There are situations where reclaim kicks in on a system with free memory.
One possible cause is a NUMA imbalance scenario where one or more nodes
are under pressure. It would help if we could easily identify such nodes.
Move the pgscan, pgsteal, and pgrefill counters from vm_event_item to
node_stat_item to provide per-node reclaim visibility. With these
counters as node stats, the values are now displayed in the per-node
section of /proc/zoneinfo, which allows for quick identification of the
affected nodes.
/proc/vmstat continues to report the same counters, aggregated across all
nodes. But the ordering of these items within the readout changes as they
move from the vm events section to the node stats section.
Memcg accounting of these counters is preserved. The relocated counters
remain visible in memory.stat alongside the existing aggregate pgscan and
pgsteal counters.
However, this change affects how the global counters are accumulated.
Previously, the global event count update was gated on !cgroup_reclaim(),
excluding memcg-based reclaim from /proc/vmstat. Now that
mod_lruvec_state() is being used to update the counters, the global
counters will include all reclaim. This is consistent with how pgdemote
counters are already tracked.
Finally, the virtio_balloon driver is updated to use
global_node_page_state() to fetch the counters, as they are no longer
accessible through the vm_events array.
Link: https://lkml.kernel.org/r/20260219235846.161910-1-jp.kobryn@linux.dev
Signed-off-by: JP Kobryn <jp.kobryn@linux.dev>
Suggested-by: Johannes Weiner <hannes@cmpxchg.org>
Acked-by: Michael S. Tsirkin <mst@redhat.com>
Reviewed-by: Vlastimil Babka (SUSE) <vbabka@kernel.org>
Acked-by: Johannes Weiner <hannes@cmpxchg.org>
Acked-by: Shakeel Butt <shakeel.butt@linux.dev>
Acked-by: Michal Hocko <mhocko@suse.com>
Cc: Alistair Popple <apopple@nvidia.com>
Cc: Axel Rasmussen <axelrasmussen@google.com>
Cc: Byungchul Park <byungchul@sk.com>
Cc: David Hildenbrand <david@kernel.org>
Cc: Eugenio Pérez <eperezma@redhat.com>
Cc: Gregory Price <gourry@gourry.net>
Cc: "Huang, Ying" <ying.huang@linux.alibaba.com>
Cc: Jason Wang <jasowang@redhat.com>
Cc: Joshua Hahn <joshua.hahnjy@gmail.com>
Cc: Liam Howlett <liam.howlett@oracle.com>
Cc: Lorenzo Stoakes <lorenzo.stoakes@oracle.com>
Cc: Mathew Brost <matthew.brost@intel.com>
Cc: Mike Rapoport <rppt@kernel.org>
Cc: Muchun Song <muchun.song@linux.dev>
Cc: Qi Zheng <zhengqi.arch@bytedance.com>
Cc: Rakie Kim <rakie.kim@sk.com>
Cc: Roman Gushchin <roman.gushchin@linux.dev>
Cc: Suren Baghdasaryan <surenb@google.com>
Cc: Wei Xu <weixugc@google.com>
Cc: Xuan Zhuo <xuanzhuo@linux.alibaba.com>
Cc: Yuanchu Xie <yuanchu@google.com>
Cc: Zi Yan <ziy@nvidia.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Stop using the maple big node for rebalance operations by changing to more
align with spanning store. The rebalance operation needs its own data
calculation in rebalance_data().
In the event of too much data, the rebalance tries to push the data using
push_data_sib(). If there is insufficient data, the rebalance operation
will rebalance against a sibling (found with rebalance_sib()).
The rebalance starts at the leaf and works its way upward in the tree
using rebalance_ascend(). Most of the code is shared with spanning store
such as the copy node having a new root, but is fundamentally different in
that the data must come from a sibling.
A parent maple state is used to track the parent location to avoid
multiple mas_ascend() calls. The maple state tree location is copied from
the parent to the mas (child) in the ascend step. Ascending itself is
done in the main loop.
