Re: [PATCH v2 2/3] mm/page_alloc: move the kswapd node skip check into a helper
From: Joshua Hahn
Date: Wed Nov 18 2026 - 16:21:32 EST
On Wed, Nov 18, 2026 at 09:16:51PM +0530, Dev Jain wrote:
>
>
> On 17/11/26 4:53 am, Joshua Hahn wrote:
> > get_page_from_freelist() checks many conditions for each zone, often
> > with a big comment block explaining the conditions.
> >
> > Move the kswapd check and its comment into should_skip_kswapd_zone() to
> > make the zone iteration cleaner to read.
> >
> > While moving this, also resolve a checkpatch warning which complains of
> > a waitqueue_active() call with no corresponding comment explaining why
> > the lockless use is OK in this case.
> >
> > No functional change intended.
> >
> > Reviewed-by: SJ Park <sj@xxxxxxxxxx>
> > Reviewed-by: Chris Down <chris@xxxxxxxxxxxxxx>
> > Reviewed-by: Zi Yan <ziy@xxxxxxxxxx>
> > Tested-by: Chris Down <chris@xxxxxxxxxxxxxx>
> > Signed-off-by: Joshua Hahn <joshua.hahnjy@xxxxxxxxx>
> > ---
>
> Thanks looks better.
>
> Reviewed-by: Dev Jain <dev.jain@xxxxxxx>
>
> But see below.
>
>
> > mm/page_alloc.c | 47 +++++++++++++++++++++++++++++------------------
> > 1 file changed, 29 insertions(+), 18 deletions(-)
> >
> > diff --git a/mm/page_alloc.c b/mm/page_alloc.c
> > index 4d829e5eac49d..43929faffb989 100644
> > --- a/mm/page_alloc.c
> > +++ b/mm/page_alloc.c
> > @@ -3785,6 +3785,32 @@ static inline unsigned int alloc_flags_cma(gfp_t gfp_mask)
> > return ALLOC_DEFAULT;
> > }
> >
> > +/*
> > + * If kswapd is already active on a node and the zone is below its kswapd
> > + * watermark for this request, keep looking for other nodes that might be idle.
> > + * This can happen if another process has NUMA bindings and is causing kswapd
> > + * wakeups on only some nodes. Avoid accidental "node_reclaim_mode"-like
> > + * behavior in this case.
> > + */
> > +static bool should_skip_kswapd_zone(struct zone *zone, unsigned int order,
> > + const struct alloc_context *ac)
>
> Make this inline? This is the very hot page allocator code that's why :)
> I may be overthinking but in page_alloc.c I do see functions with similar
> size being marked inline.
Very reasonable! I think it's just a few checks and bitwise comparisons
at most (apart from the zone watermark checks). I'll make that change.
Thanks Dev!
Joshua
> > +{
> > + unsigned long mark;
> > +
> > + /*
> > + * This means kswapd is asleep, so don't skip the zone. This is a
> > + * lockless check, since a stale result only affects zone selection.
> > + */
> > + if (waitqueue_active(&zone->zone_pgdat->kswapd_wait))
> > + return false;
> > +
> > + if (sysctl_numa_balancing_mode & NUMA_BALANCING_MEMORY_TIERING)
> > + mark = promo_wmark_pages(zone);
> > + else
> > + mark = high_wmark_pages(zone);
> > + return !zone_watermark_fast(zone, order, mark, ac->highest_zoneidx, 0);
> > +}
> > +
> > /*
> > * get_page_from_freelist goes through the zonelist trying to allocate
> > * a page.
> > @@ -3796,9 +3822,6 @@ get_page_from_freelist(gfp_t gfp_mask, unsigned int order, int alloc_flags,
> > struct zoneref *z;
> > struct zone *zone;
> > struct pglist_data *last_pgdat = NULL;
> > - enum zone_watermarks kswapd_watermark =
> > - sysctl_numa_balancing_mode & NUMA_BALANCING_MEMORY_TIERING ?
> > - WMARK_PROMO : WMARK_HIGH;
> > bool last_pgdat_dirty_ok = false;
> > bool no_fallback;
> > bool skip_kswapd_nodes = nr_online_nodes > 1;
> > @@ -3865,22 +3888,10 @@ get_page_from_freelist(gfp_t gfp_mask, unsigned int order, int alloc_flags,
> > }
> > }
> >
> > - /*
> > - * If kswapd is already active on a node and the zone is below
> > - * its kswapd watermark for this request, keep looking for
> > - * other nodes that might be idle. This can happen if another
> > - * process has NUMA bindings and is causing kswapd wakeups on
> > - * only some nodes. Avoid accidental "node_reclaim_mode"-like
> > - * behavior in this case.
> > - */
> > if (skip_kswapd_nodes &&
> > - !waitqueue_active(&zone->zone_pgdat->kswapd_wait)) {
> > - mark = wmark_pages(zone, kswapd_watermark);
> > - if (!zone_watermark_fast(zone, order, mark,
> > - ac->highest_zoneidx, 0)) {
> > - skipped_kswapd_nodes = true;
> > - continue;
> > - }
> > + should_skip_kswapd_zone(zone, order, ac)) {
> > + skipped_kswapd_nodes = true;
> > + continue;
> > }
> >
> > cond_accept_memory(zone, order, alloc_flags);
>