[dev.regabi] runtime: use g register in some assembly functions on AMD64

Now that we have a g register, just use it.

Note: functions that can be called from ABI0 context (e.g.
morestack) is unchanged. Functions that switch g is also
unchanged, because we need to set the new g in both the register
and TLS.

TODO: other OSes.

Change-Id: I692a82a7caa8417ff620a59676a6275f56747b94
Reviewed-on: https://go-review.googlesource.com/c/go/+/289718
Trust: Cherry Zhang <cherryyz@google.com>
Run-TryBot: Cherry Zhang <cherryyz@google.com>
TryBot-Result: Go Bot <gobot@golang.org>
Reviewed-by: Michael Knyszek <mknyszek@google.com>

src/runtime/asm_amd64.s

@@ -585,18 +585,20 @@ 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)
+#ifndef GOEXPERIMENT_REGABI
-	MOVQ	g(R8), R8
+	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	R9, (g_sched+gobuf_sp)(R14)
-	MOVQ	$0, (g_sched+gobuf_ret)(R8)
+	MOVQ	$0, (g_sched+gobuf_ret)(R14)
-	MOVQ	BP, (g_sched+gobuf_bp)(R8)
+	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)
@@ -1391,9 +1393,13 @@ TEXT runtime·gcWriteBarrier<ABIInternal>(SB),NOSPLIT,$112
 	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), R12
 	// Increment wbBuf.next position.

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

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