2009-03-30 01:01:07 -06:00
|
|
|
// Copyright 2009 The Go Authors. All rights reserved.
|
|
|
|
// Use of this source code is governed by a BSD-style
|
|
|
|
// license that can be found in the LICENSE file.
|
|
|
|
|
2011-12-19 13:51:13 -07:00
|
|
|
#include "zasm_GOOS_GOARCH.h"
|
2013-07-16 14:24:09 -06:00
|
|
|
#include "funcdata.h"
|
2009-06-17 16:15:55 -06:00
|
|
|
|
2013-06-11 14:49:24 -06:00
|
|
|
TEXT _rt0_go(SB),7,$0
|
2009-03-30 01:01:07 -06:00
|
|
|
// copy arguments forward on an even stack
|
2013-03-07 20:57:10 -07:00
|
|
|
MOVL argc+0(FP), AX
|
|
|
|
MOVL argv+4(FP), BX
|
2009-03-30 01:01:07 -06:00
|
|
|
SUBL $128, SP // plenty of scratch
|
2009-10-03 11:37:12 -06:00
|
|
|
ANDL $~15, SP
|
2009-03-30 01:01:07 -06:00
|
|
|
MOVL AX, 120(SP) // save argc, argv away
|
|
|
|
MOVL BX, 124(SP)
|
|
|
|
|
2011-12-07 06:53:17 -07:00
|
|
|
// set default stack bounds.
|
2013-02-28 14:24:38 -07:00
|
|
|
// _cgo_init may update stackguard.
|
2011-12-07 06:53:17 -07:00
|
|
|
MOVL $runtime·g0(SB), BP
|
|
|
|
LEAL (-64*1024+104)(SP), BX
|
|
|
|
MOVL BX, g_stackguard(BP)
|
2013-06-03 02:28:24 -06:00
|
|
|
MOVL BX, g_stackguard0(BP)
|
2011-12-07 06:53:17 -07:00
|
|
|
MOVL SP, g_stackbase(BP)
|
|
|
|
|
2013-03-12 11:47:44 -06:00
|
|
|
// find out information about the processor we're on
|
|
|
|
MOVL $0, AX
|
|
|
|
CPUID
|
|
|
|
CMPL AX, $0
|
|
|
|
JE nocpuinfo
|
|
|
|
MOVL $1, AX
|
|
|
|
CPUID
|
|
|
|
MOVL CX, runtime·cpuid_ecx(SB)
|
|
|
|
MOVL DX, runtime·cpuid_edx(SB)
|
|
|
|
nocpuinfo:
|
|
|
|
|
2013-02-28 14:24:38 -07:00
|
|
|
// if there is an _cgo_init, call it to let it
|
2009-10-03 11:37:12 -06:00
|
|
|
// initialize and to set up GS. if not,
|
|
|
|
// we set up GS ourselves.
|
2013-02-28 14:24:38 -07:00
|
|
|
MOVL _cgo_init(SB), AX
|
2009-10-03 11:37:12 -06:00
|
|
|
TESTL AX, AX
|
2011-11-09 13:11:48 -07:00
|
|
|
JZ needtls
|
2013-03-25 16:14:02 -06:00
|
|
|
MOVL $setmg_gcc<>(SB), BX
|
|
|
|
MOVL BX, 4(SP)
|
2013-03-10 22:51:42 -06:00
|
|
|
MOVL BP, 0(SP)
|
2009-10-03 11:37:12 -06:00
|
|
|
CALL AX
|
2013-06-03 02:28:24 -06:00
|
|
|
// update stackguard after _cgo_init
|
|
|
|
MOVL $runtime·g0(SB), CX
|
|
|
|
MOVL g_stackguard0(CX), AX
|
|
|
|
MOVL AX, g_stackguard(CX)
|
2013-02-28 14:24:38 -07:00
|
|
|
// skip runtime·ldt0setup(SB) and tls test after _cgo_init for non-windows
|
2011-01-20 08:22:20 -07:00
|
|
|
CMPL runtime·iswindows(SB), $0
|
|
|
|
JEQ ok
|
2011-11-09 13:11:48 -07:00
|
|
|
needtls:
|
2011-07-25 10:25:41 -06:00
|
|
|
// skip runtime·ldt0setup(SB) and tls test on Plan 9 in all cases
|
|
|
|
CMPL runtime·isplan9(SB), $1
|
|
|
|
JEQ ok
|
|
|
|
|
2009-09-22 17:28:32 -06:00
|
|
|
// set up %gs
|
runtime: ,s/[a-zA-Z0-9_]+/runtime·&/g, almost
Prefix all external symbols in runtime by runtime·,
to avoid conflicts with possible symbols of the same
name in linked-in C libraries. The obvious conflicts
are printf, malloc, and free, but hide everything to
avoid future pain.
The symbols left alone are:
** known to cgo **
_cgo_free
_cgo_malloc
libcgo_thread_start
initcgo
ncgocall
** known to linker **
_rt0_$GOARCH
_rt0_$GOARCH_$GOOS
text
etext
data
end
pclntab
epclntab
symtab
esymtab
** known to C compiler **
_divv
_modv
_div64by32
etc (arch specific)
Tested on darwin/386, darwin/amd64, linux/386, linux/amd64.
Built (but not tested) for freebsd/386, freebsd/amd64, linux/arm, windows/386.
R=r, PeterGo
CC=golang-dev
https://golang.org/cl/2899041
2010-11-04 12:00:19 -06:00
|
|
|
CALL runtime·ldt0setup(SB)
|
2009-03-30 01:01:07 -06:00
|
|
|
|
|
|
|
// store through it, to make sure it works
|
2010-01-06 18:58:55 -07:00
|
|
|
get_tls(BX)
|
|
|
|
MOVL $0x123, g(BX)
|
runtime: ,s/[a-zA-Z0-9_]+/runtime·&/g, almost
Prefix all external symbols in runtime by runtime·,
to avoid conflicts with possible symbols of the same
name in linked-in C libraries. The obvious conflicts
are printf, malloc, and free, but hide everything to
avoid future pain.
The symbols left alone are:
** known to cgo **
_cgo_free
_cgo_malloc
libcgo_thread_start
initcgo
ncgocall
** known to linker **
_rt0_$GOARCH
_rt0_$GOARCH_$GOOS
text
etext
data
end
pclntab
epclntab
symtab
esymtab
** known to C compiler **
_divv
_modv
_div64by32
etc (arch specific)
Tested on darwin/386, darwin/amd64, linux/386, linux/amd64.
Built (but not tested) for freebsd/386, freebsd/amd64, linux/arm, windows/386.
R=r, PeterGo
CC=golang-dev
https://golang.org/cl/2899041
2010-11-04 12:00:19 -06:00
|
|
|
MOVL runtime·tls0(SB), AX
|
2009-03-30 01:01:07 -06:00
|
|
|
CMPL AX, $0x123
|
|
|
|
JEQ ok
|
2009-10-03 11:37:12 -06:00
|
|
|
MOVL AX, 0 // abort
|
2009-03-30 01:01:07 -06:00
|
|
|
ok:
|
|
|
|
// set up m and g "registers"
|
2010-01-06 18:58:55 -07:00
|
|
|
get_tls(BX)
|
runtime: ,s/[a-zA-Z0-9_]+/runtime·&/g, almost
Prefix all external symbols in runtime by runtime·,
to avoid conflicts with possible symbols of the same
name in linked-in C libraries. The obvious conflicts
are printf, malloc, and free, but hide everything to
avoid future pain.
The symbols left alone are:
** known to cgo **
_cgo_free
_cgo_malloc
libcgo_thread_start
initcgo
ncgocall
** known to linker **
_rt0_$GOARCH
_rt0_$GOARCH_$GOOS
text
etext
data
end
pclntab
epclntab
symtab
esymtab
** known to C compiler **
_divv
_modv
_div64by32
etc (arch specific)
Tested on darwin/386, darwin/amd64, linux/386, linux/amd64.
Built (but not tested) for freebsd/386, freebsd/amd64, linux/arm, windows/386.
R=r, PeterGo
CC=golang-dev
https://golang.org/cl/2899041
2010-11-04 12:00:19 -06:00
|
|
|
LEAL runtime·g0(SB), CX
|
2010-01-06 18:58:55 -07:00
|
|
|
MOVL CX, g(BX)
|
runtime: ,s/[a-zA-Z0-9_]+/runtime·&/g, almost
Prefix all external symbols in runtime by runtime·,
to avoid conflicts with possible symbols of the same
name in linked-in C libraries. The obvious conflicts
are printf, malloc, and free, but hide everything to
avoid future pain.
The symbols left alone are:
** known to cgo **
_cgo_free
_cgo_malloc
libcgo_thread_start
initcgo
ncgocall
** known to linker **
_rt0_$GOARCH
_rt0_$GOARCH_$GOOS
text
etext
data
end
pclntab
epclntab
symtab
esymtab
** known to C compiler **
_divv
_modv
_div64by32
etc (arch specific)
Tested on darwin/386, darwin/amd64, linux/386, linux/amd64.
Built (but not tested) for freebsd/386, freebsd/amd64, linux/arm, windows/386.
R=r, PeterGo
CC=golang-dev
https://golang.org/cl/2899041
2010-11-04 12:00:19 -06:00
|
|
|
LEAL runtime·m0(SB), AX
|
2010-01-06 18:58:55 -07:00
|
|
|
MOVL AX, m(BX)
|
2009-03-30 01:01:07 -06:00
|
|
|
|
|
|
|
// save m->g0 = g0
|
2009-06-17 16:15:55 -06:00
|
|
|
MOVL CX, m_g0(AX)
|
2009-03-30 01:01:07 -06:00
|
|
|
|
runtime: ,s/[a-zA-Z0-9_]+/runtime·&/g, almost
Prefix all external symbols in runtime by runtime·,
to avoid conflicts with possible symbols of the same
name in linked-in C libraries. The obvious conflicts
are printf, malloc, and free, but hide everything to
avoid future pain.
The symbols left alone are:
** known to cgo **
_cgo_free
_cgo_malloc
libcgo_thread_start
initcgo
ncgocall
** known to linker **
_rt0_$GOARCH
_rt0_$GOARCH_$GOOS
text
etext
data
end
pclntab
epclntab
symtab
esymtab
** known to C compiler **
_divv
_modv
_div64by32
etc (arch specific)
Tested on darwin/386, darwin/amd64, linux/386, linux/amd64.
Built (but not tested) for freebsd/386, freebsd/amd64, linux/arm, windows/386.
R=r, PeterGo
CC=golang-dev
https://golang.org/cl/2899041
2010-11-04 12:00:19 -06:00
|
|
|
CALL runtime·emptyfunc(SB) // fault if stack check is wrong
|
2009-03-30 01:01:07 -06:00
|
|
|
|
|
|
|
// convention is D is always cleared
|
|
|
|
CLD
|
|
|
|
|
runtime: ,s/[a-zA-Z0-9_]+/runtime·&/g, almost
Prefix all external symbols in runtime by runtime·,
to avoid conflicts with possible symbols of the same
name in linked-in C libraries. The obvious conflicts
are printf, malloc, and free, but hide everything to
avoid future pain.
The symbols left alone are:
** known to cgo **
_cgo_free
_cgo_malloc
libcgo_thread_start
initcgo
ncgocall
** known to linker **
_rt0_$GOARCH
_rt0_$GOARCH_$GOOS
text
etext
data
end
pclntab
epclntab
symtab
esymtab
** known to C compiler **
_divv
_modv
_div64by32
etc (arch specific)
Tested on darwin/386, darwin/amd64, linux/386, linux/amd64.
Built (but not tested) for freebsd/386, freebsd/amd64, linux/arm, windows/386.
R=r, PeterGo
CC=golang-dev
https://golang.org/cl/2899041
2010-11-04 12:00:19 -06:00
|
|
|
CALL runtime·check(SB)
|
2009-03-30 01:01:07 -06:00
|
|
|
|
|
|
|
// saved argc, argv
|
|
|
|
MOVL 120(SP), AX
|
|
|
|
MOVL AX, 0(SP)
|
|
|
|
MOVL 124(SP), AX
|
|
|
|
MOVL AX, 4(SP)
|
runtime: ,s/[a-zA-Z0-9_]+/runtime·&/g, almost
Prefix all external symbols in runtime by runtime·,
to avoid conflicts with possible symbols of the same
name in linked-in C libraries. The obvious conflicts
are printf, malloc, and free, but hide everything to
avoid future pain.
The symbols left alone are:
** known to cgo **
_cgo_free
_cgo_malloc
libcgo_thread_start
initcgo
ncgocall
** known to linker **
_rt0_$GOARCH
_rt0_$GOARCH_$GOOS
text
etext
data
end
pclntab
epclntab
symtab
esymtab
** known to C compiler **
_divv
_modv
_div64by32
etc (arch specific)
Tested on darwin/386, darwin/amd64, linux/386, linux/amd64.
Built (but not tested) for freebsd/386, freebsd/amd64, linux/arm, windows/386.
R=r, PeterGo
CC=golang-dev
https://golang.org/cl/2899041
2010-11-04 12:00:19 -06:00
|
|
|
CALL runtime·args(SB)
|
|
|
|
CALL runtime·osinit(SB)
|
2013-03-12 11:47:44 -06:00
|
|
|
CALL runtime·hashinit(SB)
|
runtime: ,s/[a-zA-Z0-9_]+/runtime·&/g, almost
Prefix all external symbols in runtime by runtime·,
to avoid conflicts with possible symbols of the same
name in linked-in C libraries. The obvious conflicts
are printf, malloc, and free, but hide everything to
avoid future pain.
The symbols left alone are:
** known to cgo **
_cgo_free
_cgo_malloc
libcgo_thread_start
initcgo
ncgocall
** known to linker **
_rt0_$GOARCH
_rt0_$GOARCH_$GOOS
text
etext
data
end
pclntab
epclntab
symtab
esymtab
** known to C compiler **
_divv
_modv
_div64by32
etc (arch specific)
Tested on darwin/386, darwin/amd64, linux/386, linux/amd64.
Built (but not tested) for freebsd/386, freebsd/amd64, linux/arm, windows/386.
R=r, PeterGo
CC=golang-dev
https://golang.org/cl/2899041
2010-11-04 12:00:19 -06:00
|
|
|
CALL runtime·schedinit(SB)
|
2009-03-30 01:01:07 -06:00
|
|
|
|
|
|
|
// create a new goroutine to start program
|
2013-02-21 15:01:13 -07:00
|
|
|
PUSHL $runtime·main·f(SB) // entry
|
2009-06-17 16:15:55 -06:00
|
|
|
PUSHL $0 // arg size
|
2013-07-16 14:24:09 -06:00
|
|
|
ARGSIZE(8)
|
runtime: ,s/[a-zA-Z0-9_]+/runtime·&/g, almost
Prefix all external symbols in runtime by runtime·,
to avoid conflicts with possible symbols of the same
name in linked-in C libraries. The obvious conflicts
are printf, malloc, and free, but hide everything to
avoid future pain.
The symbols left alone are:
** known to cgo **
_cgo_free
_cgo_malloc
libcgo_thread_start
initcgo
ncgocall
** known to linker **
_rt0_$GOARCH
_rt0_$GOARCH_$GOOS
text
etext
data
end
pclntab
epclntab
symtab
esymtab
** known to C compiler **
_divv
_modv
_div64by32
etc (arch specific)
Tested on darwin/386, darwin/amd64, linux/386, linux/amd64.
Built (but not tested) for freebsd/386, freebsd/amd64, linux/arm, windows/386.
R=r, PeterGo
CC=golang-dev
https://golang.org/cl/2899041
2010-11-04 12:00:19 -06:00
|
|
|
CALL runtime·newproc(SB)
|
2013-07-16 14:24:09 -06:00
|
|
|
ARGSIZE(-1)
|
2009-03-30 01:01:07 -06:00
|
|
|
POPL AX
|
|
|
|
POPL AX
|
|
|
|
|
|
|
|
// start this M
|
runtime: ,s/[a-zA-Z0-9_]+/runtime·&/g, almost
Prefix all external symbols in runtime by runtime·,
to avoid conflicts with possible symbols of the same
name in linked-in C libraries. The obvious conflicts
are printf, malloc, and free, but hide everything to
avoid future pain.
The symbols left alone are:
** known to cgo **
_cgo_free
_cgo_malloc
libcgo_thread_start
initcgo
ncgocall
** known to linker **
_rt0_$GOARCH
_rt0_$GOARCH_$GOOS
text
etext
data
end
pclntab
epclntab
symtab
esymtab
** known to C compiler **
_divv
_modv
_div64by32
etc (arch specific)
Tested on darwin/386, darwin/amd64, linux/386, linux/amd64.
Built (but not tested) for freebsd/386, freebsd/amd64, linux/arm, windows/386.
R=r, PeterGo
CC=golang-dev
https://golang.org/cl/2899041
2010-11-04 12:00:19 -06:00
|
|
|
CALL runtime·mstart(SB)
|
2009-03-30 01:01:07 -06:00
|
|
|
|
|
|
|
INT $3
|
|
|
|
RET
|
|
|
|
|
2013-02-21 15:01:13 -07:00
|
|
|
DATA runtime·main·f+0(SB)/4,$runtime·main(SB)
|
|
|
|
GLOBL runtime·main·f(SB),8,$4
|
|
|
|
|
2013-07-16 14:24:09 -06:00
|
|
|
TEXT runtime·breakpoint(SB),7,$0-0
|
2009-09-22 17:28:32 -06:00
|
|
|
INT $3
|
2009-03-30 01:01:07 -06:00
|
|
|
RET
|
|
|
|
|
2013-07-16 14:24:09 -06:00
|
|
|
TEXT runtime·asminit(SB),7,$0-0
|
2013-04-02 14:45:56 -06:00
|
|
|
// Linux and MinGW start the FPU in extended double precision.
|
2012-02-13 23:23:15 -07:00
|
|
|
// Other operating systems use double precision.
|
|
|
|
// Change to double precision to match them,
|
|
|
|
// and to match other hardware that only has double.
|
|
|
|
PUSHL $0x27F
|
|
|
|
FLDCW 0(SP)
|
|
|
|
POPL AX
|
|
|
|
RET
|
|
|
|
|
2009-06-17 16:15:55 -06:00
|
|
|
/*
|
|
|
|
* go-routine
|
|
|
|
*/
|
2009-03-30 01:01:07 -06:00
|
|
|
|
runtime: scheduler, cgo reorganization
* Change use of m->g0 stack (aka scheduler stack).
* Provide runtime.mcall(f) to invoke f() on m->g0 stack.
* Replace scheduler loop entry with runtime.mcall(schedule).
Runtime.mcall eliminates the need for fake scheduler states that
exist just to run a bit of code on the m->g0 stack
(Grecovery, Gstackalloc).
The elimination of the scheduler as a loop that stops and
starts using gosave and gogo fixes a bad interaction with the
way cgo uses the m->g0 stack. Cgo runs external (gcc-compiled)
C functions on that stack, and then when calling back into Go,
it sets m->g0->sched.sp below the added call frames, so that
other uses of m->g0's stack will not interfere with those frames.
Unfortunately, gogo (longjmp) back to the scheduler loop at
this point would end up running scheduler with the lower
sp, which no longer points at a valid stack frame for
a call to scheduler. If scheduler then wrote any function call
arguments or local variables to where it expected the stack
frame to be, it would overwrite other data on the stack.
I realized this possibility while debugging a problem with
calling complex Go code in a Go -> C -> Go cgo callback.
This wasn't the bug I was looking for, it turns out, but I believe
it is a real bug nonetheless. Switching to runtime.mcall, which
only adds new frames to the stack and never jumps into
functions running in existing ones, fixes this bug.
* Move cgo-related code out of proc.c into cgocall.c.
* Add very large comment describing cgo call sequences.
* Simpilify, regularize cgo function implementations and names.
* Add test suite as misc/cgo/test.
Now the Go -> C path calls cgocall, which calls asmcgocall,
and the C -> Go path calls cgocallback, which calls cgocallbackg.
The shuffling, which affects mainly the callback case, moves
most of the callback implementation to cgocallback running
on the m->curg stack (not the m->g0 scheduler stack) and
only while accounted for with $GOMAXPROCS (between calls
to exitsyscall and entersyscall).
The previous callback code did not block in startcgocallback's
approximation to exitsyscall, so if, say, the garbage collector
were running, it would still barge in and start doing things
like call malloc. Similarly endcgocallback's approximation of
entersyscall did not call matchmg to kick off new OS threads
when necessary, which caused the bug in issue 1560.
Fixes #1560.
R=iant
CC=golang-dev
https://golang.org/cl/4253054
2011-03-07 08:37:42 -07:00
|
|
|
// void gosave(Gobuf*)
|
2009-06-17 16:15:55 -06:00
|
|
|
// save state in Gobuf; setjmp
|
2013-07-16 14:24:09 -06:00
|
|
|
TEXT runtime·gosave(SB), 7, $0-4
|
2009-06-17 16:15:55 -06:00
|
|
|
MOVL 4(SP), AX // gobuf
|
|
|
|
LEAL 4(SP), BX // caller's SP
|
|
|
|
MOVL BX, gobuf_sp(AX)
|
|
|
|
MOVL 0(SP), BX // caller's PC
|
|
|
|
MOVL BX, gobuf_pc(AX)
|
2013-06-12 13:22:26 -06:00
|
|
|
MOVL $0, gobuf_ret(AX)
|
|
|
|
MOVL $0, gobuf_ctxt(AX)
|
2010-01-06 18:58:55 -07:00
|
|
|
get_tls(CX)
|
|
|
|
MOVL g(CX), BX
|
2009-06-17 16:15:55 -06:00
|
|
|
MOVL BX, gobuf_g(AX)
|
2009-03-30 01:01:07 -06:00
|
|
|
RET
|
|
|
|
|
2013-06-12 16:05:10 -06:00
|
|
|
// void gogo(Gobuf*)
|
2009-06-17 16:15:55 -06:00
|
|
|
// restore state from Gobuf; longjmp
|
2013-07-16 14:24:09 -06:00
|
|
|
TEXT runtime·gogo(SB), 7, $0-4
|
2009-06-17 16:15:55 -06:00
|
|
|
MOVL 4(SP), BX // gobuf
|
|
|
|
MOVL gobuf_g(BX), DX
|
|
|
|
MOVL 0(DX), CX // make sure g != nil
|
2010-01-06 18:58:55 -07:00
|
|
|
get_tls(CX)
|
|
|
|
MOVL DX, g(CX)
|
2009-06-17 16:15:55 -06:00
|
|
|
MOVL gobuf_sp(BX), SP // restore SP
|
2013-06-12 13:22:26 -06:00
|
|
|
MOVL gobuf_ret(BX), AX
|
|
|
|
MOVL gobuf_ctxt(BX), DX
|
|
|
|
MOVL $0, gobuf_sp(BX) // clear to help garbage collector
|
|
|
|
MOVL $0, gobuf_ret(BX)
|
|
|
|
MOVL $0, gobuf_ctxt(BX)
|
2009-06-17 16:15:55 -06:00
|
|
|
MOVL gobuf_pc(BX), BX
|
|
|
|
JMP BX
|
|
|
|
|
runtime: scheduler, cgo reorganization
* Change use of m->g0 stack (aka scheduler stack).
* Provide runtime.mcall(f) to invoke f() on m->g0 stack.
* Replace scheduler loop entry with runtime.mcall(schedule).
Runtime.mcall eliminates the need for fake scheduler states that
exist just to run a bit of code on the m->g0 stack
(Grecovery, Gstackalloc).
The elimination of the scheduler as a loop that stops and
starts using gosave and gogo fixes a bad interaction with the
way cgo uses the m->g0 stack. Cgo runs external (gcc-compiled)
C functions on that stack, and then when calling back into Go,
it sets m->g0->sched.sp below the added call frames, so that
other uses of m->g0's stack will not interfere with those frames.
Unfortunately, gogo (longjmp) back to the scheduler loop at
this point would end up running scheduler with the lower
sp, which no longer points at a valid stack frame for
a call to scheduler. If scheduler then wrote any function call
arguments or local variables to where it expected the stack
frame to be, it would overwrite other data on the stack.
I realized this possibility while debugging a problem with
calling complex Go code in a Go -> C -> Go cgo callback.
This wasn't the bug I was looking for, it turns out, but I believe
it is a real bug nonetheless. Switching to runtime.mcall, which
only adds new frames to the stack and never jumps into
functions running in existing ones, fixes this bug.
* Move cgo-related code out of proc.c into cgocall.c.
* Add very large comment describing cgo call sequences.
* Simpilify, regularize cgo function implementations and names.
* Add test suite as misc/cgo/test.
Now the Go -> C path calls cgocall, which calls asmcgocall,
and the C -> Go path calls cgocallback, which calls cgocallbackg.
The shuffling, which affects mainly the callback case, moves
most of the callback implementation to cgocallback running
on the m->curg stack (not the m->g0 scheduler stack) and
only while accounted for with $GOMAXPROCS (between calls
to exitsyscall and entersyscall).
The previous callback code did not block in startcgocallback's
approximation to exitsyscall, so if, say, the garbage collector
were running, it would still barge in and start doing things
like call malloc. Similarly endcgocallback's approximation of
entersyscall did not call matchmg to kick off new OS threads
when necessary, which caused the bug in issue 1560.
Fixes #1560.
R=iant
CC=golang-dev
https://golang.org/cl/4253054
2011-03-07 08:37:42 -07:00
|
|
|
// void mcall(void (*fn)(G*))
|
|
|
|
// Switch to m->g0's stack, call fn(g).
|
runtime: stack split + garbage collection bug
The g->sched.sp saved stack pointer and the
g->stackbase and g->stackguard stack bounds
can change even while "the world is stopped",
because a goroutine has to call functions (and
therefore might split its stack) when exiting a
system call to check whether the world is stopped
(and if so, wait until the world continues).
