead of fs->regs.reg[X].how or fs.regs.how[X] instead of fs.regs.reg[X].how. * config/sh/linux-unwind.h (sh_fallback_frame_state): Likewise. * config/bfin/linux-unwind.h (bfin_fallback_frame_state): Likewise. * config/pa/linux-unwind.h (pa32_fallback_frame_state): Likewise. * config/pa/hpux-unwind.h (UPDATE_FS_FOR_SAR, UPDATE_FS_FOR_GR, UPDATE_FS_FOR_FR, UPDATE_FS_FOR_PC, pa_fallback_frame_state): Likewise. * config/alpha/vms-unwind.h (alpha_vms_fallback_frame_state): Likewise. * config/alpha/linux-unwind.h (alpha_fallback_frame_state): Likewise. * config/arc/linux-unwind.h (arc_fallback_frame_state, arc_frob_update_context): Likewise. * config/riscv/linux-unwind.h (riscv_fallback_frame_state): Likewise. * config/nios2/linux-unwind.h (NIOS2_REG): Likewise. * config/nds32/linux-unwind.h (NDS32_PUT_FS_REG): Likewise. * config/s390/tpf-unwind.h (s390_fallback_frame_state): Likewise. * config/s390/linux-unwind.h (s390_fallback_frame_state): Likewise. * config/sparc/sol2-unwind.h (sparc64_frob_update_context, MD_FALLBACK_FRAME_STATE_FOR): Likewise. * config/sparc/linux-unwind.h (sparc64_fallback_frame_state, sparc64_frob_update_context, sparc_fallback_frame_state): Likewise. * config/i386/sol2-unwind.h (x86_64_fallback_frame_state, x86_fallback_frame_state): Likewise. * config/i386/w32-unwind.h (i386_w32_fallback_frame_state): Likewise. * config/i386/linux-unwind.h (x86_64_fallback_frame_state, x86_fallback_frame_state): Likewise. * config/i386/freebsd-unwind.h (x86_64_freebsd_fallback_frame_state): Likewise. * config/i386/dragonfly-unwind.h (x86_64_dragonfly_fallback_frame_state): Likewise. * config/i386/gnu-unwind.h (x86_gnu_fallback_frame_state): Likewise. * config/csky/linux-unwind.h (csky_fallback_frame_state): Likewise. * config/aarch64/linux-unwind.h (aarch64_fallback_frame_state): Likewise. * config/aarch64/freebsd-unwind.h (aarch64_freebsd_fallback_frame_state): Likewise. * config/aarch64/aarch64-unwind.h (aarch64_frob_update_context): Likewise. * config/or1k/linux-unwind.h (or1k_fallback_frame_state): Likewise. * config/mips/linux-unwind.h (mips_fallback_frame_state): Likewise. * config/loongarch/linux-unwind.h (loongarch_fallback_frame_state): Likewise. * config/m68k/linux-unwind.h (m68k_fallback_frame_state): Likewise. * config/xtensa/linux-unwind.h (xtensa_fallback_frame_state): Likewise. * config/rs6000/darwin-fallback.c (set_offset): Likewise. * config/rs6000/aix-unwind.h (MD_FROB_UPDATE_CONTEXT): Likewise. * config/rs6000/linux-unwind.h (ppc_fallback_frame_state): Likewise. * config/rs6000/freebsd-unwind.h (frob_update_context): Likewise.
2022-10-06 11:04:52 +02:00
fs->regs.how[14] = REG_SAVED_OFFSET;
tm.texi (MD_UNWIND_SUPPORT): Document. * doc/tm.texi (MD_UNWIND_SUPPORT): Document. (MD_FALLBACK_FRAME_STATE_FOR): Update. * unwind-dw2.c (MD_UNWIND_SUPPORT): #include if defined. (uw_frame_state_for): Adjust MD_FALLBACK_FRAME_STATE_FOR invocation. (MD_FROB_UPDATE_CONTEXT): Remove default. (uw_update_context_1): Instead #ifdef invocation. * config/ia64/unwind-ia64.c (MD_UNWIND_SUPPORT): #include if defined. (uw_frame_state_for): Adjust MD_FALLBACK_FRAME_STATE_FOR invocation. * config/alpha/gnu.h (MD_FALLBACK_FRAME_STATE_FOR): Don't undef. (MD_UNWIND_SUPPORT): Undefine this instead. * config/i386/gnu.h: Likewise. * config/alpha/linux-unwind.h: New file, macro converted to function, extracted from.. * config/alpha/linux.h (MD_FALLBACK_FRAME_STATE_FOR): ..this. (MD_UNWIND_SUPPORT): Define. * config/alpha/vms-unwind.h, config/alpha/vms.h: Likewise. * config/i386/linux-unwind.h, config/i386/linux.h, config/i386/linux64.h: Likewise. * config/ia64/linux-unwind.h, config/ia64/linux.h: Likewise. MD_HANDLE_UNWABI too. * config/mips/linux-unwind.h, config/mips/linux.h: Likewise. * config/pa/linux-unwind.h, config/pa/pa32-linux.h: Likewise. * config/rs6000/darwin-unwind.h, config/rs6000/darwin.h: Likewise. * config/s390/linux-unwind.h, config/s390/linux.h: Likewise. * config/sparc/linux-unwind.h, config/sparc/linux.h, config/sparc/linux64.h: Likewise. * config/sh/linux-unwind.h, config/sh/linux.h: Likewise, but merge SH_FALLBACK_FRAME_FLOAT_STATE into sh_fallback_frame_state. * config/rs6000/linux-unwind.h, config/rs6000/linux.h, config/rs6000/linux64.h: Likewise. Split out get_sigcontext function. Use ARG_POINTER_REGNUM for 32-bit temp reg too. From-SVN: r87167
2004-09-08 00:17:19 +00:00
fs->regs.reg[14].loc.offset = (long)&sc->r14 - new_cfa;
libgcc: Decrease size of _Unwind_FrameState and even more size of cleared area in uw_frame_state_for The following patch implements something that has Florian found as low hanging fruit in our unwinder and has been discussed in the https://gcc.gnu.org/wiki/cauldron2022#cauldron2022talks.inprocess_unwinding_bof talk. _Unwind_FrameState type seems to be (unlike the pre-GCC 3 frame_state which has been part of ABI) private to unwind-dw2.c + unwind.inc it includes, it is always defined on the stack of some entrypoints, initialized by static uw_frame_state_for and the address of it is also passed to other static functions or the static inlines handling machine dependent unwinding, but it isn't fortunately passed to any callbacks or public functions, so I think we can safely change it any time we want. Florian mentioned that the structure is large even on x86_64, 384 bytes there, starts with 328 bytes long element with frame_state_reg_info type which then starts with an array with __LIBGCC_DWARF_FRAME_REGISTERS__ + 1 elements, each of them is 16 bytes long, on x86_64 __LIBGCC_DWARF_FRAME_REGISTERS__ is just 17 but even that is big, on say riscv __LIBGCC_DWARF_FRAME_REGISTERS__ is I think 128, on powerpc 111, on sh 153 etc. And, we memset to zero the whole fs variable with the _Unwind_FrameState type at the start of the unwinding. The reason why each element is 16 byte (on 64-bit arches) is that it contains some pointer or pointer sized integer and then an enum (with just 7 different enumerators) + padding. The following patch decreases it by moving the enum into a separate array and using just one byte for each register in that second array. We could compress it even more, say 4 bits per register, but I don't want to uglify the code for it too much and make the accesses slower. Furthermore, the clearing of the object can clear only thos how array and members after it, because REG_UNSAVED enumerator (0) doesn't actually need any pointer or pointer sized integer, it is just the other kinds that need to have there something. By doing this, on x86_64 the above numbers change to _Unwind_FrameState type being now 264 bytes long, frame_state_reg_info 208 bytes and we don't clear the first 144 bytes of the object, so the memset is 120 bytes, so ~ 31% of the old clearing size. On riscv 64-bit assuming it has same structure layout rules for the few types used there that would be ~ 2160 bytes of _Unwind_FrameState type before and ~ 1264 bytes after, with the memset previously ~ 2160 bytes and after ~ 232 bytes after. We've also talked about possibly adding a number of initially initialized regs and initializing the rest lazily, but at least for x86_64 with 18 elements in the array that doesn't seem to be worth it anymore, especially because return address column is 16 there and that is usually the first thing to be touched. It might theory help with lots of registers if they are usually untouched, but would uglify and complicate any stores to how by having to check there for the not initialized yet cases and lazy initialization, and similarly for all reads of how to do there if below last initialized one, use how, otherwise imply REG_UNSAVED. The disadvantage of the patch is that touching reg[x].loc and how[x] now means 2 cachelines rather than one as before, and I admit beyond bootstrap/regtest I haven't benchmarked it in any way. 2022-10-06 Jakub Jelinek <jakub@redhat.com> * unwind-dw2.h (REG_UNSAVED, REG_SAVED_OFFSET, REG_SAVED_REG, REG_SAVED_EXP, REG_SAVED_VAL_OFFSET, REG_SAVED_VAL_EXP, REG_UNDEFINED): New anonymous enum, moved from inside of struct frame_state_reg_info. (struct frame_state_reg_info): Remove reg[].how element and the anonymous enum there. Add how element. * unwind-dw2.c: Include stddef.h. (uw_frame_state_for): Don't clear first offsetof (_Unwind_FrameState, regs.how[0]) bytes of *fs. (execute_cfa_program, __frame_state_for, uw_update_context_1, uw_update_context): Use fs->regs.how[X] instead of fs->regs.reg[X].how or fs.regs.how[X] instead of fs.regs.reg[X].how. * config/sh/linux-unwind.h (sh_fallback_frame_state): Likewise. * config/bfin/linux-unwind.h (bfin_fallback_frame_state): Likewise. * config/pa/linux-unwind.h (pa32_fallback_frame_state): Likewise. * config/pa/hpux-unwind.h (UPDATE_FS_FOR_SAR, UPDATE_FS_FOR_GR, UPDATE_FS_FOR_FR, UPDATE_FS_FOR_PC, pa_fallback_frame_state): Likewise. * config/alpha/vms-unwind.h (alpha_vms_fallback_frame_state): Likewise. * config/alpha/linux-unwind.h (alpha_fallback_frame_state): Likewise. * config/arc/linux-unwind.h (arc_fallback_frame_state, arc_frob_update_context): Likewise. * config/riscv/linux-unwind.h (riscv_fallback_frame_state): Likewise. * config/nios2/linux-unwind.h (NIOS2_REG): Likewise. * config/nds32/linux-unwind.h (NDS32_PUT_FS_REG): Likewise. * config/s390/tpf-unwind.h (s390_fallback_frame_state): Likewise. * config/s390/linux-unwind.h (s390_fallback_frame_state): Likewise. * config/sparc/sol2-unwind.h (sparc64_frob_update_context, MD_FALLBACK_FRAME_STATE_FOR): Likewise. * config/sparc/linux-unwind.h (sparc64_fallback_frame_state, sparc64_frob_update_context, sparc_fallback_frame_state): Likewise. * config/i386/sol2-unwind.h (x86_64_fallback_frame_state, x86_fallback_frame_state): Likewise. * config/i386/w32-unwind.h (i386_w32_fallback_frame_state): Likewise. * config/i386/linux-unwind.h (x86_64_fallback_frame_state, x86_fallback_frame_state): Likewise. * config/i386/freebsd-unwind.h (x86_64_freebsd_fallback_frame_state): Likewise. * config/i386/dragonfly-unwind.h (x86_64_dragonfly_fallback_frame_state): Likewise. * config/i386/gnu-unwind.h (x86_gnu_fallback_frame_state): Likewise. * config/csky/linux-unwind.h (csky_fallback_frame_state): Likewise. * config/aarch64/linux-unwind.h (aarch64_fallback_frame_state): Likewise. * config/aarch64/freebsd-unwind.h (aarch64_freebsd_fallback_frame_state): Likewise. * config/aarch64/aarch64-unwind.h (aarch64_frob_update_context): Likewise. * config/or1k/linux-unwind.h (or1k_fallback_frame_state): Likewise. * config/mips/linux-unwind.h (mips_fallback_frame_state): Likewise. * config/loongarch/linux-unwind.h (loongarch_fallback_frame_state): Likewise. * config/m68k/linux-unwind.h (m68k_fallback_frame_state): Likewise. * config/xtensa/linux-unwind.h (xtensa_fallback_frame_state): Likewise. * config/rs6000/darwin-fallback.c (set_offset): Likewise. * config/rs6000/aix-unwind.h (MD_FROB_UPDATE_CONTEXT): Likewise. * config/rs6000/linux-unwind.h (ppc_fallback_frame_state): Likewise. * config/rs6000/freebsd-unwind.h (frob_update_context): Likewise.
