runtime/pprof: support OS X CPU profiling
Work around profiling kernel bug with signal masks.
Still broken on 64-bit Snow Leopard kernel,
but I think we can ignore that one and let people
upgrade to Lion.
Add new trivial tools addr2line and objdump to take
the place of the GNU tools of the same name, since
those are not installed on OS X.
Adapt pprof to invoke 'go tool addr2line' and
'go tool objdump' if the system tools do not exist.
Clean up disassembly of base register on amd64.
Fixes #2008.
R=golang-dev, bradfitz, mikioh.mikioh, r, iant
CC=golang-dev
https://golang.org/cl/5697066
diff --git a/src/cmd/addr2line/main.c b/src/cmd/addr2line/main.c
new file mode 100644
index 0000000..6b2fe5d
--- /dev/null
+++ b/src/cmd/addr2line/main.c
@@ -0,0 +1,68 @@
+// Copyright 2012 The Go Authors. All rights reserved.
+// Use of this source code is governed by a BSD-style
+// license that can be found in the LICENSE file.
+
+/*
+ * addr2line simulation - only enough to make pprof work on Macs
+ */
+
+#include <u.h>
+#include <libc.h>
+#include <bio.h>
+#include <mach.h>
+
+void
+usage(void)
+{
+ fprint(2, "usage: addr2line binary\n");
+ fprint(2, "reads addresses from standard input and writes two lines for each:\n");
+ fprint(2, "\tfunction name\n");
+ fprint(2, "\tfile:line\n");
+ exits("usage");
+}
+
+void
+main(int argc, char **argv)
+{
+ int fd;
+ char *p;
+ uvlong pc;
+ Symbol s;
+ Fhdr fhdr;
+ Biobuf bin, bout;
+ char file[1024];
+
+ ARGBEGIN{
+ default:
+ usage();
+ }ARGEND
+
+ if(argc != 1)
+ usage();
+
+ fd = open(argv[0], OREAD);
+ if(fd < 0)
+ sysfatal("open %s: %r", argv[0]);
+ if(crackhdr(fd, &fhdr) <= 0)
+ sysfatal("crackhdr: %r");
+ machbytype(fhdr.type);
+ if(syminit(fd, &fhdr) <= 0)
+ sysfatal("syminit: %r");
+
+ Binit(&bin, 0, OREAD);
+ Binit(&bout, 1, OWRITE);
+ for(;;) {
+ p = Brdline(&bin, '\n');
+ if(p == nil)
+ break;
+ p[Blinelen(&bin)-1] = '\0';
+ pc = strtoull(p, 0, 16);
+ if(!findsym(pc, CTEXT, &s))
+ s.name = "??";
+ if(!fileline(file, sizeof file, pc))
+ strcpy(file, "??:0");
+ Bprint(&bout, "%s\n%s\n", s.name, file);
+ }
+ Bflush(&bout);
+ exits(0);
+}
diff --git a/src/cmd/dist/build.c b/src/cmd/dist/build.c
index 5664c18..66b5c1f 100644
--- a/src/cmd/dist/build.c
+++ b/src/cmd/dist/build.c
@@ -59,7 +59,7 @@
find(char *p, char **l, int n)
{
int i;
-
+
for(i=0; i<n; i++)
if(streq(p, l[i]))
return i;
@@ -73,7 +73,7 @@
char *p;
int i;
Buf b;
-
+
binit(&b);
xgetenv(&b, "GOROOT");
@@ -126,7 +126,7 @@
xsetenv("GOROOT", goroot);
xsetenv("GOARCH", goarch);
xsetenv("GOOS", goos);
-
+
// Make the environment more predictable.
xsetenv("LANG", "C");
xsetenv("LANGUAGE", "en_US.UTF8");
@@ -170,13 +170,13 @@
int i, nrev;
Buf b, path, bmore, branch;
Vec tags;
-
+
binit(&b);
binit(&path);
binit(&bmore);
binit(&branch);
vinit(&tags);
-
+
// The $GOROOT/VERSION file takes priority, for distributions
// without the Mercurial repo.
bpathf(&path, "%s/VERSION", goroot);
@@ -232,14 +232,14 @@
bprintf(&b, "branch.%s", bstr(&branch));
tag = btake(&b);
}
-
+
if(rev[0]) {
// Tag is before the revision we're building.
// Add extra information.
run(&bmore, goroot, CheckExit, "hg", "log", "--template", " +{node|short}", "-r", rev, nil);
chomp(&bmore);
}
-
+
bprintf(&b, "%s", tag);
if(bmore.len > 0)
bwriteb(&b, &bmore);
@@ -249,14 +249,14 @@
done:
p = btake(&b);
-
-
+
+
bfree(&b);
bfree(&path);
bfree(&bmore);
bfree(&branch);
vfree(&tags);
-
+
return p;
}
@@ -325,7 +325,7 @@
xremoveall(p);
xmkdirall(p);
}
-
+
// Create object directory.
