ATI R600/700 3D Support In Fedora 12

[phoronix]

Phoronix: ATI R600/700 3D Support In Fedora 12

Fedora 12 provides "out of the box" support for kernel mode-setting with ATI R600/700 series graphics hardware, but it does not provide 3D acceleration by default. However, Red Hat's X developers have made it very easy to enable this 3D support for the ATI Radeon HD 2000, 3000, and 4000 series hardware by just installing a special Mesa package from yum. In this article we are taking a quick look at where the R600/700 3D support is at in Fedora 12.

http://www.phoronix.com/vr.php?view=14298

[m4rgin4l]

Excellent. Also, Catalyst 9.10 finally works on Fedora 11.

[dosenpfand]

Doesnt work well for me .

Made a fresh install of the 64bit beta updated it and installed the experimental package for my 2900pro.

Now it mostly hard locked in gdm or sometimes gives me about 10min before X frezes

[etymxris]

The open source radeon driver does not correctly display on Mitsubishi's new line of DLPs. It is horribly overscanned. The TV allows "1080p reduce" but then it's horribly underscanned. On Windows or with the proprietary drivers the image is slightly underscanned (without the 1080p reduce). The Catalyst Control Center then allows correcting for underscan. I have not been able to find any overscan/underscan correction for radeonhd.

[salva84]

So... does it means that it will support ati 5800 series?? Im very interested on this... beacause the open source drivers are faster than fglx in 2d acceleration. 3d acceleration doesnt care me.

[Michael]

So... does it means that it will support ati 5800 series?? Im very interested on this... beacause the open source drivers are faster than fglx in 2d acceleration. 3d acceleration doesnt care me.

Radeon HD 5800 series is R800 / Evergreen series...

[cbxbiker61]

Using XOrg Server 1.6.4, Mesa 7.6, radeon_hd 1.8, kernel 2.6.32-rc5 on a Toshiba L305D at 1280x800 is working amazingly well for me.

IoQuake3, OpenArena, WorldOfPadman, Nexuiz, Tremulous, BZFlag, SuperTuxKart are all working with no unusual artifacting. Frame rates are approximately 60 for IOQ3, OA and WOP.

Alien Arena works, but frame rates are pretty low, which is not unexpected for Alien Arena.

I build all of my programs from source, so my versions are probably not what you'd find on a standard distribution.

About the only thing that I'm hoping for is OpenGL 2.X and 3.X support.

[cbxbiker61]

I forgot to mention Urban Terror is working just fine for me also. Text displays just fine. I'm running it as a mod from my IoQuake build.:D

[droidhacker]

So... does it means that it will support ati 5800 series?? Im very interested on this... beacause the open source drivers are faster than fglx in 2d acceleration. 3d acceleration doesnt care me.

Sure, you just need to wait for support to be added. R800 is very similar to R600/700. This won't be anywhere near as long as the wait for R600 was.

[droidhacker]

The open source radeon driver does not correctly display on Mitsubishi's new line of DLPs. It is horribly overscanned. The TV allows "1080p reduce" but then it's horribly underscanned. On Windows or with the proprietary drivers the image is slightly underscanned (without the 1080p reduce). The Catalyst Control Center then allows correcting for underscan. I have not been able to find any overscan/underscan correction for radeonhd.
Typical with DLP's. Might be time to upgrade to LCD.

[etymxris]

Typical with DLP's. Might be time to upgrade to LCD.

I got the DLP because it is 73" (80" is available too). It'll be a while before LCD, LED, or plasma can hit that size at decent prices.

[acreda]

I have to say I will be looking forward to the benchmarks when .32 is released properly, I recon the advances in 33 will be a big step up..........

Imagine Ati accelatration out of the box for all distro...... reason why I brough an ATI card to show apprication (bargin as well :))

[lordmozilla]

you guys are a bit impatient.

2.6.32 isn't even out yet and this is redhat's beta backport, can't expect miracles. But this does mean by ubuntu 10.10 there should be good support. (10.04 from what I heard will have 2.6.31 which is a shame but what can you do)

But it's a nice preview article. There has been an howto on arch forums about how this works, but I'm currently on the openCL beta drivers which work well on 2.6.31.

