Increase stability of ati driver

[bojojo2020]

To increase the stability of the driver please add pci=nomsi in the kernel boot flag

The msi (message signaled interrupt) is a kernel experimental feature and should not be activate by default.

The nvidia driver doesn't use this feature except if you clearly enable it on the driver.

With this feature i can put my computer in sleep mode (S3) without any bug

[bojojo2020]

I can now type killall kdm in console and restart kdm without any problem

I have no bug when computer stop

I use the fglrx 8.721 for ubuntu

NBon't forget to verify if you have
Option "DPMS" "true"
in your xorg.conf file (section screen) or you will have a segmentation fault when starting xorg

[seebs]

This seems a bit odd -- the nomsi thing was being recommended in 2007, but normally kernel tweakery from that far back is no longer relevant.

Any concrete explanation of why this feature, which has been enabled by default even in stable distributions for more than a year, ought to be relevant?

What specific symptoms did you have that this option fixed?

Also, you have not been clear as to whether DPMS must be on to avoid segmentation faults, or turning it on will cause the segementation faults.

[bojojo2020]

The nvidia driver do not use the msi feature by default if you activate this feature (the bug are the same with the ati driver)
-you can't put your computer in sleep mode and restore your desktop all the time
-sometimes with ati driver xorg not die correctly when stopping
-you can always sweeth beetween VT1 and VT7 without any problem
-the computer do not freeze randomly.

For the dpms part:

If you dont put Option "DPMS" "true" or Option "DPMS" "false" in your xorg.conf; Xorg will simply don't start with 8.721 driver

NB: The 8.721 driver (with xorg-server 1.7) is the first ati driver where you can kill xorg and restart xorg without problem.

[stancil]

dmesg | grep -i msi
[ 0.790032] pcieport 0000:00:02.0: irq 24 for MSI/MSI-X
[ 0.790220] pcieport 0000:00:0a.0: irq 25 for MSI/MSI-X
[ 0.814207] r8169 0000:02:00.0: irq 26 for MSI/MSI-X
[ 1.198387] HDA Intel 0000:01:00.1: irq 27 for MSI/MSI-X
[ 22.548083] fglrx_pci 0000:01:00.0: irq 28 for MSI/MSI-X

Can?t confirm on Debian Squeeze with 10.3pre Kernel 2.6.33.2.
Everything is working as expected including VT switching and sleep to ram (tested multiple times in a row). The only thing not working is /etc/init.d/gdm restart but im not using that anyway. My GPU is hd 4670.

[seebs]

I don't use sleep mode, I don't exit X, I don't switch terminals. But the random freezing has happened to me.

I'm planning to try my ATI card again when catalyst 10.4 shows up, so I guess when that happens, I should also add the nomsi thing.

I'm still pretty surprised that a feature that's been standard in kernels for two or three years would be considered "experimental".

[kernelOfTruth]

NB: The 8.721 driver (with xorg-server 1.7) is the first ati driver where you can kill xorg and restart xorg without problem.

this also works with 10.1 for me (the others wouldn't work correctly -> segfaults with opengl stuff on Gentoo; Ubuntu 10.04 works fine)

so it's an continual improvement process

[bojojo2020]

this also works with 10.1 for me (the others wouldn't work correctly -> segfaults with opengl stuff on Gentoo; Ubuntu 10.04 works fine)

so it's an continual improvement process

I use sabayonlinux (limbo repository) wich is based on gentoo and there is no problem ; don't forget to use these commnds lines:
eselect opengl set ati && lafilefixer --justfixit

and verify if you have the dpms option in your xorg.conf.

[bojojo2020]

Can?t confirm on Debian Squeeze with 10.3pre Kernel 2.6.33.2.
Everything is working as expected including VT switching and sleep to ram (tested multiple times in a row). The only thing not working is /etc/init.d/gdm restart but im not using that anyway. My GPU is hd 4670.

for me /etc/init.d/gdm restart works. And
killall gdm && sleep 5 && gdm
works with the 8.721 driver . That's means you don't have memory bug.