Johannes Weiner [Fri, 19 May 2023 12:39:59 +0000 (14:39 +0200)]
mm: compaction: drop redundant watermark check in compaction_zonelist_suitable()
The watermark check in compaction_zonelist_suitable(), called from
should_compact_retry(), is sandwiched between two watermark checks
already: before, there are freelist attempts as part of direct reclaim and
direct compaction; after, there is a last-minute freelist attempt in
__alloc_pages_may_oom().
The check in compaction_zonelist_suitable() isn't necessary. Kill it.
Johannes Weiner [Fri, 19 May 2023 12:39:57 +0000 (14:39 +0200)]
mm: compaction: refactor __compaction_suitable()
__compaction_suitable() is supposed to check for available migration
targets. However, it also checks whether the operation was requested via
/proc/sys/vm/compact_memory, and whether the original allocation request
can already succeed. These don't apply to all callsites.
Move the checks out to the callers, so that later patches can deal with
them one by one. No functional change intended.
Johannes Weiner [Fri, 19 May 2023 12:39:56 +0000 (14:39 +0200)]
mm: compaction: simplify should_compact_retry()
The different branches for retry are unnecessarily complicated. There are
really only three outcomes: progress (retry n times), skipped (retry if
reclaim can help), failed (retry with higher priority).
Rearrange the branches and the retry counter to make it simpler.
The compaction result helpers encode quirks that are specific to the
allocator's retry logic. E.g. COMPACT_SUCCESS and COMPACT_COMPLETE
actually represent failures that should be retried upon, and so on. I
frequently found myself pulling up the helper implementation in order to
understand and work on the retry logic. They're not quite clean
abstractions; rather they split the retry logic into two locations.
Remove the helpers and inline the checks. Then comment on the result
interpretations directly where the decision making happens.
Liam R. Howlett [Thu, 18 May 2023 14:55:43 +0000 (10:55 -0400)]
mm: add vma_iter_{next,prev}_range() to vma iterator
Add functionality to the VMA iterator to advance and retreat one offset
within the maple tree, regardless of the value contained. This can lead
to less re-walking to find an area of interest, especially when there is
nothing in that offset.
Liam R. Howlett [Thu, 18 May 2023 14:55:41 +0000 (10:55 -0400)]
maple_tree: clear up index and last setting in single entry tree
When there is a single entry tree (range of 0-0 pointing to an entry),
then ensure the limit is either 0-0 or 1-oo, depending on where the user
walks. Ensure the correct node setting as well; either MAS_ROOT or
MAS_NONE.
Liam R. Howlett [Thu, 18 May 2023 14:55:40 +0000 (10:55 -0400)]
maple_tree: add mas_prev_range() and mas_find_range_rev interface
Some users of the maple tree may want to move to the previous range
regardless of the value stored there. Add this interface as well as the
'find' variant to support walking to the first value, then iterating over
the previous ranges.
Liam R. Howlett [Thu, 18 May 2023 14:55:39 +0000 (10:55 -0400)]
maple_tree: introduce mas_prev_slot() interface
Sometimes the user needs to revert to the previous slot, regardless of if
it is empty or not. Add an interface to go to the previous slot.
Since there can't be two consecutive NULLs in the tree, the mas_prev()
function can be implemented by calling mas_prev_slot() a maximum of 2
times. Change the underlying interface to use mas_prev_slot() to align
the code.
Liam R. Howlett [Thu, 18 May 2023 14:55:37 +0000 (10:55 -0400)]
maple_tree: add mas_next_range() and mas_find_range() interfaces
Some users of the maple tree may want to move to the next range in the
tree, even if it stores a NULL. This family of function provides that
functionality by advancing one slot at a time and returning the result,
while mas_contiguous() will iterate over the range and stop on
encountering the first NULL.
Liam R. Howlett [Thu, 18 May 2023 14:55:36 +0000 (10:55 -0400)]
maple_tree: introduce mas_next_slot() interface
Sometimes, during a tree walk, the user needs the next slot regardless of
if it is empty or not. Add an interface to get the next slot.
Since there are no consecutive NULLs allowed in the tree, the mas_next()
function can only advance two slots at most. So use the new
mas_next_slot() interface to align both implementations. Use this method
for mas_find() as well.
