604124 - Nvidia driver stops working, Firefox won't repaint

Status: RESOLVED WORKSFORME

(Core :: Graphics, defect)

Description

[Matt Caywood]

I am getting about two crashes a week on my work laptop (graphics card is NVidia Quadro NVS 160M; driver version is 8.16.11.8840. Windows 7-64)

Firefox crash report: http://crash-stats.mozilla.com/report/pending/3bea05f1-e770-4d86-9595-0825b2101013

bp-3bea05f1-e770-4d86-9595-0825b2101013

No Windows problem report however, the "problem reports" log is clear.

I'm going to upgrade the NVidia driver and see if it helps.

Comment 1

[Scoobidiver (away)]

I don't think it is related to Core Graphics section.
Can you provide your graphic section of "about:support" page ?
Once your Nvidia driver is updated, can you try with the latest nightly build:
http://nightly.mozilla.org/

Comment 2

[Scoobidiver (away)]

And also try with a new profile:
http://support.mozilla.com/en-US/kb/managing+profiles

Comment 3

[Matt Caywood]

I'm filing it in Core Graphics per Vlad's instructions at:

http://blog.vlad1.com/2010/10/12/getting-information-from-windows-vista7-display-driver-resets/

It seems like EXACTLY the kind of problem he is looking for. Please read his post.

It has nothing to do with my profile, I didn't start experiencing these crashes until I upgraded to a version of Firefox 4 that had Direct2D enabled.

Will try with latest nightly if/once I can replicate after the driver upgrade.

Comment 4

[Vladimir Vukicevic [:vlad] [:vladv] (needinfo me, slow to respond)]

Hmm, are you getting crashes?  That crash report looks like it was generated with my force-crash tool, which would imply that you're not actually getting a crash.  Are you getting the "Display driver reset" popup from windows?

The firefox crash report doesn't show anything unusual -- we're just spinning the normal event loop, waiting fort he next event; most causes of display driver crashing/window not repainting afterwards end up with firefox hung inside a direct3d call or another, but that's not the case here.  Need some more details.

Also, 8.16.11.8840 is somewhat old -- the latest version is 8.17.12.6061 (258.96).  Can you try upgrading to the latest driver version and see if that improves things?  If so, we may need to blacklist the older drivers.  Thanks!

Comment 5

[Matt Caywood]

Vlad, sorry I wasn't more clear. I was getting the "display driver reset" pop-up, and Firefox stops repainting, so I have to use the force-crash tool to get any result. No Windows problem report.

BTW, I didn't know how to report it until I saw your helpful post on planet. 

I've upgraded to the latest version Dell recommends for my laptop (8.16.11.8992) and haven't gotten any crashes in the last week, so I will see if the driver upgrade has fixed the situation.

Comment 6

[Vladimir Vukicevic [:vlad] [:vladv] (needinfo me, slow to respond)]

Huh.. no windows problem report is odd/worrying.  That should be auto generated whenever the display driver reset triggers... argh.  But yeah, keep an eye on it -- if it seems good, please just reply in this bug report and we'll blacklist the older driver versions and cross our fingers :)

Comment 7

[Brian Smith (:briansmith, :bsmith, use NEEDINFO?)]

I am having the exact same problem as Matt Caywood in comment 5, at least once a day, every day, since I received my Lenovo T510 (nVidia NVS 3100M) on Minefield.

Event log shows "Display driver nvlddmkm stopped responding and has successfully recovered." No Windows problem report.

FWIW, my laptop and I are in 650 Castro every day.

Comment 8

[Brian Smith (:briansmith, :bsmith, use NEEDINFO?)]

Because this doesn't crash Firefox but makes it unusable until it is restarted, normal crash recovery doesn't work and you risk losing all your tabs and all the data in them.

Comment 9

[Brian Smith (:briansmith, :bsmith, use NEEDINFO?)]