2022-10-06 11:04:52 +02:00
fs->regs.how[15] = REG_SAVED_OFFSET;
tm.texi (MD_UNWIND_SUPPORT): Document. * doc/tm.texi (MD_UNWIND_SUPPORT): Document. (MD_FALLBACK_FRAME_STATE_FOR): Update. * unwind-dw2.c (MD_UNWIND_SUPPORT): #include if defined. (uw_frame_state_for): Adjust MD_FALLBACK_FRAME_STATE_FOR invocation. (MD_FROB_UPDATE_CONTEXT): Remove default. (uw_update_context_1): Instead #ifdef invocation. * config/ia64/unwind-ia64.c (MD_UNWIND_SUPPORT): #include if defined. (uw_frame_state_for): Adjust MD_FALLBACK_FRAME_STATE_FOR invocation. * config/alpha/gnu.h (MD_FALLBACK_FRAME_STATE_FOR): Don't undef. (MD_UNWIND_SUPPORT): Undefine this instead. * config/i386/gnu.h: Likewise. * config/alpha/linux-unwind.h: New file, macro converted to function, extracted from.. * config/alpha/linux.h (MD_FALLBACK_FRAME_STATE_FOR): ..this. (MD_UNWIND_SUPPORT): Define. * config/alpha/vms-unwind.h, config/alpha/vms.h: Likewise. * config/i386/linux-unwind.h, config/i386/linux.h, config/i386/linux64.h: Likewise. * config/ia64/linux-unwind.h, config/ia64/linux.h: Likewise. MD_HANDLE_UNWABI too. * config/mips/linux-unwind.h, config/mips/linux.h: Likewise. * config/pa/linux-unwind.h, config/pa/pa32-linux.h: Likewise. * config/rs6000/darwin-unwind.h, config/rs6000/darwin.h: Likewise. * config/s390/linux-unwind.h, config/s390/linux.h: Likewise. * config/sparc/linux-unwind.h, config/sparc/linux.h, config/sparc/linux64.h: Likewise. * config/sh/linux-unwind.h, config/sh/linux.h: Likewise, but merge SH_FALLBACK_FRAME_FLOAT_STATE into sh_fallback_frame_state. * config/rs6000/linux-unwind.h, config/rs6000/linux.h, config/rs6000/linux64.h: Likewise. Split out get_sigcontext function. Use ARG_POINTER_REGNUM for 32-bit temp reg too. From-SVN: r87167
2004-09-08 00:17:19 +00:00
fs->regs.reg[15].loc.offset = (long)&sc->r15 - new_cfa;
libgcc: Decrease size of _Unwind_FrameState and even more size of cleared area in uw_frame_state_for The following patch implements something that has Florian found as low hanging fruit in our unwinder and has been discussed in the https://gcc.gnu.org/wiki/cauldron2022#cauldron2022talks.inprocess_unwinding_bof talk. _Unwind_FrameState type seems to be (unlike the pre-GCC 3 frame_state which has been part of ABI) private to unwind-dw2.c + unwind.inc it includes, it is always defined on the stack of some entrypoints, initialized by static uw_frame_state_for and the address of it is also passed to other static functions or the static inlines handling machine dependent unwinding, but it isn't fortunately passed to any callbacks or public functions, so I think we can safely change it any time we want. Florian mentioned that the structure is large even on x86_64, 384 bytes there, starts with 328 bytes long element with frame_state_reg_info type which then starts with an array with __LIBGCC_DWARF_FRAME_REGISTERS__ + 1 elements, each of them is 16 bytes long, on x86_64 __LIBGCC_DWARF_FRAME_REGISTERS__ is just 17 but even that is big, on say riscv __LIBGCC_DWARF_FRAME_REGISTERS__ is I think 128, on powerpc 111, on sh 153 etc. And, we memset to zero the whole fs variable with the _Unwind_FrameState type at the start of the unwinding. The reason why each element is 16 byte (on 64-bit arches) is that it contains some pointer or pointer sized integer and then an enum (with just 7 different enumerators) + padding. The following patch decreases it by moving the enum into a separate array and using just one byte for each register in that second array. We could compress it even more, say 4 bits per register, but I don't want to uglify the code for it too much and make the accesses slower. Furthermore, the clearing of the object can clear only thos how array and members after it, because REG_UNSAVED enumerator (0) doesn't actually need any pointer or pointer sized integer, it is just the other kinds that need to have there something. By doing this, on x86_64 the above numbers change to _Unwind_FrameState type being now 264 bytes long, frame_state_reg_info 208 bytes and we don't clear the first 144 bytes of the object, so the memset is 120 bytes, so ~ 31% of the old clearing size. On riscv 64-bit assuming it has same structure layout rules for the few types used there that would be ~ 2160 bytes of _Unwind_FrameState type before and ~ 1264 bytes after, with the memset previously ~ 2160 bytes and after ~ 232 bytes after. We've also talked about possibly adding a number of initially initialized regs and initializing the rest lazily, but at least for x86_64 with 18 elements in the array that doesn't seem to be worth it anymore, especially because return address column is 16 there and that is usually the first thing to be touched. It might theory help with lots of registers if they are usually untouched, but would uglify and complicate any stores to how by having to check there for the not initialized yet cases and lazy initialization, and similarly for all reads of how to do there if below last initialized one, use how, otherwise imply REG_UNSAVED. The disadvantage of the patch is that touching reg[x].loc and how[x] now means 2 cachelines rather than one as before, and I admit beyond bootstrap/regtest I haven't benchmarked it in any way. 2022-10-06 Jakub Jelinek <jakub@redhat.com> * unwind-dw2.h (REG_UNSAVED, REG_SAVED_OFFSET, REG_SAVED_REG, REG_SAVED_EXP, REG_SAVED_VAL_OFFSET, REG_SAVED_VAL_EXP, REG_UNDEFINED): New anonymous enum, moved from inside of struct frame_state_reg_info. (struct frame_state_reg_info): Remove reg[].how element and the anonymous enum there. Add how element. * unwind-dw2.c: Include stddef.h. (uw_frame_state_for): Don't clear first offsetof (_Unwind_FrameState, regs.how[0]) bytes of *fs. (execute_cfa_program, __frame_state_for, uw_update_context_1, uw_update_context): Use fs->regs.how[X] instead of fs->regs.reg[X].how or fs.regs.how[X] instead of fs.regs.reg[X].how. * config/sh/linux-unwind.h (sh_fallback_frame_state): Likewise. * config/bfin/linux-unwind.h (bfin_fallback_frame_state): Likewise. * config/pa/linux-unwind.h (pa32_fallback_frame_state): Likewise. * config/pa/hpux-unwind.h (UPDATE_FS_FOR_SAR, UPDATE_FS_FOR_GR, UPDATE_FS_FOR_FR, UPDATE_FS_FOR_PC, pa_fallback_frame_state): Likewise. * config/alpha/vms-unwind.h (alpha_vms_fallback_frame_state): Likewise. * config/alpha/linux-unwind.h (alpha_fallback_frame_state): Likewise. * config/arc/linux-unwind.h (arc_fallback_frame_state, arc_frob_update_context): Likewise. * config/riscv/linux-unwind.h (riscv_fallback_frame_state): Likewise. * config/nios2/linux-unwind.h (NIOS2_REG): Likewise. * config/nds32/linux-unwind.h (NDS32_PUT_FS_REG): Likewise. * config/s390/tpf-unwind.h (s390_fallback_frame_state): Likewise. * config/s390/linux-unwind.h (s390_fallback_frame_state): Likewise. * config/sparc/sol2-unwind.h (sparc64_frob_update_context, MD_FALLBACK_FRAME_STATE_FOR): Likewise. * config/sparc/linux-unwind.h (sparc64_fallback_frame_state, sparc64_frob_update_context, sparc_fallback_frame_state): Likewise. * config/i386/sol2-unwind.h (x86_64_fallback_frame_state, x86_fallback_frame_state): Likewise. * config/i386/w32-unwind.h (i386_w32_fallback_frame_state): Likewise. * config/i386/linux-unwind.h (x86_64_fallback_frame_state, x86_fallback_frame_state): Likewise. * config/i386/freebsd-unwind.h (x86_64_freebsd_fallback_frame_state): Likewise. * config/i386/dragonfly-unwind.h (x86_64_dragonfly_fallback_frame_state): Likewise. * config/i386/gnu-unwind.h (x86_gnu_fallback_frame_state): Likewise. * config/csky/linux-unwind.h (csky_fallback_frame_state): Likewise. * config/aarch64/linux-unwind.h (aarch64_fallback_frame_state): Likewise. * config/aarch64/freebsd-unwind.h (aarch64_freebsd_fallback_frame_state): Likewise. * config/aarch64/aarch64-unwind.h (aarch64_frob_update_context): Likewise. * config/or1k/linux-unwind.h (or1k_fallback_frame_state): Likewise. * config/mips/linux-unwind.h (mips_fallback_frame_state): Likewise. * config/loongarch/linux-unwind.h (loongarch_fallback_frame_state): Likewise. * config/m68k/linux-unwind.h (m68k_fallback_frame_state): Likewise. * config/xtensa/linux-unwind.h (xtensa_fallback_frame_state): Likewise. * config/rs6000/darwin-fallback.c (set_offset): Likewise. * config/rs6000/aix-unwind.h (MD_FROB_UPDATE_CONTEXT): Likewise. * config/rs6000/linux-unwind.h (ppc_fallback_frame_state): Likewise. * config/rs6000/freebsd-unwind.h (frob_update_context): Likewise.
2022-10-06 11:04:52 +02:00
fs->regs.how[16] = REG_SAVED_OFFSET;
tm.texi (MD_UNWIND_SUPPORT): Document. * doc/tm.texi (MD_UNWIND_SUPPORT): Document. (MD_FALLBACK_FRAME_STATE_FOR): Update. * unwind-dw2.c (MD_UNWIND_SUPPORT): #include if defined. (uw_frame_state_for): Adjust MD_FALLBACK_FRAME_STATE_FOR invocation. (MD_FROB_UPDATE_CONTEXT): Remove default. (uw_update_context_1): Instead #ifdef invocation. * config/ia64/unwind-ia64.c (MD_UNWIND_SUPPORT): #include if defined. (uw_frame_state_for): Adjust MD_FALLBACK_FRAME_STATE_FOR invocation. * config/alpha/gnu.h (MD_FALLBACK_FRAME_STATE_FOR): Don't undef. (MD_UNWIND_SUPPORT): Undefine this instead. * config/i386/gnu.h: Likewise. * config/alpha/linux-unwind.h: New file, macro converted to function, extracted from.. * config/alpha/linux.h (MD_FALLBACK_FRAME_STATE_FOR): ..this. (MD_UNWIND_SUPPORT): Define. * config/alpha/vms-unwind.h, config/alpha/vms.h: Likewise. * config/i386/linux-unwind.h, config/i386/linux.h, config/i386/linux64.h: Likewise. * config/ia64/linux-unwind.h, config/ia64/linux.h: Likewise. MD_HANDLE_UNWABI too. * config/mips/linux-unwind.h, config/mips/linux.h: Likewise. * config/pa/linux-unwind.h, config/pa/pa32-linux.h: Likewise. * config/rs6000/darwin-unwind.h, config/rs6000/darwin.h: Likewise. * config/s390/linux-unwind.h, config/s390/linux.h: Likewise. * config/sparc/linux-unwind.h, config/sparc/linux.h, config/sparc/linux64.h: Likewise. * config/sh/linux-unwind.h, config/sh/linux.h: Likewise, but merge SH_FALLBACK_FRAME_FLOAT_STATE into sh_fallback_frame_state. * config/rs6000/linux-unwind.h, config/rs6000/linux.h, config/rs6000/linux64.h: Likewise. Split out get_sigcontext function. Use ARG_POINTER_REGNUM for 32-bit temp reg too. From-SVN: r87167
2004-09-08 00:17:19 +00:00
fs->regs.reg[16].loc.offset = (long)&sc->rip - new_cfa;
fs->retaddr_column = 16;
re PR other/26208 (Serious problem with unwinding through signal frames) PR other/26208 * unwind-dw2.c (struct _Unwind_Context): Add signal_frame field. (extract_cie_info): Handle S flag in augmentation string. (execute_cfa_program): If context->signal_frame, execute also fs->pc == context->ra instructions. (uw_frame_state_for): If context->signal_frame, don't subtract one from context->ra to find FDE. (uw_update_context_1): Set context->signal_frame to fs->signal_frame. (_Unwind_GetIPInfo): New function. * unwind-dw2.h (_Unwind_FrameState): Add signal_frame field. * unwind-c.c (PERSONALITY_FUNCTION): Use _Unwind_GetIPInfo instead of _Unwind_GetIP. * unwind-sjlj.c (_Unwind_GetIPInfo): New function. * unwind-generic.h (_Unwind_GetIPInfo): New prototype. * unwind-compat.c (_Unwind_GetIPInfo): New function. * libgcc-std.ver (_Unwind_GetIPInfo): Export @@GCC_4.2.0. * config/ia64/unwind-ia64.c (_Unwind_GetIPInfo): New function. * config/arm/unwind-arm.h (_Unwind_GetIPInfo): Define. * config/i386/linux-unwind.h (x86_fallback_frame_state, x86_64_fallback_frame_state): Set fs->signal_frame. * config/rs6000/linux-unwind.h (ppc_fallback_frame_state): Likewise. (MD_FROB_UPDATE_CONTEXT): Define unconditionally. (frob_update_context): Likewise. Workaround missing S flag in Linux 2.6.12 - 2.6.16 kernel vDSOs. * config/s390/linux-unwind.h (s390_fallback_frame_state): Likewise. Remove the psw_addr + 1 hack. libjava/ * exception.cc (PERSONALITY_FUNCTION): Use _Unwind_GetIPInfo instead of _Unwind_GetIP. * include/i386-signal.h (MAKE_THROW_FRAME): Change into empty macro. (HANDLE_DIVIDE_OVERFLOW): Don't adjust _res->eip if falling through to throw. * include/x86_64-signal.h (MAKE_THROW_FRAME): Change into empty macro. * include/powerpc-signal.h (MAKE_THROW_FRAME): Change into empty macro. libstdc++-v3/ * libsupc++/eh_personality.cc (PERSONALITY_FUNCTION): Use _Unwind_GetIPInfo instead of _Unwind_GetIP. From-SVN: r111488
