On Mon, Oct 11, 2021 at 1:36 PM Sebastian Andrzej Siewior
<bigeasy@xxxxxxxxxxxxx> wrote:
>
> On 2021-10-11 15:56:09 [+0200], To Luis Claudio R. Goncalves wrote:
> > So my guess is that the _bh isn't needed.
>
> Correction: If there is at least _one_ reader in softirq all, writer
> need to disable BH which _might_ be the case here.
> The splat is probably because on read_lock holder disables BH later on
> which I don't see yet.
>
> Sebastian
The way I was able to reproduce the problem in 5.10 (and in 5.14 before -rt11)
was:
1. systemctl disable tuned.service
2. Reboot to a kernel-rt with lockdep support
3. systemctl start tuned.service
I used a few tuned profiles, but stuck with 'realtime' for obvious reasons.
If that helps in the process I can investigate which specific
action/configuration
performed by tuned is triggering the lockdep splat.
Luis
