Problem with Catalyst v8.6 and 8.7

[harre]

Hello guys,

I seem to get quite an interesting problem if I'm using the latest (8.7) and previous catalyst (8.6) driver for Linux.

The problem gone if I revert back to version 8.5

The problem I get is that I'm completely unable to run any gl-application.
If I run glxgears and fgl_glxgears I'm just presented with the window but nothing moving inside it.
glxinfo and fglxinfo just hangs and doesn't print out anything

Only thing is that my CPU is topped up on one core.

My dmesg and /var/log/Xorg.0.log doesn't say anything more than usual.

I'm running standard Ubuntu 8.04.1 (x86_64)

Is there anyone that have the same issue and any solution to this problem?

Is there anyone that have some hints on how I could get some more info out of my system to help diagnose the issue?

Poke me if you need any more details

-
Harre

[seanbarman]

Hello guys,

I seem to get quite an interesting problem if I'm using the latest (8.7) and previous catalyst (8.6) driver for Linux.

The problem gone if I revert back to version 8.5

The problem I get is that I'm completely unable to run any gl-application.
If I run glxgears and fgl_glxgears I'm just presented with the window but nothing moving inside it.
glxinfo and fglxinfo just hangs and doesn't print out anything

Only thing is that my CPU is topped up on one core.

My dmesg and /var/log/Xorg.0.log doesn't say anything more than usual.

I'm running standard Ubuntu 8.04.1 (x86_64)

Is there anyone that have the same issue and any solution to this problem?

Is there anyone that have some hints on how I could get some more info out of my system to help diagnose the issue?

Poke me if you need any more details

-
Harre

I have the same problem, I have reverted to 8.5. About the drivers I would'nt bother updating them any more as with each release they have more bugs and are also slower with each release.

I started off with 17000-1800 glxgears its now down to 14000-15000
so I would wait for people here to mess around with the drivers and report back.

I find it confusing that the people who designed these chips don't know how to make them work, may be they should borrow some off intel's or nvidias staff.

[bridgman]

Honestly we are not optimizing for glxgears. A lot of optimization work involves making important things go fast at the expense of unimportant things. If you are seeing performance slowdowns on reasonably modern apps (eg native games) please let us know.

If you are not seeing anything drawn to the screen let's try to figure out what is happening on your system. We do quite a bit of 3d testing on each release before shipping so this is probably something related to a difference between your system and our test systems.

[Kano]

You test something with 3d? Well of course no game using wine it seems...

[harre]

If you are not seeing anything drawn to the screen let's try to figure out what is happening on your system. We do quite a bit of 3d testing on each release before shipping so this is probably something related to a difference between your system and our test systems.

Aye, that's the issue I have, noting is rendered in my window.

Got any tips or pointers where to start looking for differences?

Forgot to mention what graphics card I have.
I have a Radeon HD 2900xt (r600)

[seanbarman]

People don't choose an amd graphics card to become a part of the ATI test program. Some people just expect it to work, the fact is 6 months after release my hd3870 still does not work properly and by the looks of it no one elses does.

Fact driver has more bugs than ever·

fact drivers are getting slower, they are slower in nexiuz as well

Fact most people play games with cedega or wine.

Fact your drivers now don't work with wine or cedega.

who cares about fps if wines not working.

You blaim peoples system configuration here, but 8.7 says it supports hardy heron. I know people have tried a clean install with your drivers and they still dont work, most people are still using 8.5!

[mega_mike]

Also, since the Catalyst release my system (mobility x1400) slows to a halt after 6 hours of usage. Restarting the X server at that point results in a hard lock. Why doesn't ATI address stability in their driver releases?

[Kano]

That's simple to explain. They only work between lunch and tea time then they go home and play with nvidia cards.

[mancunian]

Also, since the Catalyst release my system (mobility x1400) slows to a halt after 6 hours of usage. Restarting the X server at that point results in a hard lock. Why doesn't ATI address stability in their driver releases?

