Closed Bug 1384142 Opened 9 years ago Closed 9 years ago

Nightly win32 exception on close

Categories

(Toolkit :: Telemetry, defect, P2)

56 Branch
defect

Tracking

()

RESOLVED INVALID

People

(Reporter: rumenta71, Unassigned)

Details

(Keywords: hang, regressionwindow-wanted, Whiteboard: [fce-active-legacy])

Attachments

(2 files)

Attached image Cattura.PNG —
User Agent: Mozilla/5.0 (Windows NT 10.0; Win64; x64; rv:56.0) Gecko/20100101 Firefox/56.0 Build ID: 20170725030209 Steps to reproduce: I've Nightly 56.0a1 (2017-07-25) (64 bit) on Windows 10 build 1607 It's at least two weeks that, pretty every time I close Nightly, SO show a window that say "Nightly has stopped working." If I click on Debug a new window with tile "Visual Studio Debugger JIT" (not sure this is the correct english title, I'm working with an Italian locale) that reports "unhandled win 32 exception in Firefox.exe [8504]. Other number was 3096. I don't have Visual Studio on this machine, so I don't know what else I can report.
Severity: normal → critical
(In reply to Kohei Yoshino [:kohei] from comment #1) > Can you please use mozregression to find a regression range? > > https://mozilla.github.io/mozregression/ > https://blog.nightly.mozilla.org/2017/06/07/using-mozregression-on-windows/ Sure. Since the problem is on a working pc, I will take some time.
I'm still testing because with mozregression Firefox have to stay open some time before I can have the issue. At this moment I have find that Nightly 2017-07-01 have the problem, and Nightly 2017-06-26 should have not. I'll made other tests next week.
Update. The problem begin with Nightly 2017-06-29. Sorry for late answer but English build need to stay open more time before alert
Update. The issue is still here with Version 57.0a1 (2017-08-24) (64 bit) I've noticed that the windows "Nightly has stopped working." is not shown if pingsender.exe is not launched. I've checked with TCPView. Hope this could help.
Component: Untriaged → Widget: Win32
Product: Firefox → Core
Component: Widget: Win32 → Telemetry
Product: Core → Toolkit
Flags: needinfo?(gsvelto)
The fact that the crashreporter doesn't show up might be an instance of bug 1372830. I need more information to figure out what's going on. Can you please describe the affected computer configuration in more detail? Attaching the contents of about:support to the bug would be very helpful. Also are you using an anti-virus program on that machine? If you are, which one?
Flags: needinfo?(gsvelto) → needinfo?(rumenta71)
Attached file about:support info —
Hope this could help. the computer affected Is an Intel Core i5-4460 @3.20Ghz, Windows 10 Pro Version 1703 Build SO 15063.608 with 8 Gb ram No antivirus installed only Windows Defender Security Center. Windows Locale is Italian Nightly locale is Italian also. I've added file about_support.txt with export from about:support page. I've only obfuscated value of following pref media.webrtc.debug.aec_log_dir: C:\Users\XXXXXX~1.YYY\AppData\Local\Temp media.webrtc.debug.log_file: C:\Users\XXXXXX~1.YYY\AppData\Local\Temp\WebRTC.log With these pref I've noticed that are addressing to an non existent folder. The current windows profile is located in C:\Users\zzzzzzz.kkkkkkkkk\AppData\Local\Temp
Flags: needinfo?(rumenta71)
Thanks a lot, I'll look into this ASAP.
Priority: -- → P1
I've fiddled with the pingsender telemetry settings and even tried to inject abnormal failures in the code but I cannot seem to reproduce this on my box. (In reply to Remy Mentasti from comment #7) > media.webrtc.debug.aec_log_dir: C:\Users\XXXXXX~1.YYY\AppData\Local\Temp > media.webrtc.debug.log_file: > C:\Users\XXXXXX~1.YYY\AppData\Local\Temp\WebRTC.log > > With these pref I've noticed that are addressing to an non existent folder. > The current windows profile is located in > C:\Users\zzzzzzz.kkkkkkkkk\AppData\Local\Temp This is the only clue I have that something might be wrong. That path is obtained via the directory service and so is the pingsender path (and the ping path too) so they might be wrong. However this shouldn't cause a crash deep in JavaScript, let alone one that isn't caught by the crashreporter. That alone is rather worrisome, because we cannot tell if other people are hitting the same issue if the crashreporter isn't showing up. That being said without knowing the exact place where the crash is happening it's really hard to diagnose it. My guess is that this is happening somewhere within nsIProcess but I can't be sure. I'll try to find an alternative way of capturing a minidump so we can dig a little bit deeper into this issue.
