Patch series "page allocation tag compression", v4.
This patchset implements several improvements:
1. Gracefully handles module unloading while there are used
allocations allocated from that module;
2. Provides an option to store page allocation tag references in the
page flags, removing dependency on page extensions and eliminating the
memory overhead from storing page allocation references (~0.2% of total
system memory). This also improves page allocation performance when
CONFIG_MEM_ALLOC_PROFILING is enabled by eliminating page extension
lookup. Page allocation performance overhead is reduced from 41% to
5.5%.
Patch #1 introduces mas_for_each_rev() helper function.
Patch #2 introduces shutdown_mem_profiling() helper function to be used
when disabling memory allocation profiling.
Patch #3 copies module tags into virtually contiguous memory which
serves two purposes:
- Lets us deal with the situation when module is unloaded while there
are still live allocations from that module. Since we are using a copy
version of the tags we can safely unload the module. Space and gaps in
this contiguous memory are managed using a maple tree.
- Enables simple indexing of the tags in the later patches.
Patch #4 changes the way we allocate virtually contiguous memory for
module tags to reserve only vitrual area and populate physical pages only
as needed at module load time.
Patch #5 abstracts page allocation tag reference to simplify later
changes.
Patch #6 adds compression option to the sysctl.vm.mem_profiling boot
parameter for storing page allocation tag references inside page flags if
they fit. If the number of available page flag bits is insufficient to
address all kernel allocations, memory allocation profiling gets disabled
with an appropriate warning.
This patch (of 6):
Add mas_for_each_rev() function to iterate maple tree nodes in reverse
order.
Link: https://lkml.kernel.org/r/20241023170759.999909-1-surenb@google.com
Link: https://lkml.kernel.org/r/20241023170759.999909-2-surenb@google.com
Signed-off-by: Suren Baghdasaryan <surenb@google.com>
Suggested-by: Liam R. Howlett <Liam.Howlett@Oracle.com>
Reviewed-by: Liam R. Howlett <Liam.Howlett@Oracle.com>
Reviewed-by: Pasha Tatashin <pasha.tatashin@soleen.com>
Cc: Ard Biesheuvel <ardb@kernel.org>
Cc: Arnd Bergmann <arnd@arndb.de>
Cc: Borislav Petkov (AMD) <bp@alien8.de>
Cc: Christoph Hellwig <hch@infradead.org>
Cc: Daniel Gomez <da.gomez@samsung.com>
Cc: David Hildenbrand <david@redhat.com>
Cc: Davidlohr Bueso <dave@stgolabs.net>
Cc: David Rientjes <rientjes@google.com>
Cc: Dennis Zhou <dennis@kernel.org>
Cc: Johannes Weiner <hannes@cmpxchg.org>
Cc: John Hubbard <jhubbard@nvidia.com>
Cc: Jonathan Corbet <corbet@lwn.net>
Cc: Joonsoo Kim <iamjoonsoo.kim@lge.com>
Cc: Kalesh Singh <kaleshsingh@google.com>
Cc: Kees Cook <keescook@chromium.org>
Cc: Kent Overstreet <kent.overstreet@linux.dev>
Cc: Luis Chamberlain <mcgrof@kernel.org>
Cc: Matthew Wilcox <willy@infradead.org>
Cc: Michal Hocko <mhocko@suse.com>
Cc: Mike Rapoport (Microsoft) <rppt@kernel.org>
Cc: Minchan Kim <minchan@google.com>
Cc: Paul E. McKenney <paulmck@kernel.org>
Cc: Petr Pavlu <petr.pavlu@suse.com>
Cc: Roman Gushchin <roman.gushchin@linux.dev>
Cc: Sami Tolvanen <samitolvanen@google.com>
Cc: Sourav Panda <souravpanda@google.com>
Cc: Steven Rostedt (Google) <rostedt@goodmis.org>
Cc: Thomas Gleixner <tglx@linutronix.de>
Cc: Thomas Huth <thuth@redhat.com>
Cc: Uladzislau Rezki (Sony) <urezki@gmail.com>
Cc: Vlastimil Babka <vbabka@suse.cz>
Cc: Xiongwei Song <xiongwei.song@windriver.com>
Cc: Yu Zhao <yuzhao@google.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
By this point can_vma_merge_right() must have returned true, which implies
can_vma_merge_before() also returned true, which already asserts that the
pgoff is as expected for a merge with the following VMA, thus this
assignment is redundant.