I've got the same problem with a Radeon X1550 (RV505).

[mega_mike]

I've got the same problem with a Radeon X1550 (RV505).

I guess its somewhat comforting to know that I am not the only one, nor is the x1400 the only card that has this issue.

This was never an issue when I ran XGL on top with previous pre-catalyst drivers.

This seems like a pretty big bug if it stretches across ATI product lines. Has ATI ever responded by saying that they are aware of this issue? Looking in the release notes for the latest driver shows no indication of this being a "known issue".

[bridgman]

Is there a Bugzilla ticket open for this ?

[Melcar]

Well, the drivers work fine on my cards. My HD3850's performance on things like Nexuiz is about 95% that of the Windows driver, which is pretty darn good if you ask me. Performance on my laptop's 200M shows about the same delta. Stability is not much of a problem either; only thing that has caused a few crashes/freezes on my main system are Flash and Compiz. In fact, the only real headaches with the driver was when I messed around with different xservers while testing the open source driver on my latop. My systems are nothing special either; Ubuntu/Kubuntu Hardy with the distribution kernel.
I will, however, give you a point on Wine. Something went bad after the 8.5 release. Wine is totally unusable now with the latest driver.

[mega_mike]

Is there a Bugzilla ticket open for this ?

I did a search in bugzilla and it came up with nothing. I then decided to submit the bug. Here is the link:

http://ati.cchtml.com/show_bug.cgi?id=1222

[Melcar]

I did a search in bugzilla and it came up with nothing. I then decided to submit the bug. Here is the link:

http://ati.cchtml.com/show_bug.cgi?id=1222

Did you try taking some (if not all) of the experimental parameters off of your xorg.conf? Also, some of the stuff there you should not have to specify, like the DRI and Composite stuff, not if you're using the recent drivers. Additionally, check your physical system RAM usage; I have noticed with the recent drivers that sometimes some 3D apps. will not terminate properly, therefore "hoging" RAM and leading to performance loss over time.

[bridgman]

I did a search in bugzilla and it came up with nothing. I then decided to submit the bug. Here is the link:

http://ati.cchtml.com/show_bug.cgi?id=1222

Thanks !!! (needed three exclamation marks to meet the 10 character minimum :D)

[mega_mike]

Did you try taking some (if not all) of the experimental parameters off of your xorg.conf? Also, some of the stuff there you should not have to specify, like the DRI and Composite stuff, not if you're using the recent drivers. Additionally, check your physical system RAM usage; I have noticed with the recent drivers that sometimes some 3D apps. will not terminate properly, therefore "hoging" RAM and leading to performance loss over time.

I have actually tried with the device section bare of any options and the result is the same. The non-terminating apps seems to be a more reasonable reason. Top doesn't show anything hogging memory. There could be something latent like you said....how would I check for that anyway?

[cjr2k3]

I don't understand. I've tryed Catalist drivers in 4 diferent systems now. 2 LG laptops (mine and my girlfriend) both with x1400 and they work flawless (since the memory leak from last october). My friends x1950 on a HP desktop and today a x700 from another friend.

In all the cases 8.7 is instaled, WINE running (WoW in the x1400 and x1950, some old games on the x700), no tearing on videos etc etc...

I think it's true that people always complain but never do anything when things are working.

A poll or something would be a good thing to check about the status of fgrlx around phoronix :P

[mega_mike]

I don't understand. I've tryed Catalist drivers in 4 diferent systems now. 2 LG laptops (mine and my girlfriend) both with x1400 and they work flawless (since the memory leak from last october). My friends x1950 on a HP desktop and today a x700 from another friend.

In all the cases 8.7 is instaled, WINE running (WoW in the x1400 and x1950, some old games on the x700), no tearing on videos etc etc...

I think it's true that people always complain but never do anything when things are working.