That means the garbage collector cannot access
those values safely (without a race) for goroutines
executing system calls. Instead, save a consistent
triple in g->gcsp, g->gcstack, g->gcguard during
entersyscall and have the garbage collector refer
to those.
The old code was occasionally seeing (because of
the race) an sp and stk that did not correspond to
each other, so that stk - sp was not the number of
stack bytes following sp. In that case, if sp < stk
then the call scanblock(sp, stk - sp) scanned too
many bytes (anything between the two pointers,
which pointed into different allocation blocks).
If sp > stk then stk - sp wrapped around.
On 32-bit, stk - sp is a uintptr (uint32) converted
to int64 in the call to scanblock, so a large (~4G)
but positive number. Scanblock would try to scan
that many bytes and eventually fault accessing
unmapped memory. On 64-bit, stk - sp is a uintptr (uint64)
promoted to int64 in the call to scanblock, so a negative
number. Scanblock would not scan anything, possibly
causing in-use blocks to be freed.
In short, 32-bit platforms would have seen either
ineffective garbage collection or crashes during garbage
collection, while 64-bit platforms would have seen
either ineffective or incorrect garbage collection.
You can see the invalid arguments to scanblock in the
stack traces in issue 1620.
Fixes #1620.
Fixes #1746.
R=iant, r
CC=golang-dev
https://golang.org/cl/4437075
2011-04-27 21:21:12 -06:00
|
|
|
// Fn must never return. It should gogo(&g->sched)
|
runtime: scheduler, cgo reorganization
* Change use of m->g0 stack (aka scheduler stack).
* Provide runtime.mcall(f) to invoke f() on m->g0 stack.
* Replace scheduler loop entry with runtime.mcall(schedule).
Runtime.mcall eliminates the need for fake scheduler states that
exist just to run a bit of code on the m->g0 stack
(Grecovery, Gstackalloc).
The elimination of the scheduler as a loop that stops and
starts using gosave and gogo fixes a bad interaction with the
way cgo uses the m->g0 stack. Cgo runs external (gcc-compiled)
C functions on that stack, and then when calling back into Go,
it sets m->g0->sched.sp below the added call frames, so that
other uses of m->g0's stack will not interfere with those frames.
Unfortunately, gogo (longjmp) back to the scheduler loop at
this point would end up running scheduler with the lower
sp, which no longer points at a valid stack frame for
a call to scheduler. If scheduler then wrote any function call
arguments or local variables to where it expected the stack
frame to be, it would overwrite other data on the stack.
I realized this possibility while debugging a problem with
calling complex Go code in a Go -> C -> Go cgo callback.
This wasn't the bug I was looking for, it turns out, but I believe
it is a real bug nonetheless. Switching to runtime.mcall, which
only adds new frames to the stack and never jumps into
functions running in existing ones, fixes this bug.
* Move cgo-related code out of proc.c into cgocall.c.
* Add very large comment describing cgo call sequences.
* Simpilify, regularize cgo function implementations and names.
* Add test suite as misc/cgo/test.
Now the Go -> C path calls cgocall, which calls asmcgocall,
and the C -> Go path calls cgocallback, which calls cgocallbackg.
The shuffling, which affects mainly the callback case, moves
most of the callback implementation to cgocallback running
on the m->curg stack (not the m->g0 scheduler stack) and
only while accounted for with $GOMAXPROCS (between calls
to exitsyscall and entersyscall).
The previous callback code did not block in startcgocallback's
approximation to exitsyscall, so if, say, the garbage collector
were running, it would still barge in and start doing things
like call malloc. Similarly endcgocallback's approximation of
entersyscall did not call matchmg to kick off new OS threads
when necessary, which caused the bug in issue 1560.
Fixes #1560.
R=iant
CC=golang-dev
https://golang.org/cl/4253054
2011-03-07 08:37:42 -07:00
|
|
|
// to keep running g.
|
2013-07-16 14:24:09 -06:00
|
|
|
TEXT runtime·mcall(SB), 7, $0-4
|
runtime: scheduler, cgo reorganization
* Change use of m->g0 stack (aka scheduler stack).
* Provide runtime.mcall(f) to invoke f() on m->g0 stack.
* Replace scheduler loop entry with runtime.mcall(schedule).
Runtime.mcall eliminates the need for fake scheduler states that
exist just to run a bit of code on the m->g0 stack
(Grecovery, Gstackalloc).
The elimination of the scheduler as a loop that stops and
starts using gosave and gogo fixes a bad interaction with the
way cgo uses the m->g0 stack. Cgo runs external (gcc-compiled)
C functions on that stack, and then when calling back into Go,
it sets m->g0->sched.sp below the added call frames, so that
other uses of m->g0's stack will not interfere with those frames.
Unfortunately, gogo (longjmp) back to the scheduler loop at
this point would end up running scheduler with the lower
sp, which no longer points at a valid stack frame for
a call to scheduler. If scheduler then wrote any function call
arguments or local variables to where it expected the stack
frame to be, it would overwrite other data on the stack.
I realized this possibility while debugging a problem with
calling complex Go code in a Go -> C -> Go cgo callback.
This wasn't the bug I was looking for, it turns out, but I believe
it is a real bug nonetheless. Switching to runtime.mcall, which
only adds new frames to the stack and never jumps into
functions running in existing ones, fixes this bug.
* Move cgo-related code out of proc.c into cgocall.c.
* Add very large comment describing cgo call sequences.
* Simpilify, regularize cgo function implementations and names.
* Add test suite as misc/cgo/test.
Now the Go -> C path calls cgocall, which calls asmcgocall,
and the C -> Go path calls cgocallback, which calls cgocallbackg.
The shuffling, which affects mainly the callback case, moves
most of the callback implementation to cgocallback running
on the m->curg stack (not the m->g0 scheduler stack) and
only while accounted for with $GOMAXPROCS (between calls
to exitsyscall and entersyscall).
The previous callback code did not block in startcgocallback's
approximation to exitsyscall, so if, say, the garbage collector
were running, it would still barge in and start doing things
like call malloc. Similarly endcgocallback's approximation of
entersyscall did not call matchmg to kick off new OS threads
when necessary, which caused the bug in issue 1560.
Fixes #1560.
R=iant
CC=golang-dev
https://golang.org/cl/4253054
2011-03-07 08:37:42 -07:00
|
|
|
MOVL fn+0(FP), DI
|
|
|
|
|
|
|
|
get_tls(CX)
|
2013-06-05 05:16:53 -06:00
|
|
|
MOVL g(CX), AX // save state in g->sched
|
runtime: scheduler, cgo reorganization
* Change use of m->g0 stack (aka scheduler stack).
* Provide runtime.mcall(f) to invoke f() on m->g0 stack.
* Replace scheduler loop entry with runtime.mcall(schedule).
Runtime.mcall eliminates the need for fake scheduler states that
exist just to run a bit of code on the m->g0 stack
(Grecovery, Gstackalloc).
The elimination of the scheduler as a loop that stops and
starts using gosave and gogo fixes a bad interaction with the
way cgo uses the m->g0 stack. Cgo runs external (gcc-compiled)
C functions on that stack, and then when calling back into Go,
it sets m->g0->sched.sp below the added call frames, so that
other uses of m->g0's stack will not interfere with those frames.
Unfortunately, gogo (longjmp) back to the scheduler loop at
this point would end up running scheduler with the lower
sp, which no longer points at a valid stack frame for
a call to scheduler. If scheduler then wrote any function call
arguments or local variables to where it expected the stack
frame to be, it would overwrite other data on the stack.
I realized this possibility while debugging a problem with
calling complex Go code in a Go -> C -> Go cgo callback.
This wasn't the bug I was looking for, it turns out, but I believe
it is a real bug nonetheless. Switching to runtime.mcall, which
only adds new frames to the stack and never jumps into
functions running in existing ones, fixes this bug.
* Move cgo-related code out of proc.c into cgocall.c.
* Add very large comment describing cgo call sequences.
* Simpilify, regularize cgo function implementations and names.
* Add test suite as misc/cgo/test.
Now the Go -> C path calls cgocall, which calls asmcgocall,
and the C -> Go path calls cgocallback, which calls cgocallbackg.
The shuffling, which affects mainly the callback case, moves
most of the callback implementation to cgocallback running
on the m->curg stack (not the m->g0 scheduler stack) and
only while accounted for with $GOMAXPROCS (between calls
to exitsyscall and entersyscall).
The previous callback code did not block in startcgocallback's
approximation to exitsyscall, so if, say, the garbage collector
were running, it would still barge in and start doing things
like call malloc. Similarly endcgocallback's approximation of
entersyscall did not call matchmg to kick off new OS threads
when necessary, which caused the bug in issue 1560.
Fixes #1560.
R=iant
CC=golang-dev
https://golang.org/cl/4253054
2011-03-07 08:37:42 -07:00
|
|
|
MOVL 0(SP), BX // caller's PC
|
|
|
|
MOVL BX, (g_sched+gobuf_pc)(AX)
|
|
|
|
LEAL 4(SP), BX // caller's SP
|
|
|
|
MOVL BX, (g_sched+gobuf_sp)(AX)
|
|
|
|
MOVL AX, (g_sched+gobuf_g)(AX)
|
|
|
|
|
|
|
|
// switch to m->g0 & its stack, call fn
|
|
|
|
MOVL m(CX), BX
|
|
|
|
MOVL m_g0(BX), SI
|
|
|
|
CMPL SI, AX // if g == m->g0 call badmcall
|
|
|
|
JNE 2(PC)
|
|
|
|
CALL runtime·badmcall(SB)
|
|
|
|
MOVL SI, g(CX) // g = m->g0
|
2013-06-05 05:16:53 -06:00
|
|
|
MOVL (g_sched+gobuf_sp)(SI), SP // sp = m->g0->sched.sp
|
runtime: scheduler, cgo reorganization
* Change use of m->g0 stack (aka scheduler stack).
* Provide runtime.mcall(f) to invoke f() on m->g0 stack.
* Replace scheduler loop entry with runtime.mcall(schedule).
Runtime.mcall eliminates the need for fake scheduler states that
exist just to run a bit of code on the m->g0 stack
(Grecovery, Gstackalloc).
The elimination of the scheduler as a loop that stops and
starts using gosave and gogo fixes a bad interaction with the
way cgo uses the m->g0 stack. Cgo runs external (gcc-compiled)
C functions on that stack, and then when calling back into Go,
it sets m->g0->sched.sp below the added call frames, so that
other uses of m->g0's stack will not interfere with those frames.
Unfortunately, gogo (longjmp) back to the scheduler loop at
this point would end up running scheduler with the lower
sp, which no longer points at a valid stack frame for
a call to scheduler. If scheduler then wrote any function call
arguments or local variables to where it expected the stack
frame to be, it would overwrite other data on the stack.
I realized this possibility while debugging a problem with
calling complex Go code in a Go -> C -> Go cgo callback.
This wasn't the bug I was looking for, it turns out, but I believe
it is a real bug nonetheless. Switching to runtime.mcall, which
only adds new frames to the stack and never jumps into
functions running in existing ones, fixes this bug.
* Move cgo-related code out of proc.c into cgocall.c.
* Add very large comment describing cgo call sequences.
* Simpilify, regularize cgo function implementations and names.
* Add test suite as misc/cgo/test.
Now the Go -> C path calls cgocall, which calls asmcgocall,
and the C -> Go path calls cgocallback, which calls cgocallbackg.
The shuffling, which affects mainly the callback case, moves
most of the callback implementation to cgocallback running
on the m->curg stack (not the m->g0 scheduler stack) and
only while accounted for with $GOMAXPROCS (between calls
to exitsyscall and entersyscall).
The previous callback code did not block in startcgocallback's
approximation to exitsyscall, so if, say, the garbage collector
were running, it would still barge in and start doing things
like call malloc. Similarly endcgocallback's approximation of
entersyscall did not call matchmg to kick off new OS threads
when necessary, which caused the bug in issue 1560.
Fixes #1560.
R=iant
CC=golang-dev
https://golang.org/cl/4253054
2011-03-07 08:37:42 -07:00
|
|
|
PUSHL AX
|
|
|
|
CALL DI
|
|
|
|
POPL AX
|
|
|
|
CALL runtime·badmcall2(SB)
|
|
|
|
RET
|
|
|
|
|
2009-06-17 16:15:55 -06:00
|
|
|
/*
|
|
|
|
* support for morestack
|
|
|
|
*/
|
|
|
|
|
|
|
|
// Called during function prolog when more stack is needed.
|
2013-07-18 14:53:45 -06:00
|
|
|
//
|
|
|
|
// The traceback routines see morestack on a g0 as being
|
|
|
|
// the top of a stack (for example, morestack calling newstack
|
|
|
|
// calling the scheduler calling newm calling gc), so we must
|
|
|
|
// record an argument size. For that purpose, it has no arguments.
|
|
|
|
TEXT runtime·morestack(SB),7,$0-0
|
2009-06-17 16:15:55 -06:00
|
|
|
// Cannot grow scheduler stack (m->g0).
|
2010-01-06 18:58:55 -07:00
|
|
|
get_tls(CX)
|
|
|
|
MOVL m(CX), BX
|
2009-06-17 16:15:55 -06:00
|
|
|
MOVL m_g0(BX), SI
|
2010-01-06 18:58:55 -07:00
|
|
|
CMPL g(CX), SI
|
2009-06-17 16:15:55 -06:00
|
|
|
JNE 2(PC)
|
|
|
|
INT $3
|
|
|
|
|
2013-02-22 08:47:54 -07:00
|
|
|
// frame size in DI
|
2009-06-17 16:15:55 -06:00
|
|
|
// arg size in AX
|
|
|
|
// Save in m.
|
2013-02-22 08:47:54 -07:00
|
|
|
MOVL DI, m_moreframesize(BX)
|
2011-01-14 12:05:20 -07:00
|
|
|
MOVL AX, m_moreargsize(BX)
|
2009-06-17 16:15:55 -06:00
|
|
|
|
|
|
|
// Called from f.
|
|
|
|
// Set m->morebuf to f's caller.
|
|
|
|
MOVL 4(SP), DI // f's caller's PC
|
|
|
|
MOVL DI, (m_morebuf+gobuf_pc)(BX)
|
|
|
|
LEAL 8(SP), CX // f's caller's SP
|
|
|
|
MOVL CX, (m_morebuf+gobuf_sp)(BX)
|
2011-01-14 12:05:20 -07:00
|
|
|
MOVL CX, m_moreargp(BX)
|
2010-01-06 18:58:55 -07:00
|
|
|
get_tls(CX)
|
|
|
|
MOVL g(CX), SI
|
2009-06-17 16:15:55 -06:00
|
|
|
MOVL SI, (m_morebuf+gobuf_g)(BX)
|
|
|
|
|
runtime: record proper goroutine state during stack split
Until now, the goroutine state has been scattered during the
execution of newstack and oldstack. It's all there, and those routines
know how to get back to a working goroutine, but other pieces of
the system, like stack traces, do not. If something does interrupt
the newstack or oldstack execution, the rest of the system can't
understand the goroutine. For example, if newstack decides there
is an overflow and calls throw, the stack tracer wouldn't dump the
goroutine correctly.
For newstack to save a useful state snapshot, it needs to be able
to rewind the PC in the function that triggered the split back to
the beginning of the function. (The PC is a few instructions in, just
after the call to morestack.) To make that possible, we change the
prologues to insert a jmp back to the beginning of the function
after the call to morestack. That is, the prologue used to be roughly:
TEXT myfunc
check for split
jmpcond nosplit
call morestack
nosplit:
sub $xxx, sp
Now an extra instruction is inserted after the call:
TEXT myfunc
start:
check for split
jmpcond nosplit
call morestack
jmp start
nosplit:
sub $xxx, sp
The jmp is not executed directly. It is decoded and simulated by
runtime.rewindmorestack to discover the beginning of the function,
and then the call to morestack returns directly to the start label
instead of to the jump instruction. So logically the jmp is still
executed, just not by the cpu.
The prologue thus repeats in the case of a function that needs a
stack split, but against the cost of the split itself, the extra few
instructions are noise. The repeated prologue has the nice effect of
making a stack split double-check that the new stack is big enough:
if morestack happens to return on a too-small stack, we'll now notice
before corruption happens.
The ability for newstack to rewind to the beginning of the function
should help preemption too. If newstack decides that it was called
for preemption instead of a stack split, it now has the goroutine state
correctly paused if rescheduling is needed, and when the goroutine
can run again, it can return to the start label on its original stack
and re-execute the split check.
Here is an example of a split stack overflow showing the full
trace, without any special cases in the stack printer.
(This one was triggered by making the split check incorrect.)
runtime: newstack framesize=0x0 argsize=0x18 sp=0x6aebd0 stack=[0x6b0000, 0x6b0fa0]
morebuf={pc:0x69f5b sp:0x6aebd8 lr:0x0}
sched={pc:0x68880 sp:0x6aebd0 lr:0x0 ctxt:0x34e700}
runtime: split stack overflow: 0x6aebd0 < 0x6b0000
fatal error: runtime: split stack overflow
goroutine 1 [stack split]:
runtime.mallocgc(0x290, 0x100000000, 0x1)
/Users/rsc/g/go/src/pkg/runtime/zmalloc_darwin_amd64.c:21 fp=0x6aebd8
runtime.new()
/Users/rsc/g/go/src/pkg/runtime/zmalloc_darwin_amd64.c:682 +0x5b fp=0x6aec08
go/build.(*Context).Import(0x5ae340, 0xc210030c71, 0xa, 0xc2100b4380, 0x1b, ...)
/Users/rsc/g/go/src/pkg/go/build/build.go:424 +0x3a fp=0x6b00a0
main.loadImport(0xc210030c71, 0xa, 0xc2100b4380, 0x1b, 0xc2100b42c0, ...)
/Users/rsc/g/go/src/cmd/go/pkg.go:249 +0x371 fp=0x6b01a8
main.(*Package).load(0xc21017c800, 0xc2100b42c0, 0xc2101828c0, 0x0, 0x0, ...)
/Users/rsc/g/go/src/cmd/go/pkg.go:431 +0x2801 fp=0x6b0c98
main.loadPackage(0x369040, 0x7, 0xc2100b42c0, 0x0)
/Users/rsc/g/go/src/cmd/go/pkg.go:709 +0x857 fp=0x6b0f80
----- stack segment boundary -----
main.(*builder).action(0xc2100902a0, 0x0, 0x0, 0xc2100e6c00, 0xc2100e5750, ...)
/Users/rsc/g/go/src/cmd/go/build.go:539 +0x437 fp=0x6b14a0
main.(*builder).action(0xc2100902a0, 0x0, 0x0, 0xc21015b400, 0x2, ...)
/Users/rsc/g/go/src/cmd/go/build.go:528 +0x1d2 fp=0x6b1658
main.(*builder).test(0xc2100902a0, 0xc210092000, 0x0, 0x0, 0xc21008ff60, ...)
/Users/rsc/g/go/src/cmd/go/test.go:622 +0x1b53 fp=0x6b1f68
----- stack segment boundary -----
main.runTest(0x5a6b20, 0xc21000a020, 0x2, 0x2)
/Users/rsc/g/go/src/cmd/go/test.go:366 +0xd09 fp=0x6a5cf0
main.main()
/Users/rsc/g/go/src/cmd/go/main.go:161 +0x4f9 fp=0x6a5f78
runtime.main()
/Users/rsc/g/go/src/pkg/runtime/proc.c:183 +0x92 fp=0x6a5fa0
runtime.goexit()
/Users/rsc/g/go/src/pkg/runtime/proc.c:1266 fp=0x6a5fa8
And here is a seg fault during oldstack:
SIGSEGV: segmentation violation
PC=0x1b2a6
runtime.oldstack()
/Users/rsc/g/go/src/pkg/runtime/stack.c:159 +0x76
runtime.lessstack()
/Users/rsc/g/go/src/pkg/runtime/asm_amd64.s:270 +0x22
goroutine 1 [stack unsplit]:
fmt.(*pp).printArg(0x2102e64e0, 0xe5c80, 0x2102c9220, 0x73, 0x0, ...)
/Users/rsc/g/go/src/pkg/fmt/print.go:818 +0x3d3 fp=0x221031e6f8
fmt.(*pp).doPrintf(0x2102e64e0, 0x12fb20, 0x2, 0x221031eb98, 0x1, ...)
/Users/rsc/g/go/src/pkg/fmt/print.go:1183 +0x15cb fp=0x221031eaf0
fmt.Sprintf(0x12fb20, 0x2, 0x221031eb98, 0x1, 0x1, ...)
/Users/rsc/g/go/src/pkg/fmt/print.go:234 +0x67 fp=0x221031eb40
flag.(*stringValue).String(0x2102c9210, 0x1, 0x0)
/Users/rsc/g/go/src/pkg/flag/flag.go:180 +0xb3 fp=0x221031ebb0
flag.(*FlagSet).Var(0x2102f6000, 0x293d38, 0x2102c9210, 0x143490, 0xa, ...)
/Users/rsc/g/go/src/pkg/flag/flag.go:633 +0x40 fp=0x221031eca0
flag.(*FlagSet).StringVar(0x2102f6000, 0x2102c9210, 0x143490, 0xa, 0x12fa60, ...)
/Users/rsc/g/go/src/pkg/flag/flag.go:550 +0x91 fp=0x221031ece8
flag.(*FlagSet).String(0x2102f6000, 0x143490, 0xa, 0x12fa60, 0x0, ...)
/Users/rsc/g/go/src/pkg/flag/flag.go:563 +0x87 fp=0x221031ed38
flag.String(0x143490, 0xa, 0x12fa60, 0x0, 0x161950, ...)
/Users/rsc/g/go/src/pkg/flag/flag.go:570 +0x6b fp=0x221031ed80
testing.init()
/Users/rsc/g/go/src/pkg/testing/testing.go:-531 +0xbb fp=0x221031edc0
strings_test.init()
/Users/rsc/g/go/src/pkg/strings/strings_test.go:1115 +0x62 fp=0x221031ef70
main.init()
strings/_test/_testmain.go:90 +0x3d fp=0x221031ef78
runtime.main()
/Users/rsc/g/go/src/pkg/runtime/proc.c:180 +0x8a fp=0x221031efa0
runtime.goexit()
/Users/rsc/g/go/src/pkg/runtime/proc.c:1269 fp=0x221031efa8
goroutine 2 [runnable]:
runtime.MHeap_Scavenger()
/Users/rsc/g/go/src/pkg/runtime/mheap.c:438
runtime.goexit()
/Users/rsc/g/go/src/pkg/runtime/proc.c:1269
created by runtime.main
/Users/rsc/g/go/src/pkg/runtime/proc.c:166
rax 0x23ccc0
rbx 0x23ccc0
rcx 0x0
rdx 0x38
rdi 0x2102c0170
rsi 0x221032cfe0
rbp 0x221032cfa0
rsp 0x7fff5fbff5b0
r8 0x2102c0120
r9 0x221032cfa0
r10 0x221032c000
r11 0x104ce8
r12 0xe5c80
r13 0x1be82baac718
r14 0x13091135f7d69200
r15 0x0
rip 0x1b2a6
rflags 0x10246
cs 0x2b
fs 0x0
gs 0x0
Fixes #5723.
R=r, dvyukov, go.peter.90, dave, iant
CC=golang-dev
https://golang.org/cl/10360048
2013-06-27 09:32:01 -06:00
|
|
|
// Set g->sched to context in f.
|
|
|
|
MOVL 0(SP), AX // f's PC
|
|
|
|
MOVL AX, (g_sched+gobuf_pc)(SI)
|
|
|
|
MOVL SI, (g_sched+gobuf_g)(SI)
|
|
|
|
LEAL 4(SP), AX // f's SP
|
|
|
|
MOVL AX, (g_sched+gobuf_sp)(SI)
|
|
|
|
MOVL DX, (g_sched+gobuf_ctxt)(SI)
|
2009-06-17 16:15:55 -06:00
|
|
|
|
runtime: scheduler, cgo reorganization
* Change use of m->g0 stack (aka scheduler stack).
* Provide runtime.mcall(f) to invoke f() on m->g0 stack.
* Replace scheduler loop entry with runtime.mcall(schedule).
Runtime.mcall eliminates the need for fake scheduler states that
exist just to run a bit of code on the m->g0 stack
(Grecovery, Gstackalloc).
The elimination of the scheduler as a loop that stops and
starts using gosave and gogo fixes a bad interaction with the
way cgo uses the m->g0 stack. Cgo runs external (gcc-compiled)
C functions on that stack, and then when calling back into Go,
it sets m->g0->sched.sp below the added call frames, so that
other uses of m->g0's stack will not interfere with those frames.
Unfortunately, gogo (longjmp) back to the scheduler loop at
this point would end up running scheduler with the lower
sp, which no longer points at a valid stack frame for
a call to scheduler. If scheduler then wrote any function call
arguments or local variables to where it expected the stack
frame to be, it would overwrite other data on the stack.
I realized this possibility while debugging a problem with
calling complex Go code in a Go -> C -> Go cgo callback.
This wasn't the bug I was looking for, it turns out, but I believe
it is a real bug nonetheless. Switching to runtime.mcall, which
only adds new frames to the stack and never jumps into
functions running in existing ones, fixes this bug.
* Move cgo-related code out of proc.c into cgocall.c.
* Add very large comment describing cgo call sequences.
* Simpilify, regularize cgo function implementations and names.
* Add test suite as misc/cgo/test.
