1186755 - Notification panel freezes PC with Intel(R) G41 Express Chipset

Status: RESOLVED INCOMPLETE

(Core :: Graphics, defect)

Description

[Rod E]

User Agent: Mozilla/5.0 (Windows NT 6.1; rv:38.0) Gecko/20100101 Firefox/38.0
Build ID: 20150525141253

Steps to reproduce:

Any time Thunderbird or Firefox displays a notification panel (e.g. update available, new email arrived) the PC freezes completely.  The panel displays, but the computer freezes as the panel starts to fade out.

This is readily reproducible with Firefox and Thunderbird, and started only after Mozilla updates were installed.  All other programs function normally.


Actual results:

The PC freezes, and can only be restarted with the reset button or power cycle.

My machine is only usable when ALL notifications are turned off.  More info is available at http://forums.mozillazine.org/viewtopic.php?f=39&t=2948201&e=0


Expected results:

Normal behaviour is expected - no-one expects their PC to freeze up completely just because a notification panel has been displayed.

Comment 1

[Loic]

Is it reproducible with a fresh profile?
https://support.mozilla.org/en-US/kb/profile-manager-create-and-remove-firefox-profiles

Comment 2

[Rod E]

I don't know, and I am predictably unwilling to find out.  Anything that completely locks up the PC is potentially dangerous, and the affected machine is used for my business.  The problem has already caused much grief and I don't need more.

Note that the issue started ONLY when updates were applied to both Firefox and Thunderbird. There is exactly the same issue with both (freezing when a notification panel has displayed), so the problem is within common code.

Comment 3

[Jason Bassett]

Hello

Just registering my experience of this issue too.  My users are not using Mozilla Firefox, only Mozilla Thunderbird.

Issue started since upgrae to 38.1.0.

I have disabled all notifications for the users but they report it still freezes.  I have therefore got them running in Mozilla Thunderbird safe mode for now - this seems to work so far.

Regards

Jason

Comment 4

[Wayne Mery (:wsmwk)]

disable hardware acceleration. then run in normal mode - not safe mode.
better?

Comment 5

[Jason Bassett]

Hello Wayne

For one of my users experiencing this issue I have set gfx.direct2d.disabled true and layers.acceleration.disabled true using the Config Editor.

I will report back once confident it has resolved issue or it machine locks up again.

Regards

Jason

Comment 6

[jimgill2002]

I also have a user experiencing this and it began with the TB 38.1.0 update.

Disabling hardware acceleration solved the problem.  

He has two computers, a desktop and a laptop and he uses Mozbackup to copy the Thunderbird profile back and forth between the two machines when he leaves on business with the laptop.  Yet the laptop does not have the problem, so that seems like it can't be something in the profile.

Comment 7

[Loic]

On the machine with the issue, could you type about:support in the location bar and paste the section "graphics", please.

Comment 8

[jimgill2002]

(In reply to Loic from comment #7)
> On the machine with the issue, could you type about:support in the location
> bar and paste the section "graphics", please.

Adapter Description	Intel(R) G41 Express Chipset
Adapter Drivers igdumdx32 igd10umd32
Adapter RAM	Unknown
Asynchronous Pan/Zoom	none
ClearType Parameters	D [ Gamma: 2200 Pixel Structure: R ClearType Level: 100 Enhanced Contrast: 400 ] D [ Gamma: 2200 Pixel Structure: R]
Device ID	0x2e32
DirectWrite Enabled	false (6.2.9200.17292)
Driver Date	2-11-2011
Driver Version	8.15.10.2302
GPU #2 Active	false
GPU Accelerated Windows	0/1 Basic (OMTC)
Subsys ID	043e1028
Vendor ID	0x8086
WebGL Renderer	Google Inc. -- ANGLE (Inter(R) G41 Express Chipset Direct3D9Ex vs_3_0 ps_3_0)
windowLayerManagerRemote	true
AzureCanvasBackend	skia
AzureContentBackend	cairo
AzureFallbackCanvasBackend	cairo
AzureSkiaAccelerated	0

Comment 9

[Loic]

Is it possible to update to the latest version of the Intel(R) G41 Express Chipset drivers and make a test withHWA enabled again?
https://downloadcenter.intel.com/search?keyword=Intel%C2%AE+G41+Express+Chipset

For Vista/Win 7 32 bits, the latest version is 8.15.10.2869 (11/16/2012).

