Ubuntu 10.04 Gets A New Catalyst Pre-Release

[Tares]

It seems there are several fixes for 3D/OpenGL included. Try if your problems still exist.

I'm playing a lot of hon lately so don't have too much time to testing. But my framerate dropped quite a lot when I tried 10.3 and 10.4alpha1 ;(

[Cuppa-Chino]

Oh good

you bet my wife's laptop has just been confirmed as in production, this new continues to make linux viable in WAF terms!

[d2kx]

Just done more 3D testing. Nexuiz itself works great, but when I change settings like effects or the resolution, which then tries to restart Nexuiz, it crashes with this driver.

[smitty3268]

Ah. Thanks for sharing that John.

As the Dutch saying goes, "Nu komt de aap uit de mouw"!

So if it's not D2D, what's it going to be called? And my other question remains: what exactly will it accelerate? Will it be a low-level "engine" replacement for the Linux Catalyst EXA code?

It's a replacement for the Catalyst XAA code, they decided to use this D2D api in their drivers that is shared across platforms instead of implementing EXA. And i haven't heard of any plan to rename it, even though it sometimes causes confusion. It's really an internal driver api, though, so completely different than MS's D2D and there's no real reason users would ever need to hear about it once it's working. Just like no one really worries about what XAA is called right now in the Catalyst drivers.

[Shining Arcanine]

3D stuff is still segfaulting:

glxinfo, glxgears, compiz, xscreensaver-gl stuff, etc.

exception and only working program (probably due to 32bit and bundled own libs) is googleearth what on earth ?

system: ~amd64, gcc 4.4.3 currently moving from fPIE & fstack-protector-all to vanilla compiler, hopefully that makes a change,
around 60% is already re-compiled

card: 5850

distribution: Gentoo

for more info: http://forums.gentoo.org/viewtopic-t...ighlight-.html

many thanks in advance for any comment, fix, help, etc.


please make 10.4 usable again !

10.1 was the last usable driver for me - all others: 10.2, 10.3 lead to segfaulting apps

I believe segmentation faults occur when programs try to access stuff in memory that was not allocated to them. This can often occur if the stack is corrupted because something overwrites the return pointer in a previous stack frame, which could be caused by calling conventions mismatches in the assembly code.

I am not familiar with ATI's driver or how it interacts with the kernel (or the rest of the system), but I think we can conclude that if there was some sort of interface mismatch between the driver and the kernel, you would be getting kernel panics, rather than segmentation faults, so likely some sort of system call is being made that invokes code in the driver which expects things to have been done differently by the callee. It is possible that the callee passed a pointer to a struct on the call stack that is supposed to be a certain size and it is not big enough, causing the return pointer in the stack frame to be overwritten by the driver. Then after the kernel returned control to userland, all heck broke loose because the return pointer was corrupted.

In order to get things working with the driver, you probably will need to look at the software (and its dependencies) involved, set your CFLAGS, CXXFLAGS, CPPFLAGS and LDDFLAGS to match those used by the Ubuntu 10.04 package maintainers, emerge the exact versions that Ubuntu 10.04 uses and then if things work (in theory they should), start going back to the more optimal configuration piece by piece until things break. When you have identified the culprit, assuming it is a single piece of software, it should be possible to delve deeper into what is causing the issue by examining what changed between software versions.

[kernelOfTruth]

I believe segmentation faults occur when programs try to access stuff in memory that was not allocated to them. This can often occur if the stack is corrupted because something overwrites the return pointer in a previous stack frame, which could be caused by calling conventions mismatches in the assembly code.

I am not familiar with ATI's driver or how it interacts with the kernel (or the rest of the system), but I think we can conclude that if there was some sort of interface mismatch between the driver and the kernel, you would be getting kernel panics, rather than segmentation faults, so likely some sort of system call is being made that invokes code in the driver which expects things to have been done differently by the callee. It is possible that the callee passed a pointer to a struct on the call stack that is supposed to be a certain size and it is not big enough, causing the return pointer in the stack frame to be overwritten by the driver. Then after the kernel returned control to userland, all heck broke loose because the return pointer was corrupted.

In order to get things working with the driver, you probably will need to look at the software (and its dependencies) involved, set your CFLAGS, CXXFLAGS, CPPFLAGS and LDDFLAGS to match those used by the Ubuntu 10.04 package maintainers, emerge the exact versions that Ubuntu 10.04 uses and then if things work (in theory they should), start going back to the more optimal configuration piece by piece until things break. When you have identified the culprit, assuming it is a single piece of software, it should be possible to delve deeper into what is causing the issue by examining what changed between software versions.

thanks Shining Arcanine !

I'll do something like that but not as extensive (as if I had the time to do it several times and narrow it down ):

disabling hardened flags, switching kernel 2.6.32*, xorg-server, etc.


if it doesn't work after that I'll stick to 10.1 until middle of 2.6.35 and probably will be able to ditch fglrx (if it still doesn't work) for good when the opensource evergreen support has enough basic features (powermanagement & some moderate 3D acceleration)

[kernelOfTruth]

well, the more I think about this & eliminate potential problems and it still not working - the more I believe that it's a regression that got introduced with >10.2 so far

I'm sure I'm not the only one having this problem with segfaulting GL apps & compositing apps with error 7

if it worked 9.11, 9.12, 10.1 why shouldn't it suddenly stop working with the following drivers ?

the only change I made was some updates to a few packages that are NOT relevant for graphical stuff (simply keeping the system up-to-date)

and 10.1 is STILL working so it must be a problem with fglrx / catalyst

[kernelOfTruth]

does one still need to append nopat to the kernel-line with catalyst 10.2 - 10.4 ?

I can see no adverse affects when leaving it out



also: will the official 10.4 catalyst still include support for xorg-server 1.6* ?

thanks !

[grigi]

PAT has to still be disabled for 10.3 on my system, but since you are using Gentoo, the installer will barf if you try to install the driver through portage, and PAT is not disabled in the kernel config.

So the nopat probably does nothing for you, since PAT suport is probably not built in your kernel.

[TheWretched]

Works well on my Vaio E too...

also i have no problems with OGL software (well, I only tried Tux Racer, BillardGL and some others) either. Only activating Compiz freezes the settings window (but doesn't crash X).

I am happy for now