xf86-video-ati driver without KMS : no X

Spartakus · 2012-07-11 03:10:48

Hi everyone.

I was using the catalyst driver since the xf86-video-ati doesn't work well with my laptop. The problem is my gpu fan seems to spin at full speed all the time (i'll report soon that bug). But that's not my problem here : i really try to use the radeon driver for various reasons and so i decided to try it without KMS to see if it work without the fan problem.
With KMS, my laptop is booting just fine but, without, X doesn't work (+ no tty's) : just a black screen without errors.

My laptop is an asus N81vp and the gpu is a ATI Mobility HD4650.
Here's some info :

uname -a
Linux Spartakus 3.4.4-3-ARCH #1 SMP PREEMPT Tue Jul 3 14:36:44 UTC 2012 x86_64 GNU/Linux

pacman -Qi xf86-video-ati
Nom                   : xf86-video-ati
Version               : 1:6.14.6-1

rc.conf
dmesg

I "purge" the KMS way : remove the radeon in the modules section in mkinitpcio.conf, rebuild the initramfs and add nomodest (radeon.modeset=0 make the same black screen).
I followed the arch wiki, and change the "xorg.conf" as mentionned :

/etc/X11/xorg.conf.d/20-radeon.conf

Section "Device"
    Identifier "Radeon"
    Driver "radeon"
       Option      "DynamicPM"          "on"
       Option      "ClockGating"        "on"
EndSection

I don't know why the wiki tells us now to use create this file and not use a simple /etc/X11/xorg.conf as in the past but... i did as the wiki tells me

Here's the Xorg logs :
Xorg0.log with KMS
Xorg0.log without KMS

From what i see in there, the options in xorg.conf works :

[    15.302] (**) RADEON(0): Option "ClockGating" "on"
[    15.302] (**) RADEON(0): Option "DynamicPM" "on"

and there is a AIGLX error :

[    16.999] (II) AIGLX: Screen 0 is not DRI2 capable
[    17.327] (EE) AIGLX error: r600 does not export required DRI extension
[    17.327] (EE) AIGLX: reverting to software rendering

and at the end this

[    18.347] (II) RADEON(0): Power Mode Switch

So i bet these lines has something to do with my problem but my (little) linux knowledge stops here....
I really wanted to try this way to see if powersaving works without KMS because of the fan problem with KMS (No problem of fan with the Catalyst). And i want to know if it works before sending a bug report to know the fan problem is a driver problem or just appears with KMS enable.

Hope that someone could help me on that,
Thanx.

msx · 2012-07-11 10:02:35

Howdy archer,
these lines

Option      "DynamicPM"          "on"
Option      "ClockGating"        "on"

have nothing to do with KMS, they instruct the GPU to lower energy requeriments, here is a fine explanation of what do they both: http://phoronix.com/forums/showthread.p … -in-Radeon.

Regarding "I don't know why the wiki tells us now to use create this file and not use a simple /etc/X11/xorg.conf as in the past", that's because the last versions of Xorg server descentralized /etc/X11/xorg.conf so now the right way to configure it is adding config modules to /etc/X11/xorg.conf.d directory for each device you want to configure. Said that, you still can modify the main xorg.conf file, it's just the new way is a cleaner way to tweak xorg.conf file without messing directly with it.

Just out of curiosity: why are you trying to use the opensource module without KMS? KMS rocks!

Last edited by msx (2012-07-11 10:04:32)

Gusar · 2012-07-11 12:04:39

How do you figure you fan problems have anything to do with KMS? I don't see any connection, so I'm curious how you made it.

I don't know why the wiki tells us now to use create this file and not use a simple /etc/X11/xorg.conf as in the past

People nowadays have this weird notion that xorg.conf is some icky do-not-touch you-will-burn-in hell-if-you-do thing. When in reality there's no difference between xorg.conf and a file in xorg.conf.d/. All you need to know is that xorg.conf is parsed last. So xorg.conf and xorg.conf.d/99-something.conf are *completely* equivalent.

Saying that not using xorg.conf is "the right way" or in any way "cleaner" is pretty much nonsense.

msx · 2012-07-11 17:21:37

Gusar wrote:

How do you figure you fan problems have anything to do with KMS? I don't see any connection, so I'm curious how you made it.
I don't know why the wiki tells us now to use create this file and not use a simple /etc/X11/xorg.conf as in the past
People nowadays have this weird notion that xorg.conf is some icky do-not-touch you-will-burn-in hell-if-you-do thing. When in reality there's no difference between xorg.conf and a file in xorg.conf.d/. All you need to know is that xorg.conf is parsed last. So xorg.conf and xorg.conf.d/99-something.conf are *completely* equivalent.
Saying that not using xorg.conf is "the right way" or in any way "cleaner" is pretty much nonsense.

