all: merge branch dev.regabi (d3cd4830ad) into master

This CL merges the dev.regabi branch to the master branch. In the dev.regabi branch we have refactored the compiler, and laid some preliminary work for enabling a register-based ABI (issue #40724), including improved late call/return lowering, improved ABI wrapper generation, reflect call prepared for the new ABI, and reserving special registers in the internal ABI. The actual register-based ABI has not been enabled for the moment. The ABI-related changes are behind GOEXPERIMENT=regabi and currently off by default. Updates #40724, #44222. Fixes #44224. Change-Id: Id5de9f734d14099267ab717167aaaeef31fdba70
author: Cherry Zhang <cherryyz@google.com> 2021-02-16 16:55:30 -0500
committer: Cherry Zhang <cherryyz@google.com> 2021-02-16 17:18:37 -0500
commit: 84825599dc2592e9e9f5489323fa7b47534fc391 (patch)
tree: e5aa6d21506d430cae87e63701e89822353d0ec8 /src/runtime
parent: 5faf941df067b33485edb9cd2e880869e7feb6a3 (diff)
parent: d3cd4830adf45ce53c586a83f9d78421484737fd (diff)
download: go-84825599dc2592e9e9f5489323fa7b47534fc391.tar.xz
38 files changed, 599 insertions, 528 deletions
diff --git a/src/runtime/asm.s b/src/runtime/asm.s
index 27d8df9e06..72c744925d 100644
--- a/src/runtime/asm.s
+++ b/src/runtime/asm.s
@@ -11,3 +11,8 @@
 DATA runtime·no_pointers_stackmap+0x00(SB)/4, $2
 DATA runtime·no_pointers_stackmap+0x04(SB)/4, $0
 GLOBL runtime·no_pointers_stackmap(SB),RODATA, $8
+
+#ifndef GOARCH_amd64
+TEXT ·sigpanic0<ABIInternal>(SB),NOSPLIT,$0-0
+	JMP	·sigpanic<ABIInternal>(SB)
+#endif
diff --git a/src/runtime/asm_386.s b/src/runtime/asm_386.s
index fa3b1be339..471451df28 100644
--- a/src/runtime/asm_386.s
+++ b/src/runtime/asm_386.s
@@ -273,25 +273,6 @@ TEXT runtime·asminit(SB),NOSPLIT,$0-0
  *  go-routine
  */
 
-// void gosave(Gobuf*)
-// save state in Gobuf; setjmp
-TEXT runtime·gosave(SB), NOSPLIT, $0-4
-	MOVL	buf+0(FP), AX		// gobuf
-	LEAL	buf+0(FP), BX		// caller's SP
-	MOVL	BX, gobuf_sp(AX)
-	MOVL	0(SP), BX		// caller's PC
-	MOVL	BX, gobuf_pc(AX)
-	MOVL	$0, gobuf_ret(AX)
-	// Assert ctxt is zero. See func save.
-	MOVL	gobuf_ctxt(AX), BX
-	TESTL	BX, BX
-	JZ	2(PC)
-	CALL	runtime·badctxt(SB)
-	get_tls(CX)
-	MOVL	g(CX), BX
-	MOVL	BX, gobuf_g(AX)
-	RET
-
 // void gogo(Gobuf*)
 // restore state from Gobuf; longjmp
 TEXT runtime·gogo(SB), NOSPLIT, $8-4
@@ -477,7 +458,7 @@ TEXT runtime·morestack_noctxt(SB),NOSPLIT,$0-0
 	JMP runtime·morestack(SB)
 
 // reflectcall: call a function with the given argument list
-// func call(argtype *_type, f *FuncVal, arg *byte, argsize, retoffset uint32).
+// func call(stackArgsType *_type, f *FuncVal, stackArgs *byte, stackArgsSize, stackRetOffset, frameSize uint32, regArgs *abi.RegArgs).
 // we don't have variable-sized frames, so we use a small number
 // of constant-sized-frame functions to encode a few bits of size in the pc.
 // Caution: ugly multiline assembly macros in your future!
@@ -489,8 +470,8 @@ TEXT runtime·morestack_noctxt(SB),NOSPLIT,$0-0
 	JMP	AX
 // Note: can't just "JMP NAME(SB)" - bad inlining results.
 
-TEXT ·reflectcall(SB), NOSPLIT, $0-20
-	MOVL	argsize+12(FP), CX
+TEXT ·reflectcall(SB), NOSPLIT, $0-28
+	MOVL	frameSize+20(FP), CX
 	DISPATCH(runtime·call16, 16)
 	DISPATCH(runtime·call32, 32)
 	DISPATCH(runtime·call64, 64)
@@ -522,11 +503,11 @@ TEXT ·reflectcall(SB), NOSPLIT, $0-20
 	JMP	AX
 
 #define CALLFN(NAME,MAXSIZE)			\
-TEXT NAME(SB), WRAPPER, $MAXSIZE-20;		\
+TEXT NAME(SB), WRAPPER, $MAXSIZE-28;		\
 	NO_LOCAL_POINTERS;			\
 	/* copy arguments to stack */		\
-	MOVL	argptr+8(FP), SI;		\
-	MOVL	argsize+12(FP), CX;		\
+	MOVL	stackArgs+8(FP), SI;		\
+	MOVL	stackArgsSize+12(FP), CX;		\
 	MOVL	SP, DI;				\
 	REP;MOVSB;				\
 	/* call function */			\
@@ -535,10 +516,10 @@ TEXT NAME(SB), WRAPPER, $MAXSIZE-20;		\
 	PCDATA  $PCDATA_StackMapIndex, $0;	\
 	CALL	AX;				\
 	/* copy return values back */		\
-	MOVL	argtype+0(FP), DX;		\
-	MOVL	argptr+8(FP), DI;		\
-	MOVL	argsize+12(FP), CX;		\
-	MOVL	retoffset+16(FP), BX;		\
+	MOVL	stackArgsType+0(FP), DX;		\
+	MOVL	stackArgs+8(FP), DI;		\
+	MOVL	stackArgsSize+12(FP), CX;		\
+	MOVL	stackRetOffset+16(FP), BX;		\
 	MOVL	SP, SI;				\
 	ADDL	BX, DI;				\
 	ADDL	BX, SI;				\
@@ -550,11 +531,12 @@ TEXT NAME(SB), WRAPPER, $MAXSIZE-20;		\
 // separate function so it can allocate stack space for the arguments
 // to reflectcallmove. It does not follow the Go ABI; it expects its
 // arguments in registers.
-TEXT callRet<>(SB), NOSPLIT, $16-0
+TEXT callRet<>(SB), NOSPLIT, $20-0
 	MOVL	DX, 0(SP)
 	MOVL	DI, 4(SP)
 	MOVL	SI, 8(SP)
 	MOVL	CX, 12(SP)
+	MOVL	$0, 16(SP)
 	CALL	runtime·reflectcallmove(SB)
 	RET
 
diff --git a/src/runtime/asm_amd64.s b/src/runtime/asm_amd64.s
index 4ac87089f2..05422c9699 100644
--- a/src/runtime/asm_amd64.s
+++ b/src/runtime/asm_amd64.s
@@ -84,9 +84,7 @@ GLOBL _rt0_amd64_lib_argc<>(SB),NOPTR, $8
 DATA _rt0_amd64_lib_argv<>(SB)/8, $0
 GLOBL _rt0_amd64_lib_argv<>(SB),NOPTR, $8
 
-// Defined as ABIInternal since it does not use the stack-based Go ABI (and
-// in addition there are no calls to this entry point from Go code).
-TEXT runtime·rt0_go<ABIInternal>(SB),NOSPLIT,$0
+TEXT runtime·rt0_go(SB),NOSPLIT,$0
 	// copy arguments forward on an even stack
 	MOVQ	DI, AX		// argc
 	MOVQ	SI, BX		// argv
@@ -256,26 +254,6 @@ TEXT runtime·asminit(SB),NOSPLIT,$0-0
  *  go-routine
  */
 
-// func gosave(buf *gobuf)
-// save state in Gobuf; setjmp
-TEXT runtime·gosave(SB), NOSPLIT, $0-8
-	MOVQ	buf+0(FP), AX		// gobuf
-	LEAQ	buf+0(FP), BX		// caller's SP
-	MOVQ	BX, gobuf_sp(AX)
-	MOVQ	0(SP), BX		// caller's PC
-	MOVQ	BX, gobuf_pc(AX)
-	MOVQ	$0, gobuf_ret(AX)
-	MOVQ	BP, gobuf_bp(AX)
-	// Assert ctxt is zero. See func save.
-	MOVQ	gobuf_ctxt(AX), BX
-	TESTQ	BX, BX
-	JZ	2(PC)
-	CALL	runtime·badctxt(SB)
-	get_tls(CX)
-	MOVQ	g(CX), BX
-	MOVQ	BX, gobuf_g(AX)
-	RET
-
 // func gogo(buf *gobuf)
 // restore state from Gobuf; longjmp
 TEXT runtime·gogo(SB), NOSPLIT, $16-8
@@ -284,6 +262,7 @@ TEXT runtime·gogo(SB), NOSPLIT, $16-8
 	MOVQ	0(DX), CX		// make sure g != nil
 	get_tls(CX)
 	MOVQ	DX, g(CX)
+	MOVQ	DX, R14		// set the g register
 	MOVQ	gobuf_sp(BX), SP	// restore SP
 	MOVQ	gobuf_ret(BX), AX
 	MOVQ	gobuf_ctxt(BX), DX
@@ -320,6 +299,7 @@ TEXT runtime·mcall(SB), NOSPLIT, $0-8
 	MOVQ	$runtime·badmcall(SB), AX
 	JMP	AX
 	MOVQ	SI, g(CX)	// g = m->g0
+	MOVQ	SI, R14	// set the g register
 	MOVQ	(g_sched+gobuf_sp)(SI), SP	// sp = m->g0->sched.sp
 	PUSHQ	AX
 	MOVQ	DI, DX
@@ -366,6 +346,7 @@ TEXT runtime·systemstack(SB), NOSPLIT, $0-8
 
 	// switch to g0
 	MOVQ	DX, g(CX)
+	MOVQ	DX, R14 // set the g register
 	MOVQ	(g_sched+gobuf_sp)(DX), BX
 	// make it look like mstart called systemstack on g0, to stop traceback
 	SUBQ	$8, BX
@@ -464,8 +445,77 @@ TEXT runtime·morestack_noctxt(SB),NOSPLIT,$0
 	MOVL	$0, DX
 	JMP	runtime·morestack(SB)
 
+// REFLECTCALL_USE_REGABI is not defined. It must be defined in conjunction with the
+// register constants in the internal/abi package.
+
+#ifdef REFLECTCALL_USE_REGABI
+// spillArgs stores return values from registers to a *internal/abi.RegArgs in R12.
+TEXT spillArgs<>(SB),NOSPLIT,$0-0
+	MOVQ AX, 0(R12)
+	MOVQ BX, 8(R12)
+	MOVQ CX, 16(R12)
+	MOVQ DI, 24(R12)
+	MOVQ SI, 32(R12)
+	MOVQ R8, 40(R12)
+	MOVQ R9, 48(R12)
+	MOVQ R10, 56(R12)
+	MOVQ R11, 64(R12)
+	MOVQ X0, 72(R12)
+	MOVQ X1, 80(R12)
+	MOVQ X2, 88(R12)
+	MOVQ X3, 96(R12)
+	MOVQ X4, 104(R12)
+	MOVQ X5, 112(R12)
+	MOVQ X6, 120(R12)
+	MOVQ X7, 128(R12)
+	MOVQ X8, 136(R12)
+	MOVQ X9, 144(R12)
+	MOVQ X10, 152(R12)
+	MOVQ X11, 160(R12)
+	MOVQ X12, 168(R12)
+	MOVQ X13, 176(R12)
+	MOVQ X14, 184(R12)
+	RET
+
+// unspillArgs loads args into registers from a *internal/abi.RegArgs in R12.
+TEXT unspillArgs<>(SB),NOSPLIT,$0-0
+	MOVQ 0(R12), AX
+	MOVQ 8(R12), BX
+	MOVQ 16(R12), CX
+	MOVQ 24(R12), DI
+	MOVQ 32(R12), SI
+	MOVQ 40(R12), R8
+	MOVQ 48(R12), R9
+	MOVQ 56(R12), R10
+	MOVQ 64(R12), R11
+	MOVQ 72(R12), X0
+	MOVQ 80(R12), X1
+	MOVQ 88(R12), X2
+	MOVQ 96(R12), X3
+	MOVQ 104(R12), X4
+	MOVQ 112(R12), X5
+	MOVQ 120(R12), X6
+	MOVQ 128(R12), X7
+	MOVQ 136(R12), X8
+	MOVQ 144(R12), X9
+	MOVQ 152(R12), X10
+	MOVQ 160(R12), X11
+	MOVQ 168(R12), X12
+	MOVQ 176(R12), X13
+	MOVQ 184(R12), X14
+	RET
+#else
+// spillArgs stores return values from registers to a pointer in R12.
+TEXT spillArgs<>(SB),NOSPLIT,$0-0
+	RET
+
+// unspillArgs loads args into registers from a pointer in R12.
+TEXT unspillArgs<>(SB),NOSPLIT,$0-0
+	RET
+#endif
+
 // reflectcall: call a function with the given argument list
-// func call(argtype *_type, f *FuncVal, arg *byte, argsize, retoffset uint32).
+// func call(stackArgsType *_type, f *FuncVal, stackArgs *byte, stackArgsSize, stackRetOffset, frameSize uint32, regArgs *abi.RegArgs).
 // we don't have variable-sized frames, so we use a small number
 // of constant-sized-frame functions to encode a few bits of size in the pc.
 // Caution: ugly multiline assembly macros in your future!
@@ -477,8 +527,8 @@ TEXT runtime·morestack_noctxt(SB),NOSPLIT,$0
 	JMP	AX
 // Note: can't just "JMP NAME(SB)" - bad inlining results.
 
-TEXT ·reflectcall<ABIInternal>(SB), NOSPLIT, $0-32
-	MOVLQZX argsize+24(FP), CX
+TEXT ·reflectcall<ABIInternal>(SB), NOSPLIT, $0-48
+	MOVLQZX frameSize+32(FP), CX
 	DISPATCH(runtime·call16, 16)
 	DISPATCH(runtime·call32, 32)
 	DISPATCH(runtime·call64, 64)
@@ -510,23 +560,28 @@ TEXT ·reflectcall<ABIInternal>(SB), NOSPLIT, $0-32
 	JMP	AX
 
 #define CALLFN(NAME,MAXSIZE)			\
-TEXT NAME(SB), WRAPPER, $MAXSIZE-32;		\
+TEXT NAME(SB), WRAPPER, $MAXSIZE-48;		\
 	NO_LOCAL_POINTERS;			\
 	/* copy arguments to stack */		\
-	MOVQ	argptr+16(FP), SI;		\
-	MOVLQZX argsize+24(FP), CX;		\
+	MOVQ	stackArgs+16(FP), SI;		\
+	MOVLQZX stackArgsSize+24(FP), CX;		\
 	MOVQ	SP, DI;				\
 	REP;MOVSB;				\
+	/* set up argument registers */		\
+	MOVQ    regArgs+40(FP), R12;		\
+	CALL    unspillArgs<>(SB);		\
 	/* call function */			\
 	MOVQ	f+8(FP), DX;			\
 	PCDATA  $PCDATA_StackMapIndex, $0;	\
-	MOVQ	(DX), AX;			\
-	CALL	AX;				\
-	/* copy return values back */		\
-	MOVQ	argtype+0(FP), DX;		\
-	MOVQ	argptr+16(FP), DI;		\
-	MOVLQZX	argsize+24(FP), CX;		\
-	MOVLQZX	retoffset+28(FP), BX;		\
+	MOVQ	(DX), R12;			\
+	CALL	R12;				\
+	/* copy register return values back */		\
+	MOVQ    regArgs+40(FP), R12;		\
+	CALL    spillArgs<>(SB);		\
+	MOVLQZX	stackArgsSize+24(FP), CX;		\
+	MOVLQZX	stackRetOffset+28(FP), BX;		\
+	MOVQ	stackArgs+16(FP), DI;		\
+	MOVQ	stackArgsType+0(FP), DX;		\
 	MOVQ	SP, SI;				\
 	ADDQ	BX, DI;				\
 	ADDQ	BX, SI;				\
@@ -538,12 +593,13 @@ TEXT NAME(SB), WRAPPER, $MAXSIZE-32;		\
 // separate function so it can allocate stack space for the arguments
 // to reflectcallmove. It does not follow the Go ABI; it expects its
 // arguments in registers.
-TEXT callRet<>(SB), NOSPLIT, $32-0
+TEXT callRet<>(SB), NOSPLIT, $40-0
 	NO_LOCAL_POINTERS
 	MOVQ	DX, 0(SP)
 	MOVQ	DI, 8(SP)
 	MOVQ	SI, 16(SP)
 	MOVQ	CX, 24(SP)
+	MOVQ	R12, 32(SP)
 	CALL	runtime·reflectcallmove(SB)
 	RET
 
