allocinfo_start() only reinitializes the codetag iterator at position 0.
For subsequent reads (position > 0), it reuses cached iterator state from
the previous batch. allocinfo_stop() drops mod_lock between read batches,
which allows module unload to complete and free the module memory that the
cached iterator still references:
CPU0 (read) CPU1 (rmmod)
---- ----
allocinfo_start(pos=0)
down_read(mod_lock)
allocinfo_show()
...
allocinfo_stop()
up_read(mod_lock)
codetag_unload_module()
kfree(cmod)
release_module_tags()
...
free_mod_mem()
allocinfo_start(pos=N)
down_read(mod_lock)
// reuses cached iter, skips re-init
allocinfo_show()
ct->filename <-- UAF
After free_mod_mem() frees the module's .rodata, allocinfo_show()
dereferences ct->filename, ct->function which point there.
Save the iterator state in allocinfo_next() and resume from it in
allocinfo_start() with codetag_next_ct(), which detects module removal via
idr_find() returning NULL and skips to the next module.
Link: https://lore.kernel.org/20260604065938.105991-1-hao.ge@linux.dev
Fixes: 9f44df50fe ("alloc_tag: keep codetag iterator active between read()")
Signed-off-by: Hao Ge <hao.ge@linux.dev>
Suggested-by: Suren Baghdasaryan <surenb@google.com>
Acked-by: Suren Baghdasaryan <surenb@google.com>
Cc: Kent Overstreet <kent.overstreet@linux.dev>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Pages allocated before page_ext is available have their codetag left
uninitialized. Track these early PFNs and clear their codetag in
clear_early_alloc_pfn_tag_refs() to avoid "alloc_tag was not set" warnings
when they are freed later.
Currently a fixed-size array of 8192 entries is used, with a warning if
the limit is exceeded. However, the number of early allocations depends
on the number of CPUs and can be larger than 8192.
Replace the fixed-size array with a dynamically allocated linked list of
pfn_pool structs. Each node is allocated via alloc_page() and mapped to a
pfn_pool containing a next pointer, an atomic slot counter, and a PFN
array that fills the remainder of the page.
The tracking pages themselves are allocated via alloc_page(), which would
trigger __pgalloc_tag_add() -> alloc_tag_add_early_pfn() and recurse
indefinitely. Introduce __GFP_NO_CODETAG (reuses the %__GFP_NO_OBJ_EXT
bit) and pass gfp_flags through pgalloc_tag_add() so that the early path
can skip recording allocations that carry this flag.
Link: https://lore.kernel.org/20260604024008.46592-1-hao.ge@linux.dev
Signed-off-by: Hao Ge <hao.ge@linux.dev>
Suggested-by: Suren Baghdasaryan <surenb@google.com>
Acked-by: Suren Baghdasaryan <surenb@google.com>
Cc: Kent Overstreet <kent.overstreet@linux.dev>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
compact_gap() returns 2 << order, which is used as watermark headroom in
__compaction_suitable() and as a threshold in kswapd reclaim decisions.
The computed value scales exponentially by order. For order-9 THP
allocations this evaluates to 1024 pages, but the compaction free
scanner's working set is bounded by COMPACT_CLUSTER_MAX (32 pages). The
scanner stops isolating free pages once it matches the migration batch.
The current gap over-reserves by 32x.
On fragmented production hosts, kswapd will try to reclaim up to the gap,
but it only reaches that threshold in 18% of attempts. As a result,
reclaim continues in the majority of cases despite many lower-order free
pages being available. The over-sized gap also causes 46% of order-9
compaction suitability checks to fail unnecessarily: the zone has
sufficient free pages for the scanner to operate, but not enough to clear
the inflated threshold.
Cap compact_gap() at COMPACT_CLUSTER_MAX so the watermark headroom
reflects the scanner's actual capacity. This function is used by two key
heuristics. The first is when kswapd can stop high-order reclaim and
downgrade to order-0 balancing, allowing kcompactd to be woken for the
original higher allocation order. The second is zone suitability
checking, where the smaller gap allows compaction to start sooner.
Note that orders 0-4 are unaffected since their gap is already less than
or equal to COMPACT_CLUSTER_MAX.
A/B test on v6.13-based instagram production hosts (64GB, 60s
measurement):
Unpatched (43 hosts)
pgscan_kswapd (mean/host): ~1.6M
reclaim efficiency (steal/scan): 83.8%
per-compaction success (success/stall): 2.1%
THP success (alloc/alloc+fallback): 4.9%
forced lru_add_drain (mean/host): ~107K
Patched (59 hosts)
pgscan_kswapd (mean/host): ~449K
reclaim efficiency (steal/scan): 91.0%
per-compaction success (success/stall): 28.3%
THP success (alloc/alloc+fallback): 17.2%
forced lru_add_drain (mean/host): ~64K
Additional tests were also performed using a workload of similar shape and
based on mm-new at the time of testing. Across three 60s runs, the patch
showed improvements consistent with the previous test: reduced kswapd
reclaim and fewer THP fault fallbacks.
Unpatched
kswapd_shrink_node downgrade to order-0 (mean): 0
thp_fault_fallback (mean): 1217
pgscan_kswapd (mean): 6328
pgsteal_kswapd (mean): 5657
Patched
kswapd_shrink_node downgrade to order-0 (mean): 28
thp_fault_fallback (mean): 738
pgscan_kswapd (mean): 3773
pgsteal_kswapd (mean): 3243
Link: https://lore.kernel.org/20260604061725.13800-1-jp.kobryn@linux.dev
Signed-off-by: JP Kobryn (Meta) <jp.kobryn@linux.dev>
Reviewed-by: Vlastimil Babka (SUSE) <vbabka@kernel.org>
Cc: Brendan Jackman <jackmanb@google.com>
Cc: Johannes Weiner <hannes@cmpxchg.org>
Cc: Michal Hocko <mhocko@suse.com>
Cc: Suren Baghdasaryan <surenb@google.com>
Cc: Zi Yan <ziy@nvidia.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
In the previous commit, uswsusp was modified to pin the swap device when
the swap type is determined, ensuring the device remains valid throughout
the hibernation I/O path.
