runtime: SysMap uses MAP_FIXED if needed on 64-bit Linux

This change was adapted from gccgo's libgo/runtime/mem.c at
Ian Taylor's suggestion.  It fixes all.bash failing with
"address space conflict: map() =" on amd64 Linux with kernel
version 2.6.32.8-grsec-2.1.14-modsign-xeon-64.
With this change, SysMap will use MAP_FIXED to allocate its desired
address space, after first calling mincore to check that there is
nothing else mapped there.

R=iant, dave, n13m3y3r, rsc
CC=golang-dev
https://golang.org/cl/4438091

5 files changed, 55 insertions, 5 deletions

diff --git a/src/pkg/runtime/linux/386/sys.s b/src/pkg/runtime/linux/386/sys.s
index 868a0d901a..e8b4233242 100644
--- a/src/pkg/runtime/linux/386/sys.s
+++ b/src/pkg/runtime/linux/386/sys.s
@@ -47,6 +47,14 @@ TEXT runtime·setitimer(SB),7,$0-24
 	INT	$0x80
 	RET
 
+TEXT runtime·mincore(SB),7,$0-24
+	MOVL	$218, AX			// syscall - mincore
+	MOVL	4(SP), BX
+	MOVL	8(SP), CX
+	MOVL	12(SP), DX
+	INT	$0x80
+	RET
+
 TEXT runtime·gettime(SB), 7, $32
 	MOVL	$78, AX			// syscall - gettimeofday
 	LEAL	8(SP), BX
diff --git a/src/pkg/runtime/linux/amd64/sys.s b/src/pkg/runtime/linux/amd64/sys.s
index eadd300058..66fdab2083 100644
--- a/src/pkg/runtime/linux/amd64/sys.s
+++ b/src/pkg/runtime/linux/amd64/sys.s
@@ -53,6 +53,14 @@ TEXT runtime·setitimer(SB),7,$0-24
 	SYSCALL
 	RET
 
+TEXT runtime·mincore(SB),7,$0-24
+	MOVQ	8(SP), DI
+	MOVQ	16(SP), SI
+	MOVQ	24(SP), DX
+	MOVL	$27, AX			// syscall entry
+	SYSCALL
+	RET
+
 TEXT runtime·gettime(SB), 7, $32
 	LEAQ	8(SP), DI
 	MOVQ	$0, SI
diff --git a/src/pkg/runtime/linux/arm/sys.s b/src/pkg/runtime/linux/arm/sys.s
index 2b5365bd86..ab53498222 100644
--- a/src/pkg/runtime/linux/arm/sys.s
+++ b/src/pkg/runtime/linux/arm/sys.s
@@ -26,6 +26,7 @@
 #define SYS_exit_group (SYS_BASE + 248)
 #define SYS_munmap (SYS_BASE + 91)
 #define SYS_setitimer (SYS_BASE + 104)
+#define SYS_mincore (SYS_BASE + 219)
 #define SYS_gettid (SYS_BASE + 224)
 #define SYS_tkill (SYS_BASE + 238)
 
@@ -91,6 +92,14 @@ TEXT runtime·setitimer(SB),7,$0
 	SWI	$0
 	RET
 
+TEXT runtime·mincore(SB),7,$0
+	MOVW	0(FP), R0
+	MOVW	4(FP), R1
+	MOVW	8(FP), R2
+	MOVW	$SYS_mincore, R7
+	SWI	$0
+	RET
+
 TEXT runtime·gettime(SB),7,$32
 	/* dummy version - return 0,0 */
 	MOVW	$0, R1
diff --git a/src/pkg/runtime/linux/mem.c b/src/pkg/runtime/linux/mem.c
index 02f798732f..38ca7e4a0f 100644
--- a/src/pkg/runtime/linux/mem.c
+++ b/src/pkg/runtime/linux/mem.c
@@ -3,6 +3,30 @@
 #include "os.h"
 #include "malloc.h"
 
+enum
+{
+	ENOMEM = 12,
+};
+
+static int32
+addrspace_free(void *v, uintptr n)
+{
+	uintptr page_size = 4096;
+	uintptr off;
+	int8 one_byte;
+
+	for(off = 0; off < n; off += page_size) {
+		int32 errval = runtime·mincore((int8 *)v + off, page_size, (void *)&one_byte);
+		// errval is 0 if success, or -(error_code) if error.
+		if (errval == 0 || errval != -ENOMEM)
+			return 0;
+	}
+	USED(v);
+	USED(n);
+	return 1;
+}
+
+
 void*
 runtime·SysAlloc(uintptr n)
 {
@@ -54,11 +78,6 @@ runtime·SysReserve(void *v, uintptr n)
 	return p;
 }
 
-enum
-{
-	ENOMEM = 12,
-};
-
 void
 runtime·SysMap(void *v, uintptr n)
 {
@@ -69,6 +88,11 @@ runtime·SysMap(void *v, uintptr n)
 	// On 64-bit, we don't actually have v reserved, so tread carefully.
 	if(sizeof(void*) == 8) {
 		p = runtime·mmap(v, n, PROT_READ|PROT_WRITE|PROT_EXEC, MAP_ANON|MAP_PRIVATE, -1, 0);
+		if(p != v && addrspace_free(v, n)) {
+			// On some systems, mmap ignores v without
+			// MAP_FIXED, so retry if the address space is free.
+			p = runtime·mmap(v, n, PROT_READ|PROT_WRITE|PROT_EXEC, MAP_ANON|MAP_FIXED|MAP_PRIVATE, -1, 0);
+		}
 		if(p == (void*)ENOMEM)
 			runtime·throw("runtime: out of memory");
 		if(p != v) {
diff --git a/src/pkg/runtime/runtime.h b/src/pkg/runtime/runtime.h
index 2b2b34a3ce..f3ccff1bcd 100644
--- a/src/pkg/runtime/runtime.h
+++ b/src/pkg/runtime/runtime.h
@@ -413,6 +413,7 @@ int32	runtime·gotraceback(void);
 void	runtime·traceback(uint8 *pc, uint8 *sp, uint8 *lr, G* gp);
 void	runtime·tracebackothers(G*);
 int32	runtime·write(int32, void*, int32);
+int32	runtime·mincore(void*, uintptr, byte*);
 bool	runtime·cas(uint32*, uint32, uint32);
 bool	runtime·casp(void**, void*, void*);
 uint32	runtime·xadd(uint32 volatile*, int32);