Testing radeon KMS on Ubuntu

[tormod]

We have set up a package repository at https://launchpad.net/~xorg-edgers/+archive/radeon-kms which has kernel, libdrm, mesa and -ati packages for testing kernel mode-setting (and GEM/TTM/DRI2) on Ubuntu Karmic. The packages might install on Jaunty as well.

The above page also links to a Karmic-based live CD image (URL might change) which has all bits installed and simply boots up to the wonderful world of KMS...

Warning: This is experimental code so do not install it on your production system or anywhere where you have valuable data.

[Neo_The_User]

Be sure to have radeon.agpmode=-1 set in the kernel command line if you want to read anything. It would be rather annoying if I stated why.

[samentu]

Hello,
first of all I want to ask you if there is KMS support for an Ati Radeon x500 video card. (because this card is kinda old)
Secondly, I added the 2 entries in sources.list
But when I run apt-get update and the upgrade I don't see the kernel being upgraded and so this whole thing does not work. I am on Ubuntu 9.04
Thanks

[tormod]

Hello,
first of all I want to ask you if there is KMS support for an Ati Radeon x500 video card. (because this card is kinda old)

I am not sure there is an X500 model but there is the rv370-based X550. I have tested this on X700 which is not much newer.

Secondly, I added the 2 entries in sources.list
But when I run apt-get update and the upgrade I don't see the kernel being upgraded and so this whole thing does not work. I am on Ubuntu 9.04
Thanks
True, the kernel in the PPA does not replace the Jaunty kernel but installs side by side (which is good). Therefore you have to run:
apt-get install linux-image-2.6.30-8-generic

[tormod]

Be sure to have radeon.agpmode=-1 set in the kernel command line if you want to read anything. It would be rather annoying if I stated why.
I would not be terribly annoyed if you used a sentence on saying why instead of saying that you won't. I reckon AGP mode is broken on KMS and agpmode=-1 forces PCI mode?

[samentu]

Just upgraded to karmic and installed all your packages from ppa regarding kms.
But I have issues when resuming from suspend. And where do I add radeon.agpmode=-1? When booting the kernel? Should I append that line in the grub menu? Because I already did that and it says unknown option or something like that.
I did notice the VT's having my monitor's native resolution. That is great :) But the suspend problem is really an issue

[tormod]

Just upgraded to karmic and installed all your packages from ppa regarding kms.
But I have issues when resuming from suspend. And where do I add radeon.agpmode=-1? When booting the kernel? Should I append that line in the grub menu? Because I already did that and it says unknown option or something like that.
I did notice the VT's having my monitor's native resolution. That is great :) But the suspend problem is really an issue
My suspend is broken with the non-kms 2.6.30 kernel as well, so it is not necessarily the fault of KMS in my case. Did you test plain Karmic before installing the KMS crack?

You can add radeon.agpmode=-1 to the "kernel" line in the grub menu, or for a more permanent setting:
echo "options radeon agpmode=-1" | tee sudo /etc/modprobe.d/radeon-kms.conf
Then just delete /etc/modprobe.d/radeon-kms.conf once the AGP issues have been fixed.

[madman2k]

I tried the liveCD with an ATI Technologies Inc M56GL [Mobility FireGL V5250] which worked.
Then I tried to install the karmic packages on my jaunty installation. This did not work.
It loads VESA as default and if I force radeon via xorg.conf, it gives me:

kernel message:

Jun 7 13:25:19 deepblue kernel: [ 270.932079] [drm:radeon_fence_wait] *ERROR* fence(f646f680:0x00000002) 504ms timeout going to reset GPU
Jun 7 13:25:19 deepblue kernel: [ 270.933495] [drm] CP reset succeed (RBBM_STATUS=0x10000140)
Jun 7 13:25:19 deepblue kernel: [ 270.933503] [drm] radeon: cp idle (0x10000000)
Jun 7 13:25:19 deepblue kernel: [ 270.933513] [drm] Loading R500 Microcode
Jun 7 13:25:19 deepblue kernel: [ 270.933559] [drm] radeon: ring at 0x0000000010000000
Jun 7 13:25:19 deepblue kernel: [ 271.519871] [drm:radeon_ring_test] *ERROR* radeon: ring test failed (sracth(0x15E4)=0xCAFEDEAD)
Jun 7 13:25:19 deepblue kernel: [ 271.519877] [drm:r100_cp_init] *ERROR* radeon: cp isn't working (-22).
Jun 7 13:25:19 deepblue kernel: [ 271.519885] [drm:rv515_gpu_reset] *ERROR* Failed to reset GPU (RBBM_STATUS=0x90010140)
Jun 7 13:25:19 deepblue kernel: [ 271.519895] [drm:radeon_fence_wait] *ERROR* fence(f646f680:0x00000002) 1096ms timeout
Jun 7 13:25:19 deepblue kernel: [ 271.519900] [drm:radeon_fence_wait] *ERROR* last signaled fence(0x00000002)
Jun 7 13:25:19 deepblue kernel: [ 271.522087] [drm:radeon_ib_schedule] *ERROR* radeon: couldn't schedule IB(0).
Jun 7 13:25:19 deepblue kernel: [ 271.522094] [drm:radeon_cs_ioctl] *ERROR* Faild to schedule IB !
Jun 7 13:25:20 deepblue kernel: [ 271.870676] [drm:radeon_bo_move] *ERROR* CP is not ready use memcpy.
Jun 7 13:25:20 deepblue kernel: [ 271.876124] [drm:radeon_ib_schedule] *ERROR* radeon: couldn't schedule IB(0).
Jun 7 13:25:20 deepblue kernel: [ 271.876127] [drm:radeon_cs_ioctl] *ERROR* Faild to schedule IB !
Jun 7 13:25:21 deepblue kernel: [ 272.741028] [drm:radeon_ib_schedule] *ERROR* radeon: couldn't schedule IB(0).
Jun 7 13:25:21 deepblue kernel: [ 272.741036] [drm:radeon_cs_ioctl] *ERROR* Faild to schedule IB !
Jun 7 13:25:21 deepblue kernel: [ 273.330923] [drm:radeon_ib_schedule] *ERROR* radeon: couldn't schedule IB(0).
Jun 7 13:25:21 deepblue kernel: [ 273.330929] [drm:radeon_cs_ioctl] *ERROR* Faild to schedule IB !
Jun 7 13:25:21 deepblue kernel: [ 273.444620] [drm:radeon_bo_move] *ERROR* CP is not ready use memcpy.
Jun 7 13:25:21 deepblue kernel: [ 273.463984] [drm:radeon_ib_schedule] *ERROR* radeon: couldn't schedule IB(0).
Jun 7 13:25:21 deepblue kernel: [ 273.463989] [drm:radeon_cs_ioctl] *ERROR* Faild to schedule IB !
Jun 7 13:25:21 deepblue kernel: [ 273.481191] [drm:radeon_ib_schedule] *ERROR* radeon: couldn't schedule IB(0).
Jun 7 13:25:21 deepblue kernel: [ 273.481197] [drm:radeon_cs_ioctl] *ERROR* Faild to schedule IB !
Jun 7 13:25:21 deepblue kernel: [ 273.526445] [drm:radeon_bo_move] *ERROR* CP is not ready use memcpy.
Jun 7 13:25:21 deepblue kernel: [ 273.526580] [drm:radeon_ib_schedule] *ERROR* radeon: couldn't schedule IB(0).
Jun 7 13:25:21 deepblue kernel: [ 273.526583] [drm:radeon_cs_ioctl] *ERROR* Faild to schedule IB !
Jun 7 13:25:21 deepblue kernel: [ 273.526688] [drm:radeon_bo_move] *ERROR* CP is not ready use memcpy.
Jun 7 13:25:21 deepblue kernel: [ 273.526716] [drm:radeon_ib_schedule] *ERROR* radeon: couldn't schedule IB(0).
Jun 7 13:25:21 deepblue kernel: [ 273.526718] [drm:radeon_cs_ioctl] *ERROR* Faild to schedule IB !
Jun 7 13:25:21 deepblue kernel: [ 273.527003] [drm:radeon_bo_move] *ERROR* CP is not ready use memcpy.


xorg.log:

CS failure ...corruption... -22

while it still draws something, it is unbearable slow. It takes several seconds to just refresh the screen.

is perhaps something misconfigured missing? Id like to use DRI2, but I want to stick with jaunty.

edit:
just tried to wake up from suspend from the liveCD. It basically worked, but the radeon driver failed with the same message as when booting jaunty.