Therefore, it is no longer necessary to repeatedly get and put the swap
device reference for each swap slot allocation and free operation.
For hibernation via the sysfs interface, user-space tasks are frozen
before swap allocation begins, so swapoff cannot race with allocation.
After resume, tasks remain frozen while swap slots are freed, so
additional reference management is not required there either.
Remove the redundant swap device get/put operations from the hibernation
swap allocation and free paths.
Also remove the SWP_WRITEOK check before allocation, as the cluster
allocation logic already validates the swap device state.
Update function comments to document the caller's responsibility for
ensuring swap device stability.
Link: https://lore.kernel.org/20260323160822.1409904-3-youngjun.park@lge.com
Signed-off-by: Youngjun Park <youngjun.park@lge.com>
Reviewed-by: Kairui Song <kasong@tencent.com>
Cc: Baoquan He <bhe@redhat.com>
Cc: Barry Song <baohua@kernel.org>
Cc: Chris Li <chrisl@kernel.org>
Cc: Kemeng Shi <shikemeng@huaweicloud.com>
Cc: Nhat Pham <nphamcs@gmail.com>
Cc: "Rafael J . Wysocki" <rafael@kernel.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Patch series "mm/swap, PM: hibernate: fix swapoff race in uswsusp by
pinning swap device", v8.
Currently, in the uswsusp path, only the swap type value is retrieved at
lookup time without holding a reference. If swapoff races after the type
is acquired, subsequent slot allocations operate on a stale swap device.
Additionally, grabbing and releasing the swap device reference on every
slot allocation is inefficient across the entire hibernation swap path.
This patch series addresses these issues:
- Patch 1: Fixes the swapoff race in uswsusp by pinning the swap device
from the point it is looked up until the session completes.
- Patch 2: Removes the overhead of per-slot reference counting in alloc/free
paths and cleans up the redundant SWP_WRITEOK check.
This patch (of 2):
Hibernation via uswsusp (/dev/snapshot ioctls) has a race window: after
selecting the resume swap area but before user space is frozen, swapoff
may run and invalidate the selected swap device.
Fix this by pinning the swap device with SWP_HIBERNATION while it is in
use. The pin is exclusive, which is sufficient since hibernate_acquire()
already prevents concurrent hibernation sessions.
The kernel swsusp path (sysfs-based hibernate/resume) uses
find_hibernation_swap_type() which is not affected by the pin. It freezes
user space before touching swap, so swapoff cannot race.
Introduce dedicated helpers:
- pin_hibernation_swap_type(): Look up and pin the swap device.
Used by the uswsusp path.
- find_hibernation_swap_type(): Lookup without pinning.
Used by the kernel swsusp path.
- unpin_hibernation_swap_type(): Clear the hibernation pin.
While a swap device is pinned, swapoff is prevented from proceeding.
Link: https://lore.kernel.org/20260323160822.1409904-1-youngjun.park@lge.com
Link: https://lore.kernel.org/20260323160822.1409904-2-youngjun.park@lge.com
Signed-off-by: Youngjun Park <youngjun.park@lge.com>
Reviewed-by: Kairui Song <kasong@tencent.com>
Cc: Baoquan He <bhe@redhat.com>
Cc: Barry Song <baohua@kernel.org>
Cc: Chris Li <chrisl@kernel.org>
Cc: Kemeng Shi <shikemeng@huaweicloud.com>
Cc: Nhat Pham <nphamcs@gmail.com>
Cc: "Rafael J . Wysocki" <rafael@kernel.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Patch series "mm/sparse-vmemmap: Provide generic vmemmap_set_pmd() and
vmemmap_check_pmd()", v3.
The weak vmemmap_set_pmd() and vmemmap_check_pmd() hooks are currently
no-ops in the generic code, which leaves architectures that need PMD-level
handling to open-code the same logic locally.
This series provides generic implementations for both helpers in
mm/sparse-vmemmap.c. vmemmap_set_pmd() installs a huge PMD with
PAGE_KERNEL protection, and vmemmap_check_pmd() verifies a present leaf
PMD before reusing the existing vmemmap_verify() helper.
With those generic helpers in place, patches 2-5 remove the now redundant
arch-specific implementations from arm64, riscv, loongarch, and sparc.
This patch (of 5):
The two weak functions are currently no-ops on every architecture, forcing
each platform that needs them to duplicate the same handful of lines.
Provide a generic implementation:
- vmemmap_set_pmd() simply sets a huge PMD with PAGE_KERNEL protection.
- vmemmap_check_pmd() verifies that the PMD is present and leaf,
then calls the existing vmemmap_verify() helper.
Architectures that need special handling can continue to override the weak
symbols; everyone else gets the standard version for free.
Link: https://lore.kernel.org/20260601084845.3792171-1-songmuchun@bytedance.com
Link: https://lore.kernel.org/20260601084845.3792171-2-songmuchun@bytedance.com
Signed-off-by: Muchun Song <songmuchun@bytedance.com>
Acked-by: David Hildenbrand (Arm) <david@kernel.org>
Acked-by: Oscar Salvador (SUSE) <osalvador@kernel.org>
Cc: Albert Ou <aou@eecs.berkeley.edu>
Cc: Alexandre Ghiti <alex@ghiti.fr>
Cc: Andreas Larsson <andreas@gaisler.com>
Cc: Catalin Marinas <catalin.marinas@arm.com>
Cc: David S. Miller <davem@davemloft.net>
Cc: Huacai Chen <chenhuacai@kernel.org>
Cc: Liam R. Howlett <liam@infradead.org>
Cc: Lorenzo Stoakes <ljs@kernel.org>
Cc: Michal Hocko <mhocko@suse.com>
Cc: Mike Rapoport <rppt@kernel.org>
Cc: Palmer Dabbelt <palmer@dabbelt.com>
Cc: Suren Baghdasaryan <surenb@google.com>
Cc: Vlastimil Babka <vbabka@kernel.org>
Cc: WANG Xuerui <kernel@xen0n.name>
Cc: Will Deacon <will@kernel.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
The VMA flags bitmap is a single word today: NUM_VMA_FLAG_BITS is
BITS_PER_LONG, so on 32-bit vma_flags_t holds only 32 bits. (The bitmap
type exists so this can grow past BITS_PER_LONG later; until it does,
anything declared above the first word is out of range on 32-bit.) The bit
enum nevertheless declares some bits unconditionally above BITS_PER_LONG
-- VMA_UFFD_MINOR_BIT is 41, with VM_UFFD_MINOR == VM_NONE on 32-bit so no
VMA actually carries the bit.