Link: https://lkml.kernel.org/r/20260130205935.2559335-23-Liam.Howlett@oracle.com
Signed-off-by: Liam R. Howlett <Liam.Howlett@oracle.com>
Cc: Alice Ryhl <aliceryhl@google.com>
Cc: Andrew Ballance <andrewjballance@gmail.com>
Cc: Arnd Bergmann <arnd@arndb.de>
Cc: Christian Kujau <lists@nerdbynature.de>
Cc: Geert Uytterhoeven <geert@linux-m68k.org>
Cc: Kuninori Morimoto <kuninori.morimoto.gx@renesas.com>
Cc: Matthew Wilcox (Oracle) <willy@infradead.org>
Cc: SeongJae Park <sj@kernel.org>
Cc: Sidhartha Kumar <sidhartha.kumar@oracle.com>
Cc: Suren Baghdasaryan <surenb@google.com>
Cc: Vlastimil Babka <vbabka@suse.cz>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Stop using the maple subtree state and big node in favour of using three
destinations in the maple copy node. That is, expand the way leaves were
handled to all levels of the tree and use the maple copy node to track the
new nodes.
Extract out the sibling init into the data calculation since this is where
the insufficient data can be detected. The remainder of the sibling code
to shift the next iteration is moved to the spanning_ascend() function,
since it is not always needed.
Next introduce the dst_setup() function which will decide how many nodes
are needed to contain the data at this level. Using the destination
count, populate the copy node's dst array with the new nodes and set
d_count to the correct value. Note that this can be tricky in the case of
a leaf node with exactly enough room because of the rule against NULLs at
the end of leaves.
Once the destinations are ready, copy the data by altering the
cp_data_write() function to copy from the sources to the destinations
directly. This eliminates the use of the big node in this code path. On
node completion, node_finalise() will zero out the remaining area and set
the metadata, if necessary.
spanning_ascend() is used to decide if the operation is complete. It may
create a new root, converge into one destination, or continue upwards by
ascending the left and right write maple states.
One test case setup needed to be tweaked so that the targeted node was
surrounded by full nodes.
[akpm@linux-foundation.org: coding-style cleanups]
Link: https://lkml.kernel.org/r/20260130205935.2559335-18-Liam.Howlett@oracle.com
Signed-off-by: Liam R. Howlett <Liam.Howlett@oracle.com>
Cc: Alice Ryhl <aliceryhl@google.com>
Cc: Andrew Ballance <andrewjballance@gmail.com>
Cc: Arnd Bergmann <arnd@arndb.de>
Cc: Christian Kujau <lists@nerdbynature.de>
Cc: Geert Uytterhoeven <geert@linux-m68k.org>
Cc: Kuninori Morimoto <kuninori.morimoto.gx@renesas.com>
Cc: Matthew Wilcox (Oracle) <willy@infradead.org>
Cc: SeongJae Park <sj@kernel.org>
Cc: Sidhartha Kumar <sidhartha.kumar@oracle.com>
Cc: Suren Baghdasaryan <surenb@google.com>
Cc: Vlastimil Babka <vbabka@suse.cz>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Introduce an internal-memory only node type called maple_copy to
facilitate internal copy operations. Use it in mas_spanning_rebalance()
for just the leaf nodes. Initially, the maple_copy node is used to
configure the source nodes and copy the data into the big_node.
The maple_copy contains a list of source entries with start and end
offsets. One of the maple_copy entries can be itself with an offset of 0
to 2, representing the data where the store partially overwrites entries,
or fully overwrites the entry. The side effect is that the source nodes
no longer have to worry about partially copying the existing offset if it
is not fully overwritten.
This is in preparation of removal of the maple big_node, but for the time
being the data is copied to the big node to limit the change size.
Link: https://lkml.kernel.org/r/20260130205935.2559335-12-Liam.Howlett@oracle.com
Signed-off-by: Liam R. Howlett <Liam.Howlett@oracle.com>
Cc: Alice Ryhl <aliceryhl@google.com>
Cc: Andrew Ballance <andrewjballance@gmail.com>
Cc: Arnd Bergmann <arnd@arndb.de>
Cc: Christian Kujau <lists@nerdbynature.de>
Cc: Geert Uytterhoeven <geert@linux-m68k.org>
Cc: Kuninori Morimoto <kuninori.morimoto.gx@renesas.com>
Cc: Matthew Wilcox (Oracle) <willy@infradead.org>
Cc: SeongJae Park <sj@kernel.org>
Cc: Sidhartha Kumar <sidhartha.kumar@oracle.com>
Cc: Suren Baghdasaryan <surenb@google.com>
Cc: Vlastimil Babka <vbabka@suse.cz>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Patch series "maple_tree: Replace big node with maple copy", v3.
The big node struct was created for simplicity of splitting, rebalancing,
and spanning store operations by using a copy buffer to create the data
necessary prior to breaking it up into 256B nodes. Certain operations
were rather tricky due to the restriction of keeping NULL entries together
and never at the end of a node (except the right-most node).