@@ -604,18 +660,20 @@ TEXT runtime·jmpdefer(SB), NOSPLIT, $0-16
 	MOVQ	0(DX), BX
 	JMP	BX	// but first run the deferred function
 
-// Save state of caller into g->sched. Smashes R8, R9.
+// Save state of caller into g->sched. Smashes R9.
 TEXT gosave<>(SB),NOSPLIT,$0
-	get_tls(R8)
-	MOVQ	g(R8), R8
+#ifndef GOEXPERIMENT_REGABI
+	get_tls(R14)
+	MOVQ	g(R14), R14
+#endif
 	MOVQ	0(SP), R9
-	MOVQ	R9, (g_sched+gobuf_pc)(R8)
+	MOVQ	R9, (g_sched+gobuf_pc)(R14)
 	LEAQ	8(SP), R9
-	MOVQ	R9, (g_sched+gobuf_sp)(R8)
-	MOVQ	$0, (g_sched+gobuf_ret)(R8)
-	MOVQ	BP, (g_sched+gobuf_bp)(R8)
+	MOVQ	R9, (g_sched+gobuf_sp)(R14)
+	MOVQ	$0, (g_sched+gobuf_ret)(R14)
+	MOVQ	BP, (g_sched+gobuf_bp)(R14)
 	// Assert ctxt is zero. See func save.
-	MOVQ	(g_sched+gobuf_ctxt)(R8), R9
+	MOVQ	(g_sched+gobuf_ctxt)(R14), R9
 	TESTQ	R9, R9
 	JZ	2(PC)
 	CALL	runtime·badctxt(SB)
@@ -846,6 +904,7 @@ settls:
 TEXT setg_gcc<>(SB),NOSPLIT,$0
 	get_tls(AX)
 	MOVQ	DI, g(AX)
+	MOVQ	DI, R14 // set the g register
 	RET
 
 TEXT runtime·abort(SB),NOSPLIT,$0-0
@@ -1382,6 +1441,18 @@ TEXT runtime·addmoduledata(SB),NOSPLIT,$0-0
 	POPQ	R15
 	RET
 
+// Initialize special registers then jump to sigpanic.
+// This function is injected from the signal handler for panicking
+// signals. It is quite painful to set X15 in the signal context,
+// so we do it here.
+TEXT ·sigpanic0<ABIInternal>(SB),NOSPLIT,$0-0
+#ifdef GOEXPERIMENT_REGABI
+	get_tls(R14)
+	MOVQ	g(R14), R14
+	XORPS	X15, X15
+#endif
+	JMP	·sigpanic<ABIInternal>(SB)
+
 // gcWriteBarrier performs a heap pointer write and informs the GC.
 //
 // gcWriteBarrier does NOT follow the Go ABI. It takes two arguments:
@@ -1390,24 +1461,28 @@ TEXT runtime·addmoduledata(SB),NOSPLIT,$0-0
 // It clobbers FLAGS. It does not clobber any general-purpose registers,
 // but may clobber others (e.g., SSE registers).
 // Defined as ABIInternal since it does not use the stack-based Go ABI.
-TEXT runtime·gcWriteBarrier<ABIInternal>(SB),NOSPLIT,$120
+TEXT runtime·gcWriteBarrier<ABIInternal>(SB),NOSPLIT,$112
 	// Save the registers clobbered by the fast path. This is slightly
 	// faster than having the caller spill these.
-	MOVQ	R14, 104(SP)
-	MOVQ	R13, 112(SP)
+	MOVQ	R12, 96(SP)
+	MOVQ	R13, 104(SP)
 	// TODO: Consider passing g.m.p in as an argument so they can be shared
 	// across a sequence of write barriers.
+#ifdef GOEXPERIMENT_REGABI
+	MOVQ	g_m(R14), R13
+#else
 	get_tls(R13)
 	MOVQ	g(R13), R13
 	MOVQ	g_m(R13), R13
+#endif
 	MOVQ	m_p(R13), R13
-	MOVQ	(p_wbBuf+wbBuf_next)(R13), R14
+	MOVQ	(p_wbBuf+wbBuf_next)(R13), R12
 	// Increment wbBuf.next position.
-	LEAQ	16(R14), R14
-	MOVQ	R14, (p_wbBuf+wbBuf_next)(R13)
-	CMPQ	R14, (p_wbBuf+wbBuf_end)(R13)
+	LEAQ	16(R12), R12
+	MOVQ	R12, (p_wbBuf+wbBuf_next)(R13)
+	CMPQ	R12, (p_wbBuf+wbBuf_end)(R13)
 	// Record the write.
-	MOVQ	AX, -16(R14)	// Record value
+	MOVQ	AX, -16(R12)	// Record value
 	// Note: This turns bad pointer writes into bad
 	// pointer reads, which could be confusing. We could avoid
 	// reading from obviously bad pointers, which would
@@ -1415,12 +1490,12 @@ TEXT runtime·gcWriteBarrier<ABIInternal>(SB),NOSPLIT,$120
 	// patch this up in the signal handler, or use XCHG to
 	// combine the read and the write.
 	MOVQ	(DI), R13
-	MOVQ	R13, -8(R14)	// Record *slot
+	MOVQ	R13, -8(R12)	// Record *slot
 	// Is the buffer full? (flags set in CMPQ above)
 	JEQ	flush
 ret:
-	MOVQ	104(SP), R14
-	MOVQ	112(SP), R13
+	MOVQ	96(SP), R12
+	MOVQ	104(SP), R13
 	// Do the write.
 	MOVQ	AX, (DI)
 	RET
@@ -1450,10 +1525,10 @@ flush:
 	MOVQ	R9, 64(SP)
 	MOVQ	R10, 72(SP)
 	MOVQ	R11, 80(SP)
-	MOVQ	R12, 88(SP)
+	// R12 already saved
 	// R13 already saved
-	// R14 already saved
-	MOVQ	R15, 96(SP)
+	// R14 is g
+	MOVQ	R15, 88(SP)
 
 	// This takes arguments DI and AX
 	CALL	runtime·wbBufFlush(SB)
@@ -1469,8 +1544,7 @@ flush:
 	MOVQ	64(SP), R9
 	MOVQ	72(SP), R10
 	MOVQ	80(SP), R11
-	MOVQ	88(SP), R12
-	MOVQ	96(SP), R15
+	MOVQ	88(SP), R15
 	JMP	ret
 
 // gcWriteBarrierCX is gcWriteBarrier, but with args in DI and CX.
@@ -1732,67 +1806,67 @@ TEXT runtime·debugCallPanicked(SB),NOSPLIT,$16-16
 TEXT runtime·panicIndex<ABIInternal>(SB),NOSPLIT,$0-16
 	MOVQ	AX, x+0(FP)
 	MOVQ	CX, y+8(FP)
-	JMP	runtime·goPanicIndex(SB)
+	JMP	runtime·goPanicIndex<ABIInternal>(SB)
 TEXT runtime·panicIndexU<ABIInternal>(SB),NOSPLIT,$0-16
 	MOVQ	AX, x+0(FP)
 	MOVQ	CX, y+8(FP)
-	JMP	runtime·goPanicIndexU(SB)
+	JMP	runtime·goPanicIndexU<ABIInternal>(SB)
 TEXT runtime·panicSliceAlen<ABIInternal>(SB),NOSPLIT,$0-16
 	MOVQ	CX, x+0(FP)
 	MOVQ	DX, y+8(FP)
-	JMP	runtime·goPanicSliceAlen(SB)
+	JMP	runtime·goPanicSliceAlen<ABIInternal>(SB)
 TEXT runtime·panicSliceAlenU<ABIInternal>(SB),NOSPLIT,$0-16
 	MOVQ	CX, x+0(FP)
 	MOVQ	DX, y+8(FP)
-	JMP	runtime·goPanicSliceAlenU(SB)
+	JMP	runtime·goPanicSliceAlenU<ABIInternal>(SB)
 TEXT runtime·panicSliceAcap<ABIInternal>(SB),NOSPLIT,$0-16
 	MOVQ	CX, x+0(FP)
 	MOVQ	DX, y+8(FP)
-	JMP	runtime·goPanicSliceAcap(SB)
+	JMP	runtime·goPanicSliceAcap<ABIInternal>(SB)
 TEXT runtime·panicSliceAcapU<ABIInternal>(SB),NOSPLIT,$0-16
 	MOVQ	CX, x+0(FP)
 	MOVQ	DX, y+8(FP)
-	JMP	runtime·goPanicSliceAcapU(SB)
+	JMP	runtime·goPanicSliceAcapU<ABIInternal>(SB)
 TEXT runtime·panicSliceB<ABIInternal>(SB),NOSPLIT,$0-16
 	MOVQ	AX, x+0(FP)
 	MOVQ	CX, y+8(FP)
-	JMP	runtime·goPanicSliceB(SB)
+	JMP	runtime·goPanicSliceB<ABIInternal>(SB)
 TEXT runtime·panicSliceBU<ABIInternal>(SB),NOSPLIT,$0-16
 	MOVQ	AX, x+0(FP)
 	MOVQ	CX, y+8(FP)
-	JMP	runtime·goPanicSliceBU(SB)
+	JMP	runtime·goPanicSliceBU<ABIInternal>(SB)
 TEXT runtime·panicSlice3Alen<ABIInternal>(SB),NOSPLIT,$0-16
 	MOVQ	DX, x+0(FP)
 	MOVQ	BX, y+8(FP)
-	JMP	runtime·goPanicSlice3Alen(SB)
+	JMP	runtime·goPanicSlice3Alen<ABIInternal>(SB)
 TEXT runtime·panicSlice3AlenU<ABIInternal>(SB),NOSPLIT,$0-16
 	MOVQ	DX, x+0(FP)
 	MOVQ	BX, y+8(FP)
-	JMP	runtime·goPanicSlice3AlenU(SB)
+	JMP	runtime·goPanicSlice3AlenU<ABIInternal>(SB)
 TEXT runtime·panicSlice3Acap<ABIInternal>(SB),NOSPLIT,$0-16
 	MOVQ	DX, x+0(FP)
 	MOVQ	BX, y+8(FP)
-	JMP	runtime·goPanicSlice3Acap(SB)
+	JMP	runtime·goPanicSlice3Acap<ABIInternal>(SB)
 TEXT runtime·panicSlice3AcapU<ABIInternal>(SB),NOSPLIT,$0-16
 	MOVQ	DX, x+0(FP)
 	MOVQ	BX, y+8(FP)
-	JMP	runtime·goPanicSlice3AcapU(SB)
+	JMP	runtime·goPanicSlice3AcapU<ABIInternal>(SB)
 TEXT runtime·panicSlice3B<ABIInternal>(SB),NOSPLIT,$0-16
 	MOVQ	CX, x+0(FP)
 	MOVQ	DX, y+8(FP)
-	JMP	runtime·goPanicSlice3B(SB)
+	JMP	runtime·goPanicSlice3B<ABIInternal>(SB)
 TEXT runtime·panicSlice3BU<ABIInternal>(SB),NOSPLIT,$0-16
 	MOVQ	CX, x+0(FP)
 	MOVQ	DX, y+8(FP)
-	JMP	runtime·goPanicSlice3BU(SB)
+	JMP	runtime·goPanicSlice3BU<ABIInternal>(SB)
 TEXT runtime·panicSlice3C<ABIInternal>(SB),NOSPLIT,$0-16
 	MOVQ	AX, x+0(FP)
 	MOVQ	CX, y+8(FP)
-	JMP	runtime·goPanicSlice3C(SB)
+	JMP	runtime·goPanicSlice3C<ABIInternal>(SB)
 TEXT runtime·panicSlice3CU<ABIInternal>(SB),NOSPLIT,$0-16
 	MOVQ	AX, x+0(FP)
 	MOVQ	CX, y+8(FP)
-	JMP	runtime·goPanicSlice3CU(SB)
+	JMP	runtime·goPanicSlice3CU<ABIInternal>(SB)
 
 #ifdef GOOS_android
 // Use the free TLS_SLOT_APP slot #2 on Android Q.
diff --git a/src/runtime/asm_arm.s b/src/runtime/asm_arm.s
index c54b4eb006..23619b1408 100644
--- a/src/runtime/asm_arm.s
+++ b/src/runtime/asm_arm.s
@@ -206,23 +206,6 @@ TEXT runtime·asminit(SB),NOSPLIT,$0-0
  *  go-routine
  */
 
-// void gosave(Gobuf*)
-// save state in Gobuf; setjmp
-TEXT runtime·gosave(SB),NOSPLIT|NOFRAME,$0-4
-	MOVW	buf+0(FP), R0
-	MOVW	R13, gobuf_sp(R0)
-	MOVW	LR, gobuf_pc(R0)
-	MOVW	g, gobuf_g(R0)
-	MOVW	$0, R11
-	MOVW	R11, gobuf_lr(R0)
-	MOVW	R11, gobuf_ret(R0)
-	// Assert ctxt is zero. See func save.
-	MOVW	gobuf_ctxt(R0), R0
-	CMP	R0, R11
-	B.EQ	2(PC)
-	CALL	runtime·badctxt(SB)
-	RET
-
 // void gogo(Gobuf*)
 // restore state from Gobuf; longjmp
 TEXT runtime·gogo(SB),NOSPLIT,$8-4
@@ -421,7 +404,7 @@ TEXT runtime·morestack_noctxt(SB),NOSPLIT|NOFRAME,$0-0
 	B runtime·morestack(SB)
 
 // reflectcall: call a function with the given argument list
-// func call(argtype *_type, f *FuncVal, arg *byte, argsize, retoffset uint32).
+// func call(stackArgsType *_type, f *FuncVal, stackArgs *byte, stackArgsSize, stackRetOffset, frameSize uint32, regArgs *abi.RegArgs).
 // we don't have variable-sized frames, so we use a small number
 // of constant-sized-frame functions to encode a few bits of size in the pc.
 // Caution: ugly multiline assembly macros in your future!
@@ -432,8 +415,8 @@ TEXT runtime·morestack_noctxt(SB),NOSPLIT|NOFRAME,$0-0
 	MOVW	$NAME(SB), R1;		\
 	B	(R1)
 
-TEXT ·reflectcall(SB),NOSPLIT|NOFRAME,$0-20
-	MOVW	argsize+12(FP), R0
+TEXT ·reflectcall(SB),NOSPLIT|NOFRAME,$0-28
+	MOVW	frameSize+20(FP), R0
 	DISPATCH(runtime·call16, 16)
 	DISPATCH(runtime·call32, 32)
 	DISPATCH(runtime·call64, 64)
@@ -465,11 +448,11 @@ TEXT ·reflectcall(SB),NOSPLIT|NOFRAME,$0-20
 	B	(R1)
 
 #define CALLFN(NAME,MAXSIZE)			\
-TEXT NAME(SB), WRAPPER, $MAXSIZE-20;		\
+TEXT NAME(SB), WRAPPER, $MAXSIZE-28;		\
 	NO_LOCAL_POINTERS;			\
 	/* copy arguments to stack */		\
-	MOVW	argptr+8(FP), R0;		\
-	MOVW	argsize+12(FP), R2;		\
+	MOVW	stackArgs+8(FP), R0;		\
+	MOVW	stackArgsSize+12(FP), R2;		\
 	ADD	$4, R13, R1;			\
 	CMP	$0, R2;				\
 	B.EQ	5(PC);				\
@@ -483,10 +466,10 @@ TEXT NAME(SB), WRAPPER, $MAXSIZE-20;		\
 	PCDATA  $PCDATA_StackMapIndex, $0;	\
 	BL	(R0);				\
 	/* copy return values back */		\
-	MOVW	argtype+0(FP), R4;		\
-	MOVW	argptr+8(FP), R0;		\
-	MOVW	argsize+12(FP), R2;		\
-	MOVW	retoffset+16(FP), R3;		\
+	MOVW	stackArgsType+0(FP), R4;		\
+	MOVW	stackArgs+8(FP), R0;		\
+	MOVW	stackArgsSize+12(FP), R2;		\
+	MOVW	stackArgsRetOffset+16(FP), R3;		\
 	ADD	$4, R13, R1;			\
 	ADD	R3, R1;				\
 	ADD	R3, R0;				\
@@ -498,11 +481,13 @@ TEXT NAME(SB), WRAPPER, $MAXSIZE-20;		\
 // separate function so it can allocate stack space for the arguments
 // to reflectcallmove. It does not follow the Go ABI; it expects its
 // arguments in registers.
-TEXT callRet<>(SB), NOSPLIT, $16-0
+TEXT callRet<>(SB), NOSPLIT, $20-0
 	MOVW	R4, 4(R13)
 	MOVW	R0, 8(R13)
 	MOVW	R1, 12(R13)
 	MOVW	R2, 16(R13)
+	MOVW	$0, R7
+	MOVW	R7, 20(R13)
 	BL	runtime·reflectcallmove(SB)
 	RET
 
diff --git a/src/runtime/asm_arm64.s b/src/runtime/asm_arm64.s
index a09172f0c9..0ab92be1e4 100644
--- a/src/runtime/asm_arm64.s
+++ b/src/runtime/asm_arm64.s
@@ -113,23 +113,6 @@ TEXT runtime·asminit(SB),NOSPLIT|NOFRAME,$0-0
  *  go-routine
  */
 
-// void gosave(Gobuf*)
-// save state in Gobuf; setjmp
-TEXT runtime·gosave(SB), NOSPLIT|NOFRAME, $0-8
-	MOVD	buf+0(FP), R3
-	MOVD	RSP, R0
-	MOVD	R0, gobuf_sp(R3)
-	MOVD	R29, gobuf_bp(R3)
-	MOVD	LR, gobuf_pc(R3)
-	MOVD	g, gobuf_g(R3)
-	MOVD	ZR, gobuf_lr(R3)
-	MOVD	ZR, gobuf_ret(R3)
-	// Assert ctxt is zero. See func save.
-	MOVD	gobuf_ctxt(R3), R0
-	CBZ	R0, 2(PC)
-	CALL	runtime·badctxt(SB)
-	RET
-
 // void gogo(Gobuf*)
 // restore state from Gobuf; longjmp
 TEXT runtime·gogo(SB), NOSPLIT, $24-8
@@ -329,7 +312,7 @@ TEXT runtime·morestack_noctxt(SB),NOSPLIT|NOFRAME,$0-0
 	B runtime·morestack(SB)
 
 // reflectcall: call a function with the given argument list
-// func call(argtype *_type, f *FuncVal, arg *byte, argsize, retoffset uint32).
+// func call(stackArgsType *_type, f *FuncVal, stackArgs *byte, stackArgsSize, stackRetOffset, frameSize uint32, regArgs *abi.RegArgs).
 // we don't have variable-sized frames, so we use a small number
 // of constant-sized-frame functions to encode a few bits of size in the pc.
 // Caution: ugly multiline assembly macros in your future!
@@ -342,8 +325,8 @@ TEXT runtime·morestack_noctxt(SB),NOSPLIT|NOFRAME,$0-0
 	B	(R27)
 // Note: can't just "B NAME(SB)" - bad inlining results.
 