__VMA_UFFD_FLAGS feeds VMA_UFFD_MINOR_BIT to mk_vma_flags()
unconditionally. On 32-bit that becomes __set_bit(41, &one_long), a write
one word past the end of the single-word bitmap. The compiler folds the
out-of-bounds store with wraparound (1UL << (41 % 32) == bit 9) into the
first word; bit 9 is already in __VMA_UFFD_FLAGS so the mask happens to
come out right today, but it is an out-of-bounds write all the same, and
any high-numbered bit whose mod-BITS_PER_LONG position is otherwise unused
would silently OR an extra bit into the mask.
Rather than feed bit numbers that may not exist on the current build to
mk_vma_flags(), build the mask from whole per-mode masks that collapse to
EMPTY_VMA_FLAGS when their feature is unavailable. Add
mk_vma_flags_from_masks() for that, and define VMA_UFFD_MISSING / _WP /
_MINOR alongside the VM_UFFD_* flags, gating VMA_UFFD_MINOR on the same
config as VM_UFFD_MINOR (which implies 64BIT, where bit 41 fits). An
out-of-range bit is then never materialised, on any arch, and the in-range
fast path stays a compile-time constant.
Link: https://lore.kernel.org/20260529172331.356655-7-kas@kernel.org
Fixes: 9ea35a25d5 ("mm: introduce VMA flags bitmap type")
Signed-off-by: Kiryl Shutsemau <kas@kernel.org>
Reported-by: Sashiko AI review <sashiko-bot@kernel.org>
Suggested-by: Lorenzo Stoakes <ljs@kernel.org>
Reviewed-by: Lorenzo Stoakes <ljs@kernel.org>
Assisted-by: Claude:claude-opus-4-8
Cc: David Hildenbrand <david@kernel.org>
Cc: Michal Hocko <mhocko@suse.com>
Cc: Mike Rapoport <rppt@kernel.org>
Cc: Peter Xu <peterx@redhat.com>
Cc: Suren Baghdasaryan <surenb@google.com>
Cc: Vlastimil Babka <vbabka@kernel.org>
Cc: Balbir Singh <balbirs@nvidia.com>
Cc: <stable@vger.kernel.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
userfaultfd_must_wait() and userfaultfd_huge_must_wait() read the PTE
without taking the page table lock and then apply pte_write() /
huge_pte_write() to it. Those accessors decode bits from the present
encoding only; on a swap or migration entry they read the offset bits that
happen to share the same position and return an undefined result.
The intent of the check is "is this fault still WP-blocked?". A
non-marker swap entry means the page is in transit -- the userfault
context the original fault delivered against is no longer the same, and
the swap-in or migration completion path will re-deliver a fresh fault if
userspace still needs to handle it. Worst case under the current code the
garbage write bit says "wait", and the thread stays asleep until a
UFFDIO_WAKE that may never arrive.
Gate the writability check on pte_present() so the lockless re-check only
inspects present-PTE bits when the entry is actually present. The
non-present, non-marker case returns "don't wait" and lets the fault path
retry.
Link: https://lore.kernel.org/20260529172331.356655-6-kas@kernel.org
Fixes: 369cd2121b ("userfaultfd: hugetlbfs: userfaultfd_huge_must_wait for hugepmd ranges")
Fixes: 63b2d4174c ("userfaultfd: wp: add the writeprotect API to userfaultfd ioctl")
Signed-off-by: Kiryl Shutsemau <kas@kernel.org>
Reported-by: Sashiko AI review <sashiko-bot@kernel.org>
Reviewed-by: Lorenzo Stoakes <ljs@kernel.org>
Cc: David Hildenbrand <david@kernel.org>
Cc: Michal Hocko <mhocko@suse.com>
Cc: Mike Rapoport <rppt@kernel.org>
Cc: Peter Xu <peterx@redhat.com>
Cc: Suren Baghdasaryan <surenb@google.com>
Cc: Vlastimil Babka <vbabka@kernel.org>
Cc: Balbir Singh <balbirs@nvidia.com>
Cc: <stable@vger.kernel.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
A PAGEMAP_SCAN ioctl requesting PM_SCAN_WP_MATCHING on a hugetlb VMA hangs
the calling thread, unkillably, as soon as the scan reaches an unpopulated
part of the range:
do_pagemap_scan()
walk_page_range()
walk_hugetlb_range()
hugetlb_vma_lock_read() # take the vma lock for read ...
pagemap_scan_pte_hole() # ... ->pte_hole() for a hole
uffd_wp_range()
change_protection()
hugetlb_change_protection()
hugetlb_vma_lock_write() # ... and block taking it for write
walk_hugetlb_range() holds the hugetlb vma lock for read across the whole
walk. A present entry goes to ->hugetlb_entry(); an unpopulated one goes
to ->pte_hole(), i.e. pagemap_scan_pte_hole(). To write-protect the hole
that handler calls uffd_wp_range(), which on a hugetlb VMA reaches
hugetlb_change_protection() and takes the same vma lock for write. The
thread then blocks in down_write() waiting for the read lock it is itself
holding.
The populated path avoids this: pagemap_scan_hugetlb_entry()
write-protects the entry inline under the page-table lock and never enters
hugetlb_change_protection().
