17 files changed, 351 insertions, 442 deletions

diff --git a/libgo/go/runtime/mprof.go b/libgo/go/runtime/mprof.go
index a2701e3..cc0a673 100644
--- a/libgo/go/runtime/mprof.go
+++ b/libgo/go/runtime/mprof.go
@@ -556,7 +556,7 @@ func GoroutineProfile(p []StackRecord) (n int, ok bool) {
 	stopTheWorld("profile")
 
 	n = 1
-	for _, gp1 := range allgs() {
+	for _, gp1 := range allgs {
 		if isOK(gp1) {
 			n++
 		}
@@ -571,7 +571,7 @@ func GoroutineProfile(p []StackRecord) (n int, ok bool) {
 		r = r[1:]
 
 		// Save other goroutines.
-		for _, gp1 := range allgs() {
+		for _, gp1 := range allgs {
 			if isOK(gp1) {
 				if len(r) == 0 {
 					// Should be impossible, but better to return a
diff --git a/libgo/go/runtime/proc.go b/libgo/go/runtime/proc.go
index fa90a28..78cc6ee 100644
--- a/libgo/go/runtime/proc.go
+++ b/libgo/go/runtime/proc.go
@@ -11,15 +11,18 @@ import (
 
 // Functions temporarily called by C code.
 //go:linkname newextram runtime.newextram
+//go:linkname checkdead runtime.checkdead
+//go:linkname schedtrace runtime.schedtrace
+//go:linkname allgadd runtime.allgadd
 
 // Functions temporarily in C that have not yet been ported.
 func allocm(*p, bool, *unsafe.Pointer, *uintptr) *m
 func malg(bool, bool, *unsafe.Pointer, *uintptr) *g
-func allgadd(*g)
 
 // C functions for ucontext management.
 func setGContext()
 func makeGContext(*g, unsafe.Pointer, uintptr)
+func getTraceback(me, gp *g)
 
 // main_init_done is a signal used by cgocallbackg that initialization
 // has been completed. It is made before _cgo_notify_runtime_init_done,
@@ -27,6 +30,39 @@ func makeGContext(*g, unsafe.Pointer, uintptr)
 // it is closed, meaning cgocallbackg can reliably receive from it.
 var main_init_done chan bool
 
