Re: [PATCH 3/4] mm, page_allocator: Only use per-cpu allocator for irq-safe requests

[Hillf Danton]

> Sent: Friday, January 06, 2017 6:16 PM Mel Gorman wrote: 
> 
> On Fri, Jan 06, 2017 at 11:26:46AM +0800, Hillf Danton wrote:
> >
> > On Wednesday, January 04, 2017 7:11 PM Mel Gorman wrote:
> > > @@ -2647,9 +2644,8 @@ static struct page *rmqueue_pcplist(struct zone *preferred_zone,
> > >  	struct list_head *list;
> > >  	bool cold = ((gfp_flags & __GFP_COLD) != 0);
> > >  	struct page *page;
> > > -	unsigned long flags;
> > >
> > > -	local_irq_save(flags);
> > > +	preempt_disable();
> > >  	pcp = &this_cpu_ptr(zone->pageset)->pcp;
> > >  	list = &pcp->lists[migratetype];
> > >  	page = __rmqueue_pcplist(zone,  order, gfp_flags, migratetype,
> > > @@ -2658,7 +2654,7 @@ static struct page *rmqueue_pcplist(struct zone *preferred_zone,
> > >  		__count_zid_vm_events(PGALLOC, page_zonenum(page), 1 << order);
> > >  		zone_statistics(preferred_zone, zone, gfp_flags);
> > >  	}
> > > -	local_irq_restore(flags);
> > > +	preempt_enable();
> > >  	return page;
> > >  }
> > >
> > With PREEMPT configured, preempt_enable() adds entry point to schedule().
> > Is that needed when we try to allocate a page?
> >
> 
> Not necessarily but what are you proposing as an alternative? 

preempt_enable_no_resched() looks at first glance a choice for us to 
avoid flipping interrupts.

> get_cpu()
> is not an alternative and the point is to avoid disabling interrupts
> which is a much more expensive operation.
> 
Agree with every word.

Hillf