-TEXT ·reflectcall(SB), NOSPLIT|NOFRAME, $0-32
-	MOVWU argsize+24(FP), R16
+TEXT ·reflectcall(SB), NOSPLIT|NOFRAME, $0-48
+	MOVWU	frameSize+32(FP), R16
 	DISPATCH(runtime·call16, 16)
 	DISPATCH(runtime·call32, 32)
 	DISPATCH(runtime·call64, 64)
@@ -375,11 +358,11 @@ TEXT ·reflectcall(SB), NOSPLIT|NOFRAME, $0-32
 	B	(R0)
 
 #define CALLFN(NAME,MAXSIZE)			\
-TEXT NAME(SB), WRAPPER, $MAXSIZE-24;		\
+TEXT NAME(SB), WRAPPER, $MAXSIZE-48;		\
 	NO_LOCAL_POINTERS;			\
 	/* copy arguments to stack */		\
-	MOVD	arg+16(FP), R3;			\
-	MOVWU	argsize+24(FP), R4;		\
+	MOVD	stackArgs+16(FP), R3;			\
+	MOVWU	stackArgsSize+24(FP), R4;		\
 	ADD	$8, RSP, R5;			\
 	BIC	$0xf, R4, R6;			\
 	CBZ	R6, 6(PC);			\
@@ -405,10 +388,10 @@ TEXT NAME(SB), WRAPPER, $MAXSIZE-24;		\
 	PCDATA  $PCDATA_StackMapIndex, $0;	\
 	BL	(R0);				\
 	/* copy return values back */		\
-	MOVD	argtype+0(FP), R7;		\
-	MOVD	arg+16(FP), R3;			\
-	MOVWU	n+24(FP), R4;			\
-	MOVWU	retoffset+28(FP), R6;		\
+	MOVD	stackArgsType+0(FP), R7;		\
+	MOVD	stackArgs+16(FP), R3;			\
+	MOVWU	stackArgsSize+24(FP), R4;			\
+	MOVWU	stackRetOffset+28(FP), R6;		\
 	ADD	$8, RSP, R5;			\
 	ADD	R6, R5; 			\
 	ADD	R6, R3;				\
@@ -420,11 +403,12 @@ TEXT NAME(SB), WRAPPER, $MAXSIZE-24;		\
 // separate function so it can allocate stack space for the arguments
 // to reflectcallmove. It does not follow the Go ABI; it expects its
 // arguments in registers.
-TEXT callRet<>(SB), NOSPLIT, $40-0
+TEXT callRet<>(SB), NOSPLIT, $48-0
 	MOVD	R7, 8(RSP)
 	MOVD	R3, 16(RSP)
 	MOVD	R5, 24(RSP)
 	MOVD	R4, 32(RSP)
+	MOVD	$0, 40(RSP)
 	BL	runtime·reflectcallmove(SB)
 	RET
 
diff --git a/src/runtime/asm_mips64x.s b/src/runtime/asm_mips64x.s
index 19781f7885..694950663a 100644
--- a/src/runtime/asm_mips64x.s
+++ b/src/runtime/asm_mips64x.s
@@ -89,21 +89,6 @@ TEXT runtime·asminit(SB),NOSPLIT|NOFRAME,$0-0
  *  go-routine
  */
 
-// void gosave(Gobuf*)
-// save state in Gobuf; setjmp
-TEXT runtime·gosave(SB), NOSPLIT|NOFRAME, $0-8
-	MOVV	buf+0(FP), R1
-	MOVV	R29, gobuf_sp(R1)
-	MOVV	R31, gobuf_pc(R1)
-	MOVV	g, gobuf_g(R1)
-	MOVV	R0, gobuf_lr(R1)
-	MOVV	R0, gobuf_ret(R1)
-	// Assert ctxt is zero. See func save.
-	MOVV	gobuf_ctxt(R1), R1
-	BEQ	R1, 2(PC)
-	JAL	runtime·badctxt(SB)
-	RET
-
 // void gogo(Gobuf*)
 // restore state from Gobuf; longjmp
 TEXT runtime·gogo(SB), NOSPLIT, $16-8
@@ -279,7 +264,7 @@ TEXT runtime·morestack_noctxt(SB),NOSPLIT|NOFRAME,$0-0
 	JMP	runtime·morestack(SB)
 
 // reflectcall: call a function with the given argument list
-// func call(argtype *_type, f *FuncVal, arg *byte, argsize, retoffset uint32).
+// func call(stackArgsType *_type, f *FuncVal, stackArgs *byte, stackArgsSize, stackRetOffset, frameSize uint32, regArgs *abi.RegArgs).
 // we don't have variable-sized frames, so we use a small number
 // of constant-sized-frame functions to encode a few bits of size in the pc.
 // Caution: ugly multiline assembly macros in your future!
@@ -292,8 +277,8 @@ TEXT runtime·morestack_noctxt(SB),NOSPLIT|NOFRAME,$0-0
 	JMP	(R4)
 // Note: can't just "BR NAME(SB)" - bad inlining results.
 
-TEXT ·reflectcall(SB), NOSPLIT|NOFRAME, $0-32
-	MOVWU argsize+24(FP), R1
+TEXT ·reflectcall(SB), NOSPLIT|NOFRAME, $0-48
+	MOVWU	frameSize+32(FP), R1
 	DISPATCH(runtime·call16, 16)
 	DISPATCH(runtime·call32, 32)
 	DISPATCH(runtime·call64, 64)
@@ -325,11 +310,11 @@ TEXT ·reflectcall(SB), NOSPLIT|NOFRAME, $0-32
 	JMP	(R4)
 
 #define CALLFN(NAME,MAXSIZE)			\
-TEXT NAME(SB), WRAPPER, $MAXSIZE-24;		\
+TEXT NAME(SB), WRAPPER, $MAXSIZE-48;		\
 	NO_LOCAL_POINTERS;			\
 	/* copy arguments to stack */		\
-	MOVV	arg+16(FP), R1;			\
-	MOVWU	argsize+24(FP), R2;			\
+	MOVV	stackArgs+16(FP), R1;			\
+	MOVWU	stackArgsSize+24(FP), R2;			\
 	MOVV	R29, R3;				\
 	ADDV	$8, R3;			\
 	ADDV	R3, R2;				\
@@ -345,10 +330,10 @@ TEXT NAME(SB), WRAPPER, $MAXSIZE-24;		\
 	PCDATA  $PCDATA_StackMapIndex, $0;	\
 	JAL	(R4);				\
 	/* copy return values back */		\
-	MOVV	argtype+0(FP), R5;		\
-	MOVV	arg+16(FP), R1;			\
-	MOVWU	n+24(FP), R2;			\
-	MOVWU	retoffset+28(FP), R4;		\
+	MOVV	stackArgsType+0(FP), R5;		\
+	MOVV	stackArgs+16(FP), R1;			\
+	MOVWU	stackArgsSize+24(FP), R2;			\
+	MOVWU	stackRetOffset+28(FP), R4;		\
 	ADDV	$8, R29, R3;				\
 	ADDV	R4, R3; 			\
 	ADDV	R4, R1;				\
@@ -360,11 +345,12 @@ TEXT NAME(SB), WRAPPER, $MAXSIZE-24;		\
 // separate function so it can allocate stack space for the arguments
 // to reflectcallmove. It does not follow the Go ABI; it expects its
 // arguments in registers.
-TEXT callRet<>(SB), NOSPLIT, $32-0
+TEXT callRet<>(SB), NOSPLIT, $40-0
 	MOVV	R5, 8(R29)
 	MOVV	R1, 16(R29)
 	MOVV	R3, 24(R29)
 	MOVV	R2, 32(R29)
+	MOVV	$0, 40(R29)
 	JAL	runtime·reflectcallmove(SB)
 	RET
 
diff --git a/src/runtime/asm_mipsx.s b/src/runtime/asm_mipsx.s
index ee87d81436..8e5753d255 100644
--- a/src/runtime/asm_mipsx.s
+++ b/src/runtime/asm_mipsx.s
@@ -90,21 +90,6 @@ TEXT runtime·asminit(SB),NOSPLIT,$0-0
  *  go-routine
  */
 
-// void gosave(Gobuf*)
-// save state in Gobuf; setjmp
-TEXT runtime·gosave(SB),NOSPLIT|NOFRAME,$0-4
-	MOVW	buf+0(FP), R1
-	MOVW	R29, gobuf_sp(R1)
-	MOVW	R31, gobuf_pc(R1)
-	MOVW	g, gobuf_g(R1)
-	MOVW	R0, gobuf_lr(R1)
-	MOVW	R0, gobuf_ret(R1)
-	// Assert ctxt is zero. See func save.
-	MOVW	gobuf_ctxt(R1), R1
-	BEQ	R1, 2(PC)
-	JAL	runtime·badctxt(SB)
-	RET
-
 // void gogo(Gobuf*)
 // restore state from Gobuf; longjmp
 TEXT runtime·gogo(SB),NOSPLIT,$8-4
@@ -280,7 +265,7 @@ TEXT runtime·morestack_noctxt(SB),NOSPLIT,$0-0
 	JMP	runtime·morestack(SB)
 
 // reflectcall: call a function with the given argument list
-// func call(argtype *_type, f *FuncVal, arg *byte, argsize, retoffset uint32).
+// func call(stackArgsType *_type, f *FuncVal, stackArgs *byte, stackArgsSize, stackRetOffset, frameSize uint32, regArgs *abi.RegArgs).
 // we don't have variable-sized frames, so we use a small number
 // of constant-sized-frame functions to encode a few bits of size in the pc.
 
@@ -291,8 +276,8 @@ TEXT runtime·morestack_noctxt(SB),NOSPLIT,$0-0
 	MOVW	$NAME(SB), R4;	\
 	JMP	(R4)
 
-TEXT ·reflectcall(SB),NOSPLIT|NOFRAME,$0-20
-	MOVW	argsize+12(FP), R1
+TEXT ·reflectcall(SB),NOSPLIT|NOFRAME,$0-28
+	MOVW	frameSize+20(FP), R1
 
 	DISPATCH(runtime·call16, 16)
 	DISPATCH(runtime·call32, 32)
@@ -325,11 +310,11 @@ TEXT ·reflectcall(SB),NOSPLIT|NOFRAME,$0-20
 	JMP	(R4)
 
 #define CALLFN(NAME,MAXSIZE)	\
-TEXT NAME(SB),WRAPPER,$MAXSIZE-20;	\
+TEXT NAME(SB),WRAPPER,$MAXSIZE-28;	\
 	NO_LOCAL_POINTERS;	\
 	/* copy arguments to stack */		\
-	MOVW	arg+8(FP), R1;	\
-	MOVW	argsize+12(FP), R2;	\
+	MOVW	stackArgs+8(FP), R1;	\
+	MOVW	stackArgsSize+12(FP), R2;	\
 	MOVW	R29, R3;	\
 	ADDU	$4, R3;	\
 	ADDU	R3, R2;	\
@@ -345,10 +330,10 @@ TEXT NAME(SB),WRAPPER,$MAXSIZE-20;	\
 	PCDATA	$PCDATA_StackMapIndex, $0;	\
 	JAL	(R4);	\
 	/* copy return values back */		\
-	MOVW	argtype+0(FP), R5;	\
-	MOVW	arg+8(FP), R1;	\
-	MOVW	n+12(FP), R2;	\
-	MOVW	retoffset+16(FP), R4;	\
+	MOVW	stackArgsType+0(FP), R5;	\
+	MOVW	stackArgs+8(FP), R1;	\
+	MOVW	stackArgsSize+12(FP), R2;	\
+	MOVW	stackRetOffset+16(FP), R4;	\
 	ADDU	$4, R29, R3;	\
 	ADDU	R4, R3;	\
 	ADDU	R4, R1;	\
@@ -360,11 +345,12 @@ TEXT NAME(SB),WRAPPER,$MAXSIZE-20;	\
 // separate function so it can allocate stack space for the arguments
 // to reflectcallmove. It does not follow the Go ABI; it expects its
 // arguments in registers.
-TEXT callRet<>(SB), NOSPLIT, $16-0
+TEXT callRet<>(SB), NOSPLIT, $20-0
 	MOVW	R5, 4(R29)
 	MOVW	R1, 8(R29)
 	MOVW	R3, 12(R29)
 	MOVW	R2, 16(R29)
+	MOVW    $0, 20(R29)
 	JAL	runtime·reflectcallmove(SB)
 	RET
 
diff --git a/src/runtime/asm_ppc64x.s b/src/runtime/asm_ppc64x.s
index dc34c0e4c8..834023cce1 100644
--- a/src/runtime/asm_ppc64x.s
+++ b/src/runtime/asm_ppc64x.s
@@ -128,23 +128,6 @@ TEXT runtime·reginit(SB),NOSPLIT|NOFRAME,$0-0
  *  go-routine
  */
 
-// void gosave(Gobuf*)
-// save state in Gobuf; setjmp
-TEXT runtime·gosave(SB), NOSPLIT|NOFRAME, $0-8
-	MOVD	buf+0(FP), R3
-	MOVD	R1, gobuf_sp(R3)
-	MOVD	LR, R31
-	MOVD	R31, gobuf_pc(R3)
-	MOVD	g, gobuf_g(R3)
-	MOVD	R0, gobuf_lr(R3)
-	MOVD	R0, gobuf_ret(R3)
-	// Assert ctxt is zero. See func save.
-	MOVD	gobuf_ctxt(R3), R3
-	CMP	R0, R3
-	BEQ	2(PC)
-	BL	runtime·badctxt(SB)
-	RET
-
 // void gogo(Gobuf*)
 // restore state from Gobuf; longjmp
 TEXT runtime·gogo(SB), NOSPLIT, $16-8
@@ -356,7 +339,7 @@ TEXT runtime·morestack_noctxt(SB),NOSPLIT|NOFRAME,$0-0
 	BR	runtime·morestack(SB)
 
 // reflectcall: call a function with the given argument list
-// func call(argtype *_type, f *FuncVal, arg *byte, argsize, retoffset uint32).
+// func call(stackArgsType *_type, f *FuncVal, stackArgs *byte, stackArgsSize, stackRetOffset, frameSize uint32, regArgs *abi.RegArgs).
 // we don't have variable-sized frames, so we use a small number
 // of constant-sized-frame functions to encode a few bits of size in the pc.
 // Caution: ugly multiline assembly macros in your future!
@@ -370,8 +353,8 @@ TEXT runtime·morestack_noctxt(SB),NOSPLIT|NOFRAME,$0-0
 	BR	(CTR)
 // Note: can't just "BR NAME(SB)" - bad inlining results.
 
