I can reproduce this problem both on an AGP Radeon 9200 PRO card and on a 9200
Mobility (iBook G4), platforms AMD64 and PowerPC, kernel 2.6.17 (both) and
2.6.16 (iBook's) at least (not sure for previous versions, I just bought the
AGP card and the iBook I just used with those two kernels).
When enabling EXA acceleration (either with and without Composite), Xv support
seems not to work as intended. If you open a media player that uses Xv output
(mplayer, xine, vlc, tvtime) it seems to be using Xv (there's the colorkey
painted on the window's content), but if you try to move another window
around, not even touching the xv surface, the image stops and stands still
until you stop moving the window.
Returning on XAA acceleration the problem is gone (although there are still a
few glitches with Xv on MergedFB but that's another story).
The problem seems not reproducible on a 7500 card.
I'm not sure which information is pertinent, but I'll go attaching the
xorg.conf (currently set to use XAA, but there are the commented lines for
EXA). Both systems are Gentoo/Linux installations, using GCC 4.1.1. The
PowerPC using GLIBC 2.3.6, while the AMD64 using GLIBC 2.4. Both have Xorg
7.1, DRI enabled (using 2.6.17 kernel driver).
Created attachment 6129 [details]
Same here. Except no MergedFB, no BackingStore enabled. (because of bug #7202)
and x86. But same gcc 4.1.1 w/ X.org 7.1 My card is actually a R250Lf which I
believe is a Radeon 9200 Mobility or 9000 Mobility
(In reply to comment #0)
> Returning on XAA acceleration the problem is gone (although there are still a
> few glitches with Xv on MergedFB but that's another story).
> The problem seems not reproducible on a 7500 card.
Please attach (as opposed to paste) the full log files for all cases, preferably
the config files as well.
Created attachment 6343 [details]
Sorry for the delay, here is the updated xorg.conf.. this is for the xaa
version: the exa version as the accelmethod comment inverted and decommented
Logs coming, too.
Created attachment 6344 [details]
Xorg.0.log when starting XAA
Created attachment 6345 [details]
Xorg.0.log when starting EXA
Sorry forgot to add those two are for the 9250; if needed I can provide the
ones for the 9200 mobility, too.
I'm actually seeing this, too, but I never noticed it before, and I think 'fail'
is a little harsh. Changing title.
Well it is not only windows moves, it might even be a terminal with scrolling
output, konqueror loading a page, or something like that...
Sounds to me the thing that xv should do is failing here :)
Sounds like just a bug in radeon's exa code. moving to radeon driver.
(In reply to comment #10)
> Sounds like just a bug in radeon's exa code. moving to radeon driver.
I don't agree; I don't see anything in radeon_video.c that would explain the
difference between EXA and XAA. Diego, out of curiosity, does starting the X
server with the -dumbSched command line option make any difference?
FWIW this does not happen on an M7 (aka 7500 Mobility, uses RV200 chip).
(I use gentoo w/ xf86-video-ati and xserver from git.)
I'm told by leio that this problem might be resolved.. I haven't rebooted nor
restarted X for a while now (from before last drivers update at least), so
I'll have to check this after a reboot, which I should do tonight or tomorrow
at the very least.
With latest version of the drivers available in Gentoo, nothing changes, a
part being slower after disabling BackingStore as per bug #7202 now me too.
Will try with -dumbSched now.
With -dumbSched things change. If I move the window slowly, or just for a few
seconds, the image does not freeze/slow down.
If I move a window like crazy on the screen it does though.
Sorry about the phenomenal bug spam, guys. Adding xorg-team@ to the QA contact so bugs don't get lost in future.
For what it's worth, I've just updated to version 6.6.191, I couldn't even reach to test it as enabling EXA now makes the system run as slow as a P133 trying to render Unreal with software rendering..
(In reply to comment #17)
Which version of the server is that? If in doubt please attach the full log file.
I'd consider this bug fixed. I just upgraded to 6.9.0, re-enabled EXA and tried both kaffeine and xine with Xv output and they both work fine.