Patch series "mm/damon/sysfs-schemes: implement DAMOS tried total bytes
file".
The tried_regions directory of DAMON sysfs interface is useful for
retrieving monitoring results snapshot or DAMOS debugging. However, for
common use case that need to monitor only the total size of the scheme
tried regions (e.g., monitoring working set size), the kernel overhead for
directory construction and user overhead for reading the content could be
high if the number of monitoring region is not small. This patchset
implements DAMON sysfs files for efficient support of the use case.
The first patch implements the sysfs file to reduce the user space
overhead, and the second patch implements a command for reducing the
kernel space overhead.
The third patch adds a selftest for the new file, and following two
patches update documents.
[1] https://lore.kernel.org/damon/20230728201817.70602-1-sj@kernel.org/
This patch (of 5):
The tried_regions directory can be used for retrieving the monitoring
results snapshot for regions of specific access pattern, by setting the
scheme's action as 'stat' and the access pattern as required. While the
interface provides every detail of the monitoring results, some use cases
including working set size monitoring requires only the total size of the
regions. For such cases, users should read all the information and
calculate the total size of the regions. However, it could incur high
overhead if the number of regions is high. Add a file for retrieving only
the information, namely 'total_bytes' file. It allows users to get the
total size by reading only the file.
Link: https://lkml.kernel.org/r/20230802213222.109841-1-sj@kernel.org
Link: https://lkml.kernel.org/r/20230802213222.109841-2-sj@kernel.org
Signed-off-by: SeongJae Park <sj@kernel.org>
Cc: Jonathan Corbet <corbet@lwn.net>
Cc: Shuah Khan <shuah@kernel.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
MGLRU has a LRU list for each zone for each type (anon/file) in each
generation:
long nr_pages[MAX_NR_GENS][ANON_AND_FILE][MAX_NR_ZONES];
The min_seq (oldest generation) can progress independently for each
type but the max_seq (youngest generation) is shared for both anon and
file. This is to maintain a common frame of reference.
In order for eviction to advance the min_seq of a type, all the per-zone
lists in the oldest generation of that type must be empty.
The eviction logic only considers pages from eligible zones for
eviction or promotion.
scan_folios() {
...
for (zone = sc->reclaim_idx; zone >= 0; zone--) {
...
sort_folio(); // Promote
...
isolate_folio(); // Evict
}
...
}
Consider the system has the movable zone configured and default 4
generations. The current state of the system is as shown below
(only illustrating one type for simplicity):
Type: ANON
Zone DMA32 Normal Movable Device
Gen 0 0 0 4GB 0
Gen 1 0 1GB 1MB 0
Gen 2 1MB 4GB 1MB 0
Gen 3 1MB 1MB 1MB 0
Now consider there is a GFP_KERNEL allocation request (eligible zone
index <= Normal), evict_folios() will return without doing any work
since there are no pages to scan in the eligible zones of the oldest
generation. Reclaim won't make progress until triggered from a ZONE_MOVABLE
allocation request; which may not happen soon if there is a lot of free
memory in the movable zone. This can lead to OOM kills, although there
is 1GB pages in the Normal zone of Gen 1 that we have not yet tried to
reclaim.
This issue is not seen in the conventional active/inactive LRU since
there are no per-zone lists.
If there are no (not enough) folios to scan in the eligible zones, move
folios from ineligible zone (zone_index > reclaim_index) to the next
generation. This allows for the progression of min_seq and reclaiming
from the next generation (Gen 1).
Qualcomm, Mediatek and raspberrypi [1] discovered this issue independently.
[1] https://github.com/raspberrypi/linux/issues/5395
Link: https://lkml.kernel.org/r/20230802025606.346758-1-kaleshsingh@google.com
Fixes: ac35a49023 ("mm: multi-gen LRU: minimal implementation")
Signed-off-by: Kalesh Singh <kaleshsingh@google.com>
Reported-by: Charan Teja Kalla <quic_charante@quicinc.com>
Reported-by: Lecopzer Chen <lecopzer.chen@mediatek.com>
Tested-by: AngeloGioacchino Del Regno <angelogioacchino.delregno@collabora.com> [mediatek]
Tested-by: Charan Teja Kalla <quic_charante@quicinc.com>
Cc: Yu Zhao <yuzhao@google.com>
Cc: Barry Song <baohua@kernel.org>
Cc: Brian Geffon <bgeffon@google.com>
Cc: Jan Alexander Steffens (heftig) <heftig@archlinux.org>
Cc: Matthias Brugger <matthias.bgg@gmail.com>
Cc: Oleksandr Natalenko <oleksandr@natalenko.name>
Cc: Qi Zheng <zhengqi.arch@bytedance.com>
Cc: Steven Barrett <steven@liquorix.net>
Cc: Suleiman Souhlal <suleiman@google.com>
Cc: Suren Baghdasaryan <surenb@google.com>
Cc: Aneesh Kumar K V <aneesh.kumar@linux.ibm.com>
Cc: <stable@vger.kernel.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Before commit f53af4285d ("mm: vmscan: fix extreme overreclaim and swap
floods"), proactive reclaim will extreme overreclaim sometimes. But
proactive reclaim still inaccurate and some extent overreclaim.
