Re: [PATCH] mm/page_alloc: Use write_seqlock_irqsave() instead write_seqlock() + local_irq_save().

[Sebastian Andrzej Siewior <bigeasy@linutronix.de>]

On 2023-06-23 12:40:24 [+0200], Petr Mladek wrote:
> > but I would rather suggest lockdep annotation instead of disabling
> > either preemption or migration. If I read the comment on top right, this
> > API isn't open for the public.
> 
> It might work when mm people agree to explicitly call
> migrate_disable() in __build_all_zonelists(). I mean
> to call there:
> 
> 	migrate_disable();
> 	printk_deferred_enter();
> 	write_seqlock_irqsave(&zonelist_update_seq, flags);
> 
> Plus it would require some comments.

right but I wanted to move it after write_seqlock_irqsave().

> > To be honst, I would suggest to work towards always printing in a thread
> > with an emergency console than this deferred/ safe duckt tape.
> 
> I have used this argument many years. But it is not much convincing
> after 10 years. Note that John has sent 1st RFC 4 years ago and we
> still do not have the kthreads :-/

Yes, it makes me very sad. It feels longer than that. We have in RT ever
since…

> Best Regards,
> Petr

Sebastian