runtime: simplify CPU profiling code

This makes Go's CPU profiling code somewhat more idiomatic; e.g.,
using := instead of forward declaring variables, using "int" for
element counts instead of "uintptr", and slices instead of C-style
pointer+length.  This makes the code easier to read and eliminates a
lot of type conversion clutter.

Additionally, in sigprof we can collect just maxCPUProfStack stack
frames, as cpuprof won't use more than that anyway.

Change-Id: I0235b5ae552191bcbb453b14add6d8c01381bd06
Reviewed-on: https://go-review.googlesource.com/6072
Run-TryBot: Matthew Dempsky <mdempsky@google.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Dmitry Vyukov <dvyukov@google.com>

src/runtime/cpuprof.go

@@ -30,8 +30,8 @@
 // The state of this dance between the signal handler and the goroutine
 // is encoded in the Profile.handoff field.  If handoff == 0, then the goroutine
 // is not using either log half and is waiting (or will soon be waiting) for
-// a new piece by calling notesleep(&p->wait).  If the signal handler
+// a new piece by calling notesleep(&p.wait).  If the signal handler
-// changes handoff from 0 to non-zero, it must call notewakeup(&p->wait)
+// changes handoff from 0 to non-zero, it must call notewakeup(&p.wait)
 // to wake the goroutine.  The value indicates the number of entries in the
 // log half being handed off.  The goroutine leaves the non-zero value in
 // place until it has finished processing the log half and then flips the number
@@ -61,7 +61,7 @@ const (
 
 type cpuprofEntry struct {
 	count uintptr
-	depth uintptr
+	depth int
 	stack [maxCPUProfStack]uintptr
 }
 
@@ -81,7 +81,7 @@ type cpuProfile struct {
 	// Signal handler has filled log[toggle][:nlog].
 	// Goroutine is writing log[1-toggle][:handoff].
 	log     [2][logSize / 2]uintptr
-	nlog    uintptr
+	nlog    int
 	toggle  int32
 	handoff uint32
 
@@ -167,7 +167,7 @@ func SetCPUProfileRate(hz int) {
 		cpuprof.on = false
 
 		// Now add is not running anymore, and getprofile owns the entire log.
-		// Set the high bit in prof->handoff to tell getprofile.
+		// Set the high bit in cpuprof.handoff to tell getprofile.
 		for {
 			n := cpuprof.handoff
 			if n&0x80000000 != 0 {
@@ -185,20 +185,16 @@ func SetCPUProfileRate(hz int) {
 	unlock(&cpuprofLock)
 }
 
-func cpuproftick(pc *uintptr, n int32) {
-	if n > maxCPUProfStack {
-		n = maxCPUProfStack
-	}
-	s := (*[maxCPUProfStack]uintptr)(unsafe.Pointer(pc))[:n]
-	cpuprof.add(s)
-}
-
 // add adds the stack trace to the profile.
 // It is called from signal handlers and other limited environments
 // and cannot allocate memory or acquire locks that might be
 // held at the time of the signal, nor can it use substantial amounts
 // of stack.  It is allowed to call evict.
 func (p *cpuProfile) add(pc []uintptr) {
+	if len(pc) > maxCPUProfStack {
+		pc = pc[:maxCPUProfStack]
+	}
+
 	// Compute hash.
 	h := uintptr(0)
 	for _, x := range pc {
@@ -212,7 +208,7 @@ func (p *cpuProfile) add(pc []uintptr) {
 Assoc:
 	for i := range b.entry {
 		e := &b.entry[i]
-		if e.depth != uintptr(len(pc)) {
+		if e.depth != len(pc) {
 			continue
 		}
 		for j := range pc {
@@ -241,7 +237,7 @@ Assoc:
 	}
 