+var (
+	allgs    []*g
+	allglock mutex
+)
+
+func allgadd(gp *g) {
+	if readgstatus(gp) == _Gidle {
+		throw("allgadd: bad status Gidle")
+	}
+
+	lock(&allglock)
+	allgs = append(allgs, gp)
+	allglen = uintptr(len(allgs))
+
+	// Grow GC rescan list if necessary.
+	if len(allgs) > cap(work.rescan.list) {
+		lock(&work.rescan.lock)
+		l := work.rescan.list
+		// Let append do the heavy lifting, but keep the
+		// length the same.
+		work.rescan.list = append(l[:cap(l)], 0)[:len(l)]
+		unlock(&work.rescan.lock)
+	}
+	unlock(&allglock)
+}
+
+// All reads and writes of g's status go through readgstatus, casgstatus
+// castogscanstatus, casfrom_Gscanstatus.
+//go:nosplit
+func readgstatus(gp *g) uint32 {
+	return atomic.Load(&gp.atomicstatus)
+}
+
 // If asked to move to or from a Gscanstatus this will throw. Use the castogscanstatus
 // and casfrom_Gscanstatus instead.
 // casgstatus will loop if the g->atomicstatus is in a Gscan status until the routine that
@@ -328,3 +364,170 @@ func lockextra(nilokay bool) *m {
 func unlockextra(mp *m) {
 	atomic.Storeuintptr(&extram, uintptr(unsafe.Pointer(mp)))
 }
+
+// Check for deadlock situation.
+// The check is based on number of running M's, if 0 -> deadlock.
+func checkdead() {
+	// For -buildmode=c-shared or -buildmode=c-archive it's OK if
+	// there are no running goroutines. The calling program is
+	// assumed to be running.
+	if islibrary || isarchive {
+		return
+	}
+
+	// If we are dying because of a signal caught on an already idle thread,
+	// freezetheworld will cause all running threads to block.
+	// And runtime will essentially enter into deadlock state,
+	// except that there is a thread that will call exit soon.
+	if panicking > 0 {
+		return
+	}
+
+	// -1 for sysmon
+	run := sched.mcount - sched.nmidle - sched.nmidlelocked - 1
+	if run > 0 {
+		return
+	}
+	if run < 0 {
+		print("runtime: checkdead: nmidle=", sched.nmidle, " nmidlelocked=", sched.nmidlelocked, " mcount=", sched.mcount, "\n")
+		throw("checkdead: inconsistent counts")
+	}
+
+	grunning := 0
+	lock(&allglock)
+	for i := 0; i < len(allgs); i++ {
+		gp := allgs[i]
+		if isSystemGoroutine(gp) {
+			continue
+		}
+		s := readgstatus(gp)
+		switch s &^ _Gscan {
+		case _Gwaiting:
+			grunning++
+		case _Grunnable,
+			_Grunning,
+			_Gsyscall:
+			unlock(&allglock)
+			print("runtime: checkdead: find g ", gp.goid, " in status ", s, "\n")
+			throw("checkdead: runnable g")
+		}
+	}
+	unlock(&allglock)
+	if grunning == 0 { // possible if main goroutine calls runtime·Goexit()
+		throw("no goroutines (main called runtime.Goexit) - deadlock!")
+	}
+
+	// Maybe jump time forward for playground.
+	gp := timejump()
+	if gp != nil {
+		// Temporarily commented out for gccgo.
+		// For gccgo this code will never run anyhow.
+		// casgstatus(gp, _Gwaiting, _Grunnable)
+		// globrunqput(gp)
+		// _p_ := pidleget()
+		// if _p_ == nil {
+		// 	throw("checkdead: no p for timer")
+		// }
+		// mp := mget()
+		// if mp == nil {
+		// 	// There should always be a free M since
+		// 	// nothing is running.
+		// 	throw("checkdead: no m for timer")
+		// }
+		// nmp.nextp.set(_p_)
+		// notewakeup(&mp.park)
+		// return
+	}
+
+	getg().m.throwing = -1 // do not dump full stacks
+	throw("all goroutines are asleep - deadlock!")
+}
+
+var starttime int64
+
+func schedtrace(detailed bool) {
+	now := nanotime()
+	if starttime == 0 {
+		starttime = now
+	}
+
+	gomaxprocs := int32(GOMAXPROCS(0))
+
+	lock(&sched.lock)
+	print("SCHED ", (now-starttime)/1e6, "ms: gomaxprocs=", gomaxprocs, " idleprocs=", sched.npidle, " threads=", sched.mcount, " spinningthreads=", sched.nmspinning, " idlethreads=", sched.nmidle, " runqueue=", sched.runqsize)
+	if detailed {
+		print(" gcwaiting=", sched.gcwaiting, " nmidlelocked=", sched.nmidlelocked, " stopwait=", sched.stopwait, " sysmonwait=", sched.sysmonwait, "\n")
+	}
+	// We must be careful while reading data from P's, M's and G's.
+	// Even if we hold schedlock, most data can be changed concurrently.
+	// E.g. (p->m ? p->m->id : -1) can crash if p->m changes from non-nil to nil.
+	for i := int32(0); i < gomaxprocs; i++ {
+		_p_ := allp[i]
+		if _p_ == nil {
+			continue
+		}
+		mp := _p_.m.ptr()
+		h := atomic.Load(&_p_.runqhead)
+		t := atomic.Load(&_p_.runqtail)
+		if detailed {
+			id := int32(-1)
+			if mp != nil {
+				id = mp.id
+			}
+			print("  P", i, ": status=", _p_.status, " schedtick=", _p_.schedtick, " syscalltick=", _p_.syscalltick, " m=", id, " runqsize=", t-h, " gfreecnt=", _p_.gfreecnt, "\n")
+		} else {
+			// In non-detailed mode format lengths of per-P run queues as:
+			// [len1 len2 len3 len4]
+			print(" ")
+			if i == 0 {
+				print("[")
+			}
+			print(t - h)
+			if i == gomaxprocs-1 {
+				print("]\n")
+			}
+		}
+	}
+
+	if !detailed {
+		unlock(&sched.lock)
+		return
+	}
+
+	for mp := allm(); mp != nil; mp = mp.alllink {
+		_p_ := mp.p.ptr()
+		gp := mp.curg
+		lockedg := mp.lockedg
+		id1 := int32(-1)
+		if _p_ != nil {
+			id1 = _p_.id
+		}
+		id2 := int64(-1)
+		if gp != nil {
+			id2 = gp.goid
+		}
+		id3 := int64(-1)
+		if lockedg != nil {
+			id3 = lockedg.goid
+		}
+		print("  M", mp.id, ": p=", id1, " curg=", id2, " mallocing=", mp.mallocing, " throwing=", mp.throwing, " preemptoff=", mp.preemptoff, ""+" locks=", mp.locks, " dying=", mp.dying, " helpgc=", mp.helpgc, " spinning=", mp.spinning, " blocked=", mp.blocked, " lockedg=", id3, "\n")
+	}
+
+	lock(&allglock)
+	for gi := 0; gi < len(allgs); gi++ {
+		gp := allgs[gi]
+		mp := gp.m
+		lockedm := gp.lockedm
+		id1 := int32(-1)
+		if mp != nil {
+			id1 = mp.id
+		}
+		id2 := int32(-1)
+		if lockedm != nil {
+			id2 = lockedm.id
+		}
+		print("  G", gp.goid, ": status=", readgstatus(gp), "(", gp.waitreason, ") m=", id1, " lockedm=", id2, "\n")
+	}
+	unlock(&allglock)
+	unlock(&sched.lock)
+}
diff --git a/libgo/go/runtime/runtime2.go b/libgo/go/runtime/runtime2.go
index 6686e1f..571972c 100644
--- a/libgo/go/runtime/runtime2.go
+++ b/libgo/go/runtime/runtime2.go
@@ -755,9 +755,13 @@ const _TracebackMaxFrames = 100
 
 var (
 	//	emptystring string
-	//	allglen     uintptr
+
+	allglen uintptr
+
 	//	allm        *m
-	//	allp        [_MaxGomaxprocs + 1]*p
+
+	allp [_MaxGomaxprocs + 1]*p
+
 	//	gomaxprocs  int32
 
 	panicking uint32
diff --git a/libgo/go/runtime/stubs.go b/libgo/go/runtime/stubs.go
index c299ae0..3d18408 100644
--- a/libgo/go/runtime/stubs.go
+++ b/libgo/go/runtime/stubs.go
@@ -476,12 +476,6 @@ func UnlockOSThread()
 func lockOSThread()
 func unlockOSThread()
 func allm() *m
-func allgs() []*g
-
-//go:nosplit
-func readgstatus(gp *g) uint32 {
-	return atomic.Load(&gp.atomicstatus)
-}
 
 // Temporary for gccgo until we port malloc.go
 func persistentalloc(size, align uintptr, sysStat *uint64) unsafe.Pointer
@@ -489,9 +483,6 @@ func persistentalloc(size, align uintptr, sysStat *uint64) unsafe.Pointer
 // Temporary for gccgo until we port mheap.go
 func setprofilebucket(p unsafe.Pointer, b *bucket)
 
