runtime: use conservative scanning for debug calls

A debugger can inject a call at almost any PC, which causes
significant complications with stack scanning and growth. Currently,
the runtime solves this using precise stack maps and register maps at
nearly all PCs, but these extra maps require roughly 5% of the binary.
These extra maps were originally considered worth this space because
they were intended to be used for non-cooperative preemption, but are
now used only for debug call injection.

This CL switches from using precise maps to instead using conservative
frame scanning, much like how non-cooperative preemption works. When a
call is injected, the runtime flushes all potential pointer registers
to the stack, and then treats that frame as well as the interrupted
frame conservatively.

The limitation of conservative frame scanning is that we cannot grow
the goroutine stack. That's doable because the previous CL switched to
performing debug calls on a new goroutine, where they are free to grow
the stack.

With this CL, there are no remaining uses of precise register maps
(though we still use the unsafe-point information that's encoded in
the register map PCDATA stream), and stack maps are only used at call
sites.

For #36365.

Change-Id: Ie217b6711f3741ccc437552d8ff88f961a73cee0
Reviewed-on: https://go-review.googlesource.com/c/go/+/229300
Run-TryBot: Austin Clements <austin@google.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Cherry Zhang <cherryyz@google.com>

src/runtime/asm_amd64.s

@@ -1560,10 +1560,8 @@ GLOBL	debugCallFrameTooLarge<>(SB), RODATA, $20	// Size duplicated below
 // a stack pointer to an escaping argument. debugCallV1 cannot check
 // this invariant.
 TEXT runtime·debugCallV1(SB),NOSPLIT,$152-0
-	// Save all registers that may contain pointers in GC register
+	// Save all registers that may contain pointers so they can be
-	// map order (see ssa.registersAMD64). This makes it possible
+	// conservatively scanned.
-	// to copy the stack while updating pointers currently held in
-	// registers, and for the GC to find roots in registers.
 	//
 	// We can't do anything that might clobber any of these
 	// registers before this.

src/runtime/debugcall.go

@@ -99,6 +99,11 @@ func debugCallCheck(pc uintptr) string {
 // the calling goroutine. On the goroutine, it prepares to recover
 // panics from the debug call, and then calls the call dispatching
 // function at PC dispatch.
+//
+// This must be deeply nosplit because there are untyped values on the
+// stack from debugCallV1.
+//
+//go:nosplit
 func debugCallWrap(dispatch uintptr) {
 	var lockedm bool
 	var lockedExt uint32
@@ -139,6 +144,12 @@ func debugCallWrap(dispatch uintptr) {
 			gp.lockedm = 0
 		}
 
+		// Mark the calling goroutine as being at an async
+		// safe-point, since it has a few conservative frames
+		// at the bottom of the stack. This also prevents
+		// stack shrinks.
+		gp.asyncSafePoint = true
+
 		// Stash newg away so we can execute it below (mcall's
 		// closure can't capture anything).
 		gp.schedlink.set(newg)
@@ -175,6 +186,8 @@ func debugCallWrap(dispatch uintptr) {
 		mp.lockedg.set(gp)
 		gp.lockedm.set(mp)
 	}
+
+	gp.asyncSafePoint = false
 }
 
 // debugCallWrap1 is the continuation of debugCallWrap on the callee

src/runtime/mgcmark.go

@@ -961,7 +961,8 @@ func scanframeworker(frame *stkframe, state *stackScanState, gcw *gcWork) {
 	}
 
 	isAsyncPreempt := frame.fn.valid() && frame.fn.funcID == funcID_asyncPreempt
-	if state.conservative || isAsyncPreempt {
+	isDebugCall := frame.fn.valid() && frame.fn.funcID == funcID_debugCallV1
+	if state.conservative || isAsyncPreempt || isDebugCall {
 		if debugScanConservative {
 			println("conservatively scanning function", funcname(frame.fn), "at PC", hex(frame.continpc))
 		}
@@ -988,7 +989,7 @@ func scanframeworker(frame *stkframe, state *stackScanState, gcw *gcWork) {
 			scanConservative(frame.argp, frame.arglen, nil, gcw, state)
 		}
 
-		if isAsyncPreempt {
+		if isAsyncPreempt || isDebugCall {
 			// This function's frame contained the
 			// registers for the asynchronously stopped
 			// parent frame. Scan the parent

src/runtime/stack.go

@@ -1216,29 +1216,8 @@ func getStackMap(frame *stkframe, cache *pcvalueCache, debug bool) (locals, args
 		minsize = sys.MinFrameSize
 	}
 	if size > minsize {
-		var stkmap *stackmap
 		stackid := pcdata
-		if f.funcID != funcID_debugCallV1 {
+		stkmap := (*stackmap)(funcdata(f, _FUNCDATA_LocalsPointerMaps))
-			stkmap = (*stackmap)(funcdata(f, _FUNCDATA_LocalsPointerMaps))
-		} else {
-			// debugCallV1's stack map is the register map
-			// at its call site.
-			callerPC := frame.lr
-			caller := findfunc(callerPC)
-			if !caller.valid() {
-				println("runtime: debugCallV1 called by unknown caller", hex(callerPC))
-				throw("bad debugCallV1")
-			}
-			stackid = int32(-1)
-			if callerPC != caller.entry {
-				callerPC--
-				stackid = pcdatavalue(caller, _PCDATA_RegMapIndex, callerPC, cache)
-			}
-			if stackid == -1 {
-				stackid = 0 // in prologue
-			}
-			stkmap = (*stackmap)(funcdata(caller, _FUNCDATA_RegPointerMaps))
-		}
 		if stkmap == nil || stkmap.n <= 0 {
 			print("runtime: frame ", funcname(f), " untyped locals ", hex(frame.varp-size), "+", hex(size), "\n")
 			throw("missing stackmap")