// We keep it in pkg/ so that all the generated binaries
// are in one tree. If pkg/obj/libgc.a exists, it is a dreg from
@@ -393,7 +393,7 @@
static Vec gccargs;
// deptab lists changes to the default dependencies for a given prefix.
-// deps ending in /* read the whole directory; deps beginning with -
+// deps ending in /* read the whole directory; deps beginning with -
// exclude files with that prefix.
static struct {
char *prefix; // prefix of target
@@ -559,7 +559,7 @@
vinit(&clean);
vinit(&lib);
vinit(&extra);
-
+
// path = full path to dir.
bpathf(&path, "%s/src/%s", goroot, dir);
name = lastelem(dir);
@@ -599,7 +599,7 @@
exe = "";
if(streq(gohostos, "windows"))
exe = ".exe";
-
+
// Start final link command line.
// Note: code below knows that link.p[targ] is the target.
if(islib) {
@@ -696,13 +696,13 @@
}
files.len = n;
continue;
- }
+ }
vadd(&files, p);
}
}
}
vuniq(&files);
-
+
// Convert to absolute paths.
for(i=0; i<files.len; i++) {
if(!isabs(files.p[i])) {
@@ -740,11 +740,11 @@
files.p[n++] = files.p[i];
}
files.len = n;
-
+
for(i=0; i<lib.len && !stale; i++)
if(mtime(lib.p[i]) > ttarg)
stale = 1;
-
+
if(!stale)
goto out;
@@ -792,9 +792,9 @@
copy(bpathf(&b, "%s/zasm_GOOS_GOARCH.h", workdir),
bpathf(&b1, "%s/zasm_%s_%s.h", bstr(&path), goos, goarch), 0);
}
-
+
// Generate .c files from .goc files.
- if(streq(dir, "pkg/runtime")) {
+ if(streq(dir, "pkg/runtime")) {
for(i=0; i<files.len; i++) {
p = files.p[i];
if(!hassuffix(p, ".goc"))
@@ -808,7 +808,7 @@
}
vuniq(&files);
}
-
+
if((!streq(goos, gohostos) || !streq(goarch, gohostarch)) && isgo) {
// We've generated the right files; the go command can do the build.
if(vflag > 1)
@@ -833,13 +833,13 @@
vadd(&compile, "-m32");
if(streq(dir, "lib9"))
vadd(&compile, "-DPLAN9PORT");
-
+
vadd(&compile, "-I");
vadd(&compile, bpathf(&b, "%s/include", goroot));
-
+
vadd(&compile, "-I");
vadd(&compile, bstr(&path));
-
+
// lib9/goos.c gets the default constants hard-coded.
if(streq(name, "goos.c")) {
vadd(&compile, bprintf(&b, "-DGOOS=\"%s\"", goos));
@@ -849,7 +849,7 @@
vadd(&compile, bprintf(&b, "-DGOROOT=\"%s\"", bstr(&b1)));
vadd(&compile, bprintf(&b, "-DGOVERSION=\"%s\"", goversion));
}
-
+
// gc/lex.c records the GOEXPERIMENT setting used during the build.
if(streq(name, "lex.c")) {
xgetenv(&b, "GOEXPERIMENT");
@@ -867,7 +867,7 @@
vadd(&compile, workdir);
vadd(&compile, bprintf(&b, "-DGOOS_%s", goos));
vadd(&compile, bprintf(&b, "-DGOARCH_%s", goarch));
- }
+ }
bpathf(&b, "%s/%s", workdir, lastelem(files.p[i]));
doclean = 1;
@@ -893,7 +893,7 @@
vadd(&clean, bstr(&b));
}
bgwait();
-
+
if(isgo) {
// The last loop was compiling individual files.
// Hand the Go files to the compiler en masse.
@@ -905,16 +905,16 @@
vadd(&compile, bstr(&b));
vadd(&clean, bstr(&b));
vadd(&link, bstr(&b));
-
+
vadd(&compile, "-p");
if(hasprefix(dir, "pkg/"))
vadd(&compile, dir+4);
else
vadd(&compile, "main");
-
+
if(streq(dir, "pkg/runtime"))
vadd(&compile, "-+");
-
+
vcopy(&compile, go.p, go.len);
runv(nil, bstr(&path), CheckExit, &compile);
@@ -980,7 +980,7 @@
int i, j, ret;
Buf b;
Vec lines, fields;
-
+
// Check file name for GOOS or GOARCH.
name = lastelem(file);
for(i=0; i<nelem(okgoos); i++)
@@ -989,11 +989,11 @@
for(i=0; i<nelem(okgoarch); i++)
if(contains(name, okgoarch[i]) && !streq(okgoarch[i], goarch))
return 0;
-
+
// Omit test files.
