cgo: fix g0 stack guard

Fixes crash when cgo consumes more than 8K
of stack and makes a callback.

Fixes #1328.

R=golang-dev, rogpeppe, rsc
CC=golang-dev, mpimenov
https://golang.org/cl/5371042

misc/cgo/test/callback_c.c

@@ -8,5 +8,10 @@
 void
 callback(void *f)
 {
+	// use some stack space
+	volatile char data[64*1024];
+
+	data[0] = 0;
 	goCallback(f);
+        data[sizeof(data)-1] = 0;
 }

src/pkg/runtime/386/asm.s

@@ -26,12 +26,14 @@ TEXT _rt0_386(SB),7,$0
 	// we set up GS ourselves.
 	MOVL	initcgo(SB), AX
 	TESTL	AX, AX
-	JZ	4(PC)
+	JZ	needtls
+	PUSHL	$runtime·g0(SB)
 	CALL	AX
+	POPL	AX
 	// skip runtime·ldt0setup(SB) and tls test after initcgo for non-windows
 	CMPL runtime·iswindows(SB), $0
 	JEQ ok
-
+needtls:
 	// skip runtime·ldt0setup(SB) and tls test on Plan 9 in all cases
 	CMPL	runtime·isplan9(SB), $1
 	JEQ	ok
@@ -58,9 +60,15 @@ ok:
 	MOVL	CX, m_g0(AX)
 
 	// create istack out of the OS stack
+	// if there is an initcgo, it had setup stackguard for us
+	MOVL	initcgo(SB), AX
+	TESTL	AX, AX
+	JNZ	stackok
 	LEAL	(-64*1024+104)(SP), AX	// TODO: 104?
 	MOVL	AX, g_stackguard(CX)
+stackok:
 	MOVL	SP, g_stackbase(CX)
+
 	CALL	runtime·emptyfunc(SB)	// fault if stack check is wrong
 
 	// convention is D is always cleared

src/pkg/runtime/amd64/asm.s

@@ -17,6 +17,7 @@ TEXT _rt0_amd64(SB),7,$-8
 	MOVQ	initcgo(SB), AX
 	TESTQ	AX, AX
 	JZ	needtls
+	LEAQ	runtime·g0(SB), DI
 	CALL	AX
 	CMPL	runtime·iswindows(SB), $0
 	JEQ ok
@@ -44,8 +45,13 @@ ok:
 	MOVQ	CX, m_g0(AX)
 
 	// create istack out of the given (operating system) stack
+	// if there is an initcgo, it had setup stackguard for us
+	MOVQ	initcgo(SB), AX
+	TESTQ	AX, AX
+	JNZ	stackok
 	LEAQ	(-8192+104)(SP), AX
 	MOVQ	AX, g_stackguard(CX)
+stackok:
 	MOVQ	SP, g_stackbase(CX)
 
 	CLD				// convention is D is always left cleared

src/pkg/runtime/cgo/darwin_386.c

@@ -100,12 +100,20 @@ inittls(void)
 }
 
 static void
-xinitcgo(void)
+xinitcgo(G *g)
 {
+	pthread_attr_t attr;
+	size_t size;
+
+	pthread_attr_init(&attr);
+	pthread_attr_getstacksize(&attr, &size);
+	g->stackguard = (uintptr)&attr - size + 4096;
+	pthread_attr_destroy(&attr);
+
 	inittls();
 }
 
-void (*initcgo)(void) = xinitcgo;
+void (*initcgo)(G*) = xinitcgo;
 
 void
 libcgo_sys_thread_start(ThreadStart *ts)

src/pkg/runtime/cgo/darwin_amd64.c

@@ -70,12 +70,20 @@ inittls(void)
 }
 
 void
-xinitcgo(void)
+xinitcgo(G *g)
 {
+	pthread_attr_t attr;
+	size_t size;
+
+	pthread_attr_init(&attr);
+	pthread_attr_getstacksize(&attr, &size);
+	g->stackguard = (uintptr)&attr - size + 4096;
+	pthread_attr_destroy(&attr);
+
 	inittls();
 }
 