Below is a more detail explanation.
Current definition of can_vma_merge_right() is:
static bool can_vma_merge_right(struct vma_merge_struct *vmg,
bool can_merge_left)
{
if (!vmg->next || vmg->end != vmg->next->vm_start ||
!can_vma_merge_before(vmg))
return false;
...
}
And:
static bool can_vma_merge_before(struct vma_merge_struct *vmg)
{
pgoff_t pglen = PHYS_PFN(vmg->end - vmg->start);
...
if (vmg->next->vm_pgoff == vmg->pgoff + pglen)
return true;
...
}
Which implies vmg->pgoff == vmg->next->vm_pgoff - pglen.
None of these values are changed between the check and prior assignment,
so this was an entirely redundant assignment.
[akpm@linux-foundation.org: remove now-unused local]
[lorenzo.stoakes@oracle.com: rephrase the changelog]
Link: https://lkml.kernel.org/r/20241024093347.18057-1-richard.weiyang@gmail.com
Signed-off-by: Wei Yang <richard.weiyang@gmail.com>
Reviewed-by: Lorenzo Stoakes <lorenzo.stoakes@oracle.com>
Cc: Jann Horn <jannh@google.com>
Cc: Vlastimil Babka <vbabka@suse.cz>
Cc: Liam R. Howlett <Liam.Howlett@oracle.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
The only place where this was used was in mmap_region(), which we have now
adjusted to not require this to be performed (we reset ourselves in
effect).
It also created a dangerous assumption that VMG state could be safely
reused after a merge, at which point it may have been mutated in
unexpected ways, leading to subtle bugs.
Note that it was discovered by Wei Yang that there was also an error in
this code - we are comparing vmg->vma with prev after setting it to NULL.
This however had no impact, as we previously reset VMA iterator state
before attempting merge again, but it was useless effort.
In any case, this patch removes all of the logic so also eliminates this
wasted effort.
Link: https://lkml.kernel.org/r/5d9a59eee6498ae017cc87d89aa723de7179f75d.1729858176.git.lorenzo.stoakes@oracle.com
Signed-off-by: Lorenzo Stoakes <lorenzo.stoakes@oracle.com>
Reviewed-by: Vlastimil Babka <vbabka@suse.cz>
Cc: Jann Horn <jannh@google.com>
Cc: Liam R. Howlett <Liam.Howlett@Oracle.com>
Cc: Linus Torvalds <torvalds@linux-foundation.org>
Cc: Peter Xu <peterx@redhat.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
We have seen bugs and resource leaks arise from the complexity of the
__mmap_region() function. This, and the generally deeply fragile error
handling logic and complexity which makes understanding the function
difficult make it highly desirable to refactor it into something readable.
Achieve this by separating the function into smaller logical parts which
are easier to understand and follow, and which importantly very
significantly simplify the error handling.
Note that we now call vms_abort_munmap_vmas() in more error paths than we
used to, however in cases where no abort need occur, vms->nr_pages will be
equal to zero and we simply exit this function without doing more than we
would have done previously.
Importantly, the invocation of the driver mmap hook via mmap_file() now
has very simple and obvious handling (this was previously the most
problematic part of the mmap() operation).
Use a generalised stack-based 'mmap state' to thread through values and
also retrieve state as needed.
Also avoid ever relying on vma merge (vmg) state after a merge is
attempted, instead maintain meaningful state in the mmap state and
establish vmg state as and when required.
This avoids any subtle bugs arising from merge logic mutating this state
and mmap_region() logic later relying upon it.