Do the same for holes. Fault in the page table and install the uffd-wp
marker directly with make_uffd_wp_huge_pte() under the page-table lock,
rather than routing through uffd_wp_range(). That is the same sequence
hugetlb_change_protection() runs for an unpopulated entry, minus the vma
write lock -- which is safe to skip because PMD sharing is disabled on
uffd-wp VMAs (hugetlb_unshare_all_pmds() runs at registration), leaving
nothing for that lock to serialise against.
Link: https://lore.kernel.org/20260529172331.356655-4-kas@kernel.org
Fixes: 52526ca7fd ("fs/proc/task_mmu: implement IOCTL to get and optionally clear info about PTEs")
Signed-off-by: Kiryl Shutsemau <kas@kernel.org>
Reported-by: Sashiko AI review <sashiko-bot@kernel.org>
Assisted-by: Claude:claude-opus-4-8
Cc: David Hildenbrand <david@kernel.org>
Cc: Lorenzo Stoakes <ljs@kernel.org>
Cc: Michal Hocko <mhocko@suse.com>
Cc: Mike Rapoport <rppt@kernel.org>
Cc: Peter Xu <peterx@redhat.com>
Cc: Suren Baghdasaryan <surenb@google.com>
Cc: Vlastimil Babka <vbabka@kernel.org>
Cc: Balbir Singh <balbirs@nvidia.com>
Cc: <stable@vger.kernel.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Patch series "userfaultfd/pagemap: pre-existing fixes".
These are pre-existing bug fixes that were carried at the front of the
userfaultfd RWP working-set-tracking series up to v5 [1]. Per review
feedback that fixes should not sit in the middle of a feature series, they
are split out and sent on their own; the RWP series is reposted rebased on
top of this.
All six were flagged by the Sashiko AI review of the RWP series and carry
Reported-by: Sashiko AI review <sashiko-bot@kernel.org>. They are
independent of RWP, apply to mm-new directly, and carry Cc: stable@.
1: fs/proc/task_mmu: a missing huge_ptep_modify_prot_start() in
make_uffd_wp_huge_pte() can lose hardware Dirty/Accessed updates
when PAGEMAP_SCAN write-protects a hugetlb PTE.
2: fs/proc/task_mmu: pagemap_scan_hugetlb_entry() compares the range
against HPAGE_SIZE rather than the hstate page size, so it never
write-protects gigantic hugetlb pages.
3: fs/proc/task_mmu: PAGEMAP_SCAN with PM_SCAN_WP_MATCHING over an
unpopulated hugetlb range self-deadlocks -- pagemap_scan_pte_hole()
calls uffd_wp_range() while walk_hugetlb_range() holds the hugetlb
vma lock for read, and hugetlb_change_protection() then takes it
for write. Install the marker inline instead.
4: mm/huge_memory: change_non_present_huge_pmd() drops pmd_swp_uffd_wp
on a device-private PMD permission downgrade, silently losing the
uffd-wp marker.
5: userfaultfd: must_wait() applies pte_write() to a locklessly read
PTE without checking pte_present(), so swap/migration entries
decode random offset bits and a thread can stay parked on a stale
fault.
6: userfaultfd: __VMA_UFFD_FLAGS feeds VMA_UFFD_MINOR_BIT (41) to
mk_vma_flags() unconditionally, an out-of-bounds write into the
single-word vma_flags_t on 32-bit. Build the mask from config-gated
per-mode masks so an unavailable bit is never materialised.
This patch (of 6):
make_uffd_wp_huge_pte() arms the UFFD_WP bit on a present HugeTLB PTE by
calling huge_ptep_modify_prot_commit() with a ptent snapshot that was
fetched without the corresponding huge_ptep_modify_prot_start(). The
start helper is what atomically clears the entry so the kernel-owned
snapshot stays consistent until the commit; without it, the hardware may
set Dirty or Accessed in the live PTE between the original read and the
commit, and huge_ptep_modify_prot_commit() (whose generic implementation
just calls set_huge_pte_at()) then writes the stale snapshot back over the
live hardware bits, losing the update.
The non-hugetlb sibling make_uffd_wp_pte() does this correctly via
ptep_modify_prot_start() / ptep_modify_prot_commit(). Mirror that pattern
for the present-PTE branch. The migration case stays as-is -- migration
entries are non-present, so there's no hardware update to race against.
Link: https://lore.kernel.org/20260529172331.356655-1-kas@kernel.org
Link: https://lore.kernel.org/20260529172331.356655-2-kas@kernel.org
Link: https://lore.kernel.org/all/20260526130509.2748441-1-kirill@shutemov.name/ [1]
Fixes: 52526ca7fd ("fs/proc/task_mmu: implement IOCTL to get and optionally clear info about PTEs")
Signed-off-by: Kiryl Shutsemau <kas@kernel.org>
Reported-by: Sashiko AI review <sashiko-bot@kernel.org>
Reviewed-by: Lorenzo Stoakes <ljs@kernel.org>
Reviewed-by: Dev Jain <dev.jain@arm.com>
Cc: David Hildenbrand <david@kernel.org>
Cc: Michal Hocko <mhocko@suse.com>
Cc: Mike Rapoport <rppt@kernel.org>
Cc: Peter Xu <peterx@redhat.com>
Cc: Suren Baghdasaryan <surenb@google.com>
Cc: Vlastimil Babka <vbabka@kernel.org>
Cc: Balbir Singh <balbirs@nvidia.com>
Cc: <stable@vger.kernel.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
COMPACT_DEFERRED means compaction did not start because past failures
caused the zone to be deferred. try_to_compact_pages() returns the
maximum result seen while walking the zonelist, so a final
COMPACT_DEFERRED result means no later zone reported that compaction
actually ran.
__alloc_pages_direct_compact() skips COMPACTSTALL and COMPACTFAIL
accounting when try_to_compact_pages() returns COMPACT_SKIPPED, but not
when it returns COMPACT_DEFERRED. A deferred-only direct compaction
attempt can therefore look like a stall, and then a failure if the
allocation still cannot be satisfied.
