XBMC UI random framerate drops from 30 FPS to 15FPS

[ungoliant]

Just installed XBMC following instrucctions from rsalveti (http://rsalveti.wordpress.com/2012/01/06/hw-video-decode-and-xbmc-ubuntu-linaro/) and i'm experiencing random FPS drops in both the UI and video decode.

You can get a better idea in this video:
http://www.youtube.com/watch?v=4DcdATFduII&context=C3f58d6bADOEgsToPDskIqTI_an28joYlQPhPC6vvG

[Ricardo Salveti (rsalveti)]

Can you update to the latest driver available (1.7.10.0.1.8-1linaro1), and then disable the drm polling to see if works better?

root@linaro-ubuntu-desktop:~# echo 0 > /sys/module/drm_kms_helper/parameters/poll

After disabling polling SGX is quite stable with my system, and it also behaves better when playing a video (without the sensation of loosing frames from time to time).

As robclark was saying at the IRC, this is probably because the way DVI is detected is by polling the i2c bus checking for a valid EDID, and for some reason the system will wait a timeout at the i2c driver, causing the frame drop.

[warmcat (andy-warmcat)]

If disabling drm poll works around it I have a good idea what it might be, and a possible fix.

[Ricardo Salveti (rsalveti)]

Guess if we're able to probe DVI by a GPIO or something, the code logic can be changed to avoid I2C, but would need to check the schematic. Otherwise we'd need to disable dvi hotplug (or see if we can fix at the i2c code).

[ungoliant (nobill)]

Yes, disabling polling fixes the framerate drops. What effect does the polling disable have on the system?

[warmcat (andy-warmcat)]

DVI panel and so detect polling are disabled in tilt kernels, it's likely something else on hdmi poll path.

[Ricardo Salveti (rsalveti)]

On Fri, Jan 13, 2012 at 12:19 AM, warmcat <email address hidden> wrote:
> DVI panel and so detect polling are disabled in tilt kernels, it's
> likely something else on hdmi poll path.

True, DVI is not even working with this kernel.

[Ricardo Salveti (rsalveti)]

On Fri, Jan 13, 2012 at 12:18 AM, ungoliant <email address hidden> wrote:
> Yes, disabling polling fixes the framerate drops. What effect does the
> polling disable have on the system?

None, and afaik it'll only affect monitor hotplug, so if you don't
change your monitor, you should be fine.

[warmcat (andy-warmcat)]

Right no real effect. But I think I have a fix for this shortly that won't require poll disable.

[ungoliant (nobill)]

Is it possible that this could be causing my LCD to go suddenly go black after a couple minutes? Sound still works, and If i'm using the UI I still hear the sound made by moving around, but the only way to get image back is by restarting lightdm.

[ungoliant (nobill)]

Sorry, by "this" I meant the polling disable workaround, not the bug itself (I've been experiencing this issue since I disabled drm polling on boot)

[Keyan Mobli (persiansown)]

I am also getting screen blackouts. The image returned one time, but it usually doesn't. The image will go black within 10 minutes usually and audio still works.

I don't know if it occurred without the polling change because as soon as I saw the sudden, frequent framerate drops I disabled it, but it definitely occurs with it disabled..

[ungoliant (nobill)]

I found out that the image going out within 10 minutes issue is caused by DPMS. There's some info in arch linux wiki on how to turn DPMS off, I just added it to the xbmc start-up script, and image stays always on on my xbmc.

Of course, if you use XBMC with keyboard and mouse, this shouldn't be the problem.

[Keyan Mobli (persiansown)]

@ungoliant: I tried doing that, but it doesn't seem to work for me. (xset -dpms s off)

Can you show me what you attempted to do to prevent your screen from going black and not returning?

[Anmar Oueja (anmar)]

Andy: Did you put in the fix from comment #8

[ungoliant (nobill)]

Sorry Keyan, hadn't notice your question till now: what I did was adding both at rc.local and xbmc startup script: xset -dpms;xset s off
Its working for me now.

[warmcat (andy-warmcat)]

We also changed the way we adapt DSS in tilt-3.4 that will remove delays on drm poll and edid request that should also impact this.

[Ricardo Salveti (rsalveti)]

This is not the case anymore with the 3.4 based kernel from TI LT's, as commented by Andy on #16. Marking it as fixed released.