Now the Go -> C path calls cgocall, which calls asmcgocall,
and the C -> Go path calls cgocallback, which calls cgocallbackg.
The shuffling, which affects mainly the callback case, moves
most of the callback implementation to cgocallback running
on the m->curg stack (not the m->g0 scheduler stack) and
only while accounted for with $GOMAXPROCS (between calls
to exitsyscall and entersyscall).
The previous callback code did not block in startcgocallback's
approximation to exitsyscall, so if, say, the garbage collector
were running, it would still barge in and start doing things
like call malloc. Similarly endcgocallback's approximation of
entersyscall did not call matchmg to kick off new OS threads
when necessary, which caused the bug in issue 1560.
Fixes #1560.
R=iant
CC=golang-dev
https://golang.org/cl/4253054
2011-03-07 08:37:42 -07:00
|
|
|
// Call newstack on m->g0's stack.
|
2009-06-17 16:15:55 -06:00
|
|
|
MOVL m_g0(BX), BP
|
2010-01-06 18:58:55 -07:00
|
|
|
MOVL BP, g(CX)
|
runtime: scheduler, cgo reorganization
* Change use of m->g0 stack (aka scheduler stack).
* Provide runtime.mcall(f) to invoke f() on m->g0 stack.
* Replace scheduler loop entry with runtime.mcall(schedule).
Runtime.mcall eliminates the need for fake scheduler states that
exist just to run a bit of code on the m->g0 stack
(Grecovery, Gstackalloc).
The elimination of the scheduler as a loop that stops and
starts using gosave and gogo fixes a bad interaction with the
way cgo uses the m->g0 stack. Cgo runs external (gcc-compiled)
C functions on that stack, and then when calling back into Go,
it sets m->g0->sched.sp below the added call frames, so that
other uses of m->g0's stack will not interfere with those frames.
Unfortunately, gogo (longjmp) back to the scheduler loop at
this point would end up running scheduler with the lower
sp, which no longer points at a valid stack frame for
a call to scheduler. If scheduler then wrote any function call
arguments or local variables to where it expected the stack
frame to be, it would overwrite other data on the stack.
I realized this possibility while debugging a problem with
calling complex Go code in a Go -> C -> Go cgo callback.
This wasn't the bug I was looking for, it turns out, but I believe
it is a real bug nonetheless. Switching to runtime.mcall, which
only adds new frames to the stack and never jumps into
functions running in existing ones, fixes this bug.
* Move cgo-related code out of proc.c into cgocall.c.
* Add very large comment describing cgo call sequences.
* Simpilify, regularize cgo function implementations and names.
* Add test suite as misc/cgo/test.
Now the Go -> C path calls cgocall, which calls asmcgocall,
and the C -> Go path calls cgocallback, which calls cgocallbackg.
The shuffling, which affects mainly the callback case, moves
most of the callback implementation to cgocallback running
on the m->curg stack (not the m->g0 scheduler stack) and
only while accounted for with $GOMAXPROCS (between calls
to exitsyscall and entersyscall).
The previous callback code did not block in startcgocallback's
approximation to exitsyscall, so if, say, the garbage collector
were running, it would still barge in and start doing things
like call malloc. Similarly endcgocallback's approximation of
entersyscall did not call matchmg to kick off new OS threads
when necessary, which caused the bug in issue 1560.
Fixes #1560.
R=iant
CC=golang-dev
https://golang.org/cl/4253054
2011-03-07 08:37:42 -07:00
|
|
|
MOVL (g_sched+gobuf_sp)(BP), AX
|
2010-12-07 15:19:36 -07:00
|
|
|
MOVL -4(AX), BX // fault if CALL would, before smashing SP
|
|
|
|
MOVL AX, SP
|
runtime: ,s/[a-zA-Z0-9_]+/runtime·&/g, almost
Prefix all external symbols in runtime by runtime·,
to avoid conflicts with possible symbols of the same
name in linked-in C libraries. The obvious conflicts
are printf, malloc, and free, but hide everything to
avoid future pain.
The symbols left alone are:
** known to cgo **
_cgo_free
_cgo_malloc
libcgo_thread_start
initcgo
ncgocall
** known to linker **
_rt0_$GOARCH
_rt0_$GOARCH_$GOOS
text
etext
data
end
pclntab
epclntab
symtab
esymtab
** known to C compiler **
_divv
_modv
_div64by32
etc (arch specific)
Tested on darwin/386, darwin/amd64, linux/386, linux/amd64.
Built (but not tested) for freebsd/386, freebsd/amd64, linux/arm, windows/386.
R=r, PeterGo
CC=golang-dev
https://golang.org/cl/2899041
2010-11-04 12:00:19 -06:00
|
|
|
CALL runtime·newstack(SB)
|
2009-06-17 16:15:55 -06:00
|
|
|
MOVL $0, 0x1003 // crash if newstack returns
|
2009-03-30 01:01:07 -06:00
|
|
|
RET
|
|
|
|
|
2009-07-08 19:16:09 -06:00
|
|
|
// Called from reflection library. Mimics morestack,
|
|
|
|
// reuses stack growth code to create a frame
|
|
|
|
// with the desired args running the desired function.
|
|
|
|
//
|
|
|
|
// func call(fn *byte, arg *byte, argsize uint32).
|
2013-07-16 14:24:09 -06:00
|
|
|
TEXT reflect·call(SB), 7, $0-12
|
2010-01-06 18:58:55 -07:00
|
|
|
get_tls(CX)
|
|
|
|
MOVL m(CX), BX
|
2009-07-08 19:16:09 -06:00
|
|
|
|
|
|
|
// Save our caller's state as the PC and SP to
|
|
|
|
// restore when returning from f.
|
|
|
|
MOVL 0(SP), AX // our caller's PC
|
|
|
|
MOVL AX, (m_morebuf+gobuf_pc)(BX)
|
|
|
|
LEAL 4(SP), AX // our caller's SP
|
|
|
|
MOVL AX, (m_morebuf+gobuf_sp)(BX)
|
2010-01-06 18:58:55 -07:00
|
|
|
MOVL g(CX), AX
|
2009-07-08 19:16:09 -06:00
|
|
|
MOVL AX, (m_morebuf+gobuf_g)(BX)
|
|
|
|
|
2013-06-27 14:51:06 -06:00
|
|
|
// Save our own state as the PC and SP to restore
|
|
|
|
// if this goroutine needs to be restarted.
|
|
|
|
MOVL $reflect·call(SB), (g_sched+gobuf_pc)(AX)
|
|
|
|
MOVL SP, (g_sched+gobuf_sp)(AX)
|
|
|
|
|
2009-07-08 19:16:09 -06:00
|
|
|
// Set up morestack arguments to call f on a new stack.
|
2010-03-29 22:48:22 -06:00
|
|
|
// We set f's frame size to 1, as a hint to newstack
|
|
|
|
// that this is a call from reflect·call.
|
|
|
|
// If it turns out that f needs a larger frame than
|
|
|
|
// the default stack, f's usual stack growth prolog will
|
|
|
|
// allocate a new segment (and recopy the arguments).
|
2009-07-08 19:16:09 -06:00
|
|
|
MOVL 4(SP), AX // fn
|
|
|
|
MOVL 8(SP), DX // arg frame
|
|
|
|
MOVL 12(SP), CX // arg size
|
|
|
|
|
runtime: record proper goroutine state during stack split
Until now, the goroutine state has been scattered during the
execution of newstack and oldstack. It's all there, and those routines
know how to get back to a working goroutine, but other pieces of
the system, like stack traces, do not. If something does interrupt
the newstack or oldstack execution, the rest of the system can't
understand the goroutine. For example, if newstack decides there
is an overflow and calls throw, the stack tracer wouldn't dump the
goroutine correctly.
For newstack to save a useful state snapshot, it needs to be able
to rewind the PC in the function that triggered the split back to
the beginning of the function. (The PC is a few instructions in, just
after the call to morestack.) To make that possible, we change the
prologues to insert a jmp back to the beginning of the function
after the call to morestack. That is, the prologue used to be roughly:
TEXT myfunc
check for split
jmpcond nosplit
call morestack
nosplit:
sub $xxx, sp
Now an extra instruction is inserted after the call:
TEXT myfunc
start:
check for split
jmpcond nosplit
call morestack
jmp start
nosplit:
sub $xxx, sp
The jmp is not executed directly. It is decoded and simulated by
runtime.rewindmorestack to discover the beginning of the function,
and then the call to morestack returns directly to the start label
instead of to the jump instruction. So logically the jmp is still
executed, just not by the cpu.
The prologue thus repeats in the case of a function that needs a
stack split, but against the cost of the split itself, the extra few
instructions are noise. The repeated prologue has the nice effect of
making a stack split double-check that the new stack is big enough:
if morestack happens to return on a too-small stack, we'll now notice
before corruption happens.
The ability for newstack to rewind to the beginning of the function
should help preemption too. If newstack decides that it was called
for preemption instead of a stack split, it now has the goroutine state
correctly paused if rescheduling is needed, and when the goroutine
can run again, it can return to the start label on its original stack
and re-execute the split check.
Here is an example of a split stack overflow showing the full
trace, without any special cases in the stack printer.
(This one was triggered by making the split check incorrect.)
runtime: newstack framesize=0x0 argsize=0x18 sp=0x6aebd0 stack=[0x6b0000, 0x6b0fa0]
morebuf={pc:0x69f5b sp:0x6aebd8 lr:0x0}
sched={pc:0x68880 sp:0x6aebd0 lr:0x0 ctxt:0x34e700}
runtime: split stack overflow: 0x6aebd0 < 0x6b0000
fatal error: runtime: split stack overflow
goroutine 1 [stack split]:
runtime.mallocgc(0x290, 0x100000000, 0x1)
/Users/rsc/g/go/src/pkg/runtime/zmalloc_darwin_amd64.c:21 fp=0x6aebd8
runtime.new()
/Users/rsc/g/go/src/pkg/runtime/zmalloc_darwin_amd64.c:682 +0x5b fp=0x6aec08
go/build.(*Context).Import(0x5ae340, 0xc210030c71, 0xa, 0xc2100b4380, 0x1b, ...)
/Users/rsc/g/go/src/pkg/go/build/build.go:424 +0x3a fp=0x6b00a0
main.loadImport(0xc210030c71, 0xa, 0xc2100b4380, 0x1b, 0xc2100b42c0, ...)
/Users/rsc/g/go/src/cmd/go/pkg.go:249 +0x371 fp=0x6b01a8
main.(*Package).load(0xc21017c800, 0xc2100b42c0, 0xc2101828c0, 0x0, 0x0, ...)
/Users/rsc/g/go/src/cmd/go/pkg.go:431 +0x2801 fp=0x6b0c98
main.loadPackage(0x369040, 0x7, 0xc2100b42c0, 0x0)
/Users/rsc/g/go/src/cmd/go/pkg.go:709 +0x857 fp=0x6b0f80
----- stack segment boundary -----
main.(*builder).action(0xc2100902a0, 0x0, 0x0, 0xc2100e6c00, 0xc2100e5750, ...)
/Users/rsc/g/go/src/cmd/go/build.go:539 +0x437 fp=0x6b14a0
main.(*builder).action(0xc2100902a0, 0x0, 0x0, 0xc21015b400, 0x2, ...)
/Users/rsc/g/go/src/cmd/go/build.go:528 +0x1d2 fp=0x6b1658
main.(*builder).test(0xc2100902a0, 0xc210092000, 0x0, 0x0, 0xc21008ff60, ...)
/Users/rsc/g/go/src/cmd/go/test.go:622 +0x1b53 fp=0x6b1f68
----- stack segment boundary -----
main.runTest(0x5a6b20, 0xc21000a020, 0x2, 0x2)
/Users/rsc/g/go/src/cmd/go/test.go:366 +0xd09 fp=0x6a5cf0
main.main()
/Users/rsc/g/go/src/cmd/go/main.go:161 +0x4f9 fp=0x6a5f78
runtime.main()
/Users/rsc/g/go/src/pkg/runtime/proc.c:183 +0x92 fp=0x6a5fa0
runtime.goexit()
/Users/rsc/g/go/src/pkg/runtime/proc.c:1266 fp=0x6a5fa8
And here is a seg fault during oldstack:
SIGSEGV: segmentation violation
PC=0x1b2a6
runtime.oldstack()
/Users/rsc/g/go/src/pkg/runtime/stack.c:159 +0x76
runtime.lessstack()
/Users/rsc/g/go/src/pkg/runtime/asm_amd64.s:270 +0x22
goroutine 1 [stack unsplit]:
fmt.(*pp).printArg(0x2102e64e0, 0xe5c80, 0x2102c9220, 0x73, 0x0, ...)
/Users/rsc/g/go/src/pkg/fmt/print.go:818 +0x3d3 fp=0x221031e6f8
fmt.(*pp).doPrintf(0x2102e64e0, 0x12fb20, 0x2, 0x221031eb98, 0x1, ...)
/Users/rsc/g/go/src/pkg/fmt/print.go:1183 +0x15cb fp=0x221031eaf0
fmt.Sprintf(0x12fb20, 0x2, 0x221031eb98, 0x1, 0x1, ...)
/Users/rsc/g/go/src/pkg/fmt/print.go:234 +0x67 fp=0x221031eb40
flag.(*stringValue).String(0x2102c9210, 0x1, 0x0)
/Users/rsc/g/go/src/pkg/flag/flag.go:180 +0xb3 fp=0x221031ebb0
flag.(*FlagSet).Var(0x2102f6000, 0x293d38, 0x2102c9210, 0x143490, 0xa, ...)
/Users/rsc/g/go/src/pkg/flag/flag.go:633 +0x40 fp=0x221031eca0
flag.(*FlagSet).StringVar(0x2102f6000, 0x2102c9210, 0x143490, 0xa, 0x12fa60, ...)
/Users/rsc/g/go/src/pkg/flag/flag.go:550 +0x91 fp=0x221031ece8
flag.(*FlagSet).String(0x2102f6000, 0x143490, 0xa, 0x12fa60, 0x0, ...)
/Users/rsc/g/go/src/pkg/flag/flag.go:563 +0x87 fp=0x221031ed38
flag.String(0x143490, 0xa, 0x12fa60, 0x0, 0x161950, ...)
/Users/rsc/g/go/src/pkg/flag/flag.go:570 +0x6b fp=0x221031ed80
testing.init()
/Users/rsc/g/go/src/pkg/testing/testing.go:-531 +0xbb fp=0x221031edc0
strings_test.init()
/Users/rsc/g/go/src/pkg/strings/strings_test.go:1115 +0x62 fp=0x221031ef70
main.init()
strings/_test/_testmain.go:90 +0x3d fp=0x221031ef78
runtime.main()
/Users/rsc/g/go/src/pkg/runtime/proc.c:180 +0x8a fp=0x221031efa0
runtime.goexit()
/Users/rsc/g/go/src/pkg/runtime/proc.c:1269 fp=0x221031efa8
goroutine 2 [runnable]:
runtime.MHeap_Scavenger()
/Users/rsc/g/go/src/pkg/runtime/mheap.c:438
runtime.goexit()
/Users/rsc/g/go/src/pkg/runtime/proc.c:1269
created by runtime.main
/Users/rsc/g/go/src/pkg/runtime/proc.c:166
rax 0x23ccc0
rbx 0x23ccc0
rcx 0x0
rdx 0x38
rdi 0x2102c0170
rsi 0x221032cfe0
rbp 0x221032cfa0
rsp 0x7fff5fbff5b0
r8 0x2102c0120
r9 0x221032cfa0
r10 0x221032c000
r11 0x104ce8
r12 0xe5c80
r13 0x1be82baac718
r14 0x13091135f7d69200
r15 0x0
rip 0x1b2a6
rflags 0x10246
cs 0x2b
fs 0x0
gs 0x0
Fixes #5723.
R=r, dvyukov, go.peter.90, dave, iant
CC=golang-dev
https://golang.org/cl/10360048
2013-06-27 09:32:01 -06:00
|
|
|
MOVL AX, m_cret(BX) // f's PC
|
2011-01-14 12:05:20 -07:00
|
|
|
MOVL DX, m_moreargp(BX) // f's argument pointer
|
|
|
|
MOVL CX, m_moreargsize(BX) // f's argument size
|
|
|
|
MOVL $1, m_moreframesize(BX) // f's frame size
|
2009-07-08 19:16:09 -06:00
|
|
|
|
runtime: scheduler, cgo reorganization
* Change use of m->g0 stack (aka scheduler stack).
* Provide runtime.mcall(f) to invoke f() on m->g0 stack.
* Replace scheduler loop entry with runtime.mcall(schedule).
Runtime.mcall eliminates the need for fake scheduler states that
exist just to run a bit of code on the m->g0 stack
(Grecovery, Gstackalloc).
The elimination of the scheduler as a loop that stops and
starts using gosave and gogo fixes a bad interaction with the
way cgo uses the m->g0 stack. Cgo runs external (gcc-compiled)
C functions on that stack, and then when calling back into Go,
it sets m->g0->sched.sp below the added call frames, so that
other uses of m->g0's stack will not interfere with those frames.
Unfortunately, gogo (longjmp) back to the scheduler loop at
this point would end up running scheduler with the lower
sp, which no longer points at a valid stack frame for
a call to scheduler. If scheduler then wrote any function call
arguments or local variables to where it expected the stack
frame to be, it would overwrite other data on the stack.
I realized this possibility while debugging a problem with
calling complex Go code in a Go -> C -> Go cgo callback.
This wasn't the bug I was looking for, it turns out, but I believe
it is a real bug nonetheless. Switching to runtime.mcall, which
only adds new frames to the stack and never jumps into
functions running in existing ones, fixes this bug.
* Move cgo-related code out of proc.c into cgocall.c.
* Add very large comment describing cgo call sequences.
* Simpilify, regularize cgo function implementations and names.
* Add test suite as misc/cgo/test.
Now the Go -> C path calls cgocall, which calls asmcgocall,
and the C -> Go path calls cgocallback, which calls cgocallbackg.
The shuffling, which affects mainly the callback case, moves
most of the callback implementation to cgocallback running
on the m->curg stack (not the m->g0 scheduler stack) and
only while accounted for with $GOMAXPROCS (between calls
to exitsyscall and entersyscall).
The previous callback code did not block in startcgocallback's
approximation to exitsyscall, so if, say, the garbage collector
were running, it would still barge in and start doing things
like call malloc. Similarly endcgocallback's approximation of
entersyscall did not call matchmg to kick off new OS threads
when necessary, which caused the bug in issue 1560.
Fixes #1560.
R=iant
CC=golang-dev
https://golang.org/cl/4253054
2011-03-07 08:37:42 -07:00
|
|
|
// Call newstack on m->g0's stack.
|
2009-07-08 19:16:09 -06:00
|
|
|
MOVL m_g0(BX), BP
|
2010-01-06 18:58:55 -07:00
|
|
|
get_tls(CX)
|
|
|
|
MOVL BP, g(CX)
|
runtime: scheduler, cgo reorganization
* Change use of m->g0 stack (aka scheduler stack).
* Provide runtime.mcall(f) to invoke f() on m->g0 stack.
* Replace scheduler loop entry with runtime.mcall(schedule).
Runtime.mcall eliminates the need for fake scheduler states that
exist just to run a bit of code on the m->g0 stack
(Grecovery, Gstackalloc).
The elimination of the scheduler as a loop that stops and
starts using gosave and gogo fixes a bad interaction with the
way cgo uses the m->g0 stack. Cgo runs external (gcc-compiled)
C functions on that stack, and then when calling back into Go,
it sets m->g0->sched.sp below the added call frames, so that
other uses of m->g0's stack will not interfere with those frames.
Unfortunately, gogo (longjmp) back to the scheduler loop at
this point would end up running scheduler with the lower
sp, which no longer points at a valid stack frame for
a call to scheduler. If scheduler then wrote any function call
arguments or local variables to where it expected the stack
frame to be, it would overwrite other data on the stack.
I realized this possibility while debugging a problem with
calling complex Go code in a Go -> C -> Go cgo callback.
This wasn't the bug I was looking for, it turns out, but I believe
it is a real bug nonetheless. Switching to runtime.mcall, which
only adds new frames to the stack and never jumps into
functions running in existing ones, fixes this bug.
* Move cgo-related code out of proc.c into cgocall.c.
* Add very large comment describing cgo call sequences.
* Simpilify, regularize cgo function implementations and names.
* Add test suite as misc/cgo/test.
Now the Go -> C path calls cgocall, which calls asmcgocall,
and the C -> Go path calls cgocallback, which calls cgocallbackg.
The shuffling, which affects mainly the callback case, moves
most of the callback implementation to cgocallback running
on the m->curg stack (not the m->g0 scheduler stack) and
only while accounted for with $GOMAXPROCS (between calls
to exitsyscall and entersyscall).
The previous callback code did not block in startcgocallback's
approximation to exitsyscall, so if, say, the garbage collector
were running, it would still barge in and start doing things
like call malloc. Similarly endcgocallback's approximation of
entersyscall did not call matchmg to kick off new OS threads
when necessary, which caused the bug in issue 1560.
Fixes #1560.
R=iant
CC=golang-dev
https://golang.org/cl/4253054
2011-03-07 08:37:42 -07:00
|
|
|
MOVL (g_sched+gobuf_sp)(BP), SP
|
runtime: ,s/[a-zA-Z0-9_]+/runtime·&/g, almost
Prefix all external symbols in runtime by runtime·,
to avoid conflicts with possible symbols of the same
name in linked-in C libraries. The obvious conflicts
are printf, malloc, and free, but hide everything to
avoid future pain.
The symbols left alone are:
** known to cgo **
_cgo_free
_cgo_malloc
libcgo_thread_start
initcgo
ncgocall
** known to linker **
_rt0_$GOARCH
_rt0_$GOARCH_$GOOS
text
etext
data
end
pclntab
epclntab
symtab
esymtab
** known to C compiler **
_divv
_modv
_div64by32
etc (arch specific)
Tested on darwin/386, darwin/amd64, linux/386, linux/amd64.
Built (but not tested) for freebsd/386, freebsd/amd64, linux/arm, windows/386.
R=r, PeterGo
CC=golang-dev
https://golang.org/cl/2899041
2010-11-04 12:00:19 -06:00
|
|
|
CALL runtime·newstack(SB)
|
2009-07-08 19:16:09 -06:00
|
|
|
MOVL $0, 0x1103 // crash if newstack returns
|
|
|
|
RET
|
|
|
|
|
|
|
|
|
2009-06-17 16:15:55 -06:00
|
|
|
// Return point when leaving stack.
|
2013-07-18 14:53:45 -06:00
|
|
|
//
|
|
|
|
// Lessstack can appear in stack traces for the same reason
|
|
|
|
// as morestack; in that context, it has 0 arguments.
|
|
|
|
TEXT runtime·lessstack(SB), 7, $0-0
|
2009-06-17 16:15:55 -06:00
|
|
|
// Save return value in m->cret
|
2010-01-06 18:58:55 -07:00
|
|
|
get_tls(CX)
|
|
|
|
MOVL m(CX), BX
|
2009-06-17 16:15:55 -06:00
|
|
|
MOVL AX, m_cret(BX)
|
|
|
|
|
runtime: scheduler, cgo reorganization
* Change use of m->g0 stack (aka scheduler stack).
* Provide runtime.mcall(f) to invoke f() on m->g0 stack.
* Replace scheduler loop entry with runtime.mcall(schedule).
Runtime.mcall eliminates the need for fake scheduler states that
exist just to run a bit of code on the m->g0 stack
(Grecovery, Gstackalloc).
The elimination of the scheduler as a loop that stops and
starts using gosave and gogo fixes a bad interaction with the
way cgo uses the m->g0 stack. Cgo runs external (gcc-compiled)
C functions on that stack, and then when calling back into Go,
it sets m->g0->sched.sp below the added call frames, so that
other uses of m->g0's stack will not interfere with those frames.
Unfortunately, gogo (longjmp) back to the scheduler loop at
this point would end up running scheduler with the lower
sp, which no longer points at a valid stack frame for
a call to scheduler. If scheduler then wrote any function call
arguments or local variables to where it expected the stack
frame to be, it would overwrite other data on the stack.
I realized this possibility while debugging a problem with
calling complex Go code in a Go -> C -> Go cgo callback.
This wasn't the bug I was looking for, it turns out, but I believe
it is a real bug nonetheless. Switching to runtime.mcall, which
only adds new frames to the stack and never jumps into
functions running in existing ones, fixes this bug.
* Move cgo-related code out of proc.c into cgocall.c.
* Add very large comment describing cgo call sequences.
* Simpilify, regularize cgo function implementations and names.
* Add test suite as misc/cgo/test.
Now the Go -> C path calls cgocall, which calls asmcgocall,
and the C -> Go path calls cgocallback, which calls cgocallbackg.
The shuffling, which affects mainly the callback case, moves
most of the callback implementation to cgocallback running
on the m->curg stack (not the m->g0 scheduler stack) and
only while accounted for with $GOMAXPROCS (between calls
to exitsyscall and entersyscall).
The previous callback code did not block in startcgocallback's
approximation to exitsyscall, so if, say, the garbage collector
were running, it would still barge in and start doing things
like call malloc. Similarly endcgocallback's approximation of
entersyscall did not call matchmg to kick off new OS threads
when necessary, which caused the bug in issue 1560.
Fixes #1560.
