[RFC][PATCH 0/3] Tracing, Idle, RCU and such goodness

From: Peter Zijlstra
Date: Fri Aug 07 2020 - 15:34:31 EST


Hi,

Syzbot report:

https://lkml.kernel.org/r/000000000000e3068105ac405407@xxxxxxxxxx

triggered a whole bunch of fallout.

These 3 patches are in reverse order of discovery. With them applied the
reproducer no longer triggers and the machine boots without errors.

I also spend half the day trying to rewrite TRACE_IRQFLAGS, but so far haven't
had success. I have the feeling that there's something slightly simpler hiding
in there, but so far it doesn't want to come out.

Please, look carefully, this is all tricky code, and it's been a long and hot
day here, mistakes are almost guaranteed :/

I also want to remove the _rcuidle() from trace_preemptirq.c, but given
this was quite enough, I didn't touch those.