OSS Radeon Drivers in Debian Testing are a letdown

[Tillin9]

Finally got around to testing out the OSS Radeon drivers. I started with the defaults in Debian Testing (Squeeze) with kernel 2.6.30 and radeon 6.12.3.1 and my Radeon HD4600.

While it did worked out of the box (mode setting, video), 3D was non-existent and 2D was rather slow.

I see there is 6.12.4 ddx driver release, but it didn't seem to express any big changes for 2D or anything for 3D.

I went back to fglrx for now, since it has been working great for me for a few releases, but wanted to know if this was me (Debian is known to be a little conservative) or things just aren't there yet.

P.S. - I had to reboot after installing fglrx, something was preventing the module from loading properly. I didn't see anything else drm related via lsmod and was also under the impression that fglrx and the OSS drm could be loaded at the same time now. While this was marginally annoying (kwin4 has more dependencies and needs more of the video stack working) it may suggest a bigger issue.

[agd5f]

You need kernel 2.6.32 for 3D on r6xx/r7xx hardware. You need to completely remove fglrx to use acceleration on the open source driver.

[chithanh]

Debian testing is probably not the best choice if you want bleeding edge drivers.

If you use Fedora 12, installing mesa-dri-drivers-experimental will give you instant 3D acceleration on that card.

Other distros will work too, but require a bit more effort to get 3D, such as installing pre-release kernels and packages from external repositories.

[Tillin9]

Yep, thought that could be the case. I'm downloading the Fedora 12 Live cd now to give that a whirl.

[panda84]

Yep, thought that could be the case. I'm downloading the Fedora 12 Live cd now to give that a whirl.

The Live CD will not give you 3D as mesa-dri-drivers-experimental is not installed by default. Also mind that "experimental" indeed means experimental. :)
When Mesa 7.7 will be out (probably by the end of the year) the situation will start to get better. Not without bugs of course, but will start to get usable somehow.
BTW, if you like living-on-the-edge take a look at the latest Mesa builds in koji:
http://koji.fedoraproject.org/koji/packageinfo?packageID=184

[Panix]

The Live CD will not give you 3D as mesa-dri-drivers-experimental is not installed by default. Also mind that "experimental" indeed means experimental. :)
When Mesa 7.7 will be out (probably by the end of the year) the situation will start to get better. Not without bugs of course, but will start to get usable somehow.
BTW, if you like living-on-the-edge take a look at the latest Mesa builds in koji:
http://koji.fedoraproject.org/koji/packageinfo?packageID=184
What does that mean? How are you supposed to enable the support then?

I have a different card than the OP but I think there is a problem with older ATI hardware that uses the open source driver. I couldn't get Ubuntu 9.10 Karmic to work with 3D effects without it crashing my laptop and needing a hard restart. Trying the latest sidux DVD, a similar situation but with a black screen freeze and white mouse cursor. No keystrokes released the freeze. Aren't those recent enough distros? Also, Fedora 12 (release) was usable but I had what I call pixelation issues or some screen artifacts. But, at least I could enable 3D.

Only Mandriva 2010 had what one could call a 'smooth experience' after enabling 3D desktop effects. I don't know if anyone has had a similar overall experience but there seems to be a pattern of similarity in some cases. Just speculating though.

Sorry to interrupt the thread. I'm just interested in the solution and no matter what I tried in xorg.conf, I couldn't find a solution.

[Tillin9]

I read the article here: http://www.phoronix.com/scan.php?page=article&item=fedora_r600_3d&num=1 there is an experimental package in the repository. I assume I install Fedora to a partition, then update to the experimental support.

Note: This is to just feel it out easily. I know I can build everything myself on Debian, I just don't want to spend days doing this if it only gives me a system useful for driver development. I really appreciate the work you guys do, but can't really contribute anything useful and want a working system.

P.S. What card do you have Panix? I've had a lot of success with the r300 OSS driver on my older cards. I'm surprised to hear about your issues.

[Panix]

P.S. What card do you have Panix? I've had a lot of success with the r300 OSS driver on my older cards. I'm surprised to hear about your issues.
It's in a laptop, a Radeon Mobility 9000, RV250 card.

