Re: [PATCH v2] sched/all: Change all BUG_ON() instances in the scheduler to WARN_ON_ONCE()

[Mel Gorman]

On Sun, Aug 21, 2022 at 01:28:58PM +0200, Ingo Molnar wrote:
> 
> * Mel Gorman <mgorman@xxxxxxx> wrote:
> 
> > For the rest, I didn't see obvious recovery paths that would allow the 
> > system to run predictably. Any of them firing will have unpredictable 
> > consequences (e.g. move_queued_task firing would be fun if it was a 
> > per-cpu kthread). Depending on which warning triggers, the remaining life 
> > of the system may be very short but maybe long enough to be logged even 
> > if system locks up shortly afterwards.
> 
> Correct. I'd prefer to keep all these warnings 'simple' - i.e. no attempted 
> recovery & control flow, unless we ever expect these to trigger.
> 

I generally hope we never expect a warning to trigger until of course it
does :P

> I.e. instead of adding a 'goto' I'd prefer if we removed most of the ones 
> you highlighted. But wanted to keep this first patch simple.
> 

The only one I would push back on is claming cpupri_find_fitness() so it
does not access outside the bounds of an array after a warning triggers.

For the rest, I've no strong objection to recovering given the nature of
the BUGs you are converting. If any of them *did* trigger, it would be
more important to fix the issue than recover the warning. So other than
the out-of-bounds array access which I'd like to see clamped just in case;

Acked-by: Mel Gorman <mgorman@xxxxxxx>

-- 
Mel Gorman
SUSE Labs