R=iant
CC=golang-dev
https://golang.org/cl/4253054
2011-03-07 08:37:42 -07:00
|
|
|
// Call oldstack on m->g0's stack.
|
|
|
|
MOVL m_g0(BX), BP
|
|
|
|
MOVL BP, g(CX)
|
|
|
|
MOVL (g_sched+gobuf_sp)(BP), SP
|
runtime: ,s/[a-zA-Z0-9_]+/runtime·&/g, almost
Prefix all external symbols in runtime by runtime·,
to avoid conflicts with possible symbols of the same
name in linked-in C libraries. The obvious conflicts
are printf, malloc, and free, but hide everything to
avoid future pain.
The symbols left alone are:
** known to cgo **
_cgo_free
_cgo_malloc
libcgo_thread_start
initcgo
ncgocall
** known to linker **
_rt0_$GOARCH
_rt0_$GOARCH_$GOOS
text
etext
data
end
pclntab
epclntab
symtab
esymtab
** known to C compiler **
_divv
_modv
_div64by32
etc (arch specific)
Tested on darwin/386, darwin/amd64, linux/386, linux/amd64.
Built (but not tested) for freebsd/386, freebsd/amd64, linux/arm, windows/386.
R=r, PeterGo
CC=golang-dev
https://golang.org/cl/2899041
2010-11-04 12:00:19 -06:00
|
|
|
CALL runtime·oldstack(SB)
|
2009-06-17 16:15:55 -06:00
|
|
|
MOVL $0, 0x1004 // crash if oldstack returns
|
2009-03-30 01:01:07 -06:00
|
|
|
RET
|
|
|
|
|
2009-06-17 16:15:55 -06:00
|
|
|
|
2009-03-30 01:01:07 -06:00
|
|
|
// bool cas(int32 *val, int32 old, int32 new)
|
|
|
|
// Atomically:
|
|
|
|
// if(*val == old){
|
|
|
|
// *val = new;
|
|
|
|
// return 1;
|
|
|
|
// }else
|
|
|
|
// return 0;
|
2013-07-16 14:24:09 -06:00
|
|
|
TEXT runtime·cas(SB), 7, $0-12
|
2009-03-30 01:01:07 -06:00
|
|
|
MOVL 4(SP), BX
|
|
|
|
MOVL 8(SP), AX
|
|
|
|
MOVL 12(SP), CX
|
|
|
|
LOCK
|
|
|
|
CMPXCHGL CX, 0(BX)
|
|
|
|
JZ 3(PC)
|
|
|
|
MOVL $0, AX
|
|
|
|
RET
|
|
|
|
MOVL $1, AX
|
|
|
|
RET
|
|
|
|
|
2013-07-11 22:42:46 -06:00
|
|
|
// bool runtime·cas64(uint64 *val, uint64 old, uint64 new)
|
2012-04-05 08:47:43 -06:00
|
|
|
// Atomically:
|
|
|
|
// if(*val == *old){
|
|
|
|
// *val = new;
|
|
|
|
// return 1;
|
|
|
|
// } else {
|
|
|
|
// return 0;
|
|
|
|
// }
|
2013-07-16 14:24:09 -06:00
|
|
|
TEXT runtime·cas64(SB), 7, $0-20
|
2012-04-05 08:47:43 -06:00
|
|
|
MOVL 4(SP), BP
|
2013-07-11 22:42:46 -06:00
|
|
|
MOVL 8(SP), AX
|
|
|
|
MOVL 12(SP), DX
|
|
|
|
MOVL 16(SP), BX
|
|
|
|
MOVL 20(SP), CX
|
2012-04-05 08:47:43 -06:00
|
|
|
LOCK
|
|
|
|
CMPXCHG8B 0(BP)
|
|
|
|
JNZ cas64_fail
|
|
|
|
MOVL $1, AX
|
|
|
|
RET
|
|
|
|
cas64_fail:
|
2012-04-05 08:59:50 -06:00
|
|
|
MOVL $0, AX
|
2012-04-05 08:47:43 -06:00
|
|
|
RET
|
|
|
|
|
2010-01-06 18:58:55 -07:00
|
|
|
// bool casp(void **p, void *old, void *new)
|
|
|
|
// Atomically:
|
|
|
|
// if(*p == old){
|
|
|
|
// *p = new;
|
|
|
|
// return 1;
|
|
|
|
// }else
|
|
|
|
// return 0;
|
2013-07-16 14:24:09 -06:00
|
|
|
TEXT runtime·casp(SB), 7, $0-12
|
2010-01-06 18:58:55 -07:00
|
|
|
MOVL 4(SP), BX
|
|
|
|
MOVL 8(SP), AX
|
|
|
|
MOVL 12(SP), CX
|
|
|
|
LOCK
|
|
|
|
CMPXCHGL CX, 0(BX)
|
|
|
|
JZ 3(PC)
|
|
|
|
MOVL $0, AX
|
|
|
|
RET
|
|
|
|
MOVL $1, AX
|
|
|
|
RET
|
|
|
|
|
2011-07-15 09:27:16 -06:00
|
|
|
// uint32 xadd(uint32 volatile *val, int32 delta)
|
|
|
|
// Atomically:
|
|
|
|
// *val += delta;
|
|
|
|
// return *val;
|
2013-07-16 14:24:09 -06:00
|
|
|
TEXT runtime·xadd(SB), 7, $0-8
|
2011-07-15 09:27:16 -06:00
|
|
|
MOVL 4(SP), BX
|
|
|
|
MOVL 8(SP), AX
|
|
|
|
MOVL AX, CX
|
|
|
|
LOCK
|
|
|
|
XADDL AX, 0(BX)
|
|
|
|
ADDL CX, AX
|
|
|
|
RET
|
|
|
|
|
2013-07-16 14:24:09 -06:00
|
|
|
TEXT runtime·xchg(SB), 7, $0-8
|
runtime: improve Linux mutex
The implementation is hybrid active/passive spin/blocking mutex.
The design minimizes amount of context switches and futex calls.
The idea is that all critical sections in runtime are intentially
small, so pure blocking mutex behaves badly causing
a lot of context switches, thread parking/unparking and kernel calls.
Note that some synthetic benchmarks become somewhat slower,
that's due to increased contention on other data structures,
it should not affect programs that do any real work.
On 2 x Intel E5620, 8 HT cores, 2.4GHz
benchmark old ns/op new ns/op delta
BenchmarkSelectContended 521.00 503.00 -3.45%
BenchmarkSelectContended-2 661.00 320.00 -51.59%
BenchmarkSelectContended-4 1139.00 629.00 -44.78%
BenchmarkSelectContended-8 2870.00 878.00 -69.41%
BenchmarkSelectContended-16 5276.00 818.00 -84.50%
BenchmarkChanContended 112.00 103.00 -8.04%
BenchmarkChanContended-2 631.00 174.00 -72.42%
BenchmarkChanContended-4 682.00 272.00 -60.12%
BenchmarkChanContended-8 1601.00 520.00 -67.52%
BenchmarkChanContended-16 3100.00 372.00 -88.00%
BenchmarkChanSync 253.00 239.00 -5.53%
BenchmarkChanSync-2 5030.00 4648.00 -7.59%
BenchmarkChanSync-4 4826.00 4694.00 -2.74%
BenchmarkChanSync-8 4778.00 4713.00 -1.36%
BenchmarkChanSync-16 5289.00 4710.00 -10.95%
BenchmarkChanProdCons0 273.00 254.00 -6.96%
BenchmarkChanProdCons0-2 599.00 400.00 -33.22%
BenchmarkChanProdCons0-4 1168.00 659.00 -43.58%
BenchmarkChanProdCons0-8 2831.00 1057.00 -62.66%
BenchmarkChanProdCons0-16 4197.00 1037.00 -75.29%
BenchmarkChanProdCons10 150.00 140.00 -6.67%
BenchmarkChanProdCons10-2 607.00 268.00 -55.85%
BenchmarkChanProdCons10-4 1137.00 404.00 -64.47%
BenchmarkChanProdCons10-8 2115.00 828.00 -60.85%
BenchmarkChanProdCons10-16 4283.00 855.00 -80.04%
BenchmarkChanProdCons100 117.00 110.00 -5.98%
BenchmarkChanProdCons100-2 558.00 218.00 -60.93%
BenchmarkChanProdCons100-4 722.00 287.00 -60.25%
BenchmarkChanProdCons100-8 1840.00 431.00 -76.58%
BenchmarkChanProdCons100-16 3394.00 448.00 -86.80%
BenchmarkChanProdConsWork0 2014.00 1996.00 -0.89%
BenchmarkChanProdConsWork0-2 1207.00 1127.00 -6.63%
BenchmarkChanProdConsWork0-4 1913.00 611.00 -68.06%
BenchmarkChanProdConsWork0-8 3016.00 949.00 -68.53%
BenchmarkChanProdConsWork0-16 4320.00 1154.00 -73.29%
BenchmarkChanProdConsWork10 1906.00 1897.00 -0.47%
BenchmarkChanProdConsWork10-2 1123.00 1033.00 -8.01%
BenchmarkChanProdConsWork10-4 1076.00 571.00 -46.93%
BenchmarkChanProdConsWork10-8 2748.00 1096.00 -60.12%
BenchmarkChanProdConsWork10-16 4600.00 1105.00 -75.98%
BenchmarkChanProdConsWork100 1884.00 1852.00 -1.70%
BenchmarkChanProdConsWork100-2 1235.00 1146.00 -7.21%
BenchmarkChanProdConsWork100-4 1217.00 619.00 -49.14%
BenchmarkChanProdConsWork100-8 1534.00 509.00 -66.82%
BenchmarkChanProdConsWork100-16 4126.00 918.00 -77.75%
BenchmarkSyscall 34.40 33.30 -3.20%
BenchmarkSyscall-2 160.00 121.00 -24.38%
BenchmarkSyscall-4 131.00 136.00 +3.82%
BenchmarkSyscall-8 139.00 131.00 -5.76%
BenchmarkSyscall-16 161.00 168.00 +4.35%
BenchmarkSyscallWork 950.00 950.00 +0.00%
BenchmarkSyscallWork-2 481.00 480.00 -0.21%
BenchmarkSyscallWork-4 268.00 270.00 +0.75%
BenchmarkSyscallWork-8 156.00 169.00 +8.33%
BenchmarkSyscallWork-16 188.00 184.00 -2.13%
BenchmarkSemaSyntNonblock 36.40 35.60 -2.20%
BenchmarkSemaSyntNonblock-2 81.40 45.10 -44.59%
BenchmarkSemaSyntNonblock-4 126.00 108.00 -14.29%
BenchmarkSemaSyntNonblock-8 112.00 112.00 +0.00%
BenchmarkSemaSyntNonblock-16 110.00 112.00 +1.82%
BenchmarkSemaSyntBlock 35.30 35.30 +0.00%
BenchmarkSemaSyntBlock-2 118.00 124.00 +5.08%
BenchmarkSemaSyntBlock-4 105.00 108.00 +2.86%
BenchmarkSemaSyntBlock-8 101.00 111.00 +9.90%
BenchmarkSemaSyntBlock-16 112.00 118.00 +5.36%
BenchmarkSemaWorkNonblock 810.00 811.00 +0.12%
BenchmarkSemaWorkNonblock-2 476.00 414.00 -13.03%
BenchmarkSemaWorkNonblock-4 238.00 228.00 -4.20%
BenchmarkSemaWorkNonblock-8 140.00 126.00 -10.00%
BenchmarkSemaWorkNonblock-16 117.00 116.00 -0.85%
BenchmarkSemaWorkBlock 810.00 811.00 +0.12%
BenchmarkSemaWorkBlock-2 454.00 466.00 +2.64%
BenchmarkSemaWorkBlock-4 243.00 241.00 -0.82%
BenchmarkSemaWorkBlock-8 145.00 137.00 -5.52%
BenchmarkSemaWorkBlock-16 132.00 123.00 -6.82%
BenchmarkContendedSemaphore 123.00 102.00 -17.07%
BenchmarkContendedSemaphore-2 34.80 34.90 +0.29%
BenchmarkContendedSemaphore-4 34.70 34.80 +0.29%
BenchmarkContendedSemaphore-8 34.70 34.70 +0.00%
BenchmarkContendedSemaphore-16 34.80 34.70 -0.29%
BenchmarkMutex 26.80 26.00 -2.99%
BenchmarkMutex-2 108.00 45.20 -58.15%
BenchmarkMutex-4 103.00 127.00 +23.30%
BenchmarkMutex-8 109.00 147.00 +34.86%
BenchmarkMutex-16 102.00 152.00 +49.02%
BenchmarkMutexSlack 27.00 26.90 -0.37%
BenchmarkMutexSlack-2 149.00 165.00 +10.74%
BenchmarkMutexSlack-4 121.00 209.00 +72.73%
BenchmarkMutexSlack-8 101.00 158.00 +56.44%
BenchmarkMutexSlack-16 97.00 129.00 +32.99%
BenchmarkMutexWork 792.00 794.00 +0.25%
BenchmarkMutexWork-2 407.00 409.00 +0.49%
BenchmarkMutexWork-4 220.00 209.00 -5.00%
BenchmarkMutexWork-8 267.00 160.00 -40.07%
BenchmarkMutexWork-16 315.00 300.00 -4.76%
BenchmarkMutexWorkSlack 792.00 793.00 +0.13%
BenchmarkMutexWorkSlack-2 406.00 404.00 -0.49%
BenchmarkMutexWorkSlack-4 225.00 212.00 -5.78%
BenchmarkMutexWorkSlack-8 268.00 136.00 -49.25%
BenchmarkMutexWorkSlack-16 300.00 300.00 +0.00%
BenchmarkRWMutexWrite100 27.10 27.00 -0.37%
BenchmarkRWMutexWrite100-2 33.10 40.80 +23.26%
BenchmarkRWMutexWrite100-4 113.00 88.10 -22.04%
BenchmarkRWMutexWrite100-8 119.00 95.30 -19.92%
BenchmarkRWMutexWrite100-16 148.00 109.00 -26.35%
BenchmarkRWMutexWrite10 29.60 29.40 -0.68%
BenchmarkRWMutexWrite10-2 111.00 61.40 -44.68%
BenchmarkRWMutexWrite10-4 270.00 208.00 -22.96%
BenchmarkRWMutexWrite10-8 204.00 185.00 -9.31%
BenchmarkRWMutexWrite10-16 261.00 190.00 -27.20%
BenchmarkRWMutexWorkWrite100 1040.00 1036.00 -0.38%
BenchmarkRWMutexWorkWrite100-2 593.00 580.00 -2.19%
BenchmarkRWMutexWorkWrite100-4 470.00 365.00 -22.34%
BenchmarkRWMutexWorkWrite100-8 468.00 289.00 -38.25%
BenchmarkRWMutexWorkWrite100-16 604.00 374.00 -38.08%
BenchmarkRWMutexWorkWrite10 951.00 951.00 +0.00%
BenchmarkRWMutexWorkWrite10-2 1001.00 928.00 -7.29%
BenchmarkRWMutexWorkWrite10-4 1555.00 1006.00 -35.31%
BenchmarkRWMutexWorkWrite10-8 2085.00 1171.00 -43.84%
BenchmarkRWMutexWorkWrite10-16 2082.00 1614.00 -22.48%
R=rsc, iant, msolo, fw, iant
CC=golang-dev
https://golang.org/cl/4711045
2011-07-29 10:44:06 -06:00
|
|
|
MOVL 4(SP), BX
|
|
|
|
MOVL 8(SP), AX
|
|
|
|
XCHGL AX, 0(BX)
|
|
|
|
RET
|
|
|
|
|
2013-07-16 14:24:09 -06:00
|
|
|
TEXT runtime·procyield(SB),7,$0-0
|
runtime: improve Linux mutex
The implementation is hybrid active/passive spin/blocking mutex.
The design minimizes amount of context switches and futex calls.
The idea is that all critical sections in runtime are intentially
small, so pure blocking mutex behaves badly causing
a lot of context switches, thread parking/unparking and kernel calls.
Note that some synthetic benchmarks become somewhat slower,
that's due to increased contention on other data structures,
it should not affect programs that do any real work.
On 2 x Intel E5620, 8 HT cores, 2.4GHz
benchmark old ns/op new ns/op delta
BenchmarkSelectContended 521.00 503.00 -3.45%
BenchmarkSelectContended-2 661.00 320.00 -51.59%
BenchmarkSelectContended-4 1139.00 629.00 -44.78%
BenchmarkSelectContended-8 2870.00 878.00 -69.41%
BenchmarkSelectContended-16 5276.00 818.00 -84.50%
BenchmarkChanContended 112.00 103.00 -8.04%
BenchmarkChanContended-2 631.00 174.00 -72.42%
BenchmarkChanContended-4 682.00 272.00 -60.12%
BenchmarkChanContended-8 1601.00 520.00 -67.52%
BenchmarkChanContended-16 3100.00 372.00 -88.00%
BenchmarkChanSync 253.00 239.00 -5.53%
BenchmarkChanSync-2 5030.00 4648.00 -7.59%
BenchmarkChanSync-4 4826.00 4694.00 -2.74%
BenchmarkChanSync-8 4778.00 4713.00 -1.36%
BenchmarkChanSync-16 5289.00 4710.00 -10.95%
BenchmarkChanProdCons0 273.00 254.00 -6.96%
BenchmarkChanProdCons0-2 599.00 400.00 -33.22%
BenchmarkChanProdCons0-4 1168.00 659.00 -43.58%
BenchmarkChanProdCons0-8 2831.00 1057.00 -62.66%
BenchmarkChanProdCons0-16 4197.00 1037.00 -75.29%
BenchmarkChanProdCons10 150.00 140.00 -6.67%
BenchmarkChanProdCons10-2 607.00 268.00 -55.85%
BenchmarkChanProdCons10-4 1137.00 404.00 -64.47%
BenchmarkChanProdCons10-8 2115.00 828.00 -60.85%
BenchmarkChanProdCons10-16 4283.00 855.00 -80.04%
BenchmarkChanProdCons100 117.00 110.00 -5.98%
BenchmarkChanProdCons100-2 558.00 218.00 -60.93%
BenchmarkChanProdCons100-4 722.00 287.00 -60.25%
BenchmarkChanProdCons100-8 1840.00 431.00 -76.58%
BenchmarkChanProdCons100-16 3394.00 448.00 -86.80%
BenchmarkChanProdConsWork0 2014.00 1996.00 -0.89%
BenchmarkChanProdConsWork0-2 1207.00 1127.00 -6.63%
BenchmarkChanProdConsWork0-4 1913.00 611.00 -68.06%
BenchmarkChanProdConsWork0-8 3016.00 949.00 -68.53%
BenchmarkChanProdConsWork0-16 4320.00 1154.00 -73.29%
BenchmarkChanProdConsWork10 1906.00 1897.00 -0.47%
BenchmarkChanProdConsWork10-2 1123.00 1033.00 -8.01%
BenchmarkChanProdConsWork10-4 1076.00 571.00 -46.93%
BenchmarkChanProdConsWork10-8 2748.00 1096.00 -60.12%
BenchmarkChanProdConsWork10-16 4600.00 1105.00 -75.98%
BenchmarkChanProdConsWork100 1884.00 1852.00 -1.70%
BenchmarkChanProdConsWork100-2 1235.00 1146.00 -7.21%
BenchmarkChanProdConsWork100-4 1217.00 619.00 -49.14%
BenchmarkChanProdConsWork100-8 1534.00 509.00 -66.82%
BenchmarkChanProdConsWork100-16 4126.00 918.00 -77.75%
BenchmarkSyscall 34.40 33.30 -3.20%
BenchmarkSyscall-2 160.00 121.00 -24.38%
BenchmarkSyscall-4 131.00 136.00 +3.82%
BenchmarkSyscall-8 139.00 131.00 -5.76%
BenchmarkSyscall-16 161.00 168.00 +4.35%
BenchmarkSyscallWork 950.00 950.00 +0.00%
BenchmarkSyscallWork-2 481.00 480.00 -0.21%
BenchmarkSyscallWork-4 268.00 270.00 +0.75%
BenchmarkSyscallWork-8 156.00 169.00 +8.33%
BenchmarkSyscallWork-16 188.00 184.00 -2.13%
BenchmarkSemaSyntNonblock 36.40 35.60 -2.20%
BenchmarkSemaSyntNonblock-2 81.40 45.10 -44.59%
BenchmarkSemaSyntNonblock-4 126.00 108.00 -14.29%
BenchmarkSemaSyntNonblock-8 112.00 112.00 +0.00%
BenchmarkSemaSyntNonblock-16 110.00 112.00 +1.82%
BenchmarkSemaSyntBlock 35.30 35.30 +0.00%
BenchmarkSemaSyntBlock-2 118.00 124.00 +5.08%
BenchmarkSemaSyntBlock-4 105.00 108.00 +2.86%
BenchmarkSemaSyntBlock-8 101.00 111.00 +9.90%
BenchmarkSemaSyntBlock-16 112.00 118.00 +5.36%
BenchmarkSemaWorkNonblock 810.00 811.00 +0.12%
BenchmarkSemaWorkNonblock-2 476.00 414.00 -13.03%
BenchmarkSemaWorkNonblock-4 238.00 228.00 -4.20%
BenchmarkSemaWorkNonblock-8 140.00 126.00 -10.00%
BenchmarkSemaWorkNonblock-16 117.00 116.00 -0.85%
BenchmarkSemaWorkBlock 810.00 811.00 +0.12%
BenchmarkSemaWorkBlock-2 454.00 466.00 +2.64%
BenchmarkSemaWorkBlock-4 243.00 241.00 -0.82%
BenchmarkSemaWorkBlock-8 145.00 137.00 -5.52%
BenchmarkSemaWorkBlock-16 132.00 123.00 -6.82%
BenchmarkContendedSemaphore 123.00 102.00 -17.07%
BenchmarkContendedSemaphore-2 34.80 34.90 +0.29%
BenchmarkContendedSemaphore-4 34.70 34.80 +0.29%
BenchmarkContendedSemaphore-8 34.70 34.70 +0.00%
BenchmarkContendedSemaphore-16 34.80 34.70 -0.29%
BenchmarkMutex 26.80 26.00 -2.99%
BenchmarkMutex-2 108.00 45.20 -58.15%
BenchmarkMutex-4 103.00 127.00 +23.30%
BenchmarkMutex-8 109.00 147.00 +34.86%
BenchmarkMutex-16 102.00 152.00 +49.02%
BenchmarkMutexSlack 27.00 26.90 -0.37%
BenchmarkMutexSlack-2 149.00 165.00 +10.74%
BenchmarkMutexSlack-4 121.00 209.00 +72.73%
BenchmarkMutexSlack-8 101.00 158.00 +56.44%
BenchmarkMutexSlack-16 97.00 129.00 +32.99%
BenchmarkMutexWork 792.00 794.00 +0.25%
BenchmarkMutexWork-2 407.00 409.00 +0.49%
BenchmarkMutexWork-4 220.00 209.00 -5.00%
BenchmarkMutexWork-8 267.00 160.00 -40.07%
BenchmarkMutexWork-16 315.00 300.00 -4.76%
BenchmarkMutexWorkSlack 792.00 793.00 +0.13%
BenchmarkMutexWorkSlack-2 406.00 404.00 -0.49%
BenchmarkMutexWorkSlack-4 225.00 212.00 -5.78%
BenchmarkMutexWorkSlack-8 268.00 136.00 -49.25%
BenchmarkMutexWorkSlack-16 300.00 300.00 +0.00%
BenchmarkRWMutexWrite100 27.10 27.00 -0.37%
BenchmarkRWMutexWrite100-2 33.10 40.80 +23.26%
BenchmarkRWMutexWrite100-4 113.00 88.10 -22.04%
BenchmarkRWMutexWrite100-8 119.00 95.30 -19.92%
BenchmarkRWMutexWrite100-16 148.00 109.00 -26.35%
BenchmarkRWMutexWrite10 29.60 29.40 -0.68%
BenchmarkRWMutexWrite10-2 111.00 61.40 -44.68%
BenchmarkRWMutexWrite10-4 270.00 208.00 -22.96%
BenchmarkRWMutexWrite10-8 204.00 185.00 -9.31%
BenchmarkRWMutexWrite10-16 261.00 190.00 -27.20%
BenchmarkRWMutexWorkWrite100 1040.00 1036.00 -0.38%
BenchmarkRWMutexWorkWrite100-2 593.00 580.00 -2.19%
BenchmarkRWMutexWorkWrite100-4 470.00 365.00 -22.34%
BenchmarkRWMutexWorkWrite100-8 468.00 289.00 -38.25%
BenchmarkRWMutexWorkWrite100-16 604.00 374.00 -38.08%
BenchmarkRWMutexWorkWrite10 951.00 951.00 +0.00%
BenchmarkRWMutexWorkWrite10-2 1001.00 928.00 -7.29%
BenchmarkRWMutexWorkWrite10-4 1555.00 1006.00 -35.31%
BenchmarkRWMutexWorkWrite10-8 2085.00 1171.00 -43.84%
BenchmarkRWMutexWorkWrite10-16 2082.00 1614.00 -22.48%
R=rsc, iant, msolo, fw, iant
CC=golang-dev
https://golang.org/cl/4711045
2011-07-29 10:44:06 -06:00
|
|
|
MOVL 4(SP), AX
|
|
|
|
again:
|
|
|
|
PAUSE
|
|
|
|
SUBL $1, AX
|
|
|
|
JNZ again
|
|
|
|
RET
|
|
|
|
|
2013-07-16 14:24:09 -06:00
|
|
|
TEXT runtime·atomicstorep(SB), 7, $0-8
|
2011-07-13 12:22:41 -06:00
|
|
|
MOVL 4(SP), BX
|
|
|
|
MOVL 8(SP), AX
|
|
|
|
XCHGL AX, 0(BX)
|
|
|
|
RET
|
|
|
|
|
2013-07-16 14:24:09 -06:00
|
|
|
TEXT runtime·atomicstore(SB), 7, $0-8
|
2011-07-29 11:47:24 -06:00
|
|
|
MOVL 4(SP), BX
|
|
|
|
MOVL 8(SP), AX
|
|
|
|
XCHGL AX, 0(BX)
|
|
|
|
RET
|
|
|
|
|
2012-04-05 08:47:43 -06:00
|
|
|
// uint64 atomicload64(uint64 volatile* addr);
|
|
|
|
// so actually
|
|
|
|
// void atomicload64(uint64 *res, uint64 volatile *addr);
|
2013-07-16 14:24:09 -06:00
|
|
|
TEXT runtime·atomicload64(SB), 7, $0-8
|
2012-04-05 08:47:43 -06:00
|
|
|
MOVL 4(SP), BX
|
|
|
|
MOVL 8(SP), AX
|
|
|
|
// MOVQ (%EAX), %MM0
|
|
|
|
BYTE $0x0f; BYTE $0x6f; BYTE $0x00
|
|
|
|
// MOVQ %MM0, 0(%EBX)
|
|
|
|
BYTE $0x0f; BYTE $0x7f; BYTE $0x03
|
|
|
|
// EMMS
|
|
|
|
BYTE $0x0F; BYTE $0x77
|
|
|
|
RET
|
|
|
|
|
|
|
|
// void runtime·atomicstore64(uint64 volatile* addr, uint64 v);
|
2013-07-16 14:24:09 -06:00
|
|
|
TEXT runtime·atomicstore64(SB), 7, $0-12
|
2012-04-05 08:47:43 -06:00
|
|
|
MOVL 4(SP), AX
|
|
|
|
// MOVQ and EMMS were introduced on the Pentium MMX.
|
|
|
|
// MOVQ 0x8(%ESP), %MM0
|
|
|
|
BYTE $0x0f; BYTE $0x6f; BYTE $0x44; BYTE $0x24; BYTE $0x08
|
|
|
|
// MOVQ %MM0, (%EAX)
|
|
|
|
BYTE $0x0f; BYTE $0x7f; BYTE $0x00
|
|
|
|
// EMMS
|
|
|
|
BYTE $0x0F; BYTE $0x77
|
|
|
|
// This is essentially a no-op, but it provides required memory fencing.
|
|
|
|
// It can be replaced with MFENCE, but MFENCE was introduced only on the Pentium4 (SSE2).
|
|
|
|
MOVL $0, AX
|
|
|
|
LOCK
|
|
|
|
XADDL AX, (SP)
|
|
|
|
RET
|
|
|
|
|
2009-06-03 00:02:12 -06:00
|
|
|
// void jmpdefer(fn, sp);
|
|
|
|
// called from deferreturn.
|
2009-03-30 01:01:07 -06:00
|
|
|
// 1. pop the caller
|
|
|
|
// 2. sub 5 bytes from the callers return
|
|
|
|
// 3. jmp to the argument
|
runtime: ,s/[a-zA-Z0-9_]+/runtime·&/g, almost
Prefix all external symbols in runtime by runtime·,
to avoid conflicts with possible symbols of the same
name in linked-in C libraries. The obvious conflicts
are printf, malloc, and free, but hide everything to
avoid future pain.
