From 45cff65502ace2783f05cf27383d807f07627cf2 Mon Sep 17 00:00:00 2001
From: Alex Brainman <alex.brainman@gmail.com>
Date: Tue, 16 Jul 2013 12:36:05 +1000
Subject: [PATCH] runtime: switch to os stack in windows osyield and usleep

Fixes #5831

R=golang-dev, rsc
CC=golang-dev
https://golang.org/cl/11266043
---
 src/pkg/runtime/os_windows.c        | 17 +++++++++
 src/pkg/runtime/sys_windows_386.s   | 59 +++++++++++++++--------------
 src/pkg/runtime/sys_windows_amd64.s | 52 +++++++++++++++----------
 3 files changed, 79 insertions(+), 49 deletions(-)

diff --git a/src/pkg/runtime/os_windows.c b/src/pkg/runtime/os_windows.c
index 60160c3e16..5dcbd2c1e1 100644
--- a/src/pkg/runtime/os_windows.c
+++ b/src/pkg/runtime/os_windows.c
@@ -275,6 +275,23 @@ runtime·stdcall(void *fn, int32 count, ...)
 	return (void*)c.r1;
 }
 
+extern void runtime·usleep1(uint32);
+
+#pragma textflag 7
+void
+runtime·osyield(void)
+{
+	runtime·usleep1(1);
+}
+
+#pragma textflag 7
+void
+runtime·usleep(uint32 us)
+{
+	// Have 1us units; want 100ns units.
+	runtime·usleep1(10*us);
+}
+
 uint32
 runtime·issigpanic(uint32 code)
 {
diff --git a/src/pkg/runtime/sys_windows_386.s b/src/pkg/runtime/sys_windows_386.s
index d232bd115d..d6f1259818 100644
--- a/src/pkg/runtime/sys_windows_386.s
+++ b/src/pkg/runtime/sys_windows_386.s
@@ -327,28 +327,39 @@ TEXT runtime·remove_exception_handler(SB),7,$0
 
 	RET
 
-TEXT runtime·osyield(SB),7,$20
-	// Tried NtYieldExecution but it doesn't yield hard enough.
-	// NtWaitForSingleObject being used here as Sleep(0).
-	MOVL	runtime·NtWaitForSingleObject(SB), AX
-	MOVL	$-1, hi-4(SP)
-	MOVL	$-1, lo-8(SP)
-	LEAL	lo-8(SP), BX
-	MOVL	BX, ptime-12(SP)
-	MOVL	$0, alertable-16(SP)
-	MOVL	$-1, handle-20(SP)
-	MOVL	SP, BP
-	CALL	checkstack4<>(SB)
+// Sleep duration is in 100ns units.
+TEXT runtime·usleep1(SB),7,$0
+	MOVL	duration+0(FP), BX
+	MOVL	$runtime·usleep2(SB), AX // to hide from 8l
+
+	// Execute call on m->g0 stack, in case we are not actually
+	// calling a system call wrapper, like when running under WINE.
+	get_tls(CX)
+	CMPL	CX, $0
+	JNE	3(PC)
+	// Not a Go-managed thread. Do not switch stack.
 	CALL	AX
-	MOVL	BP, SP
 	RET
 
