Package Base Details: linux-lqx

Git Clone URL: (read-only, click to copy)
Submitter: akurei
Maintainer: sir_lucjan (damentz)
Last Packager: damentz
Votes: 128
Popularity: 1.61
First Submitted: 2011-08-08 16:08
Last Updated: 2020-04-02 15:28

Latest Comments

« First ‹ Previous ... 4 5 6 7 8 9 10 11 12 13 14 ... Next › Last »

damentz commented on 2018-11-20 14:43

@Terence, I briefly experimented again with 1000hz but it was a huge drain on battery life. MuQSS 0.173 (and possibly 0.180 with 4.19), doesn't support tickless idle correctly, and ends up feeding cpufreq with bad idle data. The periodic forces cpufreq to ingest idle time accounting data, letting the cpu properly rest while idle, but the periodic tick prevents the CPU from staying in the CPU's lowest state (C6 / C7). It's measurable, and I mentioned some data points in my commit reverting back to 250hz here:

Until I figure out a better compromise with 1000hz, you'll need to manually go in on every build and raise the frequency to 1000hz for your system. It's perfectly fine for desktops but a huge problem on laptops. Both my personal and work laptops lose about 20%-30% battery life for idle / casual loads. In my testing, 250hz preserves cpu temperatures as well as tickless idle, so that'll work for now.

Also, I re-enabled IRQ time accounting as it seems that it was originally added so that the process scheduler (CFS in this case), had more accurate data about how much cpu time each task consumed. It should have some benefits for MuQSS as well.

Agafron commented on 2018-11-19 19:30

@sir_lucjan, ok, thanks!

sir_lucjan commented on 2018-11-19 19:26


I don't think so.

Agafron commented on 2018-11-19 19:24

@sir_lucjan, ok, need I reinstall package?

$ uname -r


sir_lucjan commented on 2018-11-19 19:12



Agafron commented on 2018-11-19 19:06

@damentz, I have some diff of versions:

1 aur/linux-lqx 4.18.19_2-2 [installed: 4.18.20_1-1] (114) (1,86)

A desktop oriented kernel and modules with Liquorix patches

2 aur/linux-lqx-docs 4.18.19_2-2 [installed: 4.18.20_1-1] (114) (1,86)

Kernel hackers manual - HTML documentation that comes with the linux-lqx.

3 aur/linux-lqx-headers 4.18.19_2-2 [installed: 4.18.20_1-1] (114) (1,86)

Its PKGBUILD mistake?

damentz commented on 2018-11-18 18:39

@Terence, that's good information. While you were comparing results, I was measuring the overhead of 1000hz vs 250hz.

With 250hz, the timer consumes about 2-3ms, every second. Increasing this to 1000hz almost quadruples it. At idle, the timer tick consumes 8ms/s, but 5ms/s when the system is under full load (highest frequency). 8ms is less than 1% of the cpu usage of one core on a processor.

On the flip-side, one change I made with switching to 250hz is to re-instate a sampling down factor on ondemand. While the system is idle, it consumes about 3-5ms/s worth of processing to determine the next frequency. While the system is under full load, or near full load, ondemand consumes less than 1ms/s.

One thing I was wondering if you could try is adding rqshare=mc to your kernel parameters. Liquorix is currently configured with smt runqueue sharing, which improves throughput by looking for tasks in the order of cache locality. This might be influencing your underruns in a bad way.

Terence commented on 2018-11-18 15:22

@damentz I tried your suggestion but it didn't seem to improve things. I have between 3 and 4 times less xruns when using 1000Hz tick rate compared to the now default 250Hz. My test settings were a sample rate of 48000 Hz with 64 as the buffer size giving a latency of 1.3ms. I used the performance governor. I normally use 128 as the buffer size but I increased it in order to get more frequent xruns. An other observation is the schedutil is closer to the performance governor in terms of xruns at 1000Hz but there is a bigger gap between both at 250Hz.

I'm still wondering if it wouldn't make sense to keep 250hz as the default as my usage is quite an edge case I would say.

damentz commented on 2018-11-17 19:15

@Terence, try setting rr_interval back to 3 in /proc/sys/kernel. If that doesn't help, then it's the kernel tick frequency.

And thanks for reporting. Con has mentioned that the kernel tick frequency shouldn't matter and that MuQSS is "tickless". This may prove the opposite and that you still need a high frequency tick rate to support soft-realtime applications properly.

Let me know which of the two settings helps you the most - I'll update the configuration accordingly.

Terence commented on 2018-11-16 12:27

@damentz your changes introduced by while being saner defaults degrades realtime audio/midi performances (more frequent xruns). Would only changing back the tick to 1000hz locally enough?