ATI R600/700 OSS 3D Driver Reaches Gears Milestone

[forum1793]

With mesa pointed to master, it compiles OK.

Slackware-current (both 32 and 64 bit) upgraded xorg-server to 1.6.2 along with other changes.

I'm not sure if those changes alone or in conjunction with radeon changes but something is causing X to lock up a short while upon X starting. I've dropped back to slackware-12.2 with fglrx to type this. So far I don't see anything in the other partitions syslog or dmesg (in that var/log area). I'll have to try dropping/reinstalling the drm/mesa/ati drivers in those and see.

The 64 bit I successfully compiled latest gits pointing to master and installed and it locked probably within a minute and when opening firefox.

The 32 bit was not yet upgraded so it had the code from last night pointing to r6xx-rewrite and was working this AM. So I suspect it might be the xorg-server changes. It locked up after opening terminal in X.

[forum1793]

Recompiled [edit]drm-3d, mesa, ati, and radeonhd with the slackware-current 32 bit and still locking up.

mesa branch points to master, git after "add fine grained syncing support".

Recent slackware package changes are:

/var/log/packages/dri2proto-2.1-i486-1
/var/log/packages/inputproto-1.5.1-noarch-1
/var/log/packages/libFS-1.0.2-i486-1
/var/log/packages/libXaw-1.0.6-i486-1
/var/log/packages/libXt-1.0.6-i486-1
/var/log/packages/mozilla-firefox-3.5.1-i686-1
/var/log/packages/pixman-0.15.16-i486-1
/var/log/packages/util-macros-1.2.2-noarch-1
/var/log/packages/xf86-video-cirrus-1.3.1-i486-1
/var/log/packages/xf86-video-dummy-0.3.2-i486-1
...
/var/log/packages/xf86-video-voodoo-1.2.2-i486-1
/var/log/packages/xorg-server-1.6.2-i486-1
/var/log/packages/xorg-server-xephyr-1.6.2-i486-1
/var/log/packages/xorg-server-xnest-1.6.2-i486-1
/var/log/packages/xorg-server-xvfb-1.6.2-i486-1

[forum1793]

Still locks up on radeon if using Option "AccelMethod" "none".
If using Option "DRI" "off", it no longer locks up.

I'll try radeonhd in a minute.

[nanonyme]

Still locks up on radeon if using Option "AccelMethod" "none".
If using Option "DRI" "off", it no longer locks up.

The thing is, without DRI you're using software rasterizer. You need to disable EXA hooks in the source code for the right effect nowadays, apparently. agd5f gave a simple patch for testing (doesn't fix anything, just enables to check if disabling EXA changes anything) and to show what the necessary hooks are, located here http://www.botchco.com/alex/xorg/r67_no_exa.diff. Note that this is for xf86-video-ati and doesn't probably apply as is for radeonhd.

[forum1793]

Lock-up occurs with radeonhd also. [edit] with accelMethod EXA and DRI on.

So far I think every lock-up occurs after opening window in X, such as Konsole terminal or firefox, and trying to drag window. When I click in body of terminal and type, it does not lock up. Only when clicking on top bar of window and dragging.

Not sure if anyone recognizes this behavior in relation to the changed packages I posted earlier.

Is anyone else using xorg-server 1.6.2 with an hd3200 or similar and getting these lockups?

[bridgman]

agd5f gave a simple patch for testing (doesn't fix anything, just enables to check if disabling EXA changes anything) and to show what the necessary hooks are, located here http://www.botchco.com/alex/xorg/r67_no_exa.diff. Note that this is for xf86-video-ati and doesn't probably apply as is for radeonhd.

http://www.botchco.com/alex/xorg/r67_no_exa.diff

Is anyone else using xorg-server 1.6.2 with an hd3200 or similar and getting these lockups?

Alex is testing on an HD3200 and not seeing lockups. IIRC he is running a fairly recent X server but probably not 1.6.2.

The problem with disabling DRI is that the 3D driver needs it as well. You need the X driver to initialize DRI but not actually use it for acceleration, which is a wierd combination. You can't use shadowfb either because that keeps the master frame buffer in system memory where the GPU can't get to it. It's kind of a useless operating mode except for the specific purpose of debugging potential interactions between 2D and 3D acceleration

[forum1793]

The thing is, without DRI you're using software rasterizer. You need to disable EXA hooks in the source code for the right effect nowadays, apparently. agd5f gave a simple patch for testing (doesn't fix anything, just enables to check if disabling EXA changes anything) and to show what the necessary hooks are, located here http://www.botchco.com/alex/xorg/r67_no_exa.diff. Note that this is for xf86-video-ati and doesn't probably apply as is for radeonhd.

If your commenting about the dri off: I did this to try and isolate radeon from other slackware changes.

[forum1793]

Others at linuxquestion.org seeing this with slackware and even on intel cards. Seems like an xorg-server/slackware issue.

[bridgman]

Interesting... sure looks like it.

[Nille]

mh Ok if im disable EXA everything works an the desktop doesn't freeze. with glxgears the problem with black screen still exists but after one click its works.

some mesa demos work and the terrain demo kill the system but i think this is not a beag deal.



Now i have only one problem left. If im using the "radeon" driver the whole desktop is yellow.