Liam R. Howlett [Thu, 18 May 2023 14:55:34 +0000 (10:55 -0400)]
maple_tree: revise limit checks in mas_empty_area{_rev}()
Since the maple tree is inclusive in range, ensure that a range of 1 (min
= max) works for searching for a gap in either direction, and make sure
the size is at least 1 but not larger than the delta between min and max.
This commit also updates the testing. Unfortunately there isn't a way to
safely update the tests and code without a test failure.
Liam R. Howlett [Thu, 18 May 2023 14:55:33 +0000 (10:55 -0400)]
maple_tree: try harder to keep active node with mas_prev()
Keep a reference to the node when possible with mas_prev(). This will
avoid re-walking the tree. In keeping a reference to the node, keep the
last/index accurate to the range being referenced. This means the limit
may be within the range, but the range may extend outside of the limit.
Also fix the single entry tree to respect the range (of 0), or set the
node to MAS_NONE in the case of shifting beyond 0.
Liam R. Howlett [Thu, 18 May 2023 14:55:31 +0000 (10:55 -0400)]
mm/mmap: change do_vmi_align_munmap() for maple tree iterator changes
The maple tree iterator clean up is incompatible with the way
do_vmi_align_munmap() expects it to behave. Update the expected behaviour
to map now since the change will work currently.
Liam R. Howlett [Thu, 18 May 2023 14:55:29 +0000 (10:55 -0400)]
maple_tree: remove unnecessary check from mas_destroy()
mas_destroy currently checks if mas->node is MAS_START prior to calling
mas_start(), but this is unnecessary as mas_start() will do nothing if the
node is anything but MAS_START.
Liam R. Howlett [Thu, 18 May 2023 14:55:28 +0000 (10:55 -0400)]
maple_tree: add __init and __exit to test module
The test functions are not needed after the module is removed, so mark
them as such. Add __exit to the module removal function. Some other
variables have been marked as const static as well.
Liam R. Howlett [Thu, 18 May 2023 14:55:27 +0000 (10:55 -0400)]
mm: update vma_iter_store() to use MAS_WARN_ON()
MAS_WARN_ON() will provide more information on the maple state and can be
more useful for debugging. Use this version of WARN_ON() in the debugging
code when storing to the tree.
Update the printk to a pr_warn(), but this will only be printed when maple
tree debug is enabled anyways.
Making all print statements into one will keep them together on a busy
terminal.
Liam R. Howlett [Thu, 18 May 2023 14:55:25 +0000 (10:55 -0400)]
maple_tree: make test code work without debug enabled
The test code is less useful without debug, but can still do general
validations. Define mt_dump(), mas_dump() and mas_wr_dump() as a noop if
debug is not enabled and document it in the test module information that
more information can be obtained with another kernel config option.
MT_BUG_ON() will report a failures without tree dumps, and the output will
be less useful.
Liam R. Howlett [Thu, 18 May 2023 14:55:24 +0000 (10:55 -0400)]
maple_tree: return error on mte_pivots() out of range
Rename mte_pivots() to mas_pivots() and pass through the ma_state to set
the error code to -EIO when the offset is out of range for the node type.
Change the WARN_ON() to MAS_WARN_ON() to log the maple state.
Liam R. Howlett [Thu, 18 May 2023 14:55:22 +0000 (10:55 -0400)]
maple_tree: use MAS_WR_BUG_ON() in mas_store_prealloc()
mas_store_prealloc() should never fail, but if it does due to internal
tree issues then get as much debug information as possible prior to
crashing the kernel.
Liam R. Howlett [Thu, 18 May 2023 14:55:20 +0000 (10:55 -0400)]
maple_tree: use MAS_BUG_ON() in mas_set_height()
Use MAS_BUG_ON() instead of MT_BUG_ON() to get the maple state
information. In the unlikely event of a tree height of > 31, try to
increase the probability of useful information being logged.
Liam R. Howlett [Thu, 18 May 2023 14:55:18 +0000 (10:55 -0400)]
maple_tree: convert debug code to use MT_WARN_ON() and MAS_WARN_ON()
Using MT_WARN_ON() allows for the removal of if statements before logging.