2006-02-27 18:26:26 +01:00
fs->signal_frame = 1;
tm.texi (MD_UNWIND_SUPPORT): Document. * doc/tm.texi (MD_UNWIND_SUPPORT): Document. (MD_FALLBACK_FRAME_STATE_FOR): Update. * unwind-dw2.c (MD_UNWIND_SUPPORT): #include if defined. (uw_frame_state_for): Adjust MD_FALLBACK_FRAME_STATE_FOR invocation. (MD_FROB_UPDATE_CONTEXT): Remove default. (uw_update_context_1): Instead #ifdef invocation. * config/ia64/unwind-ia64.c (MD_UNWIND_SUPPORT): #include if defined. (uw_frame_state_for): Adjust MD_FALLBACK_FRAME_STATE_FOR invocation. * config/alpha/gnu.h (MD_FALLBACK_FRAME_STATE_FOR): Don't undef. (MD_UNWIND_SUPPORT): Undefine this instead. * config/i386/gnu.h: Likewise. * config/alpha/linux-unwind.h: New file, macro converted to function, extracted from.. * config/alpha/linux.h (MD_FALLBACK_FRAME_STATE_FOR): ..this. (MD_UNWIND_SUPPORT): Define. * config/alpha/vms-unwind.h, config/alpha/vms.h: Likewise. * config/i386/linux-unwind.h, config/i386/linux.h, config/i386/linux64.h: Likewise. * config/ia64/linux-unwind.h, config/ia64/linux.h: Likewise. MD_HANDLE_UNWABI too. * config/mips/linux-unwind.h, config/mips/linux.h: Likewise. * config/pa/linux-unwind.h, config/pa/pa32-linux.h: Likewise. * config/rs6000/darwin-unwind.h, config/rs6000/darwin.h: Likewise. * config/s390/linux-unwind.h, config/s390/linux.h: Likewise. * config/sparc/linux-unwind.h, config/sparc/linux.h, config/sparc/linux64.h: Likewise. * config/sh/linux-unwind.h, config/sh/linux.h: Likewise, but merge SH_FALLBACK_FRAME_FLOAT_STATE into sh_fallback_frame_state. * config/rs6000/linux-unwind.h, config/rs6000/linux.h, config/rs6000/linux64.h: Likewise. Split out get_sigcontext function. Use ARG_POINTER_REGNUM for 32-bit temp reg too. From-SVN: r87167
2004-09-08 00:17:19 +00:00
return _URC_NO_REASON;
}
#else /* ifdef __x86_64__ */
#define MD_FALLBACK_FRAME_STATE_FOR x86_fallback_frame_state
static _Unwind_Reason_Code
x86_fallback_frame_state (struct _Unwind_Context *context,
_Unwind_FrameState *fs)
{
unsigned char *pc = context->ra;
struct sigcontext *sc;
long new_cfa;
/* popl %eax ; movl $__NR_sigreturn,%eax ; int $0x80 */
if (*(unsigned short *)(pc+0) == 0xb858
&& *(unsigned int *)(pc+2) == 119
&& *(unsigned short *)(pc+6) == 0x80cd)
sc = context->cfa + 4;
/* movl $__NR_rt_sigreturn,%eax ; int $0x80 */
else if (*(unsigned char *)(pc+0) == 0xb8
&& *(unsigned int *)(pc+1) == 173
&& *(unsigned short *)(pc+5) == 0x80cd)
{
struct rt_sigframe {
int sig;
siginfo_t *pinfo;
tm.texi (MD_UNWIND_SUPPORT): Document. * doc/tm.texi (MD_UNWIND_SUPPORT): Document. (MD_FALLBACK_FRAME_STATE_FOR): Update. * unwind-dw2.c (MD_UNWIND_SUPPORT): #include if defined. (uw_frame_state_for): Adjust MD_FALLBACK_FRAME_STATE_FOR invocation. (MD_FROB_UPDATE_CONTEXT): Remove default. (uw_update_context_1): Instead #ifdef invocation. * config/ia64/unwind-ia64.c (MD_UNWIND_SUPPORT): #include if defined. (uw_frame_state_for): Adjust MD_FALLBACK_FRAME_STATE_FOR invocation. * config/alpha/gnu.h (MD_FALLBACK_FRAME_STATE_FOR): Don't undef. (MD_UNWIND_SUPPORT): Undefine this instead. * config/i386/gnu.h: Likewise. * config/alpha/linux-unwind.h: New file, macro converted to function, extracted from.. * config/alpha/linux.h (MD_FALLBACK_FRAME_STATE_FOR): ..this. (MD_UNWIND_SUPPORT): Define. * config/alpha/vms-unwind.h, config/alpha/vms.h: Likewise. * config/i386/linux-unwind.h, config/i386/linux.h, config/i386/linux64.h: Likewise. * config/ia64/linux-unwind.h, config/ia64/linux.h: Likewise. MD_HANDLE_UNWABI too. * config/mips/linux-unwind.h, config/mips/linux.h: Likewise. * config/pa/linux-unwind.h, config/pa/pa32-linux.h: Likewise. * config/rs6000/darwin-unwind.h, config/rs6000/darwin.h: Likewise. * config/s390/linux-unwind.h, config/s390/linux.h: Likewise. * config/sparc/linux-unwind.h, config/sparc/linux.h, config/sparc/linux64.h: Likewise. * config/sh/linux-unwind.h, config/sh/linux.h: Likewise, but merge SH_FALLBACK_FRAME_FLOAT_STATE into sh_fallback_frame_state. * config/rs6000/linux-unwind.h, config/rs6000/linux.h, config/rs6000/linux64.h: Likewise. Split out get_sigcontext function. Use ARG_POINTER_REGNUM for 32-bit temp reg too. From-SVN: r87167
2004-09-08 00:17:19 +00:00
void *puc;
siginfo_t info;
ucontext_t uc;
tm.texi (MD_UNWIND_SUPPORT): Document. * doc/tm.texi (MD_UNWIND_SUPPORT): Document. (MD_FALLBACK_FRAME_STATE_FOR): Update. * unwind-dw2.c (MD_UNWIND_SUPPORT): #include if defined. (uw_frame_state_for): Adjust MD_FALLBACK_FRAME_STATE_FOR invocation. (MD_FROB_UPDATE_CONTEXT): Remove default. (uw_update_context_1): Instead #ifdef invocation. * config/ia64/unwind-ia64.c (MD_UNWIND_SUPPORT): #include if defined. (uw_frame_state_for): Adjust MD_FALLBACK_FRAME_STATE_FOR invocation. * config/alpha/gnu.h (MD_FALLBACK_FRAME_STATE_FOR): Don't undef. (MD_UNWIND_SUPPORT): Undefine this instead. * config/i386/gnu.h: Likewise. * config/alpha/linux-unwind.h: New file, macro converted to function, extracted from.. * config/alpha/linux.h (MD_FALLBACK_FRAME_STATE_FOR): ..this. (MD_UNWIND_SUPPORT): Define. * config/alpha/vms-unwind.h, config/alpha/vms.h: Likewise. * config/i386/linux-unwind.h, config/i386/linux.h, config/i386/linux64.h: Likewise. * config/ia64/linux-unwind.h, config/ia64/linux.h: Likewise. MD_HANDLE_UNWABI too. * config/mips/linux-unwind.h, config/mips/linux.h: Likewise. * config/pa/linux-unwind.h, config/pa/pa32-linux.h: Likewise. * config/rs6000/darwin-unwind.h, config/rs6000/darwin.h: Likewise. * config/s390/linux-unwind.h, config/s390/linux.h: Likewise. * config/sparc/linux-unwind.h, config/sparc/linux.h, config/sparc/linux64.h: Likewise. * config/sh/linux-unwind.h, config/sh/linux.h: Likewise, but merge SH_FALLBACK_FRAME_FLOAT_STATE into sh_fallback_frame_state. * config/rs6000/linux-unwind.h, config/rs6000/linux.h, config/rs6000/linux64.h: Likewise. Split out get_sigcontext function. Use ARG_POINTER_REGNUM for 32-bit temp reg too. From-SVN: r87167
2004-09-08 00:17:19 +00:00
} *rt_ = context->cfa;
/* The void * cast is necessary to avoid an aliasing warning.
The aliasing warning is correct, but should not be a problem
because it does not alias anything. */
sc = (struct sigcontext *) (void *) &rt_->uc.uc_mcontext;
tm.texi (MD_UNWIND_SUPPORT): Document. * doc/tm.texi (MD_UNWIND_SUPPORT): Document. (MD_FALLBACK_FRAME_STATE_FOR): Update. * unwind-dw2.c (MD_UNWIND_SUPPORT): #include if defined. (uw_frame_state_for): Adjust MD_FALLBACK_FRAME_STATE_FOR invocation. (MD_FROB_UPDATE_CONTEXT): Remove default. (uw_update_context_1): Instead #ifdef invocation. * config/ia64/unwind-ia64.c (MD_UNWIND_SUPPORT): #include if defined. (uw_frame_state_for): Adjust MD_FALLBACK_FRAME_STATE_FOR invocation. * config/alpha/gnu.h (MD_FALLBACK_FRAME_STATE_FOR): Don't undef. (MD_UNWIND_SUPPORT): Undefine this instead. * config/i386/gnu.h: Likewise. * config/alpha/linux-unwind.h: New file, macro converted to function, extracted from.. * config/alpha/linux.h (MD_FALLBACK_FRAME_STATE_FOR): ..this. (MD_UNWIND_SUPPORT): Define. * config/alpha/vms-unwind.h, config/alpha/vms.h: Likewise. * config/i386/linux-unwind.h, config/i386/linux.h, config/i386/linux64.h: Likewise. * config/ia64/linux-unwind.h, config/ia64/linux.h: Likewise. MD_HANDLE_UNWABI too. * config/mips/linux-unwind.h, config/mips/linux.h: Likewise. * config/pa/linux-unwind.h, config/pa/pa32-linux.h: Likewise. * config/rs6000/darwin-unwind.h, config/rs6000/darwin.h: Likewise. * config/s390/linux-unwind.h, config/s390/linux.h: Likewise. * config/sparc/linux-unwind.h, config/sparc/linux.h, config/sparc/linux64.h: Likewise. * config/sh/linux-unwind.h, config/sh/linux.h: Likewise, but merge SH_FALLBACK_FRAME_FLOAT_STATE into sh_fallback_frame_state. * config/rs6000/linux-unwind.h, config/rs6000/linux.h, config/rs6000/linux64.h: Likewise. Split out get_sigcontext function. Use ARG_POINTER_REGNUM for 32-bit temp reg too. From-SVN: r87167
2004-09-08 00:17:19 +00:00
}
else
return _URC_END_OF_STACK;
new_cfa = sc->esp;
fs->regs.cfa_how = CFA_REG_OFFSET;
fs->regs.cfa_reg = 4;
fs->regs.cfa_offset = new_cfa - (long) context->cfa;
tm.texi (MD_UNWIND_SUPPORT): Document. * doc/tm.texi (MD_UNWIND_SUPPORT): Document. (MD_FALLBACK_FRAME_STATE_FOR): Update. * unwind-dw2.c (MD_UNWIND_SUPPORT): #include if defined. (uw_frame_state_for): Adjust MD_FALLBACK_FRAME_STATE_FOR invocation. (MD_FROB_UPDATE_CONTEXT): Remove default. (uw_update_context_1): Instead #ifdef invocation. * config/ia64/unwind-ia64.c (MD_UNWIND_SUPPORT): #include if defined. (uw_frame_state_for): Adjust MD_FALLBACK_FRAME_STATE_FOR invocation. * config/alpha/gnu.h (MD_FALLBACK_FRAME_STATE_FOR): Don't undef. (MD_UNWIND_SUPPORT): Undefine this instead. * config/i386/gnu.h: Likewise. * config/alpha/linux-unwind.h: New file, macro converted to function, extracted from.. * config/alpha/linux.h (MD_FALLBACK_FRAME_STATE_FOR): ..this. (MD_UNWIND_SUPPORT): Define. * config/alpha/vms-unwind.h, config/alpha/vms.h: Likewise. * config/i386/linux-unwind.h, config/i386/linux.h, config/i386/linux64.h: Likewise. * config/ia64/linux-unwind.h, config/ia64/linux.h: Likewise. MD_HANDLE_UNWABI too. * config/mips/linux-unwind.h, config/mips/linux.h: Likewise. * config/pa/linux-unwind.h, config/pa/pa32-linux.h: Likewise. * config/rs6000/darwin-unwind.h, config/rs6000/darwin.h: Likewise. * config/s390/linux-unwind.h, config/s390/linux.h: Likewise. * config/sparc/linux-unwind.h, config/sparc/linux.h, config/sparc/linux64.h: Likewise. * config/sh/linux-unwind.h, config/sh/linux.h: Likewise, but merge SH_FALLBACK_FRAME_FLOAT_STATE into sh_fallback_frame_state. * config/rs6000/linux-unwind.h, config/rs6000/linux.h, config/rs6000/linux64.h: Likewise. Split out get_sigcontext function. Use ARG_POINTER_REGNUM for 32-bit temp reg too. From-SVN: r87167
2004-09-08 00:17:19 +00:00
/* The SVR4 register numbering macros aren't usable in libgcc. */