Comment 10

[Rod E]

IMO, the above is a most unreasonable request.  It should not be necessary to change the config of the PC to try to fix something that was broken in an update.

Neither Thunderbird nor Firefox was broken before the latest update, and now users are expected to make changes that might affect other software, to 'fix' a problem that Mozilla created.  This indicates that testing was not done properly before the update was released.

It seems that no-one is really taking this seriously, and the approach to fixing the issues created is deemed to be the responsibility of users.  Not good enough!

Comment 11

[Loic]

Rod, if you think it's a regression in FF38, you can install the tool mozregression to find a regression range.
See the link http://mozilla.github.io/mozregression/ for details.

When it's installed, run the command "mozregression --good-release 37" and stop when you get the pushlog in the console output (copy the link here, no need to bisect).

If Mozregression doesn't work on your machine, you can do that manually by dichotomy by downloading the nightly builds from the FTP.
Each build is stored in the directory "mozilla-central" and you can find standalone version for Windows (as .zip, no install process)
FF37 builds started in January: http://ftp.mozilla.org/pub/mozilla.org/firefox/nightly/2015/01/

Comment 12

[Jason Bassett]

Hello

Disabling hardware acceleration solved the problem as follows: 

For one of my users experiencing this issue I have set gfx.direct2d.disabled true and layers.acceleration.disabled true using the Config Editor.

Regards

Jason

Comment 13

[Loic]

We know that, but we need a regression range to know which recent changes in the implementation of HWA in FF have implied this issue.

Comment 14

[jimgill2002]

(In reply to Loic from comment #9)
> Is it possible to update to the latest version of the Intel(R) G41 Express
> Chipset drivers and make a test withHWA enabled again?
> https://downloadcenter.intel.com/
> search?keyword=Intel%C2%AE+G41+Express+Chipset
> 
> For Vista/Win 7 32 bits, the latest version is 8.15.10.2869 (11/16/2012).

Digging deeper into that version says it's for the mobile chipset version.

Comment 15

[Loic]

Could you use the tool mozregression as I suggested?

Comment 16

[jimgill2002]

(In reply to Loic from comment #15)
> Could you use the tool mozregression as I suggested?

The client who is actually having the issue is out of town so I ran it on a test system so I could see how it worked and be prepared when they return.  

It got to the point where it said it was starting the older version from the temp folder but it never did.  I used your command line (well, adding the -gui because it was the Windows GUI version) so I don't know what's wrong.  I then tried it on a second system and the same thing happened.

Suggestions?

Comment 17

[Wayne Mery (:wsmwk)]

Jason, jimgill, what was your last previous version that worked before updating to 38.1.0?  Was it 31.x?  Or 38.0.1?
If 38.0.1 then we have a much easier job of finding the cause for your case.


(In reply to Rod E from comment #10)
> IMO, the above is a most unreasonable request.  It should not be necessary
> to change the config of the PC to try to fix something that was broken in an
> update.
> 
> Neither Thunderbird nor Firefox was broken before the latest update, and now
> users are expected to make changes that might affect other software, to
> 'fix' a problem that Mozilla created.  This indicates that testing was not
> done properly before the update was released.
>
> It seems that no-one is really taking this seriously, and the approach to
> fixing the issues created is deemed to be the responsibility of users.  Not
> good enough!