-// Currently in proc.c.
-func tracebackothers(*g)
-
 // Temporary for gccgo until we port mgc.go.
 func setgcpercent(int32) int32
 
@@ -530,9 +521,7 @@ func getZerobase() *uintptr {
 // Temporary for gccgo until we port proc.go.
 func sigprof()
 func mcount() int32
-func gcount() int32
 func goexit1()
-func schedtrace(bool)
 func freezetheworld()
 
 // Get signal trampoline, written in C.
@@ -562,6 +551,30 @@ func getCgoHasExtraM() *bool {
 	return &cgoHasExtraM
 }
 
+// Temporary for gccgo until we port proc.go.
+//go:linkname getAllP runtime.getAllP
+func getAllP() **p {
+	return &allp[0]
+}
+
+// Temporary for gccgo until we port proc.go.
+//go:linkname allocg runtime.allocg
+func allocg() *g {
+	return new(g)
+}
+
+// Temporary for gccgo until we port the garbage collector.
+//go:linkname getallglen runtime.getallglen
+func getallglen() uintptr {
+	return allglen
+}
+
+// Temporary for gccgo until we port the garbage collector.
+//go:linkname getallg runtime.getallg
+func getallg(i int) *g {
+	return allgs[i]
+}
+
 // Throw and rethrow an exception.
 func throwException()
 func rethrowException()
@@ -579,3 +592,27 @@ func getPanicking() uint32 {
 
 // Temporary for gccgo until we port mcache.go.
 func allocmcache() *mcache
+
+// Temporary for gccgo until we port mgc.go.
+// This is just so that allgadd will compile.
+var work struct {
+	rescan struct {
+		lock mutex
+		list []guintptr
+	}
+}
+
+// gcount is temporary for gccgo until more of proc.go is ported.
+// This is a copy of the C function we used to use.
+func gcount() int32 {
+	n := int32(0)
+	lock(&allglock)
+	for _, gp := range allgs {
+		s := readgstatus(gp)
+		if s == _Grunnable || s == _Grunning || s == _Gsyscall || s == _Gwaiting {
+			n++
+		}
+	}
+	unlock(&allglock)
+	return n
+}
diff --git a/libgo/go/runtime/traceback_gccgo.go b/libgo/go/runtime/traceback_gccgo.go
index f61f9a0..611aba9 100644
--- a/libgo/go/runtime/traceback_gccgo.go
+++ b/libgo/go/runtime/traceback_gccgo.go
@@ -171,3 +171,58 @@ func isSystemGoroutine(gp *g) bool {
 	// FIXME.
 	return false
 }
+
+func tracebackothers(me *g) {
+	var tb tracebackg
+	tb.gp = me
+
+	level, _, _ := gotraceback()
+
+	// Show the current goroutine first, if we haven't already.
+	g := getg()
+	gp := g.m.curg
+	if gp != nil && gp != me {
+		print("\n")
+		goroutineheader(gp)
+		gp.traceback = &tb
+		getTraceback(me, gp)
+		printtrace(tb.locbuf[:tb.c], nil)
+		printcreatedby(gp)
+	}
+
+	lock(&allglock)
+	for _, gp := range allgs {
+		if gp == me || gp == g.m.curg || readgstatus(gp) == _Gdead || isSystemGoroutine(gp) && level < 2 {
+			continue
+		}
+		print("\n")
+		goroutineheader(gp)
+
+		// gccgo's only mechanism for doing a stack trace is
+		// _Unwind_Backtrace.  And that only works for the
+		// current thread, not for other random goroutines.
+		// So we need to switch context to the goroutine, get
+		// the backtrace, and then switch back.
+		//
+		// This means that if g is running or in a syscall, we
+		// can't reliably print a stack trace.  FIXME.
+
+		// Note: gp.m == g.m occurs when tracebackothers is
+		// called from a signal handler initiated during a
+		// systemstack call. The original G is still in the
+		// running state, and we want to print its stack.
+		if gp.m != g.m && readgstatus(gp)&^_Gscan == _Grunning {
+			print("\tgoroutine running on other thread; stack unavailable\n")
+			printcreatedby(gp)
+		} else if readgstatus(gp)&^_Gscan == _Gsyscall {
+			print("\tgoroutine in C code; stack unavailable\n")
+			printcreatedby(gp)
+		} else {
+			gp.traceback = &tb
+			getTraceback(me, gp)
+			printtrace(tb.locbuf[:tb.c], nil)
+			printcreatedby(gp)
+		}
+	}
+	unlock(&allglock)
+}
diff --git a/libgo/runtime/go-libmain.c b/libgo/runtime/go-libmain.c
index a404eb9..8e07e90 100644
--- a/libgo/runtime/go-libmain.c
+++ b/libgo/runtime/go-libmain.c
@@ -13,7 +13,6 @@
 #include <unistd.h>
 
 #include "runtime.h"
-#include "go-alloc.h"
 #include "array.h"
 #include "arch.h"
 #include "malloc.h"
diff --git a/libgo/runtime/go-main.c b/libgo/runtime/go-main.c
index b8c9af1..dba8085 100644
--- a/libgo/runtime/go-main.c
+++ b/libgo/runtime/go-main.c
@@ -15,7 +15,6 @@
 #endif
 
 #include "runtime.h"
-#include "go-alloc.h"
 #include "array.h"
 #include "arch.h"
 #include "malloc.h"
diff --git a/libgo/runtime/go-new.c b/libgo/runtime/go-new.c
index 01bc2af..da44074 100644
--- a/libgo/runtime/go-new.c
+++ b/libgo/runtime/go-new.c
@@ -4,7 +4,6 @@
    Use of this source code is governed by a BSD-style
    license that can be found in the LICENSE file.  */
 