Experiences with distros and trying 3D effects:
Ubuntu* - boots in default mode and then crashes after starting Firefox, Synaptic etc.
Kubuntu - goes to black screen and freezes/crashes
sidux - Momos - same
OpenSUSE 11.2 - OpenGL 3D setting will not enable (a msg states so) - XRender works
Mandriva 2010 - 3D desktop effects works
Fedora 12 - desktop effects can be enabled but there is 'pixellation', flickering and screen artifacts

*attempts at editing/configuring xorg.conf in Ubuntu have been met with failure. Various crashes, flickering and corrupted screen/windows has provoked me to abandon any further experimentation until I find some evidence or info of what to put there

I suppose the point of this post is pure speculation on my part but imho, it does seem like the Debian-based distros are not as successful in implementing or utilizing the Radeon OSS driver?

[bridgman]

AGP is still a challenge for the driver devs, since chipset and non-graphics drivers can have a big impact on the reliability of the graphics subsystem.

If you are getting corruption with an AGP card the first thing to try is usually disabling accelerated block transfers, eg turning off AccelDFS. If you are getting crashes as well, then first thing to try is reducing AGP bus speed as low as possible or forcing to PCI mode.

[Kano]

Yes, maybe you should send the devs a HD 2600 AGP too, this card does not work too...

[bridgman]

There are quite a few AGP cards in developer hands already (unless they were tossed in a moment of AGP-hate, which is always a risk ;)).

The issues tend to be chipset- and SBIOS-specific, unfortunately, not specific to the GPU or graphics card. If the AGP bus is not reliable, there are limits to what the driver itself can do. That's why most of the recommendations involve slowing down AGP bus speed or disabling use of the AGP features which are most likely to be unreliable.

Now that motherboard vendors have stopped supporting AGP mobos with driver and SBIOS updates, there's an argument that it's time to have the driver "assume the worst" for AGP hardware and default to lowest speed and disabling optimizations by default. The problem there is that on most systems the current defaults seem to run well, so changing the defaults would force most users to edit their xorg.conf files in order to get back current behaviour.

[Panix]

Why not just buy a bunch of refurbished computers with this hardware?

There are still older HP, Thinkpads, Dell Latitudes etc. etc. and they all have either Intel or ATI graphics. Many have ATI hardware. If X, xorg, OpenGL, etc. etc. (anything to do with video) is constantly changing and with the kernel, then it just stands to reason that you might need to integrate info and ideas with the various respective distribution developers who work on video issues and whenever the need arises to integrate within the kernel. I assume this is done already but from my reading, it seems a slow progression. I confess to ignorance and inexperience but people still use this older hardware. Maybe the driver should be integrated as one so all resources can be pooled together?

[monraaf]

IMHO with the limited manpower available, AMD paid devs should be focusing their development effort on the R600/R700/R800 driver and leave the obsolete cards and obsolete technologies such as AGP to the community.

[Panix]

IMHO with the limited manpower available, AMD paid devs should be focusing their development effort on the R600/R700/R800 driver and leave the obsolete cards and obsolete technologies such as AGP to the community.
I don't agree from a marketing standpoint, though. If you were moving to a newer laptop and the older one you couldn't get ati drivers optimized for 3D, you might not consider a newer machine with newer ati hardware just from prior experience. Btw, is it not possible to integrate the driver and have it detect the hardware and then pull in the related and required code from there?

If not, I think it is asking a lot to rely on the community. I've read so much conflicted info related to my 'obsolete' video card and only found one distro in which it seems to work (we're talking 3D here. I haven't had any 2D issues no matter the distro although it doesn't help if you wanted to run GoogleEarth, say).

[Hoodlum]

I don't agree from a marketing standpoint, though. If you were moving to a newer laptop and the older one you couldn't get ati drivers optimized for 3D, you might not consider a newer machine with newer ati hardware just from prior experience. Btw, is it not possible to integrate the driver and have it detect the hardware and then pull in the related and required code from there?


Surely it is worse from a marketing standpoint to buy a new card and find it's simply not supported because the devs are too busy working on radeon x800 drivers? Would this not detract from buying their hardware in future? even moreso?

I understand your frustration....but from a logical standpoint it will not be long before the R600/R700/R800s will be legacy also. Having good support for them sooner rather than later allows the devs to focus on the next line of hardware rather than dealing with R300-R500s and in 6 months, a years time dealing with the R600/R700/R800 which would by that time already be obselete and thus forever be behind the curve. At some point the priority has to shift to "lets actually make the new hardware work". If we start "now" and from this point onwards all hardware is supported well. in the next 5 years we will be in a *much* better place. If we keep looking back and spend the majority of effort on already legacy hardware in 5 years time we will still be as behind as we are now. This does not help.