I can't speak for Firefox, but for Thunderbird, with hundreds of video cards and a wide variety of drivers for each I'm sure you can appreciate it's not financially possible nor is there enough time for Thunderbird volunteers to test all combination so that you can have free software.  And even when an issue is completely and well reported we often cannot reproduce what a user sees.  We therefore must rely on generous and cooperative users to help us better understand how a problem occurs on their system, and on which date a code change affected the behavior. In short, we don't just dump on users because we like to.

Comment 18

[Loic]

(In reply to jimgill2002 from comment #16)
> (In reply to Loic from comment #15)
> > Could you use the tool mozregression as I suggested?
> 
> The client who is actually having the issue is out of town so I ran it on a
> test system so I could see how it worked and be prepared when they return.  
> 
> It got to the point where it said it was starting the older version from the
> temp folder but it never did.  I used your command line (well, adding the
> -gui because it was the Windows GUI version) so I don't know what's wrong. 
> I then tried it on a second system and the same thing happened.
> 
> Suggestions?

The GUI is not necessary, you can go in the folder C:\mozilla-build and start the console by double clicking on the file "start-l10n.bat" then you run the command.

Comment 19

[Wayne Mery (:wsmwk)]

https://support.mozilla.org/en-US/questions/1068495 also relates to HWA and notifications

Comment 20

[jimgill2002]

(In reply to Wayne Mery (:wsmwk, use Needinfo for questions) from comment #17)
> Jason, jimgill, what was your last previous version that worked before
> updating to 38.1.0?  Was it 31.x?  Or 38.0.1?
> If 38.0.1 then we have a much easier job of finding the cause for your case.

I interrupted his vacation with a text msg and he _thinks_ it was 38.0.1 as he says he always updates when notified.

I'll still plan on getting on his machine and running the regression tool when he returns unless you come up with anything else in the meantime.

Comment 21

[jimgill2002]

(In reply to Loic from comment #18)
> (In reply to jimgill2002 from comment #16)
> > (In reply to Loic from comment #15)
> > > Could you use the tool mozregression as I suggested?
> > 
> > The client who is actually having the issue is out of town so I ran it on a
> > test system so I could see how it worked and be prepared when they return.  
> > 
> > It got to the point where it said it was starting the older version from the
> > temp folder but it never did.  I used your command line (well, adding the
> > -gui because it was the Windows GUI version) so I don't know what's wrong. 
> > I then tried it on a second system and the same thing happened.
> > 
> > Suggestions?
> 
> The GUI is not necessary, you can go in the folder C:\mozilla-build and
> start the console by double clicking on the file "start-l10n.bat" then you
> run the command.

I don't have that folder or that batch file anyplace on the test system I'd using for this.  What part of the process creates that folder, maybe I haven't done the right thing.

Comment 22

[Jeremy Hermet]

Hello.

i'm a french user and i have the same probleme on 250 pc "aures ODYSSE II Evolution".

i have chosse a test pc and i have the g41 windows driver 8.15.10.2302 (dated 11/02/2011).
1 year ago, i have passed all my pc in this driver because the old driver make BSOD.

i can give you a few details if you want.

Comment 23

[Thor]

Here are my test results. 

Grafikk
Adapter-beskrivelse    Intel(R) G41 Express Chipset
Adapter-drivere    igdumd64 igd10umd64 igdumdx32 igd10umd32
Bruk Direct2D    Blokkert av grafikkdriveren. Prøv å oppdatere grafikkdriveren til versjon 8.1500.1000.2202 eller nyere.
Bruk DirectWrite    false (6.2.9200.17461)
Driverdato    7-28-2009
Driverversjon    8.15.10.1855
Enhet-ID    0x2e32
GPU #2 aktiv    false
GPU-akselererte vinduer    1/1 Direct3D 11 WARP (OMTC)
Leverandør-ID    0x8086
RAM    Unknown
Subsys ID    305217aa
WebGL-renderer    Google Inc. -- ANGLE (Intel(R) G41 Express Chipset Direct3D9Ex vs_3_0 ps_3_0)
windowLayerManagerRemote    true
AzureCanvasBackend    skia
AzureContentBackend    cairo
AzureFallbackCanvasBackend    cairo
AzureSkiaAccelerated    0

using this page to test the problem:
https://developer.mozilla.org/en-US/docs/Web/API/notification


with newer intel driver firefox does not crash, but does not display the notification popup.
When I use latest lenovo provided intel driver, that is older (28.07.2009 8.15.10.1855), firefox freezes when running latest ESR version. I did try the mozregression tool, and nightly build 2015-03-18 shows the popup and pc does not freeze.