-void (*initcgo) = xinitcgo;
+void (*initcgo)(G*) = xinitcgo;
 
 void
 libcgo_sys_thread_start(ThreadStart *ts)

src/pkg/runtime/cgo/freebsd_386.c

@@ -8,11 +8,18 @@
 static void* threadentry(void*);
 
 static void
-xinitcgo(void)
+xinitcgo(G *g)
 {
+	pthread_attr_t attr;
+	size_t size;
+
+	pthread_attr_init(&attr);
+	pthread_attr_getstacksize(&attr, &size);
+	g->stackguard = (uintptr)&attr - size + 4096;
+	pthread_attr_destroy(&attr);
 }
 
-void (*initcgo)(void) = xinitcgo;
+void (*initcgo)(G*) = xinitcgo;
 
 void
 libcgo_sys_thread_start(ThreadStart *ts)

src/pkg/runtime/cgo/freebsd_amd64.c

@@ -8,11 +8,18 @@
 static void* threadentry(void*);
 
 static void
-xinitcgo(void)
+xinitcgo(G *g)
 {
+	pthread_attr_t attr;
+	size_t size;
+
+	pthread_attr_init(&attr);
+	pthread_attr_getstacksize(&attr, &size);
+	g->stackguard = (uintptr)&attr - size + 4096;
+	pthread_attr_destroy(&attr);
 }
 
-void (*initcgo)(void) = xinitcgo;
+void (*initcgo)(G*) = xinitcgo;
 
 void
 libcgo_sys_thread_start(ThreadStart *ts)

src/pkg/runtime/cgo/linux_386.c

@@ -9,11 +9,18 @@
 static void *threadentry(void*);
 
 static void
-xinitcgo(void)
+xinitcgo(G *g)
 {
+	pthread_attr_t attr;
+	size_t size;
+
+	pthread_attr_init(&attr);
+	pthread_attr_getstacksize(&attr, &size);
+	g->stackguard = (uintptr)&attr - size + 4096;
+	pthread_attr_destroy(&attr);
 }
 
-void (*initcgo) = xinitcgo;
+void (*initcgo)(G*) = xinitcgo;
 
 void
 libcgo_sys_thread_start(ThreadStart *ts)

src/pkg/runtime/cgo/linux_amd64.c

@@ -3,16 +3,24 @@
 // license that can be found in the LICENSE file.
 
 #include <pthread.h>
+#include <string.h> // strerror
 #include "libcgo.h"
 
 static void* threadentry(void*);
 
 void
-xinitcgo(void)
+xinitcgo(G* g)
 {
+	pthread_attr_t attr;
+	size_t size;
+
+	pthread_attr_init(&attr);
+	pthread_attr_getstacksize(&attr, &size);
+	g->stackguard = (uintptr)&attr - size + 4096;
+	pthread_attr_destroy(&attr);
 }
 
-void (*initcgo)(void) = xinitcgo;
+void (*initcgo)(G*) = xinitcgo;
 
 void
 libcgo_sys_thread_start(ThreadStart *ts)

src/pkg/runtime/cgo/linux_arm.c

@@ -5,11 +5,12 @@
 #include "libcgo.h"
 
 static void
-xinitcgo(void)
+xinitcgo(G *g)
 {
+	// unimplemented
 }
 
-void (*initcgo)(void) = xinitcgo;
+void (*initcgo)(G*) = xinitcgo;
 
 void
 libcgo_sys_thread_start(ThreadStart *ts)

src/pkg/runtime/cgo/windows_386.c

@@ -13,11 +13,13 @@ static void *threadentry(void*);
 #define STACKSIZE (1*1024*1024)
 
 static void
-xinitcgo(void)
+xinitcgo(G *g)
 {
+	int tmp;
+	g->stackguard = (uintptr)&tmp - STACKSIZE + 4096;
 }
 
-void (*initcgo)(void) = xinitcgo;
+void (*initcgo)(G*) = xinitcgo;
 
 void
 libcgo_sys_thread_start(ThreadStart *ts)

src/pkg/runtime/cgo/windows_amd64.c

@@ -13,11 +13,13 @@ static void *threadentry(void*);
 #define STACKSIZE (2*1024*1024)
 
 static void
-xinitcgo(void)
+xinitcgo(G *g)
 {
+	int tmp;
+	g->stackguard = (uintptr)&tmp - STACKSIZE + 4096;
 }
 
-void (*initcgo)(void) = xinitcgo;
+void (*initcgo)(G*) = xinitcgo;
 
 void
 libcgo_sys_thread_start(ThreadStart *ts)