794206 - Enabling the profiler hits JS_ASSERT(installed());

Status: RESOLVED FIXED

(Core :: Gecko Profiler, defect)

Description

[Dave Hylands [:dhylands]]

I tried building from Mozilla-Inbound and when I run b2g with MOZ_PROFILER_STARTUP=1 then I hit the assert at:

js/src/vm/SPSProfiler.cpp:58

Backtrace:

#0  js::SPSProfiler::enable (this=0xa5a5b025, enabled=true) at /home/work/mozilla-inbound/js/src/vm/SPSProfiler.cpp:58
#1  0x418715b0 in js::EnableRuntimeProfilingStack (rt=<value optimized out>, enabled=true) at /home/work/mozilla-inbound/js/src/jsfriendapi.cpp:900
#2  0x4133ae28 in ProfileStack::enableJSSampling (aProfileEntries=<value optimized out>, aInterval=<value optimized out>, aFeatures=<value optimized out>, aFeatureCount=1)
    at /home/work/mozilla-inbound/tools/profiler/sps_sampler.h:351
#3  mozilla_sampler_start (aProfileEntries=<value optimized out>, aInterval=<value optimized out>, aFeatures=<value optimized out>, aFeatureCount=1)
    at /home/work/mozilla-inbound/tools/profiler/TableTicker.cpp:1100
#4  0x4133af7a in mozilla_sampler_init () at /home/work/mozilla-inbound/tools/profiler/TableTicker.cpp:1012
#5  0x4082f1c8 in XREMain::XRE_main (this=0xa5a5b025, argc=1, argv=0x3, aAppData=0x1) at /home/work/mozilla-inbound/toolkit/xre/nsAppRunner.cpp:3798
#6  0x4082f52a in XRE_main (argc=1, argv=0xbebaebb4, aAppData=0xb1c0, aFlags=<value optimized out>) at /home/work/mozilla-inbound/toolkit/xre/nsAppRunner.cpp:3923
#7  0x00008af4 in do_main (argc=1, argv=0xbebaebb4) at /home/work/mozilla-inbound/b2g/app/nsBrowserApp.cpp:153
#8  main (argc=1, argv=0xbebaebb4) at /home/work/mozilla-inbound/b2g/app/nsBrowserApp.cpp:229

From what I can gather, SPSProfiler::setProfilingStack or SetRuntimeProfilingStack needs to be called. I couldn't find any calls to either of these functions outside of the ones in js/src/jsapi-tests/testProfileStrings.cpp, or a call to EnableSPSProfilingAssertions (or the JS enableSPSProfilingAssertions)

Comment 1

[Dave Hylands [:dhylands]]

I should mention that I hit the assert when using a DEBUG build. Everything seems to work fine with a non-DEBUG build, which suggests that the ASSERT may be superfluous.

Comment 2

[Benoit Girard (:BenWa)]

We shouldn't be calling js::EnableRuntimeProfilingStack this early on startup. We need to wait until the JS engine is ready and has given us a context. See:
http://mxr.mozilla.org/mozilla-central/source/tools/profiler/sps_sampler.h#349

I think we just forgot to init mRuntime to null. I'm surprised this hasn't come up in fennec before.

Comment 3

[Benoit Girard (:BenWa)]

Attachment: patch

I'm traveling and don't have access to b2g hardware. Can you test this patch for me?

Comment 4

[Dave Hylands [:dhylands]]

I removed the errant semi-colon from the patch, and it still asserts in the exact same place.

Comment 5

[Benoit Girard (:BenWa)]

Alright well I think we want this patch anyways so let's land it but keep going with this bug.

Comment 6

[Benoit Girard (:BenWa)]

Comment on attachment 664855 [details] [diff] [review]
patch

Ignore the extra ';'

Comment 7

[Benoit Girard (:BenWa)]

Attachment: Initialize mRuntime

Carry forward r+

Comment 8

[Benoit Girard (:BenWa)]

https://hg.mozilla.org/integration/mozilla-inbound/rev/85f3e07f71ce

Comment 9

[Ryan VanderMeulen [:RyanVM]]

https://hg.mozilla.org/mozilla-central/rev/85f3e07f71ce

Comment 10

[Dave Hylands [:dhylands]]

I thought I'd mention that I'm not seeing the assert any more.

I think it's possible I applied the patch in a source tree different from the one I was building - Sigh.