Treat COMPACT_DEFERRED like COMPACT_SKIPPED in this accounting path. If a
later zone runs compaction and returns a result above COMPACT_DEFERRED, or
compact_zone_order() reports COMPACT_SUCCESS for a captured page, the
final result is not COMPACT_DEFERRED and the existing accounting still
runs.
Link: https://lore.kernel.org/tencent_368AF1F3821E46232637BE16D65C45CF3308@qq.com
Fixes: 06dac2f467 ("mm: compaction: update the COMPACT[STALL|FAIL] events properly")
Signed-off-by: fujunjie <fujunjie1@qq.com>
Reviewed-by: Vlastimil Babka (SUSE) <vbabka@kernel.org>
Cc: Brendan Jackman <jackmanb@google.com>
Cc: Johannes Weiner <hannes@cmpxchg.org>
Cc: Michal Hocko <mhocko@suse.com>
Cc: Suren Baghdasaryan <surenb@google.com>
Cc: Zi Yan <ziy@nvidia.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
The force_thp_readahead path in do_sync_mmap_readahead() is gated on
HPAGE_PMD_ORDER <= MAX_PAGECACHE_ORDER and always requests HPAGE_PMD_ORDER
/ HPAGE_PMD_NR. On configurations where HPAGE_PMD_ORDER exceeds
MAX_PAGECACHE_ORDER, notably arm64 with a 64K base page size, VM_HUGEPAGE
mappings cannot use this path and fall back to the non-forced mmap
readahead path even when the mapping supports useful large folios.
Enable forced readahead for mappings that support large folios and request
the max folio order supported by the mapping, capped at 2M. 2MB is chosen
as the cap because it matches the PMD size on x86_64 and on arm64 with 4K
base pages, so the size/memory-pressure tradeoff for folios of that size
is already well understood. On arm64 with 16K and 64K base page sizes,
2MB is also the contiguous-PTE (contpte) block size, so the resulting
folios coalesce into a single TLB entry and reduce TLB pressure on the
readahead path. This will result in 32M folios not being faulted in with
16K base page size for arm64, but with contpte, the performance difference
should be negligible.
The final allocation order may still be clamped by page_cache_ra_order()
to the mapping and request geometry, but this gives VM_HUGEPAGE mappings
on such configurations a large-folio readahead request instead of dropping
back to base-page readahead.
Link: https://lore.kernel.org/20260601102205.3985788-3-usama.arif@linux.dev
Signed-off-by: Usama Arif <usama.arif@linux.dev>
Reviewed-by: Jan Kara <jack@suse.cz>
Reviewed-by: Pedro Falcato <pfalcato@suse.de>
Cc: Alistair Popple <apopple@nvidia.com>
Cc: Al Viro <viro@zeniv.linux.org.uk>
Cc: Baolin Wang <baolin.wang@linux.alibaba.com>
Cc: Barry Song <baohua@kernel.org>
Cc: Catalin Marinas <catalin.marinas@arm.com>
Cc: Christian Brauner <brauner@kernel.org>
Cc: David Hildenbrand <david@kernel.org>
Cc: Dev Jain <dev.jain@arm.com>
Cc: Heiher <r@hev.cc>
Cc: Johannes Weiner <hannes@cmpxchg.org>
Cc: Kees Cook <kees@kernel.org>
Cc: Kevin Brodsky <kevin.brodsky@arm.com>
Cc: Lance Yang <lance.yang@linux.dev>
Cc: Liam R. Howlett <liam@infradead.org>
Cc: Lorenzo Stoakes <ljs@kernel.org>
Cc: Matthew Wilcox (Oracle) <willy@infradead.org>
Cc: Michal Hocko <mhocko@suse.com>
Cc: Mike Rapoport <rppt@kernel.org>
Cc: Nico Pache <npache@redhat.com>
Cc: Pasha Tatashin <pasha.tatashin@soleen.com>
Cc: Rohan McLure <rmclure@linux.ibm.com>
Cc: Ryan Roberts <ryan.roberts@arm.com>
Cc: Shakeel Butt <shakeel.butt@linux.dev>
Cc: Suren Baghdasaryan <surenb@google.com>
Cc: Vlastimil Babka <vbabka@kernel.org>
Cc: Zi Yan <ziy@nvidia.com>
Cc: Kiryl Shutsemau (Meta) <kas@kernel.org>
Cc: Oscar Salvador (SUSE) <osalvador@kernel.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Patch series "mm: improve large folio readahead for exec memory", v7.
Two checks in do_sync_mmap_readahead() limit large-folio readahead:
1. The mmap_miss heuristic is meant to throttle wasteful speculative
readahead. It is currently also applied to the VM_EXEC readahead
path, which is targeted rather than speculative. Once mmap_miss exceeds
MMAP_LOTSAMISS, exec readahead - including the large-folio
order requested by exec_folio_order() - is disabled. On
configurations where the mmap_miss decrement paths are not
active (see patch 1) the counter only grows, so exec readahead
is permanently disabled after the first 100 faults.
2. The force_thp_readahead path is gated only on
HPAGE_PMD_ORDER <= MAX_PAGECACHE_ORDER and always drives the
readahead at HPAGE_PMD_ORDER. Configurations where
HPAGE_PMD_ORDER exceeds MAX_PAGECACHE_ORDER never reach this
path, even when the mapping itself supports usefully large
folios well below the cap.
Both issues are most visible on arm64 with a 64K base page size, where
HPAGE_PMD_ORDER is 13 (512MB) -- above MAX_PAGECACHE_ORDER (11) -- and
where fault_around_pages collapses to 1 disabling should_fault_around()
(one of the two mmap_miss decrement sites). However the fixes are
architecture-agnostic: patch 1 reflects the nature of VM_EXEC readahead
regardless of base page size, and patch 2 generalises the gate so any
mapping advertising a usefully large maximum folio order can benefit.