if(contains(name, "_test"))
return 0;
-
+
// cmd/go/doc.go has a giant /* */ comment before
// it gets to the important detail that it is not part of
// package main. We don't parse those comments,
@@ -1046,7 +1046,7 @@
bfree(&b);
vfree(&lines);
vfree(&fields);
-
+
return ret;
}
@@ -1055,7 +1055,7 @@
copy(char *dst, char *src, int exec)
{
Buf b;
-
+
if(vflag > 1)
xprintf("cp %s %s\n", src, dst);
@@ -1070,11 +1070,13 @@
"lib9",
"libbio",
"libmach",
-
+
"misc/pprof",
+ "cmd/addr2line",
"cmd/cov",
"cmd/nm",
+ "cmd/objdump",
"cmd/pack",
"cmd/prof",
@@ -1122,12 +1124,12 @@
"pkg/go/scanner",
"pkg/go/ast",
"pkg/go/parser",
- "pkg/go/build",
"pkg/os/exec",
"pkg/net/url",
"pkg/text/template/parse",
"pkg/text/template",
"pkg/go/doc",
+ "pkg/go/build",
"cmd/go",
};
@@ -1147,11 +1149,13 @@
"cmd/8c",
"cmd/8g",
"cmd/8l",
+ "cmd/addr2line",
"cmd/cc",
"cmd/cov",
"cmd/gc",
"cmd/go",
"cmd/nm",
+ "cmd/objdump",
"cmd/pack",
"cmd/prof",
"lib9",
@@ -1204,11 +1208,11 @@
int i, j, k;
Buf b, path;
Vec dir;
-
+
binit(&b);
binit(&path);
vinit(&dir);
-
+
for(i=0; i<nelem(cleantab); i++) {
bpathf(&path, "%s/src/%s", goroot, cleantab[i]);
xreaddir(&dir, bstr(&path));
@@ -1227,7 +1231,7 @@
if(rebuildall) {
// Remove object tree.
xremoveall(bpathf(&b, "%s/pkg/obj/%s_%s", goroot, gohostos, gohostarch));
-
+
// Remove installed packages and tools.
xremoveall(bpathf(&b, "%s/pkg/%s_%s", goroot, gohostos, gohostarch));
xremoveall(bpathf(&b, "%s/pkg/%s_%s", goroot, goos, goarch));
@@ -1294,7 +1298,7 @@
if(argc > 0)
usage();
-
+
xprintf(format, "GOROOT", goroot);
xprintf(format, "GOBIN", gobin);
xprintf(format, "GOARCH", goarch);
@@ -1346,7 +1350,7 @@
clean();
goversion = findgoversion();
setup();
-
+
// For the main bootstrap, building for host os/arch.
oldgoos = goos;
oldgoarch = goarch;
@@ -1356,7 +1360,7 @@
gochar = gohostchar;
xsetenv("GOARCH", goarch);
xsetenv("GOOS", goos);
-
+
for(i=0; i<nelem(buildorder); i++) {
install(bprintf(&b, buildorder[i], gohostchar));
if(!streq(oldgochar, gohostchar) && xstrstr(buildorder[i], "%s"))
@@ -1381,7 +1385,7 @@
{
char *p;
Buf pwd, src, real_src;
-
+
binit(&pwd);
binit(&src);
binit(&real_src);
@@ -1404,7 +1408,7 @@
bfree(&pwd);
bfree(&src);
bfree(&real_src);
-
+
return p;
}
@@ -1421,7 +1425,7 @@
default:
usage();
}ARGEND
-
+
if(argc == 0)
install(defaulttarg());
@@ -1469,7 +1473,7 @@
binit(&b);
binit(&b1);
binit(&search);
-
+
xprintf("\n");
xprintf("---\n");
xprintf("Installed Go for %s/%s in %s\n", goos, goarch, goroot);
@@ -1490,7 +1494,7 @@
"On OS X the debuggers must be installed setgrp procmod.\n"
"Read and run ./sudo.bash to install the debuggers.\n");
}
-
+
if(!streq(goroot_final, goroot)) {
xprintf("\n"
"The binaries expect %s to be copied or moved to %s\n",
diff --git a/src/cmd/objdump/main.c b/src/cmd/objdump/main.c
new file mode 100644
index 0000000..b684be7
--- /dev/null
+++ b/src/cmd/objdump/main.c
@@ -0,0 +1,68 @@
+// Copyright 2012 The Go Authors. All rights reserved.
+// Use of this source code is governed by a BSD-style
+// license that can be found in the LICENSE file.
+
+/*
+ * objdump simulation - only enough to make pprof work on Macs
+ */
+
+#include <u.h>
+#include <libc.h>
+#include <bio.h>
+#include <mach.h>
+
+void
+usage(void)
+{
+ fprint(2, "usage: objdump binary start stop\n");
+ fprint(2, "Disassembles binary from PC start up to stop.\n");
+ exits("usage");
+}
+
+void
+main(int argc, char **argv)
+{
+ int fd, n;
+ uvlong pc, start, stop;
+ Fhdr fhdr;
+ Biobuf bout;
+ char buf[1024];
+ Map *text;
+
+ ARGBEGIN{
+ default:
+ usage();
+ }ARGEND
+
+ if(argc != 3)
+ usage();
+ start = strtoull(argv[1], 0, 16);
+ stop = strtoull(argv[2], 0, 16);
+
+ fd = open(argv[0], OREAD);
+ if(fd < 0)
+ sysfatal("open %s: %r", argv[0]);
+ if(crackhdr(fd, &fhdr) <= 0)
+ sysfatal("crackhdr: %r");
+ machbytype(fhdr.type);
+ if(syminit(fd, &fhdr) <= 0)
+ sysfatal("syminit: %r");
+ text = loadmap(nil, fd, &fhdr);
+ if(text == nil)
+ sysfatal("loadmap: %r");
+
+ Binit(&bout, 1, OWRITE);
+ for(pc=start; pc<stop; ) {
+ if(fileline(buf, sizeof buf, pc))
+ Bprint(&bout, "%s\n", buf);
+ buf[0] = '\0';
+ machdata->das(text, pc, 0, buf, sizeof buf);
+ Bprint(&bout, " %llx: %s\n", pc, buf);
+ n = machdata->instsize(text, pc);
+ if(n <= 0)
+ break;
+ pc += n;
+ }
+ Bflush(&bout);
+ exits(0);
+}
diff --git a/src/libmach/8db.c b/src/libmach/8db.c
index 3101e13..ce1b4dd 100644
--- a/src/libmach/8db.c
+++ b/src/libmach/8db.c
@@ -1,11 +1,11 @@
// Inferno libmach/8db.c
// http://code.google.com/p/inferno-os/source/browse/utils/libmach/8db.c
//
-// Copyright © 1994-1999 Lucent Technologies Inc.
-// Power PC support Copyright © 1995-2004 C H Forsyth (forsyth@terzarima.net).
-// Portions Copyright © 1997-1999 Vita Nuova Limited.
-// Portions Copyright © 2000-2007 Vita Nuova Holdings Limited (www.vitanuova.com).
-// Revisions Copyright © 2000-2004 Lucent Technologies Inc. and others.
+// Copyright © 1994-1999 Lucent Technologies Inc.
+// Power PC support Copyright © 1995-2004 C H Forsyth (forsyth@terzarima.net).
+// Portions Copyright © 1997-1999 Vita Nuova Limited.
+// Portions Copyright © 2000-2007 Vita Nuova Holdings Limited (www.vitanuova.com).
+// Revisions Copyright © 2000-2004 Lucent Technologies Inc. and others.
// Portions Copyright © 2009 The Go Authors. All rights reserved.
//
// Permission is hereby granted, free of charge, to any person obtaining a copy
@@ -2088,18 +2088,23 @@
static void
pea(Instr *ip)
{
+ int base;
+
+ base = ip->base;
+ if(base >= 0 && (ip->rex & REXB))
+ base += 8;
+
if (ip->mod == 3) {
if (ip->osize == 'B')
bprint(ip, (ip->rex & REXB? breg64: breg)[(uchar)ip->base]);
- else if(ip->rex & REXB)
- bprint(ip, "%s%s", ANAME(ip), reg[ip->base+8]);
else
- bprint(ip, "%s%s", ANAME(ip), reg[(uchar)ip->base]);
+ bprint(ip, "%s%s", ANAME(ip), reg[base]);
return;
}
+
if (ip->segment)
bprint(ip, ip->segment);
- if (ip->asize == 'E' && ip->base == SP)
+ if (ip->asize == 'E' && base == SP)
plocal(ip);
else {
if (ip->base < 0)
diff --git a/src/pkg/runtime/lock_futex.c b/src/pkg/runtime/lock_futex.c
index 6ec4aee..b4465bf 100644
--- a/src/pkg/runtime/lock_futex.c
+++ b/src/pkg/runtime/lock_futex.c
@@ -118,8 +118,12 @@
void
runtime·notesleep(Note *n)
{
+ if(m->profilehz > 0)
+ runtime·setprof(false);
while(runtime·atomicload(&n->key) == 0)
runtime·futexsleep(&n->key, 0, -1);
+ if(m->profilehz > 0)
+ runtime·setprof(true);
}
void
@@ -135,14 +139,18 @@
if(runtime·atomicload(&n->key) != 0)
return;
+ if(m->profilehz > 0)
+ runtime·setprof(false);
deadline = runtime·nanotime() + ns;
for(;;) {
runtime·futexsleep(&n->key, 0, ns);
if(runtime·atomicload(&n->key) != 0)
- return;
+ break;
now = runtime·nanotime();
if(now >= deadline)
- return;
+ break;
ns = deadline - now;
}
+ if(m->profilehz > 0)
+ runtime·setprof(true);
}
diff --git a/src/pkg/runtime/lock_sema.c b/src/pkg/runtime/lock_sema.c
index 28d2c32..1d9c37f 100644
--- a/src/pkg/runtime/lock_sema.c
+++ b/src/pkg/runtime/lock_sema.c
@@ -154,7 +154,11 @@
return;
}
// Queued. Sleep.
+ if(m->profilehz > 0)
+ runtime·setprof(false);
runtime·semasleep(-1);
+ if(m->profilehz > 0)
+ runtime·setprof(true);
}
void
@@ -178,12 +182,16 @@
return;
}
+ if(m->profilehz > 0)
+ runtime·setprof(false);
deadline = runtime·nanotime() + ns;
for(;;) {
// Registered. Sleep.
if(runtime·semasleep(ns) >= 0) {
// Acquired semaphore, semawakeup unregistered us.
// Done.
+ if(m->profilehz > 0)
+ runtime·setprof(true);
return;
}
@@ -196,6 +204,9 @@
ns = deadline - now;
}
+ if(m->profilehz > 0)
+ runtime·setprof(true);
+
// Deadline arrived. Still registered. Semaphore not acquired.
// Want to give up and return, but have to unregister first,
// so that any notewakeup racing with the return does not
diff --git a/src/pkg/runtime/os_darwin.h b/src/pkg/runtime/os_darwin.h
index 071a547..eb5d2da 100644
--- a/src/pkg/runtime/os_darwin.h
+++ b/src/pkg/runtime/os_darwin.h
@@ -38,4 +38,7 @@
#define NSIG 32
#define SI_USER 0 /* empirically true, but not what headers say */
+#define SIG_BLOCK 1
+#define SIG_UNBLOCK 2
#define SIG_SETMASK 3
+
diff --git a/src/pkg/runtime/os_freebsd.h b/src/pkg/runtime/os_freebsd.h
index da1d8de..5e8de54 100644
--- a/src/pkg/runtime/os_freebsd.h
+++ b/src/pkg/runtime/os_freebsd.h
@@ -9,7 +9,6 @@
void runtime·sigaction(int32, struct sigaction*, struct sigaction*);
void runtime·sigprocmask(Sigset *, Sigset *);
void runtime·setsig(int32, void(*)(int32, Siginfo*, void*, G*), bool);
-void runtiem·setitimerval(int32, Itimerval*, Itimerval*);
void runtime·setitimer(int32, Itimerval*, Itimerval*);
int32 runtime·sysctl(uint32*, uint32, byte*, uintptr*, byte*, uintptr);
diff --git a/src/pkg/runtime/os_netbsd.h b/src/pkg/runtime/os_netbsd.h
index 67c58ec..4ecf78d 100644
--- a/src/pkg/runtime/os_netbsd.h
+++ b/src/pkg/runtime/os_netbsd.h
@@ -12,7 +12,6 @@
void runtime·sigaction(int32, struct sigaction*, struct sigaction*);
void runtime·setsig(int32, void(*)(int32, Siginfo*, void*, G*), bool);
void runtime·sighandler(int32 sig, Siginfo *info, void *context, G *gp);
-void runtime·setitimerval(int32, Itimerval*, Itimerval*);
void runtime·setitimer(int32, Itimerval*, Itimerval*);
int32 runtime·sysctl(uint32*, uint32, byte*, uintptr*, byte*, uintptr);
diff --git a/src/pkg/runtime/os_openbsd.h b/src/pkg/runtime/os_openbsd.h
index 67c58ec..4ecf78d 100644
--- a/src/pkg/runtime/os_openbsd.h
+++ b/src/pkg/runtime/os_openbsd.h
@@ -12,7 +12,6 @@
void runtime·sigaction(int32, struct sigaction*, struct sigaction*);
void runtime·setsig(int32, void(*)(int32, Siginfo*, void*, G*), bool);
void runtime·sighandler(int32 sig, Siginfo *info, void *context, G *gp);
-void runtime·setitimerval(int32, Itimerval*, Itimerval*);
void runtime·setitimer(int32, Itimerval*, Itimerval*);
int32 runtime·sysctl(uint32*, uint32, byte*, uintptr*, byte*, uintptr);
diff --git a/src/pkg/runtime/pprof/pprof.go b/src/pkg/runtime/pprof/pprof.go
index 099bb6a..f67e8a8 100644
--- a/src/pkg/runtime/pprof/pprof.go
+++ b/src/pkg/runtime/pprof/pprof.go
@@ -20,8 +20,8 @@
"text/tabwriter"
)
-// BUG(rsc): CPU profiling is broken on OS X, due to an Apple kernel bug.
-// For details, see http://code.google.com/p/go/source/detail?r=35b716c94225.
+// BUG(rsc): A bug in the OS X Snow Leopard 64-bit kernel prevents
+// CPU profiling from giving accurate results on that system.
// A Profile is a collection of stack traces showing the call sequences
// that led to instances of a particular event, such as allocation.
@@ -156,7 +156,7 @@
}
// Add adds the current execution stack to the profile, associated with value.
-// Add stores value in an internal map, so value must be suitable for use as
+// Add stores value in an internal map, so value must be suitable for use as
// a map key and will not be garbage collected until the corresponding
// call to Remove. Add panics if the profile already contains a stack for value.