Using MAS_WARN_ON() will provide more information when issues are
encountered.
Liam R. Howlett [Thu, 18 May 2023 14:55:13 +0000 (10:55 -0400)]
maple_tree: clean up mas_dfs_postorder()
Convert loop type to ensure all variables are set to make the compiler
happy, and use the mas_is_none() function instead of explicitly checking
the node in the maple state.
Liam R. Howlett [Thu, 18 May 2023 14:55:12 +0000 (10:55 -0400)]
maple_tree: avoid unnecessary ascending
The maple tree node limits are implied by the parent. When walking up the
tree, the limit may not be known until a slot that does not have implied
limits are encountered. However, if the node is the left-most or
right-most node, the walking up to find that limit can be skipped.
This commit also fixes the debug/testing code that was not setting the
limit on walking down the tree as that optimization is not compatible with
this change.
Liam R. Howlett [Thu, 18 May 2023 14:55:11 +0000 (10:55 -0400)]
maple_tree: clean up mas_parent_enum() and rename to mas_parent_type()
mas_parent_enum() is a simple wrapper for mte_parent_enum() which is only
called from that wrapper. Remove the wrapper and inline mte_parent_enum()
into mas_parent_enum().
At the same time, clean up the bit masking of the root pointer since it
cannot be set by the time the bit masking occurs. Change the check on the
root bit to a WARN_ON(), and fix the verification code to not trigger the
WARN_ON() before checking if the node is root.
Align the name to mas_parent_type() since mas_node_type() exists already.
Liam R. Howlett [Thu, 18 May 2023 14:55:10 +0000 (10:55 -0400)]
maple_tree: fix static analyser cppcheck issue
Patch series "Maple tree mas_{next,prev}_range() and cleanup", v4.
This patchset contains a number of clean ups to the code to make it more
usable (next/prev range), the addition of debug output formatting, the
addition of printing the maple state information in the WARN_ON/BUG_ON
code.
There is also work done here to keep nodes active during iterations to
reduce the necessity of re-walking the tree.
Finally, there is a new interface added to move to the next or previous
range in the tree, even if it is empty.
The organisation of the patches is as follows:
0001-0004 - Small clean ups
0005-0018 - Additional debug options and WARN_ON/BUG_ON changes
0019 - Test module __init and __exit addition
0020-0021 - More functional clean ups
0022-0026 - Changes to keep nodes active
0027-0034 - Add new mas_{prev,next}_range()
0035 - Use new mas_{prev,next}_range() in mmap_region()
This patch (of 35):
Static analyser of the maple tree code noticed that the split variable is
being used to dereference into an array prior to checking the variable
itself. Fix this issue by changing the order of the statement to check
the variable first.
Almost all of the callers & implementors of migrate_pages() were already
converted to use folios. compaction_alloc() & compaction_free() are
trivial to convert a part of this patch and not worth splitting out.
Lorenzo Stoakes [Wed, 17 May 2023 19:25:48 +0000 (20:25 +0100)]
mm/gup: remove vmas array from internal GUP functions
Now we have eliminated all callers to GUP APIs which use the vmas
parameter, eliminate it altogether.
This eliminates a class of bugs where vmas might have been kept around
longer than the mmap_lock and thus we need not be concerned about locks
being dropped during this operation leaving behind dangling pointers.
This simplifies the GUP API and makes it considerably clearer as to its
purpose - follow flags are applied and if pinning, an array of pages is
returned.
Lorenzo Stoakes [Wed, 17 May 2023 19:25:45 +0000 (20:25 +0100)]
mm/gup: remove vmas parameter from pin_user_pages()
We are now in a position where no caller of pin_user_pages() requires the
vmas parameter at all, so eliminate this parameter from the function and
all callers.
This clears the way to removing the vmas parameter from GUP altogether.
Lorenzo Stoakes [Wed, 17 May 2023 19:25:42 +0000 (20:25 +0100)]
io_uring: rsrc: delegate VMA file-backed check to GUP
Now that the GUP explicitly checks FOLL_LONGTERM pin_user_pages() for
broken file-backed mappings in "mm/gup: disallow FOLL_LONGTERM GUP-nonfast
writing to file-backed mappings", there is no need to explicitly check VMAs
for this condition, so simply remove this logic from io_uring altogether.