A poll or something would be a good thing to check about the status of fgrlx around phoronix :P

I guess I didn't explain this properly, performance at boot is flawless, no issues whatsoever. Performance is perfect, no gradual loss or any other issues for 6 hours. The issue is that after roughly 6 hours of usage, BOOM, the system process meter goes high at idle, the graphics get choppy and the system hard locks on an attempted X restart.

[bridgman]

Thanks for the extra detail.

The wierd thing is that a few users have seen this with the open source drivers (matching your description exactly IIRC) but I hadn't heard about it happening on fglrx before.

Jerome (glisse) was looking into the issue with the open drivers -- from what I remember last week he had fixed perhaps 1/2 dozen different lockup scenarios but this one was still unresolved. Anyways, I guess it could still be a memory leak running over some boundary but it seems unlikely we would have memory leaks causing the same behaviour on both open and closed drivers since the code is so different.

[seanbarman]

Hi,


Why not ask that Bridgman guy, believe or not he actually works for ATI, he has nothing to say about the bugs, or offer any help, just quotes release notes all the time.

He has everyone here practically beta testing for him, month after month.
send me your xorg log and rubbish like that. All we get is more bugs.It makes you wonder if they test them at all or leave all to beta testers.

These driver errors are really embarrassing, they are worse than beta drivers.

[d2kx]

Hi,


Why not ask that Bridgman guy, believe or not he actually works for ATI, he has nothing to say about the bugs, or offer any help, just quotes release notes all the time.

He has everyone here practically beta testing for him, month after month.
send me your xorg log and rubbish like that. All we get is more bugs.It makes you wonder if they test them at all or leave all to beta testers.

These driver errors are really embarrassing, they are worse than beta drivers.

You've posted in the wrong thread/forum. I'll help you out:

http://thevistaforums.com/

[Melcar]

You've posted in the wrong thread/forum. I'll help you out:

http://thevistaforums.com/



:p:D

Anyway, they do test the drivers, and work really hard at getting them working. To say otherwise is simply insulting.

[rbmorse]

I think he was trying to be insulting.

[smlbstcbr]

Hi,


Why not ask that Bridgman guy, believe or not he actually works for ATI, he has nothing to say about the bugs, or offer any help, just quotes release notes all the time.

He has everyone here practically beta testing for him, month after month.
send me your xorg log and rubbish like that. All we get is more bugs.It makes you wonder if they test them at all or leave all to beta testers.

These driver errors are really embarrassing, they are worse than beta drivers.

Ouch, very immature, very odd from someone who uses Linux. :confused::rolleyes:
I believe that the problems seen with fglrx are (at my impression) in almost all cases, related with the Linux distro. I've been using fglrx since 7-12 with my Gentoo Linux (and kernel) and fglrx has been doing very well. No major issues.

[shishir]

I have fglrx 8.49.7, or 8.5, installed, and am running openSUSE 11.0, x86_64, kernel 2.6.25.11-0.1-default.

$fglrxinfo
display: :0.0 screen: 0
OpenGL vendor string: ATI Technologies Inc.
OpenGL renderer string: ATI MOBILITY FireGL V5250
OpenGL version string: 2.1.7537 FireGL Release

----------------------------------------------------------
I have AIGLX enabled, and running compiz. Somehow, the video quality on full screen is poor. I am forced to use X11 video output on vlc (all other options crash it), which causes the picture quality to go down when i increase the size of the window.

I am not able to run Xgl, as it shows mesa libs as being used, when I run fglrxinfo. Firefox,with AIGLX for eg, gets stuck, and then jumps to fullscreen mode, instead of being smooth, like with Xgl. But, I cant play videos on Xgl as it becomes too choppy.

http://phoronix.com/forums/showthread.php?t=6585&page=25&p=36189


edit: I tried disabling compiz, but no improvement.

[shishir]

I have fglrx 8.49.7, or 8.5, installed, and am running openSUSE 11.0, x86_64, kernel 2.6.25.11-0.1-default.