[samentu]

Thanks for the quick replies.
I did not tested Karmic before installing the KMS kernel. My bad :)
I created the radeon-kms.conf file but I see no change. So I don't know what this should fix
And another thing ... do you guys have plymouth packages for karmic in ppa?

[samentu]

Resuming from suspend I get: [drm:radeon_bo_move] *ERROR* CP is not ready use memcpy.
If I switch back and forth with CTRL_ALT+F1 and CTRL+ALT+F7 I manage to get the password prompt but on enter nothing happens. So I go back to VT1 and then back to X and I am loggen in but still can't see nothing but window borders and a lot of white and grey. And going back and forth again displays a new program if you managed to start one blindly from a terminal and so on.
And at about this time I get a lot of "*ERROR* Failed to schedule IB !"

All this only on resume.

[tormod]

Then I tried to install the karmic packages on my jaunty installation. This did not work.
It loads VESA as default
So you did not read the note on the PPA page, obviously.
and if I force radeon via xorg.conf, it gives me:

Are you actually booting the kernel from the PPA?

[tormod]

Thanks for the quick replies.
I did not tested Karmic before installing the KMS kernel. My bad :)
I created the radeon-kms.conf file but I see no change. So I don't know what this should fix

It should turn off AGP mode (and use PCI mode instead) on AGP cards. But you have a PCI-E card, right?

And another thing ... do you guys have plymouth packages for karmic in ppa?
The plan for Karmic is to not use plymouth nor usplash, but rather make the boot so fast that you would not care :)

There is a PPA somewhere with plymouth packages IIRC. Search for it.

[tormod]

I added a warning to the original post. For instance, if you are sticking to Jaunty for some reason, maybe you should not install these packages for the same reason.

I can confirm that both suspend and hibernation is broken, but I am not sure if it is KMS or another kernel problem. I even had bits of my .bash_history and firefox profile ending up in /var/log/syslog so there is apparently the possibility of major file (system) corruption.

[madman2k]

So you did not read the note on the PPA page, obviously.
nope, but I figured out myself :D


Are you actually booting the kernel from the PPA?
yup I do - Im getting high res console and usplash gets kicked out. probably something is not getting initialized properly - something that should have been also re-initialzed on wakeup from suspend.

[samentu]

It should turn off AGP mode (and use PCI mode instead) on AGP cards. But you have a PCI-E card, right?

The plan for Karmic is to not use plymouth nor usplash, but rather make the boot so fast that you would not care :)

There is a PPA somewhere with plymouth packages IIRC. Search for it.

Yes, indeed I have a PCI-Express card.
I've read about the boot in under 10 seconds for karmic ... I just hope they do that for this next release.
Thanks again

[osiris]

I can confirm that both suspend and hibernation is broken, but I am not sure if it is KMS or another kernel problem. I even had bits of my .bash_history and firefox profile ending up in /var/log/syslog so there is apparently the possibility of major file (system) corruption.

I think that Ubuntu is using vbetool to POST gpu when resuming from suspend to RAM - that may lead to crash or hang when KMS is enabled. There are possibly also other issues.

[zika]

I have ASUS AH3650 (a.k.a. ATI Radeon HD 3650 512M). Machine is 64/bit AMD Phenom X3 4950 (I think).
I am using Jaunty with 2.6.30./rc8 and 2.6.30-8.999 daily kernel and radeon driver from tormod`s ppa ...
I`ve tried Your Live/CD (it is 32/bit?)(I am writing this while on it) and (compiz is not woking but I do not care) I would like to try it on Jaunty. What is the right way to do that. My 2.6.30/rc8 kernel is not the right one? I should add Your repository, update what is updateable automatically and then manually install that kernel? What problems should I expect on that route to ... ?
update: I've upgraded to Karmic in the meantime. So I will try to follow instructions on the ppa site. The only thing that bugs me is: what happens if I try to use different (read newer) kernel that is not from that ppa? Will the kernel be updated on Your ppa in the same rhythm as it is coming on in mainline ... ?
update-1: It was worth of all the "trouble" ... :) Glxgears almost doubled, compiz is not working (who cares) but card is, at last, coming near some of its previous performances in Ubuntu... Thank You for all the work You are doing for our little community.
update-2: For those who might be interested, it works with 2.6.30-8.999 (from mainline, latest daily) but with crippled speed (the same as it was in Karmic before I applied updates from Your ppa) but it boots in 5 sconds less with 2.6.30-8.999 than with Your version of 2.6.30-8-generic. I can live with that 5 seconds more ... :)

[super.rad]

Will this work with RS780 chips (HD3200), probably been asked already or mentioned somewhere but I haven't been keeping up with all the latest radeon KMS and 3D stuff

[DF5JT]

The above page also links to a Karmic-based live CD image (URL might change) which has all bits installed and simply boots up to the wonderful world of KMS...