[d2kx]

you guys are a bit impatient.

2.6.32 isn't even out yet and this is redhat's beta backport, can't expect miracles. But this does mean by ubuntu 10.10 there should be good support. (10.04 from what I heard will have 2.6.31 which is a shame but what can you do)

Ubuntu 10.04 will probably get Linux 2.6.33 or 2.6.34.

[lordmozilla]

Ubuntu 10.04 will probably get Linux 2.6.33 or 2.6.34.

my bad seems you're correct.

[Gaius]

Ok 3D is good...but...what about simple things like 2D acceleration available for EVERY users connected ... for now, only the first user to login has direct acceleration....

[bugmenot]

I didn't even needed to restart the system after installing the package. AMD64 and radeonhd3200/3850.

[MichaelPalin]

Is this Mesa build exclusive to Fedora or will be available for other distributions too? I'm one of those GNU/Linux users that wouldn't know what to do outside of Ubuntu, :D

[Louise]

@Phoronix: Did you test Apple trailer playback?

That would be really interesting to hear about as well.

[JeanPaul145]

you guys are a bit impatient.

2.6.32 isn't even out yet and this is redhat's beta backport, can't expect miracles. But this does mean by ubuntu 10.10 there should be good support. (10.04 from what I heard will have 2.6.31 which is a shame but what can you do)

But it's a nice preview article. There has been an howto on arch forums about how this works, but I'm currently on the openCL beta drivers which work well on 2.6.31.

Could you maybe provide the source for that?


Ubuntu 10.04 will probably get Linux 2.6.33 or 2.6.34.

That's exactly what I figured as well, and it wouldn't make any sense to stick with 2.6.31 for Lucid. It's not like the kernel has had major stability issues for final releases AFAIK.

[mattdm]

Freezes here too with Radeon 4850 :( Can't even login ---> unusable

This is a known bug. Boot with "pcie_aspm=off" on the kernel command line (edit /etc/grub.conf) to work around the problem.

https://bugzilla.redhat.com/show_bug.cgi?id=517625

[mattdm]

Freezes here too with Radeon 4850 :( Can't even login ---> unusable

This is a known bug. Boot with "pcie_aspm=off" on the kernel command line (edit /etc/grub.conf) to work around the problem.

For details, look at Red Hat bugzilla #517625. (I can't post the direct link yet since this is a new account. Moderator, go ahead and delete this posting once the other one goes through. Thanks!)

[frantaylor]

Some apps I'd like to see tested on this setup:

- Google Earth
- VMware Workstation with a Windows guest and DirectX support enabled
- Java3D demos

I currently don't have an ATI card but I will be interested in getting one if these apps work.

[d2kx]

Is this Mesa build exclusive to Fedora or will be available for other distributions too? I'm one of those GNU/Linux users that wouldn't know what to do outside of Ubuntu, :D

Ubuntu mainline kernel-ppa: http://kernel.ubuntu.com/~kernel-ppa/mainline/
Ubuntu xorg-edgers ppa: https://launchpad.net/~xorg-edgers

I am not commenting on these because you should only mess with them if you know what you're doing! It's not exactly hard though.

Using Ubuntu Karmic and xorg-edgers, I am able to run Nexuiz just fine and without visual artifacts/corruptions on my HD 4850. Compiz/KWin4 run amazingly well, too. I am really surprised about how good the Mesa R600 3D driver already runs.

[jbrown96]

I know this article is about 3d, but how is 2d? I have a Radeon HD3200, and I watch a lot of HD videos (mostly <1080p). How would this perform on my system? I'd rather use the open source; would opengl output work well like the catalyst drivers?

[GoremanX]

Using XOrg Server 1.6.4, Mesa 7.6, radeon_hd 1.8, kernel 2.6.32-rc5 on a Toshiba L305D at 1280x800 is working amazingly well for me.
...
I build all of my programs from source, so my versions are probably not what you'd find on a standard distribution.