Hi Gusar, I don't want to enter any discussion but I don't agree with some of the things you say:
1. "People nowadays have this weird notion that xorg.conf is some icky do-not-touch you-will-burn-in hell-if-you-do thing."
While this is true, it's easy to understand the fear a newcomer to GNU/Linux may have to touch this particular file since if it's wrong edited the X server will never show up, and for a newcomer who's still learning his/her first steps at the console, it's a killer that "will render its system unusable" - I know it's not that much, but still is how most of them perceive the whole thing.
2. "Saying that not using xorg.conf is "the right way" or in any way "cleaner" is pretty much nonsense."
I'm still convinced that in this particular case order defeats chaos, understanding "order" by using a modular way to configure Xorg and chaos by throwing every possible option used to the same file. It never hurts keep things organized, it's allways easier to make changes, track errors, correct typos and so on.

Said that I'm still carry an ancient xorg.conf with everything-but-the-kitchen-sink in it I first configured some years ago and to which I do some tweaking from time to time but never bothered to "modularize" it, LOL - the golden rule "if it's not broken, don't fix it" applies here.

Last edited by msx (2012-07-11 17:28:32)

AndyRTR · 2012-07-11 17:27:27

Make sure you get it running in KMS mode. UMS won't be supported anymore very soon.

Gusar · 2012-07-11 18:06:48

msx wrote:

1. "People nowadays have this weird notion that xorg.conf is some icky do-not-touch you-will-burn-in hell-if-you-do thing."
While this is true, it's easy to understand the fear a newcomer to GNU/Linux may have to touch this particular file since if it's wrong edited the X server will never show up

It makes no difference if a faulty edit is in xorg.conf or xorg.conf.d/90-something.conf, so why give xorg.conf special meaning?
The only way it would make sense is if a user would encounter an already written xorg.conf and would be vary of editing that, instead opting to write a new file so as to not disturb the original. But that's not the case, there is no xorg.conf by default.

msx wrote:

2. "Saying that not using xorg.conf is "the right way" or in any way "cleaner" is pretty much nonsense."
I'm still convinced that in this particular case order defeats chaos, understanding "order" by using a modular way to configure Xorg and chaos by throwing every possible option used to the same file.

This would have merit if there was tons to configure. But there isn't - a user will write at most two sections. Not much difference regarding order and chaos if these two sections are in xorg.conf or somewhere in xorg.conf.d/
Of course there's a certain tidiness in a split config, but a unified one is far from chaos.

Spartakus · 2012-07-11 18:39:08

Hi there,

First thanx for the xorg.conf explanation.

msx wrote:

these lines
Option "DynamicPM" "on"
Option "ClockGating" "on"
have nothing to do with KMS, they instruct the GPU to lower energy requeriments

For that part, these options doesn't work with KMS :

From Xorg0.log with KMS

[    16.978] (WW) RADEON(0): Option "DynamicPM" is not used
[    16.978] (WW) RADEON(0): Option "ClockGating" is not used

msx wrote:

Just out of curiosity: why are you trying to use the opensource module without KMS? KMS rocks!

As I said, i'm experiencing some powersavings problems with KMS, so i tried to figured if maybe the opensource driver works for me with a different powersaving method (without KMS). My problem with the open source driver isn't new (https://bbs.archlinux.org/viewtopic.php?id=122842), that's why i was using the catalyst since, and i'm trying to use the opensource driver without KMS now.
I can developp the problem if you wanted, but that didn't solve my problem here without KMS .

Gusar · 2012-07-11 18:59:26

KMS isn't a powersaving method. It stands for "kernel modesetting", and does what it says - modesetting.

Spartakus · 2012-07-11 19:08:51

I know that, but still, powersaving doesn't work the same way with an without KMS no ?

Arch wiki ATI wrote:

Powersaving
The powersaving part is totally different with and without KMS.

I want to "isolate" my fan screaming problem and powersaving clearly doesn't work with KMS for me, so i figured maybe i'll not have the same problem without KMS, that's all.
But maybe it's just stupid assumption so tell me !

Gusar · 2012-07-11 19:35:37

Oh, now I get it: It's not about KMS, it's about powersave in the kernel module vs. powersave in the userspace driver.

The userspace part is outdated, it hasn't been worked on since forever and is for sure less capable than the kernel part. So it's not the solution you're looking for. As UMS is gone upstream (the 6.14.6 release is the last one that still has it), it wouldn't surprise me if userspace powersave is gone with it.

