Kernel 6.12 with PREEMPT_RT patches enabled - yet POSIX non-realtime (LM22.1)

Yeah I posted that patch in my first post.
You taught me the way royka

A big thanks to royka for making that patch.

Just built 6.13 for x86 (which needed that patch)

Seem like it's got quite good latency on my test machine
(noticeably better than the 6.11 kernel I was running earlier)
 

Peter, 
we have found that they are still issuing patches at kernel.org. Instead of Rokyas patch, try this official 6.13 patch
www.kernel.org/pub/linux/kernel/projects/rt/6.13/
Let us know how it goes.

I did, I built 6.13-rc6-rt3 from kernel.org with the RT patches applied
but roykas patch was still needed.

How do I know it was needed? I stupidly thought I could create /sys/kernel/realtime
like normal file so for experimenting I did not need the patch. So I built the kernel
with the RT patch but not roykas patch first, and it failed in the expected
(POSIX non-realtime) way...
 

PCW wrote:

I did, I built 6.13-rc6-rt3 from kernel.org with the RT patches applied
but roykas patch was still needed.

Sounds like the issue is linuxCNC's method of detecting a RT kernel. Maybe we need to raise an issue and find a better way  at the source rather than Roykas work around.

I just used the sys_fs patch for realtime, that's all that is really required.

What I've notice with the patch set you end up having to set low_latency_desktop within the context of the preempt menu and then full preempt that shows up in the main section of general setup.

Without the patch set all the fully preempt option is in the preempt section under general along with the others as per previous kernel versions.

This happens whether building an arm kernel or a x86_64 kernel.

I went thru this royka during the course of quite a few posts. The guy's right on the money.

rodw wrote:

PCW wrote:

I did, I built 6.13-rc6-rt3 from kernel.org with the RT patches applied
but roykas patch was still needed.
Sounds like the issue is linuxCNC's method of detecting a RT kernel. Maybe we need to raise an issue and find a better way  at the source rather than Roykas work around.
 

I've seen the sys_fs patch in the kernel rt patchset. Name is sysfs__Add__sys_kernel_realtime_entry.patch

Well, I built the stock 6.13-rc6-rt3 with RT patches from kernel.org and it does not create /sys/kernel/realtime

Thats annoying.

I'm sure this is a linuxcnc issue. If you install rt-tests with apt, you could test RT performance using cyclictest which will eliminate linuxcnc's shenanigans.

I have found results although different are broadly simlar to latency-histogram

This patch for LinuxCNC should be sufficient, it will still look for the PREEMPT_RT in uname but removes the check for /sys/kernel/realtime
 

With this environment setting you don't need any patch by the way
LINUXCNC_FORCE_REALTIME=1  latency-histogram