runtime: round stack size to power of 2.

Fixes build on windows/386 and plan9/386.
Fixes #7487.

LGTM=mattn.jp, dvyukov, rsc
R=golang-codereviews, mattn.jp, dvyukov, 0intro, rsc
CC=golang-codereviews
https://golang.org/cl/72360043

src/pkg/runtime/proc.c

@@ -1751,12 +1751,13 @@ runtime·malg(int32 stacksize)
 
 	newg = runtime·malloc(sizeof(G));
 	if(stacksize >= 0) {
+		stacksize = runtime·round2(StackSystem + stacksize);
 		if(g == m->g0) {
 			// running on scheduler stack already.
-			stk = runtime·stackalloc(newg, StackSystem + stacksize);
+			stk = runtime·stackalloc(newg, stacksize);
 		} else {
 			// have to call stackalloc on scheduler stack.
-			newg->stacksize = StackSystem + stacksize;
+			newg->stacksize = stacksize;
 			g->param = newg;
 			runtime·mcall(mstackalloc);
 			stk = g->param;
@@ -1765,7 +1766,7 @@ runtime·malg(int32 stacksize)
 		newg->stack0 = (uintptr)stk;
 		newg->stackguard = (uintptr)stk + StackGuard;
 		newg->stackguard0 = newg->stackguard;
-		newg->stackbase = (uintptr)stk + StackSystem + stacksize - sizeof(Stktop);
+		newg->stackbase = (uintptr)stk + stacksize - sizeof(Stktop);
 	}
 	return newg;
 }

src/pkg/runtime/runtime.c

@@ -3,6 +3,7 @@
 // license that can be found in the LICENSE file.
 
 #include "runtime.h"
+#include "stack.h"
 #include "arch_GOARCH.h"
 #include "../../cmd/ld/textflag.h"
 
@@ -256,6 +257,9 @@ runtime·check(void)
 		runtime·throw("float32nan3");
 
 	TestAtomic64();
+
+	if(FixedStack != runtime·round2(FixedStack))
+		runtime·throw("FixedStack is not power-of-2");
 }
 
 uint32

src/pkg/runtime/runtime.h

@@ -900,6 +900,7 @@ void	runtime·mcall(void(*)(G*));
 uint32	runtime·fastrand1(void);
 void	runtime·rewindmorestack(Gobuf*);
 int32	runtime·timediv(int64, int32, int32*);
+int32	runtime·round2(int32 x); // round x up to a power of 2.
 
 // atomic operations
 bool	runtime·cas(uint32*, uint32, uint32);

src/pkg/runtime/stack.c

@@ -555,8 +555,8 @@ copystack(G *gp, uintptr nframes, uintptr newsize)
 }
 
 // round x up to a power of 2.
-static int32
+int32
-round2(int32 x)
+runtime·round2(int32 x)
 {
 	int32 s;
 
@@ -683,7 +683,7 @@ runtime·newstack(void)
 	if(framesize < StackMin)
 		framesize = StackMin;
 	framesize += StackSystem;
-	framesize = round2(framesize);
+	framesize = runtime·round2(framesize);
 	stk = runtime·stackalloc(gp, framesize);
 	if(gp->stacksize > runtime·maxstacksize) {
 		runtime·printf("runtime: goroutine stack exceeds %D-byte limit\n", (uint64)runtime·maxstacksize);

src/pkg/runtime/stack.h

@@ -57,15 +57,13 @@ enum {
 	// to each stack below the usual guard area for OS-specific
 	// purposes like signal handling. Used on Windows and on
 	// Plan 9 because they do not use a separate stack.
-	// The new stack code requires stacks to be a power of two,
-	// and the default start size is 4k, so make StackSystem also 4k
-	// to keep the sum a power of two. StackSystem used to be
-	// 512*sizeof(uintptr) on Windows and 512 bytes on Plan 9.
 #ifdef GOOS_windows
-	StackSystem = 4096,
+	StackSystem = 512 * sizeof(uintptr),
 #else
 #ifdef GOOS_plan9
-	StackSystem = 4096,
+	// The size of the note handler frame varies among architectures,
+	// but 512 bytes should be enough for every implementation.
+	StackSystem = 512,
 #else
 	StackSystem = 0,
 #endif	// Plan 9
@@ -79,7 +77,8 @@ enum {
 	// If the amount needed for the splitting frame + StackExtra
 	// is less than this number, the stack will have this size instead.
 	StackMin = 4096,
-	FixedStack = StackMin + StackSystem,
+	StackSystemRounded = StackSystem + (-StackSystem & (StackMin-1)),
+	FixedStack = StackMin + StackSystemRounded,
 
 	// Functions that need frames bigger than this use an extra
 	// instruction to do the stack split check, to avoid overflow