Problematic case is easy to construct. Allocate lots of anonymous memory
(e.g., 20G) in a memcg, then swapping by writing memory.recalim and there
is a certain probability of overreclaim. For example, request 1G by
writing memory.reclaim will eventually reclaim 1.7G or other values more
than 1G.
The reason is that reclaimer may have already reclaimed part of requested
memory in one loop, but before adjust sc->nr_to_reclaim in outer loop,
call shrink_lruvec() again will still follow the current sc->nr_to_reclaim
to work. It will eventually lead to overreclaim. In theory, the amount
of reclaimed would be in [request, 2 * request).
Reclaimer usually tends to reclaim more than request. But either direct
or kswapd reclaim have much smaller nr_to_reclaim targets, so it is less
noticeable and not have much impact.
Proactive reclaim can usually come in with a larger value, so the error is
difficult to ignore. Considering proactive reclaim is usually low
frequency, handle the batching into smaller chunks is a better approach.
Link: https://lkml.kernel.org/r/20230721014116.3388-1-yangyifei03@kuaishou.com
Signed-off-by: Efly Young <yangyifei03@kuaishou.com>
Suggested-by: Johannes Weiner <hannes@cmpxchg.org>
Acked-by: Johannes Weiner <hannes@cmpxchg.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Support using multiple zpools of the same type in zswap, for concurrency
purposes. A fixed number of 32 zpools is suggested by this commit, which
was determined empirically. It can be later changed or made into a config
option if needed.
On a setup with zswap and zsmalloc, comparing a single zpool to 32 zpools
shows improvements in the zsmalloc lock contention, especially on the swap
out path.
The following shows the perf analysis of the swapout path when 10
workloads are simultaneously reclaiming and refaulting tmpfs pages. There
are some improvements on the swap in path as well, but less significant.
1 zpool:
|--28.99%--zswap_frontswap_store
|
<snip>
|
|--8.98%--zpool_map_handle
| |
| --8.98%--zs_zpool_map
| |
| --8.95%--zs_map_object
| |
| --8.38%--_raw_spin_lock
| |
| --7.39%--queued_spin_lock_slowpath
|
|--8.82%--zpool_malloc
| |
| --8.82%--zs_zpool_malloc
| |
| --8.80%--zs_malloc
| |
| |--7.21%--_raw_spin_lock
| | |
| | --6.81%--queued_spin_lock_slowpath
<snip>
32 zpools:
|--16.73%--zswap_frontswap_store
|
<snip>
|
|--1.81%--zpool_malloc
| |
| --1.81%--zs_zpool_malloc
| |
| --1.79%--zs_malloc
| |
| --0.73%--obj_malloc
|
|--1.06%--zswap_update_total_size
|
|--0.59%--zpool_map_handle
| |
| --0.59%--zs_zpool_map
| |
| --0.57%--zs_map_object
| |
| --0.51%--_raw_spin_lock
<snip>
Link: https://lkml.kernel.org/r/20230620194644.3142384-1-yosryahmed@google.com
Signed-off-by: Yosry Ahmed <yosryahmed@google.com>
Suggested-by: Yu Zhao <yuzhao@google.com>
Acked-by: Chris Li (Google) <chrisl@kernel.org>
Reviewed-by: Nhat Pham <nphamcs@gmail.com>
Tested-by: Nhat Pham <nphamcs@gmail.com>
Cc: Dan Streetman <ddstreet@ieee.org>
Cc: Domenico Cerasuolo <cerasuolodomenico@gmail.com>
Cc: Johannes Weiner <hannes@cmpxchg.org>
Cc: Konrad Rzeszutek Wilk <konrad.wilk@oracle.com>
Cc: Seth Jennings <sjenning@redhat.com>
Cc: Vitaly Wool <vitaly.wool@konsulko.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Arm disabled hugetlb vmemmap optimization [1] because hugetlb vmemmap
optimization includes an update of both the permissions (writeable to
read-only) and the output address (pfn) of the vmemmap ptes. That is not
supported without unmapping of pte(marking it invalid) by some
architectures.