The big node struct is incompatible with future features that are
currently in development. Specifically different node types and different
data type sizes for pivots.
The big node struct was also a stack variable, which caused issues with
certain configurations of kernel build.
This series removes big node by introducing another node type which will
never be written to the tree: maple_copy. The maple copy node operates
more like a scatter/gather operation with a number of sources and
destinations of allocated nodes.
The sources are copied to the destinations, in turn, until the sources are
exhausted. The destination is changed if it is filled or the split
location is reached prior to the source data end.
New data is inserted by using the maple copy node itself as a source with
up to 3 slots and pivots. The data in the maple copy node is the data
being written to the tree along with any fragment of the range(s) being
overwritten.
As with all nodes, the maple copy node is of size 256B. Using a node type
allows for the copy operation to treat the new data stored in the maple
copy node the same as any other source node.
Analysis of the runtime shows no regression or benefit of removing the
larger stack structure. The motivation is the ground work to use new node
types and to help those with odd configurations that have had issues.
The change was tested by myself using mm_tests on amd64 and by Suren on
android (arm64). Limited testing on s390 qemu was also performed using
stress-ng on the virtual memory, which should cover many corner cases.
This patch (of 30):
Use RCU_INIT_POINTER to initialize an rcu pointer to an initial value
since there are no readers within the tree being created during
duplication. There is no risk of readers seeing the initialized or
uninitialized value until after the synchronization call in
mas_dup_buld().
Link: https://lkml.kernel.org/r/20260130205935.2559335-1-Liam.Howlett@oracle.com
Link: https://lkml.kernel.org/r/20260130205935.2559335-2-Liam.Howlett@oracle.com
Signed-off-by: Liam R. Howlett <Liam.Howlett@oracle.com>
Cc: Alice Ryhl <aliceryhl@google.com>
Cc: Andrew Ballance <andrewjballance@gmail.com>
Cc: Arnd Bergmann <arnd@arndb.de>
Cc: Christian Kujau <lists@nerdbynature.de>
Cc: Geert Uytterhoeven <geert@linux-m68k.org>
Cc: Kuninori Morimoto <kuninori.morimoto.gx@renesas.com>
Cc: Matthew Wilcox (Oracle) <willy@infradead.org>
Cc: SeongJae Park <sj@kernel.org>
Cc: Sidhartha Kumar <sidhartha.kumar@oracle.com>
Cc: Suren Baghdasaryan <surenb@google.com>
Cc: Vlastimil Babka <vbabka@suse.cz>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Currently, mmu_notifier_clear_young() is called immediately after
pmdp_clear_young_notify(), which already calls mmu_notifier_clear_young()
internally. This results in a redundant notifier call.
Replace pmdp_clear_young_notify() with the non-notify variant to avoid the
duplicate call and make the pmdp path consistent with the corresponding
ptep_mkold() code.
Link: https://lkml.kernel.org/r/20260203095400.2465255-1-qin.yuA@h3c.com
Signed-off-by: qinyu <qin.yuA@h3c.com>
Reviewed-by: SeongJae Park <sj@kernel.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Pull kvm fixes from Paolo Bonzini:
"ARM:
- Clear the pending exception state from a vcpu coming out of reset,
as it could otherwise affect the first instruction executed in the
guest
- Fix pointer arithmetic in address translation emulation, so that
the Hardware Access bit is set on the correct PTE instead of some
other location
s390:
- Fix deadlock in new memory management
- Properly handle kernel faults on donated memory
- Fix bounds checking for irq routing, with selftest
- Fix invalid machine checks and log all of them"
* tag 'for-linus' of git://git.kernel.org/pub/scm/virt/kvm/kvm:
KVM: arm64: Fix the descriptor address in __kvm_at_swap_desc()
KVM: s390: vsie: Avoid injecting machine check on signal
KVM: s390: log machine checks more aggressively
KVM: s390: selftests: Add IRQ routing address offset tests
KVM: s390: Limit adapter indicator access to mapped page
s390/mm: Add missing secure storage access fixups for donated memory
KVM: arm64: Discard PC update state on vcpu reset
KVM: s390: Fix a deadlock
Pull Compute Express Link (CXL) fixes from Dave Jiang:
- Adjust the startup priority of cxl_pmem to be higher than that of
cxl_acpi
- Use proper endpoint validity check upon sanitize
- Avoid incorrect DVSEC fallback when HDM decoders are enabled
- Fix CXL_ACPI and CXL_PMEM Kconfig tristate mismatch
- Fix leakage in __construct_region()
- Fix use after free of parent_port in cxl_detach_ep()
* tag 'cxl-fixes-7.0-rc6' of git://git.kernel.org/pub/scm/linux/kernel/git/cxl/cxl:
cxl: Adjust the startup priority of cxl_pmem to be higher than that of cxl_acpi
cxl/mbox: Use proper endpoint validity check upon sanitize
cxl/hdm: Avoid incorrect DVSEC fallback when HDM decoders are enabled
cxl/acpi: Fix CXL_ACPI and CXL_PMEM Kconfig tristate mismatch
cxl/region: Fix leakage in __construct_region()
cxl/port: Fix use after free of parent_port in cxl_detach_ep()
KVM/arm64 fixes for 7.0, take #4
- Clear the pending exception state from a vcpu coming out of
reset, as it could otherwise affect the first instruction
executed in the guest.