I thought the latest version of the radeonhd driver was 1.3? It doesn't matter if you compile from source, how did you end up with a 1.8? Seems like an odd typo to make, the 8 is nowhere near the 3.

[Louise]

I know this article is about 3d, but how is 2d? I have a Radeon HD3200, and I watch a lot of HD videos (mostly <1080p). How would this perform on my system? I'd rather use the open source; would opengl output work well like the catalyst drivers?

Yes, the Apple trailers would be a good benchmark, as they come in 3 resolutions.

Of course the benchmark would have to be made for an action move, with lots of FPS, and a drama with close to constant FPS.

[airlied]

Ok 3D is good...but...what about simple things like 2D acceleration available for EVERY users connected ... for now, only the first user to login has direct acceleration....

Fedora 12 does that as well, kernel modesetting enables it.

Dave.

[cbxbiker61]

I thought the latest version of the radeonhd driver was 1.3? It doesn't matter if you compile from source, how did you end up with a 1.8? Seems like an odd typo to make, the 8 is nowhere near the 3.

Hehe, it wasn't a typo it was a brain-fart. 1.3.0 is the correct version.

[RahulSundaram]

Is this Mesa build exclusive to Fedora or will be available for other distributions too? I'm one of those GNU/Linux users that wouldn't know what to do outside of Ubuntu, :D

It is not that hard to get familiar with Fedora enough to try it out. The Beta Live cd or the nightly builds at http://alt.fedoraproject.org/pub/alt/nightly-composes/ are available for easily trying the latest stuff.

Otherwise wait for another six months, I am sure other distributions will catch up. Fedora doesn't hoard patches and all these changes are merged upstream which will eventually trickle in everywhere.

[Gaius]

Fedora 12 does that as well, kernel modesetting enables it.

Dave.

Thanks ;) So as I'm using Ubuntu, do you think this howto >> https://wiki.ubuntu.com/X/RadeonKMS could solve this problem ?

[Yfrwlf]

I got the DLP because it is 73" (80" is available too). It'll be a while before LCD, LED, or plasma can hit that size at decent prices.

I have the exact same TV you do and have the same problems with overscan/underscan. So far both ATI and Nvidia cards do the same thing on that TV, and I've simply chalked it up to a crappy TV. I've seen a lot of newer TVs which display computer outputs perfectly, both LCD and LED TVs. I'm just waiting until they're low enough in price and big enough then I'll upgrade, but yes it is stupid that these DLP TVs don't display computer output correctly and can function as a big monitor. We usually leave it overscanning unless there's a real reason to see the boarders, and then change it to reduced mode.

First thing I'm doing is bringing a laptop with me for when I make my next TV purchase to make sure this will no longer be an issue.

[AdamW]

There appears to be more than one bug manifesting as the system hanging shortly after login. If pcie_aspm=off as a kernel parameter works around the problem for you, that is indeed:

https://bugzilla.redhat.com/show_bug.cgi?id=517625

if that parameter doesn't seem to have any effect, take a look at:

https://bugzilla.redhat.com/show_bug.cgi?id=531147

or:

https://bugzilla.redhat.com/show_bug.cgi?id=528593

those are the ones I'm currently tracking. 'nomodeset' may suffice as a workaround, try that.

[iVistux]

Are there any experiences with Radeon3D and KDE4 (Kwin)??

Btw. if someone successfully tests games, drop it here: http://www.x.org/wiki/RadeonProgram

[brada]

Hello, can anybody who get it work post here the output of glxgears -info?
(GL_EXTENSIONS can be wiped out).

week ago I have tried to get my Radeon HD 3650 on gentoo work. i took latest mesa, kernel drivers etc from live trees and after linking everything together my glxinfo output still showed mesa software rasterizer. is this supposed to be so with the open source drivers?

i was not able to get my ati to reasonable work either with opensource or proprietary drivers.

Proprietary: perfect 3d, hard problems with 2d - similar to reported problems with KDE4 (but e16 here and without compositing) - the movement of windows and between desktop has unbearable delays - waiting for 2s for logout window etc... i tried the patch for xorg-server which was suppposed to help KDE4 but no success here. (btw is there any chance to get support from ati people on such issues?)