-#include "go-alloc.h"
 #include "runtime.h"
 #include "arch.h"
 #include "malloc.h"
diff --git a/libgo/runtime/go-reflect-call.c b/libgo/runtime/go-reflect-call.c
index dc2f605..6a9a7f3 100644
--- a/libgo/runtime/go-reflect-call.c
+++ b/libgo/runtime/go-reflect-call.c
@@ -9,7 +9,6 @@
 #include <stdlib.h>
 
 #include "runtime.h"
-#include "go-alloc.h"
 #include "go-assert.h"
 #include "go-type.h"
 
diff --git a/libgo/runtime/go-unwind.c b/libgo/runtime/go-unwind.c
index 9e85b4b..4c9fb49 100644
--- a/libgo/runtime/go-unwind.c
+++ b/libgo/runtime/go-unwind.c
@@ -14,7 +14,6 @@
 #include "unwind-pe.h"
 
 #include "runtime.h"
-#include "go-alloc.h"
 
 /* The code for a Go exception.  */
 
diff --git a/libgo/runtime/heapdump.c b/libgo/runtime/heapdump.c
index 80d2b7b..c050541 100644
--- a/libgo/runtime/heapdump.c
+++ b/libgo/runtime/heapdump.c
@@ -311,8 +311,8 @@ dumpgs(void)
 	uint32 i;
 
 	// goroutines & stacks
-	for(i = 0; i < runtime_allglen; i++) {
-		gp = runtime_allg[i];
+	for(i = 0; i < runtime_getallglen(); i++) {
+		gp = runtime_getallg(i);
 		switch(gp->atomicstatus){
 		default:
 			runtime_printf("unexpected G.status %d\n", gp->atomicstatus);
diff --git a/libgo/runtime/malloc.goc b/libgo/runtime/malloc.goc
index 5cbdc46..1e6704c 100644
--- a/libgo/runtime/malloc.goc
+++ b/libgo/runtime/malloc.goc
@@ -10,7 +10,6 @@ package runtime
 #include <stddef.h>
 #include <errno.h>
 #include <stdlib.h>
-#include "go-alloc.h"
 #include "runtime.h"
 #include "arch.h"
 #include "malloc.h"
@@ -308,107 +307,6 @@ runtime_profilealloc(void *v, uintptr size)
 	runtime_MProf_Malloc(v, size);
 }
 
-void*
-__go_alloc(uintptr size)
-{
-	return runtime_mallocgc(size, 0, FlagNoInvokeGC);
-}
-
-// Free the object whose base pointer is v.
-void
-__go_free(void *v)
-{
-	M *m;
-	int32 sizeclass;
-	MSpan *s;
-	MCache *c;
-	uintptr size;
-
-	if(v == nil)
-		return;
-	
-	// If you change this also change mgc0.c:/^sweep,
-	// which has a copy of the guts of free.
-
-	m = runtime_m();
-	if(m->mallocing)
-		runtime_throw("malloc/free - deadlock");
-	m->mallocing = 1;
-
-	if(!runtime_mlookup(v, nil, nil, &s)) {
-		runtime_printf("free %p: not an allocated block\n", v);
-		runtime_throw("free runtime_mlookup");
-	}
-	size = s->elemsize;
-	sizeclass = s->sizeclass;
-	// Objects that are smaller than TinySize can be allocated using tiny alloc,
-	// if then such object is combined with an object with finalizer, we will crash.
-	if(size < TinySize)
-		runtime_throw("freeing too small block");
-
-	if(runtime_debug.allocfreetrace)
-		runtime_tracefree(v, size);
-
-	// Ensure that the span is swept.
-	// If we free into an unswept span, we will corrupt GC bitmaps.
-	runtime_MSpan_EnsureSwept(s);
-
-	if(s->specials != nil)
-		runtime_freeallspecials(s, v, size);
-
-	c = m->mcache;
-	if(sizeclass == 0) {
-		// Large object.
-		s->needzero = 1;
-		// Must mark v freed before calling unmarkspan and MHeap_Free:
-		// they might coalesce v into other spans and change the bitmap further.
-		runtime_markfreed(v);
-		runtime_unmarkspan(v, 1<<PageShift);
-		// NOTE(rsc,dvyukov): The original implementation of efence
-		// in CL 22060046 used SysFree instead of SysFault, so that
-		// the operating system would eventually give the memory
-		// back to us again, so that an efence program could run
-		// longer without running out of memory. Unfortunately,
-		// calling SysFree here without any kind of adjustment of the
-		// heap data structures means that when the memory does
-		// come back to us, we have the wrong metadata for it, either in
-		// the MSpan structures or in the garbage collection bitmap.
-		// Using SysFault here means that the program will run out of
-		// memory fairly quickly in efence mode, but at least it won't
-		// have mysterious crashes due to confused memory reuse.
-		// It should be possible to switch back to SysFree if we also 
-		// implement and then call some kind of MHeap_DeleteSpan.
-		if(runtime_debug.efence)
-			runtime_SysFault((void*)(s->start<<PageShift), size);
-		else
-			runtime_MHeap_Free(&runtime_mheap, s, 1);
-		c->local_nlargefree++;
-		c->local_largefree += size;
-	} else {
-		// Small object.
-		if(size > 2*sizeof(uintptr))
-			((uintptr*)v)[1] = (uintptr)0xfeedfeedfeedfeedll;	// mark as "needs to be zeroed"
-		else if(size > sizeof(uintptr))
-			((uintptr*)v)[1] = 0;
-		// Must mark v freed before calling MCache_Free:
-		// it might coalesce v and other blocks into a bigger span
-		// and change the bitmap further.
-		c->local_nsmallfree[sizeclass]++;
-		c->local_cachealloc -= size;
-		if(c->alloc[sizeclass] == s) {
-			// We own the span, so we can just add v to the freelist
-			runtime_markfreed(v);
-			((MLink*)v)->next = s->freelist;
-			s->freelist = v;
-			s->ref--;
-		} else {
-			// Someone else owns this span.  Add to free queue.
-			runtime_MCache_Free(c, v, sizeclass, size);
-		}
-	}
-	m->mallocing = 0;
-}
-
 int32
 runtime_mlookup(void *v, byte **base, uintptr *size, MSpan **sp)
 {
@@ -628,9 +526,6 @@ runtime_mallocinit(void)
 	// Initialize the rest of the allocator.	
 	runtime_MHeap_Init(&runtime_mheap);
 	runtime_m()->mcache = runtime_allocmcache();
-
-	// See if it works.
-	runtime_free(runtime_malloc(TinySize));
 }
 
