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
diff --git a/src/pkg/runtime/amd64/asm.s b/src/pkg/runtime/amd64/asm.s
index cc05435..a611985 100644
--- a/src/pkg/runtime/amd64/asm.s
+++ b/src/pkg/runtime/amd64/asm.s
@@ -89,7 +89,7 @@
* go-routine
*/
-// uintptr gosave(Gobuf*)
+// void gosave(Gobuf*)
// save state in Gobuf; setjmp
TEXT runtime·gosave(SB), 7, $0
MOVQ 8(SP), AX // gobuf
@@ -100,7 +100,6 @@
get_tls(CX)
MOVQ g(CX), BX
MOVQ BX, gobuf_g(AX)
- MOVL $0, AX // return 0
RET
// void gogo(Gobuf*, uintptr)
@@ -132,6 +131,35 @@
JMP AX
POPQ BX // not reached
+// void mcall(void (*fn)(G*))
+// Switch to m->g0's stack, call fn(g).
+// Fn must never return. It should gogo(&g->gobuf)
+// to keep running g.
+TEXT runtime·mcall(SB), 7, $0
+ MOVQ fn+0(FP), DI
+
+ get_tls(CX)
+ MOVQ g(CX), AX // save state in g->gobuf
+ MOVQ 0(SP), BX // caller's PC
+ MOVQ BX, (g_sched+gobuf_pc)(AX)
+ LEAQ 8(SP), BX // caller's SP
+ MOVQ BX, (g_sched+gobuf_sp)(AX)
+ MOVQ AX, (g_sched+gobuf_g)(AX)
+
+ // switch to m->g0 & its stack, call fn
+ MOVQ m(CX), BX
+ MOVQ m_g0(BX), SI
+ CMPQ SI, AX // if g == m->g0 call badmcall
+ JNE 2(PC)
+ CALL runtime·badmcall(SB)
+ MOVQ SI, g(CX) // g = m->g0
+ MOVQ (g_sched+gobuf_sp)(SI), SP // sp = m->g0->gobuf.sp
+ PUSHQ AX
+ CALL DI
+ POPQ AX
+ CALL runtime·badmcall2(SB)
+ RET
+
/*
* support for morestack
*/
@@ -160,10 +188,10 @@
MOVQ 0(SP), AX
MOVQ AX, m_morepc(BX)
- // Call newstack on m's scheduling stack.
+ // Call newstack on m->g0's stack.
MOVQ m_g0(BX), BP
MOVQ BP, g(CX)
- MOVQ (m_sched+gobuf_sp)(BX), SP
+ MOVQ (g_sched+gobuf_sp)(BP), SP
CALL runtime·newstack(SB)
MOVQ $0, 0x1003 // crash if newstack returns
RET
@@ -201,11 +229,11 @@
MOVL CX, m_moreargsize(BX) // f's argument size
MOVL $1, m_moreframesize(BX) // f's frame size
- // Call newstack on m's scheduling stack.
+ // Call newstack on m->g0's stack.
MOVQ m_g0(BX), BP
get_tls(CX)
MOVQ BP, g(CX)
- MOVQ (m_sched+gobuf_sp)(BX), SP
+ MOVQ (g_sched+gobuf_sp)(BP), SP
CALL runtime·newstack(SB)
MOVQ $0, 0x1103 // crash if newstack returns
RET
@@ -217,10 +245,10 @@
MOVQ m(CX), BX
MOVQ AX, m_cret(BX)
- // Call oldstack on m's scheduling stack.
- MOVQ m_g0(BX), DX
- MOVQ DX, g(CX)
- MOVQ (m_sched+gobuf_sp)(BX), SP
+ // Call oldstack on m->g0's stack.
+ MOVQ m_g0(BX), BP
+ MOVQ BP, g(CX)
+ MOVQ (g_sched+gobuf_sp)(BP), SP
CALL runtime·oldstack(SB)
MOVQ $0, 0x1004 // crash if oldstack returns
RET
@@ -336,7 +364,6 @@
MOVL $1, AX
RET
-
// void jmpdefer(fn, sp);
// called from deferreturn.
// 1. pop the caller
@@ -349,68 +376,119 @@
SUBQ $5, (SP) // return to CALL again
JMP AX // but first run the deferred function
-// runcgo(void(*fn)(void*), void *arg)
+// Dummy function to use in saved gobuf.PC,
+// to match SP pointing at a return address.
+// The gobuf.PC is unused by the contortions here
+// but setting it to return will make the traceback code work.
+TEXT return<>(SB),7,$0
+ RET
+
+// asmcgocall(void(*fn)(void*), void *arg)
// Call fn(arg) on the scheduler stack,
// aligned appropriately for the gcc ABI.
-TEXT runtime·runcgo(SB),7,$32
- MOVQ fn+0(FP), R12
- MOVQ arg+8(FP), R13
- MOVQ SP, CX
+// See cgocall.c for more details.
+TEXT runtime·asmcgocall(SB),7,$0
+ MOVQ fn+0(FP), AX
+ MOVQ arg+8(FP), BX
+ MOVQ SP, DX
// Figure out if we need to switch to m->g0 stack.
- get_tls(DI)
- MOVQ m(DI), DX
- MOVQ m_g0(DX), SI
- CMPQ g(DI), SI
- JEQ 2(PC)
- MOVQ (m_sched+gobuf_sp)(DX), SP
+ // We get called to create new OS threads too, and those
+ // come in on the m->g0 stack already.
