>=334.21 Redrawing problems in Gnome 3.10/3.12 - GTX 580

Hi,

I reported a bug a while ago where the UI of Gnome would flicker with the newest beta driver at that time.

It turned out to be a bug in clutter.

In the meantime the patch was applyied to clutter on Arch Linux fixing the extreme flickering. Around the same time I started noticing redrawing problems.

Issue:

The UI doesn’t update instantly when UI elements have changed for example a folder and its contents was loaded or a filename edited.

Example:

Several applications don’t redraw properly. This is especially noticable in gedit and nautilus when editing text.

In this example notice how it doesn’t display that the first character of the folder name is deleted despite hitting backspace like a maniac. This is also obvious when typing “This is a test” and the last character only is displayed when clicking inside nautilus to finish the renaming process (probably triggering a new redraw or something).

This will also happens when editing the path in nautilus leading to tons of erros, when trying to correct the path.

This will also give the impression that nautilus takes unusually long to open a folder completely, since the content isn’t drawn as soon as it appears.

Why I’m not sure that it’s again clutters fault:

The PC in the example is using a GTX 580 with the 337.12 driver and Gnome 3.12 from [testing] on Arch Linux.

I have a laptop with a 6150 Go also using Gnome 3.12 from [testing] using the 304.121 driver and it doesn’t have this issue. When using nouveau on this laptop everything is fine aswell.

Packages affected:

The example is using Gnome 3.12 with clutter 1.18 using nvidia 337.12. When downgrading to the latest versions in [extra] (Gnome 3.10 with clutter 1.16.4 and nvidia 334.21) the issue is still there.

This does not happen in non-gnome applications though suggesting that it might be still Gnomes fault.

Reports from other affected users:

https://bbs.archlinux.org/viewtopic.php?id=179900

This seems to be related to 334.21 Repainting issue GeForce 9400M - Linux - NVIDIA Developer Forums .

Reproducing the folder renaming example might not work on the first try. I rebooted and was able to create and rename a folder without a problem. When trying to create a second folder called “This is another test” the redrawing issue started to appear first noticed by the perceived slow reaction when hitting backspace. I repeated this a few times and the bug just seems to kick in at some point after which it is always present.

Unfortunately for some reason I’m not able to downgrade to linux 3.13.8 and nvidia 331.49 from the Arch Rollback Machine. All packages are downgraded properly and the nvidia module is loaded and the old kernel is booted but X fails to start. GDM reports itself as “active” but all that happens is that the monitor powers on a few times, before giving up and dropping me to the terminal. X.Org log doesn’t tell me anything. Same when trying startx with exec gnome-session in .xinitrc. Updating to 337.12 and Linux 3.14.1 let’s me boot to desktop again.

Sorry for the double post, but I think I found the source of this bug.

As someone reported in the Gnome bugreport for this redrawing issue it seems to be linked to the adaptive clocking of the card.

https://bugzilla.gnome.org/show_bug.cgi?id=728464

My performance levels are 782 | 405 | 50 Mhz and when the card is in performance mode there is no way for me to reproduce the bug shown in the video above. When I wait a little while the card clocks down to 405 and quickly after that to 50 Mhz. Trying rename folders is now very slow and it takes close to second for the characters to appear. However I can’t get the exact same condition where nautilus is always one step behind what actually should be displayed. Maybe this is a corner case.

At least the adaptive clocking of the card would explain the behavior I described in my 3rd post.

“I repeated this a few times and the bug just seems to kick in at some point after which it is always present.”

Hi,
It seems that GNOME developers have some patches ready but there is a technical problem preventing them from merging the patches in the mainline codebase.

https://bugzilla.gnome.org/show_bug.cgi?id=728464

They tell us that they sent all the useful information to nVidia in order to correct the problem,
When will the fix ship in a driver release?

I’ve been quiet all this time because I thought a company like nVidia would react quickly, but it seems that it’s taking a bit too long… I’ve had this bug since March and the 334.x driver series and it is really annoying in a working situation.

Is there any ETA for a fix on nVidia’s side of things?

340.24 still suffers from this issue with either gnome-shell or cinnamon.

nvidia-bug-report.log.gz (118 KB)

For me 331.49 was clean and now after 334.21 and all newer, i’ve a random cursor issue interrupt drawing, stop moving and restart at a bad position.
This appears to be the same issue for me.
And I can’t downgrade too.
I’ve no issues with my intel gpu, it’s only with nvidia module.
This bug disapears after upgrade a new kernel or power off the computer / reset bios.
After that, it come randomly, it’s really annoying.
I can’t use nouveau because is not support my gpu.
This happens with no desktop manager, only X running, it’s a nvidia kernel module issue.

I have the exact same problem with GeForce 310M on GNOME Shell (Arch Linux, kernel 3.15.5, nvidia 340.24)

I’m running Arch and just updated to nvidia 340.24 from 337.xx hoping that it would fix this issue, but no such luck. These redraw issues don’t appear when I’m running my dwm, openbox, or kde boxes, and only show up when running gnome, or cinnamon. I can manually set the Promizer option in the nvidia settings to “Prefer Maximum Performance” and the issue goes away. Rather than have to set it each time I log in, I’m using the following command in the .xinitrc file to set Promizer for maximum performance with each boot:

(sleep 5 && DISPLAY=":0.0" nvidia-settings -a [gpu:0]/GPUPowerMizerMode=1) &

So far, that’s the best temporary fix I’ve found. Oh, and my card is the GTX760 in case it matters.

I think we should ask whether nVidia or GNOME if the bugs preventing the merging of GNOME’s patches have been fixed in the 340.24 driver… I don’t know if asking GNOME at the time would be acceptable as we have already been requesting answers from them recently. Personally I don’t want to be annoying towards them, as the work has already been done and they are currently working hard on the next major release.

I keep an eye on this forum and on every new driver releases. You can be sure that whenever it gets fixed the patches will be pushed. You can subscribe to the bug in bugzilla to be notified about it.

Thanks for the quick reply :-)
Well it seems we will have to bear with it as long as nVidia isn’t fixing their driver… quite sad.
Hopefully they’ll fix it in their next release…

Why should nvidia fix something in the driver, when Gnome is broken obviously? Otherwise there wouldn’t be any patches to mutter on the Gnome bugtracker.

BECAUSE THEY INTRODUCED A REGRESSION IN THE 334.XX + BRANCH!!

331.89 Works fine!

The reply is in both my initial message and GNOME’s bugtracker: there is a bug in the driver preventing the merging of the patch. See the details below.

Well that is not really the case. In fact nvidia changed the behaviour of the extensions used by GNOME’s compositing method. There has already been a patch in clutter for correcting the most annoying bugs, but there are still screen update problems with mutter. When patching mutter for using the newly recommended behaviour, compilation fails.

P.S.: @rtcm, correct me if I’m making wrong statements.

Guess what you can change something in your driver and do everything right and expose bugs in other software…

Pretty much what happend here:

As stated just before your post, here the bug in GNOME can actually be fixed but the patch can’t be merged due to a bug in the driver. No one ever said that the bug in clutter was due to nvidia.

So, still no answer from nVidia? :-(

Incidentally, yesterday they told me that they can reproduce the problems and are working on them.

I guess you need to be a bit patient. Software is hard. Software + hardware is even harder ;-)

Yes I think so. Good to know :-) Thanks for the update.