 void*
diff --git a/libgo/runtime/mgc0.c b/libgo/runtime/mgc0.c
index aa8404e..5d6275a 100644
--- a/libgo/runtime/mgc0.c
+++ b/libgo/runtime/mgc0.c
@@ -1279,7 +1279,6 @@ markroot(ParFor *desc, uint32 i)
 		// For gccgo we use this for all the other global roots.
 		enqueue1(&wbuf, (Obj){(byte*)&runtime_m0, sizeof runtime_m0, 0});
 		enqueue1(&wbuf, (Obj){(byte*)&runtime_g0, sizeof runtime_g0, 0});
-		enqueue1(&wbuf, (Obj){(byte*)&runtime_allg, sizeof runtime_allg, 0});
 		enqueue1(&wbuf, (Obj){(byte*)&runtime_allm, sizeof runtime_allm, 0});
 		enqueue1(&wbuf, (Obj){(byte*)&runtime_allp, sizeof runtime_allp, 0});
 		enqueue1(&wbuf, (Obj){(byte*)&work, sizeof work, 0});
@@ -1334,9 +1333,9 @@ markroot(ParFor *desc, uint32 i)
 
 	default:
 		// the rest is scanning goroutine stacks
-		if(i - RootCount >= runtime_allglen)
+		if(i - RootCount >= runtime_getallglen())
 			runtime_throw("markroot: bad index");
-		gp = runtime_allg[i - RootCount];
+		gp = runtime_getallg(i - RootCount);
 		// remember when we've first observed the G blocked
 		// needed only to output in traceback
 		if((gp->atomicstatus == _Gwaiting || gp->atomicstatus == _Gsyscall) && gp->waitsince == 0)
@@ -2243,7 +2242,7 @@ gc(struct gc_args *args)
 	work.nwait = 0;
 	work.ndone = 0;
 	work.nproc = runtime_gcprocs();
-	runtime_parforsetup(work.markfor, work.nproc, RootCount + runtime_allglen, false, &markroot_funcval);
+	runtime_parforsetup(work.markfor, work.nproc, RootCount + runtime_getallglen(), false, &markroot_funcval);
 	if(work.nproc > 1) {
 		runtime_noteclear(&work.alldone);
 		runtime_helpgc(work.nproc);
diff --git a/libgo/runtime/parfor.c b/libgo/runtime/parfor.c
index ede921b..c2f10b9 100644
--- a/libgo/runtime/parfor.c
+++ b/libgo/runtime/parfor.c
@@ -5,6 +5,7 @@
 // Parallel for algorithm.
 
 #include "runtime.h"
+#include "malloc.h"
 #include "arch.h"
 
 struct ParForThread
@@ -27,7 +28,7 @@ runtime_parforalloc(uint32 nthrmax)
 
 	// The ParFor object is followed by CacheLineSize padding
 	// and then nthrmax ParForThread.
-	desc = (ParFor*)runtime_malloc(sizeof(ParFor) + CacheLineSize + nthrmax * sizeof(ParForThread));
+	desc = (ParFor*)runtime_mallocgc(sizeof(ParFor) + CacheLineSize + nthrmax * sizeof(ParForThread), 0, FlagNoInvokeGC);
 	desc->thr = (ParForThread*)((byte*)(desc+1) + CacheLineSize);
 	desc->nthrmax = nthrmax;
 	return desc;
diff --git a/libgo/runtime/proc.c b/libgo/runtime/proc.c
index 586a632..8a7a2d7 100644
--- a/libgo/runtime/proc.c
+++ b/libgo/runtime/proc.c
@@ -361,6 +361,10 @@ enum
 extern Sched* runtime_getsched() __asm__ (GOSYM_PREFIX "runtime.getsched");
 extern bool* runtime_getCgoHasExtraM()
   __asm__ (GOSYM_PREFIX "runtime.getCgoHasExtraM");
+extern P** runtime_getAllP()
+  __asm__ (GOSYM_PREFIX "runtime.getAllP");
+extern G* allocg(void)
+  __asm__ (GOSYM_PREFIX "runtime.allocg");
 
 Sched*	runtime_sched;
 int32	runtime_gomaxprocs;
@@ -374,11 +378,6 @@ int32	runtime_ncpu;
 bool	runtime_precisestack;
 static int32	newprocs;
 
-static	Lock allglock;	// the following vars are protected by this lock or by stoptheworld
-G**	runtime_allg;
-uintptr runtime_allglen;
-static	uintptr allgcap;
-
 bool	runtime_isarchive;
 
 void* runtime_mstart(void*);
@@ -403,7 +402,6 @@ static void startlockedm(G*);
 static void sysmon(void);
 static uint32 retake(int64);
 static void incidlelocked(int32);
-static void checkdead(void);
 static void exitsyscall0(G*);
 static void park0(G*);
 static void goexit0(G*);
@@ -421,6 +419,8 @@ static bool exitsyscallfast(void);
 
 void allgadd(G*)
   __asm__(GOSYM_PREFIX "runtime.allgadd");
+void checkdead(void)
+  __asm__(GOSYM_PREFIX "runtime.checkdead");
 
 bool runtime_isstarted;
 