Lorenzo Stoakes [Wed, 17 May 2023 19:25:39 +0000 (20:25 +0100)]
mm/gup: remove vmas parameter from get_user_pages_remote()
The only instances of get_user_pages_remote() invocations which used the
vmas parameter were for a single page which can instead simply look up the
VMA directly. In particular:-
- __update_ref_ctr() looked up the VMA but did nothing with it so we simply
remove it.
- __access_remote_vm() was already using vma_lookup() when the original
lookup failed so by doing the lookup directly this also de-duplicates the
code.
We are able to perform these VMA operations as we already hold the
mmap_lock in order to be able to call get_user_pages_remote().
As part of this work we add get_user_page_vma_remote() which abstracts the
VMA lookup, error handling and decrementing the page reference count should
the VMA lookup fail.
This forms part of a broader set of patches intended to eliminate the vmas
parameter altogether.
Lorenzo Stoakes [Wed, 17 May 2023 19:25:36 +0000 (20:25 +0100)]
mm/gup: remove unused vmas parameter from pin_user_pages_remote()
No invocation of pin_user_pages_remote() uses the vmas parameter, so
remove it. This forms part of a larger patch set eliminating the use of
the vmas parameters altogether.
Lorenzo Stoakes [Wed, 17 May 2023 19:25:33 +0000 (20:25 +0100)]
mm/gup: remove unused vmas parameter from get_user_pages()
Patch series "remove the vmas parameter from GUP APIs", v6.
(pin_/get)_user_pages[_remote]() each provide an optional output parameter
for an array of VMA objects associated with each page in the input range.
These provide the means for VMAs to be returned, as long as mm->mmap_lock
is never released during the GUP operation (i.e. the internal flag
FOLL_UNLOCKABLE is not specified).
In addition, these VMAs can only be accessed with the mmap_lock held and
become invalidated the moment it is released.
The vast majority of invocations do not use this functionality and of
those that do, all but one case retrieve a single VMA to perform checks
upon.
It is not egregious in the single VMA cases to simply replace the
operation with a vma_lookup(). In these cases we duplicate the (fast)
lookup on a slow path already under the mmap_lock, abstracted to a new
get_user_page_vma_remote() inline helper function which also performs
error checking and reference count maintenance.
The special case is io_uring, where io_pin_pages() specifically needs to
assert that the VMAs underlying the range do not result in broken
long-term GUP file-backed mappings.
As GUP now internally asserts that FOLL_LONGTERM mappings are not
file-backed in a broken fashion (i.e. requiring dirty tracking) - as
implemented in "mm/gup: disallow FOLL_LONGTERM GUP-nonfast writing to
file-backed mappings" - this logic is no longer required and so we can
simply remove it altogether from io_uring.
Eliminating the vmas parameter eliminates an entire class of danging
pointer errors that might have occured should the lock have been
incorrectly released.
In addition, the API is simplified and now clearly expresses what it is
intended for - applying the specified GUP flags and (if pinning) returning
pinned pages.
This change additionally opens the door to further potential improvements
in GUP and the possible marrying of disparate code paths.
I have run this series against gup_test with no issues.
Thanks to Matthew Wilcox for suggesting this refactoring!
This patch (of 6):
No invocation of get_user_pages() use the vmas parameter, so remove it.
The GUP API is confusing and caveated. Recent changes have done much to
improve that, however there is more we can do. Exporting vmas is a prime
target as the caller has to be extremely careful to preclude their use
after the mmap_lock has expired or otherwise be left with dangling
pointers.
Removing the vmas parameter focuses the GUP functions upon their primary
purpose - pinning (and outputting) pages as well as performing the actions
implied by the input flags.
This is part of a patch series aiming to remove the vmas parameter
altogether.
Kefeng Wang [Tue, 16 May 2023 06:38:20 +0000 (14:38 +0800)]
mm: page_alloc: move sysctls into it own fils
This moves all page alloc related sysctls to its own file, as part of the
kernel/sysctl.c spring cleaning, also move some functions declarations
from mm.h into internal.h.