$fglrxinfo
display: :0.0 screen: 0
OpenGL vendor string: ATI Technologies Inc.
OpenGL renderer string: ATI MOBILITY FireGL V5250
OpenGL version string: 2.1.7537 FireGL Release

----------------------------------------------------------
I have AIGLX enabled, and running compiz. Somehow, the video quality on full screen is poor. I am forced to use X11 video output on vlc (all other options crash it), which causes the picture quality to go down when i increase the size of the window.

I am not able to run Xgl, as it shows mesa libs as being used, when I run fglrxinfo. Firefox,with AIGLX for eg, gets stuck, and then jumps to fullscreen mode, instead of being smooth, like with Xgl. But, I cant play videos on Xgl as it becomes too choppy.

http://phoronix.com/forums/showthread.php?t=6585&page=25&p=36189


edit: I tried disabling compiz, but no improvement.


Ok, so I did some digging around, and found that in case of Xgl, MESA-PROXy is being used, and fglrx_dri.so is not being loaded.

Xorg.93.log snippet :

(II) Initializing built-in extension XEVIE
(II) Loading sub module "GLcore"
(II) LoadModule: "GLcore"
(II) Loading /usr/lib64/xorg/modules//extensions/libGLcore.so
(II) Module GLcore: vendor="X.Org Foundation"
compiled for 1.4.0.90, module version = 1.0.0
ABI class: X.Org Server Extension, version 0.3
(II) GLX: Initialized MESA-PROXY GL provider for screen 0

[mancunian]

The wierd thing is that a few users have seen this with the open source drivers (matching your description exactly IIRC) but I hadn't heard about it happening on fglrx before.

A couple of years back I (and quite a few other people, too) had the same problem with the nvidia driver for a geforce 2500FX card. I remember that it took nvidia a few months to solve it.So it was a case of deja vu when I experienced it for the first time with my radeon X1550-64.

Anyways, I checked the kdm-logfiles, and the crash does actually leave a trace there:

Fatal server error:
Caught signal 11. Server aborting

Every time the crash occurs I can not restart the xserver, the system also locks up completely. The only way out is a hard-reset.

[shishir]

Ok, so I did some digging around, and found that in case of Xgl, MESA-PROXy is being used, and fglrx_dri.so is not being loaded.

Xorg.93.log snippet :

(II) Initializing built-in extension XEVIE
(II) Loading sub module "GLcore"
(II) LoadModule: "GLcore"
(II) Loading /usr/lib64/xorg/modules//extensions/libGLcore.so
(II) Module GLcore: vendor="X.Org Foundation"
compiled for 1.4.0.90, module version = 1.0.0
ABI class: X.Org Server Extension, version 0.3


Ok, so some more digging :), I changed the Xgl options, removing "/usr/bin/Xgl -accel glx -accel xv:pbuffer", instead of using pbuffer.

fglrxinfo :
display: :0.0 screen: 0
OpenGL vendor string: ATI Technologies Inc.
OpenGL renderer string: ATI MOBILITY FireGL V5250
OpenGL version string: 1.2 (2.1.7769 FireGL Release)

Now, I am not able to fgl_glxgears, or glxgears(it runs with no gears shown), and resizing it crashes the X session. Also, do see MESA-PROXY still being used.

[ktraglin]

fglrxinfo
display: :0.0 screen: 0
OpenGL vendor string: ATI Technologies Inc.
OpenGL renderer string: Radeon X1550 64-bit
OpenGL version string: 2.1.7769 Release

Ubuntu 8.04.1 (64-bit)

Motherboard: Biostar TA770 A2+ (with latest BIOS update).

I had similar problems with the past couple versions of the ATI drivers. What I experienced was an xserver crash, that often would lock-up the entire system. Playing around with several xorg settings only helped temporarily (less than 24 hours). At some point, I realized that these problems would happen mostly when running Miro, which I use pretty much every day. Miro would crash frequently. Using the ATI drivers, when Miro crashed it also took the xserver down with it, and often the machine would just lock-up.