Let's try to use WER to grab a minidump, Remy could you please do the following: 1) In the Windows registry create the HKEY_LOCAL_MACHINE\SOFTWARE\Microsoft\Windows\Windows Error Reporting\LocalDumps key and leave it empty 2) Re-enable the pingsender if you disabled it 3) Use Firefox until it crashes 4) Once Firefox has crashed dismiss the debugger window you're presented and go into %LOCALAPPDATA%\CrashDumps (e.g. C:\Users\<name>\AppData\Local|CrashDumps) you should find a file called firefox.exe.<number>.dmp there 5) Send me the .dmp file via e-mail, do not attach it to the bug since it might contain sensitive information If something went wrong and no .dmp file was generated try again but this time launch firefox manually via cmd.exe with the MOZ_CRASHREPORTER_DISABLE=1 variable set, e.g.: set MOZ_CRASHREPORTER_DISABLE=1 cd "C:\Program Files\Mozilla Firefox\" # Or the correct path to Firefox installation directory firefox.exe Once you're done you can delete the LocalDumps registry key you created, this will prevent WER from gathering more minidumps. See this page for more information on this: https://msdn.microsoft.com/en-us/library/windows/desktop/bb787181(v=vs.85).aspx
Flags: needinfo?(rumenta71)
We can have a look at support.mozilla.org and other online forums to see if this is a widespread complaint as 56 goes to a bigger audience.
Keywords: regression
Whiteboard: [fce-active]
(In reply to Gabriele Svelto [:gsvelto] from comment #10) > Let's try to use WER to grab a minidump, Remy could you please do the Hi, The key was already in Windows registry, so I have two dumps. I'll send you as soon as possible.
Flags: needinfo?(rumenta71)
I've inspected the dumps that Remy sent me and discovered that the crash happens in the ig75icd64.dll module which is part of the Intel graphics driver version 20.19.15.4531. I've looked up the type of error and relevant symbols/instruction offsets and found out that the issue Remy is hitting is a bug in the Intel graphics driver for which there is already a detailed crash report: https://software.intel.com/en-us/forums/graphics-driver-bug-reporting/topic/731358 The type of error, addresses and stack trace match exactly the ones in the report so I'm pretty confident this is not a problem on our side. As to why the pingsender is triggering it is hard to say but it appears to be a race so launching an application during shutdown might make it more likely to happen.
So... is this resolved?
Flags: needinfo?(gsvelto)
I'd say yes. I've discussed it with the reporter who updated his graphics drivers in the meantime. He hasn't experienced the bug anymore but he also mentioned that he hasn't used the browser that much so he wanted to wait a few more days to report back. In the meantime I think we can safely drop the priority of this bug.
Flags: needinfo?(gsvelto)
Priority: P1 → P2
Update. Last Monday I've updated Intel HD Graphics 4600 driver to version 15.40.36.4703 with Intel Driver and Support Assistant. Info windows on ig75icd64.dll file report version 20.19.15.4703 From this moment, I've not experienced the bug anymore. Thanks to Gabriele and to everybody for helping me.
(In reply to Remy Mentasti from comment #16) > Update. > > Last Monday I've updated Intel HD Graphics 4600 driver to version > 15.40.36.4703 with Intel Driver and Support Assistant. > > Info windows on ig75icd64.dll file report version 20.19.15.4703 > > From this moment, I've not experienced the bug anymore. > > Thanks to Gabriele and to everybody for helping me. Thanks for the feedback. I'll close this bug as INVALID since we've pretty much proved that it's not an issue on our side.
Status: UNCONFIRMED → RESOLVED
Closed: 9 years ago
Resolution: --- → INVALID
Whiteboard: [fce-active] → [fce-active-legacy]
You need to log in before you can comment on or make changes to this bug.

Attachment

General

Creator:
Created:
Updated:
Size: