Open Bug 665915 Opened 14 years ago Updated 3 years ago

NVIDIA driver does not recover well from resets, sometimes blue-screens

Categories

(Core :: Graphics, defect)

Product:
Core
Component:
Graphics
Platform:
All
Windows 7
Type:
defect

Tracking

()

People

(Reporter: bjacob, Unassigned)

References

(
URL
)

Details

This WebGL test forces a driver reset on Win7: https://cvs.khronos.org/svn/repos/registry/trunk/public/webgl/sdk/tests/extra/lots-of-polys-example.html The problem is that NVIDIA driver often fails to recover well from that reset. After a few tries it's easy to get a blue screen. Can also be triggered by a silverlight 5 demo: http://people.mozilla.org/~bjacob/SilverLight5DOSJustLikeWebGL/HelloWorld3D/Bin/Debug/HelloWorld3DTestPage.html Actual results: sometimes blue screen, sometimes garbage on the screen Expected results: consistently flawless recovery from driver reset
I tried to reproduce this on my system; Win7-64 and GeForce 8800. I have some questions: 1) I installed the latest Firefox 5 but I noticed WebGL wasn't available. I had to do about:config webgl.force-enabled=true. I'm using an internal driver, which might be the cause. Do white-list or black-list drivers? It's possible you're logic thought to disable due to my yet-to-be-released driver number. 2) When I went to the site and ran the webpage I saw the screen flicker after 2s and the system recover. It seemed stable after recovering. In the description you say "After a few tries". Do I just hit refesh on the page or do I need to do more to try again? Hitting refresh for me didn't perform another GPU reset, which makes me wonder whether I need to restart the browser.
3) Another question. Does FF5 use OpenGL or D3D for the WebGL implementation?
(In reply to comment #1) > 1) I installed the latest Firefox 5 but I noticed WebGL wasn't available. I > had to do about:config webgl.force-enabled=true. I'm using an internal > driver, which might be the cause. Do white-list or black-list drivers? It's > possible you're logic thought to disable due to my yet-to-be-released driver > number. On Windows, we have a whitelist of PCI Vendor IDs, and then we get the driver version from the Windows registry, parse it, and require it to be at least 8.17.12.5721 (on Windows 7). Can you confirm that you're using an official 32bit build of Firefox? 64bit Firefox builds on Windows are not yet supported. > 2) When I went to the site and ran the webpage I saw the screen flicker > after 2s and the system recover. It seemed stable after recovering. In the > description you say "After a few tries". Do I just hit refesh on the page or > do I need to do more to try again? As far as I remember, just hitting refresh was enough. > Hitting refresh for me didn't perform > another GPU reset, which makes me wonder whether I need to restart the > browser. That suggests that after the first time it just failed to create a WebGL context. I will try again in Windows ASAP. (In reply to comment #2) > 3) Another question. Does FF5 use OpenGL or D3D for the WebGL implementation? By default, Firefox uses Direct3D 9, via ANGLE (which implements OpenGL ES 2.0 on top of D3D9). You can make Firefox use the OpenGL driver by setting webgl.prefer-native-gl=true. I observed the same problems with this Silverlight 5 page (warning, runs immediately): http://people.mozilla.org/~bjacob/SilverLight5DOSJustLikeWebGL/HelloWorld3D/Bin/Debug/HelloWorld3DTestPage.html
I reproduced the corruption problem with a GeForce GTS 260. I'm filing an internal bug to get this fixed.
Thank you very much!
Note, I have been reproducing this on a Quadro FX 880M with the 275.33 driver on Windows 7 64bit (but Firefox itself is 32bit).
I was also using the 32bit version of FF5.0. Sorry I didn't answer your question earlier.
Win7-64 Bit Ultimate Firefox 6.0.2 nVidia Driver 8.17.12.8026 running on SLi GTX280m Firefox causes the nVidia system tray icon to report that the nVidia drivers need to recover and the screen flashes - but this process can often end up looping until the system becomes completely unresponsive, or crashes to BSOD. This is also caused by the WebGL test linked in the first post. It happens whether or not "Use hardware acceleration when available" is checked or not It does not create a Firefox crash report as seen in 'about:crashes' It does cause a System Event reporting a "Bug Check" with the following details : "The computer has rebooted from a bugcheck. The bugcheck was: 0x00000116 (0xfffffa8007ff74e0, 0xfffff8800f43bcd8, 0x0000000000000000, 0x000000000000000d). A dump was saved in: C:\Windows\Minidump\092511-20670-01.dmp. Report Id: 092511-20670-01." I have the Minidumps available if anyone wants to see them? This bug is now becoming too frequent and onerous for me to keep using Firefox as my main browser.
(In reply to JimC from comment #8) > This bug is now becoming too frequent and onerous for me to keep using > Firefox as my main browser. Please try the following: go to about:config and set all of the following: layers.acceleration.disabled=true gfx.direct2d.disabled=true webgl.disabled=true gfx.font_rendering.directwrite.enabled=false Also disable Flash plugin, and restart Firefox. If that removes the driver crashes, can you please investigate which of the above makes the difference.
(In reply to Benoit Jacob [:bjacob] from comment #9) > (In reply to JimC from comment #8) > > This bug is now becoming too frequent and onerous for me to keep using > > Firefox as my main browser. > > Please try the following: go to about:config and set all of the following: > layers.acceleration.disabled=true > gfx.direct2d.disabled=true > webgl.disabled=true > gfx.font_rendering.directwrite.enabled=false > > Also disable Flash plugin, and restart Firefox. > > If that removes the driver crashes, can you please investigate which of the > above makes the difference. Well, obviously ;) , with webgl.disabled=true, the WebGL test mentioned above is safe - because it can't find a 'context'... With WebGL.disabled=false, then no matter what I do with the other settings - the computer eventually crashes to BSOD. I usually drive an external (HDMI) monitor from my laptop, but I reverted to the native screen for some tests, and still got BSOD. I even created a new profile for testing with no AddOns, all Plugins disabled, and all the settings you mentioned above except WebGL enabled - and still got BSOD. I have been getting these Firefox crashes with previous nVidia drivers also. Is the WebGL test ever supposed to work? A quick test on Firefox 5.0.1 on my Mac OSX Snow Leopard turned out to be the only application that has EVER completely crashed/frozen my Mac ! Anyway, I have left the settings you mentioned disabled except webgl, and Flash, and I will see if that gets rid of my more usual video driver crashes....
Wait, I don't get it. Because you wrote this: (In reply to JimC from comment #8) > This bug is now becoming too frequent and onerous for me to keep using > Firefox as my main browser. I thought that that meant that you were getting frequent crashes during normal usage, but this: (In reply to JimC from comment #10) > Well, obviously ;) , with webgl.disabled=true, the WebGL test mentioned > above is safe - because it can't find a 'context'... Means that you only get crashes while running the WebGL DOS testcase?? I'm confused.
(In reply to Benoit Jacob [:bjacob] from comment #11) > Wait, I don't get it. Because you wrote this: > > (In reply to JimC from comment #8) > > This bug is now becoming too frequent and onerous for me to keep using > > Firefox as my main browser. > > I thought that that meant that you were getting frequent crashes during > normal usage, but this: > > (In reply to JimC from comment #10) > > Well, obviously ;) , with webgl.disabled=true, the WebGL test mentioned > > above is safe - because it can't find a 'context'... > > Means that you only get crashes while running the WebGL DOS testcase?? > > I'm confused. Rarely a day goes by when my computer doesn't crash at least once because of Firefox - regardless of the WebGL test; Often 2 or 3 times a day. For quite some time now (several weeks/months), I couldn't figure out what was causing it, because I simply didn't believe that a web browser would be the culprit, and I spent a great deal of time testing different video drivers from nVidia & Dell. However, in the last week or so it has become clear that with only Firefox running, the crashes still occurred. If anecdotal evidence can be trusted to any degree, I would say Firefox causes more crashes when the Steam (game platform) is also running, or if Windows Media Player is also running. However, I do not have a 'specifically repeatable' test case set of circumstances guaranteed to cause a crash. Or at least I didn't until I saw the WebGL test mentioned in the first post above; The WebGL test consistantly crashes my video drivers, resulting in a BSOD, in a seemingly identical way to which Firefox crashes my PC during normal use. Obviously with WebGL disabled, the WebGL test itself doesn't even get the chance to cause a crash - but my testing of the effect of WebGL during general usage might end up taking a lot longer to confirm one way or the other, as I mention in the last sentence of my previous post. Perhaps some people wouldn't think that crashing on average once a day is 'onerous'; But it is in my opinion. A web browser should not be regularly causing graphics driver failures that need a system reboot and if I can't pin down the issue, I'll move to Opera or Chrome. (I'm not a dev., just a long time fan of Firefox, who is nearing the end of his patience ;)
Just as a thought, is it possible that you might have a hardware issue with that card? Even though the GTX280M is rare I'd still expect some others to also have this issue. Do you ever see any issues when playing video games or in other ways stressing your video card?
Jim, it seems that you have a serious problem on that machine, and it's unrelated to the problem that the WebGL test is exposing. The WebGL test just exposes a problem between WebGL and current graphics drivers, namely that one can freeze the whole display by issuing an exceedingly long drawing command. The present bug is solely for discussing a NVIDIA-specific bug that aggravates this issue by giving blue screens. The general problem that you have on your machine probably has nothing to do with that, so I'd look elsewhere. In comment 9 I gave a few ideas of stuff to try disabling, but if that doesn't make your system more secure, that means that your problem is probably not Graphics-related. FYI, a major cause of Firefox crashes in Windows is random malware. If you go to about:crashes, you can see recent crash reports that you sent. If some of them mentions something about "Winsock LSP" then it is rather probable that you have malware.
Benoit, he is also experiencing this bug, because he says "The WebGL test consistantly crashes my video drivers, resulting in a BSOD." JimC, when your computer crashes in those other cases (not the WebGL testcase above), does the screen turn black (i.e. is the driver resetting). If so, then I think that is another, more serious, instance of this bug. (More serious, because it indicates a "real world" problem vs. a synthetic test or maliciously written WebGL script.)
OK, some extra details. My system is currently stable under high graphics load - I can complete Unigine benchmarks, and play Crysis 2, Mafia II without crashes. I don't overclock my processor or graphics cards. I don't seem to have any issues with Winsock 2 LSP, and a utility called LSP-Fix couldn't see any layers that needed removal either. (Although I will re-iterate that these are graphics driver crashes, rather than Firefox crashes, so about:crashes seems less relevant than Windows Minidumps.) Forgetting about WebGL tests for a moment the problem I encounter during normal everyday usage is this : Firefox will seem to freeze, at which point I realise the entire desktop/computer has become unresponsive. After a couple of seconds the screen will flicker, and go black. (When driving an external monitor, the monitor lights flash to indicate no graphics signal is being received any longer, despite cables still being secure.) Often the screen never recovers. Sometimes the screen will flicker back to life, and a System Tray Notification speech bubble will inform me that the nVidia Graphics Drivers have recovered from an error. Often in this situation, the screen will again flicker off, and loop through a crash/recovery cycle, until finally hitting a BSOD. Windows Event Viewer will record that the system has recovered after a BugCheck, and that a Windows Minidump is available. So why did I post in this forum thread? The thread title seemed to sum up the error I am getting (with no mention of WebGL); And after attempting the WebGL test mentioned in the OP, it seemed to be the exact same error I was getting in general usage. So I thought I'd report my experience in this thread. If I'm the only one, then I guess there probably is some uniquely unfortunate confluence of hardware/settings that causes this error for me and no one else. Either way, I will try combinations of the settings that were suggested above in general usage and if I can report anything useful, I will.
(In reply to JimC from comment #16) > OK, some extra details. > > My system is currently stable under high graphics load - I can complete > Unigine benchmarks, and play Crysis 2, Mafia II without crashes. > I don't overclock my processor or graphics cards. > > I don't seem to have any issues with Winsock 2 LSP, and a utility called > LSP-Fix couldn't see any layers that needed removal either. > (Although I will re-iterate that these are graphics driver crashes, rather > than Firefox crashes, so about:crashes seems less relevant than Windows > Minidumps.) > > > > Forgetting about WebGL tests for a moment the problem I encounter during > normal everyday usage is this : > > Firefox will seem to freeze, at which point I realise the entire > desktop/computer has become unresponsive. After a couple of seconds the > screen will flicker, and go black. (When driving an external monitor, the > monitor lights flash to indicate no graphics signal is being received any > longer, despite cables still being secure.) > Often the screen never recovers. > Sometimes the screen will flicker back to life, and a System Tray > Notification speech bubble will inform me that the nVidia Graphics Drivers > have recovered from an error. > Often in this situation, the screen will again flicker off, and loop through > a crash/recovery cycle, until finally hitting a BSOD. > Windows Event Viewer will record that the system has recovered after a > BugCheck, and that a Windows Minidump is available. > > > > So why did I post in this forum thread? > > The thread title seemed to sum up the error I am getting (with no mention of > WebGL); And after attempting the WebGL test mentioned in the OP, it seemed > to be the exact same error I was getting in general usage. > > So I thought I'd report my experience in this thread. If I'm the only one, > then I guess there probably is some uniquely unfortunate confluence of > hardware/settings that causes this error for me and no one else. > > Either way, I will try combinations of the settings that were suggested > above in general usage and if I can report anything useful, I will. Sadly, the only idea I have here is for you to switch off hardware acceleration in Firefox, and report the issue to NVidia. Regardless of what we might be doing to 'tickle' this, it is a crash in the NVidia Kernel Mode driver and in the end the real issue is at their side (or faulty hardware, but considering your benchmarks are succesful that seems unlikely). What might be interesting is to see if the same happens if you use IE9, it should be using Direct2D as well and might very well tickle the same bug.
(In reply to JimC from comment #16) > The thread title seemed to sum up the error I am getting (with no mention of > WebGL); And after attempting the WebGL test mentioned in the OP, it seemed > to be the exact same error I was getting in general usage. This bug is only here to track the issue that "NVIDIA driver does not recover well from resets, sometimes blue-screens" with NVIDIA. Normally you shouldn't get frequent driver resets at all, so if you do, that is a separate bug that should be filed separately. The WebGL testcase here is just a means to trigger a driver reset. It's annoying that these can be triggered by WebGL, but that's a separate issue. This test is _supposed_ to force your driver to reset, so the fact that it does is not specific to your machine and is not related to the problem you're experiencing. In comment 9 I gave a few ideas of stuff to disable to prevent Firefox from 'tickling' your driver. This will not help at all with the WebGL testcase (unless you disable WebGL) but normally solves "Firefox crashes my system due to graphics stuff" kind of bugs. If the steps in comment 9 don't fix your problem and your graphics driver keeps crashing when you use Firefox, then that's extremely worrying as that means that just regular non-accelerated stuff crashes your driver; so in that case, I don't even know where to start looking. In any case, let's keep this bug focused on the issue described in its title. Separate bugs can be filed separately.
Severity: normal → S3
You need to log in before you can comment on or make changes to this bug.