//
diff --git a/src/pkg/runtime/pprof/pprof_test.go b/src/pkg/runtime/pprof/pprof_test.go
index 5f128c0..994ec9d 100644
--- a/src/pkg/runtime/pprof/pprof_test.go
+++ b/src/pkg/runtime/pprof/pprof_test.go
@@ -7,6 +7,7 @@
import (
"bytes"
"hash/crc32"
+ "os/exec"
"runtime"
. "runtime/pprof"
"strings"
@@ -17,8 +18,15 @@
func TestCPUProfile(t *testing.T) {
switch runtime.GOOS {
case "darwin":
- // see Apple Bug Report #9177434 (copied into change description)
- return
+ out, err := exec.Command("uname", "-a").CombinedOutput()
+ if err != nil {
+ t.Fatal(err)
+ }
+ vers := string(out)
+ t.Logf("uname -a: %v", vers)
+ if strings.Contains(vers, "Darwin Kernel Version 10.8.0") && strings.Contains(vers, "root:xnu-1504.15.3~1/RELEASE_X86_64") {
+ t.Logf("skipping test on known-broken kernel (64-bit Snow Leopard)")
+ }
case "plan9":
// unimplemented
return
diff --git a/src/pkg/runtime/proc.c b/src/pkg/runtime/proc.c
index d94bec8..ddac048 100644
--- a/src/pkg/runtime/proc.c
+++ b/src/pkg/runtime/proc.c
@@ -338,7 +338,7 @@
m->mcache = runtime·allocmcache();
runtime·callers(1, m->createstack, nelem(m->createstack));
-
+
// Add to runtime·allm so garbage collector doesn't free m
// when it is just in a register or thread-local storage.
m->alllink = runtime·allm;
@@ -728,7 +728,6 @@
// so other calls can reuse this stack space.
runtime·gosave(&m->g0->sched);
m->g0->sched.pc = (void*)-1; // make sure it is never used
-
runtime·asminit();
runtime·minit();
schedule(nil);
@@ -916,6 +915,9 @@
{
uint32 v;
+ if(m->profilehz > 0)
+ runtime·setprof(false);
+
// Leave SP around for gc and traceback.
runtime·gosave(&g->sched);
g->gcsp = g->sched.sp;
@@ -979,6 +981,9 @@
// Garbage collector isn't running (since we are),
// so okay to clear gcstack.
g->gcstack = nil;
+
+ if(m->profilehz > 0)
+ runtime·setprof(true);
return;
}
diff --git a/src/pkg/runtime/runtime.h b/src/pkg/runtime/runtime.h
index 3b0f505..f2669fd 100644
--- a/src/pkg/runtime/runtime.h
+++ b/src/pkg/runtime/runtime.h
@@ -730,3 +730,13 @@
void runtime·ifaceE2I(struct InterfaceType*, Eface, Iface*);
uintptr runtime·memlimit(void);
+
+// If appropriate, ask the operating system to control whether this
+// thread should receive profiling signals. This is only necessary on OS X.
+// An operating system should not deliver a profiling signal to a
+// thread that is not actually executing (what good is that?), but that's
+// what OS X prefers to do. When profiling is turned on, we mask
+// away the profiling signal when threads go to sleep, so that OS X
+// is forced to deliver the signal to a thread that's actually running.
+// This is a no-op on other systems.
+void runtime·setprof(bool);
diff --git a/src/pkg/runtime/signal_darwin_386.c b/src/pkg/runtime/signal_darwin_386.c
index 1844f68..9e98635 100644
--- a/src/pkg/runtime/signal_darwin_386.c
+++ b/src/pkg/runtime/signal_darwin_386.c
@@ -40,7 +40,8 @@
r = &mc->ss;
if(sig == SIGPROF) {
- runtime·sigprof((uint8*)r->eip, (uint8*)r->esp, nil, gp);
+ if(gp != m->g0 && gp != m->gsignal)
+ runtime·sigprof((uint8*)r->eip, (uint8*)r->esp, nil, gp);
return;
}
@@ -58,7 +59,7 @@
if(pc[0] == 0xF6 || pc[0] == 0xF7)
info->si_code = FPE_INTDIV;
}
-
+
// Make it look like a call to the signal func.
// Have to pass arguments out of band since
// augmenting the stack frame would break
diff --git a/src/pkg/runtime/signal_darwin_amd64.c b/src/pkg/runtime/signal_darwin_amd64.c
index 32c7308..d9c5f48 100644
--- a/src/pkg/runtime/signal_darwin_amd64.c
+++ b/src/pkg/runtime/signal_darwin_amd64.c
@@ -48,7 +48,8 @@
r = &mc->ss;
if(sig == SIGPROF) {
- runtime·sigprof((uint8*)r->rip, (uint8*)r->rsp, nil, gp);
+ if(gp != m->g0 && gp != m->gsignal)
+ runtime·sigprof((uint8*)r->rip, (uint8*)r->rsp, nil, gp);
return;
}
@@ -68,7 +69,7 @@
if(pc[0] == 0xF6 || pc[0] == 0xF7)
info->si_code = FPE_INTDIV;
}
-
+
// Make it look like a call to the signal func.
// Have to pass arguments out of band since
// augmenting the stack frame would break
@@ -77,7 +78,7 @@
gp->sigcode0 = info->si_code;
gp->sigcode1 = (uintptr)info->si_addr;
gp->sigpc = r->rip;
-
+
// Only push runtime·sigpanic if r->rip != 0.
