nVidia driver 1.0-96xx causes glitches, artifacts and crashes on Ubuntu Edgy/Feisty

[judebert]

OK, I've got a similar problem. I'm running on Debian unstable (which is essentially the same as Ubuntu), with a dual-boot to Windows XP. I've got an old Athlon 1.7GHz with an nForce motherboard. (No, I'm not l337. I'm a programmer.)

When Debian updated, so did I. I lost my nVidia drivers at the same time. I tried to upgrade, but couldn't, due to that GPL error. I downloaded the kernel source, modified paravirt.c to change EXPORT_SYMBOL_GPL to EXPORT_SYMBOL, copied the standard Debian kernel config, and recompiled. But the standard Debian packages won't install; something about overwriting glx.so. So I'm using the standard nVidia installer instead.

So much for the setup.

When I run the "nv" driver, everything is fine, but slow, and my kids like their games too much to endure that situation.

When I run the 9613 driver from the nVidia download page, I get screen corruption. The splash screen is beautiful, but GDM has chunks of un-redrawn screen. It's definitely a redraw problem: when I log in, it does a complete redraw that looks fine. But once the XFCE4 splash screen goes away, pieces of it are left behind. Dragging a window leaves its edges behind, and sometimes loses the window decorations, too. A screenshot, bug report, and xorg.conf (all tagged "default") are attached.

When I tweak the configuration as described in this forum, there are no detectable changes. (I used /etc/init.d/gdm restart to re-read the conf file and restart the X server.) A bug report and xorg.conf (tagged "tweaked") are attached.

When I run the 7185 driver from the nVidia download page, everything runs as expected. (No bug report attached. I can make one if you want it.)

Windows is running an older driver, but it works fine, too.

All told, it looks like the hardware is not the problem, since it works under some drivers. It appears there's a bug in the new driver, perhaps some feature that my old hardware just doesn't support.

If there's anything else I can provide for your debugging, just let me know.

[judebert]

Is anybody still working on this?

[judebert]

Oh, c'mon. Did I do all that work for nothing? I would certainly appreciate at least an acknowledgment.

[ctheo]

you should give this workaround a try:
http://www.nvnews.net/vbulletin/showthread.php?t=91641
(I know you don't have exactly the same issue, but it won't hurt you either to try)

[judebert]

Thanks; I'll check it out and see how it works for me.

[Licaon]

looks like 9639 does not fix these problems for me
neither is this patch

Does the problem persist if you extract the .run package with the -x command line option, change to the newly created directory, edit the file usr/src/nv/nv-vm.c to make the change below and then install the driver with ./nvidia-installer?

original:
#if defined(KERNEL_2_4) || defined(NV_CPA_DF_LIST_BUG)
#define NV_CPA_NEEDS_FLUSHING 1
#endif
modified:
#define NV_CPA_NEEDS_FLUSHING 1

[anandvp]

I have exactly the same problem as judebert. (I came across this thread searching for "NVIDIA(0): Error recovery failed"). I have a ASUS A7N266-VM motherboard with an Athlon 1.7GHz CPU. I am running Fedora 7 with the nvidia 9639 driver. I have tried most of the tips suggested earlier but none work (recompile with NV_CPA_NEEDS_FLUSHING, Disable Composite Extension, set various options of the nvidia driver in the Device section of xorg.conf, remove from /etc/modprobe.d/blacklist, acpi=off noapic kernel parameters) - the windows are not refreshed properly. The only way I can use the nvidia driver and have the screen drawn properly is if I revert to the 7185 version.

My system was working properly with Redhat 9 and still works well with Windows XP. I am attaching the bug-report log file. It also seems that the other people with this problem are using the ASUS A7N266-VM motherboard (users judebert, IntuitiveNipple, drbisio).

[IntuitiveNipple]

As I said in my comment above, I solved the issue by adjusting an entry in the module blacklist and the 8776 build.

[JulianoZaban]

Hi MartinE!