Opensource: perfect 2d. once i enable DRI, e16 pager stops working, instead of windows snapshots only black squares visible. i'm not sure i got 3D work, all xorg logs report success, but glxinfo still reports the software randerizer (see my question in the begining), but if yes then compared to proprietary drivers we are still nowhere. eg. googleearth still unusably slow - this seems to be in accordance to what devs themselves state on r600/r700 drivers wiki page.

[soulrebel]

What about power usage? The free driver can slow frequencies down when in desktop usage?

[rmartinez]

Hello, just to share my happiness! My computer (M51Ta) has an hybrid graphic configuration:

01:05.0 VGA compatible controller: ATI Technologies Inc RS780M/RS780MN [Radeon HD 3200 Graphics]
02:00.0 VGA compatible controller: ATI Technologies Inc Mobility Radeon HD 3650

I have been trying many things with Ubuntu to make it work with 3D acceleration with no good results (last fglrx, radeonhd...).
And finally, thanks to Kano's suggestion in other post (http://phoronix.com/forums/showthread.php?p=97427#post97427) I installed fedora 12 beta nightly-compose of 26/10/2009 and then installed mesa experimental drivers:
su
yum install mesa-dri-drivers-experimental


It works great!!! And I use lots of 3D stuff! Although I didn't have time to workaround a lot, so far GREAT advance!

[blindfrog]

This is a known bug. Boot with "pcie_aspm=off" on the kernel command line (edit /etc/grub.conf) to work around the problem.

For details, look at Red Hat bugzilla #517625. (I can't post the direct link yet since this is a new account. Moderator, go ahead and delete this posting once the other one goes through. Thanks!)

Heh the freezing was caused by pulseaudio+audigy combination :D I finally removed my unused audigy card and everything started to work... One hell of a surprise fix :D

[cbxbiker61]

Hello, can anybody who get it work post here the output of glxgears -info?
(GL_EXTENSIONS can be wiped out).

week ago I have tried to get my Radeon HD 3650 on gentoo work. i took latest mesa, kernel drivers etc from live trees and after linking everything together my glxinfo output still showed mesa software rasterizer. is this supposed to be so with the open source drivers?

i was not able to get my ati to reasonable work either with opensource or proprietary drivers.

Proprietary: perfect 3d, hard problems with 2d - similar to reported problems with KDE4 (but e16 here and without compositing) - the movement of windows and between desktop has unbearable delays - waiting for 2s for logout window etc... i tried the patch for xorg-server which was suppposed to help KDE4 but no success here. (btw is there any chance to get support from ati people on such issues?)

Opensource: perfect 2d. once i enable DRI, e16 pager stops working, instead of windows snapshots only black squares visible. i'm not sure i got 3D work, all xorg logs report success, but glxinfo still reports the software randerizer (see my question in the begining), but if yes then compared to proprietary drivers we are still nowhere. eg. googleearth still unusably slow - this seems to be in accordance to what devs themselves state on r600/r700 drivers wiki page.

You need to make sure r600_dri.so is getting built in Mesa. If it hasn't been installed you'll get a warning in Xorg.0.log. If it doesn't load r600 it will default to swrast. My build script patches the default dri configs before building Mesa.


--- ./configs/linux-dri.orig 2009-06-17 12:41:53.000000000 -0600
+++ ./configs/linux-dri 2009-10-11 21:54:48.910490712 -0600
@@ -63,5 +63,5 @@ GALLIUM_WINSYS_DIRS = drm
GALLIUM_WINSYS_DRM_DIRS = intel
GALLIUM_STATE_TRACKERS_DIRS = egl

-DRI_DIRS = i810 i915 i965 mach64 mga r128 r200 r300 radeon s3v \
+DRI_DIRS = i810 i915 i965 mach64 mga r128 r200 r300 r600 radeon s3v \
savage sis tdfx trident unichrome ffb swrast

