libmach:

	* heuristic to go farther during stack traces.
	* significantly improved Linux thread handing.

acid:
	* update to new libmach interface.

prof:
	* use new libmach interface.
	* multiple thread support (derived from Rob's copy).
	* first steps toward pprof-like graphs:
	  keep counters indexed by pc,callerpc pairs.

R=r
DELTA=909  (576 added, 123 deleted, 210 changed)
OCL=24240
CL=24259

include/mach_amd64.h

@@ -415,7 +415,7 @@ int		ctlproc(int pid, char *msg);
 void		detachproc(Map *m);
 int		procnotes(int pid, char ***pnotes);
 char*		proctextfile(int pid);
-int		procthreadpids(int pid, int **thread);
+int		procthreadpids(int pid, int *tid, int ntid);
 char*	procstatus(int);
 
 Maprw	fdrw;

src/libmach_amd64/8db.c

@@ -145,7 +145,7 @@ static int
 i386trace(Map *map, uvlong pc, uvlong sp, uvlong link, Tracer trace)
 {
 	int i;
-	uvlong osp;
+	uvlong osp, pc1;
 	Symbol s, f, s1;
 	extern Mach mamd64;
 	int isamd64;
@@ -189,19 +189,30 @@ i386trace(Map *map, uvlong pc, uvlong sp, uvlong link, Tracer trace)
 			break;
 		}
 		s1 = s;
-
+		pc1 = 0;
 		if(pc != s.value) {	/* not at first instruction */
 			if(findlocal(&s, FRAMENAME, &f) == 0)
 				break;
+			geta(map, sp, &pc1);
 			sp += f.value-mach->szaddr;
 		}
-		if (geta(map, sp, &pc) < 0)
+		if(geta(map, sp, &pc) < 0)
 			break;
 
+		// If PC is not valid, assume we caught the function
+		// before it moved the stack pointer down or perhaps
+		// after it moved the stack pointer back up.
+		// Try the PC we'd have gotten without the stack
+		// pointer adjustment above (pc != s.value).
+		// This only matters for the first frame, and it is only
+		// a heuristic, but it does help.
+		if(!findsym(pc, CTEXT, &s) || strcmp(s.name, "etext") == 0)
+			pc = pc1;
+
 		if(pc == 0)
 			break;
 
-		if (pc != retfromnewstack)
+		if(pc != retfromnewstack)
 			(*trace)(map, pc, sp, &s1);
 		sp += mach->szaddr;
 

src/libmach_amd64/darwin.c

@@ -341,11 +341,10 @@ attachproc(int id, Fhdr *fp)
 
 // Return list of ids for threads in id.
 int
-procthreadpids(int id, int **thread)
+procthreadpids(int id, int *out, int nout)
 {
 	Thread *t;
 	int i, n, pid;
-	int *out;
 
 	t = idtotable(id);
 	if(t == nil)
@@ -353,17 +352,13 @@ procthreadpids(int id, int **thread)
 	pid = t->pid;
 	addpid(pid, 1);	// force refresh of thread list
 	n = 0;
-	for(i=0; i<nthr; i++)
-		if(thr[i].pid == pid)
+	for(i=0; i<nthr; i++) {
+		if(thr[i].pid == pid) {
+			if(n < nout)
+				out[n] = -(i+1);
 			n++;
-	out = malloc(n*sizeof out[0]);
-	if(out == nil)
-		return -1;
-	n = 0;
-	for(i=0; i<nthr; i++)
-		if(thr[i].pid == pid)
-			out[n++] = -(i+1);
-	*thread = out;
+		}
+	}
 	return n;
 }