@@ -482,7 +482,7 @@ runtime_schedinit(void)
 			n = _MaxGomaxprocs;
 		procs = n;
 	}
-	runtime_allp = runtime_malloc((_MaxGomaxprocs+1)*sizeof(runtime_allp[0]));
+	runtime_allp = runtime_getAllP();
 	procresize(procs);
 
 	// Can not enable GC until all roots are registered.
@@ -586,85 +586,25 @@ runtime_main(void* dummy __attribute__((unused)))
 		*(int32*)0 = 0;
 }
 
-void
-runtime_tracebackothers(G * volatile me)
-{
-	G * volatile gp;
-	Traceback tb;
-	int32 traceback;
-	Slice slice;
-	volatile uintptr i;
-
-	tb.gp = me;
-	traceback = runtime_gotraceback(nil);
-	
-	// Show the current goroutine first, if we haven't already.
-	if((gp = g->m->curg) != nil && gp != me) {
-		runtime_printf("\n");
-		runtime_goroutineheader(gp);
-		gp->traceback = &tb;
-
-#ifdef USING_SPLIT_STACK
-		__splitstack_getcontext(&me->stackcontext[0]);
-#endif
-		getcontext(ucontext_arg(&me->context[0]));
-
-		if(gp->traceback != nil) {
-		  runtime_gogo(gp);
-		}
-
-		slice.__values = &tb.locbuf[0];
-		slice.__count = tb.c;
-		slice.__capacity = tb.c;
-		runtime_printtrace(slice, nil);
-		runtime_printcreatedby(gp);
-	}
-
-	runtime_lock(&allglock);
-	for(i = 0; i < runtime_allglen; i++) {
-		gp = runtime_allg[i];
-		if(gp == me || gp == g->m->curg || gp->atomicstatus == _Gdead)
-			continue;
-		if(gp->issystem && traceback < 2)
-			continue;
-		runtime_printf("\n");
-		runtime_goroutineheader(gp);
-
-		// Our only mechanism for doing a stack trace is
-		// _Unwind_Backtrace.  And that only works for the
-		// current thread, not for other random goroutines.
-		// So we need to switch context to the goroutine, get
-		// the backtrace, and then switch back.
-
-		// This means that if g is running or in a syscall, we
-		// can't reliably print a stack trace.  FIXME.
-
-		if(gp->atomicstatus == _Grunning) {
-			runtime_printf("\tgoroutine running on other thread; stack unavailable\n");
-			runtime_printcreatedby(gp);
-		} else if(gp->atomicstatus == _Gsyscall) {
-			runtime_printf("\tgoroutine in C code; stack unavailable\n");
-			runtime_printcreatedby(gp);
-		} else {
-			gp->traceback = &tb;
+void getTraceback(G*, G*) __asm__(GOSYM_PREFIX "runtime.getTraceback");
 
+// getTraceback stores a traceback of gp in the g's traceback field
+// and then returns to me.  We expect that gp's traceback is not nil.
+// It works by saving me's current context, and checking gp's traceback field.
+// If gp's traceback field is not nil, it starts running gp.
+// In places where we call getcontext, we check the traceback field.
+// If it is not nil, we collect a traceback, and then return to the
+// goroutine stored in the traceback field, which is me.
+void getTraceback(G* me, G* gp)
+{
 #ifdef USING_SPLIT_STACK
-			__splitstack_getcontext(&me->stackcontext[0]);
+	__splitstack_getcontext(&me->stackcontext[0]);
 #endif
-			getcontext(ucontext_arg(&me->context[0]));
+	getcontext(ucontext_arg(&me->stackcontext[0]));
 
-			if(gp->traceback != nil) {
-				runtime_gogo(gp);
-			}
-
-			slice.__values = &tb.locbuf[0];
-			slice.__count = tb.c;
-			slice.__capacity = tb.c;
-			runtime_printtrace(slice, nil);
-			runtime_printcreatedby(gp);
-		}
+	if (gp->traceback != nil) {
+		runtime_gogo(gp);
 	}
-	runtime_unlock(&allglock);
 }
 
 static void
@@ -1067,22 +1007,6 @@ runtime_allocm(P *p, bool allocatestack, byte** ret_g0_stack, uintptr* ret_g0_st
 	return mp;
 }
 
-static G*
-allocg(void)
-{
-	G *gp;
-	// static Type *gtype;
-	
-	// if(gtype == nil) {
-	// 	Eface e;
-	// 	runtime_gc_g_ptr(&e);
-	// 	gtype = ((PtrType*)e.__type_descriptor)->__element_type;
-	// }
-	// gp = runtime_cnew(gtype);
-	gp = runtime_malloc(sizeof(G));
-	return gp;
-}
-
 void setGContext(void) __asm__ (GOSYM_PREFIX "runtime.setGContext");
 
 // setGContext sets up a new goroutine context for the current g.
@@ -2129,6 +2053,7 @@ __go_go(void (*fn)(void*), void* arg)
 
 		newg = runtime_malg(true, false, &sp, &malsize);
 		spsize = (size_t)malsize;
+		newg->atomicstatus = _Gdead;
 		allgadd(newg);
 	}
 