--- ./configs/linux-dri-x86-64.orig 2009-10-12 00:59:12.616275651 -0600
+++ ./configs/linux-dri-x86-64 2009-10-12 01:04:00.971276520 -0600
@@ -20,7 +20,7 @@ EXTRA_LIB_PATH=-L/usr/X11R6/lib64
# the new interface. i810 are missing because there is no x86-64
# system where they could *ever* be used.
#
-DRI_DIRS = i915 i965 mach64 mga r128 r200 r300 radeon savage tdfx unichrome
+DRI_DIRS = i915 i965 mach64 mga r128 r200 r300 r600 radeon savage swrast tdfx unichrome

INSTALL_DIR = /usr
EXTRA_LIB_PATH = -L/usr/lib64

[AdamW]

blindfrog: did you file a bug for that?

There was another guy having odd interactions between radeon driver and his sound card. I wonder if there's something deeper going on than just resource conflicts...

[chrisr]

blindfrog: did you file a bug for that?

There was another guy having odd interactions between radeon driver and his sound card. I wonder if there's something deeper going on than just resource conflicts...

And then there's me:

http://phoronix.com/forums/showthread.php?p=97468#post97468

I have a Radeon 9550, and am running a 2.6.31.5 kernel with Fedora 11 and without KMS. And although I say "suddenly", it would be more accurate to say that things have suddenly become a lot worse. My "noise" problem used to only happen for a minute or so after start-up, and would then go away.

[AdamW]

And then there's me:

http://phoronix.com/forums/showthread.php?p=97468#post97468

I have a Radeon 9550, and am running a 2.6.31.5 kernel with Fedora 11 and without KMS. And although I say "suddenly", it would be more accurate to say that things have suddenly become a lot worse. My "noise" problem used to only happen for a minute or so after start-up, and would then go away.

Interesting, but not the same problem. I'd have a tough time guessing what your problem actually is, though. Try moving your Audigy to a different PCI slot, if it's an expansion card and not on-board.

[brada]

> If it doesn't load r600 it will default to swrast.

could you post me your output of:
glxinfo |grep -E 'OpenGL|server|client'
?

[dernob]

Are there any experiences with Radeon3D and KDE4 (Kwin)??

Btw. if someone successfully tests games, drop it here: http://www.x.org/wiki/RadeonProgram

I tried it in the current Fedora 12 Beta and it .. started. There were some minor glitches (like wrong transparency) and the performance was so medium. My gfx card is a radeon 4830.

[ewood]

I'd like to know whether my Power Mac G5's Radeon x800 will be working under Fedora 12? It wasn't with 11.

[bridgman]

Is there an active bug ticket open for the problem ? If not, there's a good chance nobody is aware something needs to be done.

[droidhacker]

I'd like to know whether my Power Mac G5's Radeon x800 will be working under Fedora 12? It wasn't with 11.

Right... x800 should have been working since F...9?
As bridgman above says, might be a bug that just doesn't happen to be known, i.e., it doesn't affect very many people. Be sure to report it, i.e. to the bug trackers of both Fedora as well as Xorg.

Could also be something to do with the KMS stuff that was really in a "not sure if its coming or going" state in F11, might want to try to just force-disable KMS. I think the kernel parameter is "nomodeset" (or something like that).

Might also help if you describe what it does (or does not) do in more details... "not working" doesn't give anybody much to go on.

[DanL]

and the performance was so medium...

Aren't Fedora pre-releases built with debugging code? That could slow it down..

[agd5f]

I'd like to know whether my Power Mac G5's Radeon x800 will be working under Fedora 12? It wasn't with 11.

KMS was missing big endian support when F11 was released. It should be working now, although it gets less testing than x86.

[AdamW]

DanL: the kernel from F12 Beta still had debugging code built-in, the latest one from updates does not. Testing indicates a noticeable performance improvement for Radeon/KMS with the non-debug kernel.

[ewood]

Right... x800 should have been working since F...9?
As bridgman above says, might be a bug that just doesn't happen to be known, i.e., it doesn't affect very many people. Be sure to report it, i.e. to the bug trackers of both Fedora as well as Xorg.

Could also be something to do with the KMS stuff that was really in a "not sure if its coming or going" state in F11, might want to try to just force-disable KMS. I think the kernel parameter is "nomodeset" (or something like that).