-TEXT ·reflectcall(SB), NOSPLIT|NOFRAME, $0-32
-	MOVWZ argsize+24(FP), R3
+TEXT ·reflectcall(SB), NOSPLIT|NOFRAME, $0-48
+	MOVWZ	frameSize+32(FP), R3
 	DISPATCH(runtime·call16, 16)
 	DISPATCH(runtime·call32, 32)
 	DISPATCH(runtime·call64, 64)
@@ -404,11 +387,11 @@ TEXT ·reflectcall(SB), NOSPLIT|NOFRAME, $0-32
 	BR	(CTR)
 
 #define CALLFN(NAME,MAXSIZE)			\
-TEXT NAME(SB), WRAPPER, $MAXSIZE-24;		\
+TEXT NAME(SB), WRAPPER, $MAXSIZE-48;		\
 	NO_LOCAL_POINTERS;			\
 	/* copy arguments to stack */		\
-	MOVD	arg+16(FP), R3;			\
-	MOVWZ	argsize+24(FP), R4;			\
+	MOVD	stackArgs+16(FP), R3;			\
+	MOVWZ	stackArgsSize+24(FP), R4;			\
 	MOVD    R1, R5;				\
 	CMP	R4, $8;				\
 	BLT	tailsetup;			\
@@ -456,10 +439,10 @@ callfn: \
 	MOVD	24(R1), R2;			\
 #endif						\
 	/* copy return values back */		\
-	MOVD	argtype+0(FP), R7;		\
-	MOVD	arg+16(FP), R3;			\
-	MOVWZ	n+24(FP), R4;			\
-	MOVWZ	retoffset+28(FP), R6;		\
+	MOVD	stackArgsType+0(FP), R7;		\
+	MOVD	stackArgs+16(FP), R3;			\
+	MOVWZ	stackArgsSize+24(FP), R4;			\
+	MOVWZ	stackRetOffset+28(FP), R6;		\
 	ADD	$FIXED_FRAME, R1, R5;		\
 	ADD	R6, R5; 			\
 	ADD	R6, R3;				\
@@ -471,11 +454,12 @@ callfn: \
 // separate function so it can allocate stack space for the arguments
 // to reflectcallmove. It does not follow the Go ABI; it expects its
 // arguments in registers.
-TEXT callRet<>(SB), NOSPLIT, $32-0
+TEXT callRet<>(SB), NOSPLIT, $40-0
 	MOVD	R7, FIXED_FRAME+0(R1)
 	MOVD	R3, FIXED_FRAME+8(R1)
 	MOVD	R5, FIXED_FRAME+16(R1)
 	MOVD	R4, FIXED_FRAME+24(R1)
+	MOVD	$0, FIXED_FRAME+32(R1)
 	BL	runtime·reflectcallmove(SB)
 	RET
 
diff --git a/src/runtime/asm_riscv64.s b/src/runtime/asm_riscv64.s
index 01b42dc3de..31e324d677 100644
--- a/src/runtime/asm_riscv64.s
+++ b/src/runtime/asm_riscv64.s
@@ -297,21 +297,6 @@ TEXT runtime·mcall(SB), NOSPLIT|NOFRAME, $0-8
 	JALR	RA, T1
 	JMP	runtime·badmcall2(SB)
 
-// func gosave(buf *gobuf)
-// save state in Gobuf; setjmp
-TEXT runtime·gosave(SB), NOSPLIT|NOFRAME, $0-8
-	MOV	buf+0(FP), T1
-	MOV	X2, gobuf_sp(T1)
-	MOV	RA, gobuf_pc(T1)
-	MOV	g, gobuf_g(T1)
-	MOV	ZERO, gobuf_lr(T1)
-	MOV	ZERO, gobuf_ret(T1)
-	// Assert ctxt is zero. See func save.
-	MOV	gobuf_ctxt(T1), T1
-	BEQ	T1, ZERO, 2(PC)
-	CALL	runtime·badctxt(SB)
-	RET
-
 // Save state of caller into g->sched. Smashes X31.
 TEXT gosave<>(SB),NOSPLIT|NOFRAME,$0
 	MOV	X1, (g_sched+gobuf_pc)(g)
@@ -374,7 +359,7 @@ TEXT runtime·asminit(SB),NOSPLIT|NOFRAME,$0-0
 	RET
 
 // reflectcall: call a function with the given argument list
-// func call(argtype *_type, f *FuncVal, arg *byte, argsize, retoffset uint32).
+// func call(stackArgsType *_type, f *FuncVal, stackArgs *byte, stackArgsSize, stackRetOffset, frameSize uint32, regArgs *abi.RegArgs).
 // we don't have variable-sized frames, so we use a small number
 // of constant-sized-frame functions to encode a few bits of size in the pc.
 // Caution: ugly multiline assembly macros in your future!
@@ -386,13 +371,13 @@ TEXT runtime·asminit(SB),NOSPLIT|NOFRAME,$0-0
 	JALR	ZERO, T2
 // Note: can't just "BR NAME(SB)" - bad inlining results.
 
-// func call(argtype *rtype, fn, arg unsafe.Pointer, n uint32, retoffset uint32)
+// func call(stackArgsType *rtype, fn, stackArgs unsafe.Pointer, stackArgsSize, stackRetOffset, frameSize uint32, regArgs *abi.RegArgs).
 TEXT reflect·call(SB), NOSPLIT, $0-0
 	JMP	·reflectcall(SB)
 
-// func reflectcall(argtype *_type, fn, arg unsafe.Pointer, argsize uint32, retoffset uint32)
-TEXT ·reflectcall(SB), NOSPLIT|NOFRAME, $0-32
-	MOVWU argsize+24(FP), T0
+// func call(stackArgsType *_type, fn, stackArgs unsafe.Pointer, stackArgsSize, stackRetOffset, frameSize uint32, regArgs *abi.RegArgs).
+TEXT ·reflectcall(SB), NOSPLIT|NOFRAME, $0-48
+	MOVWU	frameSize+32(FP), T0
 	DISPATCH(runtime·call16, 16)
 	DISPATCH(runtime·call32, 32)
 	DISPATCH(runtime·call64, 64)
@@ -424,11 +409,11 @@ TEXT ·reflectcall(SB), NOSPLIT|NOFRAME, $0-32
 	JALR	ZERO, T2
 
 #define CALLFN(NAME,MAXSIZE)			\
-TEXT NAME(SB), WRAPPER, $MAXSIZE-24;		\
+TEXT NAME(SB), WRAPPER, $MAXSIZE-48;		\
 	NO_LOCAL_POINTERS;			\
 	/* copy arguments to stack */		\
-	MOV	arg+16(FP), A1;			\
-	MOVWU	argsize+24(FP), A2;		\
+	MOV	stackArgs+16(FP), A1;			\
+	MOVWU	stackArgsSize+24(FP), A2;		\
 	MOV	X2, A3;				\
 	ADD	$8, A3;				\
 	ADD	A3, A2;				\
@@ -444,10 +429,10 @@ TEXT NAME(SB), WRAPPER, $MAXSIZE-24;		\
 	PCDATA  $PCDATA_StackMapIndex, $0;	\
 	JALR	RA, A4;				\
 	/* copy return values back */		\
-	MOV	argtype+0(FP), A5;		\
-	MOV	arg+16(FP), A1;			\
-	MOVWU	n+24(FP), A2;			\
-	MOVWU	retoffset+28(FP), A4;		\
+	MOV	stackArgsType+0(FP), A5;		\
+	MOV	stackArgs+16(FP), A1;			\
+	MOVWU	stackArgsSize+24(FP), A2;			\
+	MOVWU	stackRetOffset+28(FP), A4;		\
 	ADD	$8, X2, A3;			\
 	ADD	A4, A3; 			\
 	ADD	A4, A1;				\
@@ -459,11 +444,12 @@ TEXT NAME(SB), WRAPPER, $MAXSIZE-24;		\
 // separate function so it can allocate stack space for the arguments
 // to reflectcallmove. It does not follow the Go ABI; it expects its
 // arguments in registers.
-TEXT callRet<>(SB), NOSPLIT, $32-0
+TEXT callRet<>(SB), NOSPLIT, $40-0
 	MOV	A5, 8(X2)
 	MOV	A1, 16(X2)
 	MOV	A3, 24(X2)
 	MOV	A2, 32(X2)
+	MOV	$0, 40(X2)
 	CALL	runtime·reflectcallmove(SB)
 	RET
 
diff --git a/src/runtime/asm_s390x.s b/src/runtime/asm_s390x.s
index 7baef37324..fbd185c353 100644
--- a/src/runtime/asm_s390x.s
+++ b/src/runtime/asm_s390x.s
@@ -174,21 +174,6 @@ TEXT runtime·asminit(SB),NOSPLIT|NOFRAME,$0-0
  *  go-routine
  */
 
-// void gosave(Gobuf*)
-// save state in Gobuf; setjmp
-TEXT runtime·gosave(SB), NOSPLIT, $-8-8
-	MOVD	buf+0(FP), R3
-	MOVD	R15, gobuf_sp(R3)
-	MOVD	LR, gobuf_pc(R3)
-	MOVD	g, gobuf_g(R3)
-	MOVD	$0, gobuf_lr(R3)
-	MOVD	$0, gobuf_ret(R3)
-	// Assert ctxt is zero. See func save.
-	MOVD	gobuf_ctxt(R3), R3
-	CMPBEQ	R3, $0, 2(PC)
-	BL	runtime·badctxt(SB)
-	RET
-
 // void gogo(Gobuf*)
 // restore state from Gobuf; longjmp
 TEXT runtime·gogo(SB), NOSPLIT, $16-8
@@ -368,7 +353,7 @@ TEXT runtime·morestack_noctxt(SB),NOSPLIT|NOFRAME,$0-0
 	BR	runtime·morestack(SB)
 
 // reflectcall: call a function with the given argument list
-// func call(argtype *_type, f *FuncVal, arg *byte, argsize, retoffset uint32).
+// func call(stackArgsType *_type, f *FuncVal, stackArgs *byte, stackArgsSize, stackRetOffset, frameSize uint32, regArgs *abi.RegArgs).
 // we don't have variable-sized frames, so we use a small number
 // of constant-sized-frame functions to encode a few bits of size in the pc.
 // Caution: ugly multiline assembly macros in your future!
@@ -381,8 +366,8 @@ TEXT runtime·morestack_noctxt(SB),NOSPLIT|NOFRAME,$0-0
 	BR	(R5)
 // Note: can't just "BR NAME(SB)" - bad inlining results.
 
-TEXT ·reflectcall(SB), NOSPLIT, $-8-32
-	MOVWZ argsize+24(FP), R3
+TEXT ·reflectcall(SB), NOSPLIT, $-8-48
+	MOVWZ	frameSize+32(FP), R3
 	DISPATCH(runtime·call16, 16)
 	DISPATCH(runtime·call32, 32)
 	DISPATCH(runtime·call64, 64)
@@ -414,11 +399,11 @@ TEXT ·reflectcall(SB), NOSPLIT, $-8-32
 	BR	(R5)
 
 #define CALLFN(NAME,MAXSIZE)			\
-TEXT NAME(SB), WRAPPER, $MAXSIZE-24;		\
+TEXT NAME(SB), WRAPPER, $MAXSIZE-48;		\
 	NO_LOCAL_POINTERS;			\
 	/* copy arguments to stack */		\
-	MOVD	arg+16(FP), R4;			\
-	MOVWZ	argsize+24(FP), R5;		\
+	MOVD	stackArgs+16(FP), R4;			\
+	MOVWZ	stackArgsSize+24(FP), R5;		\
 	MOVD	$stack-MAXSIZE(SP), R6;		\
 loopArgs: /* copy 256 bytes at a time */	\
 	CMP	R5, $256;			\
@@ -439,11 +424,11 @@ callFunction:					\
 	PCDATA  $PCDATA_StackMapIndex, $0;	\
 	BL	(R8);				\
 	/* copy return values back */		\
-	MOVD	argtype+0(FP), R7;		\
-	MOVD	arg+16(FP), R6;			\
-	MOVWZ	n+24(FP), R5;			\
+	MOVD	stackArgsType+0(FP), R7;		\
+	MOVD	stackArgs+16(FP), R6;			\
+	MOVWZ	stackArgsSize+24(FP), R5;			\
 	MOVD	$stack-MAXSIZE(SP), R4;		\
-	MOVWZ	retoffset+28(FP), R1;		\
+	MOVWZ	stackRetOffset+28(FP), R1;		\
 	ADD	R1, R4;				\
 	ADD	R1, R6;				\
 	SUB	R1, R5;				\
@@ -454,11 +439,12 @@ callFunction:					\
 // separate function so it can allocate stack space for the arguments
 // to reflectcallmove. It does not follow the Go ABI; it expects its
 // arguments in registers.
-TEXT callRet<>(SB), NOSPLIT, $32-0
+TEXT callRet<>(SB), NOSPLIT, $40-0
 	MOVD	R7, 8(R15)
 	MOVD	R6, 16(R15)
 	MOVD	R4, 24(R15)
 	MOVD	R5, 32(R15)
+	MOVD	$0, 40(R15)
 	BL	runtime·reflectcallmove(SB)
 	RET
 
diff --git a/src/runtime/asm_wasm.s b/src/runtime/asm_wasm.s
index fcb780f1dc..cf3d961b74 100644
--- a/src/runtime/asm_wasm.s
+++ b/src/runtime/asm_wasm.s
@@ -296,14 +296,14 @@ TEXT ·asmcgocall(SB), NOSPLIT, $0-0
 		JMP NAME(SB); \
 	End
 
-TEXT ·reflectcall(SB), NOSPLIT, $0-32
+TEXT ·reflectcall(SB), NOSPLIT, $0-48
 	I64Load fn+8(FP)
 	I64Eqz
 	If
 		CALLNORESUME runtime·sigpanic<ABIInternal>(SB)
 	End
 
-	MOVW argsize+24(FP), R0
+	MOVW frameSize+32(FP), R0
 
 	DISPATCH(runtime·call16, 16)
 	DISPATCH(runtime·call32, 32)
@@ -335,18 +335,18 @@ TEXT ·reflectcall(SB), NOSPLIT, $0-32
 	JMP runtime·badreflectcall(SB)
 
 #define CALLFN(NAME, MAXSIZE) \
-TEXT NAME(SB), WRAPPER, $MAXSIZE-32; \
+TEXT NAME(SB), WRAPPER, $MAXSIZE-48; \
 	NO_LOCAL_POINTERS; \
-	MOVW argsize+24(FP), R0; \
+	MOVW stackArgsSize+24(FP), R0; \
 	\
 	Get R0; \
 	I64Eqz; \
 	Not; \
 	If; \
 		Get SP; \
-		I64Load argptr+16(FP); \
+		I64Load stackArgs+16(FP); \
 		I32WrapI64; \
-		I64Load argsize+24(FP); \
+		I64Load stackArgsSize+24(FP); \
 		I64Const $3; \
 		I64ShrU; \
 		I32WrapI64; \
@@ -359,12 +359,12 @@ TEXT NAME(SB), WRAPPER, $MAXSIZE-32; \
 	I64Load $0; \
 	CALL; \
 	\
-	I64Load32U retoffset+28(FP); \
+	I64Load32U stackRetOffset+28(FP); \
 	Set R0; \
 	\
-	MOVD argtype+0(FP), RET0; \
+	MOVD stackArgsType+0(FP), RET0; \
 	\
-	I64Load argptr+16(FP); \
+	I64Load stackArgs+16(FP); \
 	Get R0; \
 	I64Add; \
 	Set RET1; \
@@ -375,7 +375,7 @@ TEXT NAME(SB), WRAPPER, $MAXSIZE-32; \
 	I64Add; \
 	Set RET2; \
 	\
-	I64Load32U argsize+24(FP); \
+	I64Load32U stackArgsSize+24(FP); \
 	Get R0; \
 	I64Sub; \
 	Set RET3; \
@@ -387,12 +387,13 @@ TEXT NAME(SB), WRAPPER, $MAXSIZE-32; \
 // separate function so it can allocate stack space for the arguments
 // to reflectcallmove. It does not follow the Go ABI; it expects its
 // arguments in registers.
-TEXT callRet<>(SB), NOSPLIT, $32-0
+TEXT callRet<>(SB), NOSPLIT, $40-0
 	NO_LOCAL_POINTERS
 	MOVD RET0, 0(SP)
 	MOVD RET1, 8(SP)
 	MOVD RET2, 16(SP)
 	MOVD RET3, 24(SP)
+	MOVD $0,   32(SP)
 	CALL runtime·reflectcallmove(SB)
 	RET
 
diff --git a/src/runtime/cgo/gcc_amd64.S b/src/runtime/cgo/gcc_amd64.S
index 17d9d47ef4..d75f864666 100644
--- a/src/runtime/cgo/gcc_amd64.S
+++ b/src/runtime/cgo/gcc_amd64.S
@@ -30,9 +30,14 @@ EXT(crosscall_amd64):
 	pushq %r15
 
 #if defined(_WIN64)
+	movq %r8, %rdi	/* arg of setg_gcc */
+	call *%rdx	/* setg_gcc */
 	call *%rcx	/* fn */
 #else
-	call *%rdi	/* fn */
+	movq %rdi, %rbx
+	movq %rdx, %rdi	/* arg of setg_gcc */
+	call *%rsi	/* setg_gcc */
+	call *%rbx	/* fn */
 #endif
 
 	popq %r15
diff --git a/src/runtime/cgo/gcc_darwin_amd64.c b/src/runtime/cgo/gcc_darwin_amd64.c
index 51410d5026..d5b7fd8fd8 100644
--- a/src/runtime/cgo/gcc_darwin_amd64.c
+++ b/src/runtime/cgo/gcc_darwin_amd64.c
@@ -9,13 +9,16 @@
 #include "libcgo_unix.h"
 
 static void* threadentry(void*);
+static void (*setg_gcc)(void*);
 
 void
-x_cgo_init(G *g)
+x_cgo_init(G *g, void (*setg)(void*), void **tlsg, void **tlsbase)
 {
 	pthread_attr_t attr;
 	size_t size;
 