[monraaf]

Well, as a Linux user whenever I buy hardware I always check on the Internet how well it's supported under Linux. Therefore I didn't have any ATI graphics before last year :D.

Right now I'm waiting to buy a HD 5750 or 5770 for use with the OSS driver. Currently it's not yet supported, but I'm patient. However every hour Alex or some other AMD dev spends debugging AGP problems or some obsolete cards is an hour he cannot not spent on the driver for the newer cards. So if only for my own selfish reasons I say stop wasting time debugging AGP problems :D

[Panix]

Well, if it was really the case and things are SOOOO bad that only one guy has to decide which card he's going to work on (I'm exaggerating), I'd have no choice but to get a Nvidia card and totally give up on my laptop. I'd just use Windows on it, then or sell it.

Why on earth would any Linux user buy an ATI card if they have to keep wondering/worrying whether support is going to be abandoned each time a newer generation card comes out? I know I sure wouldn't even consider it. Common sense would suggest to me to bite the bullet and get a Nvidia card.

[Kano]

I tested

ATI Technologies Inc RV630 PRO AGP [Radeon HD 2600 PRO AGP] [1002:9587]

via ssh today and it only worked with BusType PCI or PCIE override, thereforce i patched this:

diff --git a/src/radeon_driver.c b/src/radeon_driver.c
index 17253a7..4edd487 100644
--- a/src/radeon_driver.c
+++ b/src/radeon_driver.c
@@ -1964,7 +1964,7 @@ static Bool RADEONPreInitChipType(ScrnInfoPtr pScrn)
(info->ChipFamily == CHIP_FAMILY_RS480))
info->cardType = CARD_PCI;

- if ((info->ChipFamily >= CHIP_FAMILY_R600) && info->IsIGP)
+ if (info->ChipFamily >= CHIP_FAMILY_R600)
info->cardType = CARD_PCIE;

/* not sure about gart table requirements */

Maybe it helps you too...

[agd5f]

That patch disables AGP on all r6xx+ hardware. If you are having AGP problems, use:
Option "BusType" "PCIE"
in the device section of your config.

[legume]

I tested

ATI Technologies Inc RV630 PRO AGP [Radeon HD 2600 PRO AGP] [1002:9587]

via ssh today and it only worked with BusType PCI or PCIE override, thereforce i patched this:


I use a RV670 AGP and on my nforce2 mobo had to set AGP Aperture to 256 in bios to get things working.

Using BusTypePCIE would cause additional problems with ums for me -

Would need to disable exa dfs to avoid corruption, DMAForXv to avoid high CPU and wouldn't be able to use power saving to avoid a hard lock when the clock turned back up.

I haven't actually tested these for a while - now mainly using kms which works well and game perf is better than ums. With ums (kms doesn't use it yet) I have to use module param no_wb=1 to avoid 3D rendering getting blocked and freezing.

Of course I guess it's not going to be the same for every AGP setup, but if you are advocating using PCIE for your users, these are worth knowing about.

[Kano]

@agd5f

That's absolutely stupid! If a user does not see X on startup he needs help. So he has to see something. With Kanotix (lenny based) only xv is accellerated anyway with r600+, so this does not hurt at all. You should write a driver that does not need any override which a normal user do not know at all! It seems that you like to tell every single user to use an override setting instead of makeing a save default.

@legume

Do you want to test a live image with that patch? KMS will not be used anyway...

[legume]

@legume

Do you want to test a live image with that patch? KMS will not be used anyway...

I'll test if you want - it would be interesting to see if the issues I listed are still there or not. It was some time ago I last tried.

If it's like any other Live CD I've ever used testing xv will be a pain as the players never seem to work with normal mpeg2 streams for some reason.

[legume]

I use a RV670 AGP and on my nforce2 mobo had to set AGP Aperture to 256 in bios to get things working.


I just retested this using drm-next and the problem is still there, but only for UMS, KMS so far is working OK.

[Kano]

The patch is not against drm but against the ddx, as i can only compile the 6-12 branch i only tested this.

[legume]

The patch is not against drm but against the ddx, as i can only compile the 6-12 branch i only tested this.

Yea - I was just retesting the aperture thing as I hadn't done so recently - the test was with my setup using AGP - no patch.

The results do make me wonder if your user would have got a working system without possible PCIE problems if kantonix used kms or if he/she changed the aperture size.

