•   
  • 1
  • 39
  • 40
  • 41(current)
  • 42
  • 43
  • 168
  •   
  Thread Closed
OpenELEC Testbuilds for RaspberryPi (Kodi 16.0)
(2015-08-11, 06:02)Qxi144 Wrote: Except for subtitles. I don't use them. Could you add an option to turn them off? I couldn't find any such option (if there is please point me to the correct setting).

Play a video that has a subtitle. Bring up the OSD (this is Enter or M on a keyboard, the button on your remote control will vary) then click on the Audio icon (3rd from the right in the wiki page), clear the option "Enable subtitles", then click on "Set as default for all media". Now your subtitles will be disabled by default (unless they're forced, of course).
Texture Cache Maintenance Utility: Preload your texture cache for optimal UI performance. Remotely manage media libraries. Purge unused artwork to free up space. Find missing media. Configurable QA check to highlight metadata issues. Aid in diagnosis of library and cache related problems.
Ok so to sumarize:
MMAL player always stalls video but the memory does not seem to leak. Look at debugs above.

OMX player works good with or without deinterlacer enabled but the memory leaks slowly untill it reaches around few mb then kodi restarts without crashlog. Mermory leaks on OMX player no matter if i have enabled or disabled deinterlacer.
During the memory leak there is no writing into kodi.log. If i stop the stream the memory does not return back , well maybee 1mb does. Also a note. I have tried to record a video and played it then with omx player enabled and there is no memory leak with recorded one only when live stream.

Another note: I have tried with tvheadend as backend and also then with iptv simple addon and on both there is memory leakage on OMX player so it is not a tvheadend cause. On iptv simple addon there is some updating on the kodi.log file.
This slowly fills with repeated data and here is a debug log during the play with omx player http://pastebin.com/kPgLnx7D

Ok so i think this is it. Tell me if you need anything more. I hope it will reveal some secret. I assume that omx player will be easier to fix since when memory leak is found then the player will work. As far as i remember there was no leackage on Rpi1 with omx player but i can check if you will find this info usefull.
(2015-08-11, 06:33)illiac4 Wrote: OMX player works good with or without deinterlacer enabled but the memory leaks slowly untill it reaches around few mb then kodi restarts without crashlog. Mermory leaks on OMX player no matter if i have enabled or disabled deinterlacer.

Upload "journalctl --no-pager | paste" after Kodi crashes and doesn't generate a crash log, as this should confirm if the Kodi process is being killed by the OOM killer. Presumably you're monitoring free memory with something like bcmstat.sh, which should be good enough evidence of a memory leak, but journalctl would confirm it for sure (I think).

Also, do the files you are playing have subtitles? There's a known ASS subtitle memory leak (not RPi specific).
Texture Cache Maintenance Utility: Preload your texture cache for optimal UI performance. Remotely manage media libraries. Purge unused artwork to free up space. Find missing media. Configurable QA check to highlight metadata issues. Aid in diagnosis of library and cache related problems.
No stream has no subtitles. Will do the upload. But i think i will do it in the afternoon since it takes some time to bleed till the end Smile

Maybee a note: on Rpi1 omx player with the same stream does not looses memory.
Uptime 2h49m with encrypted MPEG2 channel as test and all hardware acceleration off => no video freeze so far. Kodi %CPU at 55% of a core is quite heavy though. Could not figure out if libmpeg2 or ffmpeg with NEON acceleration is used to decode MPEG2 video. Deinterlacer is sub-par, no yadif-like just bob I think. Gut still says malformed MPEG2/H.264 tripping up decoder or deinterlacer in firmware.

I let it run for some more hours to verify. Box is really really quiet, just the occasional NTP and wlan0 rekeying messages.

A next test shall be a recorded transport stream, in the hope of tripping any firmware bug just by playing a file. Got a fast 128 GB USB stick, shall be enough.

If the problem is indeed in firmware, I'll write a transport stream fuzzer, that systematically tries to crash the GPU decoder/deinterlacer. ;-)
Katastrophentourist
(2015-08-10, 03:31)Milhouse Wrote: @MrNice/@doveman2: Apologies both, for some reason I misinterpreted your posts and thought the problem was that Kodi is now going straight to shutdown instead of offering you an option from the power menu. I hadn't taken on board that Kodi is in fact locking up when the power button is pressed (and only when the shutdown function is "Shutdown", as Suspend works fine). Also "Power off" from the Power menu is working normally.

