The quilt patch titled
Subject: mm/khugepaged: record SCAN_PMD_MAPPED when scan_pmd() finds hugepage
has been removed from the -mm tree. Its filename was
mm-khugepaged-record-scan_pmd_mapped-when-scan_pmd-finds-hugepage.patch
This patch was dropped because it was merged into the mm-stable branch
of git://git.kernel.org/pub/scm/linux/kernel/git/akpm/mm
------------------------------------------------------
From: "Zach O'Keefe" <zokeefe@xxxxxxxxxx>
Subject: mm/khugepaged: record SCAN_PMD_MAPPED when scan_pmd() finds hugepage
Date: Wed, 6 Jul 2022 16:59:26 -0700
When scanning an anon pmd to see if it's eligible for collapse, return
SCAN_PMD_MAPPED if the pmd already maps a hugepage. Note that
SCAN_PMD_MAPPED is different from SCAN_PAGE_COMPOUND used in the
file-collapse path, since the latter might identify pte-mapped compound
pages. This is required by MADV_COLLAPSE which necessarily needs to know
what hugepage-aligned/sized regions are already pmd-mapped.
In order to determine if a pmd already maps a hugepage, refactor
mm_find_pmd():
Return mm_find_pmd() to it's pre-commit f72e7dcdd252 ("mm: let mm_find_pmd
fix buggy race with THP fault") behavior. ksm was the only caller that
explicitly wanted a pte-mapping pmd, so open code the pte-mapping logic
there (pmd_present() and pmd_trans_huge() checks).
Undo revert change in commit f72e7dcdd252 ("mm: let mm_find_pmd fix buggy
race with THP fault") that open-coded split_huge_pmd_address() pmd lookup
and use mm_find_pmd() instead.
Link: https://lkml.kernel.org/r/20220706235936.2197195-9-zokeefe@xxxxxxxxxx
Signed-off-by: Zach O'Keefe <zokeefe@xxxxxxxxxx>
Reviewed-by: Yang Shi <shy828301@xxxxxxxxx>
Cc: Alex Shi <alex.shi@xxxxxxxxxxxxxxxxx>
Cc: Andrea Arcangeli <aarcange@xxxxxxxxxx>
Cc: Arnd Bergmann <arnd@xxxxxxxx>
Cc: Axel Rasmussen <axelrasmussen@xxxxxxxxxx>
Cc: Chris Kennelly <ckennelly@xxxxxxxxxx>
Cc: Chris Zankel <chris@xxxxxxxxxx>
Cc: David Hildenbrand <david@xxxxxxxxxx>
Cc: David Rientjes <rientjes@xxxxxxxxxx>
Cc: Helge Deller <deller@xxxxxx>
Cc: Hugh Dickins <hughd@xxxxxxxxxx>
Cc: Ivan Kokshaysky <ink@xxxxxxxxxxxxxxxxxxxx>
Cc: James Bottomley <James.Bottomley@xxxxxxxxxxxxxxxxxxxxx>
Cc: Jens Axboe <axboe@xxxxxxxxx>
Cc: "Kirill A. Shutemov" <kirill.shutemov@xxxxxxxxxxxxxxx>
Cc: Matthew Wilcox <willy@xxxxxxxxxxxxx>
Cc: Matt Turner <mattst88@xxxxxxxxx>
Cc: Max Filippov <jcmvbkbc@xxxxxxxxx>
Cc: Miaohe Lin <linmiaohe@xxxxxxxxxx>
Cc: Michal Hocko <mhocko@xxxxxxxx>
Cc: Minchan Kim <minchan@xxxxxxxxxx>
Cc: Pasha Tatashin <pasha.tatashin@xxxxxxxxxx>
Cc: Pavel Begunkov <asml.silence@xxxxxxxxx>
Cc: Peter Xu <peterx@xxxxxxxxxx>
Cc: Rongwei Wang <rongwei.wang@xxxxxxxxxxxxxxxxx>
Cc: SeongJae Park <sj@xxxxxxxxxx>
Cc: Song Liu <songliubraving@xxxxxx>
Cc: Thomas Bogendoerfer <tsbogend@xxxxxxxxxxxxxxxx>
Cc: Vlastimil Babka <vbabka@xxxxxxx>
Cc: Zi Yan <ziy@xxxxxxxxxx>
Cc: Dan Carpenter <dan.carpenter@xxxxxxxxxx>
Cc: "Souptick Joarder (HPE)" <jrdr.linux@xxxxxxxxx>
Signed-off-by: Andrew Morton <akpm@xxxxxxxxxxxxxxxxxxxx>
---