nightly 2015-06-17 shows the popup and then freezes the pc.

Comment 24

[Loic]

As workaround, you can disable HWA in the options of Firefox.
Anyway it's probably the best solution because Intel(R) G41 Express Chipset is EOLed.

Thor, did you run Mozregression to its end? You should see a pushlog in the console output. Do you have it?

Comment 25

[Thor]

if you look at the link postet longer up here:
https://support.mozilla.org/en-US/questions/1068495
at the end there is one posting similar problem with other graphics. but looks like most of us with the problem have the same card.

anyways
if I remember my testing correctly setting
gfx.direct2d.disabled true 
layers.acceleration.disabled true
did nothing on my problem machine. it crashed all the same. I can do the testing again tomorrow and see if I see something else.

not sure what you mean running mozregression to its end. the pc crashes and you have to use the powerbutton hard.

Comment 26

[Loic]

If you know a driver that doesn't crash the entire computer, you can run again Mozregression, but you need to run it completely until you get the pushlog.

Comment 27

[Loic]

In bug https://bugzilla.mozilla.org/show_bug.cgi?id=1205771#c14 another user is crashing on a website with the same Intel GPU.

Could you test the 2 nightlies he found has a possible regression range (see my link).

Comment 28

[Thor]

I confirm that newest ESR does work when doing following:
 https://support.mozilla.org/en-US/kb/forum-response-disable-hardware-acceleration

Comment 29

[Thor]

gfx.direct2d.disabled true 
layers.acceleration.disabled true

method seems to work as well

Comment 30

[Thor]

not sure what you want me to do.
the links in: https://bugzilla.mozilla.org/show_bug.cgi?id=1205771#c14 
seems like they have been moved. not working today

Comment 31

[Loic]

(In reply to Thor from comment #30)
> not sure what you want me to do.
> the links in: https://bugzilla.mozilla.org/show_bug.cgi?id=1205771#c14 
> seems like they have been moved. not working today

Do you have Mozregression installed on your machine or not?
http://mozilla.github.io/mozregression/

If yes, update to the latest version (cmd.exe > pip install -U mozregression) and find a regression range from FF34 "mozregression --bits=32 --good-release 34" but DONT stop to the 1st builds, we need a changelog!!!

Comment 32

[Thor]

haven found a driver that works okey. to day when I'm testing even the newest driver downloaded from intel creats a crash. but before the crash I see the popup.

Comment 33

[Thor]

it crashes the entire computer every time. so not sure how I can use mozregression tool here.

Comment 34

[Loic]

Do you have an Intel driver that only crashes the browser and not the entire machine?

Comment 35

[Thor]

haven't found one yet. my previous tests done yesterday with the newest intel driver did not crash the browser or machine,but it simply didnt show a popup. today when I reinstalled the newest driver it craches the machine every time.

Comment 36

[Loic]

And ith this newest driver, do you know if previous versions (like FF34) of FF display the notification panel? IMHO, you can use this driver with Mozregression as it doesn't crash the computer.

Comment 37

[Thor]

Had to return the pc to the user. anyways since the pc crashed entirely when I did my last tests with old driver or new driver. I didn't really know how to technically troubleshoot this problem any further.