I created a benchmark that mmaps a large executable file madvises it as
huge and calls RET-stub functions at PAGE_SIZE offsets across it. "Cold"
measures fault + readahead cost. "Random" first faults in all pages with
a sequential sweep (not measured), then measures time for calling random
offsets, isolating iTLB miss cost for scattered execution.
The benchmark results on Neoverse V2 (Grace), arm64 with 64K base pages,
512MB executable file on ext4, averaged over 3 runs:
Phase | Baseline | Patched | Improvement
-----------|--------------|--------------|------------------
Cold fault | 83.4 ms | 41.3 ms | 50% faster
Random | 76.0 ms | 58.3 ms | 23% faster
This patch (of 2):
The mmap_miss heuristic is intended to stop speculative mmap readahead
when a file looks like a random-access workload. That does not fit the
VM_EXEC path very well.
VM_EXEC readahead is already constrained differently from ordinary mmap
read-around: it is bounded by the VMA, uses exec_folio_order() to choose
an order useful for executable mappings, and sets async_size to 0 so it
does not create follow-on readahead. When VM_HUGEPAGE is also present,
the larger readahead is an explicit userspace opt-in.
The mmap_miss counter is decremented from cache-hit paths in
do_async_mmap_readahead() and filemap_map_pages(). Those paths are not
always enough to balance the synchronous miss increments for executable
mappings. In particular, when fault-around is effectively disabled, such
as configurations where fault_around_pages is 1, filemap_map_pages() is
not reached from the fault path. The counter can then become a stale
throttle for VM_EXEC mappings and suppress the readahead behavior that the
executable-specific path is trying to provide.
Skip both mmap_miss increments and decrements for VM_EXEC mappings,
matching the existing VM_SEQ_READ treatment and keeping the counter
accounting symmetric.
Link: https://lore.kernel.org/20260601102205.3985788-1-usama.arif@linux.dev
Link: https://lore.kernel.org/20260601102205.3985788-2-usama.arif@linux.dev
Signed-off-by: Usama Arif <usama.arif@linux.dev>
Reviewed-by: Jan Kara <jack@suse.cz>
Reviewed-by: Kiryl Shutsemau (Meta) <kas@kernel.org>
Reviewed-by: Oscar Salvador (SUSE) <osalvador@kernel.org>
Reviewed-by: Pedro Falcato <pfalcato@suse.de>
Cc: Alistair Popple <apopple@nvidia.com>
Cc: Al Viro <viro@zeniv.linux.org.uk>
Cc: Baolin Wang <baolin.wang@linux.alibaba.com>
Cc: Barry Song <baohua@kernel.org>
Cc: Catalin Marinas <catalin.marinas@arm.com>
Cc: Christian Brauner <brauner@kernel.org>
Cc: David Hildenbrand <david@kernel.org>
Cc: Dev Jain <dev.jain@arm.com>
Cc: Heiher <r@hev.cc>
Cc: Johannes Weiner <hannes@cmpxchg.org>
Cc: Kees Cook <kees@kernel.org>
Cc: Kevin Brodsky <kevin.brodsky@arm.com>
Cc: Lance Yang <lance.yang@linux.dev>
Cc: Liam R. Howlett <liam@infradead.org>
Cc: Lorenzo Stoakes <ljs@kernel.org>
Cc: Matthew Wilcox (Oracle) <willy@infradead.org>
Cc: Michal Hocko <mhocko@suse.com>
Cc: Mike Rapoport <rppt@kernel.org>
Cc: Nico Pache <npache@redhat.com>
Cc: Pasha Tatashin <pasha.tatashin@soleen.com>
Cc: Rohan McLure <rmclure@linux.ibm.com>
Cc: Ryan Roberts <ryan.roberts@arm.com>
Cc: Shakeel Butt <shakeel.butt@linux.dev>
Cc: Suren Baghdasaryan <surenb@google.com>
Cc: Vlastimil Babka <vbabka@kernel.org>
Cc: Zi Yan <ziy@nvidia.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
should_compact_retry() handles COMPACT_SKIPPED by asking
compaction_zonelist_suitable() whether reclaim can make a later compaction
attempt worthwhile. That answer is used for the current allocation, so it
should follow the same zone eligibility rules as the allocation itself.
When cpusets are enabled, allocator slowpath decisions are marked with
ALLOC_CPUSET. The allocation path, direct compaction and reclaim retry
all skip zones rejected by __cpuset_zone_allowed().
compaction_zonelist_suitable() does not apply that filter. It only walks
ac->zonelist/ac->nodemask, so it can return true because a zone that is
not usable for the current allocation would pass __compaction_suitable().
That does not let the allocation use the disallowed zone. Later
allocation and direct compaction paths still apply cpuset filtering.
However, it can make should_compact_retry() retry based on memory that
this allocation cannot use.
Pass gfp_mask down and apply the same ALLOC_CPUSET check in
compaction_zonelist_suitable(). This keeps the retry decision aligned
with the zones that the allocation is allowed to use.
A temporary debugfs probe was also used to call the old and new
compaction_zonelist_suitable() predicates in the same two-node NUMA guest.
The task was restricted to mems=0 while ac->nodemask covered nodes 0-1.
After putting pressure on node0, node0 failed __compaction_suitable() for
order-10 and node1 passed it, but node1 was rejected by
__cpuset_zone_allowed(). In that state the old predicate returned true
and the patched predicate returned false.