Link: https://lkml.kernel.org/r/25bd2edc3275450f448cbfe0756ce2a7cd06810f.1729858176.git.lorenzo.stoakes@oracle.com
Signed-off-by: Lorenzo Stoakes <lorenzo.stoakes@oracle.com>
Reviewed-by: Vlastimil Babka <vbabka@suse.cz>
Cc: Jann Horn <jannh@google.com>
Cc: Liam R. Howlett <Liam.Howlett@Oracle.com>
Cc: Linus Torvalds <torvalds@linux-foundation.org>
Cc: Peter Xu <peterx@redhat.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Patch series "fix error handling in mmap_region() and refactor", v3.
The mmap_region() function is somewhat terrifying, with spaghetti-like
control flow and numerous means by which issues can arise and incomplete
state, memory leaks and other unpleasantness can occur.
This series goes to great lengths to simplify how mmap_region() works and
to avoid unwinding errors late on in the process of setting up the VMA for
the new mapping, and equally avoids such operations occurring while the
VMA is in an inconsistent state.
This series builds on the previously submitted hotfix patches (see link to
v2 below) which addresses the most critical issues around mmap_region(),
and further works to improve mmap_region() complexity, stability, and
testability.
This series moves the code to mm/vma.c to render it userland testable,
refactors and simplifies it into smaller functions that are significantly
more readable.
It additionally avoids performing an attempt at a second merge mid-way
through allocating a new VMA, a dubious proposition at best and one that
is highly subject to subtle bugs.
Rather than do this, we simply note that we ought to retry the merge and
do this as a final step.
This patch (of 3):
Add some additional vma_internal.h stubs in preparation for
__mmap_region() being moved to mm/vma.c. Without these the move would
result in the tests no longer compiling.
Link: https://lkml.kernel.org/r/cover.1729858176.git.lorenzo.stoakes@oracle.com
Link: https://lkml.kernel.org/r/74b27e159e261d2ac1fe66a130edad1d61fdc176.1729858176.git.lorenzo.stoakes@oracle.com
Signed-off-by: Lorenzo Stoakes <lorenzo.stoakes@oracle.com>
Cc: Jann Horn <jannh@google.com>
Cc: Liam R. Howlett <Liam.Howlett@Oracle.com>
Cc: Linus Torvalds <torvalds@linux-foundation.org>
Cc: Peter Xu <peterx@redhat.com>
Cc: Vlastimil Babka <vbabka@suse.cz>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Patch series "mm/damon/vaddr: Fix issue in
damon_va_evenly_split_region()". v2.
According to the logic of damon_va_evenly_split_region(), currently
following split case would not meet the expectation:
Suppose DAMON_MIN_REGION=0x1000,
Case: Split [0x0, 0x3000) into 2 pieces, then the result would be
acutually 3 regions:
[0x0, 0x1000), [0x1000, 0x2000), [0x2000, 0x3000)
but NOT the expected 2 regions:
[0x0, 0x1000), [0x1000, 0x3000) !!!
The root cause is that when calculating size of each split piece in
damon_va_evenly_split_region():
`sz_piece = ALIGN_DOWN(sz_orig / nr_pieces, DAMON_MIN_REGION);`
both the dividing and the ALIGN_DOWN may cause loss of precision, then
each time split one piece of size 'sz_piece' from origin 'start' to 'end'
would cause more pieces are split out than expected!!!
To fix it, count for each piece split and make sure no more than
'nr_pieces'. In addition, add above case into damon_test_split_evenly().
And add 'nr_piece == 1' check in damon_va_evenly_split_region() for better
code readability and add a corresponding kunit testcase.
This patch (of 2):
According to the logic of damon_va_evenly_split_region(), currently
following split case would not meet the expectation:
Suppose DAMON_MIN_REGION=0x1000,
Case: Split [0x0, 0x3000) into 2 pieces, then the result would be
acutually 3 regions:
[0x0, 0x1000), [0x1000, 0x2000), [0x2000, 0x3000)
but NOT the expected 2 regions:
[0x0, 0x1000), [0x1000, 0x3000) !!!