- Fix the address translation emulation icode to set the Hardware
Access bit on the correct PTE instead of some other location.
Pull MM fixes from Andrew Morton:
"6 hotfixes. 2 are cc:stable. All are for MM.
All are singletons - please see the changelogs for details"
* tag 'mm-hotfixes-stable-2026-03-23-17-56' of git://git.kernel.org/pub/scm/linux/kernel/git/akpm/mm:
mm/damon/stat: monitor all System RAM resources
mm/zswap: add missing kunmap_local()
mailmap: update email address for Muhammad Usama Anjum
zram: do not slot_free() written-back slots
mm/damon/core: avoid use of half-online-committed context
mm/rmap: clear vma->anon_vma on error
Pull perf tools fixes from Arnaldo Carvalho de Melo:
- Fix parsing 'overwrite' in command line event definitions in
big-endian machines by writing correct union member
- Fix finding default metric in 'perf stat'
- Fix relative paths for including headers in 'perf kvm stat'
- Sync header copies with the kernel sources: msr-index.h, kvm,
build_bug.h
* tag 'perf-tools-fixes-for-v7.0-2-2026-03-23' of git://git.kernel.org/pub/scm/linux/kernel/git/perf/perf-tools:
tools headers: Synchronize linux/build_bug.h with the kernel sources
tools headers UAPI: Sync x86's asm/kvm.h with the kernel sources
tools headers UAPI: Sync linux/kvm.h with the kernel sources
tools arch x86: Sync the msr-index.h copy with the kernel sources
perf kvm stat: Fix relative paths for including headers
perf parse-events: Fix big-endian 'overwrite' by writing correct union member
perf metricgroup: Fix metricgroup__has_metric_or_groups()
tools headers: Skip arm64 cputype.h check
Pull media fixes from Mauro Carvalho Chehab:
- rkvdec: fix stack usage with clang and improve handling missing
short/long term RPS
- synopsys: fix a Kconfig issue and an out-of-bounds check
- verisilicon: Fix kernel panic due to __initconst misuse
- media core: serialize REINIT and REQBUFS with req_queue_mutex
* tag 'media/v7.0-5' of git://git.kernel.org/pub/scm/linux/kernel/git/mchehab/linux-media:
media: verisilicon: Fix kernel panic due to __initconst misuse
media: rkvdec: reduce stack usage in rkvdec_init_v4l2_vp9_count_tbl()
media: rkvdec: reduce excessive stack usage in assemble_hw_pps()
media: rkvdec: Improve handling missing short/long term RPS
media: mc, v4l2: serialize REINIT and REQBUFS with req_queue_mutex
media: synopsys: csi2rx: add missing kconfig dependency
media: synopsys: csi2rx: fix out-of-bounds check for formats array
Pull xen fixes from Juergen Gross:
"Restrict the xen privcmd driver in unprivileged domU to only allow
hypercalls to target domain when using secure boot"
* tag 'xsa482-7.0-tag' of git://git.kernel.org/pub/scm/linux/kernel/git/xen/tip:
xen/privcmd: add boot control for restricted usage in domU
xen/privcmd: restrict usage in unprivileged domU
DAMON_STAT usage document (Documentation/admin-guide/mm/damon/stat.rst)
says it monitors the system's entire physical memory. But, it is
monitoring only the biggest System RAM resource of the system. When there
are multiple System RAM resources, this results in monitoring only an
unexpectedly small fraction of the physical memory. For example, suppose
the system has a 500 GiB System RAM, 10 MiB non-System RAM, and 500 GiB
System RAM resources in order on the physical address space. DAMON_STAT
will monitor only the first 500 GiB System RAM. This situation is
particularly common on NUMA systems.