Might also help if you describe what it does (or does not) do in more details... "not working" doesn't give anybody much to go on.

In this case, not working means a blank or messed-up looking screen when X.org starts during DVD boot. I get the same with Ubuntu 9.04 or YDL 6.1, though I know that the YDL site states explicitly that Radeon X800 is not yet supported.

Now that Fedora 12 is about 17 days away, I'll be burning it then and hoping for the best. If that fails, I'll have to make sure there's a bug report for it. No time to try for a beta release now on my rather poor DSL.

Thanks for the replies thus far. I'm not sure yet whether Fedora is not working on this system because of the X800 GPU or the fact that the card is a flashed PC model, either. Do ye think that would make a difference? It can be quirky in Mac OS X, having minor graphics glitches that will disappear with the installation of ATI's own driver release meant for Tiger.

[tessio]

Gallium3D have something to do with this?

[bridgman]

This is all with the "classic" mesa HW driver model so far, ie not Gallium3D.

[chrisr]

Interesting, but not the same problem. I'd have a tough time guessing what your problem actually is, though. Try moving your Audigy to a different PCI slot, if it's an expansion card and not on-board.

FYI, I have managed to work around my sound problem by replacing the S/PDIF cable connecting my CD-RW drive to my Audigy2 with an analogue one instead. Whether this means that my S/PDIF cable is bad (somehow), the S/PDIF connector on my Audigy2 is bad, or ALSA's snd-emu10k1 driver has an issue with S/PDIF and Fedora 11/Pulseaudio, I do not know...

[droidhacker]

You mean that you have a wire running directly between the cd and the soundcard? Pull it out entirely -- it is redundant. That wire was useful back when the average CPU was a 386 and disk access was PIO (programmed I/O), it was used to allow the audio to bypass the CPU. This is no longer needed. First, the CPU is so much more powerful now that the impact of audio data would be negligible, second, we no longer use the old PIO mode disk access (now DMA), so it wouldn't hit the CPU anywhere near as hard either way (i.e., it would only hit the CPU if you are doing transformations on the audio stream).

And either way, when playing an audio CD, those circuits in the drive won't even be activated, so nothing is running on those wires anyways.

See the following:
http://en.wikipedia.org/wiki/Programmed_input/output
http://en.wikipedia.org/wiki/Direct_Memory_Access

I suspect that the issue you were having is a driver fault -- probably activated when there is something detected connected to that connector.

FYI, I have managed to work around my sound problem by replacing the S/PDIF cable connecting my CD-RW drive to my Audigy2 with an analogue one instead. Whether this means that my S/PDIF cable is bad (somehow), the S/PDIF connector on my Audigy2 is bad, or ALSA's snd-emu10k1 driver has an issue with S/PDIF and Fedora 11/Pulseaudio, I do not know...

[GoremanX]

You mean that you have a wire running directly between the cd and the soundcard? Pull it out entirely -- it is redundant. That wire was useful back when the average CPU was a 386 and disk access was PIO (programmed I/O), it was used to allow the audio to bypass the CPU. This is no longer needed.

I had that wire hooked up until recently, and it was good for one thing! Up until last year, I had a super-old CD-ROM drive from 1996 that had a "Play" button on it. Pressing "Play" would play the music CD that was inside, sending the sound through that analog wire and through the sound card without any processing being done at any level. That drive finally died a few months ago :( Longest-running piece of hardware I've ever owned.

Of course, that's a fringe case and not typical at all. In most cases, that wire is completely useless and I find it amusing that there are still connectors on both drives and motherboards/sound cards for it.