I installed this on a T60p with a FireGL V5200 and it works as expected. S2RAM is broken, probably due to Ubuntu's VBE check. The rest works just smooth, with wobbly compiz, 720p videos, moving glxgears on the cube.

Different from jglisse's git version and a modified kernel on my Mandriva system, even Flash HD videos work rather well in fullscreen. This is an enormous progress ;-) [1]

Generally speaking, the performance is awesome and the advantages of KMS become more and more obvious.

Thanks for the great work!

[1] Is there chance for the near future to get
* GL_ARB_multitexture
* GL_EXT_framebuffer_object
* GL_ARB_shader_objects
* GL_ARB_shading_language_100
* GL_ARB_fragment_shader
working with my FireGL card? I *really* would like to make use of GPU accelerated flash videos.

[tormod]

On Jaunty, the VBE POST on resume quirk is configured in /usr/share/hal/fdi/information/10freedesktop/20-video-quirk-pm-*

Check /var/log/pm-suspend to see which quirks are in use.

On Karmic these quirks should be automatically disabled when KMS is in use (by /usr/lib/pm-utils/sleep.d/98smart-kernel-video).

[madman2k]

On Jaunty, the VBE POST on resume quirk is configured in /usr/share/hal/fdi/information/10freedesktop/20-video-quirk-pm-*

Check /var/log/pm-suspend to see which quirks are in use.

On Karmic these quirks should be automatically disabled when KMS is in use (by /usr/lib/pm-utils/sleep.d/98smart-kernel-video).
after disabling VBE there, I could finally start an Xserver with radeon on my jaunty setup.
At least for EXA/2D - 3D is still dog slow. Might it be that GDM starts its own Xserver which get in conflict?
Does your liveCD use GDM?

[tormod]

For Jaunty, to disable all the suspend video quirks, download http://git.debian.org/?p=collab-maint/pm-utils.git;a=blob_plain;f=pm/sleep.d/98smart-kernel-video;hb=HEAD and correct the typo in it, make it executable and copy it to /etc/pm-utils/sleep.d:
sed -i 's/add_parameter/add_parameters/' 98smart-kernel-video
chmod 755 98smart-kernel-video
sudo cp 98smart-kernel-video /etc/pm/sleep.d/
Remember to delete /etc/pm/sleep.d/98smart-kernel-video again when upgrading to Karmic.

Unfortunately, suspend and hibernation is still broken my Jaunty system.

EDIT: You can also find a backport for Jaunty of the Karmic pm-utils in the PPA.

[diverse_izzue]

I tried the CD image, on a Thinkpad T60 with a Radeon X1400 Mobile and an SXGA+ screen (probably the same machine as mentioned in the article).

Except for the boot experience not being very pretty (nor very fast, from a USB stick), it works rather well. Especially:

* VT switching works fine
* DRI2 works all right, can move around glxgears in a wobbly window without corruption
* Changing resolution up and down works, even without crashing compiz, which is a first for me.

That said, the main downside is that it's considerably slower than the current architecture, for both 2D and 3D operations it seemed to me.

[val-gaav]

I can confirm that both suspend and hibernation is broken, but I am not sure if it is KMS or another kernel problem.
For me this the first time suspend works out of the box though :)

Earlier I had to use s2ram and some hacks for it to make it work.

[tormod]

There is now a new kernel in the radeon-kms PPA, based on 2.6.30 final. It is included in the new live CD version 0.14kms. Unfortunately, suspend is still broken on my gear, even though the new kernel includes the fix "radeon: fix suspend/resume on PCIE card".

EDIT: My own suspend problem is a karmic one and not a kms one: A mainline 2.6.30 kernel suspends fine, the stock karmic kernel does not.