The root cause is that when calculating size of each split piece in
damon_va_evenly_split_region():
`sz_piece = ALIGN_DOWN(sz_orig / nr_pieces, DAMON_MIN_REGION);`
both the dividing and the ALIGN_DOWN may cause loss of precision,
then each time split one piece of size 'sz_piece' from origin 'start' to
'end' would cause more pieces are split out than expected!!!
To fix it, count for each piece split and make sure no more than
'nr_pieces'. In addition, add above case into damon_test_split_evenly().
After this patch, damon-operations test passed:
# ./tools/testing/kunit/kunit.py run damon-operations
[...]
============== damon-operations (6 subtests) ===============
[PASSED] damon_test_three_regions_in_vmas
[PASSED] damon_test_apply_three_regions1
[PASSED] damon_test_apply_three_regions2
[PASSED] damon_test_apply_three_regions3
[PASSED] damon_test_apply_three_regions4
[PASSED] damon_test_split_evenly
================ [PASSED] damon-operations =================
Link: https://lkml.kernel.org/r/20241022083927.3592237-1-zhengyejian@huaweicloud.com
Link: https://lkml.kernel.org/r/20241022083927.3592237-2-zhengyejian@huaweicloud.com
Fixes: 3f49584b26 ("mm/damon: implement primitives for the virtual memory address spaces")
Signed-off-by: Zheng Yejian <zhengyejian@huaweicloud.com>
Reviewed-by: SeongJae Park <sj@kernel.org>
Cc: Fernand Sieber <sieberf@amazon.com>
Cc: Leonard Foerster <foersleo@amazon.de>
Cc: Shakeel Butt <shakeel.butt@linux.dev>
Cc: Ye Weihua <yeweihua4@huawei.com>
Cc: <stable@vger.kernel.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Previously the seq_buf used for accumulating the memory.stat output was
sized at PAGE_SIZE. But the amount of output is invariant to PAGE_SIZE;
If 4K is enough on a 4K page system, then it should also be enough on a
64K page system, so we can save 60K on the static buffer used in
mem_cgroup_print_oom_meminfo(). Let's make it so.
This also has the beneficial side effect of removing a place in the code
that assumed PAGE_SIZE is a compile-time constant. So this helps our
quest towards supporting boot-time page size selection.
Link: https://lkml.kernel.org/r/20241021130027.3615969-1-ryan.roberts@arm.com
Signed-off-by: Ryan Roberts <ryan.roberts@arm.com>
Acked-by: Johannes Weiner <hannes@cmpxchg.org>
Acked-by: Shakeel Butt <shakeel.butt@linux.dev>
Acked-by: Michal Hocko <mhocko@suse.com>
Acked-by: Roman Gushchin <roman.gushchin@linux.dev>
Acked-by: Muchun Song <muchun.song@linux.dev>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
With the strictlimit flag, wb_thresh acts as a hard limit in
balance_dirty_pages() and wb_position_ratio(). When device write
operations are inactive, wb_thresh can drop to 0, causing writes to be
blocked. The issue occasionally occurs in fuse fs, particularly with
network backends, the write thread is blocked frequently during a period.
To address it, this patch raises the minimum wb_thresh to a controllable
level, similar to the non-strictlimit case.
Link: https://lkml.kernel.org/r/20241023100032.62952-1-jimzhao.ai@gmail.com
Signed-off-by: Jim Zhao <jimzhao.ai@gmail.com>
Cc: Matthew Wilcox <willy@infradead.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
tmpfs already supports PMD-sized large folios, but the tmpfs read
operation still performs copying at PAGE_SIZE granularity, which is
unreasonable. This patch changes tmpfs to copy data at folio granularity,
which can improve the read performance, as well as changing to use folio
related functions.
Moreover, if a large folio has a subpage that is hwpoisoned, it will
still fall back to page granularity copying.
Use 'fio bs=64k' to read a 1G tmpfs file populated with 2M THPs, and I can
see about 20% performance improvement, and no regression with bs=4k.
