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

[Michal Hocko <mhocko@suse.com>]

On Fri 23-06-23 11:58:04, Sebastian Andrzej Siewior wrote:
> On 2023-06-23 11:21:01 [+0200], Michal Hocko wrote:
> > On Fri 23-06-23 10:12:50, Sebastian Andrzej Siewior wrote:
> > [...]
> > > It would do _but_ is it really needed? All users but one (the one in
> > > __build_all_zonelists()) have interrupts already disabled. This isn't a
> > > hot path and is not used very common. Does it justify the ifdef?
> > 
> > This is not my call of course but let me just stress out that if all
> > that is just because of __build_all_zonelists then it is likely not
> > really worth it. We are talking about memory hoptlug here. I seriously
> > doubt that this is something anybody would be using with RT
> > expectations.
> 
> My current plan to repost this with a better explanation and with printk
> deferred within the locked region. But first I'm going to use
> hotplug-mem to see how it works. If it can be enabled on RT then it
> should work even if nobody is using it.

Agreed! I merely wanted to point out that this needs a based fix much
more than trying to overengineer it.

-- 
Michal Hocko
SUSE Labs