[nanonyme]

[1] Is there chance for the near future to get
* GL_ARB_multitexture
* GL_EXT_framebuffer_object
* GL_ARB_shader_objects
* GL_ARB_shading_language_100
* GL_ARB_fragment_shader
working with my FireGL card? I *really* would like to make use of GPU accelerated flash videos.I think GL_ARB_multitexture has existed for a while. GL_EXT_framebuffer_object probably comes eventually, taken its time since it apparently needs a kernel memory manager to be doable sensibly. Unsure about the rest. (Just an educated guess: at latest in Gallium; seems to me a driver that supports OpenGL 2.0 should be plenty more than sufficient for them all and Gallium should provide that)

[runrun]

The above page also links to a Karmic-based live CD image (URL might change) which has all bits installed and simply boots up to the wonderful world of KMS...

I tried this on an nforce2 motherboard with an x1950pro AGP card - unfortunately it doesn't boot:

[drm:radeon_driver_load_kms] *ERROR* Failed to initalize radeon. Disabling IOCTL

[tormod]

I tried this on an nforce2 motherboard with an x1950pro AGP card - unfortunately it doesn't boot:

[drm:radeon_driver_load_kms] *ERROR* Failed to initalize radeon. Disabling IOCTL
Did you run the latest 0.14kms CD? And did you read comment #2 in this thread?

[samentu]

There is now a new kernel in the radeon-kms PPA, based on 2.6.30 final. It is included in the new live CD version 0.14kms. Unfortunately, suspend is still broken on my gear, even though the new kernel includes the fix "radeon: fix suspend/resume on PCIE card".

Thanks a lot for sharing the news about the new kernel. Suspend works like a charm now on my box.

L.E.: hibernate does not work though. When I try to hibernate it does not turn off the PC, instead I have to push the power off button on the case and on the next boot it hangs

[highlandsun]

I tried to boot the live CD from a USB memory stick (copied using the isotostick shell script) and it failed with some complaint about squashfs version mismatch. Anyone else using it?

[cristjs]

Just finished a quick test of the live cd on my Acer Aspire 5100 with a Radeon Mobility m200. It is getting twice the fps in glxgears that I get running Kubuntu JJ under xorg 7.4 radeon open source driver... Keep up the good work folks...

[DF5JT]

We have set up a package repository at https://launchpad.net/~xorg-edgers/+archive/radeon-kms which has kernel, libdrm, mesa and -ati packages for testing kernel mode-setting (and GEM/TTM/DRI2) on Ubuntu Karmic. The packages might install on Jaunty as well.

The above page also links to a Karmic-based live CD image (URL might change) which has all bits installed and simply boots up to the wonderful world of KMS...

Warning: This is experimental code so do not install it on your production system or anywhere where you have valuable data.

This is truly outstanding work indeed. On my T60p compiz works blazingly fast in all its glory, videos run smooth and tear free, even youtube HD works, albeit a little jerky, but that's the same under Windows, so this is great success.

My only gripe:

(WW) RADEON(0): Option "ForceLowPowerMode" is not used
(WW) RADEON(0): Option "DynamicPM" is not used
(WW) RADEON(0): Option "ClockGating" is not used

Any ideas why this doesn't work?

Ah, and Wolfenstein exits with a SIG11. Haven't managed to get it to work with any version of the radeon driver, only with fglrx.

[tormod]

I tried to boot the live CD from a USB memory stick (copied using the isotostick shell script) and it failed with some complaint about squashfs version mismatch. Anyone else using it?
Sounds like you did not copy the live CD correctly and you have some conflicts with old live CD stuff on it.

1. Make sure you delete the old casper folder on the stick
2. Use usb-creator instead of whatever scripts

[Kano]

Well the usb-creator is only a script with a little gui - installing syslinux on a fat stick is a trivial task, wrote a script that works with debian live and U. But i would prefer when U would be using a more recent isolinux version + isohybird to create hybrid images you can directly put onto usb device.

[forum1793]

Ah, and Wolfenstein exits with a SIG11. Haven't managed to get it to work with any version of the radeon driver, only with fglrx.

Not sure if this applies to you, but I used to get the same error with return to castle wolfenstein (RTCW) when the sound wasn't working. RTCW used OSS but my distro used alsa. It alway died with signal 11.