libgcc: Decrease size of _Unwind_FrameState and even more size of cleared area in uw_frame_state_for The following patch implements something that has Florian found as low hanging fruit in our unwinder and has been discussed in the https://gcc.gnu.org/wiki/cauldron2022#cauldron2022talks.inprocess_unwinding_bof talk. _Unwind_FrameState type seems to be (unlike the pre-GCC 3 frame_state which has been part of ABI) private to unwind-dw2.c + unwind.inc it includes, it is always defined on the stack of some entrypoints, initialized by static uw_frame_state_for and the address of it is also passed to other static functions or the static inlines handling machine dependent unwinding, but it isn't fortunately passed to any callbacks or public functions, so I think we can safely change it any time we want. Florian mentioned that the structure is large even on x86_64, 384 bytes there, starts with 328 bytes long element with frame_state_reg_info type which then starts with an array with __LIBGCC_DWARF_FRAME_REGISTERS__ + 1 elements, each of them is 16 bytes long, on x86_64 __LIBGCC_DWARF_FRAME_REGISTERS__ is just 17 but even that is big, on say riscv __LIBGCC_DWARF_FRAME_REGISTERS__ is I think 128, on powerpc 111, on sh 153 etc. And, we memset to zero the whole fs variable with the _Unwind_FrameState type at the start of the unwinding. The reason why each element is 16 byte (on 64-bit arches) is that it contains some pointer or pointer sized integer and then an enum (with just 7 different enumerators) + padding. The following patch decreases it by moving the enum into a separate array and using just one byte for each register in that second array. We could compress it even more, say 4 bits per register, but I don't want to uglify the code for it too much and make the accesses slower. Furthermore, the clearing of the object can clear only thos how array and members after it, because REG_UNSAVED enumerator (0) doesn't actually need any pointer or pointer sized integer, it is just the other kinds that need to have there something. By doing this, on x86_64 the above numbers change to _Unwind_FrameState type being now 264 bytes long, frame_state_reg_info 208 bytes and we don't clear the first 144 bytes of the object, so the memset is 120 bytes, so ~ 31% of the old clearing size. On riscv 64-bit assuming it has same structure layout rules for the few types used there that would be ~ 2160 bytes of _Unwind_FrameState type before and ~ 1264 bytes after, with the memset previously ~ 2160 bytes and after ~ 232 bytes after. We've also talked about possibly adding a number of initially initialized regs and initializing the rest lazily, but at least for x86_64 with 18 elements in the array that doesn't seem to be worth it anymore, especially because return address column is 16 there and that is usually the first thing to be touched. It might theory help with lots of registers if they are usually untouched, but would uglify and complicate any stores to how by having to check there for the not initialized yet cases and lazy initialization, and similarly for all reads of how to do there if below last initialized one, use how, otherwise imply REG_UNSAVED. The disadvantage of the patch is that touching reg[x].loc and how[x] now means 2 cachelines rather than one as before, and I admit beyond bootstrap/regtest I haven't benchmarked it in any way. 2022-10-06 Jakub Jelinek <jakub@redhat.com> * unwind-dw2.h (REG_UNSAVED, REG_SAVED_OFFSET, REG_SAVED_REG, REG_SAVED_EXP, REG_SAVED_VAL_OFFSET, REG_SAVED_VAL_EXP, REG_UNDEFINED): New anonymous enum, moved from inside of struct frame_state_reg_info. (struct frame_state_reg_info): Remove reg[].how element and the anonymous enum there. Add how element. * unwind-dw2.c: Include stddef.h. (uw_frame_state_for): Don't clear first offsetof (_Unwind_FrameState, regs.how[0]) bytes of *fs. (execute_cfa_program, __frame_state_for, uw_update_context_1, uw_update_context): Use fs->regs.how[X] instead of fs->regs.reg[X].how or fs.regs.how[X] instead of fs.regs.reg[X].how. * config/sh/linux-unwind.h (sh_fallback_frame_state): Likewise. * config/bfin/linux-unwind.h (bfin_fallback_frame_state): Likewise. * config/pa/linux-unwind.h (pa32_fallback_frame_state): Likewise. * config/pa/hpux-unwind.h (UPDATE_FS_FOR_SAR, UPDATE_FS_FOR_GR, UPDATE_FS_FOR_FR, UPDATE_FS_FOR_PC, pa_fallback_frame_state): Likewise. * config/alpha/vms-unwind.h (alpha_vms_fallback_frame_state): Likewise. * config/alpha/linux-unwind.h (alpha_fallback_frame_state): Likewise. * config/arc/linux-unwind.h (arc_fallback_frame_state, arc_frob_update_context): Likewise. * config/riscv/linux-unwind.h (riscv_fallback_frame_state): Likewise. * config/nios2/linux-unwind.h (NIOS2_REG): Likewise. * config/nds32/linux-unwind.h (NDS32_PUT_FS_REG): Likewise. * config/s390/tpf-unwind.h (s390_fallback_frame_state): Likewise. * config/s390/linux-unwind.h (s390_fallback_frame_state): Likewise. * config/sparc/sol2-unwind.h (sparc64_frob_update_context, MD_FALLBACK_FRAME_STATE_FOR): Likewise. * config/sparc/linux-unwind.h (sparc64_fallback_frame_state, sparc64_frob_update_context, sparc_fallback_frame_state): Likewise. * config/i386/sol2-unwind.h (x86_64_fallback_frame_state, x86_fallback_frame_state): Likewise. * config/i386/w32-unwind.h (i386_w32_fallback_frame_state): Likewise. * config/i386/linux-unwind.h (x86_64_fallback_frame_state, x86_fallback_frame_state): Likewise. * config/i386/freebsd-unwind.h (x86_64_freebsd_fallback_frame_state): Likewise. * config/i386/dragonfly-unwind.h (x86_64_dragonfly_fallback_frame_state): Likewise. * config/i386/gnu-unwind.h (x86_gnu_fallback_frame_state): Likewise. * config/csky/linux-unwind.h (csky_fallback_frame_state): Likewise. * config/aarch64/linux-unwind.h (aarch64_fallback_frame_state): Likewise. * config/aarch64/freebsd-unwind.h (aarch64_freebsd_fallback_frame_state): Likewise. * config/aarch64/aarch64-unwind.h (aarch64_frob_update_context): Likewise. * config/or1k/linux-unwind.h (or1k_fallback_frame_state): Likewise. * config/mips/linux-unwind.h (mips_fallback_frame_state): Likewise. * config/loongarch/linux-unwind.h (loongarch_fallback_frame_state): Likewise. * config/m68k/linux-unwind.h (m68k_fallback_frame_state): Likewise. * config/xtensa/linux-unwind.h (xtensa_fallback_frame_state): Likewise. * config/rs6000/darwin-fallback.c (set_offset): Likewise. * config/rs6000/aix-unwind.h (MD_FROB_UPDATE_CONTEXT): Likewise. * config/rs6000/linux-unwind.h (ppc_fallback_frame_state): Likewise. * config/rs6000/freebsd-unwind.h (frob_update_context): Likewise.
2022-10-06 11:04:52 +02:00
fs->regs.how[0] = REG_SAVED_OFFSET;
fs->regs.reg[0].loc.offset = (long)&sc->eax - new_cfa;
libgcc: Decrease size of _Unwind_FrameState and even more size of cleared area in uw_frame_state_for The following patch implements something that has Florian found as low hanging fruit in our unwinder and has been discussed in the https://gcc.gnu.org/wiki/cauldron2022#cauldron2022talks.inprocess_unwinding_bof talk. _Unwind_FrameState type seems to be (unlike the pre-GCC 3 frame_state which has been part of ABI) private to unwind-dw2.c + unwind.inc it includes, it is always defined on the stack of some entrypoints, initialized by static uw_frame_state_for and the address of it is also passed to other static functions or the static inlines handling machine dependent unwinding, but it isn't fortunately passed to any callbacks or public functions, so I think we can safely change it any time we want. Florian mentioned that the structure is large even on x86_64, 384 bytes there, starts with 328 bytes long element with frame_state_reg_info type which then starts with an array with __LIBGCC_DWARF_FRAME_REGISTERS__ + 1 elements, each of them is 16 bytes long, on x86_64 __LIBGCC_DWARF_FRAME_REGISTERS__ is just 17 but even that is big, on say riscv __LIBGCC_DWARF_FRAME_REGISTERS__ is I think 128, on powerpc 111, on sh 153 etc. And, we memset to zero the whole fs variable with the _Unwind_FrameState type at the start of the unwinding. The reason why each element is 16 byte (on 64-bit arches) is that it contains some pointer or pointer sized integer and then an enum (with just 7 different enumerators) + padding. The following patch decreases it by moving the enum into a separate array and using just one byte for each register in that second array. We could compress it even more, say 4 bits per register, but I don't want to uglify the code for it too much and make the accesses slower. Furthermore, the clearing of the object can clear only thos how array and members after it, because REG_UNSAVED enumerator (0) doesn't actually need any pointer or pointer sized integer, it is just the other kinds that need to have there something. By doing this, on x86_64 the above numbers change to _Unwind_FrameState type being now 264 bytes long, frame_state_reg_info 208 bytes and we don't clear the first 144 bytes of the object, so the memset is 120 bytes, so ~ 31% of the old clearing size. On riscv 64-bit assuming it has same structure layout rules for the few types used there that would be ~ 2160 bytes of _Unwind_FrameState type before and ~ 1264 bytes after, with the memset previously ~ 2160 bytes and after ~ 232 bytes after. We've also talked about possibly adding a number of initially initialized regs and initializing the rest lazily, but at least for x86_64 with 18 elements in the array that doesn't seem to be worth it anymore, especially because return address column is 16 there and that is usually the first thing to be touched. It might theory help with lots of registers if they are usually untouched, but would uglify and complicate any stores to how by having to check there for the not initialized yet cases and lazy initialization, and similarly for all reads of how to do there if below last initialized one, use how, otherwise imply REG_UNSAVED. The disadvantage of the patch is that touching reg[x].loc and how[x] now means 2 cachelines rather than one as before, and I admit beyond bootstrap/regtest I haven't benchmarked it in any way. 2022-10-06 Jakub Jelinek <jakub@redhat.com> * unwind-dw2.h (REG_UNSAVED, REG_SAVED_OFFSET, REG_SAVED_REG, REG_SAVED_EXP, REG_SAVED_VAL_OFFSET, REG_SAVED_VAL_EXP, REG_UNDEFINED): New anonymous enum, moved from inside of struct frame_state_reg_info. (struct frame_state_reg_info): Remove reg[].how element and the anonymous enum there. Add how element. * unwind-dw2.c: Include stddef.h. (uw_frame_state_for): Don't clear first offsetof (_Unwind_FrameState, regs.how[0]) bytes of *fs. (execute_cfa_program, __frame_state_for, uw_update_context_1, uw_update_context): Use fs->regs.how[X] instead of fs->regs.reg[X].how or fs.regs.how[X] instead of fs.regs.reg[X].how. * config/sh/linux-unwind.h (sh_fallback_frame_state): Likewise. * config/bfin/linux-unwind.h (bfin_fallback_frame_state): Likewise. * config/pa/linux-unwind.h (pa32_fallback_frame_state): Likewise. * config/pa/hpux-unwind.h (UPDATE_FS_FOR_SAR, UPDATE_FS_FOR_GR, UPDATE_FS_FOR_FR, UPDATE_FS_FOR_PC, pa_fallback_frame_state): Likewise. * config/alpha/vms-unwind.h (alpha_vms_fallback_frame_state): Likewise. * config/alpha/linux-unwind.h (alpha_fallback_frame_state): Likewise. * config/arc/linux-unwind.h (arc_fallback_frame_state, arc_frob_update_context): Likewise. * config/riscv/linux-unwind.h (riscv_fallback_frame_state): Likewise. * config/nios2/linux-unwind.h (NIOS2_REG): Likewise. * config/nds32/linux-unwind.h (NDS32_PUT_FS_REG): Likewise. * config/s390/tpf-unwind.h (s390_fallback_frame_state): Likewise. * config/s390/linux-unwind.h (s390_fallback_frame_state): Likewise. * config/sparc/sol2-unwind.h (sparc64_frob_update_context, MD_FALLBACK_FRAME_STATE_FOR): Likewise. * config/sparc/linux-unwind.h (sparc64_fallback_frame_state, sparc64_frob_update_context, sparc_fallback_frame_state): Likewise. * config/i386/sol2-unwind.h (x86_64_fallback_frame_state, x86_fallback_frame_state): Likewise. * config/i386/w32-unwind.h (i386_w32_fallback_frame_state): Likewise. * config/i386/linux-unwind.h (x86_64_fallback_frame_state, x86_fallback_frame_state): Likewise. * config/i386/freebsd-unwind.h (x86_64_freebsd_fallback_frame_state): Likewise. * config/i386/dragonfly-unwind.h (x86_64_dragonfly_fallback_frame_state): Likewise. * config/i386/gnu-unwind.h (x86_gnu_fallback_frame_state): Likewise. * config/csky/linux-unwind.h (csky_fallback_frame_state): Likewise. * config/aarch64/linux-unwind.h (aarch64_fallback_frame_state): Likewise. * config/aarch64/freebsd-unwind.h (aarch64_freebsd_fallback_frame_state): Likewise. * config/aarch64/aarch64-unwind.h (aarch64_frob_update_context): Likewise. * config/or1k/linux-unwind.h (or1k_fallback_frame_state): Likewise. * config/mips/linux-unwind.h (mips_fallback_frame_state): Likewise. * config/loongarch/linux-unwind.h (loongarch_fallback_frame_state): Likewise. * config/m68k/linux-unwind.h (m68k_fallback_frame_state): Likewise. * config/xtensa/linux-unwind.h (xtensa_fallback_frame_state): Likewise. * config/rs6000/darwin-fallback.c (set_offset): Likewise. * config/rs6000/aix-unwind.h (MD_FROB_UPDATE_CONTEXT): Likewise. * config/rs6000/linux-unwind.h (ppc_fallback_frame_state): Likewise. * config/rs6000/freebsd-unwind.h (frob_update_context): Likewise.