+	setg_gcc = setg;
+
 	pthread_attr_init(&attr);
 	pthread_attr_getstacksize(&attr, &size);
 	g->stacklo = (uintptr)&attr - size + 4096;
@@ -57,10 +60,6 @@ threadentry(void *v)
 	ts = *(ThreadStart*)v;
 	free(v);
 
-	// Move the g pointer into the slot reserved in thread local storage.
-	// Constant must match the one in cmd/link/internal/ld/sym.go.
-	asm volatile("movq %0, %%gs:0x30" :: "r"(ts.g));
-
-	crosscall_amd64(ts.fn);
+	crosscall_amd64(ts.fn, setg_gcc, (void*)ts.g);
 	return nil;
 }
diff --git a/src/runtime/cgo/gcc_dragonfly_amd64.c b/src/runtime/cgo/gcc_dragonfly_amd64.c
index d25db91900..0003414bf8 100644
--- a/src/runtime/cgo/gcc_dragonfly_amd64.c
+++ b/src/runtime/cgo/gcc_dragonfly_amd64.c
@@ -61,11 +61,6 @@ threadentry(void *v)
 	ts = *(ThreadStart*)v;
 	free(v);
 
-	/*
-	 * Set specific keys.
-	 */
-	setg_gcc((void*)ts.g);
-
-	crosscall_amd64(ts.fn);
+	crosscall_amd64(ts.fn, setg_gcc, (void*)ts.g);
 	return nil;
 }
diff --git a/src/runtime/cgo/gcc_freebsd_amd64.c b/src/runtime/cgo/gcc_freebsd_amd64.c
index 514a2f8a23..6071ec3909 100644
--- a/src/runtime/cgo/gcc_freebsd_amd64.c
+++ b/src/runtime/cgo/gcc_freebsd_amd64.c
@@ -69,11 +69,6 @@ threadentry(void *v)
 	free(v);
 	_cgo_tsan_release();
 
-	/*
-	 * Set specific keys.
-	 */
-	setg_gcc((void*)ts.g);
-
-	crosscall_amd64(ts.fn);
+	crosscall_amd64(ts.fn, setg_gcc, (void*)ts.g);
 	return nil;
 }
diff --git a/src/runtime/cgo/gcc_linux_amd64.c b/src/runtime/cgo/gcc_linux_amd64.c
index f2bf6482cb..c25e7e769b 100644
--- a/src/runtime/cgo/gcc_linux_amd64.c
+++ b/src/runtime/cgo/gcc_linux_amd64.c
@@ -89,11 +89,6 @@ threadentry(void *v)
 	free(v);
 	_cgo_tsan_release();
 
-	/*
-	 * Set specific keys.
-	 */
-	setg_gcc((void*)ts.g);
-
-	crosscall_amd64(ts.fn);
+	crosscall_amd64(ts.fn, setg_gcc, (void*)ts.g);
 	return nil;
 }
diff --git a/src/runtime/cgo/gcc_netbsd_amd64.c b/src/runtime/cgo/gcc_netbsd_amd64.c
index dc966fc45b..9f4b031a08 100644
--- a/src/runtime/cgo/gcc_netbsd_amd64.c
+++ b/src/runtime/cgo/gcc_netbsd_amd64.c
@@ -62,11 +62,6 @@ threadentry(void *v)
 	ts = *(ThreadStart*)v;
 	free(v);
 
-	/*
-	 * Set specific keys.
-	 */
-	setg_gcc((void*)ts.g);
-
 	// On NetBSD, a new thread inherits the signal stack of the
 	// creating thread. That confuses minit, so we remove that
 	// signal stack here before calling the regular mstart. It's
@@ -78,6 +73,6 @@ threadentry(void *v)
 	ss.ss_flags = SS_DISABLE;
 	sigaltstack(&ss, nil);
 
-	crosscall_amd64(ts.fn);
+	crosscall_amd64(ts.fn, setg_gcc, (void*)ts.g);
 	return nil;
 }
diff --git a/src/runtime/cgo/gcc_openbsd_amd64.c b/src/runtime/cgo/gcc_openbsd_amd64.c
index 34319fb0b8..09d2750f3a 100644
--- a/src/runtime/cgo/gcc_openbsd_amd64.c
+++ b/src/runtime/cgo/gcc_openbsd_amd64.c
@@ -60,11 +60,6 @@ threadentry(void *v)
 	ts = *(ThreadStart*)v;
 	free(v);
 
-	/*
-	 * Set specific keys.
-	 */
-	setg_gcc((void*)ts.g);
-
-	crosscall_amd64(ts.fn);
+	crosscall_amd64(ts.fn, setg_gcc, (void*)ts.g);
 	return nil;
 }
diff --git a/src/runtime/cgo/gcc_solaris_amd64.c b/src/runtime/cgo/gcc_solaris_amd64.c
index 079bd12898..e89e844b1e 100644
--- a/src/runtime/cgo/gcc_solaris_amd64.c
+++ b/src/runtime/cgo/gcc_solaris_amd64.c
@@ -72,11 +72,6 @@ threadentry(void *v)
 	ts = *(ThreadStart*)v;
 	free(v);
 
-	/*
-	 * Set specific keys.
-	 */
-	setg_gcc((void*)ts.g);
-
-	crosscall_amd64(ts.fn);
+	crosscall_amd64(ts.fn, setg_gcc, (void*)ts.g);
 	return nil;
 }
diff --git a/src/runtime/cgo/gcc_windows_amd64.c b/src/runtime/cgo/gcc_windows_amd64.c
index 0f8c817f0e..25cfd086dd 100644
--- a/src/runtime/cgo/gcc_windows_amd64.c
+++ b/src/runtime/cgo/gcc_windows_amd64.c
@@ -12,10 +12,12 @@
 #include "libcgo_windows.h"
 
 static void threadentry(void*);
+static void (*setg_gcc)(void*);
 
 void
-x_cgo_init(G *g)
+x_cgo_init(G *g, void (*setg)(void*), void **tlsg, void **tlsbase)
 {
+	setg_gcc = setg;
 }
 
 
@@ -46,10 +48,8 @@ threadentry(void *v)
 	 */
 	asm volatile (
 	  "movq %0, %%gs:0x28\n"	// MOVL tls0, 0x28(GS)
-	  "movq %%gs:0x28, %%rax\n" // MOVQ 0x28(GS), tmp
-	  "movq %1, 0(%%rax)\n" // MOVQ g, 0(GS)
-	  :: "r"(ts.tls), "r"(ts.g) : "%rax"
+	  :: "r"(ts.tls)
 	);
 
-	crosscall_amd64(ts.fn);
+	crosscall_amd64(ts.fn, setg_gcc, (void*)ts.g);
 }
diff --git a/src/runtime/cgo/libcgo.h b/src/runtime/cgo/libcgo.h
index aba500a301..af4960e7e9 100644
--- a/src/runtime/cgo/libcgo.h
+++ b/src/runtime/cgo/libcgo.h
@@ -66,7 +66,7 @@ uintptr_t _cgo_wait_runtime_init_done(void);
 /*
  * Call fn in the 6c world.
  */
-void crosscall_amd64(void (*fn)(void));
+void crosscall_amd64(void (*fn)(void), void (*setg_gcc)(void*), void *g);
 
 /*
  * Call fn in the 8c world.
diff --git a/src/runtime/duff_amd64.s b/src/runtime/duff_amd64.s
index 2ff5bf6dbc..df010f5853 100644
--- a/src/runtime/duff_amd64.s
+++ b/src/runtime/duff_amd64.s
@@ -5,100 +5,100 @@
 #include "textflag.h"
 
 TEXT runtime·duffzero<ABIInternal>(SB), NOSPLIT, $0-0
-	MOVUPS	X0,(DI)
-	MOVUPS	X0,16(DI)
-	MOVUPS	X0,32(DI)
-	MOVUPS	X0,48(DI)
+	MOVUPS	X15,(DI)
+	MOVUPS	X15,16(DI)
+	MOVUPS	X15,32(DI)
+	MOVUPS	X15,48(DI)
 	LEAQ	64(DI),DI
 
-	MOVUPS	X0,(DI)
-	MOVUPS	X0,16(DI)
-	MOVUPS	X0,32(DI)
-	MOVUPS	X0,48(DI)
+	MOVUPS	X15,(DI)
+	MOVUPS	X15,16(DI)
+	MOVUPS	X15,32(DI)
+	MOVUPS	X15,48(DI)
 	LEAQ	64(DI),DI
 
-	MOVUPS	X0,(DI)
-	MOVUPS	X0,16(DI)
-	MOVUPS	X0,32(DI)
-	MOVUPS	X0,48(DI)
+	MOVUPS	X15,(DI)
+	MOVUPS	X15,16(DI)
+	MOVUPS	X15,32(DI)
+	MOVUPS	X15,48(DI)
 	LEAQ	64(DI),DI
 
-	MOVUPS	X0,(DI)
-	MOVUPS	X0,16(DI)
-	MOVUPS	X0,32(DI)
-	MOVUPS	X0,48(DI)
+	MOVUPS	X15,(DI)
+	MOVUPS	X15,16(DI)
+	MOVUPS	X15,32(DI)
+	MOVUPS	X15,48(DI)
 	LEAQ	64(DI),DI
 
-	MOVUPS	X0,(DI)
-	MOVUPS	X0,16(DI)
-	MOVUPS	X0,32(DI)
-	MOVUPS	X0,48(DI)
+	MOVUPS	X15,(DI)
+	MOVUPS	X15,16(DI)
+	MOVUPS	X15,32(DI)
+	MOVUPS	X15,48(DI)
 	LEAQ	64(DI),DI
 
-	MOVUPS	X0,(DI)
-	MOVUPS	X0,16(DI)
-	MOVUPS	X0,32(DI)
-	MOVUPS	X0,48(DI)
+	MOVUPS	X15,(DI)
+	MOVUPS	X15,16(DI)
+	MOVUPS	X15,32(DI)
+	MOVUPS	X15,48(DI)
 	LEAQ	64(DI),DI
 
-	MOVUPS	X0,(DI)
-	MOVUPS	X0,16(DI)
-	MOVUPS	X0,32(DI)
-	MOVUPS	X0,48(DI)
+	MOVUPS	X15,(DI)
+	MOVUPS	X15,16(DI)
+	MOVUPS	X15,32(DI)
+	MOVUPS	X15,48(DI)
 	LEAQ	64(DI),DI
 
-	MOVUPS	X0,(DI)
-	MOVUPS	X0,16(DI)
-	MOVUPS	X0,32(DI)
-	MOVUPS	X0,48(DI)
+	MOVUPS	X15,(DI)
+	MOVUPS	X15,16(DI)
+	MOVUPS	X15,32(DI)
+	MOVUPS	X15,48(DI)
 	LEAQ	64(DI),DI
 
-	MOVUPS	X0,(DI)
-	MOVUPS	X0,16(DI)
-	MOVUPS	X0,32(DI)
-	MOVUPS	X0,48(DI)
+	MOVUPS	X15,(DI)
+	MOVUPS	X15,16(DI)
+	MOVUPS	X15,32(DI)
+	MOVUPS	X15,48(DI)
 	LEAQ	64(DI),DI
 
-	MOVUPS	X0,(DI)
-	MOVUPS	X0,16(DI)
-	MOVUPS	X0,32(DI)
-	MOVUPS	X0,48(DI)
+	MOVUPS	X15,(DI)
+	MOVUPS	X15,16(DI)
+	MOVUPS	X15,32(DI)
+	MOVUPS	X15,48(DI)
 	LEAQ	64(DI),DI
 
-	MOVUPS	X0,(DI)
-	MOVUPS	X0,16(DI)
-	MOVUPS	X0,32(DI)
-	MOVUPS	X0,48(DI)
+	MOVUPS	X15,(DI)
+	MOVUPS	X15,16(DI)
+	MOVUPS	X15,32(DI)
+	MOVUPS	X15,48(DI)
 	LEAQ	64(DI),DI
 
-	MOVUPS	X0,(DI)
-	MOVUPS	X0,16(DI)
-	MOVUPS	X0,32(DI)
-	MOVUPS	X0,48(DI)
+	MOVUPS	X15,(DI)
+	MOVUPS	X15,16(DI)
+	MOVUPS	X15,32(DI)
+	MOVUPS	X15,48(DI)
 	LEAQ	64(DI),DI
 
-	MOVUPS	X0,(DI)
-	MOVUPS	X0,16(DI)
-	MOVUPS	X0,32(DI)
-	MOVUPS	X0,48(DI)
+	MOVUPS	X15,(DI)
+	MOVUPS	X15,16(DI)
+	MOVUPS	X15,32(DI)
+	MOVUPS	X15,48(DI)
 	LEAQ	64(DI),DI
 
-	MOVUPS	X0,(DI)
-	MOVUPS	X0,16(DI)
-	MOVUPS	X0,32(DI)
-	MOVUPS	X0,48(DI)
+	MOVUPS	X15,(DI)
+	MOVUPS	X15,16(DI)
+	MOVUPS	X15,32(DI)
+	MOVUPS	X15,48(DI)
 	LEAQ	64(DI),DI
 
-	MOVUPS	X0,(DI)
-	MOVUPS	X0,16(DI)
-	MOVUPS	X0,32(DI)
-	MOVUPS	X0,48(DI)
+	MOVUPS	X15,(DI)
+	MOVUPS	X15,16(DI)
+	MOVUPS	X15,32(DI)
+	MOVUPS	X15,48(DI)
 	LEAQ	64(DI),DI
 
-	MOVUPS	X0,(DI)
-	MOVUPS	X0,16(DI)
-	MOVUPS	X0,32(DI)
-	MOVUPS	X0,48(DI)
+	MOVUPS	X15,(DI)
+	MOVUPS	X15,16(DI)
+	MOVUPS	X15,32(DI)
+	MOVUPS	X15,48(DI)
 	LEAQ	64(DI),DI
 
 	RET
diff --git a/src/runtime/mbarrier.go b/src/runtime/mbarrier.go
index 2b5affce52..4994347bde 100644
--- a/src/runtime/mbarrier.go
+++ b/src/runtime/mbarrier.go
@@ -14,6 +14,7 @@
 package runtime
 
 import (
+	"internal/abi"
 	"runtime/internal/sys"
 	"unsafe"
 )
@@ -223,11 +224,18 @@ func reflect_typedmemmovepartial(typ *_type, dst, src unsafe.Pointer, off, size
 // stack map of reflectcall is wrong.
 //
 //go:nosplit
-func reflectcallmove(typ *_type, dst, src unsafe.Pointer, size uintptr) {
+func reflectcallmove(typ *_type, dst, src unsafe.Pointer, size uintptr, regs *abi.RegArgs) {
 	if writeBarrier.needed && typ != nil && typ.ptrdata != 0 && size >= sys.PtrSize {
 		bulkBarrierPreWrite(uintptr(dst), uintptr(src), size)
 	}
 	memmove(dst, src, size)
+
+	// Move pointers returned in registers to a place where the GC can see them.
+	for i := range regs.Ints {
+		if regs.ReturnIsPtr.Get(i) {
+			regs.Ptrs[i] = unsafe.Pointer(regs.Ints[i])
+		}
+	}
 }
 
 //go:nosplit
diff --git a/src/runtime/mfinal.go b/src/runtime/mfinal.go
index f4dbd77252..7d0313be12 100644
--- a/src/runtime/mfinal.go
+++ b/src/runtime/mfinal.go
@@ -7,6 +7,7 @@
 package runtime
 
 import (
+	"internal/abi"
 	"runtime/internal/atomic"
 	"runtime/internal/sys"
 	"unsafe"
@@ -219,7 +220,11 @@ func runfinq() {
 					throw("bad kind in runfinq")
 				}
 				fingRunning = true
-				reflectcall(nil, unsafe.Pointer(f.fn), frame, uint32(framesz), uint32(framesz))
+				// Pass a dummy RegArgs for now.
+				//
+				// TODO(mknyszek): Pass arguments in registers.
+				var regs abi.RegArgs
+				reflectcall(nil, unsafe.Pointer(f.fn), frame, uint32(framesz), uint32(framesz), uint32(framesz), &regs)
 				fingRunning = false
 