I can reproduce this with the latest build (once I set my shutdown function to Shutdown), and there's a trac ticket for this issue (trac #16146) so hopefully there will be a fix soon - it's most likely one of the ApplicationMessenger commits introduced in build #0715b.

Thanks. I avoid the problem by putting this in my user remote.xml

<power>XBMC.ActivateWindow(shutdownmenu)</power>

and would have thought that's a more user-friendly approach than having the button Shutdown or Suspend, as it gives the user a chance to back out if they press it accidentally (although I guess Suspend is easy enough to come out of) and is easier than having to navigate to the Quit button (which is less accessible on some skins than others) but I guess it's up to the OE/Kodi team if they want to make that default behaviour.
(2015-08-10, 20:31)Milhouse Wrote:
(2015-08-10, 13:01)illiac4 Wrote: Milhouse will you be so kind and compile latest tvheadend if possible so i can see if maybe new tvheadend is going to solve the problem.

New tvheadend add-on builds: RPi / RPi2

Thanks. I installed that and it got the Muxes to finally change from PEND and the Services populated. I've mapped them to channels and the Broadcast menu now shows on the skin (AppTV) but it doesn't show any EPG as of yet (no channels even, just blank page). I notice that the EPG Source column on the Channels tab is empty for all channels but maybe that's normal for Freeview. I've got only UK:Freeview ticked on the EPG Grabber tab and tried clicking "Trigger OTA EPG Grabber".

The Status - Subscriptions tab does show epgrab running on two muxes (and two other lines showing "keep", which maybe indicates it's finished the grab for those), so maybe the EPG will populate when they've finished. The log doesn't seem to show any evidence of the epggrab running.

kodi.log: https://drive.google.com/file/d/0B1fDI89...sp=sharing
(2015-08-11, 06:33)illiac4 Wrote: OMX player works good with or without deinterlacer enabled but the memory leaks slowly untill it reaches around few mb then kodi restarts without crashlog. Mermory leaks on OMX player no matter if i have enabled or disabled deinterlacer.

Okay, when gpu memory is getting low can you run:
Code:
vcgencmd cache_flush && (vcdbg reloc && vcdbg malloc)

and post output to paste site. After kodi has crashed and is in a bad state, can you stop kodi (systemctl stop kodi) and then run the same command.
That might identify the name of the buffer that has leaked.

If the output contains "corruption" then run again. The way the heap is read asynchronously means you sometimes get an inconsistent result. If you see the same output twice in a row it is likely to be believable.
(2015-08-11, 06:37)Milhouse Wrote:
(2015-08-11, 06:33)illiac4 Wrote: OMX player works good with or without deinterlacer enabled but the memory leaks slowly untill it reaches around few mb then kodi restarts without crashlog. Mermory leaks on OMX player no matter if i have enabled or disabled deinterlacer.

Upload "journalctl --no-pager | paste" after Kodi crashes and doesn't generate a crash log, as this should confirm if the Kodi process is being killed by the OOM killer. Presumably you're monitoring free memory with something like bcmstat.sh, which should be good enough evidence of a memory leak, but journalctl would confirm it for sure (I think).

Also, do the files you are playing have subtitles? There's a known ASS subtitle memory leak (not RPi specific).

There it is:
http://sprunge.us/McNM


And the output of vcgencmd cache_flush && (vcdbg reloc && vcdbg malloc) after the kodi restarted:
http://sprunge.us/FEVP


I will also try to do what popcornmix said and also paste those logs.
(2015-08-11, 06:32)Milhouse Wrote:
(2015-08-11, 06:02)Qxi144 Wrote: Except for subtitles. I don't use them. Could you add an option to turn them off? I couldn't find any such option (if there is please point me to the correct setting).

Play a video that has a subtitle. Bring up the OSD (this is Enter or M on a keyboard, the button on your remote control will vary) then click on the Audio icon (3rd from the right in the wiki page), clear the option "Enable subtitles", then click on "Set as default for all media". Now your subtitles will be disabled by default (unless they're forced, of course).

Thanks! I should have investigated it a bit more I guess. I know most people who post here have problems, but I just wanted to let you know that it works fine for me using your firmware.

Please keep up the work. It's appreciated!

Thanks again.
(2015-08-11, 14:39)popcornmix Wrote:
(2015-08-11, 06:33)illiac4 Wrote: OMX player works good with or without deinterlacer enabled but the memory leaks slowly untill it reaches around few mb then kodi restarts without crashlog. Mermory leaks on OMX player no matter if i have enabled or disabled deinterlacer.

