DVB issue since LE switched to kernel 4.9.x

  • jahutchi

    Could you try the build mentioned in post #230?


    It would also be of great help if you could do an usb dump on your x86_64 machine and report your findings to the mentioned mailing list.

    Code
    1. You can get it with wireshark/tshark or tcpdump. See:
    2. https://technolinchpin.wordpress.com/2015/10/23/usb-bus-sniffers-for-linux-system/
    3. https://wiki.wireshark.org/CaptureSetup/USB

    smp maybe you could also provide usb dump with non working 4.9/4.10 kernel (without revert and without patch)?

  • No improvement whatsoever with the buffer increase patch.

    Edited 2 times, last by smp ().

  • I tried to increase the buffer size even further

    Code
    1. stream = DVB_USB_STREAM_BULK(0x82, 128, 8192)

    No improvement as well.

  • I've uploaded a new build, #0108b, which is based on 4.14.10, no commits reverted, but includes a new patch from Linus Torvalds (email).


    RPi / RPi2 / Generic


    It would be good for everyone affected by this DVB issue to test this new build, but in particular X86_64 (Generic) users as I get the impression the kernel developers would prefer results from X86_64 systems due to the Raspberry Pi using a non-standard USB host controller driver. Any USB traces etc. would best come from X86_64 systems to eliminate RPi differences etc.


    Edit: This is the patch: DP2

  • I now tested the patch from Linus Torvalds with kernel 4.9.75 and 4.14.10 and it works great, no issues whatsoever. I can only test on RPi3.

  • Installed LibreELEC on a RPi2 myself.


    I can see that these /proc/sys/kernel/sched_*_granularity_ns files are (unfortunately) not available on this kernel...

    (The kernel have not been compile with the corresponding CONFIG option to export this)

  • I tested the build from post #232 on Raspi3 with DVBSky S960. No artefacts or errors in the stream.


    But:

    when I watch Live-TV while recording (the same channel, like timeshift) there are audio drop-outs. Only while simultaneously watching. When I stop recording and watch the file, there are no audio drop-outs (sounds like that: LibreELEC Testbuilds for RaspberryPi (Kodi 18.0)). This occured with the build from post #221 as well.

  • I tested the build from post #232 on Raspi3 with DVBSky S960. No artefacts or errors in the stream.

    I think you mean post #242, right (the Linus Torvalds patch)?


    Is anyone able to test #0108b on x86_64? jahutchi I think you have a Revo 3700 (Atom D525 quad core) - are you able to test this build?

  • Installed LibreELEC on a RPi2 myself.


    I can see that these /proc/sys/kernel/sched_*_granularity_ns files are (unfortunately) not available on this kernel...

    (The kernel have not been compile with the corresponding CONFIG option to export this)

    I enabled CONFIG_SCHED_DEBUG=y and rebuilt with kernel 4.14.10 and Linus Torvalds patch:

    Code
    1. rpi22:~ # grep -H . /proc/sys/kernel/sched_*_granularity_ns
    2. /proc/sys/kernel/sched_min_granularity_ns:2250000
    3. /proc/sys/kernel/sched_wakeup_granularity_ns:3000000

    If you wish to test this build I can upload it.

  • when I watch Live-TV while recording (the same channel, like timeshift) there are audio drop-outs.

    I noticed the audio dropouts with kernels 4.14/4.15 but didn't want to post about it in this thread because this issue is unrelated to softirq. There is no such problem with kernel 4.9 (with patched and unpatched softirq).

    I did some basic performance testing with different kernels but with otherwise identical builds. For some reason kernels 4.14/4.15 generally perform much worse than 4.9 in LE and have all sorts of audio issues (the dropouts sometimes happen even when there's not much network or i/o activity).

    Edited once, last by smp ().

  • Is anyone able to test #0108b on x86_64? jahutchi I think you have a Revo 3700 (Atom D525 quad core) - are you able to test this build


    Apologies that I've not been able to contribute as much as I would like over the last few days - work & family life getting in the way, but I have been following these threads with interest.


    I have been testing some builds...


    My testing suggests that both reverting 4cd13c2 and using the Linus Torvalds patch have the same positive effect, but that there has been a further regression in later kernels 4.14/4.15? as also suspected by smp - whereby I can no longer watch and record programs at the same time :-(


    Millhouse build #0108b:

    SHAS

    With this build I can get a stable stream when watching live tv. However, if I watch and record at the same time then I get breakups.


    Millhouse build #0101z: JYUK

    Same as above - I get a stable stream when watching live tv. However, if I watch and record at the same time then I get breakups.


    8.2.2 official (i.e. kernel 4.11.12) with just 4cd13c21b207e80ddb1144c576500098f2d5f88 reverted: ANGH

    I can watch live tv and record a further 3 programs at the same time without any breakups


    I am currently building an equivalent 8.2.2 official image with the Linus Torvalds patch included, rather than reverting 4cd13c2 - to confirm for sure whether both patches do indeed have the exact same positive effect.