The symbols left alone are:
** known to cgo **
_cgo_free
_cgo_malloc
libcgo_thread_start
initcgo
ncgocall
** known to linker **
_rt0_$GOARCH
_rt0_$GOARCH_$GOOS
text
etext
data
end
pclntab
epclntab
symtab
esymtab
** known to C compiler **
_divv
_modv
_div64by32
etc (arch specific)
Tested on darwin/386, darwin/amd64, linux/386, linux/amd64.
Built (but not tested) for freebsd/386, freebsd/amd64, linux/arm, windows/386.
R=r, PeterGo
CC=golang-dev
https://golang.org/cl/2899041
2010-11-04 12:00:19 -06:00
|
|
|
TEXT runtime·jmpdefer(SB), 7, $0
|
2013-02-22 08:47:54 -07:00
|
|
|
MOVL 4(SP), DX // fn
|
2009-06-03 00:02:12 -06:00
|
|
|
MOVL 8(SP), BX // caller sp
|
|
|
|
LEAL -4(BX), SP // caller sp after CALL
|
|
|
|
SUBL $5, (SP) // return to CALL again
|
2013-02-22 08:47:54 -07:00
|
|
|
MOVL 0(DX), BX
|
2013-02-21 15:01:13 -07:00
|
|
|
JMP BX // but first run the deferred function
|
2009-03-30 01:01:07 -06:00
|
|
|
|
2013-06-12 13:22:26 -06:00
|
|
|
// Save state of caller into g->sched.
|
|
|
|
TEXT gosave<>(SB),7,$0
|
|
|
|
PUSHL AX
|
|
|
|
PUSHL BX
|
|
|
|
get_tls(BX)
|
|
|
|
MOVL g(BX), BX
|
|
|
|
LEAL arg+0(FP), AX
|
|
|
|
MOVL AX, (g_sched+gobuf_sp)(BX)
|
|
|
|
MOVL -4(AX), AX
|
|
|
|
MOVL AX, (g_sched+gobuf_pc)(BX)
|
|
|
|
MOVL $0, (g_sched+gobuf_ret)(BX)
|
|
|
|
MOVL $0, (g_sched+gobuf_ctxt)(BX)
|
|
|
|
POPL BX
|
|
|
|
POPL AX
|
runtime: scheduler, cgo reorganization
* Change use of m->g0 stack (aka scheduler stack).
* Provide runtime.mcall(f) to invoke f() on m->g0 stack.
* Replace scheduler loop entry with runtime.mcall(schedule).
Runtime.mcall eliminates the need for fake scheduler states that
exist just to run a bit of code on the m->g0 stack
(Grecovery, Gstackalloc).
The elimination of the scheduler as a loop that stops and
starts using gosave and gogo fixes a bad interaction with the
way cgo uses the m->g0 stack. Cgo runs external (gcc-compiled)
C functions on that stack, and then when calling back into Go,
it sets m->g0->sched.sp below the added call frames, so that
other uses of m->g0's stack will not interfere with those frames.
Unfortunately, gogo (longjmp) back to the scheduler loop at
this point would end up running scheduler with the lower
sp, which no longer points at a valid stack frame for
a call to scheduler. If scheduler then wrote any function call
arguments or local variables to where it expected the stack
frame to be, it would overwrite other data on the stack.
I realized this possibility while debugging a problem with
calling complex Go code in a Go -> C -> Go cgo callback.
This wasn't the bug I was looking for, it turns out, but I believe
it is a real bug nonetheless. Switching to runtime.mcall, which
only adds new frames to the stack and never jumps into
functions running in existing ones, fixes this bug.
* Move cgo-related code out of proc.c into cgocall.c.
* Add very large comment describing cgo call sequences.
* Simpilify, regularize cgo function implementations and names.
* Add test suite as misc/cgo/test.
Now the Go -> C path calls cgocall, which calls asmcgocall,
and the C -> Go path calls cgocallback, which calls cgocallbackg.
The shuffling, which affects mainly the callback case, moves
most of the callback implementation to cgocallback running
on the m->curg stack (not the m->g0 scheduler stack) and
only while accounted for with $GOMAXPROCS (between calls
to exitsyscall and entersyscall).
The previous callback code did not block in startcgocallback's
approximation to exitsyscall, so if, say, the garbage collector
were running, it would still barge in and start doing things
like call malloc. Similarly endcgocallback's approximation of
entersyscall did not call matchmg to kick off new OS threads
when necessary, which caused the bug in issue 1560.
Fixes #1560.
R=iant
CC=golang-dev
https://golang.org/cl/4253054
2011-03-07 08:37:42 -07:00
|
|
|
RET
|
|
|
|
|
|
|
|
// asmcgocall(void(*fn)(void*), void *arg)
|
|
|
|
// Call fn(arg) on the scheduler stack,
|
|
|
|
// aligned appropriately for the gcc ABI.
|
|
|
|
// See cgocall.c for more details.
|
2013-07-16 14:24:09 -06:00
|
|
|
TEXT runtime·asmcgocall(SB),7,$0-8
|
runtime: scheduler, cgo reorganization
* Change use of m->g0 stack (aka scheduler stack).
* Provide runtime.mcall(f) to invoke f() on m->g0 stack.
* Replace scheduler loop entry with runtime.mcall(schedule).
Runtime.mcall eliminates the need for fake scheduler states that
exist just to run a bit of code on the m->g0 stack
(Grecovery, Gstackalloc).
The elimination of the scheduler as a loop that stops and
starts using gosave and gogo fixes a bad interaction with the
way cgo uses the m->g0 stack. Cgo runs external (gcc-compiled)
C functions on that stack, and then when calling back into Go,
it sets m->g0->sched.sp below the added call frames, so that
other uses of m->g0's stack will not interfere with those frames.
Unfortunately, gogo (longjmp) back to the scheduler loop at
this point would end up running scheduler with the lower
sp, which no longer points at a valid stack frame for
a call to scheduler. If scheduler then wrote any function call
arguments or local variables to where it expected the stack
frame to be, it would overwrite other data on the stack.
I realized this possibility while debugging a problem with
calling complex Go code in a Go -> C -> Go cgo callback.
This wasn't the bug I was looking for, it turns out, but I believe
it is a real bug nonetheless. Switching to runtime.mcall, which
only adds new frames to the stack and never jumps into
functions running in existing ones, fixes this bug.
* Move cgo-related code out of proc.c into cgocall.c.
* Add very large comment describing cgo call sequences.
* Simpilify, regularize cgo function implementations and names.
* Add test suite as misc/cgo/test.
Now the Go -> C path calls cgocall, which calls asmcgocall,
and the C -> Go path calls cgocallback, which calls cgocallbackg.
The shuffling, which affects mainly the callback case, moves
most of the callback implementation to cgocallback running
on the m->curg stack (not the m->g0 scheduler stack) and
only while accounted for with $GOMAXPROCS (between calls
to exitsyscall and entersyscall).
The previous callback code did not block in startcgocallback's
approximation to exitsyscall, so if, say, the garbage collector
were running, it would still barge in and start doing things
like call malloc. Similarly endcgocallback's approximation of
entersyscall did not call matchmg to kick off new OS threads
when necessary, which caused the bug in issue 1560.
Fixes #1560.
R=iant
CC=golang-dev
https://golang.org/cl/4253054
2011-03-07 08:37:42 -07:00
|
|
|
MOVL fn+0(FP), AX
|
|
|
|
MOVL arg+4(FP), BX
|
|
|
|
MOVL SP, DX
|
|
|
|
|
|
|
|
// Figure out if we need to switch to m->g0 stack.
|
|
|
|
// We get called to create new OS threads too, and those
|
|
|
|
// come in on the m->g0 stack already.
|
|
|
|
get_tls(CX)
|
|
|
|
MOVL m(CX), BP
|
|
|
|
MOVL m_g0(BP), SI
|
|
|
|
MOVL g(CX), DI
|
|
|
|
CMPL SI, DI
|
2013-06-12 13:22:26 -06:00
|
|
|
JEQ 4(PC)
|
|
|
|
CALL gosave<>(SB)
|
runtime: scheduler, cgo reorganization
* Change use of m->g0 stack (aka scheduler stack).
* Provide runtime.mcall(f) to invoke f() on m->g0 stack.
* Replace scheduler loop entry with runtime.mcall(schedule).
Runtime.mcall eliminates the need for fake scheduler states that
exist just to run a bit of code on the m->g0 stack
(Grecovery, Gstackalloc).
The elimination of the scheduler as a loop that stops and
starts using gosave and gogo fixes a bad interaction with the
way cgo uses the m->g0 stack. Cgo runs external (gcc-compiled)
C functions on that stack, and then when calling back into Go,
it sets m->g0->sched.sp below the added call frames, so that
other uses of m->g0's stack will not interfere with those frames.
Unfortunately, gogo (longjmp) back to the scheduler loop at
this point would end up running scheduler with the lower
sp, which no longer points at a valid stack frame for
a call to scheduler. If scheduler then wrote any function call
arguments or local variables to where it expected the stack
frame to be, it would overwrite other data on the stack.
I realized this possibility while debugging a problem with
calling complex Go code in a Go -> C -> Go cgo callback.
This wasn't the bug I was looking for, it turns out, but I believe
it is a real bug nonetheless. Switching to runtime.mcall, which
only adds new frames to the stack and never jumps into
functions running in existing ones, fixes this bug.
* Move cgo-related code out of proc.c into cgocall.c.
* Add very large comment describing cgo call sequences.
* Simpilify, regularize cgo function implementations and names.
* Add test suite as misc/cgo/test.
Now the Go -> C path calls cgocall, which calls asmcgocall,
and the C -> Go path calls cgocallback, which calls cgocallbackg.
The shuffling, which affects mainly the callback case, moves
most of the callback implementation to cgocallback running
on the m->curg stack (not the m->g0 scheduler stack) and
only while accounted for with $GOMAXPROCS (between calls
to exitsyscall and entersyscall).
The previous callback code did not block in startcgocallback's
approximation to exitsyscall, so if, say, the garbage collector
were running, it would still barge in and start doing things
like call malloc. Similarly endcgocallback's approximation of
entersyscall did not call matchmg to kick off new OS threads
when necessary, which caused the bug in issue 1560.
Fixes #1560.
R=iant
CC=golang-dev
https://golang.org/cl/4253054
2011-03-07 08:37:42 -07:00
|
|
|
MOVL SI, g(CX)
|
|
|
|
MOVL (g_sched+gobuf_sp)(SI), SP
|
|
|
|
|
|
|
|
// Now on a scheduling stack (a pthread-created stack).
|
|
|
|
SUBL $32, SP
|
|
|
|
ANDL $~15, SP // alignment, perhaps unnecessary
|
|
|
|
MOVL DI, 8(SP) // save g
|
|
|
|
MOVL DX, 4(SP) // save SP
|
|
|
|
MOVL BX, 0(SP) // first argument in x86-32 ABI
|
|
|
|
CALL AX
|
|
|
|
|
|
|
|
// Restore registers, g, stack pointer.
|
|
|
|
get_tls(CX)
|
|
|
|
MOVL 8(SP), DI
|
|
|
|
MOVL DI, g(CX)
|
|
|
|
MOVL 4(SP), SP
|
|
|
|
RET
|
|
|
|
|
|
|
|
// cgocallback(void (*fn)(void*), void *frame, uintptr framesize)
|
2013-02-22 14:08:56 -07:00
|
|
|
// Turn the fn into a Go func (by taking its address) and call
|
|
|
|
// cgocallback_gofunc.
|
2013-07-16 14:24:09 -06:00
|
|
|
TEXT runtime·cgocallback(SB),7,$12-12
|
2013-02-22 14:08:56 -07:00
|
|
|
LEAL fn+0(FP), AX
|
|
|
|
MOVL AX, 0(SP)
|
|
|
|
MOVL frame+4(FP), AX
|
|
|
|
MOVL AX, 4(SP)
|
|
|
|
MOVL framesize+8(FP), AX
|
|
|
|
MOVL AX, 8(SP)
|
|
|
|
MOVL $runtime·cgocallback_gofunc(SB), AX
|
|
|
|
CALL AX
|
|
|
|
RET
|
|
|
|
|
|
|
|
// cgocallback_gofunc(FuncVal*, void *frame, uintptr framesize)
|
|
|
|
// See cgocall.c for more details.
|
2013-07-24 07:01:57 -06:00
|
|
|
TEXT runtime·cgocallback_gofunc(SB),7,$12-12
|
2013-02-20 15:48:23 -07:00
|
|
|
// If m is nil, Go did not create the current thread.
|
|
|
|
// Call needm to obtain one for temporary use.
|
|
|
|
// In this case, we're running on the thread stack, so there's
|
|
|
|
// lots of space, but the linker doesn't know. Hide the call from
|
|
|
|
// the linker analysis by using an indirect call through AX.
|
runtime: scheduler, cgo reorganization
* Change use of m->g0 stack (aka scheduler stack).
* Provide runtime.mcall(f) to invoke f() on m->g0 stack.
* Replace scheduler loop entry with runtime.mcall(schedule).
Runtime.mcall eliminates the need for fake scheduler states that
exist just to run a bit of code on the m->g0 stack
(Grecovery, Gstackalloc).
The elimination of the scheduler as a loop that stops and
starts using gosave and gogo fixes a bad interaction with the
way cgo uses the m->g0 stack. Cgo runs external (gcc-compiled)
C functions on that stack, and then when calling back into Go,
it sets m->g0->sched.sp below the added call frames, so that
other uses of m->g0's stack will not interfere with those frames.
Unfortunately, gogo (longjmp) back to the scheduler loop at
this point would end up running scheduler with the lower
sp, which no longer points at a valid stack frame for
a call to scheduler. If scheduler then wrote any function call
arguments or local variables to where it expected the stack
frame to be, it would overwrite other data on the stack.
I realized this possibility while debugging a problem with
calling complex Go code in a Go -> C -> Go cgo callback.
This wasn't the bug I was looking for, it turns out, but I believe
it is a real bug nonetheless. Switching to runtime.mcall, which
only adds new frames to the stack and never jumps into
functions running in existing ones, fixes this bug.
* Move cgo-related code out of proc.c into cgocall.c.
* Add very large comment describing cgo call sequences.
* Simpilify, regularize cgo function implementations and names.
* Add test suite as misc/cgo/test.
Now the Go -> C path calls cgocall, which calls asmcgocall,
and the C -> Go path calls cgocallback, which calls cgocallbackg.
The shuffling, which affects mainly the callback case, moves
most of the callback implementation to cgocallback running
on the m->curg stack (not the m->g0 scheduler stack) and
only while accounted for with $GOMAXPROCS (between calls
to exitsyscall and entersyscall).
The previous callback code did not block in startcgocallback's
approximation to exitsyscall, so if, say, the garbage collector
were running, it would still barge in and start doing things
like call malloc. Similarly endcgocallback's approximation of
entersyscall did not call matchmg to kick off new OS threads
when necessary, which caused the bug in issue 1560.
Fixes #1560.
R=iant
CC=golang-dev
https://golang.org/cl/4253054
2011-03-07 08:37:42 -07:00
|
|
|
get_tls(CX)
|
2013-02-20 15:48:23 -07:00
|
|
|
#ifdef GOOS_windows
|
2013-07-23 16:40:02 -06:00
|
|
|
MOVL $0, BP
|
2013-02-20 15:48:23 -07:00
|
|
|
CMPL CX, $0
|
2013-07-23 20:59:32 -06:00
|
|
|
JEQ 2(PC)
|
2013-02-20 15:48:23 -07:00
|
|
|
#endif
|
runtime: scheduler, cgo reorganization
* Change use of m->g0 stack (aka scheduler stack).
* Provide runtime.mcall(f) to invoke f() on m->g0 stack.
* Replace scheduler loop entry with runtime.mcall(schedule).
Runtime.mcall eliminates the need for fake scheduler states that
exist just to run a bit of code on the m->g0 stack
(Grecovery, Gstackalloc).
The elimination of the scheduler as a loop that stops and
starts using gosave and gogo fixes a bad interaction with the
way cgo uses the m->g0 stack. Cgo runs external (gcc-compiled)
C functions on that stack, and then when calling back into Go,
it sets m->g0->sched.sp below the added call frames, so that
other uses of m->g0's stack will not interfere with those frames.
Unfortunately, gogo (longjmp) back to the scheduler loop at
this point would end up running scheduler with the lower
sp, which no longer points at a valid stack frame for
a call to scheduler. If scheduler then wrote any function call
arguments or local variables to where it expected the stack
frame to be, it would overwrite other data on the stack.
I realized this possibility while debugging a problem with
calling complex Go code in a Go -> C -> Go cgo callback.
This wasn't the bug I was looking for, it turns out, but I believe
it is a real bug nonetheless. Switching to runtime.mcall, which
only adds new frames to the stack and never jumps into
functions running in existing ones, fixes this bug.
* Move cgo-related code out of proc.c into cgocall.c.
* Add very large comment describing cgo call sequences.
* Simpilify, regularize cgo function implementations and names.
* Add test suite as misc/cgo/test.
Now the Go -> C path calls cgocall, which calls asmcgocall,
and the C -> Go path calls cgocallback, which calls cgocallbackg.
The shuffling, which affects mainly the callback case, moves
most of the callback implementation to cgocallback running
on the m->curg stack (not the m->g0 scheduler stack) and
only while accounted for with $GOMAXPROCS (between calls
to exitsyscall and entersyscall).
The previous callback code did not block in startcgocallback's
approximation to exitsyscall, so if, say, the garbage collector
were running, it would still barge in and start doing things
like call malloc. Similarly endcgocallback's approximation of
entersyscall did not call matchmg to kick off new OS threads
when necessary, which caused the bug in issue 1560.
Fixes #1560.
R=iant
CC=golang-dev
https://golang.org/cl/4253054
2011-03-07 08:37:42 -07:00
|
|
|
MOVL m(CX), BP
|
2013-07-24 07:01:57 -06:00
|
|
|
MOVL BP, DX // saved copy of oldm
|
2012-03-08 10:12:40 -07:00
|
|
|
CMPL BP, $0
|
2013-02-20 15:48:23 -07:00
|
|
|
JNE havem
|
|
|
|
needm:
|
2013-07-24 07:01:57 -06:00
|
|
|
MOVL DX, 0(SP)
|
2013-02-20 15:48:23 -07:00
|
|
|
MOVL $runtime·needm(SB), AX
|
|
|
|
CALL AX
|
2013-07-24 07:01:57 -06:00
|
|
|
MOVL 0(SP), DX
|
2013-02-20 15:48:23 -07:00
|
|
|
get_tls(CX)
|
|
|
|
MOVL m(CX), BP
|
2012-03-08 10:12:40 -07:00
|
|
|
|
2013-02-20 15:48:23 -07:00
|
|
|
havem:
|
|
|
|
// Now there's a valid m, and we're running on its m->g0.
|
|
|
|
// Save current m->g0->sched.sp on stack and then set it to SP.
|
|
|
|
// Save current sp in m->g0->sched.sp in preparation for
|
|
|
|
// switch back to m->curg stack.
|
2013-07-23 16:40:02 -06:00
|
|
|
// NOTE: unwindm knows that the saved g->sched.sp is at 0(SP).
|
2013-07-24 07:01:57 -06:00
|
|
|
// On Windows, the SEH is at 4(SP) and 8(SP).
|
runtime: scheduler, cgo reorganization
* Change use of m->g0 stack (aka scheduler stack).
* Provide runtime.mcall(f) to invoke f() on m->g0 stack.
* Replace scheduler loop entry with runtime.mcall(schedule).
Runtime.mcall eliminates the need for fake scheduler states that
exist just to run a bit of code on the m->g0 stack
(Grecovery, Gstackalloc).
The elimination of the scheduler as a loop that stops and
starts using gosave and gogo fixes a bad interaction with the
way cgo uses the m->g0 stack. Cgo runs external (gcc-compiled)
C functions on that stack, and then when calling back into Go,
it sets m->g0->sched.sp below the added call frames, so that
other uses of m->g0's stack will not interfere with those frames.
Unfortunately, gogo (longjmp) back to the scheduler loop at
this point would end up running scheduler with the lower
sp, which no longer points at a valid stack frame for
a call to scheduler. If scheduler then wrote any function call
arguments or local variables to where it expected the stack
frame to be, it would overwrite other data on the stack.
I realized this possibility while debugging a problem with
calling complex Go code in a Go -> C -> Go cgo callback.
This wasn't the bug I was looking for, it turns out, but I believe
it is a real bug nonetheless. Switching to runtime.mcall, which
only adds new frames to the stack and never jumps into
functions running in existing ones, fixes this bug.
* Move cgo-related code out of proc.c into cgocall.c.
* Add very large comment describing cgo call sequences.
* Simpilify, regularize cgo function implementations and names.
* Add test suite as misc/cgo/test.
Now the Go -> C path calls cgocall, which calls asmcgocall,
and the C -> Go path calls cgocallback, which calls cgocallbackg.
The shuffling, which affects mainly the callback case, moves
most of the callback implementation to cgocallback running
on the m->curg stack (not the m->g0 scheduler stack) and
only while accounted for with $GOMAXPROCS (between calls
to exitsyscall and entersyscall).
The previous callback code did not block in startcgocallback's
approximation to exitsyscall, so if, say, the garbage collector
were running, it would still barge in and start doing things
like call malloc. Similarly endcgocallback's approximation of
entersyscall did not call matchmg to kick off new OS threads
when necessary, which caused the bug in issue 1560.
Fixes #1560.
R=iant
CC=golang-dev
https://golang.org/cl/4253054
2011-03-07 08:37:42 -07:00
|
|
|
MOVL m_g0(BP), SI
|
2013-07-23 16:40:02 -06:00
|
|
|
MOVL (g_sched+gobuf_sp)(SI), AX
|
|
|
|
MOVL AX, 0(SP)
|
runtime: scheduler, cgo reorganization
* Change use of m->g0 stack (aka scheduler stack).
* Provide runtime.mcall(f) to invoke f() on m->g0 stack.
* Replace scheduler loop entry with runtime.mcall(schedule).
Runtime.mcall eliminates the need for fake scheduler states that
exist just to run a bit of code on the m->g0 stack
(Grecovery, Gstackalloc).
The elimination of the scheduler as a loop that stops and
starts using gosave and gogo fixes a bad interaction with the
way cgo uses the m->g0 stack. Cgo runs external (gcc-compiled)
C functions on that stack, and then when calling back into Go,
it sets m->g0->sched.sp below the added call frames, so that
other uses of m->g0's stack will not interfere with those frames.
