vblank interrupts increase during 'DPMS off'

[Johannes Hessellund]

Binary package hint: xserver-xorg-video-ati

Using freshly updated Jaunty.

Radeon driver generates more interrupts during screenblanks ('xset dpms force off') than not.

Measuring with powertop 1.11 my hardware generates approx 10 wakeups/s .
When issuing the 'xset dpms force off' wakeups increases to approx 60/s.

This seems to be radeon driver generating vblank interrupts during dpms off. Thus preventing the CPU to enter C3-C4 state.

=> increased powerconsumption during dpms!

Using repos opensource radeon driver. Compiz disabled

Hardware: Thinkpad T42, Radeon 7500.

For more info read dri-devel list: http://<email address hidden>/msg37843.html
Not sure if any solution was found.

Reproduce:
1. Disable compiz
2. On terminal: sudo powertop
3. On another terminal: xset dpms force off

[agd5f (agd5f)]

Please attach your xorg log and config. Also, what kernel and drm modules are you using?

[Johannes Hessellund (osos)]

Created an attachment (id=25001)
Xorg.log with thinkpad attached to port replicator

[Johannes Hessellund (osos)]

Created an attachment (id=25002)
xorg.conf

[Johannes Hessellund (osos)]

Kernel: Ubuntu Jaunty's Linux 2.6.28-11.42
DRM version (from 'dmesg | grep drm'):
[ 20.392854] [drm] Initialized drm 1.1.0 20060810
[ 20.418107] [drm] Initialized radeon 1.29.0 20080528 on minor 0
[ 21.055228] [drm] Setting GART location based on new memory map
[ 21.055238] [drm] Loading R100 Microcode
[ 21.055294] [drm] writeback test succeeded in 2 usecs

[Johannes Hessellund (osos)]

I noticed it has been discussed on the mailinglist:
http://<email address hidden>/msg37843.html

Did you ever find a solution?

[Johannes Hessellund (osos)]

Binary package hint: xserver-xorg-video-ati

Using freshly updated Jaunty.

Radeon driver generates more interrupts during screenblanks ('xset dpms force off') than not.

Measuring with powertop 1.11 my hardware generates approx 10 wakeups/s .
When issuing the 'xset dpms force off' wakeups increases to approx 60/s.

This seems to be radeon driver generating vblank interrupts during dpms off. Thus preventing the CPU to enter C3-C4 state.

=> increased powerconsumption during dpms!

Using repos opensource radeon driver.

Hardware: Thinkpad T42, Radeon 7500.

[Johannes Hessellund (osos)]

Upstream bugreport: https://bugs.freedesktop.org/show_bug.cgi?id=21321

[Johannes Hessellund (osos)]

Is this a bug in the xorg radeon driver, or in the drm module ?

Any options I could try to workaround this ?

Any other informations I should provide you ?

[Johannes Hessellund (osos)]

During idle use with LVDS on (brightness as low as possible), my laptop uses between 8,3W and 8.8W.
During idle with lvds turned off (DPMS off), it uses 8,4W.
Measured with powertop.

This means I have no real powersaving, by letting it turn off the monitor !

Normally I could win ½ hour on batter by letting dpms turn off the lvds.

Please look into this.

[Johannes Hessellund (osos)]

Drm version!

Is it the libdrm package you mean?

My libdrm version: 2.4.5

Is this fixed in later versions ?

[Johannes Hessellund (osos)]

This might be a bug in libdrm!?

Was this fixed in git ? I do see some vblank fixes committed!

[Michel Dänzer (michel-daenzer)]

(In reply to comment #6)
> Is this a bug in the xorg radeon driver, or in the drm module ?

The X driver I think, apparently it considers the CRTC disabled but it's still generating vertical blank interrupts.

> Any options I could try to workaround this ?

None that I know of unfortunately.

[Johannes Hessellund (osos)]

(In reply to comment #8)
> The X driver I think, apparently it considers the CRTC disabled but it's still
> generating vertical blank interrupts.

Is this only a bug on specific hardware models or a generel bug?

I think the 6.9.0 driver did properly disable my hardware (radeon 7500), but now it doesn't!