As root try this:

echo 'wolfsp.x86 0 0 direct' > /proc/asound/card0/pcm0p/oss

The start the game (probably as user) with : wolfsp.x86

[tormod]

(WW) RADEON(0): Option "ForceLowPowerMode" is not used
(WW) RADEON(0): Option "DynamicPM" is not used
(WW) RADEON(0): Option "ClockGating" is not used

Any ideas why this doesn't work?

This is because the kms-capable version of the -ati driver was forked off master a long time ago, before the power management was introduced, and it has not been synchronized since. Expect this to be fixed within the next weeks.

[highlandsun]

Well the usb-creator is only a script with a little gui - installing syslinux on a fat stick is a trivial task, wrote a script that works with debian live and U. But i would prefer when U would be using a more recent isolinux version + isohybird to create hybrid images you can directly put onto usb device.

Second that; I specifically pointed the Ubuntu folks at isolinux+isohybrid several months ago. They decided not to investigate because it didn't fit their plans for USB. Annoying... (They were only interested in ISO + USB if it allowed saving state in the unused space on the USB flash, and they already had a different approach to get there. They don't seem to care about providing a single ISO that works identically on CD and USB.)

[Kano]

I think they wait till suse gives em an new patched syslinux. The gfxboot patched syslinux is not supported by upstream. And it was definitely no U invention.

[GDJacobs]

Any reports of success from Debian users? I ask this because I'm about at my wits end troubleshooting some annoying green screen corruption under KMS. A non-KMS kernel loads Xorg normally. The KMS kernel works cleanly when using modeset=0.

I have tried working around possibilities of versioning problems by using the Edgers KMS PPA as well as building from Jerome's git repository for mesa, xorg ati+radeon servers and libdrm as per http://jglisse.livejournal.com/1822.html. For the kernel, I have used both the patched KMS kernel on edgers and the 2.6.29 tree in Jerome's private repo.

When loading KMS I receive the version mismatch error reported by MùPùF here:
http://www.phoronix.com/forums/showthread.php?t=17719

dmesg:
http://pastebin.com/f6e9dc289
X.org log:
http://pastebin.com/f2eb3bc19
lsmod:
http://pastebin.com/f27d0ec27

For what it's worth, v0.2 of the Edgers live CD works brilliantly, so I know my RS4xx IGP is at least nominally compatible. I'm guessing this is a userspace issue, but I just don't know what else to attack. Suggestions are very much appreciated.

[nanonyme]

This is because the kms-capable version of the -ati driver was forked off master a long time ago, before the power management was introduced, and it has not been synchronized since. Expect this to be fixed within the next weeks.Including which in the KMS-capable version power management is afaik supposed to be written in kernel. (which leads to better power management than before due to the centralization of modesetting and acceleration features)

[meden]

Hello,
I'm testing KMS packages under Jaunty (and yes, I read the disclaimer...) and I get very poor performances. With Compiz the system is unusable and with Metacity is quite slow (slowness is well noticeable on resizing and moving windows).

To make things work I had to install packages from radeon-kms PPA (of course) and upgrade xorg and mesa to Karmic; with only KMS packages or with radeon-kms and xorg-edgers packages I get console working, but xorg fails to start ("failed to allocate fb").

Everything works good with KMS live 0.14 (boot from usb).

How can I solve this issue? Thanks

[madman2k]

I dont have a solution, but I have the same problems using KMS on jaunty. Probably the xserver is too old or some other userspace component.
But radeon KMS is stilly quite immature - radeon-rewrite has still some major performance refressions and power management does not work at all..

[meden]

By setting drm.debug=1 I get syslog flooded by

[drm:drm_ioctl] pid=4025, cmd=0x40086409, nr=0x09, dev 0xe200, auth=1
[drm:drm_ioctl] pid=4025, cmd=0xc01c645d, nr=0x5d, dev 0xe200, auth=1
[drm:drm_ioctl] pid=4025, cmd=0xc020645e, nr=0x5e, dev 0xe200, auth=1
[drm:drm_ioctl] pid=4025, cmd=0xc0206466, nr=0x66, dev 0xe200, auth=1
[drm:drm_ioctl] pid=4025, cmd=0xc0086464, nr=0x64, dev 0xe200, auth=1

where pid=4025 is Xorg and cmd sequence is fixed.

