runtime: make asmcgocall more robust to missing G

Being able to call asmcgocall without a G is useful for code shared
between different stages of the runtime initialization and thread
creation.

Cq-Include-Trybots: luci.golang.try:gotip-darwin-arm64_15,gotip-linux-mips64le,gotip-linux-ppc64le_power10,gotip-linux-riscv64,gotip-openbsd-ppc64,gotip-openbsd-amd64
Change-Id: Ic427764de197e648e8b9987c98c3b7521512cc5c
Reviewed-on: https://go-review.googlesource.com/c/go/+/750541
Reviewed-by: Carlos Amedee <carlos@golang.org>
LUCI-TryBot-Result: Go LUCI <golang-scoped@luci-project-accounts.iam.gserviceaccount.com>
Reviewed-by: Cherry Mui <cherryyz@google.com>

src/runtime/asm_386.s

@@ -665,6 +665,13 @@ TEXT ·asmcgocall(SB),NOSPLIT,$0-12
 	// We get called to create new OS threads too, and those
 	// come in on the m->g0 stack already. Or we might already
 	// be on the m->gsignal stack.
+#ifdef GOOS_windows
+	// On Windows, get_tls might return garbage if the thread
+	// has never called into Go, so check tls_g directly.
+	MOVL	runtime·tls_g(SB), CX
+	CMPL	CX, $0
+	JEQ	nosave
+#endif
 	get_tls(CX)
 	MOVL	g(CX), DI
 	CMPL	DI, $0
@@ -741,7 +748,7 @@ loadg:
 #ifdef GOOS_windows
 	MOVL	$0, BP
 	CMPL	CX, $0
-	JEQ	2(PC) // TODO
+	JEQ	needm
 #endif
 	MOVL	g(CX), BP
 	CMPL	BP, $0

src/runtime/asm_amd64.s

@@ -1015,10 +1015,6 @@ nosave:
 	// This code is like the above sequence but without saving/restoring g
 	// and without worrying about the stack moving out from under us
 	// (because we're on a system stack, not a goroutine stack).
-	// The above code could be used directly if already on a system stack,
-	// but then the only path through this code would be a rare case on Solaris.
-	// Using this code for all "already on system stack" calls exercises it more,
-	// which should help keep it correct.
 	MOVQ	fn+0(FP), AX
 	MOVQ	arg+8(FP), BX
 	MOVQ	SP, DX

src/runtime/asm_arm.s

@@ -630,10 +630,6 @@ nosave:
 	// This code is like the above sequence but without saving/restoring g
 	// and without worrying about the stack moving out from under us
 	// (because we're on a system stack, not a goroutine stack).
-	// The above code could be used directly if already on a system stack,
-	// but then the only path through this code would be a rare case on Solaris.
-	// Using this code for all "already on system stack" calls exercises it more,
-	// which should help keep it correct.
 	SUB	$24, R13
 	BIC	$0x7, R13	// alignment for gcc ABI
 	// save null g in case someone looks during debugging.

src/runtime/asm_arm64.s

@@ -1244,10 +1244,6 @@ nosave:
 	// This code is like the above sequence but without saving/restoring g
 	// and without worrying about the stack moving out from under us
 	// (because we're on a system stack, not a goroutine stack).
-	// The above code could be used directly if already on a system stack,
-	// but then the only path through this code would be a rare case on Solaris.
-	// Using this code for all "already on system stack" calls exercises it more,
-	// which should help keep it correct.
 	MOVD	fn+0(FP), R1
 	MOVD	arg+8(FP), R0
 	MOVD	RSP, R2

src/runtime/asm_loong64.s

@@ -539,6 +539,7 @@ TEXT ·asmcgocall(SB),NOSPLIT,$0-20
 	// 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.
+	BEQ	g, R0, nosave
 	MOVV	g_m(g), R5
 	MOVV	m_gsignal(R5), R6
 	BEQ	R6, g, g0
@@ -571,6 +572,20 @@ g0:
 	MOVW	R4, ret+16(FP)
 	RET
 
+nosave:
+	// Running on a system stack, perhaps even without a g.
+	// Having no g can happen during thread creation or thread teardown.
+	MOVV	fn+0(FP), R25
+	MOVV	arg+8(FP), R4
+	MOVV	R3, R12
+	ADDV	$-16, R3
+	MOVV	R0, 0(R3)	// Where above code stores g, in case someone looks during debugging.
+	MOVV	R12, 8(R3)	// Save original stack pointer.
+	JAL	(R25)
+	MOVV	8(R3), R3	// Restore stack pointer.
+	MOVW	R4, ret+16(FP)
+	RET
+
 // func cgocallback(fn, frame unsafe.Pointer, ctxt uintptr)
 // See cgocall.go for more details.
 TEXT ·cgocallback(SB),NOSPLIT,$24-24

src/runtime/asm_mips64x.s

@@ -452,6 +452,7 @@ TEXT ·asmcgocall(SB),NOSPLIT,$0-20
 	// We get called to create new OS threads too, and those
 	// come in on the m->g0 stack already. Or we might already
 	// be on the m->gsignal stack.
+	BEQ	g, R0, nosave
 	MOVV	g_m(g), R5
 	MOVV	m_gsignal(R5), R6
 	BEQ	R6, g, g0
@@ -484,6 +485,20 @@ g0:
 	MOVW	R2, ret+16(FP)
 	RET
 