Unfortunately, gogo (longjmp) back to the scheduler loop at
this point would end up running scheduler with the lower
sp, which no longer points at a valid stack frame for
a call to scheduler. If scheduler then wrote any function call
arguments or local variables to where it expected the stack
frame to be, it would overwrite other data on the stack.
I realized this possibility while debugging a problem with
calling complex Go code in a Go -> C -> Go cgo callback.
This wasn't the bug I was looking for, it turns out, but I believe
it is a real bug nonetheless. Switching to runtime.mcall, which
only adds new frames to the stack and never jumps into
functions running in existing ones, fixes this bug.
* Move cgo-related code out of proc.c into cgocall.c.
* Add very large comment describing cgo call sequences.
* Simpilify, regularize cgo function implementations and names.
* Add test suite as misc/cgo/test.
Now the Go -> C path calls cgocall, which calls asmcgocall,
and the C -> Go path calls cgocallback, which calls cgocallbackg.
The shuffling, which affects mainly the callback case, moves
most of the callback implementation to cgocallback running
on the m->curg stack (not the m->g0 scheduler stack) and
only while accounted for with $GOMAXPROCS (between calls
to exitsyscall and entersyscall).
The previous callback code did not block in startcgocallback's
approximation to exitsyscall, so if, say, the garbage collector
were running, it would still barge in and start doing things
like call malloc. Similarly endcgocallback's approximation of
entersyscall did not call matchmg to kick off new OS threads
when necessary, which caused the bug in issue 1560.
Fixes #1560.
R=iant
CC=golang-dev
https://golang.org/cl/4253054
2011-03-07 08:37:42 -07:00
|
|
|
MOVL SP, (g_sched+gobuf_sp)(SI)
|
|
|
|
|
2013-07-23 16:40:02 -06:00
|
|
|
// Switch to m->curg stack and call runtime.cgocallbackg.
|
|
|
|
// Because we are taking over the execution of m->curg
|
|
|
|
// but *not* resuming what had been running, we need to
|
|
|
|
// save that information (m->curg->sched) so we can restore it.
|
2013-06-05 05:16:53 -06:00
|
|
|
// We can restore m->curg->sched.sp easily, because calling
|
2011-08-18 10:17:09 -06:00
|
|
|
// runtime.cgocallbackg leaves SP unchanged upon return.
|
2013-06-05 05:16:53 -06:00
|
|
|
// To save m->curg->sched.pc, we push it onto the stack.
|
runtime: scheduler, cgo reorganization
* Change use of m->g0 stack (aka scheduler stack).
* Provide runtime.mcall(f) to invoke f() on m->g0 stack.
* Replace scheduler loop entry with runtime.mcall(schedule).
Runtime.mcall eliminates the need for fake scheduler states that
exist just to run a bit of code on the m->g0 stack
(Grecovery, Gstackalloc).
The elimination of the scheduler as a loop that stops and
starts using gosave and gogo fixes a bad interaction with the
way cgo uses the m->g0 stack. Cgo runs external (gcc-compiled)
C functions on that stack, and then when calling back into Go,
it sets m->g0->sched.sp below the added call frames, so that
other uses of m->g0's stack will not interfere with those frames.
Unfortunately, gogo (longjmp) back to the scheduler loop at
this point would end up running scheduler with the lower
sp, which no longer points at a valid stack frame for
a call to scheduler. If scheduler then wrote any function call
arguments or local variables to where it expected the stack
frame to be, it would overwrite other data on the stack.
I realized this possibility while debugging a problem with
calling complex Go code in a Go -> C -> Go cgo callback.
This wasn't the bug I was looking for, it turns out, but I believe
it is a real bug nonetheless. Switching to runtime.mcall, which
only adds new frames to the stack and never jumps into
functions running in existing ones, fixes this bug.
* Move cgo-related code out of proc.c into cgocall.c.
* Add very large comment describing cgo call sequences.
* Simpilify, regularize cgo function implementations and names.
* Add test suite as misc/cgo/test.
Now the Go -> C path calls cgocall, which calls asmcgocall,
and the C -> Go path calls cgocallback, which calls cgocallbackg.
The shuffling, which affects mainly the callback case, moves
most of the callback implementation to cgocallback running
on the m->curg stack (not the m->g0 scheduler stack) and
only while accounted for with $GOMAXPROCS (between calls
to exitsyscall and entersyscall).
The previous callback code did not block in startcgocallback's
approximation to exitsyscall, so if, say, the garbage collector
were running, it would still barge in and start doing things
like call malloc. Similarly endcgocallback's approximation of
entersyscall did not call matchmg to kick off new OS threads
when necessary, which caused the bug in issue 1560.
Fixes #1560.
R=iant
CC=golang-dev
https://golang.org/cl/4253054
2011-03-07 08:37:42 -07:00
|
|
|
// This has the added benefit that it looks to the traceback
|
2011-08-18 10:17:09 -06:00
|
|
|
// routine like cgocallbackg is going to return to that
|
2013-07-23 16:40:02 -06:00
|
|
|
// PC (because the frame we allocate below has the same
|
|
|
|
// size as cgocallback_gofunc's frame declared above)
|
runtime: scheduler, cgo reorganization
* Change use of m->g0 stack (aka scheduler stack).
* Provide runtime.mcall(f) to invoke f() on m->g0 stack.
* Replace scheduler loop entry with runtime.mcall(schedule).
Runtime.mcall eliminates the need for fake scheduler states that
exist just to run a bit of code on the m->g0 stack
(Grecovery, Gstackalloc).
The elimination of the scheduler as a loop that stops and
starts using gosave and gogo fixes a bad interaction with the
way cgo uses the m->g0 stack. Cgo runs external (gcc-compiled)
C functions on that stack, and then when calling back into Go,
it sets m->g0->sched.sp below the added call frames, so that
other uses of m->g0's stack will not interfere with those frames.
Unfortunately, gogo (longjmp) back to the scheduler loop at
this point would end up running scheduler with the lower
sp, which no longer points at a valid stack frame for
a call to scheduler. If scheduler then wrote any function call
arguments or local variables to where it expected the stack
frame to be, it would overwrite other data on the stack.
I realized this possibility while debugging a problem with
calling complex Go code in a Go -> C -> Go cgo callback.
This wasn't the bug I was looking for, it turns out, but I believe
it is a real bug nonetheless. Switching to runtime.mcall, which
only adds new frames to the stack and never jumps into
functions running in existing ones, fixes this bug.
* Move cgo-related code out of proc.c into cgocall.c.
* Add very large comment describing cgo call sequences.
* Simpilify, regularize cgo function implementations and names.
* Add test suite as misc/cgo/test.
Now the Go -> C path calls cgocall, which calls asmcgocall,
and the C -> Go path calls cgocallback, which calls cgocallbackg.
The shuffling, which affects mainly the callback case, moves
most of the callback implementation to cgocallback running
on the m->curg stack (not the m->g0 scheduler stack) and
only while accounted for with $GOMAXPROCS (between calls
to exitsyscall and entersyscall).
The previous callback code did not block in startcgocallback's
approximation to exitsyscall, so if, say, the garbage collector
were running, it would still barge in and start doing things
like call malloc. Similarly endcgocallback's approximation of
entersyscall did not call matchmg to kick off new OS threads
when necessary, which caused the bug in issue 1560.
Fixes #1560.
R=iant
CC=golang-dev
https://golang.org/cl/4253054
2011-03-07 08:37:42 -07:00
|
|
|
// so that the traceback will seamlessly trace back into
|
|
|
|
// the earlier calls.
|
2013-07-23 16:40:02 -06:00
|
|
|
//
|
2013-07-24 07:01:57 -06:00
|
|
|
// In the new goroutine, 0(SP) holds the saved oldm (DX) register.
|
|
|
|
// 4(SP) and 8(SP) are unused.
|
runtime: scheduler, cgo reorganization
* Change use of m->g0 stack (aka scheduler stack).
* Provide runtime.mcall(f) to invoke f() on m->g0 stack.
* Replace scheduler loop entry with runtime.mcall(schedule).
Runtime.mcall eliminates the need for fake scheduler states that
exist just to run a bit of code on the m->g0 stack
(Grecovery, Gstackalloc).
The elimination of the scheduler as a loop that stops and
starts using gosave and gogo fixes a bad interaction with the
way cgo uses the m->g0 stack. Cgo runs external (gcc-compiled)
C functions on that stack, and then when calling back into Go,
it sets m->g0->sched.sp below the added call frames, so that
other uses of m->g0's stack will not interfere with those frames.
Unfortunately, gogo (longjmp) back to the scheduler loop at
this point would end up running scheduler with the lower
sp, which no longer points at a valid stack frame for
a call to scheduler. If scheduler then wrote any function call
arguments or local variables to where it expected the stack
frame to be, it would overwrite other data on the stack.
I realized this possibility while debugging a problem with
calling complex Go code in a Go -> C -> Go cgo callback.
This wasn't the bug I was looking for, it turns out, but I believe
it is a real bug nonetheless. Switching to runtime.mcall, which
only adds new frames to the stack and never jumps into
functions running in existing ones, fixes this bug.
* Move cgo-related code out of proc.c into cgocall.c.
* Add very large comment describing cgo call sequences.
* Simpilify, regularize cgo function implementations and names.
* Add test suite as misc/cgo/test.
Now the Go -> C path calls cgocall, which calls asmcgocall,
and the C -> Go path calls cgocallback, which calls cgocallbackg.
The shuffling, which affects mainly the callback case, moves
most of the callback implementation to cgocallback running
on the m->curg stack (not the m->g0 scheduler stack) and
only while accounted for with $GOMAXPROCS (between calls
to exitsyscall and entersyscall).
The previous callback code did not block in startcgocallback's
approximation to exitsyscall, so if, say, the garbage collector
were running, it would still barge in and start doing things
like call malloc. Similarly endcgocallback's approximation of
entersyscall did not call matchmg to kick off new OS threads
when necessary, which caused the bug in issue 1560.
Fixes #1560.
R=iant
CC=golang-dev
https://golang.org/cl/4253054
2011-03-07 08:37:42 -07:00
|
|
|
MOVL m_curg(BP), SI
|
|
|
|
MOVL SI, g(CX)
|
2013-07-23 16:40:02 -06:00
|
|
|
MOVL (g_sched+gobuf_sp)(SI), DI // prepare stack as DI
|
runtime: scheduler, cgo reorganization
* Change use of m->g0 stack (aka scheduler stack).
* Provide runtime.mcall(f) to invoke f() on m->g0 stack.
* Replace scheduler loop entry with runtime.mcall(schedule).
Runtime.mcall eliminates the need for fake scheduler states that
exist just to run a bit of code on the m->g0 stack
(Grecovery, Gstackalloc).
The elimination of the scheduler as a loop that stops and
starts using gosave and gogo fixes a bad interaction with the
way cgo uses the m->g0 stack. Cgo runs external (gcc-compiled)
C functions on that stack, and then when calling back into Go,
it sets m->g0->sched.sp below the added call frames, so that
other uses of m->g0's stack will not interfere with those frames.
Unfortunately, gogo (longjmp) back to the scheduler loop at
this point would end up running scheduler with the lower
sp, which no longer points at a valid stack frame for
a call to scheduler. If scheduler then wrote any function call
arguments or local variables to where it expected the stack
frame to be, it would overwrite other data on the stack.
I realized this possibility while debugging a problem with
calling complex Go code in a Go -> C -> Go cgo callback.
This wasn't the bug I was looking for, it turns out, but I believe
it is a real bug nonetheless. Switching to runtime.mcall, which
only adds new frames to the stack and never jumps into
functions running in existing ones, fixes this bug.
* Move cgo-related code out of proc.c into cgocall.c.
* Add very large comment describing cgo call sequences.
* Simpilify, regularize cgo function implementations and names.
* Add test suite as misc/cgo/test.
Now the Go -> C path calls cgocall, which calls asmcgocall,
and the C -> Go path calls cgocallback, which calls cgocallbackg.
The shuffling, which affects mainly the callback case, moves
most of the callback implementation to cgocallback running
on the m->curg stack (not the m->g0 scheduler stack) and
only while accounted for with $GOMAXPROCS (between calls
to exitsyscall and entersyscall).
The previous callback code did not block in startcgocallback's
approximation to exitsyscall, so if, say, the garbage collector
were running, it would still barge in and start doing things
like call malloc. Similarly endcgocallback's approximation of
entersyscall did not call matchmg to kick off new OS threads
when necessary, which caused the bug in issue 1560.
Fixes #1560.
R=iant
CC=golang-dev
https://golang.org/cl/4253054
2011-03-07 08:37:42 -07:00
|
|
|
MOVL (g_sched+gobuf_pc)(SI), BP
|
2013-07-23 16:40:02 -06:00
|
|
|
MOVL BP, -4(DI)
|
2013-07-24 07:01:57 -06:00
|
|
|
LEAL -(4+12)(DI), SP
|
|
|
|
MOVL DX, 0(SP)
|
runtime: scheduler, cgo reorganization
* Change use of m->g0 stack (aka scheduler stack).
* Provide runtime.mcall(f) to invoke f() on m->g0 stack.
* Replace scheduler loop entry with runtime.mcall(schedule).
Runtime.mcall eliminates the need for fake scheduler states that
exist just to run a bit of code on the m->g0 stack
(Grecovery, Gstackalloc).
The elimination of the scheduler as a loop that stops and
starts using gosave and gogo fixes a bad interaction with the
way cgo uses the m->g0 stack. Cgo runs external (gcc-compiled)
C functions on that stack, and then when calling back into Go,
it sets m->g0->sched.sp below the added call frames, so that
other uses of m->g0's stack will not interfere with those frames.
Unfortunately, gogo (longjmp) back to the scheduler loop at
this point would end up running scheduler with the lower
sp, which no longer points at a valid stack frame for
a call to scheduler. If scheduler then wrote any function call
arguments or local variables to where it expected the stack
frame to be, it would overwrite other data on the stack.
I realized this possibility while debugging a problem with
calling complex Go code in a Go -> C -> Go cgo callback.
This wasn't the bug I was looking for, it turns out, but I believe
it is a real bug nonetheless. Switching to runtime.mcall, which
only adds new frames to the stack and never jumps into
functions running in existing ones, fixes this bug.
* Move cgo-related code out of proc.c into cgocall.c.
* Add very large comment describing cgo call sequences.
* Simpilify, regularize cgo function implementations and names.
* Add test suite as misc/cgo/test.
Now the Go -> C path calls cgocall, which calls asmcgocall,
and the C -> Go path calls cgocallback, which calls cgocallbackg.
The shuffling, which affects mainly the callback case, moves
most of the callback implementation to cgocallback running
on the m->curg stack (not the m->g0 scheduler stack) and
only while accounted for with $GOMAXPROCS (between calls
to exitsyscall and entersyscall).
The previous callback code did not block in startcgocallback's
approximation to exitsyscall, so if, say, the garbage collector
were running, it would still barge in and start doing things
like call malloc. Similarly endcgocallback's approximation of
entersyscall did not call matchmg to kick off new OS threads
when necessary, which caused the bug in issue 1560.
Fixes #1560.
R=iant
CC=golang-dev
https://golang.org/cl/4253054
2011-03-07 08:37:42 -07:00
|
|
|
CALL runtime·cgocallbackg(SB)
|
2013-07-24 07:01:57 -06:00
|
|
|
MOVL 0(SP), DX
|
runtime: scheduler, cgo reorganization
* Change use of m->g0 stack (aka scheduler stack).
* Provide runtime.mcall(f) to invoke f() on m->g0 stack.
* Replace scheduler loop entry with runtime.mcall(schedule).
Runtime.mcall eliminates the need for fake scheduler states that
exist just to run a bit of code on the m->g0 stack
(Grecovery, Gstackalloc).
The elimination of the scheduler as a loop that stops and
starts using gosave and gogo fixes a bad interaction with the
way cgo uses the m->g0 stack. Cgo runs external (gcc-compiled)
C functions on that stack, and then when calling back into Go,
it sets m->g0->sched.sp below the added call frames, so that
other uses of m->g0's stack will not interfere with those frames.
Unfortunately, gogo (longjmp) back to the scheduler loop at
this point would end up running scheduler with the lower
sp, which no longer points at a valid stack frame for
a call to scheduler. If scheduler then wrote any function call
arguments or local variables to where it expected the stack
frame to be, it would overwrite other data on the stack.
I realized this possibility while debugging a problem with
calling complex Go code in a Go -> C -> Go cgo callback.
This wasn't the bug I was looking for, it turns out, but I believe
it is a real bug nonetheless. Switching to runtime.mcall, which
only adds new frames to the stack and never jumps into
functions running in existing ones, fixes this bug.
* Move cgo-related code out of proc.c into cgocall.c.
* Add very large comment describing cgo call sequences.
* Simpilify, regularize cgo function implementations and names.
* Add test suite as misc/cgo/test.
Now the Go -> C path calls cgocall, which calls asmcgocall,
and the C -> Go path calls cgocallback, which calls cgocallbackg.
The shuffling, which affects mainly the callback case, moves
most of the callback implementation to cgocallback running
on the m->curg stack (not the m->g0 scheduler stack) and
only while accounted for with $GOMAXPROCS (between calls
to exitsyscall and entersyscall).
The previous callback code did not block in startcgocallback's
approximation to exitsyscall, so if, say, the garbage collector
were running, it would still barge in and start doing things
like call malloc. Similarly endcgocallback's approximation of
entersyscall did not call matchmg to kick off new OS threads
when necessary, which caused the bug in issue 1560.
Fixes #1560.
R=iant
CC=golang-dev
https://golang.org/cl/4253054
2011-03-07 08:37:42 -07:00
|
|
|
|
2013-06-05 05:16:53 -06:00
|
|
|
// Restore g->sched (== m->curg->sched) from saved values.
|
runtime: scheduler, cgo reorganization
* Change use of m->g0 stack (aka scheduler stack).
* Provide runtime.mcall(f) to invoke f() on m->g0 stack.
* Replace scheduler loop entry with runtime.mcall(schedule).
Runtime.mcall eliminates the need for fake scheduler states that
exist just to run a bit of code on the m->g0 stack
(Grecovery, Gstackalloc).
The elimination of the scheduler as a loop that stops and
starts using gosave and gogo fixes a bad interaction with the
way cgo uses the m->g0 stack. Cgo runs external (gcc-compiled)
C functions on that stack, and then when calling back into Go,
it sets m->g0->sched.sp below the added call frames, so that
other uses of m->g0's stack will not interfere with those frames.
Unfortunately, gogo (longjmp) back to the scheduler loop at
this point would end up running scheduler with the lower
sp, which no longer points at a valid stack frame for
a call to scheduler. If scheduler then wrote any function call
arguments or local variables to where it expected the stack
frame to be, it would overwrite other data on the stack.
I realized this possibility while debugging a problem with
calling complex Go code in a Go -> C -> Go cgo callback.
This wasn't the bug I was looking for, it turns out, but I believe
it is a real bug nonetheless. Switching to runtime.mcall, which
only adds new frames to the stack and never jumps into
functions running in existing ones, fixes this bug.
* Move cgo-related code out of proc.c into cgocall.c.
* Add very large comment describing cgo call sequences.
* Simpilify, regularize cgo function implementations and names.
* Add test suite as misc/cgo/test.
Now the Go -> C path calls cgocall, which calls asmcgocall,
and the C -> Go path calls cgocallback, which calls cgocallbackg.
The shuffling, which affects mainly the callback case, moves
most of the callback implementation to cgocallback running
on the m->curg stack (not the m->g0 scheduler stack) and
only while accounted for with $GOMAXPROCS (between calls
to exitsyscall and entersyscall).
The previous callback code did not block in startcgocallback's
approximation to exitsyscall, so if, say, the garbage collector
were running, it would still barge in and start doing things
like call malloc. Similarly endcgocallback's approximation of
entersyscall did not call matchmg to kick off new OS threads
when necessary, which caused the bug in issue 1560.
Fixes #1560.
R=iant
CC=golang-dev
https://golang.org/cl/4253054
2011-03-07 08:37:42 -07:00
|
|
|
get_tls(CX)
|
|
|
|
MOVL g(CX), SI
|
2013-07-24 07:01:57 -06:00
|
|
|
MOVL 12(SP), BP
|
runtime: scheduler, cgo reorganization
* Change use of m->g0 stack (aka scheduler stack).
* Provide runtime.mcall(f) to invoke f() on m->g0 stack.
* Replace scheduler loop entry with runtime.mcall(schedule).
Runtime.mcall eliminates the need for fake scheduler states that
exist just to run a bit of code on the m->g0 stack
(Grecovery, Gstackalloc).
The elimination of the scheduler as a loop that stops and
starts using gosave and gogo fixes a bad interaction with the
way cgo uses the m->g0 stack. Cgo runs external (gcc-compiled)
C functions on that stack, and then when calling back into Go,
it sets m->g0->sched.sp below the added call frames, so that
other uses of m->g0's stack will not interfere with those frames.
Unfortunately, gogo (longjmp) back to the scheduler loop at
this point would end up running scheduler with the lower
sp, which no longer points at a valid stack frame for
a call to scheduler. If scheduler then wrote any function call
arguments or local variables to where it expected the stack
frame to be, it would overwrite other data on the stack.
I realized this possibility while debugging a problem with
calling complex Go code in a Go -> C -> Go cgo callback.
This wasn't the bug I was looking for, it turns out, but I believe
it is a real bug nonetheless. Switching to runtime.mcall, which
only adds new frames to the stack and never jumps into
functions running in existing ones, fixes this bug.
* Move cgo-related code out of proc.c into cgocall.c.
* Add very large comment describing cgo call sequences.
* Simpilify, regularize cgo function implementations and names.
* Add test suite as misc/cgo/test.
Now the Go -> C path calls cgocall, which calls asmcgocall,
and the C -> Go path calls cgocallback, which calls cgocallbackg.
The shuffling, which affects mainly the callback case, moves
most of the callback implementation to cgocallback running
on the m->curg stack (not the m->g0 scheduler stack) and
only while accounted for with $GOMAXPROCS (between calls
to exitsyscall and entersyscall).
The previous callback code did not block in startcgocallback's
approximation to exitsyscall, so if, say, the garbage collector
were running, it would still barge in and start doing things
like call malloc. Similarly endcgocallback's approximation of
entersyscall did not call matchmg to kick off new OS threads
when necessary, which caused the bug in issue 1560.
Fixes #1560.
R=iant
CC=golang-dev
https://golang.org/cl/4253054
2011-03-07 08:37:42 -07:00
|
|
|
MOVL BP, (g_sched+gobuf_pc)(SI)
|
2013-07-24 07:01:57 -06:00
|
|
|
LEAL (12+4)(SP), DI
|
runtime: scheduler, cgo reorganization
* Change use of m->g0 stack (aka scheduler stack).
* Provide runtime.mcall(f) to invoke f() on m->g0 stack.
* Replace scheduler loop entry with runtime.mcall(schedule).
Runtime.mcall eliminates the need for fake scheduler states that
exist just to run a bit of code on the m->g0 stack
(Grecovery, Gstackalloc).
The elimination of the scheduler as a loop that stops and
starts using gosave and gogo fixes a bad interaction with the
way cgo uses the m->g0 stack. Cgo runs external (gcc-compiled)
C functions on that stack, and then when calling back into Go,
it sets m->g0->sched.sp below the added call frames, so that
other uses of m->g0's stack will not interfere with those frames.
Unfortunately, gogo (longjmp) back to the scheduler loop at
this point would end up running scheduler with the lower
sp, which no longer points at a valid stack frame for
a call to scheduler. If scheduler then wrote any function call
arguments or local variables to where it expected the stack
frame to be, it would overwrite other data on the stack.
I realized this possibility while debugging a problem with
calling complex Go code in a Go -> C -> Go cgo callback.
This wasn't the bug I was looking for, it turns out, but I believe
it is a real bug nonetheless. Switching to runtime.mcall, which
only adds new frames to the stack and never jumps into
functions running in existing ones, fixes this bug.
* Move cgo-related code out of proc.c into cgocall.c.
* Add very large comment describing cgo call sequences.
* Simpilify, regularize cgo function implementations and names.
* Add test suite as misc/cgo/test.
Now the Go -> C path calls cgocall, which calls asmcgocall,
and the C -> Go path calls cgocallback, which calls cgocallbackg.
The shuffling, which affects mainly the callback case, moves
most of the callback implementation to cgocallback running
on the m->curg stack (not the m->g0 scheduler stack) and
only while accounted for with $GOMAXPROCS (between calls
to exitsyscall and entersyscall).
The previous callback code did not block in startcgocallback's
approximation to exitsyscall, so if, say, the garbage collector
were running, it would still barge in and start doing things
like call malloc. Similarly endcgocallback's approximation of
entersyscall did not call matchmg to kick off new OS threads
when necessary, which caused the bug in issue 1560.
Fixes #1560.
R=iant
CC=golang-dev
https://golang.org/cl/4253054
2011-03-07 08:37:42 -07:00
|
|
|
MOVL DI, (g_sched+gobuf_sp)(SI)
|
|
|
|
|
|
|
|
// Switch back to m->g0's stack and restore m->g0->sched.sp.
|
|
|
|
// (Unlike m->curg, the g0 goroutine never uses sched.pc,
|
|
|
|
// so we do not have to restore it.)
|
|
|
|
MOVL m(CX), BP
|
|
|
|
MOVL m_g0(BP), SI
|
|
|
|
MOVL SI, g(CX)
|
|
|
|
MOVL (g_sched+gobuf_sp)(SI), SP
|
2013-07-23 16:40:02 -06:00
|
|
|
MOVL 0(SP), AX
|
|
|
|
MOVL AX, (g_sched+gobuf_sp)(SI)
|
2013-02-20 15:48:23 -07:00
|
|
|
|
|
|
|
// If the m on entry was nil, we called needm above to borrow an m
|
|
|
|
// for the duration of the call. Since the call is over, return it with dropm.
|
2013-07-24 07:01:57 -06:00
|
|
|
CMPL DX, $0
|
2013-02-20 15:48:23 -07:00
|
|
|
JNE 3(PC)
|
|
|
|
MOVL $runtime·dropm(SB), AX
|
|
|
|
CALL AX
|
runtime: scheduler, cgo reorganization
* Change use of m->g0 stack (aka scheduler stack).