Kefeng Wang [Tue, 16 May 2023 06:38:18 +0000 (14:38 +0800)]
mm: page_alloc: move pm_* function into power
pm_restrict_gfp_mask()/pm_restore_gfp_mask() only used in power, let's
move them out of page_alloc.c.
Adding a general gfp_has_io_fs() function which return true if gfp with
both __GFP_IO and __GFP_FS flags, then use it inside of
pm_suspended_storage(), also the pm_suspended_storage() is moved into
suspend.h.
DEFINE_DYNAMIC_DEBUG_METADATA and DYNAMIC_DEBUG_BRANCH already has stub
definitions without dynamic debug feature, remove unnecessary
alloc_contig_dump_pages() stub.
Kefeng Wang [Tue, 16 May 2023 06:38:12 +0000 (14:38 +0800)]
mm: page_alloc: collect mem statistic into show_mem.c
Let's move show_mem.c from lib to mm, as it belongs memory subsystem, also
split some memory statistic related functions from page_alloc.c to
show_mem.c, and we cleanup some unneeded include.
Kefeng Wang [Tue, 16 May 2023 06:38:11 +0000 (14:38 +0800)]
mm: page_alloc: move set_zone_contiguous() into mm_init.c
set_zone_contiguous() is only used in mm init/hotplug, and
clear_zone_contiguous() only used in hotplug, move them from page_alloc.c
to the more appropriate file.
Kefeng Wang [Tue, 16 May 2023 06:38:10 +0000 (14:38 +0800)]
mm: page_alloc: move init_on_alloc/free() into mm_init.c
Since commit f2fc4b44ec2b ("mm: move init_mem_debugging_and_hardening() to
mm/mm_init.c"), the init_on_alloc() and init_on_free() define is better to
move there too.
Kefeng Wang [Tue, 16 May 2023 06:38:09 +0000 (14:38 +0800)]
mm: page_alloc: move mirrored_kernelcore into mm_init.c
Patch series "mm: page_alloc: misc cleanup and refactor", v2.
This aims to reduce more space in page_alloc.c, also do some cleanup, no
functional changes intended.
This patch (of 13):
Since commit 9420f89db2dd ("mm: move most of core MM initialization to
mm/mm_init.c"), mirrored_kernelcore should be moved into mm_init.c, as
most related codes are already there.
Following the discussion about direct map fragmentaion at LSF/MM [1], it
appears that direct map fragmentation has a negligible effect on kernel
data accesses. Since the only reason that warranted secretmem to be
disabled by default was concern about performance regression caused by the
direct map fragmentation, it makes perfect sense to lift this restriction
and make secretmem enabled.
secretmem obeys RLIMIT_MEMBLOCK and as such it is not expected to cause
large fragmentation of the direct map or meaningfull increase in page
tables allocated during split of the large mappings in the direct map.
The secretmem.enable parameter is retained to allow system administrators
to disable secretmem at boot.
Switch the default setting of secretmem.enable parameter to 1.
Mel Gorman [Mon, 15 May 2023 11:33:44 +0000 (12:33 +0100)]
Revert "Revert "mm/compaction: fix set skip in fast_find_migrateblock""
This reverts commit 95e7a450b819 ("Revert "mm/compaction: fix set skip in
fast_find_migrateblock"").
Commit 7efc3b726103 ("mm/compaction: fix set skip in
fast_find_migrateblock") was reverted due to bug reports about khugepaged
consuming large amounts of CPU without making progress. The underlying
bug was partially fixed by commit cfccd2e63e7e ("mm, compaction: finish
pageblocks on complete migration failure") but it only mitigated the
problem and Vlastimil Babka pointing out the same issue could
theoretically happen to kcompactd.
As pageblocks containing pages that fail to migrate should now be forcibly
rescanned to set the skip hint if skip hints are used,
fast_find_migrateblock() should no longer loop on a small subset of
pageblocks for prolonged periods of time. Revert the revert so
fast_find_migrateblock() is effective again.
Using the mmtests config workload-usemem-stress-numa-compact, the number
of unique ranges scanned was analysed for both kcompactd and !kcompactd
activity.
