Replay divergence with VMWare Fusion 8

Robert O'Callahan robert at ocallahan.org
Wed Mar 2 20:04:03 UTC 2016


On Thu, Mar 3, 2016 at 8:54 AM, Jan de Mooij <jdemooij at mozilla.com> wrote:

> On Wed, Mar 2, 2016 at 8:37 PM, Robert O'Callahan <robert at ocallahan.org>
> wrote:
>
>> One question is: are the divergence failures in Firefox clustered in one
>> phase of startup or shutdown, or are they uniformly distributed?
>>
>
> It's hard to say. The failures are not entirely random/non-deterministic
> though; sometimes replays fail at exactly the same number of ticks and I
> very often see the divergence while executing the following tight loops
> (this is a Firefox debug build):
>
>
> https://dxr.mozilla.org/mozilla-central/rev/eb25b90a05c194bfd4f498ff3ffee7440f85f1cd/toolkit/components/url-classifier/LookupCache.cpp#550
>
> https://dxr.mozilla.org/mozilla-central/rev/eb25b90a05c194bfd4f498ff3ffee7440f85f1cd/image/imgFrame.cpp#364
>
> https://dxr.mozilla.org/mozilla-central/rev/eb25b90a05c194bfd4f498ff3ffee7440f85f1cd/js/src/vm/String.cpp#991
>

Interesting. Can we dig into these? Maybe find me on IRC? I'm particularly
interested in knowing whether we diverge at the start or end of the loop,
or in the middle. If we study the disassembly of the code and check
register values we might be able to narrow down the bug.

Not sure if that's just because we execute a lot of branches there, or if
> there's another reason.
>
> Sometimes I also get this failure: Assertion
> `!use_breakpoint_optimization' failed to hold. Is that related?
>

That is also very interesting. That suggests that the problem might not be
in the tick counting (or not just in the tick counting) but we're actually
taking an unexpected control flow path. More data about that failure would
also be good.

Rob
-- 
lbir ye,ea yer.tnietoehr  rdn rdsme,anea lurpr  edna e hnysnenh hhe uresyf
toD
selthor  stor  edna  siewaoeodm  or v sstvr  esBa  kbvted,t
rdsme,aoreseoouoto
o l euetiuruewFa  kbn e hnystoivateweh uresyf tulsa rehr  rdm  or rnea
lurpr
.a war hsrer holsa rodvted,t  nenh hneireseoouot.tniesiewaoeivatewt sstvr
esn
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://mail.mozilla.org/pipermail/rr-dev/attachments/20160303/91370802/attachment-0001.html>


More information about the rr-dev mailing list