2022-10-06 11:04:52 +02:00
fs->regs.how[3] = REG_SAVED_OFFSET;
fs->regs.reg[3].loc.offset = (long)&sc->ebx - new_cfa;
libgcc: Decrease size of _Unwind_FrameState and even more size of cleared area in uw_frame_state_for The following patch implements something that has Florian found as low hanging fruit in our unwinder and has been discussed in the https://gcc.gnu.org/wiki/cauldron2022#cauldron2022talks.inprocess_unwinding_bof talk. _Unwind_FrameState type seems to be (unlike the pre-GCC 3 frame_state which has been part of ABI) private to unwind-dw2.c + unwind.inc it includes, it is always defined on the stack of some entrypoints, initialized by static uw_frame_state_for and the address of it is also passed to other static functions or the static inlines handling machine dependent unwinding, but it isn't fortunately passed to any callbacks or public functions, so I think we can safely change it any time we want. Florian mentioned that the structure is large even on x86_64, 384 bytes there, starts with 328 bytes long element with frame_state_reg_info type which then starts with an array with __LIBGCC_DWARF_FRAME_REGISTERS__ + 1 elements, each of them is 16 bytes long, on x86_64 __LIBGCC_DWARF_FRAME_REGISTERS__ is just 17 but even that is big, on say riscv __LIBGCC_DWARF_FRAME_REGISTERS__ is I think 128, on powerpc 111, on sh 153 etc. And, we memset to zero the whole fs variable with the _Unwind_FrameState type at the start of the unwinding. The reason why each element is 16 byte (on 64-bit arches) is that it contains some pointer or pointer sized integer and then an enum (with just 7 different enumerators) + padding. The following patch decreases it by moving the enum into a separate array and using just one byte for each register in that second array. We could compress it even more, say 4 bits per register, but I don't want to uglify the code for it too much and make the accesses slower. Furthermore, the clearing of the object can clear only thos how array and members after it, because REG_UNSAVED enumerator (0) doesn't actually need any pointer or pointer sized integer, it is just the other kinds that need to have there something. By doing this, on x86_64 the above numbers change to _Unwind_FrameState type being now 264 bytes long, frame_state_reg_info 208 bytes and we don't clear the first 144 bytes of the object, so the memset is 120 bytes, so ~ 31% of the old clearing size. On riscv 64-bit assuming it has same structure layout rules for the few types used there that would be ~ 2160 bytes of _Unwind_FrameState type before and ~ 1264 bytes after, with the memset previously ~ 2160 bytes and after ~ 232 bytes after. We've also talked about possibly adding a number of initially initialized regs and initializing the rest lazily, but at least for x86_64 with 18 elements in the array that doesn't seem to be worth it anymore, especially because return address column is 16 there and that is usually the first thing to be touched. It might theory help with lots of registers if they are usually untouched, but would uglify and complicate any stores to how by having to check there for the not initialized yet cases and lazy initialization, and similarly for all reads of how to do there if below last initialized one, use how, otherwise imply REG_UNSAVED. The disadvantage of the patch is that touching reg[x].loc and how[x] now means 2 cachelines rather than one as before, and I admit beyond bootstrap/regtest I haven't benchmarked it in any way. 2022-10-06 Jakub Jelinek <jakub@redhat.com> * unwind-dw2.h (REG_UNSAVED, REG_SAVED_OFFSET, REG_SAVED_REG, REG_SAVED_EXP, REG_SAVED_VAL_OFFSET, REG_SAVED_VAL_EXP, REG_UNDEFINED): New anonymous enum, moved from inside of struct frame_state_reg_info. (struct frame_state_reg_info): Remove reg[].how element and the anonymous enum there. Add how element. * unwind-dw2.c: Include stddef.h. (uw_frame_state_for): Don't clear first offsetof (_Unwind_FrameState, regs.how[0]) bytes of *fs. (execute_cfa_program, __frame_state_for, uw_update_context_1, uw_update_context): Use fs->regs.how[X] instead of fs->regs.reg[X].how or fs.regs.how[X] instead of fs.regs.reg[X].how. * config/sh/linux-unwind.h (sh_fallback_frame_state): Likewise. * config/bfin/linux-unwind.h (bfin_fallback_frame_state): Likewise. * config/pa/linux-unwind.h (pa32_fallback_frame_state): Likewise. * config/pa/hpux-unwind.h (UPDATE_FS_FOR_SAR, UPDATE_FS_FOR_GR, UPDATE_FS_FOR_FR, UPDATE_FS_FOR_PC, pa_fallback_frame_state): Likewise. * config/alpha/vms-unwind.h (alpha_vms_fallback_frame_state): Likewise. * config/alpha/linux-unwind.h (alpha_fallback_frame_state): Likewise. * config/arc/linux-unwind.h (arc_fallback_frame_state, arc_frob_update_context): Likewise. * config/riscv/linux-unwind.h (riscv_fallback_frame_state): Likewise. * config/nios2/linux-unwind.h (NIOS2_REG): Likewise. * config/nds32/linux-unwind.h (NDS32_PUT_FS_REG): Likewise. * config/s390/tpf-unwind.h (s390_fallback_frame_state): Likewise. * config/s390/linux-unwind.h (s390_fallback_frame_state): Likewise. * config/sparc/sol2-unwind.h (sparc64_frob_update_context, MD_FALLBACK_FRAME_STATE_FOR): Likewise. * config/sparc/linux-unwind.h (sparc64_fallback_frame_state, sparc64_frob_update_context, sparc_fallback_frame_state): Likewise. * config/i386/sol2-unwind.h (x86_64_fallback_frame_state, x86_fallback_frame_state): Likewise. * config/i386/w32-unwind.h (i386_w32_fallback_frame_state): Likewise. * config/i386/linux-unwind.h (x86_64_fallback_frame_state, x86_fallback_frame_state): Likewise. * config/i386/freebsd-unwind.h (x86_64_freebsd_fallback_frame_state): Likewise. * config/i386/dragonfly-unwind.h (x86_64_dragonfly_fallback_frame_state): Likewise. * config/i386/gnu-unwind.h (x86_gnu_fallback_frame_state): Likewise. * config/csky/linux-unwind.h (csky_fallback_frame_state): Likewise. * config/aarch64/linux-unwind.h (aarch64_fallback_frame_state): Likewise. * config/aarch64/freebsd-unwind.h (aarch64_freebsd_fallback_frame_state): Likewise. * config/aarch64/aarch64-unwind.h (aarch64_frob_update_context): Likewise. * config/or1k/linux-unwind.h (or1k_fallback_frame_state): Likewise. * config/mips/linux-unwind.h (mips_fallback_frame_state): Likewise. * config/loongarch/linux-unwind.h (loongarch_fallback_frame_state): Likewise. * config/m68k/linux-unwind.h (m68k_fallback_frame_state): Likewise. * config/xtensa/linux-unwind.h (xtensa_fallback_frame_state): Likewise. * config/rs6000/darwin-fallback.c (set_offset): Likewise. * config/rs6000/aix-unwind.h (MD_FROB_UPDATE_CONTEXT): Likewise. * config/rs6000/linux-unwind.h (ppc_fallback_frame_state): Likewise. * config/rs6000/freebsd-unwind.h (frob_update_context): Likewise.
2022-10-06 11:04:52 +02:00
fs->regs.how[1] = REG_SAVED_OFFSET;
fs->regs.reg[1].loc.offset = (long)&sc->ecx - new_cfa;
libgcc: Decrease size of _Unwind_FrameState and even more size of cleared area in uw_frame_state_for The following patch implements something that has Florian found as low hanging fruit in our unwinder and has been discussed in the https://gcc.gnu.org/wiki/cauldron2022#cauldron2022talks.inprocess_unwinding_bof talk. _Unwind_FrameState type seems to be (unlike the pre-GCC 3 frame_state which has been part of ABI) private to unwind-dw2.c + unwind.inc it includes, it is always defined on the stack of some entrypoints, initialized by static uw_frame_state_for and the address of it is also passed to other static functions or the static inlines handling machine dependent unwinding, but it isn't fortunately passed to any callbacks or public functions, so I think we can safely change it any time we want. Florian mentioned that the structure is large even on x86_64, 384 bytes there, starts with 328 bytes long element with frame_state_reg_info type which then starts with an array with __LIBGCC_DWARF_FRAME_REGISTERS__ + 1 elements, each of them is 16 bytes long, on x86_64 __LIBGCC_DWARF_FRAME_REGISTERS__ is just 17 but even that is big, on say riscv __LIBGCC_DWARF_FRAME_REGISTERS__ is I think 128, on powerpc 111, on sh 153 etc. And, we memset to zero the whole fs variable with the _Unwind_FrameState type at the start of the unwinding. The reason why each element is 16 byte (on 64-bit arches) is that it contains some pointer or pointer sized integer and then an enum (with just 7 different enumerators) + padding. The following patch decreases it by moving the enum into a separate array and using just one byte for each register in that second array. We could compress it even more, say 4 bits per register, but I don't want to uglify the code for it too much and make the accesses slower. Furthermore, the clearing of the object can clear only thos how array and members after it, because REG_UNSAVED enumerator (0) doesn't actually need any pointer or pointer sized integer, it is just the other kinds that need to have there something. By doing this, on x86_64 the above numbers change to _Unwind_FrameState type being now 264 bytes long, frame_state_reg_info 208 bytes and we don't clear the first 144 bytes of the object, so the memset is 120 bytes, so ~ 31% of the old clearing size. On riscv 64-bit assuming it has same structure layout rules for the few types used there that would be ~ 2160 bytes of _Unwind_FrameState type before and ~ 1264 bytes after, with the memset previously ~ 2160 bytes and after ~ 232 bytes after. We've also talked about possibly adding a number of initially initialized regs and initializing the rest lazily, but at least for x86_64 with 18 elements in the array that doesn't seem to be worth it anymore, especially because return address column is 16 there and that is usually the first thing to be touched. It might theory help with lots of registers if they are usually untouched, but would uglify and complicate any stores to how by having to check there for the not initialized yet cases and lazy initialization, and similarly for all reads of how to do there if below last initialized one, use how, otherwise imply REG_UNSAVED. The disadvantage of the patch is that touching reg[x].loc and how[x] now means 2 cachelines rather than one as before, and I admit beyond bootstrap/regtest I haven't benchmarked it in any way. 2022-10-06 Jakub Jelinek <jakub@redhat.com> * unwind-dw2.h (REG_UNSAVED, REG_SAVED_OFFSET, REG_SAVED_REG, REG_SAVED_EXP, REG_SAVED_VAL_OFFSET, REG_SAVED_VAL_EXP, REG_UNDEFINED): New anonymous enum, moved from inside of struct frame_state_reg_info. (struct frame_state_reg_info): Remove reg[].how element and the anonymous enum there. Add how element. * unwind-dw2.c: Include stddef.h. (uw_frame_state_for): Don't clear first offsetof (_Unwind_FrameState, regs.how[0]) bytes of *fs. (execute_cfa_program, __frame_state_for, uw_update_context_1, uw_update_context): Use fs->regs.how[X] instead of fs->regs.reg[X].how or fs.regs.how[X] instead of fs.regs.reg[X].how. * config/sh/linux-unwind.h (sh_fallback_frame_state): Likewise. * config/bfin/linux-unwind.h (bfin_fallback_frame_state): Likewise. * config/pa/linux-unwind.h (pa32_fallback_frame_state): Likewise. * config/pa/hpux-unwind.h (UPDATE_FS_FOR_SAR, UPDATE_FS_FOR_GR, UPDATE_FS_FOR_FR, UPDATE_FS_FOR_PC, pa_fallback_frame_state): Likewise. * config/alpha/vms-unwind.h (alpha_vms_fallback_frame_state): Likewise. * config/alpha/linux-unwind.h (alpha_fallback_frame_state): Likewise. * config/arc/linux-unwind.h (arc_fallback_frame_state, arc_frob_update_context): Likewise. * config/riscv/linux-unwind.h (riscv_fallback_frame_state): Likewise. * config/nios2/linux-unwind.h (NIOS2_REG): Likewise. * config/nds32/linux-unwind.h (NDS32_PUT_FS_REG): Likewise. * config/s390/tpf-unwind.h (s390_fallback_frame_state): Likewise. * config/s390/linux-unwind.h (s390_fallback_frame_state): Likewise. * config/sparc/sol2-unwind.h (sparc64_frob_update_context, MD_FALLBACK_FRAME_STATE_FOR): Likewise. * config/sparc/linux-unwind.h (sparc64_fallback_frame_state, sparc64_frob_update_context, sparc_fallback_frame_state): Likewise. * config/i386/sol2-unwind.h (x86_64_fallback_frame_state, x86_fallback_frame_state): Likewise. * config/i386/w32-unwind.h (i386_w32_fallback_frame_state): Likewise. * config/i386/linux-unwind.h (x86_64_fallback_frame_state, x86_fallback_frame_state): Likewise. * config/i386/freebsd-unwind.h (x86_64_freebsd_fallback_frame_state): Likewise. * config/i386/dragonfly-unwind.h (x86_64_dragonfly_fallback_frame_state): Likewise. * config/i386/gnu-unwind.h (x86_gnu_fallback_frame_state): Likewise. * config/csky/linux-unwind.h (csky_fallback_frame_state): Likewise. * config/aarch64/linux-unwind.h (aarch64_fallback_frame_state): Likewise. * config/aarch64/freebsd-unwind.h (aarch64_freebsd_fallback_frame_state): Likewise. * config/aarch64/aarch64-unwind.h (aarch64_frob_update_context): Likewise. * config/or1k/linux-unwind.h (or1k_fallback_frame_state): Likewise. * config/mips/linux-unwind.h (mips_fallback_frame_state): Likewise. * config/loongarch/linux-unwind.h (loongarch_fallback_frame_state): Likewise. * config/m68k/linux-unwind.h (m68k_fallback_frame_state): Likewise. * config/xtensa/linux-unwind.h (xtensa_fallback_frame_state): Likewise. * config/rs6000/darwin-fallback.c (set_offset): Likewise. * config/rs6000/aix-unwind.h (MD_FROB_UPDATE_CONTEXT): Likewise. * config/rs6000/linux-unwind.h (ppc_fallback_frame_state): Likewise. * config/rs6000/freebsd-unwind.h (frob_update_context): Likewise.
