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

[Hillf Danton <hdanton@sina.com>]

On Wed, 20 Aug 2025 08:13:07 -0700 Joshua Hahn wrote:
> On Wed, 20 Aug 2025 09:29:00 +0800 Hillf Danton <hdanton@sina.com> wrote:
> > On Mon, 18 Aug 2025 11:58:03 -0700 Joshua Hahn 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.
> > > 
> > Instead of the unlock/lock game, simply return with the rest left to workqueue
> > in case of lock contension. But workqueue is still unable to kill soft lockup
> > if the number of contending CPUs is large enough.
> 
> Thank you for the idea. One concern that I have is that sometimes, we do expect
> free_pcppages_bulk to actually free all of the pages that it has promised to
> do. One example is when it is called from drain_zone_pages. Of course, we can
> have a while loop that would call free_pcppages_bulk until it returns 0, but
> I think that would be reduced to unlocking / locking over and over again.
> 
In the case of drain_zone_pages(), I think adding something like the pcpu_drain_mutex
to the path updating zone counters is a cure.