runtime: refactor/fix asmcgocall/asmcgocall_errno

Instead of making asmcgocall call asmcgocall_errno,
make both load args into registers and call a shared
assembly function.

On amd64, this costs 1 word in the asmcgocall_errno path
but saves 3 words in the asmcgocall path, and the latter
is what happens on critical nosplit paths on Windows.

On arm, this fixes build failures: asmcgocall was writing
the arguments for asmcgocall_errno into the wrong
place on the stack. Passing them in registers avoids the
decision entirely.

On 386, this isn't really needed, since the nosplit paths
have twice as many words to work with, but do it for consistency.

Update #8635
Fixes arm build (except GOARM=5).

TBR=iant
CC=golang-codereviews
https://golang.org/cl/134390043

3 files changed, 25 insertions, 16 deletions

diff --git a/src/pkg/runtime/asm_386.s b/src/pkg/runtime/asm_386.s
index 2477726f51..aafe960ce3 100644
--- a/src/pkg/runtime/asm_386.s
+++ b/src/pkg/runtime/asm_386.s
@@ -680,17 +680,21 @@ TEXT gosave<>(SB),NOSPLIT,$0
 // Call fn(arg) on the scheduler stack,
 // aligned appropriately for the gcc ABI.
 // See cgocall.c for more details.
-TEXT runtime·asmcgocall(SB),NOSPLIT,$12-8
+TEXT runtime·asmcgocall(SB),NOSPLIT,$0-8
 	MOVL	fn+0(FP), AX
 	MOVL	arg+4(FP), BX
-	MOVL	AX, 0(SP)
-	MOVL	BX, 4(SP)
-	CALL	runtime·asmcgocall_errno(SB)
+	CALL	asmcgocall<>(SB)
 	RET
 
 TEXT runtime·asmcgocall_errno(SB),NOSPLIT,$0-12
 	MOVL	fn+0(FP), AX
 	MOVL	arg+4(FP), BX
+	CALL	asmcgocall<>(SB)
+	MOVL	AX, ret+8(FP)
+	RET
+
+TEXT asmcgocall<>(SB),NOSPLIT,$0-12
+	// fn in AX, arg in BX
 	MOVL	SP, DX
 
 	// Figure out if we need to switch to m->g0 stack.
@@ -720,7 +724,6 @@ TEXT runtime·asmcgocall_errno(SB),NOSPLIT,$0-12
 	MOVL	8(SP), DI
 	MOVL	DI, g(CX)
 	MOVL	4(SP), SP
-	MOVL	AX, ret+8(FP)
 	RET
 
 // cgocallback(void (*fn)(void*), void *frame, uintptr framesize)
diff --git a/src/pkg/runtime/asm_amd64.s b/src/pkg/runtime/asm_amd64.s
index d7e30de7b4..5840b56c81 100644
--- a/src/pkg/runtime/asm_amd64.s
+++ b/src/pkg/runtime/asm_amd64.s
@@ -764,17 +764,21 @@ TEXT gosave<>(SB),NOSPLIT,$0
 // Call fn(arg) on the scheduler stack,
 // aligned appropriately for the gcc ABI.
 // See cgocall.c for more details.
-TEXT runtime·asmcgocall(SB),NOSPLIT,$24-16
+TEXT runtime·asmcgocall(SB),NOSPLIT,$0-16
 	MOVQ	fn+0(FP), AX
 	MOVQ	arg+8(FP), BX
-	MOVQ	AX, 0(SP)
-	MOVQ	BX, 8(SP)
-	CALL	runtime·asmcgocall_errno(SB)
+	CALL	asmcgocall<>(SB)
 	RET
 
 TEXT runtime·asmcgocall_errno(SB),NOSPLIT,$0-20
 	MOVQ	fn+0(FP), AX
 	MOVQ	arg+8(FP), BX
+	CALL	asmcgocall<>(SB)
+	MOVL	AX, ret+16(FP)
+	RET
+
+// asmcgocall common code. fn in AX, arg in BX. returns errno in AX.
+TEXT asmcgocall<>(SB),NOSPLIT,$0-0
 	MOVQ	SP, DX
 
 	// Figure out if we need to switch to m->g0 stack.
@@ -813,7 +817,6 @@ nosave:
 	MOVQ	48(SP), DI
 	MOVQ	DI, g(CX)
 	MOVQ	40(SP), SP
-	MOVL	AX, ret+16(FP)
 	RET
 
 // cgocallback(void (*fn)(void*), void *frame, uintptr framesize)
diff --git a/src/pkg/runtime/asm_arm.s b/src/pkg/runtime/asm_arm.s
index 26b58cfb17..49a863258c 100644
--- a/src/pkg/runtime/asm_arm.s
+++ b/src/pkg/runtime/asm_arm.s
@@ -493,17 +493,21 @@ TEXT gosave<>(SB),NOSPLIT,$0
 // Call fn(arg) on the scheduler stack,
 // aligned appropriately for the gcc ABI.
 // See cgocall.c for more details.
-TEXT	runtime·asmcgocall(SB),NOSPLIT,$12-8
+TEXT	runtime·asmcgocall(SB),NOSPLIT,$0-8
 	MOVW	fn+0(FP), R1
-	MOVW	arg+4(FP), R2
-	MOVW	R1, 0(R13)
-	MOVW	R2, 4(R13)
-	BL	runtime·asmcgocall_errno(SB)
+	MOVW	arg+4(FP), R0
+	BL	asmcgocall<>(SB)
 	RET
 
 TEXT runtime·asmcgocall_errno(SB),NOSPLIT,$0-12
 	MOVW	fn+0(FP), R1
 	MOVW	arg+4(FP), R0
+	BL	asmcgocall<>(SB)
+	MOVW	R0, ret+8(FP)
+	RET
+
+TEXT asmcgocall<>(SB),NOSPLIT,$0-0
+	// fn in R1, arg in R0.
 	MOVW	R13, R2
 	MOVW	g, R5
 
@@ -529,7 +533,6 @@ TEXT runtime·asmcgocall_errno(SB),NOSPLIT,$0-12
 	// Restore registers, g, stack pointer.
 	MOVW	20(R13), g
 	MOVW	16(R13), R13
-	MOVW	R0, ret+8(FP)
 	RET
 
 // cgocallback(void (*fn)(void*), void *frame, uintptr framesize)