I get even a lot of

[drm:drm_mode_cursor_ioctl]
[drm:drm_ioctl] pid=4025, cmd=0xc01c64a3, nr=0xa3, dev 0xe200, auth=1


and in Xorg.0.log

RADEON DRM CS failure - corruptions/glitches may occur -22
bufmgr: last submission : r:64 vs g:268435456 w:94208 vs v:46861515
RADEON DRM CS failure - corruptions/glitches may occur -22
bufmgr: last submission : r:64 vs g:268435456 w:188416 vs v:46861515


What's going on?

[DF5JT]

The latest updates include 2.6.30-10-generic and with this kernel, KMS does not work anymore.

[samentu]

The latest updates include 2.6.30-10-generic and with this kernel, KMS does not work anymore.

Has this happened to somebody else? Because I'm planning to upgrade to the new kernel but if KMS does not work ...... I won't do it

[bridgman]

I think KMS and GEM/TTM for radeon went into 2.6.31, not 2.6.30...

[val-gaav]

yeah maybe the update is unpatched kernel from main ubuntu repos ?

I mean the last listed kernel here :
https://launchpad.net/~xorg-edgers/+archive/radeon-kms
is
linux - 2.6.30-9.10kms6 uploaded 2009-06-17

you should obviously stick to that if you want KMS

[highlandsun]

Sounds like you did not copy the live CD correctly and you have some conflicts with old live CD stuff on it.

1. Make sure you delete the old casper folder on the stick
2. Use usb-creator instead of whatever scripts

I had 2 other partitions with live CD isos on them. Apparently the init script doesn't just use the partition it booted from, it searches for a /casper directory on every block device it can mount. Seems a bit strange, because my other two isos boot just fine without any confusion. Anyway, I renamed the /casper on the other two partitions so they wouldn't be found, so it mounted the correct one from /dev/sdb3.

Usually I just have to toggle the bootable flag on the partition I want, using fdisk. This new casper script seems like a step backward.

Anyway, that seemed to be the only problem and so it booted successfully on my Asus M6Ne with Mobility 9700 (M300). But it only sees the LVDS and VGA, it doesn't see my S-Video connector. I guess I'm going to stick with Jaunty for a while longer.

[tormod]

I had 2 other partitions with live CD isos on them. Apparently the init script doesn't just use the partition it booted from, it searches for a /casper directory on every block device it can mount. Seems a bit strange, because my other two isos boot just fine without any confusion. Anyway, I renamed the /casper on the other two partitions so they wouldn't be found, so it mounted the correct one from /dev/sdb3.

Usually I just have to toggle the bootable flag on the partition I want, using fdisk. This new casper script seems like a step backward.
casper never looked for a bootable flag. Only the DOS MBR uses the bootable flag AFAIK. casper should verify the .disk/casper-uuid on the block device (against its conf/uuid in the initrd) to make sure it does not mount a wrong /casper directory. Did you boot with the ignore_uuid option?

Anyway, that seemed to be the only problem and so it booted successfully on my Asus M6Ne with Mobility 9700 (M300). But it only sees the LVDS and VGA, it doesn't see my S-Video connector. I guess I'm going to stick with Jaunty for a while longer.
You can always report the bug to help getting it fixed, if it is not a known issue.

[highlandsun]

casper never looked for a bootable flag. Only the DOS MBR uses the bootable flag AFAIK. casper should verify the .disk/casper-uuid on the block device (against its conf/uuid in the initrd) to make sure it does not mount a wrong /casper directory. Did you boot with the ignore_uuid option?


I didn't set any options at all. I dunno why it didn't match UUIDs properly.


You can always report the bug to help getting it fixed, if it is not a known issue.

Which bug tracker? That iso is already 2 weeks old, not even sure if the bug is current any more.

[tormod]

I didn't set any options at all. I dunno why it didn't match UUIDs properly.
Yes, that is strange. Possibly a bug.

Which bug tracker? That iso is already 2 weeks old, not even sure if the bug is current any more.
Right, and with all the stuff happening now (kms remake in -ati master, kms kernel enabled by default in karmic, assorted 2.6.31 issues) it can be some days before we get to make a new CD iso. I had to resort to vesa last night :)

[pedrodh]

Hi! I'm sorry for my ignorance, but what does this KMS feature brings for the radeon cards? I assume this will make the open source drivers better yes? I've a X600 ( RV380 GPU I think ) and since ATI have ceased support for this and other cards with their proprietary driver this could bring some joy into my life, maybe there's still hope for this laptop :P. This feature applies to older GPUs like mine, correct?