2022-10-06 11:04:52 +02:00
fs->regs.how[2] = REG_SAVED_OFFSET;
fs->regs.reg[2].loc.offset = (long)&sc->edx - new_cfa;
libgcc: Decrease size of _Unwind_FrameState and even more size of cleared area in uw_frame_state_for The following patch implements something that has Florian found as low hanging fruit in our unwinder and has been discussed in the https://gcc.gnu.org/wiki/cauldron2022#cauldron2022talks.inprocess_unwinding_bof talk. _Unwind_FrameState type seems to be (unlike the pre-GCC 3 frame_state which has been part of ABI) private to unwind-dw2.c + unwind.inc it includes, it is always defined on the stack of some entrypoints, initialized by static uw_frame_state_for and the address of it is also passed to other static functions or the static inlines handling machine dependent unwinding, but it isn't fortunately passed to any callbacks or public functions, so I think we can safely change it any time we want. Florian mentioned that the structure is large even on x86_64, 384 bytes there, starts with 328 bytes long element with frame_state_reg_info type which then starts with an array with __LIBGCC_DWARF_FRAME_REGISTERS__ + 1 elements, each of them is 16 bytes long, on x86_64 __LIBGCC_DWARF_FRAME_REGISTERS__ is just 17 but even that is big, on say riscv __LIBGCC_DWARF_FRAME_REGISTERS__ is I think 128, on powerpc 111, on sh 153 etc. And, we memset to zero the whole fs variable with the _Unwind_FrameState type at the start of the unwinding. The reason why each element is 16 byte (on 64-bit arches) is that it contains some pointer or pointer sized integer and then an enum (with just 7 different enumerators) + padding. The following patch decreases it by moving the enum into a separate array and using just one byte for each register in that second array. We could compress it even more, say 4 bits per register, but I don't want to uglify the code for it too much and make the accesses slower. Furthermore, the clearing of the object can clear only thos how array and members after it, because REG_UNSAVED enumerator (0) doesn't actually need any pointer or pointer sized integer, it is just the other kinds that need to have there something. By doing this, on x86_64 the above numbers change to _Unwind_FrameState type being now 264 bytes long, frame_state_reg_info 208 bytes and we don't clear the first 144 bytes of the object, so the memset is 120 bytes, so ~ 31% of the old clearing size. On riscv 64-bit assuming it has same structure layout rules for the few types used there that would be ~ 2160 bytes of _Unwind_FrameState type before and ~ 1264 bytes after, with the memset previously ~ 2160 bytes and after ~ 232 bytes after. We've also talked about possibly adding a number of initially initialized regs and initializing the rest lazily, but at least for x86_64 with 18 elements in the array that doesn't seem to be worth it anymore, especially because return address column is 16 there and that is usually the first thing to be touched. It might theory help with lots of registers if they are usually untouched, but would uglify and complicate any stores to how by having to check there for the not initialized yet cases and lazy initialization, and similarly for all reads of how to do there if below last initialized one, use how, otherwise imply REG_UNSAVED. The disadvantage of the patch is that touching reg[x].loc and how[x] now means 2 cachelines rather than one as before, and I admit beyond bootstrap/regtest I haven't benchmarked it in any way. 2022-10-06 Jakub Jelinek <jakub@redhat.com> * unwind-dw2.h (REG_UNSAVED, REG_SAVED_OFFSET, REG_SAVED_REG, REG_SAVED_EXP, REG_SAVED_VAL_OFFSET, REG_SAVED_VAL_EXP, REG_UNDEFINED): New anonymous enum, moved from inside of struct frame_state_reg_info. (struct frame_state_reg_info): Remove reg[].how element and the anonymous enum there. Add how element. * unwind-dw2.c: Include stddef.h. (uw_frame_state_for): Don't clear first offsetof (_Unwind_FrameState, regs.how[0]) bytes of *fs. (execute_cfa_program, __frame_state_for, uw_update_context_1, uw_update_context): Use fs->regs.how[X] instead of fs->regs.reg[X].how or fs.regs.how[X] instead of fs.regs.reg[X].how. * config/sh/linux-unwind.h (sh_fallback_frame_state): Likewise. * config/bfin/linux-unwind.h (bfin_fallback_frame_state): Likewise. * config/pa/linux-unwind.h (pa32_fallback_frame_state): Likewise. * config/pa/hpux-unwind.h (UPDATE_FS_FOR_SAR, UPDATE_FS_FOR_GR, UPDATE_FS_FOR_FR, UPDATE_FS_FOR_PC, pa_fallback_frame_state): Likewise. * config/alpha/vms-unwind.h (alpha_vms_fallback_frame_state): Likewise. * config/alpha/linux-unwind.h (alpha_fallback_frame_state): Likewise. * config/arc/linux-unwind.h (arc_fallback_frame_state, arc_frob_update_context): Likewise. * config/riscv/linux-unwind.h (riscv_fallback_frame_state): Likewise. * config/nios2/linux-unwind.h (NIOS2_REG): Likewise. * config/nds32/linux-unwind.h (NDS32_PUT_FS_REG): Likewise. * config/s390/tpf-unwind.h (s390_fallback_frame_state): Likewise. * config/s390/linux-unwind.h (s390_fallback_frame_state): Likewise. * config/sparc/sol2-unwind.h (sparc64_frob_update_context, MD_FALLBACK_FRAME_STATE_FOR): Likewise. * config/sparc/linux-unwind.h (sparc64_fallback_frame_state, sparc64_frob_update_context, sparc_fallback_frame_state): Likewise. * config/i386/sol2-unwind.h (x86_64_fallback_frame_state, x86_fallback_frame_state): Likewise. * config/i386/w32-unwind.h (i386_w32_fallback_frame_state): Likewise. * config/i386/linux-unwind.h (x86_64_fallback_frame_state, x86_fallback_frame_state): Likewise. * config/i386/freebsd-unwind.h (x86_64_freebsd_fallback_frame_state): Likewise. * config/i386/dragonfly-unwind.h (x86_64_dragonfly_fallback_frame_state): Likewise. * config/i386/gnu-unwind.h (x86_gnu_fallback_frame_state): Likewise. * config/csky/linux-unwind.h (csky_fallback_frame_state): Likewise. * config/aarch64/linux-unwind.h (aarch64_fallback_frame_state): Likewise. * config/aarch64/freebsd-unwind.h (aarch64_freebsd_fallback_frame_state): Likewise. * config/aarch64/aarch64-unwind.h (aarch64_frob_update_context): Likewise. * config/or1k/linux-unwind.h (or1k_fallback_frame_state): Likewise. * config/mips/linux-unwind.h (mips_fallback_frame_state): Likewise. * config/loongarch/linux-unwind.h (loongarch_fallback_frame_state): Likewise. * config/m68k/linux-unwind.h (m68k_fallback_frame_state): Likewise. * config/xtensa/linux-unwind.h (xtensa_fallback_frame_state): Likewise. * config/rs6000/darwin-fallback.c (set_offset): Likewise. * config/rs6000/aix-unwind.h (MD_FROB_UPDATE_CONTEXT): Likewise. * config/rs6000/linux-unwind.h (ppc_fallback_frame_state): Likewise. * config/rs6000/freebsd-unwind.h (frob_update_context): Likewise.
2022-10-06 11:04:52 +02:00
fs->regs.how[6] = REG_SAVED_OFFSET;
fs->regs.reg[6].loc.offset = (long)&sc->esi - new_cfa;
libgcc: Decrease size of _Unwind_FrameState and even more size of cleared area in uw_frame_state_for The following patch implements something that has Florian found as low hanging fruit in our unwinder and has been discussed in the https://gcc.gnu.org/wiki/cauldron2022#cauldron2022talks.inprocess_unwinding_bof talk. _Unwind_FrameState type seems to be (unlike the pre-GCC 3 frame_state which has been part of ABI) private to unwind-dw2.c + unwind.inc it includes, it is always defined on the stack of some entrypoints, initialized by static uw_frame_state_for and the address of it is also passed to other static functions or the static inlines handling machine dependent unwinding, but it isn't fortunately passed to any callbacks or public functions, so I think we can safely change it any time we want. Florian mentioned that the structure is large even on x86_64, 384 bytes there, starts with 328 bytes long element with frame_state_reg_info type which then starts with an array with __LIBGCC_DWARF_FRAME_REGISTERS__ + 1 elements, each of them is 16 bytes long, on x86_64 __LIBGCC_DWARF_FRAME_REGISTERS__ is just 17 but even that is big, on say riscv __LIBGCC_DWARF_FRAME_REGISTERS__ is I think 128, on powerpc 111, on sh 153 etc. And, we memset to zero the whole fs variable with the _Unwind_FrameState type at the start of the unwinding. The reason why each element is 16 byte (on 64-bit arches) is that it contains some pointer or pointer sized integer and then an enum (with just 7 different enumerators) + padding. The following patch decreases it by moving the enum into a separate array and using just one byte for each register in that second array. We could compress it even more, say 4 bits per register, but I don't want to uglify the code for it too much and make the accesses slower. Furthermore, the clearing of the object can clear only thos how array and members after it, because REG_UNSAVED enumerator (0) doesn't actually need any pointer or pointer sized integer, it is just the other kinds that need to have there something. By doing this, on x86_64 the above numbers change to _Unwind_FrameState type being now 264 bytes long, frame_state_reg_info 208 bytes and we don't clear the first 144 bytes of the object, so the memset is 120 bytes, so ~ 31% of the old clearing size. On riscv 64-bit assuming it has same structure layout rules for the few types used there that would be ~ 2160 bytes of _Unwind_FrameState type before and ~ 1264 bytes after, with the memset previously ~ 2160 bytes and after ~ 232 bytes after. We've also talked about possibly adding a number of initially initialized regs and initializing the rest lazily, but at least for x86_64 with 18 elements in the array that doesn't seem to be worth it anymore, especially because return address column is 16 there and that is usually the first thing to be touched. It might theory help with lots of registers if they are usually untouched, but would uglify and complicate any stores to how by having to check there for the not initialized yet cases and lazy initialization, and similarly for all reads of how to do there if below last initialized one, use how, otherwise imply REG_UNSAVED. The disadvantage of the patch is that touching reg[x].loc and how[x] now means 2 cachelines rather than one as before, and I admit beyond bootstrap/regtest I haven't benchmarked it in any way. 2022-10-06 Jakub Jelinek <jakub@redhat.com> * unwind-dw2.h (REG_UNSAVED, REG_SAVED_OFFSET, REG_SAVED_REG, REG_SAVED_EXP, REG_SAVED_VAL_OFFSET, REG_SAVED_VAL_EXP, REG_UNDEFINED): New anonymous enum, moved from inside of struct frame_state_reg_info. (struct frame_state_reg_info): Remove reg[].how element and the anonymous enum there. Add how element. * unwind-dw2.c: Include stddef.h. (uw_frame_state_for): Don't clear first offsetof (_Unwind_FrameState, regs.how[0]) bytes of *fs. (execute_cfa_program, __frame_state_for, uw_update_context_1, uw_update_context): Use fs->regs.how[X] instead of fs->regs.reg[X].how or fs.regs.how[X] instead of fs.regs.reg[X].how. * config/sh/linux-unwind.h (sh_fallback_frame_state): Likewise. * config/bfin/linux-unwind.h (bfin_fallback_frame_state): Likewise. * config/pa/linux-unwind.h (pa32_fallback_frame_state): Likewise. * config/pa/hpux-unwind.h (UPDATE_FS_FOR_SAR, UPDATE_FS_FOR_GR, UPDATE_FS_FOR_FR, UPDATE_FS_FOR_PC, pa_fallback_frame_state): Likewise. * config/alpha/vms-unwind.h (alpha_vms_fallback_frame_state): Likewise. * config/alpha/linux-unwind.h (alpha_fallback_frame_state): Likewise. * config/arc/linux-unwind.h (arc_fallback_frame_state, arc_frob_update_context): Likewise. * config/riscv/linux-unwind.h (riscv_fallback_frame_state): Likewise. * config/nios2/linux-unwind.h (NIOS2_REG): Likewise. * config/nds32/linux-unwind.h (NDS32_PUT_FS_REG): Likewise. * config/s390/tpf-unwind.h (s390_fallback_frame_state): Likewise. * config/s390/linux-unwind.h (s390_fallback_frame_state): Likewise. * config/sparc/sol2-unwind.h (sparc64_frob_update_context, MD_FALLBACK_FRAME_STATE_FOR): Likewise. * config/sparc/linux-unwind.h (sparc64_fallback_frame_state, sparc64_frob_update_context, sparc_fallback_frame_state): Likewise. * config/i386/sol2-unwind.h (x86_64_fallback_frame_state, x86_fallback_frame_state): Likewise. * config/i386/w32-unwind.h (i386_w32_fallback_frame_state): Likewise. * config/i386/linux-unwind.h (x86_64_fallback_frame_state, x86_fallback_frame_state): Likewise. * config/i386/freebsd-unwind.h (x86_64_freebsd_fallback_frame_state): Likewise. * config/i386/dragonfly-unwind.h (x86_64_dragonfly_fallback_frame_state): Likewise. * config/i386/gnu-unwind.h (x86_gnu_fallback_frame_state): Likewise. * config/csky/linux-unwind.h (csky_fallback_frame_state): Likewise. * config/aarch64/linux-unwind.h (aarch64_fallback_frame_state): Likewise. * config/aarch64/freebsd-unwind.h (aarch64_freebsd_fallback_frame_state): Likewise. * config/aarch64/aarch64-unwind.h (aarch64_frob_update_context): Likewise. * config/or1k/linux-unwind.h (or1k_fallback_frame_state): Likewise. * config/mips/linux-unwind.h (mips_fallback_frame_state): Likewise. * config/loongarch/linux-unwind.h (loongarch_fallback_frame_state): Likewise. * config/m68k/linux-unwind.h (m68k_fallback_frame_state): Likewise. * config/xtensa/linux-unwind.h (xtensa_fallback_frame_state): Likewise. * config/rs6000/darwin-fallback.c (set_offset): Likewise. * config/rs6000/aix-unwind.h (MD_FROB_UPDATE_CONTEXT): Likewise. * config/rs6000/linux-unwind.h (ppc_fallback_frame_state): Likewise. * config/rs6000/freebsd-unwind.h (frob_update_context): Likewise.
