runtime: make runtime·usleep and runtime·osyield callable from cgo callback

runtime·usleep and runtime·osyield fall back to calling an
assembly wrapper for the libc functions in the absence of a m,
so they can be called in cgo callback context.

LGTM=rsc
R=minux.ma, rsc
CC=dave, golang-codereviews
https://golang.org/cl/102620044

src/pkg/runtime/os_solaris.c

@@ -568,10 +568,13 @@ runtime·sysconf(int32 name)
 	return runtime·sysvicall6(libc·sysconf, 1, (uintptr)name);
 }
 
+extern void runtime·usleep1(uint32);
+
+#pragma textflag NOSPLIT
 void
-runtime·usleep(uint32 us)
+runtime·usleep(uint32 µs)
 {
-	runtime·sysvicall6(libc·usleep, 1, (uintptr)us);
+	runtime·usleep1(µs);
 }
 
 int32
@@ -580,8 +583,17 @@ runtime·write(uintptr fd, void* buf, int32 nbyte)
 	return runtime·sysvicall6(libc·write, 3, (uintptr)fd, (uintptr)buf, (uintptr)nbyte);
 }
 
+extern void runtime·osyield1(void);
+
+#pragma textflag NOSPLIT
 void
 runtime·osyield(void)
 {
-	runtime·sysvicall6(libc·sched_yield, 0);
+	// Check the validity of m because we might be called in cgo callback
+	// path early enough where there isn't a m available yet.
+	if(g && g->m != nil) {
+		runtime·sysvicall6(libc·sched_yield, 0);
+		return;
+	}
+	runtime·osyield1();
 }

src/pkg/runtime/sys_solaris_amd64.s

@@ -270,3 +270,55 @@ exit:
 
 	ADDQ    $184, SP
 	RET
+
+// Called from runtime·usleep (Go). Can be called on Go stack, on OS stack,
+// can also be called in cgo callback path without a g->m.
+TEXT runtime·usleep1(SB),NOSPLIT,$0
+	MOVL	us+0(FP), DI
+	MOVQ	$runtime·usleep2(SB), AX // to hide from 6l
+
+	// Execute call on m->g0.
+	get_tls(R15)
+	CMPQ	R15, $0
+	JE	usleep1_noswitch
+
+	MOVQ	g(R15), R13
+	CMPQ	R13, $0
+	JE	usleep1_noswitch
+	MOVQ	g_m(R13), R13
+	CMPQ	R13, $0
+	JE	usleep1_noswitch
+	// TODO(aram): do something about the cpu profiler here.
+
+	MOVQ	m_g0(R13), R14
+	CMPQ	g(R15), R14
+	JNE	usleep1_switch
+	// executing on m->g0 already
+	CALL	AX
+	RET
+
+usleep1_switch:
+	// Switch to m->g0 stack and back.
+	MOVQ	(g_sched+gobuf_sp)(R14), R14
+	MOVQ	SP, -8(R14)
+	LEAQ	-8(R14), SP
+	CALL	AX
+	MOVQ	0(SP), SP
+	RET
+
+usleep1_noswitch:
+	// Not a Go-managed thread. Do not switch stack.
+	CALL	AX
+	RET
+
+// Runs on OS stack. duration (in µs units) is in DI.
+TEXT runtime·usleep2(SB),NOSPLIT,$0
+	MOVQ	libc·usleep(SB), AX
+	CALL	AX
+	RET
+
+// Runs on OS stack, called from runtime·osyield.
+TEXT runtime·osyield1(SB),NOSPLIT,$0
+	MOVQ	libc·sched_yield(SB), AX
+	CALL	AX
+	RET