[PlanetCCRMA] preempt=full in 5.16.0 (was: Akmod for the realtime kernel)

[Fernando Lopez-Lezcano]

On 1/24/22 2:10 PM, Nils Tonnätt wrote:
> Hi,
> 
> I checked out the upcoming kernel for Fedora 35:
> 
> [    0.069100] Kernel command line: 
> BOOT_IMAGE=(hd2,gpt2)/boot/vmlinuz-5.16.2-200.fc35.x86_64 
> root=UUID=7812bc02-a2ea-4335-bc7b-636e1612841a ro rhgb quiet threadirqs 
> preempt=full
> [    0.069632] Dynamic Preempt: full
> 
> This week is kernel test week. So probably next week, we will have a 
> kernel with dynamic preemption in Fedora.

Yes, very good news! Thanks for following up!

I have a custom build of that one (5.16.2) running on both my home 
"music computers" (Fedora 34) and they work great! I need a custom build 
because of a problem with the motu driver - it needs a patch to be able 
to run in "vendor" mode with 64 channels I/O. I also created an akmod 
for the driver and you can add it to any kernel now (anyone interested 
let me know)...

< you can always downloaded the latest untested builds from the Fedora 
koji build server, that is where I get the .src.rpm package that I use 
as a base >

-- Fernando



> Am Di, 11. Jan 2022 um 17:55:04 -0800 schrieb Fernando Lopez-Lezcano 
> <nando at ccrma.Stanford.EDU>:
>> On 1/10/22 3:06 PM, Fernando Lopez-Lezcano wrote: ...
>>
>>     ==== System A: kernel 5.14.8-200.1.motu.fc34.ccrma (preempt=full) 
>>
>> I did a local build for fc34 of the latest Koji 5.16.0 Fedora source 
>> package and installed it in my "System B". At least in that source 
>> package CONFIG_PREEMPT_DYNAMIC is set to "y" (yay!, cross fingers it 
>> stays that way!) and I can switch to preempt=full when booting without 
>> any further tweaks. The kernel no longer prints the status when 
>> booting, but "cat /sys/kernel/debug/check/preempt" confirms the system 
>> is in preempt=full state. The latency numbers I see on my first test 
>> and reported by cyclictest are not quite as good as "System A" 
>> (running 5.14.8), but they are good. -- Fernando PS: Anyone knows how 
>> to convince pipewire to let go of an audio interface? That is, not use 
>> it anymore? I'm trying the Motu driver and I would like to remove the 
>> snd_usb_audio kernel driver to reload the motu driver - long story of 
>> why, and it is not easy as the driver is in use by pipewire...