Okay, when gpu memory is getting low can you run:
Code:
vcgencmd cache_flush && (vcdbg reloc && vcdbg malloc)

and post output to paste site. After kodi has crashed and is in a bad state, can you stop kodi (systemctl stop kodi) and then run the same command.
That might identify the name of the buffer that has leaked.

If the output contains "corruption" then run again. The way the heap is read asynchronously means you sometimes get an inconsistent result. If you see the same output twice in a row it is likely to be believable.

Well this one has freezed at around 240mb free memory. The video has stalled and after pressing stop, the gui has freezed and had to stop kodi with the command and took the log. Will try to do it again if needed. I hope that it will reveal something usefull.
Here are the results:
Ok so "vcgencmd cache_flush && (vcdbg reloc && vcdbg malloc)" outbut when memory gets low:
http://sprunge.us/MfaD


and the output after Kodi crash and manual(systemctl stop kodi):
http://sprunge.us/BPPd
I have then stopped kodi once again and took the same log again:
http://sprunge.us/YQYI
I managed to get the EPG to populate by using "Clear Data".

There's a definite improvement in Live TV, as it no longer goes berserk if I toggle full-screen back to the EPG or go to the main menu and then back to Live TV.

I did notice a fair bit of stuttering and dropped/skipped frames however. When I first checked, I think it was around 479 skipped but I figured that might have been as a result of toggling full-screen, so I restarted the channel and I had a lot less after that but still noticed some gaps in audio and stutters in the video. I'm using MMAL acceleration only and had deinterlace set to Auto. I tried with deinterlace off and that seemed to reduce them but not entirely and even when starting the channel again with deinterlace disabled, there were quite a few gaps in audio for the first few minutes, although it settled down after that. On the third try, I think there were a total of 73 skipped frames (1 dropped) by the time the programme finished and whilst 49 of those occurred with deinterlace on and after I disabled it, that only increased slightly to 52 over the next few minutes, later on there were a couple of obvious stutters/gaps and on checking it had jumped to 63 and the next time, 72.

The log is 554MB but I've compressed it down to 31MB and shared it here: https://drive.google.com/file/d/0B1fDI89...sp=sharing

I'm thinking that with a CRT TV (which this RPi v2 will be used with) there's no point in having deinterlace or sync playback to display enabled. Is that correct, as if so it might at least eliminate a couple of potential problem features and give my brother less issues?
(2015-08-11, 16:26)illiac4 Wrote: and the output after Kodi crash and manual(systemctl stop kodi):
http://sprunge.us/BPPd

Okay, there are some clues in there. The video_decode and image_fx pools are still present, which suggests one of those components hung.
The "qpu_malloc" shows that advanced deinterlace was in progress when it hung.

Can you switch from advanced deinterlace to bob and then grab the same log if it hangs. Then disable deinterlace completely and grab the same log if it hangs.
(2015-08-11, 19:13)popcornmix Wrote:
(2015-08-11, 16:26)illiac4 Wrote: and the output after Kodi crash and manual(systemctl stop kodi):
http://sprunge.us/BPPd

Okay, there are some clues in there. The video_decode and image_fx pools are still present, which suggests one of those components hung.
The "qpu_malloc" shows that advanced deinterlace was in progress when it hung.

Can you switch from advanced deinterlace to bob and then grab the same log if it hangs. Then disable deinterlace completely and grab the same log if it hangs.

Without interlacer will do tomorrow morning since it takes more than hour to hang.

With bob deinterlacer:
One log at around half of memory already leaked:
http://sprunge.us/HNHh
At around 200mb free:
http://sprunge.us/hWcR
At around 100mb free:
http://sprunge.us/gHVe
At around 50mb free:
http://sprunge.us/IaLM

At freeze it has had 36mb left and could not even ssh into pi as it has fully freezed. Picture has freezed and this is it, no response, no remote responce no ssh. So freezed till death.
Quote:2. Fix for power button shutdown

What did you mean.
Is it a predefined gpio pinout ?
And when yes, is it implemented, that another gpio to specify witch onboard instruments?

I ask you because I would like implemented hardware-keys to control the openelec-raspi in the kitchen. :-)
  •   
  • 1
  • 39
  • 40
  • 41(current)
  • 42
  • 43
  • 168
  •   
  Thread Closed
 
Thread Rating:
  • 10 Vote(s) - 5 Average



Logout Mark Read Team Forum Stats Members Help
OpenELEC Testbuilds for RaspberryPi (Kodi 16.0)510