Doing all kinds of searching on the web, I found that my motherboard uses an SB600 Southbridge. Also, using Phoronix Test Suite, I was able to find that my motherboard is NOT fully ACPI compliant. Changes to some kernel boot-options seem to have made my entire system much more stable. I added "acpi=off pci=nomsi" to the end of all the "kernel" lines in /boot/grub/menu.lst. Now, though Miro still crashes just as often, my system doesn't crash with it.

There are still some issues with "flickering" while watching videos with compiz running, but turning compiz off is my temporary remedy for that. I'm hopeful that Ubuntu 8.10, with xorg 7.4 and open-source ATI drivers will address the remaining issues. In the meantime, if anyone knows how to stop this flickering, I would be very much appreciative.

[RealNC]

I've been using fglrx since 7-12 with my Gentoo Linux (and kernel) and fglrx has been doing very well. No major issues.

I'm on Gentoo too. AMD64 (Intel Core 2). What ATI card do you have? With my X1950XT, fglrx is a disaster.

Also, what software do you run? I have problems with Wine, full-screen Flash (YouTube) and full screen Video (mplayer). X-Video is tearing extremely badly. And switching from X to a console sometimes results in the card locking up completely.

[jupbell]

690G gigabyte GA-MA69G-S3H using the IGP

OpenGL renderer string: ATI Radeon X1200 Series
OpenGL version string: 2.1.7769 Release

-- UPDATE --

8.7 is buggy on my setup. Nothing is really new, as many people have said its been buggy.

I see this in dmesg: ( seen it before )

[fglrx] GART Table is not in FRAME_BUFFER range
[fglrx] Reserved FB block: Shared offset:0, size:1000000
[fglrx] Reserved FB block: Unshared offset:7ffb000, size:5000

Is it because I'm using the IGP?

Random and reproducible crashes ( either X crashes or machine reboots as if to save itself from overheating )

when using the following: ( these bugs are not new )

firefox2 + flashplayer

Stopping a DVD using xine ( usually crashed X only )

Games like Urban Terror always crashed. I have not tried it yet with 8.7

$>fgl_glxgears
Using GLX_SGIX_pbuffer
744 frames in 5.0 seconds = 148.800 FPS
1249 frames in 5.0 seconds = 249.800 FPS
1617 frames in 5.0 seconds = 323.400 FPS
1552 frames in 5.0 seconds = 310.400 FPS
1263 frames in 5.0 seconds = 252.600 FPS

I've been watching gkrellm when using firefox2 + flashplayer and there is some point where the CPUs getting maxed.

How much CPU is the IGP allowed to use and/or if something buggy happens in the driver, why is it allowed to max out a CPU?

I'm assuming my mobo has a failsafe that is saving it from damage.

I'm also assuming if I was using an actual video card, that this would be preventable and handled without crashing things. Could I need a BIOS update?

Its odd that the kernel allows any of this to happen, but I guess thats what this means :D

fglrx: module license 'Proprietary. (C) 2002 - ATI Technologies, Starnberg, GERMANY' taints kernel.

Peace y'all

[mega_mike]

8.7 performance on a Mobility Radeon x1400 is pretty good with one exception. After 6 hours of usage my CPU usage spikes to a constant 50% (gnome system monitor and xorg processes seem to be using most of the dual core cpu).

Attempting to restart x at this point results in a hard lock. Memory usage and swap usage are low.

[bridgman]

That sounds like what the open source devs are calling a "soft lock". On WindowsXP we have a feature called VPURecover which detects this condition and more-or-less invisibly recovers from it. Vista has the same capability built into the OS and driver model (I think the events are called TDR's). Linux doesn't have anything like that today but a number of people are working towards it.

[Melcar]

Some form of event logs for fglrx would be nice too, or something that tells us "yes, fglrx caused that massive system failure you just experienced" :D.