[AdamW]

if anyone's following this bug who tried F12 Beta on an r600 card but was having hanging issues - which AREN'T related to oddball soundcard problems =) - please keep an eye on this bug report:

https://bugzilla.redhat.com/show_bug.cgi?id=528593

we're fairly sure we got the bug fixed in a recent kernel build, which you can get here:

http://koji.fedoraproject.org/koji/buildinfo?buildID=139511

all early tests are positive. If you don't have / want to have an F12 installation, there's an x86-64 live image which includes the fixed kernel here:

http://adamwill.fedorapeople.org/radeon-20091102-x86_64.iso

[gorneman]

if anyone's following this bug who tried F12 Beta on an r600 card but was having hanging issues - which AREN'T related to oddball soundcard problems =) - please keep an eye on this bug report:

https://bugzilla.redhat.com/show_bug.cgi?id=528593

we're fairly sure we got the bug fixed in a recent kernel build, which you can get here:

http://koji.fedoraproject.org/koji/buildinfo?buildID=139511

all early tests are positive. If you don't have / want to have an F12 installation, there's an x86-64 live image which includes the fixed kernel here:

http://adamwill.fedorapeople.org/radeon-20091102-x86_64.iso

Please can someone post an alternate link to this image? Thanks

[gorneman]

can someone post another link to this image (radeon-20091102-x86_64.iso) ? Thanks

[AdamW]

can someone post another link to this image (radeon-20091102-x86_64.iso) ? Thanks

It's obsolete, you don't need it any more. The current nightly - http://alt.fedoraproject.org/pub/alt/nightly-composes/desktop/ - should be fine.

[elanthis]

Is there any word on the screen corruption issues on r600/hd4770 on fedora 12? I was told that nodownloadfromscreen should be set to avoid it - if so, shouldn't the driver just enable that by default? It seems wrong to push a known broken driver to users in a stable distro release that requires manually setting an option that 99% of those users will never find out about. They'll (rightfully I
o) just assume Linux is crappy amd abandon it forever if the out of the box experience is that buggy... It would be better to just disable the r600 driver entirely if it can't automatically enable the option to fix the corruption.

My fedora bug report:
https://bugzilla.redhat.com/show_bug.cgi?id=524419

[AdamW]

Is there any word on the screen corruption issues on r600/hd4770 on fedora 12? I was told that nodownloadfromscreen should be set to avoid it - if so, shouldn't the driver just enable that by default? It seems wrong to push a known broken driver to users in a stable distro release that requires manually setting an option that 99% of those users will never find out about. They'll (rightfully I
o) just assume Linux is crappy amd abandon it forever if the out of the box experience is that buggy... It would be better to just disable the r600 driver entirely if it can't automatically enable the option to fix the corruption.

My fedora bug report:
https://bugzilla.redhat.com/show_bug.cgi?id=524419

Your comment is literally the first I've heard of this bug, but Matej is triaging it and Dave is responding to it, if they'd considered it sufficiently serious to consider as a release-critical issue they would have nominated it as such. This looks like it's specific to particular chips (I don't think the other guy reporting in the bug has the same bug as you, since his hardware is a lot older).

I happen to have a 4770 here so I'll see if I can reproduce on that, but honestly it's something we'd usually be happy with fixing as an update. We can't just go setting workaround parameters universally; there's a reason they're not there by default, even though they help in a specific case it means they would hurt in others.

[elanthis]

Your comment is literally the first I've heard of this bug, but Matej is triaging it and Dave is responding to it, if they'd considered it sufficiently serious to consider as a release-critical issue they would have nominated it as such.

They should be turned on for the hardware that needs it, was my point.

I posted about this on Phoronix some weeks back, right before I opened that bug report. Not a big deal. Thank you very much for looking at it, I appreciate it tons. The sooner I stop needing to reboot every few hours, the better. :)

I happen to have a 4770 here so I'll see if I can reproduce on that, but honestly it's something we'd usually be happy with fixing as an update.

Try opening a theming site in Firefox (gnome-look.org, for example), browse around looking at themes, and open them all in new tabs. After you get enough pages open with images simultaneously is when it starts happening for me. I tend to browse by reading through a few pages, opening all the interesting links with a middle-click, and then going through one by one after I finish the current page. I tend to end up with 40+ tabs at a time whenever I get into reading something serious, and if there's enough images (ads, screenshots, whatever), the corruption starts. Hope that helps reproduce it.

[AdamW]

They should be turned on for the hardware that needs it, was my point.