@@ -2152,31 +2077,6 @@ __go_go(void (*fn)(void*), void* arg)
 	return newg;
 }
 
-void
-allgadd(G *gp)
-{
-	G **new;
-	uintptr cap;
-
-	runtime_lock(&allglock);
-	if(runtime_allglen >= allgcap) {
-		cap = 4096/sizeof(new[0]);
-		if(cap < 2*allgcap)
-			cap = 2*allgcap;
-		new = runtime_malloc(cap*sizeof(new[0]));
-		if(new == nil)
-			runtime_throw("runtime: cannot allocate memory");
-		if(runtime_allg != nil) {
-			runtime_memmove(new, runtime_allg, runtime_allglen*sizeof(new[0]));
-			runtime_free(runtime_allg);
-		}
-		runtime_allg = new;
-		allgcap = cap;
-	}
-	runtime_allg[runtime_allglen++] = gp;
-	runtime_unlock(&allglock);
-}
-
 // Put on gfree list.
 // If local list is too long, transfer a batch to the global list.
 static void
@@ -2352,29 +2252,6 @@ runtime_lockedOSThread(void)
 }
 
 int32
-runtime_gcount(void)
-{
-	G *gp;
-	int32 n, s;
-	uintptr i;
-
-	n = 0;
-	runtime_lock(&allglock);
-	// TODO(dvyukov): runtime.NumGoroutine() is O(N).
-	// We do not want to increment/decrement centralized counter in newproc/goexit,
-	// just to make runtime.NumGoroutine() faster.
-	// Compromise solution is to introduce per-P counters of active goroutines.
-	for(i = 0; i < runtime_allglen; i++) {
-		gp = runtime_allg[i];
-		s = gp->atomicstatus;
-		if(s == _Grunnable || s == _Grunning || s == _Gsyscall || s == _Gwaiting)
-			n++;
-	}
-	runtime_unlock(&allglock);
-	return n;
-}
-
-int32
 runtime_mcount(void)
 {
 	return runtime_sched->mcount;
@@ -2638,59 +2515,6 @@ incidlelocked(int32 v)
 	runtime_unlock(&runtime_sched->lock);
 }
 
-// Check for deadlock situation.
-// The check is based on number of running M's, if 0 -> deadlock.
-static void
-checkdead(void)
-{
-	G *gp;
-	int32 run, grunning, s;
-	uintptr i;
-
-	// For -buildmode=c-shared or -buildmode=c-archive it's OK if
-	// there are no running goroutines.  The calling program is
-	// assumed to be running.
-	if(runtime_isarchive) {
-		return;
-	}
-
-	// -1 for sysmon
-	run = runtime_sched->mcount - runtime_sched->nmidle - runtime_sched->nmidlelocked - 1;
-	if(run > 0)
-		return;
-	// If we are dying because of a signal caught on an already idle thread,
-	// freezetheworld will cause all running threads to block.
-	// And runtime will essentially enter into deadlock state,
-	// except that there is a thread that will call runtime_exit soon.
-	if(runtime_panicking() > 0)
-		return;
-	if(run < 0) {
-		runtime_printf("runtime: checkdead: nmidle=%d nmidlelocked=%d mcount=%d\n",
-			runtime_sched->nmidle, runtime_sched->nmidlelocked, runtime_sched->mcount);
-		runtime_throw("checkdead: inconsistent counts");
-	}
-	grunning = 0;
-	runtime_lock(&allglock);
-	for(i = 0; i < runtime_allglen; i++) {
-		gp = runtime_allg[i];
-		if(gp->isbackground)
-			continue;
-		s = gp->atomicstatus;
-		if(s == _Gwaiting)
-			grunning++;
-		else if(s == _Grunnable || s == _Grunning || s == _Gsyscall) {
-			runtime_unlock(&allglock);
-			runtime_printf("runtime: checkdead: find g %D in status %d\n", gp->goid, s);
-			runtime_throw("checkdead: runnable g");
-		}
-	}
-	runtime_unlock(&allglock);
-	if(grunning == 0)  // possible if main goroutine calls runtime_Goexit()
-		runtime_throw("no goroutines (main called runtime.Goexit) - deadlock!");
-	g->m->throwing = -1;  // do not dump full stacks
-	runtime_throw("all goroutines are asleep - deadlock!");
-}
-
 static void
 sysmon(void)
 {
@@ -2832,94 +2656,6 @@ preemptall(void)
 	return false;
 }
 