[bridgman]

Agreed. Short term priority is simply avoiding the massive system failures but I expect we will want to make the driver more chatty over time to help troubleshoot any remaining issues. So far as long as we can repro the problem in house we can usually fix it; the verbose logs start to become a big help for problems we can't repro in house where we need to rely on details from a customer system.

[leef]

That sounds like what the open source devs are calling a "soft lock". On WindowsXP we have a feature called VPURecover which detects this condition and more-or-less invisibly recovers from it. Vista has the same capability built into the OS and driver model (I think the events are called TDR's). Linux doesn't have anything like that today but a number of people are working towards it.

Does this raise more questions?

Should I understand this to mean there is a known "lock up" condition (generally - since XP, fglrx, and oss all are susceptible) with a known recovery method (at least on the XP side)?

I can imagine that video cards are so complicated that they can't be definitively controlled by programmers, and that implementing a watchdog 'recovery' process is easier than fixing the original problem. I'd prefer if the underlying problem were fixed, rather than adding an extra process and a system hiccup every 6 hours.

*Edit: removed accidental dig.

[forum1793]

8.7 is running like a champion on my setup

The big difference so far is logging in to XFCE is super fast again. Not 100% sure this is due to the new driver.

690G gigabyte GA-MA69G-S3H using the IGP
...
fgl_glrxgears runs fast.
1635 frames in 5.0 seconds = 327.000 FPS

Are you experiencing the problems in mythtv and wine like everyone else?

[mega_mike]

That sounds like what the open source devs are calling a "soft lock". On WindowsXP we have a feature called VPURecover which detects this condition and more-or-less invisibly recovers from it. Vista has the same capability built into the OS and driver model (I think the events are called TDR's). Linux doesn't have anything like that today but a number of people are working towards it.

Is there anything I can do to help identify the source of this cpu eating problem? My Xorg.0.log file shows no errors and I don't see anything out of the ordinary in the rest of the system logs.

I would like to provide more information but I need some clues on what to look for.

If this helps, this my xorg.conf
http://pastebin.com/m43808b77

This is the Xorg.0.log that corresponds to that xorg.conf
http://pastebin.com/m71bd6070

Also, in response to the question about issues with Wine. Wine did corrupt my display at couple versions back, but there have been a couple updates to Wine since then and I have experienced no issues at all (I am using the winehq repository)

*UPDATE*
Ok, I just triggered the system slowdown at 5 hours and 20 minutes of usage. I watched my idle processor usage go from 0 to 50 and then it stays there until I reboot :( . The only thing I noticed new in my Xorg.0.log file is following appended at the end:

SetGrabKeysState - disabled
SetGrabKeysState - enabled
SetClientVersion: 0 9
SetClientVersion: 0 9
SetClientVersion: 0 9
Got unexpected buttonTimer in state 0

[bridgman]

AFAIK every GPU brand sees this problem, which is why Vista took what was previously an ATI-only feature and generalized it across all GPU vendors. It was a much bigger issue in the AGP days, where we were all trying to run reliable systems over an inherently flaky foundation. I'm only guessing that this is the issue in your case, of course.

[Cyberdemon]

Here with my Sapphire Radeon 3650 512Mb DDR3... on Ubuntu 8.04 64 bits, same problems...
I've tested drivers from 8.3 to 8.7, latest ones (8.6 and 8.7)with crappy results (screen corruption issues and low perfomance).
8.3 to 8.5 works so well but after some minutes of playing (wine+wow, or Quake3, War$ow, etc...) freezes the screen and hang the system.

Screen tearing when playing videos makes me sick...

I have no problems in windows xp.
My system specs are:
AMD Athlon(tm) Dual Core Processor 4850e
ASUS M3A78-EH
4gb MDT DDR 800 for AMD
Sapphire Radeon HD 3650

It's very sad to pay for a full AMD computer to deal with this kind of problems...
This is my first and last one...

PS: Sorry for my english, I'm from Spain.