Re: [PATCH] mm/page_alloc: Occasionally relinquish zone lock in batch freeing

[Joshua Hahn <joshua.hahnjy@gmail.com>]

On Mon, 18 Aug 2025 17:13:40 -0700 Andrew Morton <akpm@linux-foundation.org> wrote:

Hello Andrew,

Thank you for your time & feedback, as always!

> On Mon, 18 Aug 2025 11:58:03 -0700 Joshua Hahn <joshua.hahnjy@gmail.com> wrote:
> 
> > While testing workloads with high sustained memory pressure on large machines
> > (1TB memory, 316 CPUs), we saw an unexpectedly high number of softlockups.
> > Further investigation showed that the lock in free_pcppages_bulk was being held
> > for a long time, even being held while 2k+ pages were being freed.
> > 
> > Instead of holding the lock for the entirety of the freeing, check to see if
> > the zone lock is contended every pcp->batch pages. If there is contention,
> > relinquish the lock so that other processors have a change to grab the lock
> > and perform critical work.
> > 
> > In our fleet,
> 
> who is "our"?

Sorry for the ambiguity -- I work for Meta, so I was referring to their
machines. I'll make this clearer in the next version.

> > @@ -1267,12 +1270,22 @@ static void free_pcppages_bulk(struct zone *zone, int count,
> >  
> >  			/* must delete to avoid corrupting pcp list */
> >  			list_del(&page->pcp_list);
> > +			batch -= nr_pages;
> >  			count -= nr_pages;
> >  			pcp->count -= nr_pages;
> >  
> >  			__free_one_page(page, pfn, zone, order, mt, FPI_NONE);
> >  			trace_mm_page_pcpu_drain(page, order, mt);
> > -		} while (count > 0 && !list_empty(list));
> > +		} while (batch > 0 && !list_empty(list));
> > +
> > +		/*
> > +		 * Prevent starving the lock for other users; every pcp->batch
> > +		 * pages freed, relinquish the zone lock if it is contended.
> > +		 */
> > +		if (count && spin_is_contended(&zone->lock)) {
> > +			spin_unlock_irqrestore(&zone->lock, flags);
> > +			spin_lock_irqsave(&zone->lock, flags);
> > +		}
> >  	}
> 
> Pretty this isn't.
> 
> Sigh, we do so much stuff here and in __free_one_page().
> 
> What sort of guarantee do we have that the contending task will be able
> to get in and grab the spinlock in that tiny time window?

Thank you for pointing this out -- I don't think there is any guarantee.
Kiryl suggested that I put a cond_resched() here, in order to guarantee that
the contending tasks will be able to grab the spinlock. I think that's a great
idea -- I'll make this change in v2.

Thank you for your feedback, have a great day!
Joshua