include/trace/events/huge_memory.h | 1
mm/huge_memory.c | 18 --------
mm/internal.h | 2
mm/khugepaged.c | 60 +++++++++++++++++++++------
mm/ksm.c | 10 ++++
mm/rmap.c | 15 ++----
6 files changed, 67 insertions(+), 39 deletions(-)
--- a/include/trace/events/huge_memory.h~mm-khugepaged-record-scan_pmd_mapped-when-scan_pmd-finds-hugepage
+++ a/include/trace/events/huge_memory.h
@@ -11,6 +11,7 @@
EM( SCAN_FAIL, "failed") \
EM( SCAN_SUCCEED, "succeeded") \
EM( SCAN_PMD_NULL, "pmd_null") \
+ EM( SCAN_PMD_MAPPED, "page_pmd_mapped") \
EM( SCAN_EXCEED_NONE_PTE, "exceed_none_pte") \
EM( SCAN_EXCEED_SWAP_PTE, "exceed_swap_pte") \
EM( SCAN_EXCEED_SHARED_PTE, "exceed_shared_pte") \
--- a/mm/huge_memory.c~mm-khugepaged-record-scan_pmd_mapped-when-scan_pmd-finds-hugepage
+++ a/mm/huge_memory.c
@@ -2286,25 +2286,11 @@ out:
void split_huge_pmd_address(struct vm_area_struct *vma, unsigned long address,
bool freeze, struct folio *folio)
{
- pgd_t *pgd;
- p4d_t *p4d;
- pud_t *pud;
- pmd_t *pmd;
+ pmd_t *pmd = mm_find_pmd(vma->vm_mm, address);
- pgd = pgd_offset(vma->vm_mm, address);
- if (!pgd_present(*pgd))
+ if (!pmd)
return;
- p4d = p4d_offset(pgd, address);
- if (!p4d_present(*p4d))
- return;
-
- pud = pud_offset(p4d, address);
- if (!pud_present(*pud))
- return;
-
- pmd = pmd_offset(pud, address);
-
__split_huge_pmd(vma, pmd, address, freeze, folio);
}
--- a/mm/internal.h~mm-khugepaged-record-scan_pmd_mapped-when-scan_pmd-finds-hugepage
+++ a/mm/internal.h
@@ -187,7 +187,7 @@ extern void reclaim_throttle(pg_data_t *
/*
* in mm/rmap.c:
*/
-extern pmd_t *mm_find_pmd(struct mm_struct *mm, unsigned long address);
+pmd_t *mm_find_pmd(struct mm_struct *mm, unsigned long address);
/*
* in mm/page_alloc.c
--- a/mm/khugepaged.c~mm-khugepaged-record-scan_pmd_mapped-when-scan_pmd-finds-hugepage
+++ a/mm/khugepaged.c
@@ -28,6 +28,7 @@ enum scan_result {
SCAN_FAIL,
SCAN_SUCCEED,
SCAN_PMD_NULL,
+ SCAN_PMD_MAPPED,
SCAN_EXCEED_NONE_PTE,
SCAN_EXCEED_SWAP_PTE,
SCAN_EXCEED_SHARED_PTE,
@@ -877,6 +878,45 @@ static int hugepage_vma_revalidate(struc
return SCAN_SUCCEED;
}
+static int find_pmd_or_thp_or_none(struct mm_struct *mm,
+ unsigned long address,
+ pmd_t **pmd)
+{
+ pmd_t pmde;
+
+ *pmd = mm_find_pmd(mm, address);
+ if (!*pmd)
+ return SCAN_PMD_NULL;
+
+ pmde = pmd_read_atomic(*pmd);
+
+#ifdef CONFIG_TRANSPARENT_HUGEPAGE
+ /* See comments in pmd_none_or_trans_huge_or_clear_bad() */
+ barrier();
+#endif
+ if (!pmd_present(pmde))
+ return SCAN_PMD_NULL;
+ if (pmd_trans_huge(pmde))
+ return SCAN_PMD_MAPPED;
+ if (pmd_bad(pmde))
+ return SCAN_PMD_NULL;
+ return SCAN_SUCCEED;
+}
+
+static int check_pmd_still_valid(struct mm_struct *mm,
+ unsigned long address,
+ pmd_t *pmd)
+{
+ pmd_t *new_pmd;
+ int result = find_pmd_or_thp_or_none(mm, address, &new_pmd);
+
+ if (result != SCAN_SUCCEED)
+ return result;
+ if (new_pmd != pmd)
+ return SCAN_FAIL;
+ return SCAN_SUCCEED;
+}
+
/*
* Bring missing pages in from swap, to complete THP collapse.
* Only done if khugepaged_scan_pmd believes it is worthwhile.