[agd5f]

@agd5f

That's absolutely stupid! If a user does not see X on startup he needs help. So he has to see something. With Kanotix (lenny based) only xv is accellerated anyway with r600+, so this does not hurt at all. You should write a driver that does not need any override which a normal user do not know at all! It seems that you like to tell every single user to use an override setting instead of makeing a save default.


Kano, feel free to tweak the drivers as you see fit for your distro. It doesn't make sense to disable AGP for all users because one user is having a problem. Should I disable acceleration for all users since there might be a user that has a problem with acceleration?

[Kano]

Why don't add an extra option to mark cards which needs that workaround?

[legume]

Why don't add an extra option to mark cards which needs that workaround?

Card + mobo rather than just card.

You could detect agpgart loaded before X starts and use BusType PCIE if you wanted and possibly add workarounds for the problems it causes at the same time.

At least AGP users could then choose to try using their hardware as it should be.

I tried your ISO and as expected got corruption and poor xv Cpu usage.

http://www.andyqos.ukfsn.org/kanotix-dfs.png

Both of which could be sorted using xorg.conf.

I notice that you don't use the power saving options anyway, but for me they still cause hard locks with BusType PCIE.

[Kano]

Well the system i tested had VIA chipset, maybe NV is different. Did a BusType "AGP" setting fix your problems?

[bridgman]

Card + mobo rather than just card.

It's even worse than that, depending on card/GPU + mobo/NB + SBIOS version + chipset drivers at the very least.

AGP at higher speeds / lower voltages is also quite sensitive to electrical noise in the system, which tends to get worse over time as parts age.

Doing something like SmartGART (**) would be a pretty big project, but I'm wondering if it might be possible to put in some simple logic that detects when the last Xorg startup was unsuccessful and turns off acceleration next time to improve the chances of a successful startup.

** SmartGART was an AGP diagnostic that ran at startup in our Windows drivers and automatically set options like AGP bus speed and FastWrite

[Hoodlum]

Just wanted to add that if people created / maintained a work-around for every weird piece of hardware there would be significantly less time working on anything else (eg good default compatibility). Not to mention the lacking documentation situation(which has happened before) where people provided a work-around for some odd piece of hardware 5 years ago and no one has any clue what that work-around actually relates to.

[legume]

Did a BusType "AGP" setting fix your problems?

Ahh, yes it does override PCIE and fix them on your patched Kanotix.

Edit: I forgot to say, it's really nice to see a live ISO with mplayer :-)

[Melcar]

I think a good way to remedy this would be to have a way for the user to be able to easily configure driver options. Yes, you can configure xorg.conf, but I'm thinking more of something like Ubuntu's "bullet proof X", where the user would get a screen that gives him a list of relevant options for drivers and lets him make modifications.

[Panix]

I think a good way to remedy this would be to have a way for the user to be able to easily configure driver options. Yes, you can configure xorg.conf, but I'm thinking more of something like Ubuntu's "bullet proof X", where the user would get a screen that gives him a list of relevant options for drivers and lets him make modifications.
What?!? Ubuntu is probably among the worst for ATI open source drivers. AT least, it seems some people are reporting a bug in Mesa.

I have tried Ubuntu 9.10 and I get a crash when trying the Live CD. All I have to do is start Firefox and have 3D effects enabled.

I still say it's a big fat negative that they throw away the support or neglect it to that degree. Also, there's three drivers! Radeon, radeonhd and fglrx? Isn't that right? Crazy. No wonder, it's such a mess.

I'd go with Nvidia if I was getting a desktop video card. At least, I could rely on it working with 3D.

[bridgman]

Also, there's three drivers! Radeon, radeonhd and fglrx? Isn't that right? Crazy. No wonder, it's such a mess.

Three drivers seems to be the most common number for some reason. NVidia has the binary plus nv and Nouveau, Via has a binary plus openchrome and unichrome...

The radeon and radeonhd drivers use the same drm and mesa code, and that code is where most of the new development has been happening for the last year.

[bibi]

I think a good way to remedy this would be to have a way for the user to be able to easily configure driver options. Yes, you can configure xorg.conf, but I'm thinking more of something like Ubuntu's "bullet proof X", where the user would get a screen that gives him a list of relevant options for drivers and lets him make modifications.
Reply With Quote

+1, even if i dont know what is "bullet proof X" ;)

Bibi