Note that the vanilla kernel and the full series had some duplication of
ranges scanned but it was not severe and would be in line with compaction
resets when the skip hints are cleared. Just a revert of commit 7efc3b726103 ("mm/compaction: fix set skip in fast_find_migrateblock")
showed excessive rescans of the same ranges so the series should not
reintroduce bug 1206848.
Mel Gorman [Mon, 15 May 2023 11:33:43 +0000 (12:33 +0100)]
mm: compaction: update pageblock skip when first migration candidate is not at the start
isolate_migratepages_block should mark a pageblock as skip if scanning
started on an aligned pageblock boundary but it only updates the skip flag
if the first migration candidate is also aligned. Tracing during a
compaction stress load (mmtests: workload-usemem-stress-numa-compact) that
many pageblocks are not marked skip causing excessive scanning of blocks
that had been recently checked. Update pageblock skip based on
"valid_page" which is set if scanning started on a pageblock boundary.
Mel Gorman [Mon, 15 May 2023 11:33:42 +0000 (12:33 +0100)]
mm: compaction: only force pageblock scan completion when skip hints are obeyed
fast_find_migrateblock relies on skip hints to avoid rescanning a recently
selected pageblock but compact_zone() only forces the pageblock scan
completion to set the skip hint if in direct compaction. While this
prevents direct compaction repeatedly scanning a subset of blocks due to
fast_find_migrateblock(), it does not prevent proactive compaction, node
compaction and kcompactd encountering the same problem described in commit cfccd2e63e7e ("mm, compaction: finish pageblocks on complete migration
failure").
Force the scan completion of a pageblock to set the skip hint if skip
hints are obeyed to prevent fast_find_migrateblock() repeatedly selecting
a subset of pageblocks.
Mel Gorman [Mon, 15 May 2023 11:33:41 +0000 (12:33 +0100)]
mm: compaction: ensure rescanning only happens on partially scanned pageblocks
Patch series "Follow-up "Fix excessive CPU usage during compaction"".
The series "Fix excessive CPU usage during compaction" [1] attempted to
fix a bug [2] but Vlastimil noted that the fix was incomplete [3]. While
the series was merged, fast_find_migrateblock was still disabled. This
series should fix the corner cases and allow 95e7a450b819 ("Revert
"mm/compaction: fix set skip in fast_find_migrateblock"") to be safely
reverted. Details on how many pageblocks are rescanned are in the
changelog of the last patch.
"Raghavendra K T" tested this and reported "decent improvement from perf
perspective as well as compaction related data [4]
This patch (of 4):
compact_zone() intends to rescan pageblocks if there is a failure to
migrate "within the current order-aligned block". However, the pageblock
scan may already be complete and moved to the next block causing the next
pageblock to be "rescanned". Ensure only the most recent pageblock is
rescanned.
Yuanchu Xie [Mon, 15 May 2023 17:26:08 +0000 (01:26 +0800)]
mm: pagemap: restrict pagewalk to the requested range
The pagewalk in pagemap_read reads one PTE past the end of the requested
range, and stops when the buffer runs out of space. While it produces
the right result, the extra read is unnecessary and less performant.
I timed the following command before and after this patch:
dd count=100000 if=/proc/self/pagemap of=/dev/null
The results are consistently within 0.001s across 5 runs.
Before:
100000+0 records in
100000+0 records out 51200000 bytes (51 MB) copied, 0.0763159 s, 671 MB/s
real 0m0.078s
user 0m0.012s
sys 0m0.065s
After:
100000+0 records in
100000+0 records out 51200000 bytes (51 MB) copied, 0.0487928 s, 1.0 GB/s
Haifeng Xu [Mon, 8 May 2023 07:35:38 +0000 (07:35 +0000)]
mm, oom: do not check 0 mask in out_of_memory()
Since commit 60e2793d440a ("mm, oom: do not trigger out_of_memory from the
#PF"), no user sets gfp_mask to 0. Remove the 0 mask check and update the
comments.