Select a physical address range that covers all System RAM areas of the
system, to fix this issue and make it work as documented.
[sj@kernel.org: return error if monitoring target region is invalid]
Link: https://lkml.kernel.org/r/20260317053631.87907-1-sj@kernel.org
Link: https://lkml.kernel.org/r/20260316235118.873-1-sj@kernel.org
Fixes: 369c415e60 ("mm/damon: introduce DAMON_STAT module")
Signed-off-by: SeongJae Park <sj@kernel.org>
Cc: <stable@vger.kernel.org> [6.17+]
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Commit e2c3b6b21c ("mm: zswap: use SG list decompression APIs from
zsmalloc") updated zswap_decompress() to use the scatterwalk API to copy
data for uncompressed pages.
In doing so, it mapped kernel memory locally for 32-bit kernels using
kmap_local_folio(), however it never unmapped this memory.
This resulted in the linked syzbot report where a BUG_ON() is triggered
due to leaking the kmap slot.
This patch fixes the issue by explicitly unmapping the established kmap.
Also, add flush_dcache_folio() after the kunmap_local() call
I had assumed that a new folio here combined with the flush that is done at
the point of setting the PTE would suffice, but it doesn't seem that's
actually the case, as update_mmu_cache() will in many archtectures only
actually flush entries where a dcache flush was done on a range previously.
I had also wondered whether kunmap_local() might suffice, but it doesn't
seem to be the case.
Some arches do seem to actually dcache flush on unmap, parisc does it if
CONFIG_HIGHMEM is not set by setting ARCH_HAS_FLUSH_ON_KUNMAP and calling
kunmap_flush_on_unmap() from __kunmap_local(), otherwise non-CONFIG_HIGHMEM
callers do nothing here.
Otherwise arch_kmap_local_pre_unmap() is called which does:
* sparc - flush_cache_all()
* arm - if VIVT, __cpuc_flush_dcache_area()
* otherwise - nothing
Also arch_kmap_local_post_unmap() is called which does:
* arm - local_flush_tlb_kernel_page()
* csky - kmap_flush_tlb()
* microblaze, ppc - local_flush_tlb_page()
* mips - local_flush_tlb_one()
* sparc - flush_tlb_all() (again)
* x86 - arch_flush_lazy_mmu_mode()
* otherwise - nothing
But this is only if it's high memory, and doesn't cover all architectures,
so is presumably intended to handle other cache consistency concerns.
In any case, VIPT is problematic here whether low or high memory (in spite
of what the documentation claims, see [0] - 'the kernel did write to a page
that is in the page cache page and / or in high memory'), because dirty
cache lines may exist at the set indexed by the kernel direct mapping,
which won't exist in the set indexed by any subsequent userland mapping,
meaning userland might read stale data from L2 cache.
Even if the documentation is correct and low memory is fine not to be
flushed here, we can't be sure as to whether the memory is low or high
(kmap_local_folio() will be a no-op if low), and this call should be
harmless if it is low.
VIVT would require more work if the memory were shared and already mapped,
but this isn't the case here, and would anyway be handled by the dcache
flush call.
In any case, we definitely need this flush as far as I can tell.
And we should probably consider updating the documentation unless it turns
out there's somehow dcache synchronisation that happens for low
memory/64-bit kernels elsewhere?
[ljs@kernel.org: add flush_dcache_folio() after the kunmap_local() call]
Link: https://lkml.kernel.org/r/13e09a99-181f-45ac-a18d-057faf94bccb@lucifer.local
Link: https://lkml.kernel.org/r/20260316140122.339697-1-ljs@kernel.org
Link: https://docs.kernel.org/core-api/cachetlb.html [0]
Fixes: e2c3b6b21c ("mm: zswap: use SG list decompression APIs from zsmalloc")
Signed-off-by: Lorenzo Stoakes (Oracle) <ljs@kernel.org>
Reported-by: syzbot+fe426bef95363177631d@syzkaller.appspotmail.com
Closes: https://lore.kernel.org/all/69b75e2c.050a0220.12d28.015a.GAE@google.com
Acked-by: Yosry Ahmed <yosry@kernel.org>
Acked-by: Johannes Weiner <hannes@cmpxchg.org>
Reviewed-by: SeongJae Park <sj@kernel.org>
Acked-by: Yosry Ahmed <yosry@kernel.org>
Acked-by: Nhat Pham <nphamcs@gmail.com>
Cc: Chengming Zhou <chengming.zhou@linux.dev>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