@@ -988,9 +1028,8 @@ static int collapse_huge_page(struct mm_
goto out_nolock;
}
- pmd = mm_find_pmd(mm, address);
- if (!pmd) {
- result = SCAN_PMD_NULL;
+ result = find_pmd_or_thp_or_none(mm, address, &pmd);
+ if (result != SCAN_SUCCEED) {
mmap_read_unlock(mm);
goto out_nolock;
}
@@ -1018,7 +1057,8 @@ static int collapse_huge_page(struct mm_
if (result != SCAN_SUCCEED)
goto out_up_write;
/* check if the pmd is still valid */
- if (mm_find_pmd(mm, address) != pmd)
+ result = check_pmd_still_valid(mm, address, pmd);
+ if (result != SCAN_SUCCEED)
goto out_up_write;
anon_vma_lock_write(vma->anon_vma);
@@ -1121,11 +1161,9 @@ static int khugepaged_scan_pmd(struct mm
VM_BUG_ON(address & ~HPAGE_PMD_MASK);
- pmd = mm_find_pmd(mm, address);
- if (!pmd) {
- result = SCAN_PMD_NULL;
+ result = find_pmd_or_thp_or_none(mm, address, &pmd);
+ if (result != SCAN_SUCCEED)
goto out;
- }
memset(cc->node_load, 0, sizeof(cc->node_load));
pte = pte_offset_map_lock(mm, pmd, address, &ptl);
@@ -1383,8 +1421,7 @@ void collapse_pte_mapped_thp(struct mm_s
if (!PageHead(hpage))
goto drop_hpage;
- pmd = mm_find_pmd(mm, haddr);
- if (!pmd)
+ if (find_pmd_or_thp_or_none(mm, haddr, &pmd) != SCAN_SUCCEED)
goto drop_hpage;
start_pte = pte_offset_map_lock(mm, pmd, haddr, &ptl);
@@ -1502,8 +1539,7 @@ static void retract_page_tables(struct a
if (vma->vm_end < addr + HPAGE_PMD_SIZE)
continue;
mm = vma->vm_mm;
- pmd = mm_find_pmd(mm, addr);
- if (!pmd)
+ if (find_pmd_or_thp_or_none(mm, addr, &pmd) != SCAN_SUCCEED)
continue;
/*
* We need exclusive mmap_lock to retract page table.
--- a/mm/ksm.c~mm-khugepaged-record-scan_pmd_mapped-when-scan_pmd-finds-hugepage
+++ a/mm/ksm.c
@@ -1134,6 +1134,7 @@ static int replace_page(struct vm_area_s
{
struct mm_struct *mm = vma->vm_mm;
pmd_t *pmd;
+ pmd_t pmde;
pte_t *ptep;
pte_t newpte;
spinlock_t *ptl;
@@ -1148,6 +1149,15 @@ static int replace_page(struct vm_area_s
pmd = mm_find_pmd(mm, addr);
if (!pmd)
goto out;
+ /*
+ * Some THP functions use the sequence pmdp_huge_clear_flush(), set_pmd_at()
+ * without holding anon_vma lock for write. So when looking for a
+ * genuine pmde (in which to find pte), test present and !THP together.
+ */
+ pmde = *pmd;
+ barrier();
+ if (!pmd_present(pmde) || pmd_trans_huge(pmde))
+ goto out;
mmu_notifier_range_init(&range, MMU_NOTIFY_CLEAR, 0, vma, mm, addr,
addr + PAGE_SIZE);
--- a/mm/rmap.c~mm-khugepaged-record-scan_pmd_mapped-when-scan_pmd-finds-hugepage
+++ a/mm/rmap.c
@@ -767,13 +767,17 @@ unsigned long page_address_in_vma(struct
return vma_address(page, vma);
}
+/*
+ * Returns the actual pmd_t* where we expect 'address' to be mapped from, or
+ * NULL if it doesn't exist. No guarantees / checks on what the pmd_t*
+ * represents.
+ */
pmd_t *mm_find_pmd(struct mm_struct *mm, unsigned long address)
{
pgd_t *pgd;
p4d_t *p4d;
pud_t *pud;
pmd_t *pmd = NULL;
- pmd_t pmde;
pgd = pgd_offset(mm, address);
if (!pgd_present(*pgd))
@@ -788,15 +792,6 @@ pmd_t *mm_find_pmd(struct mm_struct *mm,
goto out;
pmd = pmd_offset(pud, address);
- /*
- * Some THP functions use the sequence pmdp_huge_clear_flush(), set_pmd_at()
- * without holding anon_vma lock for write. So when looking for a
- * genuine pmde (in which to find pte), test present and !THP together.
- */
- pmde = *pmd;
- barrier();
- if (!pmd_present(pmde) || pmd_trans_huge(pmde))
- pmd = NULL;
out:
return pmd;
}
_
Patches currently in -mm which might be from zokeefe@xxxxxxxxxx are
mm-shmem-add-flag-to-enforce-shmem-thp-in-hugepage_vma_check.patch
mm-khugepaged-attempt-to-map-file-shmem-backed-pte-mapped-thps-by-pmds.patch
mm-madvise-add-file-and-shmem-support-to-madv_collapse.patch
mm-khugepaged-add-tracepoint-to-hpage_collapse_scan_file.patch
selftests-vm-dedup-thp-helpers.patch
selftests-vm-modularize-thp-collapse-memory-operations.patch
selftests-vm-add-thp-collapse-file-and-tmpfs-testing.patch
selftests-vm-add-thp-collapse-shmem-testing.patch
selftests-vm-add-file-shmem-madv_collapse-selftest-for-cleared-pmd.patch
selftests-vm-add-selftest-for-madv_collapse-of-uffd-minor-memory.patch
