In a debugging session the other day, Rik noticed that node_reclaim()
was missing memstall annotations.  This means we'll miss pressure and
lost productivity resulting from reclaim on an overloaded local NUMA
node when vm.zone_reclaim_mode is enabled.
There haven't been any reports, but that's likely because
vm.zone_reclaim_mode hasn't been a commonly used feature recently, and
the intersection between such setups and psi users is probably nil.
But secondary memory such as CXL-connected DIMMS, persistent memory etc,
and the page demotion patches that handle them
(https://lore.kernel.org/lkml/
20210401183216.
443C4443@viggo.jf.intel.com/)
could soon make this a more common codepath again.
Link: https://lkml.kernel.org/r/20210818152457.35846-1-hannes@cmpxchg.org
Signed-off-by: Johannes Weiner <hannes@cmpxchg.org>
Reported-by: Rik van Riel <riel@surriel.com>
Reviewed-by: Shakeel Butt <shakeelb@google.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
                .may_swap = 1,
                .reclaim_idx = gfp_zone(gfp_mask),
        };
+       unsigned long pflags;
 
        trace_mm_vmscan_node_reclaim_begin(pgdat->node_id, order,
                                           sc.gfp_mask);
 
        cond_resched();
+       psi_memstall_enter(&pflags);
        fs_reclaim_acquire(sc.gfp_mask);
        /*
         * We need to be able to allocate from the reserves for RECLAIM_UNMAP
        current->flags &= ~PF_SWAPWRITE;
        memalloc_noreclaim_restore(noreclaim_flag);
        fs_reclaim_release(sc.gfp_mask);
+       psi_memstall_leave(&pflags);
 
        trace_mm_vmscan_node_reclaim_end(sc.nr_reclaimed);