With DAX vmemmap optimization we don't require such pte updates and
architectures can enable DAX vmemmap optimization while having hugetlb
vmemmap optimization disabled. Hence split DAX optimization support into
a different config.
s390, loongarch and riscv don't have devdax support. So the DAX config is
not enabled for them. With this change, arm64 should be able to select
DAX optimization
[1] commit 060a2c92d1 ("arm64: mm: hugetlb: Disable HUGETLB_PAGE_OPTIMIZE_VMEMMAP")
Link: https://lkml.kernel.org/r/20230724190759.483013-8-aneesh.kumar@linux.ibm.com
Signed-off-by: Aneesh Kumar K.V <aneesh.kumar@linux.ibm.com>
Cc: Catalin Marinas <catalin.marinas@arm.com>
Cc: Christophe Leroy <christophe.leroy@csgroup.eu>
Cc: Dan Williams <dan.j.williams@intel.com>
Cc: Joao Martins <joao.m.martins@oracle.com>
Cc: Michael Ellerman <mpe@ellerman.id.au>
Cc: Mike Kravetz <mike.kravetz@oracle.com>
Cc: Muchun Song <muchun.song@linux.dev>
Cc: Nicholas Piggin <npiggin@gmail.com>
Cc: Oscar Salvador <osalvador@suse.de>
Cc: Will Deacon <will@kernel.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Patch series "Add support for DAX vmemmap optimization for ppc64", v6.
This patch series implements changes required to support DAX vmemmap
optimization for ppc64. The vmemmap optimization is only enabled with
radix MMU translation and 1GB PUD mapping with 64K page size.
The patch series also splits the hugetlb vmemmap optimization as a
separate Kconfig variable so that architectures can enable DAX vmemmap
optimization without enabling hugetlb vmemmap optimization. This should
enable architectures like arm64 to enable DAX vmemmap optimization while
they can't enable hugetlb vmemmap optimization. More details of the same
are in patch "mm/vmemmap optimization: Split hugetlb and devdax vmemmap
optimization".
With 64K page size for 16384 pages added (1G) we save 14 pages
With 4K page size for 262144 pages added (1G) we save 4094 pages
With 4K page size for 512 pages added (2M) we save 6 pages
This patch (of 13):
Architectures like powerpc would like to enable transparent huge page pud
support only with radix translation. To support that add
has_transparent_pud_hugepage() helper that architectures can override.
[aneesh.kumar@linux.ibm.com: use the new has_transparent_pud_hugepage()]
Link: https://lkml.kernel.org/r/87tttrvtaj.fsf@linux.ibm.com
Link: https://lkml.kernel.org/r/20230724190759.483013-1-aneesh.kumar@linux.ibm.com
Link: https://lkml.kernel.org/r/20230724190759.483013-2-aneesh.kumar@linux.ibm.com
Signed-off-by: Aneesh Kumar K.V <aneesh.kumar@linux.ibm.com>
Reviewed-by: Christophe Leroy <christophe.leroy@csgroup.eu>
Cc: Catalin Marinas <catalin.marinas@arm.com>
Cc: Dan Williams <dan.j.williams@intel.com>
Cc: Joao Martins <joao.m.martins@oracle.com>
Cc: Michael Ellerman <mpe@ellerman.id.au>
Cc: Mike Kravetz <mike.kravetz@oracle.com>
Cc: Muchun Song <muchun.song@linux.dev>
Cc: Nicholas Piggin <npiggin@gmail.com>
Cc: Oscar Salvador <osalvador@suse.de>
Cc: Will Deacon <will@kernel.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Postpone checking the VMA_LOCK flag until we've attempted to handle faults
on PUDs. There's a mild upside to this patch in that we'll allocate the
page tables while under the VMA lock rather than the mmap lock, reducing
the hold time on the mmap lock, since the retry will find the page tables
already populated. The real purpose here is to make a commit that shows
we don't call ->huge_fault under the VMA lock. We do now handle setting
the accessed bit on a PUD fault under the VMA lock, but that doesn't seem
likely to be a measurable difference.
Link: https://lkml.kernel.org/r/20230724185410.1124082-5-willy@infradead.org
Signed-off-by: Matthew Wilcox (Oracle) <willy@infradead.org>
Cc: Arjun Roy <arjunroy@google.com>
Cc: Eric Dumazet <edumazet@google.com>
Cc: Punit Agrawal <punit.agrawal@bytedance.com>
Cc: Suren Baghdasaryan <surenb@google.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
By delaying the setting of prev/next VMA until after the write of NULL,
the probability of the prev/next VMA already being in the CPU cache is
significantly increased, especially for larger munmap operations. It
also means that prev/next will be loaded closer to when they are used.
