[dev.garbage] runtime: Add write barriers to c code

Also improve missing GC mark diagnostics.

LGTM=rsc
R=rsc
CC=golang-codereviews
https://golang.org/cl/169450043

src/runtime/mgc0.c

@@ -29,8 +29,7 @@
 //       Preempted goroutines are scanned before P schedules next goroutine.
 //  3. Set phase = GCmark.
 //  4. Wait for all P's to acknowledge phase change.
-//  5. Now write barrier marks and enqueues black or grey to white pointers. If a pointer is
-//       stored into a white slot, such pointer is not marked.
+//  5. Now write barrier marks and enqueues black, grey, or white to white pointers.
 //       Malloc still allocates white (non-marked) objects.
 //  6. Meanwhile GC transitively walks the heap marking reachable objects.
 //  7. When GC finishes marking heap, it preempts P's one-by-one and
@@ -446,7 +445,25 @@ greyobject(byte *obj, Markbits *mbits, Workbuf *wbuf)
 
 	if(checkmark) {
 		if(!ismarked(mbits)) {
+			MSpan *s;
+			pageID k;
+			uintptr x, i;
+
 			runtime·printf("runtime:greyobject: checkmarks finds unexpected unmarked object obj=%p, mbits->bits=%x, *mbits->bitp=%x\n", obj, mbits->bits, *mbits->bitp);
+
+			k = (uintptr)obj>>PageShift;
+			x = k;
+			x -= (uintptr)runtime·mheap.arena_start>>PageShift;
+			s = runtime·mheap.spans[x];
+			runtime·printf("runtime:greyobject Span: obj=%p, k=%p", obj, k);
+			if (s == nil) {
+				runtime·printf(" s=nil\n");
+			} else {
+				runtime·printf(" s->start=%p s->limit=%p, s->state=%d, s->sizeclass=%d, s->elemsize=%D \n", s->start*PageSize, s->limit, s->state, s->sizeclass, s->elemsize);
+				for(i=0; i<s->sizeclass; i++) {
+					runtime·printf(" ((uintptr*)obj)[%D]=%p\n", i, ((uintptr*)obj)[i]);
+				}
+			}
 			runtime·throw("checkmark found unmarked object");
 		}
 		if(ischeckmarked(mbits))

src/runtime/os_darwin.c

@@ -135,7 +135,10 @@ void
 runtime·mpreinit(M *mp)
 {
 	mp->gsignal = runtime·malg(32*1024);	// OS X wants >=8K, Linux >=2K
+	runtime·writebarrierptr_nostore(&mp->gsignal, mp->gsignal);
+
 	mp->gsignal->m = mp;
+	runtime·writebarrierptr_nostore(&mp->gsignal->m, mp->gsignal->m);
 }
 
 // Called to initialize a new m (including the bootstrap m).

src/runtime/os_dragonfly.c

@@ -195,7 +195,10 @@ void
 runtime·mpreinit(M *mp)
 {
 	mp->gsignal = runtime·malg(32*1024);
+	runtime·writebarrierptr_nostore(&mp->gsignal, mp->gsignal);
+
 	mp->gsignal->m = mp;
+	runtime·writebarrierptr_nostore(&mp->gsignal->m, mp->gsignal->m);
 }
 
 // Called to initialize a new m (including the bootstrap m).

src/runtime/os_freebsd.c

@@ -203,7 +203,10 @@ void
 runtime·mpreinit(M *mp)
 {
 	mp->gsignal = runtime·malg(32*1024);
+	runtime·writebarrierptr_nostore(&mp->gsignal, mp->gsignal);
+
 	mp->gsignal->m = mp;
+	runtime·writebarrierptr_nostore(&mp->gsignal->m, mp->gsignal->m);
 }
 
 // Called to initialize a new m (including the bootstrap m).

