]> git.ipfire.org Git - thirdparty/kernel/linux.git/commitdiff
mm/huge_memory: separate out the folio part of zap_huge_pmd()
authorLorenzo Stoakes (Oracle) <ljs@kernel.org>
Fri, 20 Mar 2026 18:07:27 +0000 (18:07 +0000)
committerAndrew Morton <akpm@linux-foundation.org>
Sun, 5 Apr 2026 20:53:46 +0000 (13:53 -0700)
Place the part of the logic that manipulates counters and possibly updates
the accessed bit of the folio into its own function to make zap_huge_pmd()
more readable.

Also rename flush_needed to is_present as we only require a flush for
present entries.

Additionally add comments as to why we're doing what we're doing with
respect to softleaf entries.

This also lays the ground for further refactoring.

Link: https://lkml.kernel.org/r/6c4db67952f5529da4db102a6149b9050b5dda4e.1774029655.git.ljs@kernel.org
Signed-off-by: Lorenzo Stoakes (Oracle) <ljs@kernel.org>
Reviewed-by: Baolin Wang <baolin.wang@linux.alibaba.com>
Reviewed-by: Suren Baghdasaryan <surenb@google.com>
Cc: Barry Song <baohua@kernel.org>
Cc: David Hildenbrand <david@kernel.org>
Cc: Dev Jain <dev.jain@arm.com>
Cc: Lance Yang <lance.yang@linux.dev>
Cc: Liam Howlett <liam.howlett@oracle.com>
Cc: Michal Hocko <mhocko@suse.com>
Cc: Mike Rapoport <rppt@kernel.org>
Cc: Nico Pache <npache@redhat.com>
Cc: Qi Zheng <zhengqi.arch@bytedance.com>
Cc: Ryan Roberts <ryan.roberts@arm.com>
Cc: Suren Baghdasaryan <surenb@google.com>
Cc: Zi Yan <ziy@nvidia.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
mm/huge_memory.c

index c515e293ed48283f51037baf878bc6c91c8b56ba..d1e66df05c8665319dbd35e52d45a26db49ebc6c 100644 (file)
@@ -2402,6 +2402,37 @@ static inline void zap_deposited_table(struct mm_struct *mm, pmd_t *pmd)
        mm_dec_nr_ptes(mm);
 }
 
+static void zap_huge_pmd_folio(struct mm_struct *mm, struct vm_area_struct *vma,
+               pmd_t pmdval, struct folio *folio, bool is_present,
+               bool *has_deposit)
+{
+       const bool is_device_private = folio_is_device_private(folio);
+
+       /* Present and device private folios are rmappable. */
+       if (is_present || is_device_private)
+               folio_remove_rmap_pmd(folio, &folio->page, vma);
+
+       if (folio_test_anon(folio)) {
+               *has_deposit = true;
+               add_mm_counter(mm, MM_ANONPAGES, -HPAGE_PMD_NR);
+       } else {
+               add_mm_counter(mm, mm_counter_file(folio),
+                              -HPAGE_PMD_NR);
+
+               /*
+                * Use flush_needed to indicate whether the PMD entry
+                * is present, instead of checking pmd_present() again.
+                */
+               if (is_present && pmd_young(pmdval) &&
+                   likely(vma_has_recency(vma)))
+                       folio_mark_accessed(folio);
+       }
+
+       /* Device private folios are pinned. */
+       if (is_device_private)
+               folio_put(folio);
+}
+
 /**
  * zap_huge_pmd - Zap a huge THP which is of PMD size.
  * @tlb: The MMU gather TLB state associated with the operation.
@@ -2417,7 +2448,7 @@ bool zap_huge_pmd(struct mmu_gather *tlb, struct vm_area_struct *vma,
        bool has_deposit = arch_needs_pgtable_deposit();
        struct mm_struct *mm = tlb->mm;
        struct folio *folio = NULL;
-       bool flush_needed = false;
+       bool is_present = false;
        spinlock_t *ptl;
        pmd_t orig_pmd;
 
@@ -2446,14 +2477,11 @@ bool zap_huge_pmd(struct mmu_gather *tlb, struct vm_area_struct *vma,
 
        if (pmd_present(orig_pmd)) {
                folio = pmd_folio(orig_pmd);
-
-               flush_needed = true;
-               folio_remove_rmap_pmd(folio, &folio->page, vma);
+               is_present = true;
        } else if (pmd_is_valid_softleaf(orig_pmd)) {
                const softleaf_t entry = softleaf_from_pmd(orig_pmd);
 
                folio = softleaf_to_folio(entry);
-
                if (!thp_migration_supported())
                        WARN_ONCE(1, "Non present huge pmd without pmd migration enabled!");
        } else {
@@ -2461,33 +2489,14 @@ bool zap_huge_pmd(struct mmu_gather *tlb, struct vm_area_struct *vma,
                goto out;
        }
 
-       if (folio_test_anon(folio)) {
-               has_deposit = true;
-               add_mm_counter(mm, MM_ANONPAGES, -HPAGE_PMD_NR);
-       } else {
-               add_mm_counter(mm, mm_counter_file(folio),
-                              -HPAGE_PMD_NR);
-
-               /*
-                * Use flush_needed to indicate whether the PMD entry
-                * is present, instead of checking pmd_present() again.
-                */
-               if (flush_needed && pmd_young(orig_pmd) &&
-                   likely(vma_has_recency(vma)))
-                       folio_mark_accessed(folio);
-       }
-
-       if (folio_is_device_private(folio)) {
-               folio_remove_rmap_pmd(folio, &folio->page, vma);
-               folio_put(folio);
-       }
+       zap_huge_pmd_folio(mm, vma, orig_pmd, folio, is_present, &has_deposit);
 
 out:
        if (has_deposit)
                zap_deposited_table(mm, pmd);
 
        spin_unlock(ptl);
-       if (flush_needed)
+       if (is_present)
                tlb_remove_page_size(tlb, &folio->page, HPAGE_PMD_SIZE);
        return true;
 }