Attachment: Minidump saved after attaching to process and breaking

I used Task Manager to attach VS2010 to the process and take this minidump. This minidump was *not* the result of a crash. Firefox continues executing with 100% CPU. After resuming and breaking a second time, VS2010 started sating that the Firefox process appeared to be deadlocked as no user code was executing.

Win7-x64 on a T510 with nVidia 3100M, driver version 8.17.12.5738.

Comment 10

[Brian Smith (:briansmith, :bsmith, use NEEDINFO?)]

Attachment: Screenshot of Firefox window

Comment 11

[Brian Smith (:briansmith, :bsmith, use NEEDINFO?)]

Attachment: Screenshot of blank white box that was drawn on top of all windows

This white box looks to be about the same dimensions as the address bar drop-down.

Comment 12

[Brian Smith (:briansmith, :bsmith, use NEEDINFO?)]

Attachment: Photograph of screen showing that menus appear over the white box

Comment 13

[Damon Sicore (:damons)]

Bas, can you own this?  I'm adding this to the blocker list until we get to the bottom of it.

Comment 14

[NVD]

The latest Nvidia driver is 260.80 WHQL. Please test this driver and update to the latest nightly build of Firefox 4.0.

http://www.nvidia.com/object/notebook-win7-winvista-260.89-whql-driver.html

http://www.nvidia.com/object/notebook-win7-winvista-64bit-260.89-whql-driver.html

Comment 15

[Brian Smith (:briansmith, :bsmith, use NEEDINFO?)]

That's not the driver in ThinkVantage System Update (TVSU) or Windows Update and that nVidia-provided driver won't install. None of the inf hacks I found for getting the nVidia reference driver to install would work.

I am happy to hack together some solution for getting the reference driver installed but FF4.0 should work to see if the problem is in the driver, but upgrading the driver isn't a realistic solution for users unless/until it shows up on TVSU or Windows Update.

Comment 16

[Brian Smith (:briansmith, :bsmith, use NEEDINFO?)]

Let's try again with real English: I am happy to work to hack together some solution for getting the reference driver installed to see if the problem is in the driver,  but upgrading the driver isn't a realistic solution for users unless/until it shows up on TVSU or Windows Update (because of Lenovo's modifications to the device IDs).

Here are some instructions for getting the reference drivers installed:
http://208.74.204.134/t5/T400-T500-and-newer-T-series/T410-discrete-gfx-Problem-with-Video-Card-and-Windows-Aero/m-p/201893
http://identitybase.blogspot.com/2010/07/lenovo-thinkpad-t410t410i-screen-black.html

Comment 17

[NVD]

Brian, don't use the modified driver infs. I thought the newest drivers would install since the supported products tab doesn't say that it wouldn't install on Lenovo's laptops. For example, "Other Sony VAIO notebooks are not supported at this time (please contact Sony for driver support)." So clearly you can't install on any Sony notebook.

Best thing to do is to wait for a newer driver from Nvidia that will install on the T510 or Lenovo releases a newer driver that is supported on the T510. And definitely report this bug to Nvidia.

Comment 18

[Bas Schouten (:bas.schouten)]

If you happen to be able to I'd be interested to know if the reference driver would fix the issue. Then we could simply blacklist older drivers for this device. Although I realize that would mean for the average Lenovo user with this laptop we'd get no D2D.

Comment 19

[Brian Smith (:briansmith, :bsmith, use NEEDINFO?)]

Attachment: Better results in 2010-10-22 or 2010-10-24 build

I am always running Firefox, so I don't know if Firefox is always the one causing crashes. But, in at least one case, a crash occurred when Firefox was in the background, while I was using other apps. So, it may indeed be the case that the driver is buggy. 

And, previously I had got stuck on a 2010-10-19 build which exhibited the previously-described behavior. I upgraded to -22 and then to -24. Sometime after upgrading to -22 (maybe before upgrading to -24), it looks like the Firefox's recovery from the video driver crash improved. Now, the browser stays usable but there are rendering artifacts. See the attackment.

Comment 20

[Brian Smith (:briansmith, :bsmith, use NEEDINFO?)]

