Re: BUG: Scheduler crashes with 4.19 kernels

[Vincent Guittot]

Hi Janne,

On Thu, 30 Jan 2020 at 09:09, Huttunen, Janne (Nokia - FI/Espoo)
<janne.huttunen@xxxxxxxxx> wrote:
>
>
> We have experienced somewhat reproducible scheduler crashes
> and BUGs with a 4.19 kernel. The workload is some kind of
> container related stress-test i.e. it creates and destroys
> containers in a rapid rate. Currently I don't have any more
> details about the specifics of the test case.
>
> Within a few hours of testing, the scheduler code in the
> kernel either triggers an assertion or tries to dereference
> a completely invalid pointer and crashes there.
>

[snip]

>
> Based on the similarity of our crashes and the one reported
> in the commit log, I cherry picked f6783319737f to top of
> 4.19.97 (with 5d299eabea5a as a prerequisite to make the
> fix apply more cleanly). With the patched kernel the test
> ran for three days without a crash in any of the machines.
>
> So, do you think f6783319737f is the proper fix? Should it
> be backported to 4.19.y? Does it require (or benefit from)
> some additional patches that should be backported too?

Yes I think that' you're right and both patches below should be backported
f6783319737f ('sched/fair: Fix insertion in rq->leaf_cfs_rq_list')
5d299eabea5a ('sched/fair: Add tmp_alone_branch assertion')

The other related patch has already been backported to 4.19
c40f7d74c741 sched/fair: Fix infinite loop in
update_blocked_averages() by reverting a9e7f6544b9c

and there is no other patch related to this problem

>