Link: https://lore.kernel.org/tencent_F59F2BA2CC5779308E10DF54593C736D3E0A@qq.com
Fixes: 435b3894e7 ("mm:page_alloc: fix the NULL ac->nodemask in __alloc_pages_slowpath()")
Signed-off-by: fujunjie <fujunjie1@qq.com>
Reviewed-by: Vlastimil Babka (SUSE) <vbabka@kernel.org>
Cc: Brendan Jackman <jackmanb@google.com>
Cc: Johannes Weiner <hannes@cmpxchg.org>
Cc: Michal Hocko <mhocko@suse.com>
Cc: Suren Baghdasaryan <surenb@google.com>
Cc: Zi Yan <ziy@nvidia.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
The deferred split queue handles cgroups in a suboptimal fashion. The
queue is per-NUMA node or per-cgroup, not the intersection. That means on
a cgrouped system, a node-restricted allocation entering reclaim can end
up splitting large pages on other nodes:
alloc/unmap
deferred_split_folio()
list_add_tail(memcg->split_queue)
set_shrinker_bit(memcg, node, deferred_shrinker_id)
for_each_zone_zonelist_nodemask(restricted_nodes)
mem_cgroup_iter()
shrink_slab(node, memcg)
shrink_slab_memcg(node, memcg)
if test_shrinker_bit(memcg, node, deferred_shrinker_id)
deferred_split_scan()
walks memcg->split_queue
The shrinker bit adds an imperfect guard rail. As soon as the cgroup has
a single large page on the node of interest, all large pages owned by that
memcg, including those on other nodes, will be split.
list_lru properly sets up per-node, per-cgroup lists. As a bonus, it
streamlines a lot of the list operations and reclaim walks. It's used
widely by other major shrinkers already. Convert the deferred split queue
as well.
The list_lru per-memcg heads are instantiated on demand when the first
object of interest is allocated for a cgroup, by calling
folio_memcg_alloc_deferred(). Add calls to where splittable pages are
created: anon faults, swapin faults, khugepaged collapse.
These calls create all possible node heads for the cgroup at once, so the
migration code (between nodes) doesn't need any special care.
[akpm@linux-foundation.org: fix build with CONFIG_TRANSPARENT_HUGEPAGE=n]
Link: https://lore.kernel.org/202605281620.lc3rtkBm-lkp@intel.com
[hannes@cmpxchg.org: fix cgroup.memory=nokmem handling]
Link: https://lore.kernel.org/ah9PGv12mqai84ES@cmpxchg.org
Link: https://lore.kernel.org/20260527204757.2544958-10-hannes@cmpxchg.org
Signed-off-by: Johannes Weiner <hannes@cmpxchg.org>
Reported-by: Mikhail Zaslonko <zaslonko@linux.ibm.com>
Tested-by: Mikhail Zaslonko <zaslonko@linux.ibm.com>
Acked-by: Shakeel Butt <shakeel.butt@linux.dev>
Reviewed-by: Lorenzo Stoakes (Oracle) <ljs@kernel.org>
Acked-by: Usama Arif <usama.arif@linux.dev>
Reviewed-by: Kairui Song <kasong@tencent.com>
Cc: Baolin Wang <baolin.wang@linux.alibaba.com>
Cc: Barry Song <baohua@kernel.org>
Cc: Dave Chinner <david@fromorbit.com>
Cc: David Hildenbrand (Arm) <david@kernel.org>
Cc: Dev Jain <dev.jain@arm.com>
Cc: Lance Yang <lance.yang@linux.dev>
Cc: Liam R. Howlett <liam@infradead.org>
Cc: Michal Hocko <mhocko@kernel.org>
Cc: Muchun Song <muchun.song@linux.dev>
Cc: Nico Pache <npache@redhat.com>
Cc: Roman Gushchin <roman.gushchin@linux.dev>
Cc: Ryan Roberts <ryan.roberts@arm.com>
Cc: Vasily Gorbik <gor@linux.ibm.com>
Cc: Vlastimil Babka <vbabka@kernel.org>
Cc: Zi Yan <ziy@nvidia.com>
Cc: kernel test robot <lkp@intel.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Patch series "mm: switch THP shrinker to list_lru", v5.
The open-coded deferred split queue has issues. It's not NUMA-aware (when
cgroup is enabled), and it's more complicated in the callsites interacting
with it. Switching to list_lru fixes the NUMA problem and streamlines
things. It also simplifies planned shrinker work.
Patch 1 fixes a pre-existing list_lru bug where the shrinker bit is set on
the caller's memcg rather than the ancestor whose sublist the item
actually lands on after a walk-up. Standalone, backportable; the rest of
the series depends on it.
Patches 2-5 are cleanups and small refactors in list_lru code. They're
basically independent, but make the THP shrinker conversion easier.
Patch 6 extends the list_lru API to allow the caller to control the
locking scope. The THP shrinker has private state it needs to keep
synchronized with the LRU state.
Patch 7 extends the list_lru API with a convenience helper to do list_lru
head allocation (memcg_list_lru_alloc) when coming from a folio. Anon
THPs are instantiated in several places, and with the folio reparenting
patches pending, folio_memcg() access is now a more delicate dance. This
avoids having to replicate that dance everywhere.
Patch 8 flattens the alloc_anon_folio() retry loop so the next patch's
list_lru hook lands as a clean addition rather than nested deep inside an
if (folio) block.
Patch 9 finally switches the deferred_split_queue to list_lru.
This patch (of 9):
When list_lru_add() races with cgroup deletion, the shrinker bit is set on
the wrong group and lost. This can cause a shrinker run to miss the
cgroup that actually has the object.
When the passed in memcg is dead, the function finds the first non-dead
parent from the passed in memcg and adds the object there; but the
shrinker bit is set on the memcg that was passed in.
This bug is as old as the shrinker bitmap itself.
Fix it by returning the "effective" memcg from the locking function, and
have the caller use that.
