1322211 - Crash in mozalloc_abort | NS_DebugBreak | mozilla::ipc::FatalError | mozilla::ipc::IProtocol::HandleFatalError | mozilla::ipc::IProtocol::FatalError | mozilla::plugins::PPluginInstanceChild::CallPStreamNotifyConstructor

Status: RESOLVED DUPLICATE of bug 1278161

(Core Graveyard :: Plug-ins, defect, P1)

Description

[Ting-Yu Chou [:ting] (away)]

This bug was filed from the Socorro interface and is 
report bp-0d7a1cb5-e79f-482e-a6ac-3ddd02161206.
=============================================================
#21 top crash of Nightly 20161204030210 on Windows, 6 crashes from 5 installations.

Comment 1

[Benjamin Smedberg]

Bill, do you know why we wouldn't have the abort message in the crash metadata here?

Comment 2

[Bill McCloskey [inactive unless it's an emergency] (:billm)]

This crash was submitted from the infobar. I think we don't get the crash reason since it's attached via PrepareChildExceptionTimeAnnotations. That function writes its annotations into a temp file that (I believe) doesn't get picked up by the infobar code.

I filed bug 1323097 for this. I won't have time to work on it myself though.

Comment 3

[[:philipp]]

this first started appearing on 52.0a1 build 20161110030211 and frequently thereafter.

this would be the pushlog for build 20161110030211 -1 day:
https://hg.mozilla.org/mozilla-central/pushloghtml?fromchange=783356f1476eafd8e4d6fa5f3919cf6167e84f8d&tochange=336759fad4621dfcd0a3293840edbed67018accd
...and back a further 2 days: 
https://hg.mozilla.org/mozilla-central/pushloghtml?fromchange=908557c762f798605a2f96e4c943791cbada1b50&tochange=783356f1476eafd8e4d6fa5f3919cf6167e84f8d

Comment 4

[Astley Chen (inactive)]

[Tracking Requested - why for this release]: #6 top crash on 52.0b plugin process in last 7 days.

https://crash-stats.mozilla.com/topcrashers/?product=Firefox&version=52.0b&days=7&process_type=plugin

Comment 5

[Benjamin Smedberg]

Loading a raw dump, aErrorMsg is "constructor for actor failed", which means that this call failed:

    bool sendok__;
    {
        GeckoProfilerTracingRAII syncIPCTracer(
                "IPC",
                "PPluginInstance::Msg_PStreamNotifyConstructor");
        sendok__ = (GetIPCChannel())->Call(msg__, (&(reply__)));
    }
    if ((!(sendok__))) {
        FatalError("constructor for actor failed");

I can think of the following reasons this might fail:

* the channel was destroyed. Unlikely in this case because we probably wouldn't have a crash report
* message deserialization issues
* AllocPStreamNotifyParent failure, but this is infallible-alloc so very unlikely
* AnswerPStreamNotifyConstructor failure: this is programmatically possible, see https://dxr.mozilla.org/mozilla-central/rev/1d025ac534a6333a8170a59a95a8a3673d4028ee/dom/plugins/ipc/PluginInstanceParent.cpp#581
** In the case where we the request fails immediately, we send a message to destroy the actor that is currently being constructed
** PStreamNotifyParent::Send__delete__ is failing in this case?

I'm going to try and construct a testcase where this might trigger. The most obvious choice is the automated test which already triggers this codepath, but that appears to be working: http://searchfox.org/mozilla-central/rev/d20e4431d0cf40311afa797868bc5c58c54790a2/dom/plugins/test/mochitest/test_streamNotify.html#52 (loading http://localhost:-8/)

Comment 6

[Benjamin Smedberg]

In Firefox 51, the signature for this is mozalloc_abort | NS_DebugBreak | mozilla::ipc::FatalError | mozilla::plugins::PPluginInstanceChild::FatalError | mozilla::plugins::PPluginInstanceChild::CallPStreamNotifyConstructor  and so this is a duplicate of a crash which appears to be a shutdown-only issue and therefore low priority.

Comment 7

[Astley Chen (inactive)]

Per comment 6, this is not a new crash on Firefox 52.

Comment 8

[Gerry Chang [:gchang]]

Too late for 51. Mark 50/51 won't fix.

Comment 9

[Liz Henry (:lizzard) (relman/hg->git project)]

Too late for 53 as well, also this is a duplicate.

Comment 10

[Gerry Chang [:gchang]]

Too late for 54.