 	// Reuse the newly evicted entry.
-	e.depth = uintptr(len(pc))
+	e.depth = len(pc)
 	e.count = 1
 	copy(e.stack[:], pc)
 }
@@ -256,7 +252,7 @@ func (p *cpuProfile) evict(e *cpuprofEntry) bool {
 	d := e.depth
 	nslot := d + 2
 	log := &p.log[p.toggle]
-	if p.nlog+nslot > uintptr(len(p.log[0])) {
+	if p.nlog+nslot > len(log) {
 		if !p.flushlog() {
 			return false
 		}
@@ -266,7 +262,7 @@ func (p *cpuProfile) evict(e *cpuprofEntry) bool {
 	q := p.nlog
 	log[q] = e.count
 	q++
-	log[q] = d
+	log[q] = uintptr(d)
 	q++
 	copy(log[q:], e.stack[:d])
 	q += d
@@ -287,7 +283,7 @@ func (p *cpuProfile) flushlog() bool {
 
 	p.toggle = 1 - p.toggle
 	log := &p.log[p.toggle]
-	q := uintptr(0)
+	q := 0
 	if p.lost > 0 {
 		lostPC := funcPC(lostProfileData)
 		log[0] = p.lost
@@ -360,7 +356,7 @@ func (p *cpuProfile) getprofile() []byte {
 
 	// In flush mode.
 	// Add is no longer being called.  We own the log.
-	// Also, p->handoff is non-zero, so flushlog will return false.
+	// Also, p.handoff is non-zero, so flushlog will return false.
 	// Evict the hash table into the log and return it.
 Flush:
 	for i := range p.hash {

src/runtime/extern.go

@@ -114,7 +114,7 @@ func Caller(skip int) (pc uintptr, file string, line int, ok bool) {
 	// and what it called, so that we can see if it
 	// "called" sigpanic.
 	var rpc [2]uintptr
-	if callers(1+skip-1, &rpc[0], 2) < 2 {
+	if callers(1+skip-1, rpc[:]) < 2 {
 		return
 	}
 	f := findfunc(rpc[1])
@@ -161,7 +161,7 @@ func Callers(skip int, pc []uintptr) int {
 	if len(pc) == 0 {
 		return 0
 	}
-	return callers(skip, &pc[0], len(pc))
+	return callers(skip, pc)
 }
 
 // GOROOT returns the root of the Go tree.

src/runtime/mprof.go

@@ -232,7 +232,7 @@ func mProf_GC() {
 // Called by malloc to record a profiled block.
 func mProf_Malloc(p unsafe.Pointer, size uintptr) {
 	var stk [maxStack]uintptr
-	nstk := callers(4, &stk[0], len(stk))
+	nstk := callers(4, stk[:])
 	lock(&proflock)
 	b := stkbucket(memProfile, size, stk[:nstk], true)
 	mp := b.mp()
@@ -300,9 +300,9 @@ func blockevent(cycles int64, skip int) {
 	var nstk int
 	var stk [maxStack]uintptr
 	if gp.m.curg == nil || gp.m.curg == gp {
-		nstk = callers(skip, &stk[0], len(stk))
+		nstk = callers(skip, stk[:])
 	} else {
-		nstk = gcallers(gp.m.curg, skip, &stk[0], len(stk))
+		nstk = gcallers(gp.m.curg, skip, stk[:])
 	}
 	lock(&proflock)
 	b := stkbucket(blockProfile, 0, stk[:nstk], true)

src/runtime/os1_windows.go

@@ -527,7 +527,7 @@ func profilem(mp *m) {
 	r = (*context)(unsafe.Pointer((uintptr(unsafe.Pointer(&rbuf[15]))) &^ 15))
 	r.contextflags = _CONTEXT_CONTROL
 	stdcall2(_GetThreadContext, mp.thread, uintptr(unsafe.Pointer(r)))
-	sigprof((*byte)(unsafe.Pointer(r.ip())), (*byte)(unsafe.Pointer(r.sp())), nil, gp, mp)
+	sigprof(r.ip(), r.sp(), 0, gp, mp)
 }
 
 func profileloop1() {

src/runtime/proc1.go

@@ -100,7 +100,7 @@ func mcommoninit(mp *m) {
 
 	// g0 stack won't make sense for user (and is not necessary unwindable).
 	if _g_ != _g_.m.g0 {
-		callers(1, &mp.createstack[0], len(mp.createstack))
+		callers(1, mp.createstack[:])
 	}
 
 	mp.fastrand = 0x49f6428a + uint32(mp.id) + uint32(cputicks())
@@ -2286,11 +2286,7 @@ func _GC()           { _GC() }
 var etext struct{}
 