Pasha Tatashin [Mon, 8 May 2023 23:40:59 +0000 (23:40 +0000)]
mm: hugetlb_vmemmap: provide stronger vmemmap allocation guarantees
HugeTLB pages have a struct page optimizations where struct pages for tail
pages are freed. However, when HugeTLB pages are destroyed, the memory
for struct pages (vmemmap) need to be allocated again.
Currently, __GFP_NORETRY flag is used to allocate the memory for vmemmap,
but given that this flag makes very little effort to actually reclaim
memory the returning of huge pages back to the system can be problem.
Lets use __GFP_RETRY_MAYFAIL instead. This flag is also performs graceful
reclaim without causing ooms, but at least it may perform a few retries,
and will fail only when there is genuinely little amount of unused memory
in the system.
Freeing a 1G page requires 16M of free memory. A machine might need to be
reconfigured from one task to another, and release a large number of 1G
pages back to the system if allocating 16M fails, the release won't work.
Arnd Bergmann [Tue, 9 May 2023 14:57:21 +0000 (16:57 +0200)]
kasan: use internal prototypes matching gcc-13 builtins
gcc-13 warns about function definitions for builtin interfaces that have a
different prototype, e.g.:
In file included from kasan_test.c:31:
kasan.h:574:6: error: conflicting types for built-in function '__asan_register_globals'; expected 'void(void *, long int)' [-Werror=builtin-declaration-mismatch]
574 | void __asan_register_globals(struct kasan_global *globals, size_t size);
kasan.h:577:6: error: conflicting types for built-in function '__asan_alloca_poison'; expected 'void(void *, long int)' [-Werror=builtin-declaration-mismatch]
577 | void __asan_alloca_poison(unsigned long addr, size_t size);
kasan.h:580:6: error: conflicting types for built-in function '__asan_load1'; expected 'void(void *)' [-Werror=builtin-declaration-mismatch]
580 | void __asan_load1(unsigned long addr);
kasan.h:581:6: error: conflicting types for built-in function '__asan_store1'; expected 'void(void *)' [-Werror=builtin-declaration-mismatch]
581 | void __asan_store1(unsigned long addr);
kasan.h:643:6: error: conflicting types for built-in function '__hwasan_tag_memory'; expected 'void(void *, unsigned char, long int)' [-Werror=builtin-declaration-mismatch]
643 | void __hwasan_tag_memory(unsigned long addr, u8 tag, unsigned long size);
The two problems are:
- Addresses are passes as 'unsigned long' in the kernel, but gcc-13
expects a 'void *'.
- sizes meant to use a signed ssize_t rather than size_t.
Change all the prototypes to match these. Using 'void *' consistently for
addresses gets rid of a couple of type casts, so push that down to the
leaf functions where possible.
This now passes all randconfig builds on arm, arm64 and x86, but I have
not tested it on the other architectures that support kasan, since they
tend to fail randconfig builds in other ways. This might fail if any of
the 32-bit architectures expect a 'long' instead of 'int' for the size
argument.
The __asan_allocas_unpoison() function prototype is somewhat weird, since
it uses a pointer for 'stack_top' and an size_t for 'stack_bottom'. This
looks like it is meant to be 'addr' and 'size' like the others, but the
implementation clearly treats them as 'top' and 'bottom'.
Arnd Bergmann [Tue, 9 May 2023 14:57:20 +0000 (16:57 +0200)]
kasan: add kasan_tag_mismatch prototype
The kasan sw-tags implementation contains one function that is only called
from assembler and has no prototype in a header. This causes a W=1
warning:
mm/kasan/sw_tags.c:171:6: warning: no previous prototype for 'kasan_tag_mismatch' [-Wmissing-prototypes]
171 | void kasan_tag_mismatch(unsigned long addr, unsigned long access_info,
Add a prototype in the local header to get a clean build.
Huang Ying [Wed, 10 May 2023 03:18:29 +0000 (11:18 +0800)]
migrate_pages_batch: simplify retrying and failure counting of large folios
After recent changes to the retrying and failure counting in
migrate_pages_batch(), it was found that it's unnecessary to count
retrying and failure for normal, large, and THP folios separately.
Because we don't use retrying and failure number of large folios directly.
So, in this patch, we simplified retrying and failure counting of large
folios via counting retrying and failure of normal and large folios
together. This results in the reduced line number.