// If r->rip == 0, probably panicked because of a
// call to a nil func. Not pushing that onto sp will
@@ -92,7 +93,7 @@
r->rip = (uintptr)runtime·sigpanic;
return;
}
-
+
if(info->si_code == SI_USER || (t->flags & SigNotify))
if(runtime·sigsend(sig))
return;
diff --git a/src/pkg/runtime/signal_unix.c b/src/pkg/runtime/signal_unix.c
index 1370841..0b9d2a5 100644
--- a/src/pkg/runtime/signal_unix.c
+++ b/src/pkg/runtime/signal_unix.c
@@ -55,17 +55,17 @@
runtime·resetcpuprofiler(int32 hz)
{
Itimerval it;
-
+
runtime·memclr((byte*)&it, sizeof it);
if(hz == 0) {
runtime·setitimer(ITIMER_PROF, &it, nil);
- runtime·setsig(SIGPROF, SIG_IGN, true);
+ runtime·setprof(false);
} else {
- runtime·setsig(SIGPROF, runtime·sighandler, true);
it.it_interval.tv_sec = 0;
it.it_interval.tv_usec = 1000000 / hz;
it.it_value = it.it_interval;
runtime·setitimer(ITIMER_PROF, &it, nil);
+ runtime·setprof(true);
}
m->profilehz = hz;
}
diff --git a/src/pkg/runtime/sys_darwin_386.s b/src/pkg/runtime/sys_darwin_386.s
index e235a84..c2dab89 100644
--- a/src/pkg/runtime/sys_darwin_386.s
+++ b/src/pkg/runtime/sys_darwin_386.s
@@ -100,14 +100,14 @@
IMULL $1000, BX
ADDL BX, AX
ADCL $0, DX
-
+
MOVL ret+0(FP), DI
MOVL AX, 0(DI)
MOVL DX, 4(DI)
RET
TEXT runtime·sigprocmask(SB),7,$0
- MOVL $48, AX
+ MOVL $329, AX // pthread_sigmask (on OS X, sigprocmask==entire process)
INT $0x80
JAE 2(PC)
CALL runtime·notok(SB)
diff --git a/src/pkg/runtime/sys_darwin_amd64.s b/src/pkg/runtime/sys_darwin_amd64.s
index 13882c8..4b215d0 100644
--- a/src/pkg/runtime/sys_darwin_amd64.s
+++ b/src/pkg/runtime/sys_darwin_amd64.s
@@ -96,7 +96,7 @@
MOVL 8(SP), DI
MOVQ 16(SP), SI
MOVQ 24(SP), DX
- MOVL $(0x2000000+48), AX // syscall entry
+ MOVL $(0x2000000+329), AX // pthread_sigmask (on OS X, sigprocmask==entire process)
SYSCALL
JCC 2(PC)
CALL runtime·notok(SB)
diff --git a/src/pkg/runtime/thread_darwin.c b/src/pkg/runtime/thread_darwin.c
index d170dfb..556fb67 100644
--- a/src/pkg/runtime/thread_darwin.c
+++ b/src/pkg/runtime/thread_darwin.c
@@ -11,6 +11,7 @@
static Sigset sigset_all = ~(Sigset)0;
static Sigset sigset_none;
+static Sigset sigset_prof = 1<<(SIGPROF-1);
static void
unimplemented(int8 *name)
@@ -23,7 +24,14 @@
int32
runtime·semasleep(int64 ns)
{
- return runtime·mach_semacquire(m->waitsema, ns);
+ int32 v;
+
+ if(m->profilehz > 0)
+ runtime·setprof(false);
+ v = runtime·mach_semacquire(m->waitsema, ns);
+ if(m->profilehz > 0)
+ runtime·setprof(true);
+ return v;
}
void
@@ -84,7 +92,7 @@
runtime·sigprocmask(SIG_SETMASK, &sigset_all, &oset);
errno = runtime·bsdthread_create(stk, m, g, fn);
runtime·sigprocmask(SIG_SETMASK, &oset, nil);
-
+
if(errno < 0) {
runtime·printf("runtime: failed to create new OS thread (have %d already; errno=%d)\n", runtime·mcount(), -errno);
runtime·throw("runtime.newosproc");
@@ -98,7 +106,11 @@
// Initialize signal handling.
m->gsignal = runtime·malg(32*1024); // OS X wants >=8K, Linux >=2K
runtime·signalstack(m->gsignal->stackguard - StackGuard, 32*1024);
- runtime·sigprocmask(SIG_SETMASK, &sigset_none, nil);
+
+ if(m->profilehz > 0)
+ runtime·sigprocmask(SIG_SETMASK, &sigset_none, nil);
+ else
+ runtime·sigprocmask(SIG_SETMASK, &sigset_prof, nil);
}
// Mach IPC, to get at semaphores
@@ -434,3 +446,34 @@
// the limit.
return 0;
}
+
+// NOTE(rsc): On OS X, when the CPU profiling timer expires, the SIGPROF
+// signal is not guaranteed to be sent to the thread that was executing to
+// cause it to expire. It can and often does go to a sleeping thread, which is
+// not interesting for our profile. This is filed Apple Bug Report #9177434,
+// copied to http://code.google.com/p/go/source/detail?r=35b716c94225.