 				// Drop finalizer queue heap references
diff --git a/src/runtime/mkduff.go b/src/runtime/mkduff.go
index 94ae75fbfe..ef297f073e 100644
--- a/src/runtime/mkduff.go
+++ b/src/runtime/mkduff.go
@@ -62,15 +62,15 @@ func gen(arch string, tags, zero, copy func(io.Writer)) {
 func notags(w io.Writer) { fmt.Fprintln(w) }
 
 func zeroAMD64(w io.Writer) {
-	// X0: zero
+	// X15: zero
 	// DI: ptr to memory to be zeroed
 	// DI is updated as a side effect.
-	fmt.Fprintln(w, "TEXT runtime·duffzero(SB), NOSPLIT, $0-0")
+	fmt.Fprintln(w, "TEXT runtime·duffzero<ABIInternal>(SB), NOSPLIT, $0-0")
 	for i := 0; i < 16; i++ {
-		fmt.Fprintln(w, "\tMOVUPS\tX0,(DI)")
-		fmt.Fprintln(w, "\tMOVUPS\tX0,16(DI)")
-		fmt.Fprintln(w, "\tMOVUPS\tX0,32(DI)")
-		fmt.Fprintln(w, "\tMOVUPS\tX0,48(DI)")
+		fmt.Fprintln(w, "\tMOVUPS\tX15,(DI)")
+		fmt.Fprintln(w, "\tMOVUPS\tX15,16(DI)")
+		fmt.Fprintln(w, "\tMOVUPS\tX15,32(DI)")
+		fmt.Fprintln(w, "\tMOVUPS\tX15,48(DI)")
 		fmt.Fprintln(w, "\tLEAQ\t64(DI),DI") // We use lea instead of add, to avoid clobbering flags
 		fmt.Fprintln(w)
 	}
@@ -84,7 +84,7 @@ func copyAMD64(w io.Writer) {
 	//
 	// This is equivalent to a sequence of MOVSQ but
 	// for some reason that is 3.5x slower than this code.
-	fmt.Fprintln(w, "TEXT runtime·duffcopy(SB), NOSPLIT, $0-0")
+	fmt.Fprintln(w, "TEXT runtime·duffcopy<ABIInternal>(SB), NOSPLIT, $0-0")
 	for i := 0; i < 64; i++ {
 		fmt.Fprintln(w, "\tMOVUPS\t(SI), X0")
 		fmt.Fprintln(w, "\tADDQ\t$16, SI")
diff --git a/src/runtime/os2_aix.go b/src/runtime/os2_aix.go
index 428ff7f225..abd1010be9 100644
--- a/src/runtime/os2_aix.go
+++ b/src/runtime/os2_aix.go
@@ -18,11 +18,11 @@ import (
 
 //go:cgo_import_dynamic libc___n_pthreads __n_pthreads "libpthread.a/shr_xpg5_64.o"
 //go:cgo_import_dynamic libc___mod_init __mod_init "libc.a/shr_64.o"
-//go:linkname libc___n_pthreads libc___n_pthread
+//go:linkname libc___n_pthreads libc___n_pthreads
 //go:linkname libc___mod_init libc___mod_init
 
 var (
-	libc___n_pthread,
+	libc___n_pthreads,
 	libc___mod_init libFunc
 )
 
diff --git a/src/runtime/panic.go b/src/runtime/panic.go
index 5b2ccdd874..e320eaa596 100644
--- a/src/runtime/panic.go
+++ b/src/runtime/panic.go
@@ -5,6 +5,7 @@
 package runtime
 
 import (
+	"internal/abi"
 	"runtime/internal/atomic"
 	"runtime/internal/sys"
 	"unsafe"
@@ -874,7 +875,13 @@ func reflectcallSave(p *_panic, fn, arg unsafe.Pointer, argsize uint32) {
 		p.pc = getcallerpc()
 		p.sp = unsafe.Pointer(getcallersp())
 	}
-	reflectcall(nil, fn, arg, argsize, argsize)
+	// Pass a dummy RegArgs for now since no function actually implements
+	// the register-based ABI.
+	//
+	// TODO(mknyszek): Implement this properly, setting up arguments in
+	// registers as necessary in the caller.
+	var regs abi.RegArgs
+	reflectcall(nil, fn, arg, argsize, argsize, argsize, &regs)
 	if p != nil {
 		p.pc = 0
 		p.sp = unsafe.Pointer(nil)
@@ -968,7 +975,9 @@ func gopanic(e interface{}) {
 			}
 		} else {
 			p.argp = unsafe.Pointer(getargp(0))
-			reflectcall(nil, unsafe.Pointer(d.fn), deferArgs(d), uint32(d.siz), uint32(d.siz))
+
+			var regs abi.RegArgs
+			reflectcall(nil, unsafe.Pointer(d.fn), deferArgs(d), uint32(d.siz), uint32(d.siz), uint32(d.siz), &regs)
 		}
 		p.argp = nil
 
diff --git a/src/runtime/race/output_test.go b/src/runtime/race/output_test.go
index 69496874c6..17dc32013f 100644
--- a/src/runtime/race/output_test.go
+++ b/src/runtime/race/output_test.go
@@ -7,6 +7,7 @@
 package race_test
 
 import (
+	"fmt"
 	"internal/testenv"
 	"os"
 	"os/exec"
@@ -71,9 +72,24 @@ func TestOutput(t *testing.T) {
 			"GORACE="+test.gorace,
 		)
 		got, _ := cmd.CombinedOutput()
-		if !regexp.MustCompile(test.re).MatchString(string(got)) {
-			t.Fatalf("failed test case %v, expect:\n%v\ngot:\n%s",
-				test.name, test.re, got)
+		matched := false
+		for _, re := range test.re {
+			if regexp.MustCompile(re).MatchString(string(got)) {
+				matched = true
+				break
+			}
+		}
+		if !matched {
+			exp := fmt.Sprintf("expect:\n%v\n", test.re[0])
+			if len(test.re) > 1 {
+				exp = fmt.Sprintf("expected one of %d patterns:\n",
+					len(test.re))
+				for k, re := range test.re {
+					exp += fmt.Sprintf("pattern %d:\n%v\n", k, re)
+				}
+			}
+			t.Fatalf("failed test case %v, %sgot:\n%s",
+				test.name, exp, got)
 		}
 	}
 }
@@ -84,7 +100,7 @@ var tests = []struct {
 	goos   string
 	gorace string
 	source string
-	re     string
+	re     []string
 }{
 	{"simple", "run", "", "atexit_sleep_ms=0", `
 package main
@@ -107,7 +123,7 @@ func racer(x *int, done chan bool) {
 	store(x, 42)
 	done <- true
 }
-`, `==================
+`, []string{`==================
 WARNING: DATA RACE
 Write at 0x[0-9,a-f]+ by goroutine [0-9]:
   main\.store\(\)
@@ -129,7 +145,7 @@ Goroutine [0-9] \(running\) created at:
 ==================
 Found 1 data race\(s\)
 exit status 66
-`},
+`}},
 
 	{"exitcode", "run", "", "atexit_sleep_ms=0 exitcode=13", `
 package main
@@ -143,7 +159,7 @@ func main() {
 	x = 43
 	<-done
 }
-`, `exit status 13`},
+`, []string{`exit status 13`}},
 
 	{"strip_path_prefix", "run", "", "atexit_sleep_ms=0 strip_path_prefix=/main.", `
 package main
@@ -157,9 +173,9 @@ func main() {
 	x = 43
 	<-done
 }
-`, `
+`, []string{`
       go:7 \+0x[0-9,a-f]+
-`},
+`}},
 
 	{"halt_on_error", "run", "", "atexit_sleep_ms=0 halt_on_error=1", `
 package main
@@ -173,10 +189,10 @@ func main() {
 	x = 43
 	<-done
 }
-`, `
+`, []string{`
 ==================
 exit status 66
-`},
+`}},
 
 	{"test_fails_on_race", "test", "", "atexit_sleep_ms=0", `
 package main_test
@@ -193,12 +209,12 @@ func TestFail(t *testing.T) {
 	<-done
 	t.Log(t.Failed())
 }
-`, `
+`, []string{`
 ==================
 --- FAIL: TestFail \(0...s\)
 .*main_test.go:14: true
 .*testing.go:.*: race detected during execution of test
-FAIL`},
+FAIL`}},
 
 	{"slicebytetostring_pc", "run", "", "atexit_sleep_ms=0", `
 package main
@@ -211,11 +227,11 @@ func main() {
 	data[0] = 1
 	<-done
 }
-`, `
+`, []string{`
   runtime\.slicebytetostring\(\)
       .*/runtime/string\.go:.*
   main\.main\.func1\(\)
-      .*/main.go:7`},
+      .*/main.go:7`}},
 
 	// Test for https://golang.org/issue/33309
 	{"midstack_inlining_traceback", "run", "linux", "atexit_sleep_ms=0", `
@@ -241,7 +257,7 @@ func g(c chan int) {
 func h(c chan int) {
 	c <- x
 }
-`, `==================
+`, []string{`==================
 WARNING: DATA RACE
 Read at 0x[0-9,a-f]+ by goroutine [0-9]:
   main\.h\(\)
@@ -261,7 +277,7 @@ Goroutine [0-9] \(running\) created at:
 ==================
 Found 1 data race\(s\)
 exit status 66
-`},
+`}},
 
 	// Test for https://golang.org/issue/17190
 	{"external_cgo_thread", "run", "linux", "atexit_sleep_ms=0", `
@@ -300,7 +316,25 @@ func main() {
 	racy++
 	<- done
 }
-`, `==================
+`, []string{`==================
+WARNING: DATA RACE
+Read at 0x[0-9,a-f]+ by main goroutine:
+  main\.main\(\)
+      .*/main\.go:34 \+0x[0-9,a-f]+
+
+Previous write at 0x[0-9,a-f]+ by goroutine [0-9]:
+  main\.goCallback\(\)
+      .*/main\.go:27 \+0x[0-9,a-f]+
+  _cgoexp_[0-9a-z]+_goCallback\(\)
+      .*_cgo_gotypes\.go:[0-9]+ \+0x[0-9,a-f]+
+  _cgoexp_[0-9a-z]+_goCallback\(\)
+      <autogenerated>:1 \+0x[0-9,a-f]+
+
+Goroutine [0-9] \(running\) created at:
+  runtime\.newextram\(\)
+      .*/runtime/proc.go:[0-9]+ \+0x[0-9,a-f]+
+==================`,
+		`==================
 WARNING: DATA RACE
 Read at 0x[0-9,a-f]+ by .*:
   main\..*
@@ -313,7 +347,7 @@ Previous write at 0x[0-9,a-f]+ by .*:
 Goroutine [0-9] \(running\) created at:
   runtime\.newextram\(\)
       .*/runtime/proc.go:[0-9]+ \+0x[0-9,a-f]+
-==================`},
+==================`}},
 	{"second_test_passes", "test", "", "atexit_sleep_ms=0", `
 package main_test
 import "testing"
@@ -331,11 +365,11 @@ func TestFail(t *testing.T) {
 
 func TestPass(t *testing.T) {
 }
-`, `
+`, []string{`
 ==================
 --- FAIL: TestFail \(0...s\)
 .*testing.go:.*: race detected during execution of test
-FAIL`},
+FAIL`}},
 	{"mutex", "run", "", "atexit_sleep_ms=0", `
 package main
 import (
@@ -366,7 +400,7 @@ func main() {
 	}
 	wg.Wait()
 	if (data == iterations*(threads+1)) { fmt.Println("pass") }
-}`, `pass`},
+}`, []string{`pass`}},
 	// Test for https://github.com/golang/go/issues/37355
 	{"chanmm", "run", "", "atexit_sleep_ms=0", `
 package main
@@ -395,7 +429,7 @@ func main() {
 	wg.Wait()
 	_ = data
 }
-`, `==================
+`, []string{`==================
 WARNING: DATA RACE
 Write at 0x[0-9,a-f]+ by goroutine [0-9]:
   main\.main\.func2\(\)
@@ -408,5 +442,5 @@ Previous write at 0x[0-9,a-f]+ by main goroutine:
 Goroutine [0-9] \(running\) created at:
   main\.main\(\)
       .*/main.go:[0-9]+ \+0x[0-9,a-f]+
-==================`},
+==================`}},
 }
diff --git a/src/runtime/race_amd64.s b/src/runtime/race_amd64.s
index 9818bc6ddf..c3b7bbfbfe 100644
--- a/src/runtime/race_amd64.s
+++ b/src/runtime/race_amd64.s
@@ -146,8 +146,10 @@ TEXT	runtime·racewriterangepc1(SB), NOSPLIT, $0-24
 // If addr (RARG1) is out of range, do nothing.
 // Otherwise, setup goroutine context and invoke racecall. Other arguments already set.
 TEXT	racecalladdr<>(SB), NOSPLIT, $0-0
+#ifndef GOEXPERIMENT_REGABI
 	get_tls(R12)
 	MOVQ	g(R12), R14
+#endif
 	MOVQ	g_racectx(R14), RARG0	// goroutine context
 	// Check that addr is within [arenastart, arenaend) or within [racedatastart, racedataend).
 	CMPQ	RARG1, runtime·racearenastart(SB)
@@ -183,8 +185,10 @@ TEXT	runtime·racefuncenter(SB), NOSPLIT, $0-8
 // R11 = caller's return address
 TEXT	racefuncenter<>(SB), NOSPLIT, $0-0
 	MOVQ	DX, R15		// save function entry context (for closures)
+#ifndef GOEXPERIMENT_REGABI
 	get_tls(R12)
 	MOVQ	g(R12), R14
+#endif
 	MOVQ	g_racectx(R14), RARG0	// goroutine context
 	MOVQ	R11, RARG1
 	// void __tsan_func_enter(ThreadState *thr, void *pc);
@@ -197,8 +201,10 @@ TEXT	racefuncenter<>(SB), NOSPLIT, $0-0
 // func runtime·racefuncexit()
 // Called from instrumented code.
 TEXT	runtime·racefuncexit(SB), NOSPLIT, $0-0
+#ifndef GOEXPERIMENT_REGABI
 	get_tls(R12)
 	MOVQ	g(R12), R14
+#endif
 	MOVQ	g_racectx(R14), RARG0	// goroutine context
 	// void __tsan_func_exit(ThreadState *thr);
 	MOVQ	$__tsan_func_exit(SB), AX
@@ -357,8 +363,10 @@ racecallatomic_data:
 	JAE	racecallatomic_ignore
 racecallatomic_ok:
 	// Addr is within the good range, call the atomic function.
+#ifndef GOEXPERIMENT_REGABI
 	get_tls(R12)
 	MOVQ	g(R12), R14
+#endif
 	MOVQ	g_racectx(R14), RARG0	// goroutine context
 	MOVQ	8(SP), RARG1	// caller pc
 	MOVQ	(SP), RARG2	// pc
@@ -370,8 +378,10 @@ racecallatomic_ignore:
 	// An attempt to synchronize on the address would cause crash.
 	MOVQ	AX, R15	// remember the original function
 	MOVQ	$__tsan_go_ignore_sync_begin(SB), AX
+#ifndef GOEXPERIMENT_REGABI
 	get_tls(R12)
 	MOVQ	g(R12), R14
+#endif
 	MOVQ	g_racectx(R14), RARG0	// goroutine context
 	CALL	racecall<>(SB)
 	MOVQ	R15, AX	// restore the original function
@@ -399,8 +409,10 @@ TEXT	runtime·racecall(SB), NOSPLIT, $0-0
 
 // Switches SP to g0 stack and calls (AX). Arguments already set.
 TEXT	racecall<>(SB), NOSPLIT, $0-0
+#ifndef GOEXPERIMENT_REGABI
 	get_tls(R12)
 	MOVQ	g(R12), R14
+#endif
 	MOVQ	g_m(R14), R13
 	// Switch to g0 stack.
 	MOVQ	SP, R12		// callee-saved, preserved across the CALL
@@ -412,6 +424,9 @@ call:
 	ANDQ	$~15, SP	// alignment for gcc ABI
 	CALL	AX
 	MOVQ	R12, SP
+	// Back to Go world, set special registers.
+	// The g register (R14) is preserved in C.
+	XORPS	X15, X15
 	RET
 
 // C->Go callback thunk that allows to call runtime·racesymbolize from C code.
@@ -419,7 +434,9 @@ call:
 // The overall effect of Go->C->Go call chain is similar to that of mcall.
 // RARG0 contains command code. RARG1 contains command-specific context.
 // See racecallback for command codes.
-TEXT	runtime·racecallbackthunk(SB), NOSPLIT, $56-8
+// Defined as ABIInternal so as to avoid introducing a wrapper,
+// because its address is passed to C via funcPC.
+TEXT	runtime·racecallbackthunk<ABIInternal>(SB), NOSPLIT, $56-8
 	// Handle command raceGetProcCmd (0) here.
 	// First, code below assumes that we are on curg, while raceGetProcCmd
 	// can be executed on g0. Second, it is called frequently, so will
@@ -447,12 +464,13 @@ rest:
 	PUSHQ	R15
 	// Set g = g0.
 	get_tls(R12)
-	MOVQ	g(R12), R13
-	MOVQ	g_m(R13), R14
-	MOVQ	m_g0(R14), R15
+	MOVQ	g(R12), R14
+	MOVQ	g_m(R14), R13
+	MOVQ	m_g0(R13), R15
 	CMPQ	R13, R15
 	JEQ	noswitch	// branch if already on g0
 	MOVQ	R15, g(R12)	// g = m->g0
+	MOVQ	R15, R14	// set g register
 	PUSHQ	RARG1	// func arg
 	PUSHQ	RARG0	// func arg
 	CALL	runtime·racecallback(SB)
diff --git a/src/runtime/runtime2.go b/src/runtime/runtime2.go
index 109f0da131..b7c7b4cff2 100644
--- a/src/runtime/runtime2.go
+++ b/src/runtime/runtime2.go
@@ -853,7 +853,7 @@ type funcinl struct {
 // layout of Itab known to compilers
 // allocated in non-garbage-collected memory
 // Needs to be in sync with
-// ../cmd/compile/internal/gc/reflect.go:/^func.dumptabs.
+// ../cmd/compile/internal/gc/reflect.go:/^func.WriteTabs.
 type itab struct {
 	inter *interfacetype
 	_type *_type
diff --git a/src/runtime/signal_amd64.go b/src/runtime/signal_amd64.go
index 6ab1f758c2..3eeb5e044f 100644
--- a/src/runtime/signal_amd64.go
+++ b/src/runtime/signal_amd64.go
@@ -65,11 +65,14 @@ func (c *sigctxt) preparePanic(sig uint32, gp *g) {
 	pc := uintptr(c.rip())
 	sp := uintptr(c.rsp())
 