2022-10-06 11:04:52 +02:00
fs->regs.how[7] = REG_SAVED_OFFSET;
fs->regs.reg[7].loc.offset = (long)&sc->edi - new_cfa;
libgcc: Decrease size of _Unwind_FrameState and even more size of cleared area in uw_frame_state_for The following patch implements something that has Florian found as low hanging fruit in our unwinder and has been discussed in the https://gcc.gnu.org/wiki/cauldron2022#cauldron2022talks.inprocess_unwinding_bof talk. _Unwind_FrameState type seems to be (unlike the pre-GCC 3 frame_state which has been part of ABI) private to unwind-dw2.c + unwind.inc it includes, it is always defined on the stack of some entrypoints, initialized by static uw_frame_state_for and the address of it is also passed to other static functions or the static inlines handling machine dependent unwinding, but it isn't fortunately passed to any callbacks or public functions, so I think we can safely change it any time we want. Florian mentioned that the structure is large even on x86_64, 384 bytes there, starts with 328 bytes long element with frame_state_reg_info type which then starts with an array with __LIBGCC_DWARF_FRAME_REGISTERS__ + 1 elements, each of them is 16 bytes long, on x86_64 __LIBGCC_DWARF_FRAME_REGISTERS__ is just 17 but even that is big, on say riscv __LIBGCC_DWARF_FRAME_REGISTERS__ is I think 128, on powerpc 111, on sh 153 etc. And, we memset to zero the whole fs variable with the _Unwind_FrameState type at the start of the unwinding. The reason why each element is 16 byte (on 64-bit arches) is that it contains some pointer or pointer sized integer and then an enum (with just 7 different enumerators) + padding. The following patch decreases it by moving the enum into a separate array and using just one byte for each register in that second array. We could compress it even more, say 4 bits per register, but I don't want to uglify the code for it too much and make the accesses slower. Furthermore, the clearing of the object can clear only thos how array and members after it, because REG_UNSAVED enumerator (0) doesn't actually need any pointer or pointer sized integer, it is just the other kinds that need to have there something. By doing this, on x86_64 the above numbers change to _Unwind_FrameState type being now 264 bytes long, frame_state_reg_info 208 bytes and we don't clear the first 144 bytes of the object, so the memset is 120 bytes, so ~ 31% of the old clearing size. On riscv 64-bit assuming it has same structure layout rules for the few types used there that would be ~ 2160 bytes of _Unwind_FrameState type before and ~ 1264 bytes after, with the memset previously ~ 2160 bytes and after ~ 232 bytes after. We've also talked about possibly adding a number of initially initialized regs and initializing the rest lazily, but at least for x86_64 with 18 elements in the array that doesn't seem to be worth it anymore, especially because return address column is 16 there and that is usually the first thing to be touched. It might theory help with lots of registers if they are usually untouched, but would uglify and complicate any stores to how by having to check there for the not initialized yet cases and lazy initialization, and similarly for all reads of how to do there if below last initialized one, use how, otherwise imply REG_UNSAVED. The disadvantage of the patch is that touching reg[x].loc and how[x] now means 2 cachelines rather than one as before, and I admit beyond bootstrap/regtest I haven't benchmarked it in any way. 2022-10-06 Jakub Jelinek <jakub@redhat.com> * unwind-dw2.h (REG_UNSAVED, REG_SAVED_OFFSET, REG_SAVED_REG, REG_SAVED_EXP, REG_SAVED_VAL_OFFSET, REG_SAVED_VAL_EXP, REG_UNDEFINED): New anonymous enum, moved from inside of struct frame_state_reg_info. (struct frame_state_reg_info): Remove reg[].how element and the anonymous enum there. Add how element. * unwind-dw2.c: Include stddef.h. (uw_frame_state_for): Don't clear first offsetof (_Unwind_FrameState, regs.how[0]) bytes of *fs. (execute_cfa_program, __frame_state_for, uw_update_context_1, uw_update_context): Use fs->regs.how[X] instead of fs->regs.reg[X].how or fs.regs.how[X] instead of fs.regs.reg[X].how. * config/sh/linux-unwind.h (sh_fallback_frame_state): Likewise. * config/bfin/linux-unwind.h (bfin_fallback_frame_state): Likewise. * config/pa/linux-unwind.h (pa32_fallback_frame_state): Likewise. * config/pa/hpux-unwind.h (UPDATE_FS_FOR_SAR, UPDATE_FS_FOR_GR, UPDATE_FS_FOR_FR, UPDATE_FS_FOR_PC, pa_fallback_frame_state): Likewise. * config/alpha/vms-unwind.h (alpha_vms_fallback_frame_state): Likewise. * config/alpha/linux-unwind.h (alpha_fallback_frame_state): Likewise. * config/arc/linux-unwind.h (arc_fallback_frame_state, arc_frob_update_context): Likewise. * config/riscv/linux-unwind.h (riscv_fallback_frame_state): Likewise. * config/nios2/linux-unwind.h (NIOS2_REG): Likewise. * config/nds32/linux-unwind.h (NDS32_PUT_FS_REG): Likewise. * config/s390/tpf-unwind.h (s390_fallback_frame_state): Likewise. * config/s390/linux-unwind.h (s390_fallback_frame_state): Likewise. * config/sparc/sol2-unwind.h (sparc64_frob_update_context, MD_FALLBACK_FRAME_STATE_FOR): Likewise. * config/sparc/linux-unwind.h (sparc64_fallback_frame_state, sparc64_frob_update_context, sparc_fallback_frame_state): Likewise. * config/i386/sol2-unwind.h (x86_64_fallback_frame_state, x86_fallback_frame_state): Likewise. * config/i386/w32-unwind.h (i386_w32_fallback_frame_state): Likewise. * config/i386/linux-unwind.h (x86_64_fallback_frame_state, x86_fallback_frame_state): Likewise. * config/i386/freebsd-unwind.h (x86_64_freebsd_fallback_frame_state): Likewise. * config/i386/dragonfly-unwind.h (x86_64_dragonfly_fallback_frame_state): Likewise. * config/i386/gnu-unwind.h (x86_gnu_fallback_frame_state): Likewise. * config/csky/linux-unwind.h (csky_fallback_frame_state): Likewise. * config/aarch64/linux-unwind.h (aarch64_fallback_frame_state): Likewise. * config/aarch64/freebsd-unwind.h (aarch64_freebsd_fallback_frame_state): Likewise. * config/aarch64/aarch64-unwind.h (aarch64_frob_update_context): Likewise. * config/or1k/linux-unwind.h (or1k_fallback_frame_state): Likewise. * config/mips/linux-unwind.h (mips_fallback_frame_state): Likewise. * config/loongarch/linux-unwind.h (loongarch_fallback_frame_state): Likewise. * config/m68k/linux-unwind.h (m68k_fallback_frame_state): Likewise. * config/xtensa/linux-unwind.h (xtensa_fallback_frame_state): Likewise. * config/rs6000/darwin-fallback.c (set_offset): Likewise. * config/rs6000/aix-unwind.h (MD_FROB_UPDATE_CONTEXT): Likewise. * config/rs6000/linux-unwind.h (ppc_fallback_frame_state): Likewise. * config/rs6000/freebsd-unwind.h (frob_update_context): Likewise.
2022-10-06 11:04:52 +02:00
fs->regs.how[5] = REG_SAVED_OFFSET;
fs->regs.reg[5].loc.offset = (long)&sc->ebp - new_cfa;
libgcc: Decrease size of _Unwind_FrameState and even more size of cleared area in uw_frame_state_for The following patch implements something that has Florian found as low hanging fruit in our unwinder and has been discussed in the https://gcc.gnu.org/wiki/cauldron2022#cauldron2022talks.inprocess_unwinding_bof talk. _Unwind_FrameState type seems to be (unlike the pre-GCC 3 frame_state which has been part of ABI) private to unwind-dw2.c + unwind.inc it includes, it is always defined on the stack of some entrypoints, initialized by static uw_frame_state_for and the address of it is also passed to other static functions or the static inlines handling machine dependent unwinding, but it isn't fortunately passed to any callbacks or public functions, so I think we can safely change it any time we want. Florian mentioned that the structure is large even on x86_64, 384 bytes there, starts with 328 bytes long element with frame_state_reg_info type which then starts with an array with __LIBGCC_DWARF_FRAME_REGISTERS__ + 1 elements, each of them is 16 bytes long, on x86_64 __LIBGCC_DWARF_FRAME_REGISTERS__ is just 17 but even that is big, on say riscv __LIBGCC_DWARF_FRAME_REGISTERS__ is I think 128, on powerpc 111, on sh 153 etc. And, we memset to zero the whole fs variable with the _Unwind_FrameState type at the start of the unwinding. The reason why each element is 16 byte (on 64-bit arches) is that it contains some pointer or pointer sized integer and then an enum (with just 7 different enumerators) + padding. The following patch decreases it by moving the enum into a separate array and using just one byte for each register in that second array. We could compress it even more, say 4 bits per register, but I don't want to uglify the code for it too much and make the accesses slower. Furthermore, the clearing of the object can clear only thos how array and members after it, because REG_UNSAVED enumerator (0) doesn't actually need any pointer or pointer sized integer, it is just the other kinds that need to have there something. By doing this, on x86_64 the above numbers change to _Unwind_FrameState type being now 264 bytes long, frame_state_reg_info 208 bytes and we don't clear the first 144 bytes of the object, so the memset is 120 bytes, so ~ 31% of the old clearing size. On riscv 64-bit assuming it has same structure layout rules for the few types used there that would be ~ 2160 bytes of _Unwind_FrameState type before and ~ 1264 bytes after, with the memset previously ~ 2160 bytes and after ~ 232 bytes after. We've also talked about possibly adding a number of initially initialized regs and initializing the rest lazily, but at least for x86_64 with 18 elements in the array that doesn't seem to be worth it anymore, especially because return address column is 16 there and that is usually the first thing to be touched. It might theory help with lots of registers if they are usually untouched, but would uglify and complicate any stores to how by having to check there for the not initialized yet cases and lazy initialization, and similarly for all reads of how to do there if below last initialized one, use how, otherwise imply REG_UNSAVED. The disadvantage of the patch is that touching reg[x].loc and how[x] now means 2 cachelines rather than one as before, and I admit beyond bootstrap/regtest I haven't benchmarked it in any way. 2022-10-06 Jakub Jelinek <jakub@redhat.com> * unwind-dw2.h (REG_UNSAVED, REG_SAVED_OFFSET, REG_SAVED_REG, REG_SAVED_EXP, REG_SAVED_VAL_OFFSET, REG_SAVED_VAL_EXP, REG_UNDEFINED): New anonymous enum, moved from inside of struct frame_state_reg_info. (struct frame_state_reg_info): Remove reg[].how element and the anonymous enum there. Add how element. * unwind-dw2.c: Include stddef.h. (uw_frame_state_for): Don't clear first offsetof (_Unwind_FrameState, regs.how[0]) bytes of *fs. (execute_cfa_program, __frame_state_for, uw_update_context_1, uw_update_context): Use fs->regs.how[X] instead of fs->regs.reg[X].how or fs.regs.how[X] instead of fs.regs.reg[X].how. * config/sh/linux-unwind.h (sh_fallback_frame_state): Likewise. * config/bfin/linux-unwind.h (bfin_fallback_frame_state): Likewise. * config/pa/linux-unwind.h (pa32_fallback_frame_state): Likewise. * config/pa/hpux-unwind.h (UPDATE_FS_FOR_SAR, UPDATE_FS_FOR_GR, UPDATE_FS_FOR_FR, UPDATE_FS_FOR_PC, pa_fallback_frame_state): Likewise. * config/alpha/vms-unwind.h (alpha_vms_fallback_frame_state): Likewise. * config/alpha/linux-unwind.h (alpha_fallback_frame_state): Likewise. * config/arc/linux-unwind.h (arc_fallback_frame_state, arc_frob_update_context): Likewise. * config/riscv/linux-unwind.h (riscv_fallback_frame_state): Likewise. * config/nios2/linux-unwind.h (NIOS2_REG): Likewise. * config/nds32/linux-unwind.h (NDS32_PUT_FS_REG): Likewise. * config/s390/tpf-unwind.h (s390_fallback_frame_state): Likewise. * config/s390/linux-unwind.h (s390_fallback_frame_state): Likewise. * config/sparc/sol2-unwind.h (sparc64_frob_update_context, MD_FALLBACK_FRAME_STATE_FOR): Likewise. * config/sparc/linux-unwind.h (sparc64_fallback_frame_state, sparc64_frob_update_context, sparc_fallback_frame_state): Likewise. * config/i386/sol2-unwind.h (x86_64_fallback_frame_state, x86_fallback_frame_state): Likewise. * config/i386/w32-unwind.h (i386_w32_fallback_frame_state): Likewise. * config/i386/linux-unwind.h (x86_64_fallback_frame_state, x86_fallback_frame_state): Likewise. * config/i386/freebsd-unwind.h (x86_64_freebsd_fallback_frame_state): Likewise. * config/i386/dragonfly-unwind.h (x86_64_dragonfly_fallback_frame_state): Likewise. * config/i386/gnu-unwind.h (x86_gnu_fallback_frame_state): Likewise. * config/csky/linux-unwind.h (csky_fallback_frame_state): Likewise. * config/aarch64/linux-unwind.h (aarch64_fallback_frame_state): Likewise. * config/aarch64/freebsd-unwind.h (aarch64_freebsd_fallback_frame_state): Likewise. * config/aarch64/aarch64-unwind.h (aarch64_frob_update_context): Likewise. * config/or1k/linux-unwind.h (or1k_fallback_frame_state): Likewise. * config/mips/linux-unwind.h (mips_fallback_frame_state): Likewise. * config/loongarch/linux-unwind.h (loongarch_fallback_frame_state): Likewise. * config/m68k/linux-unwind.h (m68k_fallback_frame_state): Likewise. * config/xtensa/linux-unwind.h (xtensa_fallback_frame_state): Likewise. * config/rs6000/darwin-fallback.c (set_offset): Likewise. * config/rs6000/aix-unwind.h (MD_FROB_UPDATE_CONTEXT): Likewise. * config/rs6000/linux-unwind.h (ppc_fallback_frame_state): Likewise. * config/rs6000/freebsd-unwind.h (frob_update_context): Likewise.
