Received-SPF: pass (google.com: domain of linux-kernel-owner@vger.kernel.org designates 2620:137:e000::1:20 as permitted sender) client-ip=2620:137:e000::1:20;
MIME-Version: 1.0
References: <20220304015941.1704249-1-eric.dumazet@gmail.com>
 <CANn89iJgaSDucfvTusWDKVU4vraqku3pw1BBPjNVj5cZt2U7vg@mail.gmail.com> <deae2ce6-8a43-5720-319e-cab25ee3bdb6@suse.cz>
In-Reply-To: <deae2ce6-8a43-5720-319e-cab25ee3bdb6@suse.cz>
From:   Eric Dumazet <edumazet@google.com>
Date:   Fri, 4 Mar 2022 08:50:03 -0800
Message-ID: <CANn89iJOAnSz3h36A_gAoog1xRjkaR=3i1L6iz8C_E6DC8TwUw@mail.gmail.com>
Subject: Re: [PATCH] mm/page_alloc: call check_new_pages() while zone spinlock
 is not held
To:     Vlastimil Babka <vbabka@suse.cz>
Cc:     Eric Dumazet <eric.dumazet@gmail.com>,
        Andrew Morton <akpm@linux-foundation.org>,
        linux-kernel <linux-kernel@vger.kernel.org>,
        linux-mm <linux-mm@kvack.org>,
        Mel Gorman <mgorman@techsingularity.net>,
        Michal Hocko <mhocko@kernel.org>,
        Shakeel Butt <shakeelb@google.com>,
        Wei Xu <weixugc@google.com>, Greg Thelen <gthelen@google.com>,
        Hugh Dickins <hughd@google.com>,
        David Rientjes <rientjes@google.com>
Content-Type: text/plain; charset="UTF-8"
Precedence: bulk

On Fri, Mar 4, 2022 at 12:32 AM Vlastimil Babka <vbabka@suse.cz> wrote:
>
>
> Ah, you found out already. Well maybe it could simply be moved inside the
> loop under the locked section and always done when page != NULL? I mean if
> check_new_pages() fails we just leak the problematic pages anyway so they
> are no longer free to allocate anymore and we should not count them as such.
>

Yes, and I guess that calling __mod_zone_freepage_state() for the page
about to be leaked
is just fine.

This means we can call check_newpages() without holding zone spinlock
or hard irqs being masked.

I will test something like:

diff --git a/mm/page_alloc.c b/mm/page_alloc.c
index 3589febc6d31928f850ebe5a4015ddc40e0469f3..1804287c1b792b8aa0e964b17eb002b6b1115258
100644
--- a/mm/page_alloc.c
+++ b/mm/page_alloc.c
@@ -3706,10 +3706,10 @@ struct page *rmqueue(struct zone *preferred_zone,
         * allocate greater than order-1 page units with __GFP_NOFAIL.
         */
        WARN_ON_ONCE((gfp_flags & __GFP_NOFAIL) && (order > 1));
-       spin_lock_irqsave(&zone->lock, flags);

        do {
                page = NULL;
+               spin_lock_irqsave(&zone->lock, flags);
                /*
                 * order-0 request can reach here when the pcplist is skipped
                 * due to non-CMA allocation context. HIGHATOMIC area is
@@ -3721,15 +3721,15 @@ struct page *rmqueue(struct zone *preferred_zone,
                        if (page)
                                trace_mm_page_alloc_zone_locked(page,
order, migratetype);
                }
-               if (!page)
+               if (!page) {
                        page = __rmqueue(zone, order, migratetype, alloc_flags);
-       } while (page && check_new_pages(page, order));
-       if (!page)
-               goto failed;
-
-       __mod_zone_freepage_state(zone, -(1 << order),
-                                 get_pcppage_migratetype(page));
-       spin_unlock_irqrestore(&zone->lock, flags);
+                       if (!page)
+                               goto failed;
+               }
+               __mod_zone_freepage_state(zone, -(1 << order),
+                                         get_pcppage_migratetype(page));
+               spin_unlock_irqrestore(&zone->lock, flags);
+       } while (check_new_pages(page, order));

        __count_zid_vm_events(PGALLOC, page_zonenum(page), 1 << order);
        zone_statistics(preferred_zone, zone, 1);