I will be glad to help test any solution provided?
This could give me approx. ½ hour more battery-time in my work conditions! Because I 'dpms off' as often as posible to save power.

[Bryce Harrington (bryce)]

Hi osos,

Please attach the output of `lspci -vvnn`, and attach your /var/log/Xorg.0.log (and maybe Xorg.0.log.old) file from after reproducing this issue. If you've made any customizations to your /etc/X11/xorg.conf please attach that as well.

[This is an automated message. Apologies if it has reached you inappropriately; please just reply to this message indicating so.]

[Michel Dänzer (michel-daenzer)]

(In reply to comment #9)
> (In reply to comment #8)
> > The X driver I think, apparently it considers the CRTC disabled but it's still
> > generating vertical blank interrupts.
>
> Is this only a bug on specific hardware models or a generel bug?

Not sure but I suspect it affects all Radeons, at least all pre-AVIVO ones.

> I think the 6.9.0 driver did properly disable my hardware (radeon 7500), but
> now it doesn't!

Interesting, sounds like the old driver really disabled the CRTC while the current one leaves it somewhat enabled. Alex, does this give you any ideas?

[agd5f (agd5f)]

(In reply to comment #10)
>
> Interesting, sounds like the old driver really disabled the CRTC while the
> current one leaves it somewhat enabled. Alex, does this give you any ideas?
>

We never actually disabled the crtcs, just blank them it looks like. Same thing even with older versions of the driver, so I don't know how this ever worked in the past. I've just pushed a fix:
7d9f643ae3d07e51e644a5979ca90bc2c102bc89

[Johannes Hessellund (osos)]

(In reply to comment #11)
> (In reply to comment #10)
> >
> > Interesting, sounds like the old driver really disabled the CRTC while the
> > current one leaves it somewhat enabled. Alex, does this give you any ideas?
> >
>
> We never actually disabled the crtcs, just blank them it looks like. Same
> thing even with older versions of the driver, so I don't know how this ever
> worked in the past. I've just pushed a fix:
> 7d9f643ae3d07e51e644a5979ca90bc2c102bc89
>

How do I compile a version to test with my ubuntu 9.04 setup?

[agd5f (agd5f)]

(In reply to comment #12)
>
> How do I compile a version to test with my ubuntu 9.04 setup?
>

basic instructions here:
http://www.botchco.com/agd5f/?page_id=2

[Johannes Hessellund (osos)]

This might be fixed with commit 7d9f643ae3d07e51e644a5979ca90bc2c102bc89 in radeon driver git.

This patch should be backported to ubuntu's ati driver!

[Johannes Hessellund (osos)]

Tormod Volden was so kind to put latest git in his launchpad ppa.
https://launchpad.net/~tormodvolden/+archive/ppa

I have testet your fix. WORKS GREAT! ;-)

Power usage has dropped during dpms off. My hardware now uses approx 6.5 W when dpms off, thats 2 W less than before.

Thank you for your great work.

[Johannes Hessellund (osos)]

Alex commit really did fix this bug.

Power usage during 'dpms off' on my hardware is now approx. 6.5 W. Thats 2W less than before! ;-)

The fix should really get into ubuntu's ati driver!

Until then, Tormod Volden has latest git version in his ppa:
https://launchpad.net/~tormodvolden/+archive/ppa

[Bryce Harrington (bryce)]

Uploaded to karmic.

[Launchpad Janitor (janitor)]

This bug was fixed in the package xserver-xorg-video-ati - 1:6.12.2-0ubuntu2

---------------
xserver-xorg-video-ati (1:6.12.2-0ubuntu2) karmic; urgency=low

  * Add 105_pre_avivo_vblank_interrupt.patch: For pre-avivo hardware, -ati
    only blanks crtcs. This cherrypicked patch makes it disable them.
    (LP: #364639)

 -- Bryce Harrington <email address hidden> Mon, 04 May 2009 16:30:25 -0700

[Bryce Harrington (bryce)]

If you wish this backported to jaunty, please find some additional people affected by this bug, and verify that the patch now uploaded to karmic solves it.