 // Called if we receive a SIGPROF signal.
-func sigprof(pc *uint8, sp *uint8, lr *uint8, gp *g, mp *m) {
+func sigprof(pc, sp, lr uintptr, gp *g, mp *m) {
-	var n int32
-	var traceback bool
-	var stk [100]uintptr
-
 	if prof.hz == 0 {
 		return
 	}
@@ -2370,18 +2366,18 @@ func sigprof(pc *uint8, sp *uint8, lr *uint8, gp *g, mp *m) {
 	// To recap, there are no constraints on the assembly being used for the
 	// transition. We simply require that g and SP match and that the PC is not
 	// in gogo.
-	traceback = true
+	traceback := true
-	usp := uintptr(unsafe.Pointer(sp))
 	gogo := funcPC(gogo)
 	if gp == nil || gp != mp.curg ||
-		usp < gp.stack.lo || gp.stack.hi < usp ||
+		sp < gp.stack.lo || gp.stack.hi < sp ||
-		(gogo <= uintptr(unsafe.Pointer(pc)) && uintptr(unsafe.Pointer(pc)) < gogo+_RuntimeGogoBytes) {
+		(gogo <= pc && pc < gogo+_RuntimeGogoBytes) {
 		traceback = false
 	}
 
-	n = 0
+	var stk [maxCPUProfStack]uintptr
+	n := 0
 	if traceback {
-		n = int32(gentraceback(uintptr(unsafe.Pointer(pc)), uintptr(unsafe.Pointer(sp)), uintptr(unsafe.Pointer(lr)), gp, 0, &stk[0], len(stk), nil, nil, _TraceTrap))
+		n = gentraceback(pc, sp, lr, gp, 0, &stk[0], len(stk), nil, nil, _TraceTrap)
 	}
 	if !traceback || n <= 0 {
 		// Normal traceback is impossible or has failed.
@@ -2391,21 +2387,21 @@ func sigprof(pc *uint8, sp *uint8, lr *uint8, gp *g, mp *m) {
 			// Cgo, we can't unwind and symbolize arbitrary C code,
 			// so instead collect Go stack that leads to the cgo call.
 			// This is especially important on windows, since all syscalls are cgo calls.
-			n = int32(gentraceback(mp.curg.syscallpc, mp.curg.syscallsp, 0, mp.curg, 0, &stk[0], len(stk), nil, nil, 0))
+			n = gentraceback(mp.curg.syscallpc, mp.curg.syscallsp, 0, mp.curg, 0, &stk[0], len(stk), nil, nil, 0)
 		}
 		if GOOS == "windows" && n == 0 && mp.libcallg != nil && mp.libcallpc != 0 && mp.libcallsp != 0 {
 			// Libcall, i.e. runtime syscall on windows.
 			// Collect Go stack that leads to the call.
-			n = int32(gentraceback(mp.libcallpc, mp.libcallsp, 0, mp.libcallg, 0, &stk[0], len(stk), nil, nil, 0))
+			n = gentraceback(mp.libcallpc, mp.libcallsp, 0, mp.libcallg, 0, &stk[0], len(stk), nil, nil, 0)
 		}
 		if n == 0 {
 			// If all of the above has failed, account it against abstract "System" or "GC".
 			n = 2
 			// "ExternalCode" is better than "etext".
-			if uintptr(unsafe.Pointer(pc)) > uintptr(unsafe.Pointer(&etext)) {
+			if pc > uintptr(unsafe.Pointer(&etext)) {
-				pc = (*uint8)(unsafe.Pointer(uintptr(funcPC(_ExternalCode) + _PCQuantum)))
+				pc = funcPC(_ExternalCode) + _PCQuantum
 			}
-			stk[0] = uintptr(unsafe.Pointer(pc))
+			stk[0] = pc
 			if mp.preemptoff != "" || mp.helpgc != 0 {
 				stk[1] = funcPC(_GC) + _PCQuantum
 			} else {
@@ -2420,7 +2416,7 @@ func sigprof(pc *uint8, sp *uint8, lr *uint8, gp *g, mp *m) {
 			osyield()
 		}
 		if prof.hz != 0 {
-			cpuproftick(&stk[0], n)
+			cpuprof.add(stk[:n])
 		}
 		atomicstore(&prof.lock, 0)
 	}

