runtime/pprof: reject cpu profiling with -race

[alexbrainman]

c:\>go test -race -v runtime/pprof
=== RUN TestCPUProfile
--- FAIL: TestCPUProfile (8.26 seconds)
        pprof_test.go:56: profile too short: [0x0 0x3 0x0 0x2710 0x0 0x0 0x1 0x0]
FAIL
exit status 1
FAIL    runtime/pprof   8.328s

[dvyukov]

Comment 1:

Labels changed: added racedetector.

[alberts]

Comment 2:

I'm seeing this at tip:
=== RUN TestCPUProfile
runtime: unknown argument frame size for runtime.asmcgocall
fatal error: invalid stack
goroutine 0 [idle]:
runtime.throw(0x69e27c)
    /build/go.tip/go/src/pkg/runtime/panic.c:475 +0x67
runtime.gentraceback(0x423552, 0x7fab03ffee20, 0x0, 0xc2100017e0, 0x0, ...)
    /build/go.tip/go/src/pkg/runtime/traceback_x86.c:104 +0xa7b
runtime.sigprof(0x423552, 0x7fab03ffee20, 0x0, 0xc2100017e0)
    /build/go.tip/go/src/pkg/runtime/proc.c:1818 +0xe3
runtime.sighandler(0x1b, 0xc210050bf0, 0xc210050ac0, 0xc2100017e0)
    /build/go.tip/go/src/pkg/runtime/signal_amd64.c:51 +0x81
runtime.sigtramp()
    /build/go.tip/go/src/pkg/runtime/sys_linux_amd64.s:207 +0x54
runtime: unknown argument frame size for runtime.sigreturn
panic during panic
FAIL    runtime/pprof   0.653s

[davecheney]

Comment 3:

Related, issue #5659.

[gopherbot]

Comment 4 by ray@cloudflare.com:

Hello,
We're also seeing this on some of our server code. It seems to only occur under load and
when we attempt a cpuprofile. Perhaps another data point will be helpful.
 runtime: unknown argument frame size for runtime.chansend1
 fatal error: invalid stack
 goroutine 0 [idle]:
 runtime.throw(0xc23bfc)
    /var/lib/jenkins/tools/go-latest/src/pkg/runtime/panic.c:475 +0x67
 runtime.gentraceback(0x42bad9, 0x7fa51757e9c0, 0x0, 0xc21959fd80, 0x0, ...)
    /var/lib/jenkins/tools/go-latest/src/pkg/runtime/traceback_x86.c:104 +0xa7b
 runtime.sigprof(0x42bad9, 0x7fa51757e9c0, 0x0, 0xc21959fd80)
    /var/lib/jenkins/tools/go-latest/src/pkg/runtime/proc.c:1833 +0xe3
 runtime.sighandler(0x1b, 0xc2147edd70, 0xc2147edc40, 0xc21959fd80)
    /var/lib/jenkins/tools/go-latest/src/pkg/runtime/signal_amd64.c:51 +0x81
 runtime.sigtramp()
    /var/lib/jenkins/tools/go-latest/src/pkg/runtime/sys_linux_amd64.s:207 +0x54
 runtime: unknown argument frame size for runtime.sigreturn
 panic during panic

[davecheney]

Comment 5:

@ray - thanks for the data point. Is this a released version of Go, or trunk ? which
revno ?

[gopherbot]

Comment 6 by ray@cloudflare.com:

Oops, yes, we are tracking tip (but that isn't too useful). This is the version string
we built with:
devel +2b6568891862 Mon Jun 17 15:41:17 2013 +0400

[davecheney]

Comment 7:

@ray - that is ok, I didn't think this issue was present in the 1.1 release branch.
Thanks for confirming.

[minux]

Comment 8:

Re #4, it should be issue #5736.

[rsc]

Comment 9:

Let's just make profiling refuse to run when race mode is on.

[rsc]

Comment 10:

Labels changed: added priority-later, go1.2, removed priority-triage.

[dvyukov]

Comment 11:

I would close this.
There are no principal reasons why this can not work (even if not very useful, while I
actually used profiling to understand race-induced overheads). The bugs revealed are not
race-specific, the can happen with cgo or with goroutine preemption.

[alexbrainman]

Comment 12:

Dmitriy,
Feel free to close. I don't have particular interest in this. I just reported a broken
test that I saw.
Alex

[rsc]

Comment 13:

Dmitriy, I don't understand what you mean by "close this".
People are seeing failures in runtime/pprof under race detection.
We cannot just close the issue without addressing that.
Either (1) we fix pprof or race detector, whichever is the problem, or (2) we disable
pprof with race detection.

[dvyukov]

Comment 14:

Several issues reported in this thread. Most of them are not related to race detector.
We need to file remaining issues with pprof/runtime interaction in separate issues (if
any). But this issue is misleading.

[dvyukov]

Comment 15:

This issue was closed by revision e33e476.

Status changed to Fixed.