My GForce2 MX is doing the same thing! Your screenshot represents exactly my screen behaviour... I've tried many things but nothing works. Do you have any advice for me? I'm really lost in this problem! Please, any help will be apreciated...
Thanks a lot.

[DebianCoder]

I confirm this bug. It is not an Ubuntu problem. It is not a BIOS problem. It is a GLX 1.0.9639 driver bug. Something went horribly wrong from 1.0.8776 to 1.0.9639.

I compile kernel.org and nVidia from source. I am using the right versions of everything. I do not use your shell script, I use Debian packages.

kernel

2.6.20.15 (my build, Intel x86 32-bit)

GPU

GeForce3 rev 163

driver

1.0.9639 via Debian's packages,

nvidia-kernel-legacy-96xx-source

nvidia-glx-legacy-96xx

xserver

1.3.0.0-dfsg-7 (stock Debian)

VideoBIOS

03.20.00.10.02

Debian's bugzilla shows assorted problems (all driver versions).
http://bugs.debian.org/cgi-bin/pkgreport.cgi?pkg=nvidia-glx

I do not have any of the "file missing" or "typo" problems. Those relate to your shell script. I use Debian's build system. Everything is in order. I build my own kernel module for my kernel. Your closed binary blob is the problem.

To evade the bug, I disable GLX in xorg.conf. I mean literally the "Disable" keyword. Then at least KDE/Gnome work, just without GLX. RenderAccel is also TRUE and seems to work.

Similar bugs reported for 100.14.xx suggest common GLX bugs in 100.14.xx and 1.0.9639. However I only confirm 1.0.9639.

Previously I used 1.0.8776 in kernel 2.6.18.x without problems. I've used this board in Windows and Linux. It always worked before. This is a driver bug. It is not a hardware bug or a firmware/BIOS bug.

There is no obvious place on nVidia.com to get the latest nVidia BIOS. There are only driver downloads. Please offer a section for BIOS downloads. You always ask if people have the latest BIOS, but don't put BIOS downloads on your site. If you mean motherboard BIOS, that's really a stretch. You're looking in the wrong place. The first place to look is the most recent code changes.

Sometimes 1.0.9639 corrupted the X log file during the crash. When it did not, the last line was loading GLX, so that's when the crash happens. It's either loading GLX or initializing GLX.

It hangs at the logo screen which is shown all crunched up, because mode timings are wrong. If I use "NoLogo" in xorg.conf, it just hangs with a blank screen. In any case the machine needs a hard reset.

My monitor is flat panel BUT I use KVMs and VGA cables, so it is a "CRT" callout for the driver. The KVM means I *do not* want EDID. I hard-coded monitor specs into xorg.conf and don't want your driver second-guessing them across a KVM. (Unfortunately your driver does not respect my DPI, another problem.)

I will not spend more time sending logs, filling out forms, etc. Sorry. This buggy driver has cost us a week.

Anyway, you have enough data to reproduce the bug from me and everyone else. I have read all the other reports.

You need more help? OK. Then send 6-7 free boards to Debian, Ubuntu, and RedHat, 1-2 for each driver version you offer. These three distributions are the basis for most others. Of course Ubuntu itself derives from Debian.

[stuartlevy]

Curious, DebianCoder, why does your X config include
Option "NoPowerConnectorCheck" "True"
? Hope you don't actually need that -- if you have an unpowered
connector, no wonder the card flakes out if something tries to use OpenGL!

[DebianCoder]

Originally Posted by stuartlevy

Curious, DebianCoder, why does your X config include
Option "NoPowerConnectorCheck" "True"
? Hope you don't actually need that -- if you have an unpowered
connector, no wonder the card flakes out if something tries to use OpenGL!

Look, in debugging, you turn things off to isolate. I need no check because I know I don't.

Please do not distract the nVidia staff from understanding that this driver is buggy as hell. That was half of my statement, OK? nVidia needs to stop asking sidetrack questions like yours.