src/runtime/signal_386.go

@@ -29,7 +29,7 @@ func sighandler(sig uint32, info *siginfo, ctxt unsafe.Pointer, gp *g) {
 	c := &sigctxt{info, ctxt}
 
 	if sig == _SIGPROF {
-		sigprof((*byte)(unsafe.Pointer(uintptr(c.eip()))), (*byte)(unsafe.Pointer(uintptr(c.esp()))), nil, gp, _g_.m)
+		sigprof(uintptr(c.eip()), uintptr(c.esp()), 0, gp, _g_.m)
 		return
 	}
 

src/runtime/signal_amd64x.go

@@ -42,7 +42,7 @@ func sighandler(sig uint32, info *siginfo, ctxt unsafe.Pointer, gp *g) {
 	c := &sigctxt{info, ctxt}
 
 	if sig == _SIGPROF {
-		sigprof((*byte)(unsafe.Pointer(uintptr(c.rip()))), (*byte)(unsafe.Pointer(uintptr(c.rsp()))), nil, gp, _g_.m)
+		sigprof(uintptr(c.rip()), uintptr(c.rsp()), 0, gp, _g_.m)
 		return
 	}
 

src/runtime/signal_arm.go

@@ -37,7 +37,7 @@ func sighandler(sig uint32, info *siginfo, ctxt unsafe.Pointer, gp *g) {
 	c := &sigctxt{info, ctxt}
 
 	if sig == _SIGPROF {
-		sigprof((*byte)(unsafe.Pointer(uintptr(c.pc()))), (*byte)(unsafe.Pointer(uintptr(c.sp()))), (*byte)(unsafe.Pointer(uintptr(c.lr()))), gp, _g_.m)
+		sigprof(uintptr(c.pc()), uintptr(c.sp()), uintptr(c.lr()), gp, _g_.m)
 		return
 	}
 

src/runtime/signal_ppc64x.go

@@ -55,7 +55,7 @@ func sighandler(sig uint32, info *siginfo, ctxt unsafe.Pointer, gp *g) {
 	c := &sigctxt{info, ctxt}
 
 	if sig == _SIGPROF {
-		sigprof((*byte)(unsafe.Pointer(uintptr(c.pc()))), (*byte)(unsafe.Pointer(uintptr(c.sp()))), (*byte)(unsafe.Pointer(uintptr(c.link()))), gp, _g_.m)
+		sigprof(uintptr(c.pc()), uintptr(c.sp()), uintptr(c.link()), gp, _g_.m)
 		return
 	}
 	flags := int32(_SigThrow)

src/runtime/trace.go

@@ -468,9 +468,9 @@ func traceEvent(ev byte, stack bool, args ...uint64) {
 		}
 		var nstk int
 		if gp == _g_ {
-			nstk = callers(1, &buf.stk[0], len(buf.stk))
+			nstk = callers(1, buf.stk[:])
 		} else if gp != nil {
-			nstk = gcallers(mp.curg, 1, &buf.stk[0], len(buf.stk))
+			nstk = gcallers(mp.curg, 1, buf.stk[:])
 		}
 		id := trace.stackTab.put(buf.stk[:nstk])
 		data = traceAppend(data, uint64(id))