src/runtime/os_linux.c

@@ -233,7 +233,10 @@ void
 runtime·mpreinit(M *mp)
 {
 	mp->gsignal = runtime·malg(32*1024);	// OS X wants >=8K, Linux >=2K
+	runtime·writebarrierptr_nostore(&mp->gsignal, mp->gsignal);
+
 	mp->gsignal->m = mp;
+	runtime·writebarrierptr_nostore(&mp->gsignal->m, mp->gsignal->m);
 }
 
 // Called to initialize a new m (including the bootstrap m).

src/runtime/os_nacl.c

@@ -20,7 +20,10 @@ void
 runtime·mpreinit(M *mp)
 {
 	mp->gsignal = runtime·malg(32*1024);	// OS X wants >=8K, Linux >=2K
+	runtime·writebarrierptr_nostore(&mp->gsignal, mp->gsignal);
+
 	mp->gsignal->m = mp;
+	runtime·writebarrierptr_nostore(&mp->gsignal->m, mp->gsignal->m);
 }
 
 // Called to initialize a new m (including the bootstrap m).

src/runtime/os_netbsd.c

@@ -271,7 +271,10 @@ void
 runtime·mpreinit(M *mp)
 {
 	mp->gsignal = runtime·malg(32*1024);
+	runtime·writebarrierptr_nostore(&mp->gsignal, mp->gsignal);
+
 	mp->gsignal->m = mp;
+	runtime·writebarrierptr_nostore(&mp->gsignal->m, mp->gsignal->m);
 }
 
 // Called to initialize a new m (including the bootstrap m).

src/runtime/os_openbsd.c

@@ -217,7 +217,10 @@ void
 runtime·mpreinit(M *mp)
 {
 	mp->gsignal = runtime·malg(32*1024);
+	runtime·writebarrierptr_nostore(&mp->gsignal, mp->gsignal);
+
 	mp->gsignal->m = mp;
+	runtime·writebarrierptr_nostore(&mp->gsignal->m, mp->gsignal->m);
 }
 
 // Called to initialize a new m (including the bootstrap m).

src/runtime/os_plan9.c

@@ -20,12 +20,18 @@ runtime·mpreinit(M *mp)
 {
 	// Initialize stack and goroutine for note handling.
 	mp->gsignal = runtime·malg(32*1024);
+	runtime·writebarrierptr_nostore(&mp->gsignal, mp->gsignal);
+
 	mp->gsignal->m = mp;
+	runtime·writebarrierptr_nostore(&mp->gsignal->m, mp->gsignal->m);
+
 	mp->notesig = (int8*)runtime·mallocgc(ERRMAX*sizeof(int8), nil, FlagNoScan);
+	runtime·writebarrierptr_nostore(&mp->notesig, mp->notesig);
 
 	// Initialize stack for handling strings from the
 	// errstr system call, as used in package syscall.
 	mp->errstr = (byte*)runtime·mallocgc(ERRMAX*sizeof(byte), nil, FlagNoScan);
+	runtime·writebarrierptr_nostore(&mp->errstr, mp->errstr);
 }
 
 // Called to initialize a new m (including the bootstrap m).

src/runtime/os_solaris.c

@@ -176,7 +176,10 @@ void
 runtime·mpreinit(M *mp)
 {
 	mp->gsignal = runtime·malg(32*1024);
+	runtime·writebarrierptr_nostore(&mp->gsignal, mp->gsignal);
+
 	mp->gsignal->m = mp;
+	runtime·writebarrierptr_nostore(&mp->gsignal->m, mp->gsignal->m);
 }
 
 // Called to initialize a new m (including the bootstrap m).

src/runtime/proc.c

@@ -876,7 +876,9 @@ runtime·allocm(P *p)
 		mp->g0 = runtime·malg(-1);
 	else
 		mp->g0 = runtime·malg(8192);
+	runtime·writebarrierptr_nostore(&mp->g0, mp->g0);
 	mp->g0->m = mp;
+	runtime·writebarrierptr_nostore(&mp->g0->m, mp->g0->m);
 
 	if(p == g->m->p)
 		releasep();