* Provide runtime.mcall(f) to invoke f() on m->g0 stack.
* Replace scheduler loop entry with runtime.mcall(schedule).
Runtime.mcall eliminates the need for fake scheduler states that
exist just to run a bit of code on the m->g0 stack
(Grecovery, Gstackalloc).
The elimination of the scheduler as a loop that stops and
starts using gosave and gogo fixes a bad interaction with the
way cgo uses the m->g0 stack. Cgo runs external (gcc-compiled)
C functions on that stack, and then when calling back into Go,
it sets m->g0->sched.sp below the added call frames, so that
other uses of m->g0's stack will not interfere with those frames.
Unfortunately, gogo (longjmp) back to the scheduler loop at
this point would end up running scheduler with the lower
sp, which no longer points at a valid stack frame for
a call to scheduler. If scheduler then wrote any function call
arguments or local variables to where it expected the stack
frame to be, it would overwrite other data on the stack.
I realized this possibility while debugging a problem with
calling complex Go code in a Go -> C -> Go cgo callback.
This wasn't the bug I was looking for, it turns out, but I believe
it is a real bug nonetheless. Switching to runtime.mcall, which
only adds new frames to the stack and never jumps into
functions running in existing ones, fixes this bug.
* Move cgo-related code out of proc.c into cgocall.c.
* Add very large comment describing cgo call sequences.
* Simpilify, regularize cgo function implementations and names.
* Add test suite as misc/cgo/test.
Now the Go -> C path calls cgocall, which calls asmcgocall,
and the C -> Go path calls cgocallback, which calls cgocallbackg.
The shuffling, which affects mainly the callback case, moves
most of the callback implementation to cgocallback running
on the m->curg stack (not the m->g0 scheduler stack) and
only while accounted for with $GOMAXPROCS (between calls
to exitsyscall and entersyscall).
The previous callback code did not block in startcgocallback's
approximation to exitsyscall, so if, say, the garbage collector
were running, it would still barge in and start doing things
like call malloc. Similarly endcgocallback's approximation of
entersyscall did not call matchmg to kick off new OS threads
when necessary, which caused the bug in issue 1560.
Fixes #1560.
R=iant
CC=golang-dev
https://golang.org/cl/4253054
2011-03-07 08:37:42 -07:00
|
|
|
|
|
|
|
// Done!
|
|
|
|
RET
|
|
|
|
|
2013-02-20 15:48:23 -07:00
|
|
|
// void setmg(M*, G*); set m and g. for use by needm.
|
2013-07-16 14:24:09 -06:00
|
|
|
TEXT runtime·setmg(SB), 7, $0-8
|
2013-02-20 15:48:23 -07:00
|
|
|
#ifdef GOOS_windows
|
|
|
|
MOVL mm+0(FP), AX
|
|
|
|
CMPL AX, $0
|
|
|
|
JNE settls
|
|
|
|
MOVL $0, 0x14(FS)
|
|
|
|
RET
|
|
|
|
settls:
|
|
|
|
LEAL m_tls(AX), AX
|
|
|
|
MOVL AX, 0x14(FS)
|
|
|
|
#endif
|
|
|
|
MOVL mm+0(FP), AX
|
|
|
|
get_tls(CX)
|
|
|
|
MOVL mm+0(FP), AX
|
|
|
|
MOVL AX, m(CX)
|
|
|
|
MOVL gg+4(FP), BX
|
|
|
|
MOVL BX, g(CX)
|
|
|
|
RET
|
|
|
|
|
2013-03-25 16:14:02 -06:00
|
|
|
// void setmg_gcc(M*, G*); set m and g. for use by gcc
|
2013-07-16 14:24:09 -06:00
|
|
|
TEXT setmg_gcc<>(SB), 7, $0
|
2013-03-25 16:14:02 -06:00
|
|
|
get_tls(AX)
|
|
|
|
MOVL mm+0(FP), DX
|
|
|
|
MOVL DX, m(AX)
|
|
|
|
MOVL gg+4(FP), DX
|
|
|
|
MOVL DX,g (AX)
|
|
|
|
RET
|
|
|
|
|
runtime: scheduler, cgo reorganization
* Change use of m->g0 stack (aka scheduler stack).
* Provide runtime.mcall(f) to invoke f() on m->g0 stack.
* Replace scheduler loop entry with runtime.mcall(schedule).
Runtime.mcall eliminates the need for fake scheduler states that
exist just to run a bit of code on the m->g0 stack
(Grecovery, Gstackalloc).
The elimination of the scheduler as a loop that stops and
starts using gosave and gogo fixes a bad interaction with the
way cgo uses the m->g0 stack. Cgo runs external (gcc-compiled)
C functions on that stack, and then when calling back into Go,
it sets m->g0->sched.sp below the added call frames, so that
other uses of m->g0's stack will not interfere with those frames.
Unfortunately, gogo (longjmp) back to the scheduler loop at
this point would end up running scheduler with the lower
sp, which no longer points at a valid stack frame for
a call to scheduler. If scheduler then wrote any function call
arguments or local variables to where it expected the stack
frame to be, it would overwrite other data on the stack.
I realized this possibility while debugging a problem with
calling complex Go code in a Go -> C -> Go cgo callback.
This wasn't the bug I was looking for, it turns out, but I believe
it is a real bug nonetheless. Switching to runtime.mcall, which
only adds new frames to the stack and never jumps into
functions running in existing ones, fixes this bug.
* Move cgo-related code out of proc.c into cgocall.c.
* Add very large comment describing cgo call sequences.
* Simpilify, regularize cgo function implementations and names.
* Add test suite as misc/cgo/test.
Now the Go -> C path calls cgocall, which calls asmcgocall,
and the C -> Go path calls cgocallback, which calls cgocallbackg.
The shuffling, which affects mainly the callback case, moves
most of the callback implementation to cgocallback running
on the m->curg stack (not the m->g0 scheduler stack) and
only while accounted for with $GOMAXPROCS (between calls
to exitsyscall and entersyscall).
The previous callback code did not block in startcgocallback's
approximation to exitsyscall, so if, say, the garbage collector
were running, it would still barge in and start doing things
like call malloc. Similarly endcgocallback's approximation of
entersyscall did not call matchmg to kick off new OS threads
when necessary, which caused the bug in issue 1560.
Fixes #1560.
R=iant
CC=golang-dev
https://golang.org/cl/4253054
2011-03-07 08:37:42 -07:00
|
|
|
// check that SP is in range [g->stackbase, g->stackguard)
|
2013-07-16 14:24:09 -06:00
|
|
|
TEXT runtime·stackcheck(SB), 7, $0-0
|
runtime: scheduler, cgo reorganization
* Change use of m->g0 stack (aka scheduler stack).
* Provide runtime.mcall(f) to invoke f() on m->g0 stack.
* Replace scheduler loop entry with runtime.mcall(schedule).
Runtime.mcall eliminates the need for fake scheduler states that
exist just to run a bit of code on the m->g0 stack
(Grecovery, Gstackalloc).
The elimination of the scheduler as a loop that stops and
starts using gosave and gogo fixes a bad interaction with the
way cgo uses the m->g0 stack. Cgo runs external (gcc-compiled)
C functions on that stack, and then when calling back into Go,
it sets m->g0->sched.sp below the added call frames, so that
other uses of m->g0's stack will not interfere with those frames.
Unfortunately, gogo (longjmp) back to the scheduler loop at
this point would end up running scheduler with the lower
sp, which no longer points at a valid stack frame for
a call to scheduler. If scheduler then wrote any function call
arguments or local variables to where it expected the stack
frame to be, it would overwrite other data on the stack.
I realized this possibility while debugging a problem with
calling complex Go code in a Go -> C -> Go cgo callback.
This wasn't the bug I was looking for, it turns out, but I believe
it is a real bug nonetheless. Switching to runtime.mcall, which
only adds new frames to the stack and never jumps into
functions running in existing ones, fixes this bug.
* Move cgo-related code out of proc.c into cgocall.c.
* Add very large comment describing cgo call sequences.
* Simpilify, regularize cgo function implementations and names.
* Add test suite as misc/cgo/test.
Now the Go -> C path calls cgocall, which calls asmcgocall,
and the C -> Go path calls cgocallback, which calls cgocallbackg.
The shuffling, which affects mainly the callback case, moves
most of the callback implementation to cgocallback running
on the m->curg stack (not the m->g0 scheduler stack) and
only while accounted for with $GOMAXPROCS (between calls
to exitsyscall and entersyscall).
The previous callback code did not block in startcgocallback's
approximation to exitsyscall, so if, say, the garbage collector
were running, it would still barge in and start doing things
like call malloc. Similarly endcgocallback's approximation of
entersyscall did not call matchmg to kick off new OS threads
when necessary, which caused the bug in issue 1560.
Fixes #1560.
R=iant
CC=golang-dev
https://golang.org/cl/4253054
2011-03-07 08:37:42 -07:00
|
|
|
get_tls(CX)
|
|
|
|
MOVL g(CX), AX
|
|
|
|
CMPL g_stackbase(AX), SP
|
|
|
|
JHI 2(PC)
|
|
|
|
INT $3
|
|
|
|
CMPL SP, g_stackguard(AX)
|
|
|
|
JHI 2(PC)
|
|
|
|
INT $3
|
|
|
|
RET
|
|
|
|
|
2013-07-16 14:24:09 -06:00
|
|
|
TEXT runtime·memclr(SB),7,$0-8
|
2009-03-30 01:01:07 -06:00
|
|
|
MOVL 4(SP), DI // arg 1 addr
|
|
|
|
MOVL 8(SP), CX // arg 2 count
|
2011-07-23 13:46:58 -06:00
|
|
|
MOVL CX, BX
|
|
|
|
ANDL $3, BX
|
2009-03-30 01:01:07 -06:00
|
|
|
SHRL $2, CX
|
|
|
|
MOVL $0, AX
|
|
|
|
CLD
|
|
|
|
REP
|
|
|
|
STOSL
|
2011-07-23 13:46:58 -06:00
|
|
|
MOVL BX, CX
|
|
|
|
REP
|
|
|
|
STOSB
|
2009-03-30 01:01:07 -06:00
|
|
|
RET
|
|
|
|
|
2013-07-16 14:24:09 -06:00
|
|
|
TEXT runtime·getcallerpc(SB),7,$0-4
|
2009-03-30 01:01:07 -06:00
|
|
|
MOVL x+0(FP),AX // addr of first arg
|
|
|
|
MOVL -4(AX),AX // get calling pc
|
|
|
|
RET
|
|
|
|
|
2013-07-16 14:24:09 -06:00
|
|
|
TEXT runtime·setcallerpc(SB),7,$0-8
|
2009-03-30 01:01:07 -06:00
|
|
|
MOVL x+0(FP),AX // addr of first arg
|
|
|
|
MOVL x+4(FP), BX
|
|
|
|
MOVL BX, -4(AX) // set calling pc
|
|
|
|
RET
|
|
|
|
|
2013-07-16 14:24:09 -06:00
|
|
|
TEXT runtime·getcallersp(SB), 7, $0-4
|
2010-04-05 13:51:09 -06:00
|
|
|
MOVL sp+0(FP), AX
|
|
|
|
RET
|
|
|
|
|
2012-02-02 12:09:27 -07:00
|
|
|
// int64 runtime·cputicks(void), so really
|
|
|
|
// void runtime·cputicks(int64 *ticks)
|
2013-07-16 14:24:09 -06:00
|
|
|
TEXT runtime·cputicks(SB),7,$0-4
|
2012-02-06 10:49:28 -07:00
|
|
|
RDTSC
|
2012-02-02 12:09:27 -07:00
|
|
|
MOVL ret+0(FP), DI
|
|
|
|
MOVL AX, 0(DI)
|
|
|
|
MOVL DX, 4(DI)
|
|
|
|
RET
|
|
|
|
|
2013-07-16 14:24:09 -06:00
|
|
|
TEXT runtime·ldt0setup(SB),7,$16-0
|
2009-03-30 01:01:07 -06:00
|
|
|
// set up ldt 7 to point at tls0
|
|
|
|
// ldt 1 would be fine on Linux, but on OS X, 7 is as low as we can go.
|
2009-09-22 17:28:32 -06:00
|
|
|
// the entry number is just a hint. setldt will set up GS with what it used.
|
2009-03-30 01:01:07 -06:00
|
|
|
MOVL $7, 0(SP)
|
runtime: ,s/[a-zA-Z0-9_]+/runtime·&/g, almost
Prefix all external symbols in runtime by runtime·,
to avoid conflicts with possible symbols of the same
name in linked-in C libraries. The obvious conflicts
are printf, malloc, and free, but hide everything to
avoid future pain.
The symbols left alone are:
** known to cgo **
_cgo_free
_cgo_malloc
libcgo_thread_start
initcgo
ncgocall
** known to linker **
_rt0_$GOARCH
_rt0_$GOARCH_$GOOS
text
etext
data
end
pclntab
epclntab
symtab
esymtab
** known to C compiler **
_divv
_modv
_div64by32
etc (arch specific)
Tested on darwin/386, darwin/amd64, linux/386, linux/amd64.
Built (but not tested) for freebsd/386, freebsd/amd64, linux/arm, windows/386.
R=r, PeterGo
CC=golang-dev
https://golang.org/cl/2899041
2010-11-04 12:00:19 -06:00
|
|
|
LEAL runtime·tls0(SB), AX
|
2009-03-30 01:01:07 -06:00
|
|
|
MOVL AX, 4(SP)
|
|
|
|
MOVL $32, 8(SP) // sizeof(tls array)
|
runtime: ,s/[a-zA-Z0-9_]+/runtime·&/g, almost
Prefix all external symbols in runtime by runtime·,
to avoid conflicts with possible symbols of the same
name in linked-in C libraries. The obvious conflicts
are printf, malloc, and free, but hide everything to
avoid future pain.
The symbols left alone are:
** known to cgo **
_cgo_free
_cgo_malloc
libcgo_thread_start
initcgo
ncgocall
** known to linker **
_rt0_$GOARCH
_rt0_$GOARCH_$GOOS
text
etext
data
end
pclntab
epclntab
symtab
esymtab
** known to C compiler **
_divv
_modv
_div64by32
etc (arch specific)
Tested on darwin/386, darwin/amd64, linux/386, linux/amd64.
Built (but not tested) for freebsd/386, freebsd/amd64, linux/arm, windows/386.
R=r, PeterGo
CC=golang-dev
https://golang.org/cl/2899041
2010-11-04 12:00:19 -06:00
|
|
|
CALL runtime·setldt(SB)
|
2009-03-30 01:01:07 -06:00
|
|
|
RET
|
|
|
|
|
2013-07-16 14:24:09 -06:00
|
|
|
TEXT runtime·emptyfunc(SB),0,$0-0
|
2009-03-30 01:01:07 -06:00
|
|
|
RET
|
|
|
|
|
2013-07-16 14:24:09 -06:00
|
|
|
TEXT runtime·abort(SB),7,$0-0
|
2009-03-30 01:01:07 -06:00
|
|
|
INT $0x3
|
2009-10-03 11:37:12 -06:00
|
|
|
|
2013-07-16 14:24:09 -06:00
|
|
|
TEXT runtime·stackguard(SB),7,$0-8
|
2012-03-15 13:22:30 -06:00
|
|
|
MOVL SP, DX
|
|
|
|
MOVL DX, sp+0(FP)
|
|
|
|
get_tls(CX)
|
|
|
|
MOVL g(CX), BX
|
|
|
|
MOVL g_stackguard(BX), DX
|
2013-03-22 10:57:55 -06:00
|
|
|
MOVL DX, limit+4(FP)
|
2012-03-15 13:22:30 -06:00
|
|
|
RET
|
|
|
|
|
runtime: ,s/[a-zA-Z0-9_]+/runtime·&/g, almost
Prefix all external symbols in runtime by runtime·,
to avoid conflicts with possible symbols of the same
name in linked-in C libraries. The obvious conflicts
are printf, malloc, and free, but hide everything to
avoid future pain.
The symbols left alone are:
** known to cgo **
_cgo_free
_cgo_malloc
libcgo_thread_start
initcgo
ncgocall
** known to linker **
_rt0_$GOARCH
_rt0_$GOARCH_$GOOS
text
etext
data
end
pclntab
epclntab
symtab
esymtab
** known to C compiler **
_divv
_modv
_div64by32
etc (arch specific)
Tested on darwin/386, darwin/amd64, linux/386, linux/amd64.
Built (but not tested) for freebsd/386, freebsd/amd64, linux/arm, windows/386.
R=r, PeterGo
CC=golang-dev
https://golang.org/cl/2899041
2010-11-04 12:00:19 -06:00
|
|
|
GLOBL runtime·tls0(SB), $32
|
2013-03-12 11:47:44 -06:00
|
|
|
|
|
|
|
// hash function using AES hardware instructions
|
2013-07-16 14:24:09 -06:00
|
|
|
TEXT runtime·aeshash(SB),7,$0-12
|
2013-03-12 11:47:44 -06:00
|
|
|
MOVL 4(SP), DX // ptr to hash value
|
|
|
|
MOVL 8(SP), CX // size
|
|
|
|
MOVL 12(SP), AX // ptr to data
|
|
|
|
JMP runtime·aeshashbody(SB)
|
|
|
|
|
2013-07-16 14:24:09 -06:00
|
|
|
TEXT runtime·aeshashstr(SB),7,$0-12
|
2013-03-12 11:47:44 -06:00
|
|
|
MOVL 4(SP), DX // ptr to hash value
|
|
|
|
MOVL 12(SP), AX // ptr to string struct
|
|
|
|
MOVL 4(AX), CX // length of string
|
|
|
|
MOVL (AX), AX // string data
|
|
|
|
JMP runtime·aeshashbody(SB)
|
|
|
|
|
|
|
|
// AX: data
|
|
|
|
// CX: length
|
|
|
|
// DX: ptr to seed input / hash output
|
2013-07-16 14:24:09 -06:00
|
|
|
TEXT runtime·aeshashbody(SB),7,$0-12
|
2013-03-12 11:47:44 -06:00
|
|
|
MOVL (DX), X0 // seed to low 32 bits of xmm0
|
|
|
|
PINSRD $1, CX, X0 // size to next 32 bits of xmm0
|
2013-03-20 15:34:26 -06:00
|
|
|
MOVO runtime·aeskeysched+0(SB), X2
|
|
|
|
MOVO runtime·aeskeysched+16(SB), X3
|
2013-05-15 10:40:14 -06:00
|
|
|
CMPL CX, $16
|
|
|
|
JB aessmall
|
2013-03-12 11:47:44 -06:00
|
|
|
aesloop:
|
|
|
|
CMPL CX, $16
|
2013-05-15 10:40:14 -06:00
|
|
|
JBE aesloopend
|
2013-03-12 11:47:44 -06:00
|
|
|
MOVOU (AX), X1
|
|
|
|
AESENC X2, X0
|
|
|
|
AESENC X1, X0
|
|
|
|
SUBL $16, CX
|
|
|
|
ADDL $16, AX
|
|
|
|
JMP aesloop
|
2013-05-15 10:40:14 -06:00
|
|
|
// 1-16 bytes remaining
|
2013-03-12 11:47:44 -06:00
|
|
|
aesloopend:
|
2013-05-15 10:40:14 -06:00
|
|
|
// This load may overlap with the previous load above.
|
|
|
|
// We'll hash some bytes twice, but that's ok.
|
|
|
|
MOVOU -16(AX)(CX*1), X1
|
|
|
|
JMP partial
|
|
|
|
// 0-15 bytes
|
|
|
|
aessmall:
|
2013-03-12 11:47:44 -06:00
|
|
|
TESTL CX, CX
|
2013-05-15 10:40:14 -06:00
|
|
|
JE finalize // 0 bytes
|
2013-03-12 11:47:44 -06:00
|
|
|
|
2013-05-15 10:40:14 -06:00
|
|
|
CMPB AX, $0xf0
|
|
|
|
JA highpartial
|
2013-03-12 11:47:44 -06:00
|
|
|
|
2013-05-15 10:40:14 -06:00
|
|
|
// 16 bytes loaded at this address won't cross
|
|
|
|
// a page boundary, so we can load it directly.
|
2013-03-12 11:47:44 -06:00
|
|
|
MOVOU (AX), X1
|
|
|
|
ADDL CX, CX
|
2013-07-16 14:24:09 -06:00
|
|
|
PAND masks<>(SB)(CX*8), X1
|
2013-03-12 11:47:44 -06:00
|
|
|
JMP partial
|
|
|
|
highpartial:
|
2013-05-15 10:40:14 -06:00
|
|
|
// address ends in 1111xxxx. Might be up against
|
2013-03-12 11:47:44 -06:00
|
|
|
// a page boundary, so load ending at last byte.
|
|
|
|
// Then shift bytes down using pshufb.
|
|
|
|
MOVOU -16(AX)(CX*1), X1
|
|
|
|
ADDL CX, CX
|
2013-07-16 14:24:09 -06:00
|
|
|
PSHUFB shifts<>(SB)(CX*8), X1
|
2013-03-12 11:47:44 -06:00
|
|
|
partial:
|
|
|
|
// incorporate partial block into hash
|
|
|
|
AESENC X3, X0
|
|
|
|
AESENC X1, X0
|
|
|
|
finalize:
|
|
|
|
// finalize hash
|
|
|
|
AESENC X2, X0
|
|
|
|
AESENC X3, X0
|
|
|
|
AESENC X2, X0
|
|
|
|
MOVL X0, (DX)
|
|
|
|
RET
|
|
|
|
|
2013-07-16 14:24:09 -06:00
|
|
|
TEXT runtime·aeshash32(SB),7,$0-12
|
2013-03-12 11:47:44 -06:00
|
|
|
MOVL 4(SP), DX // ptr to hash value
|
|
|
|
MOVL 12(SP), AX // ptr to data
|
|
|
|
MOVL (DX), X0 // seed
|
|
|
|
PINSRD $1, (AX), X0 // data
|
2013-03-20 15:34:26 -06:00
|
|
|
AESENC runtime·aeskeysched+0(SB), X0
|
|
|
|
AESENC runtime·aeskeysched+16(SB), X0
|
|
|
|
AESENC runtime·aeskeysched+0(SB), X0
|
2013-03-12 11:47:44 -06:00
|
|
|
MOVL X0, (DX)
|
|
|
|
RET
|
|
|
|
|
2013-07-16 14:24:09 -06:00
|
|
|
TEXT runtime·aeshash64(SB),7,$0-12
|
2013-03-12 11:47:44 -06:00
|
|
|
MOVL 4(SP), DX // ptr to hash value
|
|
|
|
MOVL 12(SP), AX // ptr to data
|
|
|
|
MOVQ (AX), X0 // data
|
|
|
|
PINSRD $2, (DX), X0 // seed
|
2013-03-20 15:34:26 -06:00
|
|
|
AESENC runtime·aeskeysched+0(SB), X0
|
|
|
|
AESENC runtime·aeskeysched+16(SB), X0
|
|
|
|
AESENC runtime·aeskeysched+0(SB), X0
|
2013-03-12 11:47:44 -06:00
|
|
|
MOVL X0, (DX)
|
|
|
|
RET
|
|
|
|
|
|
|
|
// simple mask to get rid of data in the high part of the register.