-TEXT runtime·usleep(SB),7,$20
-	MOVL	runtime·NtWaitForSingleObject(SB), AX 
-	// Have 1us units; need negative 100ns units.
-	// Assume multiply by 10 will not overflow 32-bit word.
-	MOVL	usec+0(FP), BX
-	IMULL	$10, BX
+	MOVL	m(CX), BP
+	MOVL	m_g0(BP), SI
+	CMPL	g(CX), SI
+	JNE	3(PC)
+	// executing on m->g0 already
+	CALL	AX
+	RET
+
+	// Switch to m->g0 stack and back.
+	MOVL	(g_sched+gobuf_sp)(SI), SI
+	MOVL	SP, -4(SI)
+	LEAL	-4(SI), SP
+	CALL	AX
+	MOVL	0(SP), SP
+	RET
+
+// Runs on OS stack. duration (in 100ns units) is in BX.
+TEXT runtime·usleep2(SB),7,$20
+	// Want negative 100ns units.
 	NEGL	BX
 	MOVL	$-1, hi-4(SP)
 	MOVL	BX, lo-8(SP)
@@ -357,15 +368,7 @@ TEXT runtime·usleep(SB),7,$20
 	MOVL	$0, alertable-16(SP)
 	MOVL	$-1, handle-20(SP)
 	MOVL	SP, BP
-	CALL	checkstack4<>(SB)
+	MOVL	runtime·NtWaitForSingleObject(SB), AX
 	CALL	AX
 	MOVL	BP, SP
 	RET
-
-// This function requires 4 bytes of stack,
-// to simulate what calling NtWaitForSingleObject will use.
-// (It is just a CALL to the system call dispatch.)
-// If the linker okays the call to checkstack4 (a NOSPLIT function)
-// then the call to NtWaitForSingleObject is okay too.
-TEXT checkstack4<>(SB),7,$4
-	RET
diff --git a/src/pkg/runtime/sys_windows_amd64.s b/src/pkg/runtime/sys_windows_amd64.s
index 4837a02a5a..a8953d5390 100644
--- a/src/pkg/runtime/sys_windows_amd64.s
+++ b/src/pkg/runtime/sys_windows_amd64.s
@@ -322,34 +322,44 @@ TEXT runtime·install_exception_handler(SB),7,$0
 TEXT runtime·remove_exception_handler(SB),7,$0
 	RET
 
-TEXT runtime·osyield(SB),7,$8
-	// Tried NtYieldExecution but it doesn't yield hard enough.
-	// NtWaitForSingleObject being used here as Sleep(0).
-	// The CALL is safe because NtXxx is a system call wrapper:
-	// it puts the right system call number in AX, then does
-	// a SYSENTER and a RET.
-	MOVQ	runtime·NtWaitForSingleObject(SB), AX
-	MOVQ	$1, BX
-	NEGQ	BX
-	MOVQ	SP, R8 // ptime
-	MOVQ	BX, (R8)
-	MOVQ	$-1, CX // handle
-	MOVQ	$0, DX // alertable
+// Sleep duration is in 100ns units.
+TEXT runtime·usleep1(SB),7,$0
+	MOVL	duration+0(FP), BX
+	MOVQ	$runtime·usleep2(SB), AX // to hide from 6l
+
+	// Execute call on m->g0 stack, in case we are not actually
+	// calling a system call wrapper, like when running under WINE.
+	get_tls(R15)
+	CMPQ	R15, $0
+	JNE	3(PC)
+	// Not a Go-managed thread. Do not switch stack.
 	CALL	AX
 	RET
 
-TEXT runtime·usleep(SB),7,$8
-	// The CALL is safe because NtXxx is a system call wrapper:
-	// it puts the right system call number in AX, then does
-	// a SYSENTER and a RET.
-	MOVQ	runtime·NtWaitForSingleObject(SB), AX
-	// Have 1us units; want negative 100ns units.
-	MOVL	usec+0(FP), BX
-	IMULQ	$10, BX
+	MOVQ	m(R15), R14
+	MOVQ	m_g0(R14), R14
+	CMPQ	g(R15), R14
+	JNE	3(PC)
+	// executing on m->g0 already
+	CALL	AX
+	RET
+
+	// 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
+
+// Runs on OS stack. duration (in 100ns units) is in BX.
+TEXT runtime·usleep2(SB),7,$8
+	// Want negative 100ns units.
 	NEGQ	BX
 	MOVQ	SP, R8 // ptime
 	MOVQ	BX, (R8)
 	MOVQ	$-1, CX // handle
 	MOVQ	$0, DX // alertable
+	MOVQ	runtime·NtWaitForSingleObject(SB), AX
 	CALL	AX
 	RET
-- 
2.48.1