EXA causing X to crash - any workarounds?

Gullible Jones · 2006-12-20 23:08:56

Enabling EXA is still causing X to freeze on start, and I still don't understand why. In xorg.conf's Device section I have been trying this:

        Identifier  "Card0"
        Driver      "via"
#        Option      "EnableAGPDMA"
#        Option      "XAANoOffscreenPixmaps"
        Option      "AccelMethod" "EXA"
#        Option      "EXANoComposite"
        BusID       "PCI:1:0:0"

Tried disabling AGPDMA after I discovered that it screws up AIGLX. Still crashed. Tried disabling composite acceleration from EXA. Still crashed. Tried turning IPv6 back on, in case XOrg 7.1 minds when it's turned off. Didn't get rid of the crash.

I took a look at the log file:

http://pastebin.archlinux.org/496

Nothing very fishy lept out. The one thing that struck me as suspicious was this:

(II) Loading sub module "exa"
(II) LoadModule: "exa"
(II) Loading /usr/lib/xorg/modules/libexa.so
(II) Module exa: vendor="X.Org Foundation"
        compiled for 7.1.1, module version = 2.0.0
        ABI class: X.Org Video Driver, version 1.0
(II) Loading sub module "xaa"
(II) LoadModule: "xaa"
(II) Loading /usr/lib/xorg/modules/libxaa.so
(II) Module xaa: vendor="X.Org Foundation"
        compiled for 7.1.1, module version = 1.2.0
        ABI class: X.Org Video Driver, version 1.0

I would think that XAA shouldn't load if EXA is loaded, but I'm not really sure. To be frank I'm pretty clueless about the inner workings of X.

kth5 · 2006-12-21 00:20:10

First of all, you can force either of the two 2D acceleration mechanisms, not both. Even though the X startup loads both that doesn't mean it uses both or the two conflict, it's just a method of letting it fall back gracefully just in case something goes wrong.

Generally EXA is not yet faster than XAA on most cards. On my ATI 9200 I've seen huge slowdowns, especially when large areas of the screen change (read: non-Xv or OpenGL) like when you scroll a webpage in Firefox/BonEcho. It seems to me that EXA still has a long way to go in most drivers and thus should only be experimented with as it is now.

byte · 2006-12-21 05:46:50

The only thing I recognized when I enabled EXA (after reading 'man radeon') for a test was that it totally messed up font rendering. They got speckled, strokes were missing, really ugly.

deadlydeathcone · 2006-12-21 07:03:09

EXA compositing just recently got enabled in the via drivers, so to get it working you're going to have to use CVS grabs of the openchrome drivers in addition to Xorg 7.1 with a few fixes patched in that haven't made it to head. I went through the same process in Edgy Eft (I don't currently use Arch, I'm just browsing the forums here as Arch looks like a very nice distro), but it's worth it as even though the compositing code is new it's very quick and stable.

Instructions can be found at the Openchrome wiki, or if you dual boot I could provide some debian packages.

Gullible Jones · 2006-12-21 16:18:47

deadlydeathcone wrote:

EXA compositing just recently got enabled in the via drivers, so to get it working you're going to have to use CVS grabs of the openchrome drivers in addition to Xorg 7.1 with a few fixes patched in that haven't made it to head. I went through the same process in Edgy Eft (I don't currently use Arch, I'm just browsing the forums here as Arch looks like a very nice distro), but it's worth it as even though the compositing code is new it's very quick and stable.

All the necessary stuff is supposedly in the main repos. Arch is already using 7.1, the CVS openchrome drivers are in Extra (and crash just like the via and unichrome ones), etc.

(What fixes are you talking about though? There are unpatched bugs in the current version of the server that prevent EXA from working properly?)

byte wrote:

The only thing I recognized when I enabled EXA (after reading 'man radeon') for a test was that it totally messed up font rendering. They got speckled, strokes were missing, really ugly.

That's interesting. I've used EXA before, and didn't find it to provide any noticable increase in the speed of compositing effects, but it definitely didn't mess up fonts like that. If I get it working I guess I'll be able to tell you if it's specific to radeon...

kth5 wrote:

First of all, you can force either of the two 2D acceleration mechanisms, not both. Even though the X startup loads both that doesn't mean it uses both or the two conflict, it's just a method of letting it fall back gracefully just in case something goes wrong.

Okay, thanks for clearing that up.

deadlydeathcone · 2006-12-21 21:12:30

Gullible Jones wrote:

All the necessary stuff is supposedly in the main repos. Arch is already using 7.1, the CVS openchrome drivers are in Extra (and crash just like the via and unichrome ones), etc.
(What fixes are you talking about though? There are unpatched bugs in the current version of the server that prevent EXA from working properly?)

Yes, there are still outstanding bugs without fixes in Xorg 7.1. It's all detailed in the link that I provided above.

Gullible Jones · 2006-12-21 22:09:17

Hmm. Maybe I'll post a bug report about that.

Wait a minute... The one causing the crash is 6818, and it's supposed to be fixed in the most recent version according to that website. Which package is the relevant one for this, xorg-server? And is the fix they refer to in the 1.1.99.x/1.2.99.x (7.2rc1/7.2rc2?) versions?

deadlydeathcone · 2006-12-22 07:39:27

Gullible Jones wrote:

Wait a minute... The one causing the crash is 6818, and it's supposed to be fixed in the most recent version according to that website.

The major one is 7639, which hasn't gotten patched as they're currently arguing over the implementation. The rest are optimizations which while not necessary should really be used as otherwise EXA is really, REALLY slow.

Gullible Jones wrote:

Which package is the relevant one for this, xorg-server? And is the fix they refer to in the 1.1.99.x/1.2.99.x (7.2rc1/7.2rc2?) versions?

After looking through your repository that looks like the right one. I patched against Xorg 7.1 (xorg-server 1.1.1-4 for Arch), so I'm not sure if they'll apply cleanly to anything after that, but it's worth a shot.

Gullible Jones · 2006-12-22 16:12:08

Thanks. You guys think I should file a feature request for patches for these bugs?

byte · 2007-01-03 18:02:14

I tried EXA again after all those Xorg related upgrades lately and it's fine now for me with the radeon driver.

Arch Linux

#1 2006-12-20 23:08:56

EXA causing X to crash - any workarounds?

#2 2006-12-21 00:20:10

Re: EXA causing X to crash - any workarounds?

#3 2006-12-21 05:46:50

Re: EXA causing X to crash - any workarounds?

#4 2006-12-21 07:03:09

Re: EXA causing X to crash - any workarounds?

#5 2006-12-21 16:18:47

Re: EXA causing X to crash - any workarounds?

#6 2006-12-21 21:12:30

Re: EXA causing X to crash - any workarounds?

#7 2006-12-21 22:09:17

Re: EXA causing X to crash - any workarounds?

#8 2006-12-22 07:39:27

Re: EXA causing X to crash - any workarounds?

#9 2006-12-22 16:12:08

Re: EXA causing X to crash - any workarounds?

#10 2007-01-03 18:02:14

Re: EXA causing X to crash - any workarounds?

Board footer