-void
-runtime_schedtrace(bool detailed)
-{
-	static int64 starttime;
-	int64 now;
-	int64 id1, id2, id3;
-	int32 i, t, h;
-	uintptr gi;
-	const char *fmt;
-	M *mp, *lockedm;
-	G *gp, *lockedg;
-	P *p;
-
-	now = runtime_nanotime();
-	if(starttime == 0)
-		starttime = now;
-
-	runtime_lock(&runtime_sched->lock);
-	runtime_printf("SCHED %Dms: gomaxprocs=%d idleprocs=%d threads=%d idlethreads=%d runqueue=%d",
-		(now-starttime)/1000000, runtime_gomaxprocs, runtime_sched->npidle, runtime_sched->mcount,
-		runtime_sched->nmidle, runtime_sched->runqsize);
-	if(detailed) {
-		runtime_printf(" gcwaiting=%d nmidlelocked=%d nmspinning=%d stopwait=%d sysmonwait=%d\n",
-			runtime_sched->gcwaiting, runtime_sched->nmidlelocked, runtime_sched->nmspinning,
-			runtime_sched->stopwait, runtime_sched->sysmonwait);
-	}
-	// We must be careful while reading data from P's, M's and G's.
-	// Even if we hold schedlock, most data can be changed concurrently.
-	// E.g. (p->m ? p->m->id : -1) can crash if p->m changes from non-nil to nil.
-	for(i = 0; i < runtime_gomaxprocs; i++) {
-		p = runtime_allp[i];
-		if(p == nil)
-			continue;
-		mp = (M*)p->m;
-		h = runtime_atomicload(&p->runqhead);
-		t = runtime_atomicload(&p->runqtail);
-		if(detailed)
-			runtime_printf("  P%d: status=%d schedtick=%d syscalltick=%d m=%d runqsize=%d gfreecnt=%d\n",
-				i, p->status, p->schedtick, p->syscalltick, mp ? mp->id : -1, t-h, p->gfreecnt);
-		else {
-			// In non-detailed mode format lengths of per-P run queues as:
-			// [len1 len2 len3 len4]
-			fmt = " %d";
-			if(runtime_gomaxprocs == 1)
-				fmt = " [%d]\n";
-			else if(i == 0)
-				fmt = " [%d";
-			else if(i == runtime_gomaxprocs-1)
-				fmt = " %d]\n";
-			runtime_printf(fmt, t-h);
-		}
-	}
-	if(!detailed) {
-		runtime_unlock(&runtime_sched->lock);
-		return;
-	}
-	for(mp = runtime_allm; mp; mp = mp->alllink) {
-		p = (P*)mp->p;
-		gp = mp->curg;
-		lockedg = mp->lockedg;
-		id1 = -1;
-		if(p)
-			id1 = p->id;
-		id2 = -1;
-		if(gp)
-			id2 = gp->goid;
-		id3 = -1;
-		if(lockedg)
-			id3 = lockedg->goid;
-		runtime_printf("  M%d: p=%D curg=%D mallocing=%d throwing=%d gcing=%d"
-			" locks=%d dying=%d helpgc=%d spinning=%d blocked=%d lockedg=%D\n",
-			mp->id, id1, id2,
-			mp->mallocing, mp->throwing, mp->gcing, mp->locks, mp->dying, mp->helpgc,
-			mp->spinning, mp->blocked, id3);
-	}
-	runtime_lock(&allglock);
-	for(gi = 0; gi < runtime_allglen; gi++) {
-		gp = runtime_allg[gi];
-		mp = gp->m;
-		lockedm = gp->lockedm;
-		runtime_printf("  G%D: status=%d(%S) m=%d lockedm=%d\n",
-			gp->goid, gp->atomicstatus, gp->waitreason, mp ? mp->id : -1,
-			lockedm ? lockedm->id : -1);
-	}
-	runtime_unlock(&allglock);
-	runtime_unlock(&runtime_sched->lock);
-}
-
 // Put mp on midle list.
 // Sched must be locked.
 static void
@@ -3357,20 +3093,6 @@ runtime_go_allm()
 	return &runtime_allm;
 }
 
-extern Slice runtime_go_allgs(void)
-  __asm__ (GOSYM_PREFIX "runtime.allgs");
-
-Slice
-runtime_go_allgs()
-{
-	Slice s;
-
-	s.__values = runtime_allg;
-	s.__count = runtime_allglen;
-	s.__capacity = allgcap;
-	return s;
-}
-
 intgo NumCPU(void) __asm__ (GOSYM_PREFIX "runtime.NumCPU");
 
 intgo
diff --git a/libgo/runtime/runtime.h b/libgo/runtime/runtime.h
index 54bdcf8..c8f490f 100644
--- a/libgo/runtime/runtime.h
+++ b/libgo/runtime/runtime.h
@@ -7,6 +7,7 @@
 #include "go-assert.h"
 #include <complex.h>
 #include <signal.h>
+#include <stdint.h>
 #include <stdio.h>
 #include <stdlib.h>
 #include <string.h>
@@ -22,8 +23,6 @@
 #include <sys/mman.h>
 #endif
 
-#include "go-alloc.h"
-
 #define _STRINGIFY2_(x) #x
 #define _STRINGIFY_(x) _STRINGIFY2_(x)
 #define GOSYM_PREFIX _STRINGIFY_(__USER_LABEL_PREFIX__)
@@ -233,8 +232,10 @@ enum
  */
 extern	uintptr* runtime_getZerobase(void)
   __asm__(GOSYM_PREFIX "runtime.getZerobase");
-extern	G**	runtime_allg;
-extern	uintptr runtime_allglen;
+extern G* runtime_getallg(intgo)
+  __asm__(GOSYM_PREFIX "runtime.getallg");
+extern uintptr runtime_getallglen(void)
+  __asm__(GOSYM_PREFIX "runtime.getallglen");
 extern	G*	runtime_lastg;
 extern	M*	runtime_allm;
 extern	P**	runtime_allp;
@@ -309,13 +310,9 @@ MCache*	runtime_allocmcache(void)
 void	runtime_freemcache(MCache*);
 void	runtime_mallocinit(void);
 void	runtime_mprofinit(void);
-#define runtime_malloc(s) __go_alloc(s)
-#define runtime_free(p) __go_free(p)
 #define runtime_getcallersp(p) __builtin_frame_address(0)
 int32	runtime_mcount(void)
   __asm__ (GOSYM_PREFIX "runtime.mcount");
-int32	runtime_gcount(void)
-  __asm__ (GOSYM_PREFIX "runtime.gcount");
 void	runtime_mcall(void(*)(G*));
 uint32	runtime_fastrand1(void) __asm__ (GOSYM_PREFIX "runtime.fastrand1");
 int32	runtime_timediv(int64, int32, int32*)
diff --git a/libgo/runtime/runtime_c.c b/libgo/runtime/runtime_c.c
index 4645312..67d5ee2 100644
--- a/libgo/runtime/runtime_c.c
+++ b/libgo/runtime/runtime_c.c
@@ -25,7 +25,8 @@ extern volatile intgo runtime_MemProfileRate
 
 struct gotraceback_ret {
 	int32 level;
-	bool crash;
+	bool  all;
+	bool  crash;
 };
 
 extern struct gotraceback_ret gotraceback(void)