mm/gup: use gup_can_follow_protnone() also in GUP-fast
authorDavid Hildenbrand <david@redhat.com>
Thu, 25 Aug 2022 16:46:58 +0000 (18:46 +0200)
committerAndrew Morton <akpm@linux-foundation.org>
Tue, 27 Sep 2022 02:46:28 +0000 (19:46 -0700)
There seems to be no reason why FOLL_FORCE during GUP-fast would have to
fallback to the slow path when stumbling over a PROT_NONE mapped page.  We
only have to trigger hinting faults in case FOLL_FORCE is not set, and any
kind of fault handling naturally happens from the slow path -- where NUMA
hinting accounting/handling would be performed.

Note that the comment regarding THP migration is outdated: commit
2b4847e73004 ("mm: numa: serialise parallel get_user_page against THP
migration") described that this was required for THP due to lack of PMD
migration entries.  Nowadays, we do have proper PMD migration entries in
place -- see set_pmd_migration_entry(), which does a proper
pmdp_invalidate() when placing the migration entry.

So let's just reuse gup_can_follow_protnone() here to make it consistent
and drop the somewhat outdated comments.

Link: https://lkml.kernel.org/r/20220825164659.89824-3-david@redhat.com
Signed-off-by: David Hildenbrand <david@redhat.com>
Cc: Andrea Arcangeli <aarcange@redhat.com>
Cc: Hugh Dickins <hughd@google.com>
Cc: Jason Gunthorpe <jgg@nvidia.com>
Cc: John Hubbard <jhubbard@nvidia.com>
Cc: Matthew Wilcox <willy@infradead.org>
Cc: Mel Gorman <mgorman@suse.de>
Cc: Peter Xu <peterx@redhat.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
mm/gup.c

index f06770e03549556da12c5d28ae964e21781d888a..ce00a4c40da8a7c75cd1cf48e3439a416433e2ca 100644 (file)
--- a/mm/gup.c
+++ b/mm/gup.c
@@ -2420,11 +2420,7 @@ static int gup_pte_range(pmd_t pmd, pmd_t *pmdp, unsigned long addr,
                struct page *page;
                struct folio *folio;
 
-               /*
-                * Similar to the PMD case below, NUMA hinting must take slow
-                * path using the pte_protnone check.
-                */
-               if (pte_protnone(pte))
+               if (pte_protnone(pte) && !gup_can_follow_protnone(flags))
                        goto pte_unmap;
 
                if (!pte_access_permitted(pte, flags & FOLL_WRITE))
@@ -2808,12 +2804,8 @@ static int gup_pmd_range(pud_t *pudp, pud_t pud, unsigned long addr, unsigned lo
 
                if (unlikely(pmd_trans_huge(pmd) || pmd_huge(pmd) ||
                             pmd_devmap(pmd))) {
-                       /*
-                        * NUMA hinting faults need to be handled in the GUP
-                        * slowpath for accounting purposes and so that they
-                        * can be serialised against THP migration.
-                        */
-                       if (pmd_protnone(pmd))
+                       if (pmd_protnone(pmd) &&
+                           !gup_can_follow_protnone(flags))
                                return 0;
 
                        if (!gup_huge_pmd(pmd, pmdp, addr, next, flags,