[bridgman]

KMS on its own allows :

- faster and more reliable suspend/resume/VT-switch

- the ability to run the X server without root privileges (so less likelihood of wedging your system),

- a single graphics driver that is used everywhere. The pre-KMS environment has a variety of different drivers taking turns controlling the hardware, which frequently works well but sometimes doesn't -- suing a single driver for everything also means bootup can be smoother and kernel errors can display messages reliably

Equally important to you is that KMS is built on top of a common in-kernel memory manager for graphics (GEM/TTM), which is the pre-requisite for a number of other desireable features :

- DRI2 / Redirected Direct Rendering (flicker-free compositing of 3D windows)

- higher levels of GL support in the 3D drivers (allows apps/games run which could not run before, allows some already running games to take faster code paths)

- Gallium3D (think of it as a new, improved hardware driver model inside Mesa), which offers both higher potential performance and a framework capable of supporting other acceleration APIs including video decode

[pedrodh]

Thank you for the info. I hope this get done soon, I've had problems with resume in the past, and well, the current open-source driver lacks power when it comes to games. Good luck for the developers :) ( even though they might not be reading this :P ).

[PapadakosPanagiotis]

Anyone expreriences random small freezes with kms? This is on a ATI Mobility Radeon X2300

[agd5f]

Anyone expreriences random small freezes with kms? This is on a ATI Mobility Radeon X2300

This is probably due to buffer migrations. The new mm code isn't well tuned yet.

[Fenix-TX]

I'm trying to get kms with latest karmic kernel and using ppa repo from this post, but i think that is not working because on dmesg i have these lines:

[ 13.118824] [drm] radeon default to kernel modesetting DISABLED.
[ 13.119111] [drm] Initialized radeon 1.30.0 20080528 for 0000:01:00.0 on minor 0

[madman2k]

I'm trying to get kms with latest karmic kernel and using ppa repo from this post, but i think that is not working because on dmesg i have these lines:

[ 13.118824] [drm] radeon default to kernel modesetting DISABLED.
[ 13.119111] [drm] Initialized radeon 1.30.0 20080528 for 0000:01:00.0 on minor 0

read the comment on the PPA page :p

[runrun]

I'm trying to boot the latest 0.16 livecd hosted on your link on a machine with an agp x1950pro.
I'm getting a 'Failed to initalize radeon. Disabling IOCTL' error during the boot process and the system stops.
Post #2 says add 'radeon.agpmode=-1' in the kernel command line and ppa says(?) 'radeon.modeset=1'
I assume this is where I'm going wrong trying to boot the live system, pushing f6 when the cd first boots and adding 'radeon.modeset=1' etc returns an unknown option error as you boot - and I don't know how to continue.

[tormod]

Boot with radeon.modeset=1 to use KMS and with radeon.modeset=0 to run without. Official Ubuntu kernels currently default to 0, the xorg-edgers kernels default to 1.

If you have an AGP card, you can force the AGP mode with radeon.agpmode=X where X is 1,2,4,8 (same as the old AGPMode option in xorg.conf, see "man radeon"). You can turn off AGP (and force PCI instead) with radeon.agpmode=-1.

"modinfo radeon" lists all options for the radeon module. For general boot options, see https://help.ubuntu.com/community/BootOptions

Ignore the "unknown option" warning, it is a false warning.

[madman2k]

Boot with radeon.modeset=1 to use KMS and with radeon.modeset=0 to run without. Official Ubuntu kernels currently default to 0, the xorg-edgers kernels default to 1.

If you have an AGP card, you can force the AGP mode with radeon.agpmode=X where X is 1,2,4,8 (same as the old AGPMode option in xorg.conf, see "man radeon"). You can turn off AGP (and force PCI instead) with radeon.agpmode=-1.

"modinfo radeon" lists all options for the radeon module. For general boot options, see https://help.ubuntu.com/community/BootOptions

Ignore the "unknown option" warning, it is a false warning.
try writing agpmode=-1 - at least for me radeon.modeset=1 did not work for the official kernel, but modeset=1 did.

[nanonyme]

Boot with radeon.modeset=1 to use KMS and with radeon.modeset=0 to run without. Official Ubuntu kernels currently default to 0, the xorg-edgers kernels default to 1.And official Linus' tree kernels default to 1 as well, aight?