I posted about this on Phoronix some weeks back, right before I opened that bug report. Not a big deal. Thank you very much for looking at it, I appreciate it tons. The sooner I stop needing to reboot every few hours, the better. :)



Try opening a theming site in Firefox (gnome-look.org, for example), browse around looking at themes, and open them all in new tabs. After you get enough pages open with images simultaneously is when it starts happening for me. I tend to browse by reading through a few pages, opening all the interesting links with a middle-click, and then going through one by one after I finish the current page. I tend to end up with 40+ tabs at a time whenever I get into reading something serious, and if there's enough images (ads, screenshots, whatever), the corruption starts. Hope that helps reproduce it.

Thanks, that's useful (and explains why I haven't hit it on regular testing of that system, I don't open that many tabs!)

We have a general policy of not spending too much time on implementing blacklists of hardware-specific workarounds and suchlike things in Fedora, instead spending the time working on just fixing the underlying problems; it can lead to a bit more pain in the short term but helps in the long term, you wind up with a clean code base with real fixes rather than a gigantic pile of workarounds you can't find the end of :)

[elanthis]

Thanks, that's useful (and explains why I haven't hit it on regular testing of that system, I don't open that many tabs!)

We have a general policy of not spending too much time on implementing blacklists of hardware-specific workarounds and suchlike things in Fedora, instead spending the time working on just fixing the underlying problems; it can lead to a bit more pain in the short term but helps in the long term, you wind up with a clean code base with real fixes rather than a gigantic pile of workarounds you can't find the end of :)

Ideally, you're right. Practically, if the issue can't be fixed in an acceptable time scale, a quick blacklist that you can fix properly after release is a far better experience for the end user. End users are much happier with a "it at least works right for now and will work better later" than with a "it doesn't work right now but if you just suck it up and deal with it it'll work right later." ;)

[AdamW]

Ideally, you're right. Practically, if the issue can't be fixed in an acceptable time scale, a quick blacklist that you can fix properly after release is a far better experience for the end user. End users are much happier with a "it at least works right for now and will work better later" than with a "it doesn't work right now but if you just suck it up and deal with it it'll work right later." ;)

As I said, in the short term, yeah. In the long term, it can get ugly. (I spent hours maintaining Mandriva's gigantic private list of graphics card PCI IDs which matches cards to drivers, with specific workarounds for problematic cards...half the workarounds had been written eight years ago by people who'd left seven years ago and it was impossible to tell what the hell they were for...)

[Korri]

Hi,

I've been running Fedora 12 on my HP 6735b laptop (RS780M/RS780MN [Radeon HD 3200 Graphics]) for a couple of days now and I've run into few problems. The only way to get the system into a usable state is to use nomodeset and the radeon driver. With this setup everything works, except suspend. When starting up from suspend, the screen stays blank. The system seems to start up ok, since it shuts down if I press the the power button.

Without nomodeset the machine crashes immediately when the fedora boot screen (plymouth?) is drawn.

So has anyone got suspend working with RS780M? What about KMS? Should it even work with my hardware?

[agd5f]

Works fine on all my r6xx/r7xx cards including my rs780. I use it daily. KMS should work fine on all r1xx-r7xx radeons.

[AdamW]

Hi,

I've been running Fedora 12 on my HP 6735b laptop (RS780M/RS780MN [Radeon HD 3200 Graphics]) for a couple of days now and I've run into few problems. The only way to get the system into a usable state is to use nomodeset and the radeon driver. With this setup everything works, except suspend. When starting up from suspend, the screen stays blank. The system seems to start up ok, since it shuts down if I press the the power button.

Without nomodeset the machine crashes immediately when the fedora boot screen (plymouth?) is drawn.

So has anyone got suspend working with RS780M? What about KMS? Should it even work with my hardware?

We want KMS to work with all hardware; the idea is for UMS (the old way) to go away in future leaving KMS as the only path. So we definitely need reports of any case where KMS fails but UMS works. Please file a report, including the information from your post and also everything mentioned in:

https://fedoraproject.org/wiki/How_to_debug_Xorg_problems

thanks!