+	// In case we are panicking from external code, we need to initialize
+	// Go special registers. We inject sigpanic0 (instead of sigpanic),
+	// which takes care of that.
 	if shouldPushSigpanic(gp, pc, *(*uintptr)(unsafe.Pointer(sp))) {
-		c.pushCall(funcPC(sigpanic), pc)
+		c.pushCall(funcPC(sigpanic0), pc)
 	} else {
 		// Not safe to push the call. Just clobber the frame.
-		c.set_rip(uint64(funcPC(sigpanic)))
+		c.set_rip(uint64(funcPC(sigpanic0)))
 	}
 }
 
diff --git a/src/runtime/stubs.go b/src/runtime/stubs.go
index 2ee2c74dfe..c0cc95ec65 100644
--- a/src/runtime/stubs.go
+++ b/src/runtime/stubs.go
@@ -4,7 +4,10 @@
 
 package runtime
 
-import "unsafe"
+import (
+	"internal/abi"
+	"unsafe"
+)
 
 // Should be a built-in for unsafe.Pointer?
 //go:nosplit
@@ -167,7 +170,6 @@ func noescape(p unsafe.Pointer) unsafe.Pointer {
 // pointer-declared arguments.
 func cgocallback(fn, frame, ctxt uintptr)
 func gogo(buf *gobuf)
-func gosave(buf *gobuf)
 
 //go:noescape
 func jmpdefer(fv *funcval, argp uintptr)
@@ -175,19 +177,50 @@ func asminit()
 func setg(gg *g)
 func breakpoint()
 
-// reflectcall calls fn with a copy of the n argument bytes pointed at by arg.
-// After fn returns, reflectcall copies n-retoffset result bytes
-// back into arg+retoffset before returning. If copying result bytes back,
-// the caller should pass the argument frame type as argtype, so that
-// call can execute appropriate write barriers during the copy.
+// reflectcall calls fn with arguments described by stackArgs, stackArgsSize,
+// frameSize, and regArgs.
 //
-// Package reflect always passes a frame type. In package runtime,
-// Windows callbacks are the only use of this that copies results
-// back, and those cannot have pointers in their results, so runtime
-// passes nil for the frame type.
+// Arguments passed on the stack and space for return values passed on the stack
+// must be laid out at the space pointed to by stackArgs (with total length
+// stackArgsSize) according to the ABI.
+//
+// stackRetOffset must be some value <= stackArgsSize that indicates the
+// offset within stackArgs where the return value space begins.
+//
+// frameSize is the total size of the argument frame at stackArgs and must
+// therefore be >= stackArgsSize. It must include additional space for spilling
+// register arguments for stack growth and preemption.
+//
+// TODO(mknyszek): Once we don't need the additional spill space, remove frameSize,
+// since frameSize will be redundant with stackArgsSize.
+//
+// Arguments passed in registers must be laid out in regArgs according to the ABI.
+// regArgs will hold any return values passed in registers after the call.
+//
+// reflectcall copies stack arguments from stackArgs to the goroutine stack, and
+// then copies back stackArgsSize-stackRetOffset bytes back to the return space
+// in stackArgs once fn has completed. It also "unspills" argument registers from
+// regArgs before calling fn, and spills them back into regArgs immediately
+// following the call to fn. If there are results being returned on the stack,
+// the caller should pass the argument frame type as stackArgsType so that
+// reflectcall can execute appropriate write barriers during the copy.
+//
+// reflectcall expects regArgs.ReturnIsPtr to be populated indicating which
+// registers on the return path will contain Go pointers. It will then store
+// these pointers in regArgs.Ptrs such that they are visible to the GC.
+//
+// Package reflect passes a frame type. In package runtime, there is only
+// one call that copies results back, in callbackWrap in syscall_windows.go, and it
+// does NOT pass a frame type, meaning there are no write barriers invoked. See that
+// call site for justification.
 //
 // Package reflect accesses this symbol through a linkname.
-func reflectcall(argtype *_type, fn, arg unsafe.Pointer, argsize uint32, retoffset uint32)
+//
+// Arguments passed through to reflectcall do not escape. The type is used
+// only in a very limited callee of reflectcall, the stackArgs are copied, and
+// regArgs is only used in the reflectcall frame.
+//go:noescape
+func reflectcall(stackArgsType *_type, fn, stackArgs unsafe.Pointer, stackArgsSize, stackRetOffset, frameSize uint32, regArgs *abi.RegArgs)
 
 func procyield(cycles uint32)
 
@@ -357,3 +390,7 @@ func duffcopy()
 
 // Called from linker-generated .initarray; declared for go vet; do NOT call from Go.
 func addmoduledata()
+
+// Injected by the signal handler for panicking signals. On many platforms it just
+// jumps to sigpanic.
+func sigpanic0()
diff --git a/src/runtime/sys_darwin_amd64.s b/src/runtime/sys_darwin_amd64.s
index 630fb5df64..0fe8c7e172 100644
--- a/src/runtime/sys_darwin_amd64.s
+++ b/src/runtime/sys_darwin_amd64.s
@@ -5,6 +5,8 @@
 // System calls and other sys.stuff for AMD64, Darwin
 // System calls are implemented in libSystem, this file contains
 // trampolines that convert from Go to C calling convention.
+// The trampolines are ABIInternal as they are referenced from
+// Go code with funcPC.
 
 #include "go_asm.h"
 #include "go_tls.h"
@@ -13,7 +15,7 @@
 #define CLOCK_REALTIME		0
 
 // Exit the entire program (like C exit)
-TEXT runtime·exit_trampoline(SB),NOSPLIT,$0
+TEXT runtime·exit_trampoline<ABIInternal>(SB),NOSPLIT,$0
 	PUSHQ	BP
 	MOVQ	SP, BP
 	MOVL	0(DI), DI		// arg 1 exit status
@@ -22,7 +24,7 @@ TEXT runtime·exit_trampoline(SB),NOSPLIT,$0
 	POPQ	BP
 	RET
 
-TEXT runtime·open_trampoline(SB),NOSPLIT,$0
+TEXT runtime·open_trampoline<ABIInternal>(SB),NOSPLIT,$0
 	PUSHQ	BP
 	MOVQ	SP, BP
 	MOVL	8(DI), SI		// arg 2 flags
@@ -33,7 +35,7 @@ TEXT runtime·open_trampoline(SB),NOSPLIT,$0
 	POPQ	BP
 	RET
 
-TEXT runtime·close_trampoline(SB),NOSPLIT,$0
+TEXT runtime·close_trampoline<ABIInternal>(SB),NOSPLIT,$0
 	PUSHQ	BP
 	MOVQ	SP, BP
 	MOVL	0(DI), DI		// arg 1 fd
@@ -41,7 +43,7 @@ TEXT runtime·close_trampoline(SB),NOSPLIT,$0
 	POPQ	BP
 	RET
 
-TEXT runtime·read_trampoline(SB),NOSPLIT,$0
+TEXT runtime·read_trampoline<ABIInternal>(SB),NOSPLIT,$0
 	PUSHQ	BP
 	MOVQ	SP, BP
 	MOVQ	8(DI), SI		// arg 2 buf
@@ -57,7 +59,7 @@ noerr:
 	POPQ	BP
 	RET
 
-TEXT runtime·write_trampoline(SB),NOSPLIT,$0
+TEXT runtime·write_trampoline<ABIInternal>(SB),NOSPLIT,$0
 	PUSHQ	BP
 	MOVQ	SP, BP
 	MOVQ	8(DI), SI		// arg 2 buf
@@ -73,7 +75,7 @@ noerr:
 	POPQ	BP
 	RET
 
-TEXT runtime·pipe_trampoline(SB),NOSPLIT,$0
+TEXT runtime·pipe_trampoline<ABIInternal>(SB),NOSPLIT,$0
 	PUSHQ	BP
 	MOVQ	SP, BP
 	CALL	libc_pipe(SB)		// pointer already in DI
@@ -84,7 +86,7 @@ TEXT runtime·pipe_trampoline(SB),NOSPLIT,$0
 	POPQ	BP
 	RET
 
-TEXT runtime·setitimer_trampoline(SB),NOSPLIT,$0
+TEXT runtime·setitimer_trampoline<ABIInternal>(SB),NOSPLIT,$0
 	PUSHQ	BP
 	MOVQ	SP, BP
 	MOVQ	8(DI), SI		// arg 2 new
@@ -94,7 +96,7 @@ TEXT runtime·setitimer_trampoline(SB),NOSPLIT,$0
 	POPQ	BP
 	RET
 
-TEXT runtime·madvise_trampoline(SB), NOSPLIT, $0
+TEXT runtime·madvise_trampoline<ABIInternal>(SB), NOSPLIT, $0
 	PUSHQ	BP
 	MOVQ	SP, BP
 	MOVQ	8(DI), SI	// arg 2 len
@@ -105,12 +107,12 @@ TEXT runtime·madvise_trampoline(SB), NOSPLIT, $0
 	POPQ	BP
 	RET
 
-TEXT runtime·mlock_trampoline(SB), NOSPLIT, $0
+TEXT runtime·mlock_trampoline<ABIInternal>(SB), NOSPLIT, $0
 	UNDEF // unimplemented
 
 GLOBL timebase<>(SB),NOPTR,$(machTimebaseInfo__size)
 
-TEXT runtime·nanotime_trampoline(SB),NOSPLIT,$0
+TEXT runtime·nanotime_trampoline<ABIInternal>(SB),NOSPLIT,$0
 	PUSHQ	BP
 	MOVQ	SP, BP
 	MOVQ	DI, BX
@@ -139,7 +141,7 @@ initialized:
 	POPQ	BP
 	RET
 
-TEXT runtime·walltime_trampoline(SB),NOSPLIT,$0
+TEXT runtime·walltime_trampoline<ABIInternal>(SB),NOSPLIT,$0
 	PUSHQ	BP			// make a frame; keep stack aligned
 	MOVQ	SP, BP
 	MOVQ	DI, SI			// arg 2 timespec
@@ -148,7 +150,7 @@ TEXT runtime·walltime_trampoline(SB),NOSPLIT,$0
 	POPQ	BP
 	RET
 
-TEXT runtime·sigaction_trampoline(SB),NOSPLIT,$0
+TEXT runtime·sigaction_trampoline<ABIInternal>(SB),NOSPLIT,$0
 	PUSHQ	BP
 	MOVQ	SP, BP
 	MOVQ	8(DI), SI		// arg 2 new
@@ -161,7 +163,7 @@ TEXT runtime·sigaction_trampoline(SB),NOSPLIT,$0
 	POPQ	BP
 	RET
 
-TEXT runtime·sigprocmask_trampoline(SB),NOSPLIT,$0
+TEXT runtime·sigprocmask_trampoline<ABIInternal>(SB),NOSPLIT,$0
 	PUSHQ	BP
 	MOVQ	SP, BP
 	MOVQ	8(DI), SI	// arg 2 new
@@ -174,7 +176,7 @@ TEXT runtime·sigprocmask_trampoline(SB),NOSPLIT,$0
 	POPQ	BP
 	RET
 
-TEXT runtime·sigaltstack_trampoline(SB),NOSPLIT,$0
+TEXT runtime·sigaltstack_trampoline<ABIInternal>(SB),NOSPLIT,$0
 	PUSHQ	BP
 	MOVQ	SP, BP
 	MOVQ	8(DI), SI		// arg 2 old
@@ -186,7 +188,7 @@ TEXT runtime·sigaltstack_trampoline(SB),NOSPLIT,$0
 	POPQ	BP
 	RET
 
-TEXT runtime·raiseproc_trampoline(SB),NOSPLIT,$0
+TEXT runtime·raiseproc_trampoline<ABIInternal>(SB),NOSPLIT,$0
 	PUSHQ	BP
 	MOVQ	SP, BP
 	MOVL	0(DI), BX	// signal
@@ -212,7 +214,7 @@ TEXT runtime·sigfwd(SB),NOSPLIT,$0-32
 
 // This is the function registered during sigaction and is invoked when
 // a signal is received. It just redirects to the Go function sigtrampgo.
-TEXT runtime·sigtramp(SB),NOSPLIT,$0
+TEXT runtime·sigtramp<ABIInternal>(SB),NOSPLIT,$0
 	// This runs on the signal stack, so we have lots of stack available.
 	// We allocate our own stack space, because if we tell the linker
 	// how much we're using, the NOSPLIT check fails.
@@ -246,7 +248,7 @@ TEXT runtime·sigtramp(SB),NOSPLIT,$0
 
 // Used instead of sigtramp in programs that use cgo.
 // Arguments from kernel are in DI, SI, DX.
-TEXT runtime·cgoSigtramp(SB),NOSPLIT,$0
+TEXT runtime·cgoSigtramp<ABIInternal>(SB),NOSPLIT,$0
 	// If no traceback function, do usual sigtramp.
 	MOVQ	runtime·cgoTraceback(SB), AX
 	TESTQ	AX, AX
@@ -289,12 +291,12 @@ TEXT runtime·cgoSigtramp(SB),NOSPLIT,$0
 	// The first three arguments, and the fifth, are already in registers.
 	// Set the two remaining arguments now.
 	MOVQ	runtime·cgoTraceback(SB), CX
-	MOVQ	$runtime·sigtramp(SB), R9
+	MOVQ	$runtime·sigtramp<ABIInternal>(SB), R9
 	MOVQ	_cgo_callers(SB), AX
 	JMP	AX
 
 sigtramp:
-	JMP	runtime·sigtramp(SB)
+	JMP	runtime·sigtramp<ABIInternal>(SB)
 
 sigtrampnog:
 	// Signal arrived on a non-Go thread. If this is SIGPROF, get a
@@ -320,7 +322,7 @@ sigtrampnog:
 	MOVQ	_cgo_callers(SB), AX
 	JMP	AX
 
-TEXT runtime·mmap_trampoline(SB),NOSPLIT,$0
+TEXT runtime·mmap_trampoline<ABIInternal>(SB),NOSPLIT,$0
 	PUSHQ	BP			// make a frame; keep stack aligned
 	MOVQ	SP, BP
 	MOVQ	DI, BX
@@ -343,7 +345,7 @@ ok:
 	POPQ	BP
 	RET
 
-TEXT runtime·munmap_trampoline(SB),NOSPLIT,$0
+TEXT runtime·munmap_trampoline<ABIInternal>(SB),NOSPLIT,$0
 	PUSHQ	BP
 	MOVQ	SP, BP
 	MOVQ	8(DI), SI		// arg 2 len
@@ -355,7 +357,7 @@ TEXT runtime·munmap_trampoline(SB),NOSPLIT,$0
 	POPQ	BP
 	RET
 
-TEXT runtime·usleep_trampoline(SB),NOSPLIT,$0
+TEXT runtime·usleep_trampoline<ABIInternal>(SB),NOSPLIT,$0
 	PUSHQ	BP
 	MOVQ	SP, BP
 	MOVL	0(DI), DI	// arg 1 usec
@@ -367,7 +369,7 @@ TEXT runtime·settls(SB),NOSPLIT,$32
 	// Nothing to do on Darwin, pthread already set thread-local storage up.
 	RET
 
-TEXT runtime·sysctl_trampoline(SB),NOSPLIT,$0
+TEXT runtime·sysctl_trampoline<ABIInternal>(SB),NOSPLIT,$0
 	PUSHQ	BP
 	MOVQ	SP, BP
 	MOVL	8(DI), SI		// arg 2 miblen
@@ -380,7 +382,7 @@ TEXT runtime·sysctl_trampoline(SB),NOSPLIT,$0
 	POPQ	BP
 	RET
 
-TEXT runtime·sysctlbyname_trampoline(SB),NOSPLIT,$0
+TEXT runtime·sysctlbyname_trampoline<ABIInternal>(SB),NOSPLIT,$0
 	PUSHQ	BP
 	MOVQ	SP, BP
 	MOVQ	8(DI), SI		// arg 2 oldp
@@ -392,14 +394,14 @@ TEXT runtime·sysctlbyname_trampoline(SB),NOSPLIT,$0
 	POPQ	BP
 	RET
 
-TEXT runtime·kqueue_trampoline(SB),NOSPLIT,$0
+TEXT runtime·kqueue_trampoline<ABIInternal>(SB),NOSPLIT,$0
 	PUSHQ	BP
 	MOVQ	SP, BP
 	CALL	libc_kqueue(SB)
 	POPQ	BP
 	RET
 