+// To work around this bug, we disable receipt of the profiling signal on
+// a thread while in blocking system calls. This forces the kernel to deliver
+// the profiling signal to an executing thread.
+//
+// The workaround fails on OS X machines using a 64-bit Snow Leopard kernel.
+// In that configuration, the kernel appears to want to deliver SIGPROF to the
+// sleeping threads regardless of signal mask and, worse, does not deliver
+// the signal until the thread wakes up on its own.
+//
+// If necessary, we can switch to using ITIMER_REAL for OS X and handle
+// the kernel-generated SIGALRM by generating our own SIGALRMs to deliver
+// to all the running threads. SIGALRM does not appear to be affected by
+// the 64-bit Snow Leopard bug. However, as of this writing Mountain Lion
+// is in preview, making Snow Leopard two versions old, so it is unclear how
+// much effort we need to spend on one buggy kernel.
+
+// Control whether profiling signal can be delivered to this thread.
+void
+runtime·setprof(bool on)
+{
+ if(on)
+ runtime·sigprocmask(SIG_UNBLOCK, &sigset_prof, nil);
+ else
+ runtime·sigprocmask(SIG_BLOCK, &sigset_prof, nil);
+}
diff --git a/src/pkg/runtime/thread_freebsd.c b/src/pkg/runtime/thread_freebsd.c
index 7871827..77e8bb3 100644
--- a/src/pkg/runtime/thread_freebsd.c
+++ b/src/pkg/runtime/thread_freebsd.c
@@ -189,3 +189,9 @@
return rl.rlim_cur - used;
}
+
+void
+runtime·setprof(bool on)
+{
+ USED(on);
+}
diff --git a/src/pkg/runtime/thread_linux.c b/src/pkg/runtime/thread_linux.c
index d406a71..6b42844 100644
--- a/src/pkg/runtime/thread_linux.c
+++ b/src/pkg/runtime/thread_linux.c
@@ -228,7 +228,7 @@
Rlimit rl;
extern byte text[], end[];
uintptr used;
-
+
if(runtime·getrlimit(RLIMIT_AS, &rl) != 0)
return 0;
if(rl.rlim_cur >= 0x7fffffff)
@@ -249,3 +249,9 @@
return rl.rlim_cur - used;
}
+
+void
+runtime·setprof(bool on)
+{
+ USED(on);
+}
diff --git a/src/pkg/runtime/thread_netbsd.c b/src/pkg/runtime/thread_netbsd.c
index 7d14e5c..62e133c 100644
--- a/src/pkg/runtime/thread_netbsd.c
+++ b/src/pkg/runtime/thread_netbsd.c
@@ -207,3 +207,9 @@
{
return 0;
}
+
+void
+runtime·setprof(bool on)
+{
+ USED(on);
+}
diff --git a/src/pkg/runtime/thread_openbsd.c b/src/pkg/runtime/thread_openbsd.c
index 704d95a..bee0c57 100644
--- a/src/pkg/runtime/thread_openbsd.c
+++ b/src/pkg/runtime/thread_openbsd.c
@@ -207,3 +207,9 @@
{
return 0;
}
+
+void
+runtime·setprof(bool on)
+{
+ USED(on);
+}
diff --git a/src/pkg/runtime/thread_plan9.c b/src/pkg/runtime/thread_plan9.c
index 7d5c38f..aaed505 100644
--- a/src/pkg/runtime/thread_plan9.c
+++ b/src/pkg/runtime/thread_plan9.c
@@ -87,7 +87,7 @@
// The naïve implementation (without the cached
// file descriptor) is roughly four times slower
// in 9vx on a 2.16 GHz Intel Core 2 Duo.
-
+
if(fd < 0 && (fd = runtime·open((byte*)"/dev/bintime", OREAD|OCEXEC)) < 0)
return 0;
if(runtime·pread(fd, b, sizeof b, 0) != sizeof b)
@@ -241,3 +241,9 @@
{
return 0;
}
+
+void
+runtime·setprof(bool on)
+{
+ USED(on);
+}
diff --git a/src/pkg/runtime/thread_windows.c b/src/pkg/runtime/thread_windows.c
index 8feac97..8a448bc 100644
--- a/src/pkg/runtime/thread_windows.c
+++ b/src/pkg/runtime/thread_windows.c
@@ -226,7 +226,7 @@
time·now(int64 sec, int32 usec)
{
int64 ns;
-
+
ns = runtime·nanotime();
sec = ns / 1000000000LL;
usec = ns - sec * 1000000000LL;
@@ -431,3 +431,9 @@
{
return 0;
}
+
+void
+runtime·setprof(bool on)
+{
+ USED(on);
+}