I did write a little script to automaticly turn off hardware accelleration. 
Can come in handy when you just don't have time to troubleshoot the problem, and you serve alot of computers like I do....
script underneath disables acceleration for :
[ deviceID = "0x2e32", vendorID = "0x8086" ]  // Intel G41 express graphics - http://pcidatabase.com/search.php?device_search_str=0x2E32


I really hope someone that has time to troubleshoot can give Loic the data he needs to pinpoint what goes wrong.



Script:
C:\Program Files (x86)\Mozilla Firefox\defaults\pref\local-settings.js
    pref('general.config.obscure_value', 0);
    pref("general.config.filename", "mozilla.cfg");

C:\Program Files (x86)\Mozilla Firefox\mozilla.cfg
    function deactivateHardwareAcc(blockGraphicCardArray){ 
        // Disable Firefox's Hardware Acceleration, based on graphiccard=>vendorid and cardid
        try {
            // nsIGfxInfo may not be implemented on some platforms.
             var gfxInfo  = Cc["@mozilla.org/gfx/info;1"].getService(Ci.nsIGfxInfo);
             var deviceID = ""; var vendorID = "";
                        
            for (blockGraphicCard of blockGraphicCardArray) {
                deviceID = blockGraphicCard[0]; vendorID = blockGraphicCard[1];
                if(
                    (( gfxInfo["adapterVendorID"]==vendorID) && ( gfxInfo["adapterDeviceID"]==deviceID)) ||
                    ((gfxInfo["adapterVendorID2"]==vendorID) && (gfxInfo["adapterDeviceID2"]==deviceID))
                   ){
                     pref("gfx.direct2d.disabled", true);
                     pref("layers.acceleration.disabled", true);
                }
            }
        }catch (e) {}
    }

    //----------------------------------Firefox's Hardware Acceleration--------------------<start>-------------------------------------------------
        try {
            //Enable hardwareaccellartion 
             pref("gfx.direct2d.disabled", false);
             pref("layers.acceleration.disabled", false);
             
            // Deactivate hardwareaccelleration for spesific graphic cards 
             var blockGraphicCardArray = [ 
                                            [ deviceID = "0x2e32", vendorID = "0x8086" ]  // Intel G41 express graphics - http://pcidatabase.com/search.php?device_search_str=0x2E32
                                        ]
             deactivateHardwareAcc( blockGraphicCardArray );
        }catch (e) {
            //displayError("Hardware Acceleration", e);
        }
    //----------------------------------Firefox's Hardware Acceleration--------------------<slutt>-------------------------------------------------

Comment 38

[Ryan VanderMeulen [:RyanVM]]

Rod, some changes have been made to driver blacklisting and I'm wondering if things have gotten any better for you with more-recent releases?

Comment 39

[Rod E]

To be honest, I've not been game to try a later version since the last update caused so much mayhem.  I use my machine for business and can't afford to lose data because of a bug in mail or browser software.

Comment 40

[Ryan VanderMeulen [:RyanVM]]

Completely understandable, unfortunate as that is :(

If you find some time to come back to it, please do report back here. Dealing with driver issues has been a high priority for the graphics team, so feedback as to how they're doing is very helpful.

Comment 41

[Sotaro Ikeda [:sotaro]]

(In reply to Thor from comment #23)
> GPU-akselererte vinduer    1/1 Direct3D 11 WARP (OMTC)

It seems like a problme of WARP on Win 7. It is disabled by Bug 1179504.

Comment 42

[Ryan VanderMeulen [:RyanVM]]

Hi Rod, just wondering if you've been willing to try a more recently release. According to one of our graphics developers, this may have been fixed in Fx40+.

Comment 43

[Rod E]

Ryan, I'm running the latest version but the notifications are turned off by default.  The problem only showed up on my primary machine that is used for business, and a serious crash isn't something I'm willing to risk if I can help it.

Comment 44

[Ryan VanderMeulen [:RyanVM]]

Given what we know and the lack of any recent reports, I'm closing this out as incomplete then. If you run into this bug again in the future, feel free to comment here and we can reopen the bug.