Thursday, 18 February 2021

linux-5.11-ck1, MuQSS version 0.208 for linux-5.11

Announcing a new -ck release, 5.11-ck1  with the latest version of the Multiple Queue Skiplist Scheduler, version 0.208 These are patches designed to improve system responsiveness and interactivity with specific emphasis on the desktop, but configurable for any workload.

This was a resync and bugfix from 5.10-ck1. 

 linux-5.11-ck1:

patch-5.11-ck1.xz

Git tree:

5.11-ck


MuQSS only:

0001-MultiQueue-Skiplist-Scheduler-v0.208.patch

Git tree:

5.11-muqss


Web: kernel.kolivas.org

Enjoy!
お楽しみ下さい
-ck

12 comments:

  1. Thank you very much for this update! This new version works perfectly.

    ReplyDelete
  2. Thank you very much.
    Nice low latency ;)

    ReplyDelete
  3. I have a problem when compile Patch ck

    [root@localhost linux-5.11]# make
    SYNC include/config/auto.conf.cmd
    CC arch/x86/kernel/asm-offsets.s
    UPD include/generated/asm-offsets.h
    CALL scripts/checksyscalls.sh
    CALL scripts/atomic/check-atomics.sh
    DESCEND objtool
    DESCEND bpf/resolve_btfids
    HDRINST usr/include/linux/sched.h
    error: include/uapi/linux/sched.h: leak CONFIG_SCHED_MUQSS to user-space
    make[1]: *** [scripts/Makefile.headersinst:63: usr/include/linux/sched.h] Error 1
    make: *** [Makefile:1293: headers] Error 2

    Helpme pls

    ReplyDelete
    Replies
    1. https://github.com/zen-kernel/zen-kernel/commit/59473b1177948ea41df9340c9010ab958008c880

      Delete
  4. sorry a question: ashmem_linux and binder_linux forr anbox?

    ReplyDelete
  5. Hi, I tried 5.11-ck1 on my Ryzen 9 5950 (16 core/32 threads).
    Here is two observations: schedutil on vanilla 5.11 clocks down to 2,2GHz. -ck1 wont go below 3,4GHz.
    Also with -ck1 I experience Mesa-related micro freezes, especially when the system is unloaded. It runs fine while compiling the kernel ...
    I guess at some interaction with Mesa 21 because the freezes disappear then disabling kwin's compositing. Video in Firefox keeps stuttering anyways.
    My (oldish) dual-core notebook runs nice and snappy though :)

    ReplyDelete
    Replies
    1. Which runqueue sharing have this issue (mc is the default)?
      Since you have Ryzen, plz try rqshare=llc (add this to kernel commandline at boot).
      You can try to experiment with smt too.

      Delete
    2. Could also be that sched idle mesa bug which I submitted a patch for. I don't think it's yet part of the main mesa code. Also don't use schedutil with muqss, you'll never throttle down - use ondemand.

      Delete
    3. Could you point me towards that mesa patch? Also I already run rqshare=llc. Thanks for the hints, I will try ondemand.

      Delete
    4. https://ck-hack.blogspot.com/2020/05/linux-56-ck2-muqss-version-02-for-linux.html

      Delete
    5. That patch seems to have been applied upstream for mesa 21:

      348 if (queue->flags & UTIL_QUEUE_INIT_USE_MINIMUM_PRIORITY) {
      349 #if defined(__linux__) && defined(SCHED_BATCH)
      350 struct sched_param sched_param = {0};
      351
      352 /* The nice() function can only set a maximum of 19.
      353 * SCHED_BATCH gives the scheduler a hint that this is a latency
      354 * insensitive thread.
      355 *
      356 * Note that Linux only allows decreasing the priority. The original
      357 * priority can't be restored.
      358 */
      359 pthread_setschedparam(queue->threads[index], SCHED_BATCH, &sched_param);
      360 #endif
      361 }
      362 return true;
      363 }

      Frequency scaling behaves as expected with ondemand :)
      Also the micro freezes aren't as bad. Also I was running NO_HZ_FULL. Now I switched to NO_HZ_IDLE, but that doesn't change anything. The display (including X11 mouse cursor) still freezes occasionally for like one to two seconds.

      Delete
    6. Unknown, could you tell us more about your base system? Distribution, release, desktop environment, version of Xorg, version of Mesa, and your GPU as well?

      Delete