|
2013-07-16 14:24:09 -06:00
|
|
|
DATA masks<>+0x00(SB)/4, $0x00000000
|
|
|
|
DATA masks<>+0x04(SB)/4, $0x00000000
|
|
|
|
DATA masks<>+0x08(SB)/4, $0x00000000
|
|
|
|
DATA masks<>+0x0c(SB)/4, $0x00000000
|
2013-03-12 11:47:44 -06:00
|
|
|
|
2013-07-16 14:24:09 -06:00
|
|
|
DATA masks<>+0x10(SB)/4, $0x000000ff
|
|
|
|
DATA masks<>+0x14(SB)/4, $0x00000000
|
|
|
|
DATA masks<>+0x18(SB)/4, $0x00000000
|
|
|
|
DATA masks<>+0x1c(SB)/4, $0x00000000
|
2013-03-12 11:47:44 -06:00
|
|
|
|
2013-07-16 14:24:09 -06:00
|
|
|
DATA masks<>+0x20(SB)/4, $0x0000ffff
|
|
|
|
DATA masks<>+0x24(SB)/4, $0x00000000
|
|
|
|
DATA masks<>+0x28(SB)/4, $0x00000000
|
|
|
|
DATA masks<>+0x2c(SB)/4, $0x00000000
|
2013-03-12 11:47:44 -06:00
|
|
|
|
2013-07-16 14:24:09 -06:00
|
|
|
DATA masks<>+0x30(SB)/4, $0x00ffffff
|
|
|
|
DATA masks<>+0x34(SB)/4, $0x00000000
|
|
|
|
DATA masks<>+0x38(SB)/4, $0x00000000
|
|
|
|
DATA masks<>+0x3c(SB)/4, $0x00000000
|
2013-03-12 11:47:44 -06:00
|
|
|
|
2013-07-16 14:24:09 -06:00
|
|
|
DATA masks<>+0x40(SB)/4, $0xffffffff
|
|
|
|
DATA masks<>+0x44(SB)/4, $0x00000000
|
|
|
|
DATA masks<>+0x48(SB)/4, $0x00000000
|
|
|
|
DATA masks<>+0x4c(SB)/4, $0x00000000
|
2013-03-12 11:47:44 -06:00
|
|
|
|
2013-07-16 14:24:09 -06:00
|
|
|
DATA masks<>+0x50(SB)/4, $0xffffffff
|
|
|
|
DATA masks<>+0x54(SB)/4, $0x000000ff
|
|
|
|
DATA masks<>+0x58(SB)/4, $0x00000000
|
|
|
|
DATA masks<>+0x5c(SB)/4, $0x00000000
|
2013-03-12 11:47:44 -06:00
|
|
|
|
2013-07-16 14:24:09 -06:00
|
|
|
DATA masks<>+0x60(SB)/4, $0xffffffff
|
|
|
|
DATA masks<>+0x64(SB)/4, $0x0000ffff
|
|
|
|
DATA masks<>+0x68(SB)/4, $0x00000000
|
|
|
|
DATA masks<>+0x6c(SB)/4, $0x00000000
|
2013-03-12 11:47:44 -06:00
|
|
|
|
2013-07-16 14:24:09 -06:00
|
|
|
DATA masks<>+0x70(SB)/4, $0xffffffff
|
|
|
|
DATA masks<>+0x74(SB)/4, $0x00ffffff
|
|
|
|
DATA masks<>+0x78(SB)/4, $0x00000000
|
|
|
|
DATA masks<>+0x7c(SB)/4, $0x00000000
|
2013-03-12 11:47:44 -06:00
|
|
|
|
2013-07-16 14:24:09 -06:00
|
|
|
DATA masks<>+0x80(SB)/4, $0xffffffff
|
|
|
|
DATA masks<>+0x84(SB)/4, $0xffffffff
|
|
|
|
DATA masks<>+0x88(SB)/4, $0x00000000
|
|
|
|
DATA masks<>+0x8c(SB)/4, $0x00000000
|
2013-03-12 11:47:44 -06:00
|
|
|
|
2013-07-16 14:24:09 -06:00
|
|
|
DATA masks<>+0x90(SB)/4, $0xffffffff
|
|
|
|
DATA masks<>+0x94(SB)/4, $0xffffffff
|
|
|
|
DATA masks<>+0x98(SB)/4, $0x000000ff
|
|
|
|
DATA masks<>+0x9c(SB)/4, $0x00000000
|
2013-03-12 11:47:44 -06:00
|
|
|
|
2013-07-16 14:24:09 -06:00
|
|
|
DATA masks<>+0xa0(SB)/4, $0xffffffff
|
|
|
|
DATA masks<>+0xa4(SB)/4, $0xffffffff
|
|
|
|
DATA masks<>+0xa8(SB)/4, $0x0000ffff
|
|
|
|
DATA masks<>+0xac(SB)/4, $0x00000000
|
2013-03-12 11:47:44 -06:00
|
|
|
|
2013-07-16 14:24:09 -06:00
|
|
|
DATA masks<>+0xb0(SB)/4, $0xffffffff
|
|
|
|
DATA masks<>+0xb4(SB)/4, $0xffffffff
|
|
|
|
DATA masks<>+0xb8(SB)/4, $0x00ffffff
|
|
|
|
DATA masks<>+0xbc(SB)/4, $0x00000000
|
2013-03-12 11:47:44 -06:00
|
|
|
|
2013-07-16 14:24:09 -06:00
|
|
|
DATA masks<>+0xc0(SB)/4, $0xffffffff
|
|
|
|
DATA masks<>+0xc4(SB)/4, $0xffffffff
|
|
|
|
DATA masks<>+0xc8(SB)/4, $0xffffffff
|
|
|
|
DATA masks<>+0xcc(SB)/4, $0x00000000
|
2013-03-12 11:47:44 -06:00
|
|
|
|
2013-07-16 14:24:09 -06:00
|
|
|
DATA masks<>+0xd0(SB)/4, $0xffffffff
|
|
|
|
DATA masks<>+0xd4(SB)/4, $0xffffffff
|
|
|
|
DATA masks<>+0xd8(SB)/4, $0xffffffff
|
|
|
|
DATA masks<>+0xdc(SB)/4, $0x000000ff
|
2013-03-12 11:47:44 -06:00
|
|
|
|
2013-07-16 14:24:09 -06:00
|
|
|
DATA masks<>+0xe0(SB)/4, $0xffffffff
|
|
|
|
DATA masks<>+0xe4(SB)/4, $0xffffffff
|
|
|
|
DATA masks<>+0xe8(SB)/4, $0xffffffff
|
|
|
|
DATA masks<>+0xec(SB)/4, $0x0000ffff
|
2013-03-12 11:47:44 -06:00
|
|
|
|
2013-07-16 14:24:09 -06:00
|
|
|
DATA masks<>+0xf0(SB)/4, $0xffffffff
|
|
|
|
DATA masks<>+0xf4(SB)/4, $0xffffffff
|
|
|
|
DATA masks<>+0xf8(SB)/4, $0xffffffff
|
|
|
|
DATA masks<>+0xfc(SB)/4, $0x00ffffff
|
|
|
|
|
|
|
|
GLOBL masks<>(SB),8,$256
|
|
|
|
|
|
|
|
// these are arguments to pshufb. They move data down from
|
|
|
|
// the high bytes of the register to the low bytes of the register.
|
|
|
|
// index is how many bytes to move.
|
|
|
|
DATA shifts<>+0x00(SB)/4, $0x00000000
|
|
|
|
DATA shifts<>+0x04(SB)/4, $0x00000000
|
|
|
|
DATA shifts<>+0x08(SB)/4, $0x00000000
|
|
|
|
DATA shifts<>+0x0c(SB)/4, $0x00000000
|
2013-03-12 11:47:44 -06:00
|
|
|
|
2013-07-16 14:24:09 -06:00
|
|
|
DATA shifts<>+0x10(SB)/4, $0xffffff0f
|
|
|
|
DATA shifts<>+0x14(SB)/4, $0xffffffff
|
|
|
|
DATA shifts<>+0x18(SB)/4, $0xffffffff
|
|
|
|
DATA shifts<>+0x1c(SB)/4, $0xffffffff
|
2013-03-12 11:47:44 -06:00
|
|
|
|
2013-07-16 14:24:09 -06:00
|
|
|
DATA shifts<>+0x20(SB)/4, $0xffff0f0e
|
|
|
|
DATA shifts<>+0x24(SB)/4, $0xffffffff
|
|
|
|
DATA shifts<>+0x28(SB)/4, $0xffffffff
|
|
|
|
DATA shifts<>+0x2c(SB)/4, $0xffffffff
|
2013-03-12 11:47:44 -06:00
|
|
|
|
2013-07-16 14:24:09 -06:00
|
|
|
DATA shifts<>+0x30(SB)/4, $0xff0f0e0d
|
|
|
|
DATA shifts<>+0x34(SB)/4, $0xffffffff
|
|
|
|
DATA shifts<>+0x38(SB)/4, $0xffffffff
|
|
|
|
DATA shifts<>+0x3c(SB)/4, $0xffffffff
|
2013-03-12 11:47:44 -06:00
|
|
|
|
2013-07-16 14:24:09 -06:00
|
|
|
DATA shifts<>+0x40(SB)/4, $0x0f0e0d0c
|
|
|
|
DATA shifts<>+0x44(SB)/4, $0xffffffff
|
|
|
|
DATA shifts<>+0x48(SB)/4, $0xffffffff
|
|
|
|
DATA shifts<>+0x4c(SB)/4, $0xffffffff
|
2013-03-12 11:47:44 -06:00
|
|
|
|
2013-07-16 14:24:09 -06:00
|
|
|
DATA shifts<>+0x50(SB)/4, $0x0e0d0c0b
|
|
|
|
DATA shifts<>+0x54(SB)/4, $0xffffff0f
|
|
|
|
DATA shifts<>+0x58(SB)/4, $0xffffffff
|
|
|
|
DATA shifts<>+0x5c(SB)/4, $0xffffffff
|
2013-03-12 11:47:44 -06:00
|
|
|
|
2013-07-16 14:24:09 -06:00
|
|
|
DATA shifts<>+0x60(SB)/4, $0x0d0c0b0a
|
|
|
|
DATA shifts<>+0x64(SB)/4, $0xffff0f0e
|
|
|
|
DATA shifts<>+0x68(SB)/4, $0xffffffff
|
|
|
|
DATA shifts<>+0x6c(SB)/4, $0xffffffff
|
2013-03-12 11:47:44 -06:00
|
|
|
|
2013-07-16 14:24:09 -06:00
|
|
|
DATA shifts<>+0x70(SB)/4, $0x0c0b0a09
|
|
|
|
DATA shifts<>+0x74(SB)/4, $0xff0f0e0d
|
|
|
|
DATA shifts<>+0x78(SB)/4, $0xffffffff
|
|
|
|
DATA shifts<>+0x7c(SB)/4, $0xffffffff
|
2013-03-12 11:47:44 -06:00
|
|
|
|
2013-07-16 14:24:09 -06:00
|
|
|
DATA shifts<>+0x80(SB)/4, $0x0b0a0908
|
|
|
|
DATA shifts<>+0x84(SB)/4, $0x0f0e0d0c
|
|
|
|
DATA shifts<>+0x88(SB)/4, $0xffffffff
|
|
|
|
DATA shifts<>+0x8c(SB)/4, $0xffffffff
|
2013-03-12 11:47:44 -06:00
|
|
|
|
2013-07-16 14:24:09 -06:00
|
|
|
DATA shifts<>+0x90(SB)/4, $0x0a090807
|
|
|
|
DATA shifts<>+0x94(SB)/4, $0x0e0d0c0b
|
|
|
|
DATA shifts<>+0x98(SB)/4, $0xffffff0f
|
|
|
|
DATA shifts<>+0x9c(SB)/4, $0xffffffff
|
2013-03-12 11:47:44 -06:00
|
|
|
|
2013-07-16 14:24:09 -06:00
|
|
|
DATA shifts<>+0xa0(SB)/4, $0x09080706
|
|
|
|
DATA shifts<>+0xa4(SB)/4, $0x0d0c0b0a
|
|
|
|
DATA shifts<>+0xa8(SB)/4, $0xffff0f0e
|
|
|
|
DATA shifts<>+0xac(SB)/4, $0xffffffff
|
2013-03-12 11:47:44 -06:00
|
|
|
|
2013-07-16 14:24:09 -06:00
|
|
|
DATA shifts<>+0xb0(SB)/4, $0x08070605
|
|
|
|
DATA shifts<>+0xb4(SB)/4, $0x0c0b0a09
|
|
|
|
DATA shifts<>+0xb8(SB)/4, $0xff0f0e0d
|
|
|
|
DATA shifts<>+0xbc(SB)/4, $0xffffffff
|
2013-03-12 11:47:44 -06:00
|
|
|
|
2013-07-16 14:24:09 -06:00
|
|
|
DATA shifts<>+0xc0(SB)/4, $0x07060504
|
|
|
|
DATA shifts<>+0xc4(SB)/4, $0x0b0a0908
|
|
|
|
DATA shifts<>+0xc8(SB)/4, $0x0f0e0d0c
|
|
|
|
DATA shifts<>+0xcc(SB)/4, $0xffffffff
|
2013-03-12 11:47:44 -06:00
|
|
|
|
2013-07-16 14:24:09 -06:00
|
|
|
DATA shifts<>+0xd0(SB)/4, $0x06050403
|
|
|
|
DATA shifts<>+0xd4(SB)/4, $0x0a090807
|
|
|
|
DATA shifts<>+0xd8(SB)/4, $0x0e0d0c0b
|
|
|
|
DATA shifts<>+0xdc(SB)/4, $0xffffff0f
|
2013-03-12 11:47:44 -06:00
|
|
|
|
2013-07-16 14:24:09 -06:00
|
|
|
DATA shifts<>+0xe0(SB)/4, $0x05040302
|
|
|
|
DATA shifts<>+0xe4(SB)/4, $0x09080706
|
|
|
|
DATA shifts<>+0xe8(SB)/4, $0x0d0c0b0a
|
|
|
|
DATA shifts<>+0xec(SB)/4, $0xffff0f0e
|
2013-03-12 11:47:44 -06:00
|
|
|
|
2013-07-16 14:24:09 -06:00
|
|
|
DATA shifts<>+0xf0(SB)/4, $0x04030201
|
|
|
|
DATA shifts<>+0xf4(SB)/4, $0x08070605
|
|
|
|
DATA shifts<>+0xf8(SB)/4, $0x0c0b0a09
|
|
|
|
DATA shifts<>+0xfc(SB)/4, $0xff0f0e0d
|
|
|
|
|
|
|
|
GLOBL shifts<>(SB),8,$256
|
2013-03-12 11:47:44 -06:00
|
|
|
|
2013-07-16 14:24:09 -06:00
|
|
|
TEXT runtime·memeq(SB),7,$0-12
|
2013-04-02 17:26:15 -06:00
|
|
|
MOVL a+0(FP), SI
|
|
|
|
MOVL b+4(FP), DI
|
|
|
|
MOVL count+8(FP), BX
|
|
|
|
JMP runtime·memeqbody(SB)
|
|
|
|
|
2013-07-16 14:24:09 -06:00
|
|
|
TEXT bytes·Equal(SB),7,$0-25
|
2013-04-02 17:26:15 -06:00
|
|
|
MOVL a_len+4(FP), BX
|
|
|
|
MOVL b_len+16(FP), CX
|
|
|
|
XORL AX, AX
|
|
|
|
CMPL BX, CX
|
|
|
|
JNE eqret
|
|
|
|
MOVL a+0(FP), SI
|
|
|
|
MOVL b+12(FP), DI
|
|
|
|
CALL runtime·memeqbody(SB)
|
|
|
|
eqret:
|
|
|
|
MOVB AX, ret+24(FP)
|
|
|
|
RET
|
|
|
|
|
|
|
|
// a in SI
|
|
|
|
// b in DI
|
|
|
|
// count in BX
|
2013-07-16 14:24:09 -06:00
|
|
|
TEXT runtime·memeqbody(SB),7,$0-0
|
2013-04-02 17:26:15 -06:00
|
|
|
XORL AX, AX
|
|
|
|
|
|
|
|
CMPL BX, $4
|
|
|
|
JB small
|
|
|
|
|
|
|
|
// 64 bytes at a time using xmm registers
|
|
|
|
hugeloop:
|
|
|
|
CMPL BX, $64
|
|
|
|
JB bigloop
|
|
|
|
TESTL $0x4000000, runtime·cpuid_edx(SB) // check for sse2
|
|
|
|
JE bigloop
|
|
|
|
MOVOU (SI), X0
|
|
|
|
MOVOU (DI), X1
|
|
|
|
MOVOU 16(SI), X2
|
|
|
|
MOVOU 16(DI), X3
|
|
|
|
MOVOU 32(SI), X4
|
|
|
|
MOVOU 32(DI), X5
|
|
|
|
MOVOU 48(SI), X6
|
|
|
|
MOVOU 48(DI), X7
|
|
|
|
PCMPEQB X1, X0
|
|
|
|
PCMPEQB X3, X2
|
|
|
|
PCMPEQB X5, X4
|
|
|
|
PCMPEQB X7, X6
|
|
|
|
PAND X2, X0
|
|
|
|
PAND X6, X4
|
|
|
|
PAND X4, X0
|
|
|
|
PMOVMSKB X0, DX
|
|
|
|
ADDL $64, SI
|
|
|
|
ADDL $64, DI
|
|
|
|
SUBL $64, BX
|
|
|
|
CMPL DX, $0xffff
|
|
|
|
JEQ hugeloop
|
|
|
|
RET
|
|
|
|
|
|
|
|
// 4 bytes at a time using 32-bit register
|
|
|
|
bigloop:
|
|
|
|
CMPL BX, $4
|
|
|
|
JBE leftover
|
|
|
|
MOVL (SI), CX
|
|
|
|
MOVL (DI), DX
|
|
|
|
ADDL $4, SI
|
|
|
|
ADDL $4, DI
|
|
|
|
SUBL $4, BX
|
|
|
|
CMPL CX, DX
|
|
|
|
JEQ bigloop
|
|
|
|
RET
|
|
|
|
|
|
|
|
// remaining 0-4 bytes
|
|
|
|
leftover:
|
|
|
|
MOVL -4(SI)(BX*1), CX
|
|
|
|
MOVL -4(DI)(BX*1), DX
|
|
|
|
CMPL CX, DX
|
|
|
|
SETEQ AX
|
|
|
|
RET
|
|
|
|
|
|
|
|
small:
|
|
|
|
CMPL BX, $0
|
|
|
|
JEQ equal
|
|
|
|
|
|
|
|
LEAL 0(BX*8), CX
|
|
|
|
NEGL CX
|
|
|
|
|
|
|
|
MOVL SI, DX
|
|
|
|
CMPB DX, $0xfc
|
|
|
|
JA si_high
|
|
|
|
|
|
|
|
// load at SI won't cross a page boundary.
|
|
|
|
MOVL (SI), SI
|
|
|
|
JMP si_finish
|
|
|
|
si_high:
|
|
|
|
// address ends in 111111xx. Load up to bytes we want, move to correct position.
|
|
|
|
MOVL -4(SI)(BX*1), SI
|
|
|
|
SHRL CX, SI
|
|
|
|
si_finish:
|
|
|
|
|
|
|
|
// same for DI.
|
|
|
|
MOVL DI, DX
|
|
|
|
CMPB DX, $0xfc
|
|
|
|
JA di_high
|
|
|
|
MOVL (DI), DI
|
|
|
|
JMP di_finish
|
|
|
|
di_high:
|
|
|
|
MOVL -4(DI)(BX*1), DI
|
|
|
|
SHRL CX, DI
|
|
|
|
di_finish:
|
|
|
|
|
|
|
|
SUBL SI, DI
|
|
|
|
SHLL CX, DI
|
|
|
|
equal:
|
|
|
|
SETEQ AX
|
|
|
|
RET
|
2013-05-14 17:05:51 -06:00
|
|
|
|
2013-07-16 14:24:09 -06:00
|
|
|
TEXT runtime·cmpstring(SB),7,$0-20
|
2013-05-14 17:05:51 -06:00
|
|
|
MOVL s1+0(FP), SI
|
|
|
|
MOVL s1+4(FP), BX
|
|
|
|
MOVL s2+8(FP), DI
|
|
|
|
MOVL s2+12(FP), DX
|
|
|
|
CALL runtime·cmpbody(SB)
|
|
|
|
MOVL AX, res+16(FP)
|
|
|
|
RET
|
|
|
|
|
2013-07-16 14:24:09 -06:00
|
|
|
TEXT bytes·Compare(SB),7,$0-28
|
2013-05-14 17:05:51 -06:00
|
|
|
MOVL s1+0(FP), SI
|
|
|
|
MOVL s1+4(FP), BX
|
|
|
|
MOVL s2+12(FP), DI
|
|
|
|
MOVL s2+16(FP), DX
|
|
|
|
CALL runtime·cmpbody(SB)
|
|
|
|
MOVL AX, res+24(FP)
|
|
|
|
RET
|
|
|
|
|
2013-08-01 17:11:19 -06:00
|
|
|
TEXT bytes·IndexByte(SB),7,$0
|
|
|
|
MOVL s+0(FP), SI
|
|
|
|
MOVL s_len+4(FP), CX
|
|
|
|
MOVB c+12(FP), AL
|
|
|
|
MOVL SI, DI
|
|
|
|
CLD; REPN; SCASB
|
|
|
|
JZ 3(PC)
|
|
|
|
MOVL $-1, ret+16(FP)
|
|
|
|
RET
|
|
|
|
SUBL SI, DI
|
|
|
|
SUBL $1, DI
|
|
|
|
MOVL DI, ret+16(FP)
|
|
|
|
RET
|
|
|
|
|
2013-05-14 17:05:51 -06:00
|
|
|
// input:
|
|
|
|
// SI = a
|
|
|
|
// DI = b
|
|
|
|
// BX = alen
|
|
|
|
// DX = blen
|
|
|
|
// output:
|
|
|
|
// AX = 1/0/-1
|
2013-07-16 14:24:09 -06:00
|
|
|
TEXT runtime·cmpbody(SB),7,$0-0
|
2013-05-14 17:05:51 -06:00
|
|
|
CMPL SI, DI
|
|
|
|
JEQ cmp_allsame
|
|
|
|
CMPL BX, DX
|
|
|
|
MOVL DX, BP
|
|
|
|
CMOVLLT BX, BP // BP = min(alen, blen)
|
|
|
|
CMPL BP, $4
|
|
|
|
JB cmp_small
|
|
|
|
TESTL $0x4000000, runtime·cpuid_edx(SB) // check for sse2
|
|
|
|
JE cmp_mediumloop
|
|
|
|
cmp_largeloop:
|
|
|
|
CMPL BP, $16
|
|
|
|
JB cmp_mediumloop
|
|
|
|
MOVOU (SI), X0
|
|
|
|
MOVOU (DI), X1
|
|
|
|
PCMPEQB X0, X1
|
|
|
|
PMOVMSKB X1, AX
|
|
|
|
XORL $0xffff, AX // convert EQ to NE
|
|
|
|
JNE cmp_diff16 // branch if at least one byte is not equal
|
|
|
|
ADDL $16, SI
|
|
|
|
ADDL $16, DI
|
|
|
|
SUBL $16, BP
|
|
|
|
JMP cmp_largeloop
|
|
|
|
|
|
|
|
cmp_diff16:
|
|
|
|
BSFL AX, BX // index of first byte that differs
|
|
|
|
XORL AX, AX
|
|
|
|
MOVB (SI)(BX*1), CX
|
|
|
|
CMPB CX, (DI)(BX*1)
|
|
|
|
SETHI AX
|
|
|
|
LEAL -1(AX*2), AX // convert 1/0 to +1/-1
|
|
|
|
RET
|
|
|
|
|
|
|
|
cmp_mediumloop:
|
|
|
|
CMPL BP, $4
|
|
|
|
JBE cmp_0through4
|
|
|
|
MOVL (SI), AX
|
|
|
|
MOVL (DI), CX
|
|
|
|
CMPL AX, CX
|
|
|
|
JNE cmp_diff4
|
|
|
|
ADDL $4, SI
|
|
|
|
ADDL $4, DI
|
|
|
|
SUBL $4, BP
|
|
|
|
JMP cmp_mediumloop
|
|
|
|
|
|
|
|
cmp_0through4:
|
|
|
|
MOVL -4(SI)(BP*1), AX
|
|
|
|
MOVL -4(DI)(BP*1), CX
|
|
|
|
CMPL AX, CX
|
|
|
|
JEQ cmp_allsame
|
|
|
|
|
|
|
|
cmp_diff4:
|
|
|
|
BSWAPL AX // reverse order of bytes
|
|
|
|
BSWAPL CX
|
|
|
|
XORL AX, CX // find bit differences
|
|
|
|
BSRL CX, CX // index of highest bit difference
|
|
|
|
SHRL CX, AX // move a's bit to bottom
|
|
|
|
ANDL $1, AX // mask bit
|
|
|
|
LEAL -1(AX*2), AX // 1/0 => +1/-1
|
|
|
|
RET
|
|
|
|
|
|
|
|
// 0-3 bytes in common
|
|
|
|
cmp_small:
|
|
|
|
LEAL (BP*8), CX
|
|
|
|
NEGL CX
|
|
|
|
JEQ cmp_allsame
|
|
|
|
|
|
|
|
// load si
|
|
|
|
CMPB SI, $0xfc
|
|
|
|
JA cmp_si_high
|
|
|
|
MOVL (SI), SI
|
|
|
|
JMP cmp_si_finish
|
|
|
|
cmp_si_high:
|
|
|
|
MOVL -4(SI)(BP*1), SI
|
|
|
|
SHRL CX, SI
|
|
|
|
cmp_si_finish:
|
|
|
|
SHLL CX, SI
|
|
|
|
|
|
|
|
// same for di
|
|
|
|
CMPB DI, $0xfc
|
|
|
|
JA cmp_di_high
|
|
|
|
MOVL (DI), DI
|
|
|
|
JMP cmp_di_finish
|
|
|
|
cmp_di_high:
|
|
|
|
MOVL -4(DI)(BP*1), DI
|
|
|
|
SHRL CX, DI
|
|
|
|
cmp_di_finish:
|
|
|
|
SHLL CX, DI
|
|
|
|
|
|
|
|
BSWAPL SI // reverse order of bytes
|
|
|
|
BSWAPL DI
|
|
|
|
XORL SI, DI // find bit differences
|
|
|
|
JEQ cmp_allsame
|
|
|
|
BSRL DI, CX // index of highest bit difference
|
|
|
|
SHRL CX, SI // move a's bit to bottom
|
|
|
|
ANDL $1, SI // mask bit
|
|
|
|
LEAL -1(SI*2), AX // 1/0 => +1/-1
|
|
|
|
RET
|
|
|
|
|
|
|
|
// all the bytes in common are the same, so we just need
|
|
|
|
// to compare the lengths.
|
|
|
|
cmp_allsame:
|
|
|
|
XORL AX, AX
|
|
|
|
XORL CX, CX
|
|
|
|
CMPL BX, DX
|
|
|
|
SETGT AX // 1 if alen > blen
|
|
|
|
SETEQ CX // 1 if alen == blen
|
|
|
|
LEAL -1(CX)(AX*2), AX // 1,0,-1 result
|
|
|
|
RET
|