-TEXT runtime·kevent_trampoline(SB),NOSPLIT,$0
+TEXT runtime·kevent_trampoline<ABIInternal>(SB),NOSPLIT,$0
 	PUSHQ	BP
 	MOVQ	SP, BP
 	MOVQ	8(DI), SI		// arg 2 keventt
@@ -418,7 +420,7 @@ ok:
 	POPQ	BP
 	RET
 
-TEXT runtime·fcntl_trampoline(SB),NOSPLIT,$0
+TEXT runtime·fcntl_trampoline<ABIInternal>(SB),NOSPLIT,$0
 	PUSHQ	BP
 	MOVQ	SP, BP
 	MOVL	4(DI), SI		// arg 2 cmd
@@ -475,7 +477,7 @@ TEXT runtime·mstart_stub(SB),NOSPLIT,$0
 // A pointer to the arguments is passed in DI.
 // A single int32 result is returned in AX.
 // (For more results, make an args/results structure.)
-TEXT runtime·pthread_attr_init_trampoline(SB),NOSPLIT,$0
+TEXT runtime·pthread_attr_init_trampoline<ABIInternal>(SB),NOSPLIT,$0
 	PUSHQ	BP	// make frame, keep stack 16-byte aligned.
 	MOVQ	SP, BP
 	MOVQ	0(DI), DI // arg 1 attr
@@ -483,7 +485,7 @@ TEXT runtime·pthread_attr_init_trampoline(SB),NOSPLIT,$0
 	POPQ	BP
 	RET
 
-TEXT runtime·pthread_attr_getstacksize_trampoline(SB),NOSPLIT,$0
+TEXT runtime·pthread_attr_getstacksize_trampoline<ABIInternal>(SB),NOSPLIT,$0
 	PUSHQ	BP
 	MOVQ	SP, BP
 	MOVQ	8(DI), SI	// arg 2 size
@@ -492,7 +494,7 @@ TEXT runtime·pthread_attr_getstacksize_trampoline(SB),NOSPLIT,$0
 	POPQ	BP
 	RET
 
-TEXT runtime·pthread_attr_setdetachstate_trampoline(SB),NOSPLIT,$0
+TEXT runtime·pthread_attr_setdetachstate_trampoline<ABIInternal>(SB),NOSPLIT,$0
 	PUSHQ	BP
 	MOVQ	SP, BP
 	MOVQ	8(DI), SI	// arg 2 state
@@ -501,7 +503,7 @@ TEXT runtime·pthread_attr_setdetachstate_trampoline(SB),NOSPLIT,$0
 	POPQ	BP
 	RET
 
-TEXT runtime·pthread_create_trampoline(SB),NOSPLIT,$0
+TEXT runtime·pthread_create_trampoline<ABIInternal>(SB),NOSPLIT,$0
 	PUSHQ	BP
 	MOVQ	SP, BP
 	SUBQ	$16, SP
@@ -514,7 +516,7 @@ TEXT runtime·pthread_create_trampoline(SB),NOSPLIT,$0
 	POPQ	BP
 	RET
 
-TEXT runtime·raise_trampoline(SB),NOSPLIT,$0
+TEXT runtime·raise_trampoline<ABIInternal>(SB),NOSPLIT,$0
 	PUSHQ	BP
 	MOVQ	SP, BP
 	MOVL	0(DI), DI	// arg 1 signal
@@ -522,7 +524,7 @@ TEXT runtime·raise_trampoline(SB),NOSPLIT,$0
 	POPQ	BP
 	RET
 
-TEXT runtime·pthread_mutex_init_trampoline(SB),NOSPLIT,$0
+TEXT runtime·pthread_mutex_init_trampoline<ABIInternal>(SB),NOSPLIT,$0
 	PUSHQ	BP
 	MOVQ	SP, BP
 	MOVQ	8(DI), SI	// arg 2 attr
@@ -531,7 +533,7 @@ TEXT runtime·pthread_mutex_init_trampoline(SB),NOSPLIT,$0
 	POPQ	BP
 	RET
 
-TEXT runtime·pthread_mutex_lock_trampoline(SB),NOSPLIT,$0
+TEXT runtime·pthread_mutex_lock_trampoline<ABIInternal>(SB),NOSPLIT,$0
 	PUSHQ	BP
 	MOVQ	SP, BP
 	MOVQ	0(DI), DI	// arg 1 mutex
@@ -539,7 +541,7 @@ TEXT runtime·pthread_mutex_lock_trampoline(SB),NOSPLIT,$0
 	POPQ	BP
 	RET
 
-TEXT runtime·pthread_mutex_unlock_trampoline(SB),NOSPLIT,$0
+TEXT runtime·pthread_mutex_unlock_trampoline<ABIInternal>(SB),NOSPLIT,$0
 	PUSHQ	BP
 	MOVQ	SP, BP
 	MOVQ	0(DI), DI	// arg 1 mutex
@@ -547,7 +549,7 @@ TEXT runtime·pthread_mutex_unlock_trampoline(SB),NOSPLIT,$0
 	POPQ	BP
 	RET
 
-TEXT runtime·pthread_cond_init_trampoline(SB),NOSPLIT,$0
+TEXT runtime·pthread_cond_init_trampoline<ABIInternal>(SB),NOSPLIT,$0
 	PUSHQ	BP
 	MOVQ	SP, BP
 	MOVQ	8(DI), SI	// arg 2 attr
@@ -556,7 +558,7 @@ TEXT runtime·pthread_cond_init_trampoline(SB),NOSPLIT,$0
 	POPQ	BP
 	RET
 
-TEXT runtime·pthread_cond_wait_trampoline(SB),NOSPLIT,$0
+TEXT runtime·pthread_cond_wait_trampoline<ABIInternal>(SB),NOSPLIT,$0
 	PUSHQ	BP
 	MOVQ	SP, BP
 	MOVQ	8(DI), SI	// arg 2 mutex
@@ -565,7 +567,7 @@ TEXT runtime·pthread_cond_wait_trampoline(SB),NOSPLIT,$0
 	POPQ	BP
 	RET
 
-TEXT runtime·pthread_cond_timedwait_relative_np_trampoline(SB),NOSPLIT,$0
+TEXT runtime·pthread_cond_timedwait_relative_np_trampoline<ABIInternal>(SB),NOSPLIT,$0
 	PUSHQ	BP
 	MOVQ	SP, BP
 	MOVQ	8(DI), SI	// arg 2 mutex
@@ -575,7 +577,7 @@ TEXT runtime·pthread_cond_timedwait_relative_np_trampoline(SB),NOSPLIT,$0
 	POPQ	BP
 	RET
 
-TEXT runtime·pthread_cond_signal_trampoline(SB),NOSPLIT,$0
+TEXT runtime·pthread_cond_signal_trampoline<ABIInternal>(SB),NOSPLIT,$0
 	PUSHQ	BP
 	MOVQ	SP, BP
 	MOVQ	0(DI), DI	// arg 1 cond
@@ -583,7 +585,7 @@ TEXT runtime·pthread_cond_signal_trampoline(SB),NOSPLIT,$0
 	POPQ	BP
 	RET
 
-TEXT runtime·pthread_self_trampoline(SB),NOSPLIT,$0
+TEXT runtime·pthread_self_trampoline<ABIInternal>(SB),NOSPLIT,$0
 	PUSHQ	BP
 	MOVQ	SP, BP
 	MOVQ	DI, BX		// BX is caller-save
@@ -592,7 +594,7 @@ TEXT runtime·pthread_self_trampoline(SB),NOSPLIT,$0
 	POPQ	BP
 	RET
 
-TEXT runtime·pthread_kill_trampoline(SB),NOSPLIT,$0
+TEXT runtime·pthread_kill_trampoline<ABIInternal>(SB),NOSPLIT,$0
 	PUSHQ	BP
 	MOVQ	SP, BP
 	MOVQ	8(DI), SI	// arg 2 sig
@@ -617,7 +619,7 @@ TEXT runtime·pthread_kill_trampoline(SB),NOSPLIT,$0
 //
 // syscall expects a 32-bit result and tests for 32-bit -1
 // to decide there was an error.
-TEXT runtime·syscall(SB),NOSPLIT,$0
+TEXT runtime·syscall<ABIInternal>(SB),NOSPLIT,$0
 	PUSHQ	BP
 	MOVQ	SP, BP
 	SUBQ	$16, SP
@@ -667,7 +669,7 @@ ok:
 //
 // syscallX is like syscall but expects a 64-bit result
 // and tests for 64-bit -1 to decide there was an error.
-TEXT runtime·syscallX(SB),NOSPLIT,$0
+TEXT runtime·syscallX<ABIInternal>(SB),NOSPLIT,$0
 	PUSHQ	BP
 	MOVQ	SP, BP
 	SUBQ	$16, SP
@@ -703,7 +705,7 @@ ok:
 
 // syscallPtr is like syscallX except that the libc function reports an
 // error by returning NULL and setting errno.
-TEXT runtime·syscallPtr(SB),NOSPLIT,$0
+TEXT runtime·syscallPtr<ABIInternal>(SB),NOSPLIT,$0
 	PUSHQ	BP
 	MOVQ	SP, BP
 	SUBQ	$16, SP
@@ -756,7 +758,7 @@ ok:
 //
 // syscall6 expects a 32-bit result and tests for 32-bit -1
 // to decide there was an error.
-TEXT runtime·syscall6(SB),NOSPLIT,$0
+TEXT runtime·syscall6<ABIInternal>(SB),NOSPLIT,$0
 	PUSHQ	BP
 	MOVQ	SP, BP
 	SUBQ	$16, SP
@@ -809,7 +811,7 @@ ok:
 //
 // syscall6X is like syscall6 but expects a 64-bit result
 // and tests for 64-bit -1 to decide there was an error.
-TEXT runtime·syscall6X(SB),NOSPLIT,$0
+TEXT runtime·syscall6X<ABIInternal>(SB),NOSPLIT,$0
 	PUSHQ	BP
 	MOVQ	SP, BP
 	SUBQ	$16, SP
@@ -845,7 +847,7 @@ ok:
 
 // syscallNoErr is like syscall6 but does not check for errors, and
 // only returns one value, for use with standard C ABI library functions.
-TEXT runtime·syscallNoErr(SB),NOSPLIT,$0
+TEXT runtime·syscallNoErr<ABIInternal>(SB),NOSPLIT,$0
 	PUSHQ	BP
 	MOVQ	SP, BP
 	SUBQ	$16, SP
diff --git a/src/runtime/sys_linux_amd64.s b/src/runtime/sys_linux_amd64.s
index 37cb8dad03..d48573c2c5 100644
--- a/src/runtime/sys_linux_amd64.s
+++ b/src/runtime/sys_linux_amd64.s
@@ -215,9 +215,13 @@ TEXT runtime·walltime1(SB),NOSPLIT,$16-12
 
 	MOVQ	SP, R12	// Save old SP; R12 unchanged by C code.
 
+#ifdef GOEXPERIMENT_REGABI
+	MOVQ	g_m(R14), BX // BX unchanged by C code.
+#else
 	get_tls(CX)
 	MOVQ	g(CX), AX
 	MOVQ	g_m(AX), BX // BX unchanged by C code.
+#endif
 
 	// Set vdsoPC and vdsoSP for SIGPROF traceback.
 	// Save the old values on stack and restore them on exit,
@@ -232,7 +236,11 @@ TEXT runtime·walltime1(SB),NOSPLIT,$16-12
 	MOVQ	CX, m_vdsoPC(BX)
 	MOVQ	DX, m_vdsoSP(BX)
 
+#ifdef GOEXPERIMENT_REGABI
+	CMPQ	R14, m_curg(BX)	// Only switch if on curg.
+#else
 	CMPQ	AX, m_curg(BX)	// Only switch if on curg.
+#endif
 	JNE	noswitch
 
 	MOVQ	m_g0(BX), DX
@@ -275,9 +283,13 @@ TEXT runtime·nanotime1(SB),NOSPLIT,$16-8
 
 	MOVQ	SP, R12	// Save old SP; R12 unchanged by C code.
 
+#ifdef GOEXPERIMENT_REGABI
+	MOVQ	g_m(R14), BX // BX unchanged by C code.
+#else
 	get_tls(CX)
 	MOVQ	g(CX), AX
 	MOVQ	g_m(AX), BX // BX unchanged by C code.
+#endif
 
 	// Set vdsoPC and vdsoSP for SIGPROF traceback.
 	// Save the old values on stack and restore them on exit,
@@ -292,7 +304,11 @@ TEXT runtime·nanotime1(SB),NOSPLIT,$16-8
 	MOVQ	CX, m_vdsoPC(BX)
 	MOVQ	DX, m_vdsoSP(BX)
 
+#ifdef GOEXPERIMENT_REGABI
+	CMPQ	R14, m_curg(BX)	// Only switch if on curg.
+#else
 	CMPQ	AX, m_curg(BX)	// Only switch if on curg.
+#endif
 	JNE	noswitch
 
 	MOVQ	m_g0(BX), DX
@@ -632,6 +648,7 @@ nog1:
 	get_tls(CX)
 	MOVQ	R13, g_m(R9)
 	MOVQ	R9, g(CX)
+	MOVQ	R9, R14 // set g register
 	CALL	runtime·stackcheck(SB)
 
 nog2:
diff --git a/src/runtime/syscall2_solaris.go b/src/runtime/syscall2_solaris.go
index e098e8006a..3310489202 100644
--- a/src/runtime/syscall2_solaris.go
+++ b/src/runtime/syscall2_solaris.go
@@ -15,7 +15,6 @@ import _ "unsafe" // for go:linkname
 //go:cgo_import_dynamic libc_gethostname gethostname "libc.so"
 //go:cgo_import_dynamic libc_getpid getpid "libc.so"
 //go:cgo_import_dynamic libc_ioctl ioctl "libc.so"
-//go:cgo_import_dynamic libc_pipe pipe "libc.so"
 //go:cgo_import_dynamic libc_setgid setgid "libc.so"
 //go:cgo_import_dynamic libc_setgroups setgroups "libc.so"
 //go:cgo_import_dynamic libc_setsid setsid "libc.so"
@@ -33,7 +32,6 @@ import _ "unsafe" // for go:linkname
 //go:linkname libc_gethostname libc_gethostname
 //go:linkname libc_getpid libc_getpid
 //go:linkname libc_ioctl libc_ioctl
-//go:linkname libc_pipe libc_pipe
 //go:linkname libc_setgid libc_setgid
 //go:linkname libc_setgroups libc_setgroups
 //go:linkname libc_setsid libc_setsid
diff --git a/src/runtime/syscall_windows.go b/src/runtime/syscall_windows.go
index 7835b492f7..add40bb0b3 100644
--- a/src/runtime/syscall_windows.go
+++ b/src/runtime/syscall_windows.go
@@ -5,6 +5,7 @@
 package runtime
 
 import (
+	"internal/abi"
 	"runtime/internal/sys"
 	"unsafe"
 )
@@ -242,7 +243,11 @@ func callbackWrap(a *callbackArgs) {
 
 	// Even though this is copying back results, we can pass a nil
 	// type because those results must not require write barriers.
-	reflectcall(nil, unsafe.Pointer(c.fn), noescape(goArgs), uint32(c.retOffset)+sys.PtrSize, uint32(c.retOffset))
+	//
+	// Pass a dummy RegArgs for now.
+	// TODO(mknyszek): Pass arguments in registers.
+	var regs abi.RegArgs
+	reflectcall(nil, unsafe.Pointer(c.fn), noescape(goArgs), uint32(c.retOffset)+sys.PtrSize, uint32(c.retOffset), uint32(c.retOffset)+sys.PtrSize, &regs)
 
 	// Extract the result.
 	a.result = *(*uintptr)(unsafe.Pointer(&frame[c.retOffset]))
diff --git a/src/runtime/textflag.h b/src/runtime/textflag.h
index daca36d948..e727208cd0 100644
--- a/src/runtime/textflag.h
+++ b/src/runtime/textflag.h
@@ -35,3 +35,5 @@
 // Function is the top of the call stack. Call stack unwinders should stop
 // at this function.
 #define TOPFRAME 2048
+// Function is an ABI wrapper.
+#define ABIWRAPPER 4096
diff --git a/src/runtime/type.go b/src/runtime/type.go
index 81455f3532..18fc4bbfad 100644
--- a/src/runtime/type.go
+++ b/src/runtime/type.go
@@ -383,7 +383,7 @@ type maptype struct {
 }
 
 // Note: flag values must match those used in the TMAP case
-// in ../cmd/compile/internal/gc/reflect.go:dtypesym.
+// in ../cmd/compile/internal/gc/reflect.go:writeType.
 func (mt *maptype) indirectkey() bool { // store ptr to key instead of key itself
 	return mt.flags&1 != 0
 }
author	Cherry Zhang <cherryyz@google.com>	2021-02-16 16:55:30 -0500
committer	Cherry Zhang <cherryyz@google.com>	2021-02-16 17:18:37 -0500
commit	84825599dc2592e9e9f5489323fa7b47534fc391 (patch)
tree	e5aa6d21506d430cae87e63701e89822353d0ec8 /src/runtime
parent	5faf941df067b33485edb9cd2e880869e7feb6a3 (diff)
parent	d3cd4830adf45ce53c586a83f9d78421484737fd (diff)
download	go-84825599dc2592e9e9f5489323fa7b47534fc391.tar.xz