This requires changing the loop type when gathering the VMAs that will
be freed.
Since prev will be set later in the function, it is better to reverse
the splitting direction of the start VMA (modify the new_below argument
to __split_vma).
Using the vma_iter_prev_range() to walk back to the correct location in
the tree will, on the most part, mean walking within the CPU cache.
Usually, this is two steps vs a node reset and a tree re-walk.
Link: https://lkml.kernel.org/r/20230724183157.3939892-16-Liam.Howlett@oracle.com
Signed-off-by: Liam R. Howlett <Liam.Howlett@oracle.com>
Cc: Peng Zhang <zhangpeng.00@bytedance.com>
Cc: Suren Baghdasaryan <surenb@google.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
The majority of the calls to munmap a vm range is within a single vma.
The maple tree is able to store a single entry at 0, with a size of 1 as
a pointer and avoid any allocations. Change do_vmi_align_munmap() to
store the VMAs being munmap()'ed into a tree indexed by the count. This
will leverage the ability to store the first entry without a node
allocation.
Storing the entries into a tree by the count and not the vma start and
end means changing the functions which iterate over the entries. Update
unmap_vmas() and free_pgtables() to take a maple state and a tree end
address to support this functionality.
Passing through the same maple state to unmap_vmas() and free_pgtables()
means the state needs to be reset between calls. This happens in the
static unmap_region() and exit_mmap().
Link: https://lkml.kernel.org/r/20230724183157.3939892-4-Liam.Howlett@oracle.com
Signed-off-by: Liam R. Howlett <Liam.Howlett@oracle.com>
Cc: Peng Zhang <zhangpeng.00@bytedance.com>
Cc: Suren Baghdasaryan <surenb@google.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Despite its name, mm_drop_all_locks() does not drop _all_ locks; the mmap
lock is held write-locked by the caller, and the caller is responsible for
dropping the mmap lock at a later point (which will also release the VMA
locks).
Calling vma_end_write_all() here is dangerous because the caller might
have write-locked a VMA with the expectation that it will stay
write-locked until the mmap_lock is released, as usual.
This _almost_ becomes a problem in the following scenario:
An anonymous VMA A and an SGX VMA B are mapped adjacent to each other.
Userspace calls munmap() on a range starting at the start address of A and
ending in the middle of B.
Hypothetical call graph with additional notes in brackets:
do_vmi_align_munmap
[begin first for_each_vma_range loop]
vma_start_write [on VMA A]
vma_mark_detached [on VMA A]
__split_vma [on VMA B]
sgx_vma_open [== new->vm_ops->open]
sgx_encl_mm_add
__mmu_notifier_register [luckily THIS CAN'T ACTUALLY HAPPEN]
mm_take_all_locks
mm_drop_all_locks
vma_end_write_all [drops VMA lock taken on VMA A before]
vma_start_write [on VMA B]
vma_mark_detached [on VMA B]
[end first for_each_vma_range loop]
vma_iter_clear_gfp [removes VMAs from maple tree]
mmap_write_downgrade
unmap_region
mmap_read_unlock
In this hypothetical scenario, while do_vmi_align_munmap() thinks it still
holds a VMA write lock on VMA A, the VMA write lock has actually been
invalidated inside __split_vma().
The call from sgx_encl_mm_add() to __mmu_notifier_register() can't
actually happen here, as far as I understand, because we are duplicating
an existing SGX VMA, but sgx_encl_mm_add() only calls
__mmu_notifier_register() for the first SGX VMA created in a given
process. So this could only happen in fork(), not on munmap(). But in my
view it is just pure luck that this can't happen.
Also, we wouldn't actually have any bad consequences from this in
do_vmi_align_munmap(), because by the time the bug drops the lock on VMA
A, we've already marked VMA A as detached, which makes it completely
ineligible for any VMA-locked page faults. But again, that's just pure
luck.
So remove the vma_end_write_all(), so that VMA write locks are only ever
released on mmap_write_unlock() or mmap_write_downgrade().
Also add comments to document the locking rules established by this patch.
Link: https://lkml.kernel.org/r/20230720193436.454247-1-jannh@google.com
Fixes: eeff9a5d47 ("mm/mmap: prevent pagefault handler from racing with mmu_notifier registration")
Signed-off-by: Jann Horn <jannh@google.com>
Reviewed-by: Suren Baghdasaryan <surenb@google.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