+ get_tls(CX)
+ MOVQ m(CX), BP
+ MOVQ m_g0(BP), SI
+ MOVQ g(CX), DI
+ CMPQ SI, DI
+ JEQ 6(PC)
+ MOVQ SP, (g_sched+gobuf_sp)(DI)
+ MOVQ $return<>(SB), (g_sched+gobuf_pc)(DI)
+ MOVQ DI, (g_sched+gobuf_g)(DI)
+ MOVQ SI, g(CX)
+ MOVQ (g_sched+gobuf_sp)(SI), SP
// Now on a scheduling stack (a pthread-created stack).
SUBQ $32, SP
ANDQ $~15, SP // alignment for gcc ABI
- MOVQ g(DI), BP
- MOVQ BP, 16(SP)
- MOVQ SI, g(DI)
- MOVQ CX, 8(SP)
- MOVQ R13, DI // DI = first argument in AMD64 ABI
- CALL R12
+ MOVQ DI, 16(SP) // save g
+ MOVQ DX, 8(SP) // save SP
+ MOVQ BX, DI // DI = first argument in AMD64 ABI
+ CALL AX
// Restore registers, g, stack pointer.
- get_tls(DI)
- MOVQ 16(SP), SI
- MOVQ SI, g(DI)
+ get_tls(CX)
+ MOVQ 16(SP), DI
+ MOVQ DI, g(CX)
MOVQ 8(SP), SP
RET
-// runcgocallback(G *g1, void* sp, void (*fn)(void))
-// Switch to g1 and sp, call fn, switch back. fn's arguments are on
-// the new stack.
-TEXT runtime·runcgocallback(SB),7,$48
- MOVQ g1+0(FP), DX
- MOVQ sp+8(FP), AX
- MOVQ fp+16(FP), BX
+// cgocallback(void (*fn)(void*), void *frame, uintptr framesize)
+// See cgocall.c for more details.
+TEXT runtime·cgocallback(SB),7,$24
+ MOVQ fn+0(FP), AX
+ MOVQ frame+8(FP), BX
+ MOVQ framesize+16(FP), DX
- // We are running on m's scheduler stack. Save current SP
- // into m->sched.sp so that a recursive call to runcgo doesn't
- // clobber our stack, and also so that we can restore
- // the SP when the call finishes. Reusing m->sched.sp
- // for this purpose depends on the fact that there is only
- // one possible gosave of m->sched.
+ // Save current m->g0->sched.sp on stack and then set it to SP.
get_tls(CX)
- MOVQ DX, g(CX)
- MOVQ m(CX), CX
- MOVQ SP, (m_sched+gobuf_sp)(CX)
+ MOVQ m(CX), BP
+ MOVQ m_g0(BP), SI
+ PUSHQ (g_sched+gobuf_sp)(SI)
+ MOVQ SP, (g_sched+gobuf_sp)(SI)
- // Set new SP, call fn
- MOVQ AX, SP
- CALL BX
+ // Switch to m->curg stack and call runtime.cgocallback
+ // with the three arguments. 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->gobuf)
+ // so that we can restore it when we're done.
+ // We can restore m->curg->gobuf.sp easily, because calling
+ // runtime.cgocallback leaves SP unchanged upon return.
+ // To save m->curg->gobuf.pc, we push it onto the stack.
+ // This has the added benefit that it looks to the traceback
+ // routine like cgocallback is going to return to that
+ // PC (because we defined cgocallback to have
+ // a frame size of 24, the same amount that we use below),
+ // so that the traceback will seamlessly trace back into
+ // the earlier calls.
+ MOVQ m_curg(BP), SI
+ MOVQ SI, g(CX)
+ MOVQ (g_sched+gobuf_sp)(SI), DI // prepare stack as DI
- // Restore old g and SP, return
+ // Push gobuf.pc
+ MOVQ (g_sched+gobuf_pc)(SI), BP
+ SUBQ $8, DI
+ MOVQ BP, 0(DI)
+
+ // Push arguments to cgocallbackg.
+ // Frame size here must match the frame size above
+ // to trick traceback routines into doing the right thing.
+ SUBQ $24, DI
+ MOVQ AX, 0(DI)
+ MOVQ BX, 8(DI)
+ MOVQ DX, 16(DI)
+
+ // Switch stack and make the call.
+ MOVQ DI, SP
+ CALL runtime·cgocallbackg(SB)
+
+ // Restore g->gobuf (== m->curg->gobuf) from saved values.
get_tls(CX)
- MOVQ m(CX), DX
- MOVQ m_g0(DX), BX
- MOVQ BX, g(CX)
- MOVQ (m_sched+gobuf_sp)(DX), SP
+ MOVQ g(CX), SI
+ MOVQ 24(SP), BP
+ MOVQ BP, (g_sched+gobuf_pc)(SI)
+ LEAQ (24+8)(SP), DI
+ MOVQ 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.)
+ MOVQ m(CX), BP
+ MOVQ m_g0(BP), SI
+ MOVQ SI, g(CX)
+ MOVQ (g_sched+gobuf_sp)(SI), SP
+ POPQ (g_sched+gobuf_sp)(SI)
+
+ // Done!
RET
// check that SP is in range [g->stackbase, g->stackguard)