Link: https://lore.kernel.org/20260527204757.2544958-1-hannes@cmpxchg.org
Link: https://lore.kernel.org/20260527204757.2544958-2-hannes@cmpxchg.org
Fixes: fae91d6d8b ("mm/list_lru.c: set bit in memcg shrinker bitmap on first list_lru item appearance")
Signed-off-by: Johannes Weiner <hannes@cmpxchg.org>
Reported-by: Usama Arif <usama.arif@linux.dev>
Reported-by: Sashiko
Acked-by: Usama Arif <usama.arif@linux.dev>
Reviewed-by: Wei Yang <richard.weiyang@gmail.com>
Cc: Baolin Wang <baolin.wang@linux.alibaba.com>
Cc: Barry Song <baohua@kernel.org>
Cc: Dave Chinner <david@fromorbit.com>
Cc: David Hildenbrand <david@kernel.org>
Cc: Dev Jain <dev.jain@arm.com>
Cc: Kairui Song <ryncsn@gmail.com>
Cc: Lance Yang <lance.yang@linux.dev>
Cc: Liam R. Howlett <liam@infradead.org>
Cc: Lorenzo Stoakes <ljs@kernel.org>
Cc: Michal Hocko <mhocko@kernel.org>
Cc: Mikhail Zaslonko <zaslonko@linux.ibm.com>
Cc: Muchun Song <muchun.song@linux.dev>
Cc: Nico Pache <npache@redhat.com>
Cc: Roman Gushchin <roman.gushchin@linux.dev>
Cc: Ryan Roberts <ryan.roberts@arm.com>
Cc: Shakeel Butt <shakeel.butt@linux.dev>
Cc: Vasily Gorbik <gor@linux.ibm.com>
Cc: Vlastimil Babka <vbabka@kernel.org>
Cc: Zi Yan <ziy@nvidia.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Patch series "zram: writeback fixes", v2.
Brian (privately) reported a "leak" of writeback bitmap in certain cases,
so that backing device can store less pages; and a theoretical data leak
in the trailing bytes of compressed writeback pages. Both issues are low
risk.
This patch (of 2):
When compressed writeback is available writtenback pages contain "garbage"
in PAGE_SIZE - obj_size trailing bytes. That "garbage" is, basically,
whatever data that page held before we got it for writeback. To get
advantage of it an attacker needs to be able to read from active backing
swap device, which is already catastrophic. Still, just in case, zero out
those trailing bytes before writeback to a backing device so that we only
store swap-ed out data there.
Link: https://lore.kernel.org/20260526022754.2377730-1-senozhatsky@chromium.org
Link: https://lore.kernel.org/20260526022754.2377730-3-senozhatsky@chromium.org
Fixes: d38fab605c ("zram: introduce compressed data writeback")
Signed-off-by: Sergey Senozhatsky <senozhatsky@chromium.org>
Suggested-by: Brian Geffon <bgeffon@google.com>
Cc: Jens Axboe <axboe@kernel.dk>
Cc: Minchan Kim <minchan@kernel.org>
Cc: Richard Chang <richardycc@google.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
zram_writeback_slots() loop can terminate with valid reserved backing
device blk_idx. The problem is that cleanup code doesn't release that
reserved blk_idx before zram_writeback_slots() returns, which leads to
blk_idx leak (it becomes permanently busy and can not be used for actual
writeback.) This does not lead to any system instabilities, it only means
that we can writeback less pages. The scenario is hard to hit in practice
as it requires writeabck to race with modification (slot-free or
overwrite) of the final post-processing slot.
Release reserved but unused blk_idx before returning from
zram_writeback_slots().
Link: https://lore.kernel.org/20260526022754.2377730-2-senozhatsky@chromium.org
Fixes: f405066a1f ("zram: introduce writeback bio batching")
Signed-off-by: Sergey Senozhatsky <senozhatsky@chromium.org>
Suggested-by: Brian Geffon <bgeffon@google.com>
Cc: Jens Axboe <axboe@kernel.dk>
Cc: Minchan Kim <minchan@kernel.org>
Cc: Richard Chang <richardycc@google.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Commit 01b9da291c ("mm: memcontrol: convert objcg to be per-memcg
per-node type") split a memcg's single obj_cgroup into one per NUMA node
so that reparenting LRU folios can take per-node lru locks. As a side
effect, the per-CPU obj_stock_pcp -- which caches exactly one cached_objcg
-- thrashes on workloads where threads of the same memcg run on different
NUMA nodes. The kernel test robot reported a 67.7% regression on
stress-ng.switch.ops_per_sec from this pattern.
Mirror the multi-slot pattern already used by memcg_stock_pcp: turn
nr_bytes and cached_objcg into NR_OBJ_STOCK-element arrays, scan all slots
on consume/refill/account, prefer empty slots when inserting, and evict a
slot round-robin only when full. With multiple slots a CPU can hold the
per-node objcg variants of one memcg plus a few siblings without ever
forcing a drain.
A single int8_t index records which slot the cached slab stats belong to;
the stats are flushed on slot or pgdat change. With NR_OBJ_STOCK = 5 the
layout (verified with pahole) is:
offset 0 : lock(1) + index(1) + node_id(2) + slab stats(4) = 8B
offset 8 : nr_bytes[5] = 10B
offset 18 : padding = 6B
offset 24 : cached[5] = 40B
offset 64 : (line 2) work_struct + flags (cold)
so consume_obj_stock, refill_obj_stock and the slab account path each
touch exactly one 64-byte cache line on non-debug 64-bit builds.
Link: https://lore.kernel.org/20260526033931.1760588-5-shakeel.butt@linux.dev
Signed-off-by: Shakeel Butt <shakeel.butt@linux.dev>
Reported-by: kernel test robot <oliver.sang@intel.com>
Closes: https://lore.kernel.org/oe-lkp/202605121641.b6a60cb0-lkp@intel.com
Fixes: 01b9da291c ("mm: memcontrol: convert objcg to be per-memcg per-node type")
Tested-by: kernel test robot <oliver.sang@intel.com>
Reviewed-by: Harry Yoo (Oracle) <harry@kernel.org>
Cc: Alexandre Ghiti <alex@ghiti.fr>
Cc: Johannes Weiner <hannes@cmpxchg.org>
Cc: Joshua Hahn <joshua.hahnjy@gmail.com>
Cc: Michal Hocko <mhocko@kernel.org>
Cc: Muchun Song <muchun.song@linux.dev>
Cc: Qi Zheng <qi.zheng@linux.dev>
Cc: Roman Gushchin <roman.gushchin@linux.dev>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