(In reply to comment #18)
> If you happen to be able to I'd be interested to know if the reference driver
> would fix the issue. Then we could simply blacklist older drivers for this
> device. Although I realize that would mean for the average Lenovo user with
> this laptop we'd get no D2D.

I will give the machine to QA so they can help with this issue, as Damon suggested. I will try to help them get the reference driver installed if they have trouble with getting it installed.

Comment 21

[NVD]

Nvidia has released new 260.99 WHQL drivers today, don't know if they will install or not though.

http://www.nvidia.com/object/notebook-win7-winvista-260.99-whql-driver.html

http://www.nvidia.com/object/notebook-win7-winvista-64bit-260.99-whql-driver.html

Comment 22

[Bas Schouten (:bas.schouten)]

(In reply to comment #19)
> Created attachment 485660 [details]
> Better results in 2010-10-22 or 2010-10-24 build
> 
> I am always running Firefox, so I don't know if Firefox is always the one
> causing crashes. But, in at least one case, a crash occurred when Firefox was
> in the background, while I was using other apps. So, it may indeed be the case
> that the driver is buggy. 
> 
> And, previously I had got stuck on a 2010-10-19 build which exhibited the
> previously-described behavior. I upgraded to -22 and then to -24. Sometime
> after upgrading to -22 (maybe before upgrading to -24), it looks like the
> Firefox's recovery from the video driver crash improved. Now, the browser stays
> usable but there are rendering artifacts. See the attackment.

Ah, I know why this is, we need to blow out the shadow cache when our device is destroyed. I'll work on this.

Comment 23

[Matt Caywood]

Just FYI, I haven't had a crash in 10 days of normal browsing since upgrading to 8.16.11.8992. So far it's looking like a bug in 8.16.11.8840 was to blame.

Comment 25

[Masayuki Nakano [:masayuki] (he/him)(JST, +0900)]

Looks like bug 606983 fix improves this bug. The client area of Windows work fine for me.

However, when the Firefox button is on title bar, Aero glass of the window isn't recovered. And also I cannot use minimize, maximize and close button on the title bar (they are not painted and nothing happens when I clicked their position).

If I hide the Firefox button by menubar, all okay for me now.

Comment 26

[Jim Mathies [:jimm]]

(In reply to comment #25)
> Looks like bug 606983 fix improves this bug. The client area of Windows work
> fine for me.
> 
> However, when the Firefox button is on title bar, Aero glass of the window
> isn't recovered. And also I cannot use minimize, maximize and close button on
> the title bar (they are not painted and nothing happens when I clicked their
> position).
> 
> If I hide the Firefox button by menubar, all okay for me now.

That's bug 545892 and should be fixed. If you've found a new, reliable way to reproduce that please file a new bug.

Comment 27

[Bas Schouten (:bas.schouten)]

I had a crash on my NVidia machine earlier and it seemed to recover beautifully back into glass eventually.

Comment 28

[Masayuki Nakano [:masayuki] (he/him)(JST, +0900)]

(In reply to comment #26)
> That's bug 545892 and should be fixed. If you've found a new, reliable way to
> reproduce that please file a new bug.

filed bug 608515.

Comment 29

[Bas Schouten (:bas.schouten)]

Are you still seeing this? I believe it will be fixed by several improvements we made to our recovery logic, but please do confirm!

Comment 30

[Brian Smith (:briansmith, :bsmith, use NEEDINFO?)]

I have not seen this for weeks and I haven't seen any problems with recovering glass after a driver crash for at least a week. (The only problem I've noticed recently was with switching from basic->Aero while Minefield is running.)

Comment 31

[Brian Smith (:briansmith, :bsmith, use NEEDINFO?)]

(In reply to comment #30)
> I have not seen this for weeks and I haven't seen any problems with recovering
> glass after a driver crash for at least a week. (The only problem I've noticed
> recently was with switching from basic->Aero while Minefield is running.)

Actually, I just tried it and I don't see this problem any more either.