Before the patch:
READ: bw=10.0GiB/s
After the patch:
READ: bw=12.0GiB/s
Link: https://lkml.kernel.org/r/2129a21a5b9f77d3bb7ddec152c009ce7c5653c4.1729218573.git.baolin.wang@linux.alibaba.com
Signed-off-by: Baolin Wang <baolin.wang@linux.alibaba.com>
Reviewed-by: Yang Shi <shy828301@gmail.com>
Cc: David Hildenbrand <david@redhat.com>
Cc: Hugh Dickins <hughd@google.com>
Cc: Kefeng Wang <wangkefeng.wang@huawei.com>
Cc: Matthew Wilcox <willy@infradead.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Patch series "Improve the tmpfs large folio read performance", v2.
tmpfs already supports PMD-sized large folios, but the tmpfs read
operation still performs copying at PAGE_SIZE granularity, which is not
perfect. This patchset changes tmpfs to copy data at the folio
granularity, which can improve the read performance.
Use 'fio bs=64k' to read a 1G tmpfs file populated with 2M THPs, and I can
see about 20% performance improvement, and no regression with bs=4k. I
also did some functional testing with the xfstests suite, and I did not
find any regressions with the following xfstests config:
FSTYP=tmpfs
export TEST_DIR=/mnt/tempfs_mnt
export TEST_DEV=/mnt/tempfs_mnt
export SCRATCH_MNT=/mnt/scratchdir
export SCRATCH_DEV=/mnt/scratchdir
This patch (of 2):
Using iocb->ki_pos to check if the read bytes exceeds the file size and to
calculate the bytes to be read can help simplify the code logic.
Meanwhile, this is also a preparation for improving tmpfs large folios
read performance in the following patch.
Link: https://lkml.kernel.org/r/cover.1729218573.git.baolin.wang@linux.alibaba.com
Link: https://lkml.kernel.org/r/e8863e289577e0dc1e365b5419bf2d1c9a24ae3d.1729218573.git.baolin.wang@linux.alibaba.com
Signed-off-by: Baolin Wang <baolin.wang@linux.alibaba.com>
Reviewed-by: Yang Shi <shy828301@gmail.com>
Cc: David Hildenbrand <david@redhat.com>
Cc: Hugh Dickins <hughd@google.com>
Cc: Kefeng Wang <wangkefeng.wang@huawei.com>
Cc: Matthew Wilcox <willy@infradead.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
The readahead flag is set on a folio based on the lookahead_size and
nr_to_read. For example, when the readahead happens from index to index +
nr_to_read, then the readahead `mark` offset from index is set at
nr_to_read - lookahead_size.
There are some scenarios where the lookahead_size > nr_to_read. For
example, readahead window was created, but the file was truncated before
the readahead starts. do_page_cache_ra() will clamp the nr_to_read if the
readahead window extends beyond EOF after truncation. If this happens,
readahead flag should not be set on any folio on the current readahead
window.
The current calculation for `mark` with mapping_min_order > 0 gives
incorrect results when lookahead_size > nr_to_read due to rounding up
operation:
index = 128
nr_to_read = 16
lookahead_size = 28
mapping_min_order = 4 (16 pages)
ra_folio_index = round_up(128 + 16 - 28, 16) = 128;
mark = 128 - 128 = 0; # offset from index to set RA flag
In the above example, the lookahead_size is actually lying outside the
current readahead window. Without this patch, RA flag will be set
incorrectly on the folio at index 128. This can lead to marking the
readahead flag on the wrong folio, therefore, triggering a readahead when
it is not necessary.
Explicitly initialize `mark` to be ULONG_MAX and only calculate it when
lookahead_size is within the readahead window.
Link: https://lkml.kernel.org/r/20241017062342.478973-1-kernel@pankajraghav.com
Fixes: 26cfdb395e ("readahead: allocate folios with mapping_min_order in readahead")
Signed-off-by: Pankaj Raghav <p.raghav@samsung.com>
Cc: Luis Chamberlain <mcgrof@kernel.org>
Cc: Matthew Wilcox <willy@infradead.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