2022-10-06 11:04:52 +02:00
fs->regs.how[8] = REG_SAVED_OFFSET;
fs->regs.reg[8].loc.offset = (long)&sc->eip - new_cfa;
tm.texi (MD_UNWIND_SUPPORT): Document. * doc/tm.texi (MD_UNWIND_SUPPORT): Document. (MD_FALLBACK_FRAME_STATE_FOR): Update. * unwind-dw2.c (MD_UNWIND_SUPPORT): #include if defined. (uw_frame_state_for): Adjust MD_FALLBACK_FRAME_STATE_FOR invocation. (MD_FROB_UPDATE_CONTEXT): Remove default. (uw_update_context_1): Instead #ifdef invocation. * config/ia64/unwind-ia64.c (MD_UNWIND_SUPPORT): #include if defined. (uw_frame_state_for): Adjust MD_FALLBACK_FRAME_STATE_FOR invocation. * config/alpha/gnu.h (MD_FALLBACK_FRAME_STATE_FOR): Don't undef. (MD_UNWIND_SUPPORT): Undefine this instead. * config/i386/gnu.h: Likewise. * config/alpha/linux-unwind.h: New file, macro converted to function, extracted from.. * config/alpha/linux.h (MD_FALLBACK_FRAME_STATE_FOR): ..this. (MD_UNWIND_SUPPORT): Define. * config/alpha/vms-unwind.h, config/alpha/vms.h: Likewise. * config/i386/linux-unwind.h, config/i386/linux.h, config/i386/linux64.h: Likewise. * config/ia64/linux-unwind.h, config/ia64/linux.h: Likewise. MD_HANDLE_UNWABI too. * config/mips/linux-unwind.h, config/mips/linux.h: Likewise. * config/pa/linux-unwind.h, config/pa/pa32-linux.h: Likewise. * config/rs6000/darwin-unwind.h, config/rs6000/darwin.h: Likewise. * config/s390/linux-unwind.h, config/s390/linux.h: Likewise. * config/sparc/linux-unwind.h, config/sparc/linux.h, config/sparc/linux64.h: Likewise. * config/sh/linux-unwind.h, config/sh/linux.h: Likewise, but merge SH_FALLBACK_FRAME_FLOAT_STATE into sh_fallback_frame_state. * config/rs6000/linux-unwind.h, config/rs6000/linux.h, config/rs6000/linux64.h: Likewise. Split out get_sigcontext function. Use ARG_POINTER_REGNUM for 32-bit temp reg too. From-SVN: r87167
2004-09-08 00:17:19 +00:00
fs->retaddr_column = 8;
re PR other/26208 (Serious problem with unwinding through signal frames) PR other/26208 * unwind-dw2.c (struct _Unwind_Context): Add signal_frame field. (extract_cie_info): Handle S flag in augmentation string. (execute_cfa_program): If context->signal_frame, execute also fs->pc == context->ra instructions. (uw_frame_state_for): If context->signal_frame, don't subtract one from context->ra to find FDE. (uw_update_context_1): Set context->signal_frame to fs->signal_frame. (_Unwind_GetIPInfo): New function. * unwind-dw2.h (_Unwind_FrameState): Add signal_frame field. * unwind-c.c (PERSONALITY_FUNCTION): Use _Unwind_GetIPInfo instead of _Unwind_GetIP. * unwind-sjlj.c (_Unwind_GetIPInfo): New function. * unwind-generic.h (_Unwind_GetIPInfo): New prototype. * unwind-compat.c (_Unwind_GetIPInfo): New function. * libgcc-std.ver (_Unwind_GetIPInfo): Export @@GCC_4.2.0. * config/ia64/unwind-ia64.c (_Unwind_GetIPInfo): New function. * config/arm/unwind-arm.h (_Unwind_GetIPInfo): Define. * config/i386/linux-unwind.h (x86_fallback_frame_state, x86_64_fallback_frame_state): Set fs->signal_frame. * config/rs6000/linux-unwind.h (ppc_fallback_frame_state): Likewise. (MD_FROB_UPDATE_CONTEXT): Define unconditionally. (frob_update_context): Likewise. Workaround missing S flag in Linux 2.6.12 - 2.6.16 kernel vDSOs. * config/s390/linux-unwind.h (s390_fallback_frame_state): Likewise. Remove the psw_addr + 1 hack. libjava/ * exception.cc (PERSONALITY_FUNCTION): Use _Unwind_GetIPInfo instead of _Unwind_GetIP. * include/i386-signal.h (MAKE_THROW_FRAME): Change into empty macro. (HANDLE_DIVIDE_OVERFLOW): Don't adjust _res->eip if falling through to throw. * include/x86_64-signal.h (MAKE_THROW_FRAME): Change into empty macro. * include/powerpc-signal.h (MAKE_THROW_FRAME): Change into empty macro. libstdc++-v3/ * libsupc++/eh_personality.cc (PERSONALITY_FUNCTION): Use _Unwind_GetIPInfo instead of _Unwind_GetIP. From-SVN: r111488
2006-02-27 18:26:26 +01:00
fs->signal_frame = 1;
tm.texi (MD_UNWIND_SUPPORT): Document. * doc/tm.texi (MD_UNWIND_SUPPORT): Document. (MD_FALLBACK_FRAME_STATE_FOR): Update. * unwind-dw2.c (MD_UNWIND_SUPPORT): #include if defined. (uw_frame_state_for): Adjust MD_FALLBACK_FRAME_STATE_FOR invocation. (MD_FROB_UPDATE_CONTEXT): Remove default. (uw_update_context_1): Instead #ifdef invocation. * config/ia64/unwind-ia64.c (MD_UNWIND_SUPPORT): #include if defined. (uw_frame_state_for): Adjust MD_FALLBACK_FRAME_STATE_FOR invocation. * config/alpha/gnu.h (MD_FALLBACK_FRAME_STATE_FOR): Don't undef. (MD_UNWIND_SUPPORT): Undefine this instead. * config/i386/gnu.h: Likewise. * config/alpha/linux-unwind.h: New file, macro converted to function, extracted from.. * config/alpha/linux.h (MD_FALLBACK_FRAME_STATE_FOR): ..this. (MD_UNWIND_SUPPORT): Define. * config/alpha/vms-unwind.h, config/alpha/vms.h: Likewise. * config/i386/linux-unwind.h, config/i386/linux.h, config/i386/linux64.h: Likewise. * config/ia64/linux-unwind.h, config/ia64/linux.h: Likewise. MD_HANDLE_UNWABI too. * config/mips/linux-unwind.h, config/mips/linux.h: Likewise. * config/pa/linux-unwind.h, config/pa/pa32-linux.h: Likewise. * config/rs6000/darwin-unwind.h, config/rs6000/darwin.h: Likewise. * config/s390/linux-unwind.h, config/s390/linux.h: Likewise. * config/sparc/linux-unwind.h, config/sparc/linux.h, config/sparc/linux64.h: Likewise. * config/sh/linux-unwind.h, config/sh/linux.h: Likewise, but merge SH_FALLBACK_FRAME_FLOAT_STATE into sh_fallback_frame_state. * config/rs6000/linux-unwind.h, config/rs6000/linux.h, config/rs6000/linux64.h: Likewise. Split out get_sigcontext function. Use ARG_POINTER_REGNUM for 32-bit temp reg too. From-SVN: r87167
2004-09-08 00:17:19 +00:00
return _URC_NO_REASON;
}
#define MD_FROB_UPDATE_CONTEXT x86_frob_update_context
/* Fix up for kernels that have vDSO, but don't have S flag in it. */
static void
x86_frob_update_context (struct _Unwind_Context *context,
_Unwind_FrameState *fs ATTRIBUTE_UNUSED)
{
unsigned char *pc = context->ra;
/* movl $__NR_rt_sigreturn,%eax ; {int $0x80 | syscall} */
if (*(unsigned char *)(pc+0) == 0xb8
&& *(unsigned int *)(pc+1) == 173
&& (*(unsigned short *)(pc+5) == 0x80cd
|| *(unsigned short *)(pc+5) == 0x050f))
_Unwind_SetSignalFrame (context, 1);
}
tm.texi (MD_UNWIND_SUPPORT): Document. * doc/tm.texi (MD_UNWIND_SUPPORT): Document. (MD_FALLBACK_FRAME_STATE_FOR): Update. * unwind-dw2.c (MD_UNWIND_SUPPORT): #include if defined. (uw_frame_state_for): Adjust MD_FALLBACK_FRAME_STATE_FOR invocation. (MD_FROB_UPDATE_CONTEXT): Remove default. (uw_update_context_1): Instead #ifdef invocation. * config/ia64/unwind-ia64.c (MD_UNWIND_SUPPORT): #include if defined. (uw_frame_state_for): Adjust MD_FALLBACK_FRAME_STATE_FOR invocation. * config/alpha/gnu.h (MD_FALLBACK_FRAME_STATE_FOR): Don't undef. (MD_UNWIND_SUPPORT): Undefine this instead. * config/i386/gnu.h: Likewise. * config/alpha/linux-unwind.h: New file, macro converted to function, extracted from.. * config/alpha/linux.h (MD_FALLBACK_FRAME_STATE_FOR): ..this. (MD_UNWIND_SUPPORT): Define. * config/alpha/vms-unwind.h, config/alpha/vms.h: Likewise. * config/i386/linux-unwind.h, config/i386/linux.h, config/i386/linux64.h: Likewise. * config/ia64/linux-unwind.h, config/ia64/linux.h: Likewise. MD_HANDLE_UNWABI too. * config/mips/linux-unwind.h, config/mips/linux.h: Likewise. * config/pa/linux-unwind.h, config/pa/pa32-linux.h: Likewise. * config/rs6000/darwin-unwind.h, config/rs6000/darwin.h: Likewise. * config/s390/linux-unwind.h, config/s390/linux.h: Likewise. * config/sparc/linux-unwind.h, config/sparc/linux.h, config/sparc/linux64.h: Likewise. * config/sh/linux-unwind.h, config/sh/linux.h: Likewise, but merge SH_FALLBACK_FRAME_FLOAT_STATE into sh_fallback_frame_state. * config/rs6000/linux-unwind.h, config/rs6000/linux.h, config/rs6000/linux64.h: Likewise. Split out get_sigcontext function. Use ARG_POINTER_REGNUM for 32-bit temp reg too. From-SVN: r87167
2004-09-08 00:17:19 +00:00
#endif /* ifdef __x86_64__ */
#endif /* ifdef inhibit_libc */