Spartakus · 2012-07-11 19:53:01

Hmm that's interesting. So maybe the wiki is outdated too ?
I'm thinking that kinda everybody goes with KMS now because it's the way to go. So for me, no solution without KMS ....
I'm gonna report my bug (the part that with KMS i've got a screaming gpu fan and no working powersaving) and hope there's a fixing to that.
Thx for the reply man.

Spartakus · 2012-07-11 20:07:45

As you said :

http://www.phoronix.com/scan.php?page=n … px=MTEyOTg

ConnorBehan · 2012-07-11 20:46:46

Spartakus wrote:

I "purge" the KMS way : remove the radeon in the modules section in mkinitpcio.conf, rebuild the initramfs and add nomodest (radeon.modeset=0 make the same black screen).

FWIW, you seem to be missing a step - adding the radeon module back to rc.conf. Because the way you describe it, the radeon.modeset=0 option would not even be used because radeon would not get loaded at all, leading to quite a broken experience with X.

Also about your issue, try changing the value of radeon.dynclks (this looked like the most relevant option for you in the output of "modinfo radeon"). The userspace power management options do not get used with KMS because one would've already had to already give this information to the kernel module.

Spartakus · 2012-07-11 23:51:39

I thought Xorg was loading the radeon module withtout putting radeon in modules in the rc.conf
And indeed it xas loaded in my test without KMS.

From Xorg0.log without KMS

[    15.171] (II) LoadModule: "radeon"

But i tried anyway with radeon in rc.conf
So no more AIGLX error

[    15.674] (==) AIGLX enabled

and no more

Spartakus wrote:

[ 18.347] (II) RADEON(0): Power Mode Switch

But still a nice black screen, so...

About the dynclks option, i think there's two way of adding it no ?

Either in the /boot/grub/menu.lst like that : radeon.dynclks=1
Or in a /etc/modprobe.d/radeon.conf like that : options radeon dynclks=1

I think you are talking about my fan problem no ? (With KMS)
First way works fine for me (i mean no crash ) but there's no change in terms of powersaving and the use of both dynpm and profile powersaing methods keeps crashing my laptop with that option.

ConnorBehan · 2012-07-12 00:07:06

Spartakus wrote:

I thought Xorg was loading the radeon module withtout putting radeon in modules in the rc.conf
And indeed it xas loaded in my test without KMS.
From Xorg0.log without KMS
[    15.171] (II) LoadModule: "radeon"

Ah, ambiguous wording there. By "module" the log file just means "/usr/lib/xorg/modules/drivers/radeon_drv.so". X doesn't load kernel modules.

Spartakus wrote:

About the dynclks option, i think there's two way of adding it no ?
Either in the /boot/grub/menu.lst like that : radeon.dynclks=1
Or in a /etc/modprobe.d/radeon.conf like that : options radeon dynclks=1
I think you are talking about my fan problem no ? (With KMS)
First way works fine for me (i mean no crash ) but there's no change in terms of powersaving and the use of both dynpm and profile powersaing methods keeps crashing my laptop with that option.

Yep, those two ways of adding it should do the same thing. I was just guessing that it could help your fan problem but I wasn't sure.

Spartakus · 2012-07-12 00:18:39

Thx for trying.
For my part, i will report my bug to http://bugs.freedesktop.org. That's the right place to do so, no ? (First bug report for me )

Arch Linux

#1 2012-07-11 03:10:48

xf86-video-ati driver without KMS : no X

#2 2012-07-11 10:02:35

Re: xf86-video-ati driver without KMS : no X

#3 2012-07-11 12:04:39

Re: xf86-video-ati driver without KMS : no X

#4 2012-07-11 17:21:37

Re: xf86-video-ati driver without KMS : no X

#5 2012-07-11 17:27:27

Re: xf86-video-ati driver without KMS : no X

#6 2012-07-11 18:06:48

Re: xf86-video-ati driver without KMS : no X

#7 2012-07-11 18:39:08

Re: xf86-video-ati driver without KMS : no X

#8 2012-07-11 18:59:26

Re: xf86-video-ati driver without KMS : no X

#9 2012-07-11 19:08:51

Re: xf86-video-ati driver without KMS : no X

#10 2012-07-11 19:35:37

Re: xf86-video-ati driver without KMS : no X

#11 2012-07-11 19:53:01

Re: xf86-video-ati driver without KMS : no X

#12 2012-07-11 20:07:45

Re: xf86-video-ati driver without KMS : no X

#13 2012-07-11 20:46:46

Re: xf86-video-ati driver without KMS : no X

#14 2012-07-11 23:51:39

Re: xf86-video-ati driver without KMS : no X

#15 2012-07-12 00:07:06

Re: xf86-video-ati driver without KMS : no X

#16 2012-07-12 00:18:39

Re: xf86-video-ati driver without KMS : no X

Board footer