+nosave:
+	// Running on a system stack, perhaps even without a g.
+	// Having no g can happen during thread creation or thread teardown.
+	MOVV	fn+0(FP), R25
+	MOVV	arg+8(FP), R4
+	MOVV	R29, R3
+	ADDV	$-16, R29
+	MOVV	R0, 0(R29)	// Where above code stores g, in case someone looks during debugging.
+	MOVV	R3, 8(R29)	// Save original stack pointer.
+	JAL	(R25)
+	MOVV	8(R29), R29	// Restore stack pointer.
+	MOVW	R2, ret+16(FP)
+	RET
+
 // func cgocallback(fn, frame unsafe.Pointer, ctxt uintptr)
 // See cgocall.go for more details.
 TEXT ·cgocallback(SB),NOSPLIT,$24-24

src/runtime/asm_ppc64x.s

@@ -782,10 +782,6 @@ nosave:
 	// This code is like the above sequence but without saving/restoring g
 	// and without worrying about the stack moving out from under us
 	// (because we're on a system stack, not a goroutine stack).
-	// The above code could be used directly if already on a system stack,
-	// but then the only path through this code would be a rare case.
-	// Using this code for all "already on system stack" calls exercises it more,
-	// which should help keep it correct.
 
 	SUB	$(asmcgocallSaveOffset+8), R1, R10
 	RLDCR	$0, R10, $~15, R1		// 16-byte alignment for gcc ABI

src/runtime/asm_riscv64.s

@@ -398,6 +398,7 @@ TEXT ·asmcgocall(SB),NOSPLIT,$0-20
 	// We get called to create new OS threads too, and those
 	// come in on the m->g0 stack already. Or we might already
 	// be on the m->gsignal stack.
+	BEQZ	g, nosave
 	MOV	g_m(g), X6
 	MOV	m_gsignal(X6), X7
 	BEQ	X7, g, g0
@@ -431,6 +432,20 @@ g0:
 	MOVW	X10, ret+16(FP)
 	RET
 
+nosave:
+	// Running on a system stack, perhaps even without a g.
+	// Having no g can happen during thread creation or thread teardown.
+	MOV	fn+0(FP), X11
+	MOV	arg+8(FP), X10
+	MOV	X2, X8
+	SUB	$16, X2
+	MOV	ZERO, 0(X2)	// Where above code stores g, in case someone looks during debugging.
+	MOV	X8, 8(X2)	// Save original stack pointer.
+	JALR	RA, (X11)
+	MOV	8(X2), X2	// Restore stack pointer.
+	MOVW	X10, ret+16(FP)
+	RET
+
 // func asminit()
 TEXT runtime·asminit(SB),NOSPLIT|NOFRAME,$0-0
 	RET

src/runtime/asm_s390x.s

@@ -552,6 +552,7 @@ TEXT ·asmcgocall(SB),NOSPLIT,$0-20
 	// We get called to create new OS threads too, and those
 	// come in on the m->g0 stack already. Or we might already
 	// be on the m->gsignal stack.
+	CMPBEQ	g, $0, nosave
 	MOVD	g_m(g), R6
 	MOVD	m_gsignal(R6), R7
 	CMPBEQ	R7, g, g0
@@ -589,6 +590,25 @@ g0:
 	MOVW	R2, ret+16(FP)
 	RET
 
+nosave:
+	// Running on a system stack, perhaps even without a g.
+	// Having no g can happen during thread creation or thread teardown.
+	MOVD	fn+0(FP), R3
+	MOVD	arg+8(FP), R4
+	MOVD	R15, R2
+	SUB	$176, R15
+	MOVD	$~7, R6
+	AND	R6, R15
+	MOVD	$0, 168(R15)	// Where above code stores g, in case someone looks during debugging.
+	MOVD	R2, 160(R15)	// Save original stack pointer.
+	MOVD	$0, 0(R15)	// clear back chain pointer
+	MOVD	R4, R2		// arg in R2
+	BL	R3
+	XOR	R0, R0
+	MOVD	160(R15), R15	// Restore stack pointer.
+	MOVW	R2, ret+16(FP)
+	RET
+
 // cgocallback(fn, frame unsafe.Pointer, ctxt uintptr)
 // See cgocall.go for more details.
 TEXT ·cgocallback(SB),NOSPLIT,$24-24

src/runtime/sys_openbsd.go

@@ -85,7 +85,7 @@ func sched_yield_trampoline()
 
 //go:nosplit
 func osyield_no_g() {
-	asmcgocall_no_g(unsafe.Pointer(abi.FuncPCABI0(sched_yield_trampoline)), unsafe.Pointer(nil))
+	asmcgocall(unsafe.Pointer(abi.FuncPCABI0(sched_yield_trampoline)), unsafe.Pointer(nil))
 }
 
 // This is exported via linkname to assembly in runtime/cgo.