runtime: decouple stack bounds and stack allocation size

Currently the runtime assumes that the allocation for the stack is
exactly [stack.lo, stack.hi). We're about to steal a small part of
this allocation for per-stack GC metadata. To prepare for this, this
commit adds a field to the G for the allocated size of the stack.
With this change, stack.lo and stack.hi continue to act as the true
bounds on the stack, but are no longer also used as the bounds on the
stack allocation.

(I also tried this the other way around, where stack.lo and stack.hi
remained the allocation bounds and I introduced a new top of stack.
However, there are far more places that assume stack.hi is the true
top of the stack than there are places that assume it's the top of the
allocation.)

Change-Id: Ifa9d956753be53d286d09cbc73d47fb34a18c0c6
Reviewed-on: https://go-review.googlesource.com/10312Reviewed-by: Russ Cox <rsc@golang.org>

src/runtime/proc1.go

@@ -2161,6 +2161,7 @@ func malg(stacksize int32) *g {
 		})
 		newg.stackguard0 = newg.stack.lo + _StackGuard
 		newg.stackguard1 = ^uintptr(0)
+		newg.stackAlloc = uintptr(stacksize)
 	}
 	return newg
 }
@@ -2276,11 +2277,11 @@ func gfput(_p_ *p, gp *g) {
 		throw("gfput: bad status (not Gdead)")
 	}
 
-	stksize := gp.stack.hi - gp.stack.lo
+	stksize := gp.stackAlloc
 
 	if stksize != _FixedStack {
 		// non-standard stack size - free it.
-		stackfree(gp.stack)
+		stackfree(gp.stack, gp.stackAlloc)
 		gp.stack.lo = 0
 		gp.stack.hi = 0
 		gp.stackguard0 = 0
@@ -2330,9 +2331,10 @@ retry:
 				gp.stack = stackalloc(_FixedStack)
 			})
 			gp.stackguard0 = gp.stack.lo + _StackGuard
+			gp.stackAlloc = _FixedStack
 		} else {
 			if raceenabled {
-				racemalloc(unsafe.Pointer(gp.stack.lo), gp.stack.hi-gp.stack.lo)
+				racemalloc(unsafe.Pointer(gp.stack.lo), gp.stackAlloc)
 			}
 		}
 	}

src/runtime/runtime2.go

@@ -216,6 +216,7 @@ type g struct {
 
 	_panic       *_panic // innermost panic - offset known to liblink
 	_defer       *_defer // innermost defer
+	stackAlloc   uintptr // stack allocation is [stack.lo,stack.lo+stackAlloc)
 	sched        gobuf
 	syscallsp    uintptr        // if status==Gsyscall, syscallsp = sched.sp to use during gc
 	syscallpc    uintptr        // if status==Gsyscall, syscallpc = sched.pc to use during gc

src/runtime/stack1.go

@@ -246,13 +246,15 @@ func stackalloc(n uint32) stack {
 	return stack{uintptr(v), uintptr(v) + uintptr(n)}
 }
 
-func stackfree(stk stack) {
+func stackfree(stk stack, n uintptr) {
 	gp := getg()
-	n := stk.hi - stk.lo
 	v := (unsafe.Pointer)(stk.lo)
 	if n&(n-1) != 0 {
 		throw("stack not a power of 2")
 	}
+	if stk.lo+n < stk.hi {
+		throw("bad stack size")
+	}
 	if stackDebug >= 1 {
 		println("stackfree", v, n)
 		memclr(v, n) // for testing, clobber stack data
@@ -584,14 +586,16 @@ func copystack(gp *g, newsize uintptr) {
 	gp.stack = new
 	gp.stackguard0 = new.lo + _StackGuard // NOTE: might clobber a preempt request
 	gp.sched.sp = new.hi - used
+	oldsize := gp.stackAlloc
+	gp.stackAlloc = newsize
 
 	// free old stack
 	if stackPoisonCopy != 0 {
 		fillstack(old, 0xfc)
 	}
-	if newsize > old.hi-old.lo {
+	if newsize > oldsize {
 		// growing, free stack immediately
-		stackfree(old)
+		stackfree(old, oldsize)
 	} else {
 		// shrinking, queue up free operation.  We can't actually free the stack
 		// just yet because we might run into the following situation:
@@ -604,6 +608,7 @@ func copystack(gp *g, newsize uintptr) {
 		// By not freeing, we prevent step #4 until GC is done.
 		lock(&stackpoolmu)
 		*(*stack)(unsafe.Pointer(old.lo)) = stackfreequeue
+		*(*uintptr)(unsafe.Pointer(old.lo + ptrSize)) = oldsize
 		stackfreequeue = old
 		unlock(&stackpoolmu)
 	}
@@ -743,7 +748,7 @@ func newstack() {
 	}
 
 	// Allocate a bigger segment and move the stack.
-	oldsize := int(gp.stack.hi - gp.stack.lo)
+	oldsize := int(gp.stackAlloc)
 	newsize := oldsize * 2
 	if uintptr(newsize) > maxstacksize {
 		print("runtime: goroutine stack exceeds ", maxstacksize, "-byte limit\n")
@@ -786,7 +791,7 @@ func shrinkstack(gp *g) {
 		if gp.stack.lo != 0 {
 			// Free whole stack - it will get reallocated
 			// if G is used again.
-			stackfree(gp.stack)
+			stackfree(gp.stack, gp.stackAlloc)
 			gp.stack.lo = 0
 			gp.stack.hi = 0
 		}
@@ -796,7 +801,7 @@ func shrinkstack(gp *g) {
 		throw("missing stack in shrinkstack")
 	}
 
-	oldsize := gp.stack.hi - gp.stack.lo
+	oldsize := gp.stackAlloc
 	newsize := oldsize / 2
 	if newsize < _FixedStack {
 		return // don't shrink below the minimum-sized stack
@@ -832,7 +837,8 @@ func shrinkfinish() {
 	unlock(&stackpoolmu)
 	for s.lo != 0 {
 		t := *(*stack)(unsafe.Pointer(s.lo))
-		stackfree(s)
+		n := *(*uintptr)(unsafe.Pointer(s.lo + ptrSize))
+		stackfree(s, n)
 		s = t
 	}
 }