src/runtime/traceback.go

@@ -104,7 +104,7 @@ func tracebackdefers(gp *g, callback func(*stkframe, unsafe.Pointer) bool, v uns
 // the runtime.Callers function (pcbuf != nil), as well as the garbage
 // collector (callback != nil).  A little clunky to merge these, but avoids
 // duplicating the code and all its subtlety.
-func gentraceback(pc0 uintptr, sp0 uintptr, lr0 uintptr, gp *g, skip int, pcbuf *uintptr, max int, callback func(*stkframe, unsafe.Pointer) bool, v unsafe.Pointer, flags uint) int {
+func gentraceback(pc0, sp0, lr0 uintptr, gp *g, skip int, pcbuf *uintptr, max int, callback func(*stkframe, unsafe.Pointer) bool, v unsafe.Pointer, flags uint) int {
 	if goexitPC == 0 {
 		throw("gentraceback before goexitPC initialization")
 	}
@@ -367,7 +367,7 @@ func gentraceback(pc0 uintptr, sp0 uintptr, lr0 uintptr, gp *g, skip int, pcbuf
 		}
 	}
 
-	if pcbuf == nil && callback == nil {
+	if printing {
 		n = nprint
 	}
 
@@ -474,7 +474,7 @@ func printcreatedby(gp *g) {
 	}
 }
 
-func traceback(pc uintptr, sp uintptr, lr uintptr, gp *g) {
+func traceback(pc, sp, lr uintptr, gp *g) {
 	traceback1(pc, sp, lr, gp, 0)
 }
 
@@ -484,11 +484,11 @@ func traceback(pc uintptr, sp uintptr, lr uintptr, gp *g) {
 // the initial PC must not be rewound to the previous instruction.
 // (All the saved pairs record a PC that is a return address, so we
 // rewind it into the CALL instruction.)
-func tracebacktrap(pc uintptr, sp uintptr, lr uintptr, gp *g) {
+func tracebacktrap(pc, sp, lr uintptr, gp *g) {
 	traceback1(pc, sp, lr, gp, _TraceTrap)
 }
 
-func traceback1(pc uintptr, sp uintptr, lr uintptr, gp *g, flags uint) {
+func traceback1(pc, sp, lr uintptr, gp *g, flags uint) {
 	var n int
 	if readgstatus(gp)&^_Gscan == _Gsyscall {
 		// Override registers if blocked in system call.
@@ -508,18 +508,18 @@ func traceback1(pc uintptr, sp uintptr, lr uintptr, gp *g, flags uint) {
 	printcreatedby(gp)
 }
 
-func callers(skip int, pcbuf *uintptr, m int) int {
+func callers(skip int, pcbuf []uintptr) int {
 	sp := getcallersp(unsafe.Pointer(&skip))
 	pc := uintptr(getcallerpc(unsafe.Pointer(&skip)))
 	var n int
 	systemstack(func() {
-		n = gentraceback(pc, sp, 0, getg(), skip, pcbuf, m, nil, nil, 0)
+		n = gentraceback(pc, sp, 0, getg(), skip, &pcbuf[0], len(pcbuf), nil, nil, 0)
 	})
 	return n
 }
 
-func gcallers(gp *g, skip int, pcbuf *uintptr, m int) int {
+func gcallers(gp *g, skip int, pcbuf []uintptr) int {
-	return gentraceback(^uintptr(0), ^uintptr(0), 0, gp, skip, pcbuf, m, nil, nil, 0)
+	return gentraceback(^uintptr(0), ^uintptr(0), 0, gp, skip, &pcbuf[0], len(pcbuf), nil, nil, 0)
 }
 
 func showframe(f *_func, gp *g) bool {