On 2021-01-28 07:28:35 (+0100), Ahmed S. Darwish wrote:

On Sat, Jan 23, 2021 at 12:59:29PM +0100, David Runge wrote:
...
Can you please send the full kernel log, with boot parameter
"firewire_ohci.debug=-1", for mainline v5.4.91 vs. v5.4.91-rt50?

kernel log for mainline:
https://pkgbuild.com/~dvzrv/bugs/2021/01/linux-5.4.91-kernel.log

kernel log for rt (no dice :-/):
https://pkgbuild.com/~dvzrv/bugs/2021/01/linux-5.4.91.50.arch1-kernel.log

Please also make sure to use the exact same defconfig for both, except
with CONFIG_PREEMPT_RT=y for the -rt kenrel of course.

config diff:
https://pkgbuild.com/~dvzrv/bugs/2021/01/linux-5.4.91-lts_changed_vs_rt-lts.patch

This will help in comparing apples to apples, as you were posting the
results of different base kernels (vs. the -rt versions), and different
defconfigs (vs. -rt) within such kernels.

The configuration for the 5.4.91 stock kernel has not been updated since
the 5.4.83, which is maybe a bit misleading.
Generally I enable anything that I can in the config for the realtime
kernel. However, some settings of course get switched off, as they are
not compatible.
The diff above is for a custom version I just made, just to make sure
and clear things out.

Good luck,

Thanks, seems it's needed!

P.S. Since this maybe is a timing issue, and ohci bus reset handling
occurs at SCHED_OTHER workqueu context, were you by any chance running
unreasonably-intensive realtime workloads by the time the driver was
loaded?

Not that I am aware of. The driver neither initializes the device during
boot nor when later reloading.
There are no intense services running. The system is fairly
multi-purpose, so most things are started on demand.

Thanks!

David

-- 
https://sleepmap.de