Previously, in migrate_pages_batch we need to track whether the source
folio is large/THP before splitting. So is_large is used to cache
folio_test_large() result. Now, we don't need that variable any more
because we don't count retrying and failure of large folios (only counting
that of THP folios). So, in this patch, is_large is removed to simplify
the code.
This is just code cleanup, no functionality changes are expected.
Rick Wertenbroek [Wed, 10 May 2023 09:07:57 +0000 (11:07 +0200)]
mm: memory_hotplug: fix format string in warnings
The format string in __add_pages and __remove_pages has a typo and prints
e.g., "Misaligned __add_pages start: 0xfc605 end: #fc609" instead of
"Misaligned __add_pages start: 0xfc605 end: 0xfc609" Fix "#%lx" => "%#lx"
Peng Zhang [Sat, 6 May 2023 02:47:52 +0000 (10:47 +0800)]
maple_tree: fix potential out-of-bounds access in mas_wr_end_piv()
Check the write offset end bounds before using it as the offset into the
pivot array. This avoids a possible out-of-bounds access on the pivot
array if the write extends to the last slot in the node, in which case the
node maximum should be used as the end pivot.
akpm: this doesn't affect any current callers, but new users of mapletree
may encounter this problem if backported into earlier kernels, so let's
fix it in -stable kernels in case of this.
Lorenzo Stoakes [Sat, 6 May 2023 14:05:25 +0000 (15:05 +0100)]
mm/gup: add missing gup_must_unshare() check to gup_huge_pgd()
All other instances of gup_huge_pXd() perform the unshare check, so update
the PGD-specific function to do so as well.
While checking pgd_write() might seem unusual, this function already
performs such a check via pgd_access_permitted() so this is in line with
the existing implementation.
David said:
: This change makes unshare handling across all GUP-fast variants
: consistent, which is desirable as GUP-fast is complicated enough
: already even when consistent.
:
: This function was the only one I seemed to have missed (or left out and
: forgot why -- maybe because it's really dead code for now). The COW
: selftest would identify the problem, so far there was no report.
: Either the selftest wasn't run on corresponding architectures with that
: hugetlb size, or that code is still dead code and unused by
: architectures.
:
: the original commit(s) that added unsharing explain why we care about
: these checks:
:
: a7f226604170acd6 ("mm/gup: trigger FAULT_FLAG_UNSHARE when R/O-pinning a possibly shared anonymous page")
: 84209e87c6963f92 ("mm/gup: reliable R/O long-term pinning in COW mappings")
Nhat Pham [Wed, 3 May 2023 01:36:08 +0000 (18:36 -0700)]
selftests: add selftests for cachestat
Test cachestat on a newly created file, /dev/ files, /proc/ files and a
directory. Also test on a shmem file (which can also be tested with
huge pages since tmpfs supports huge pages).
Nhat Pham [Wed, 3 May 2023 01:36:07 +0000 (18:36 -0700)]
cachestat: implement cachestat syscall
There is currently no good way to query the page cache state of large file
sets and directory trees. There is mincore(), but it scales poorly: the
kernel writes out a lot of bitmap data that userspace has to aggregate,
when the user really doesn not care about per-page information in that
case. The user also needs to mmap and unmap each file as it goes along,
which can be quite slow as well.
Some use cases where this information could come in handy:
* Allowing database to decide whether to perform an index scan or
direct table queries based on the in-memory cache state of the
index.
* Visibility into the writeback algorithm, for performance issues
diagnostic.
* Workload-aware writeback pacing: estimating IO fulfilled by page
cache (and IO to be done) within a range of a file, allowing for
more frequent syncing when and where there is IO capacity, and
batching when there is not.
* Computing memory usage of large files/directory trees, analogous to
the du tool for disk usage.
More information about these use cases could be found in the following
thread:
This patch implements a new syscall that queries cache state of a file and
summarizes the number of cached pages, number of dirty pages, number of
pages marked for writeback, number of (recently) evicted pages, etc. in a
given range. Currently, the syscall is only wired in for x86
architecture.
NAME
cachestat - query the page cache statistics of a file.
int cachestat(unsigned int fd, struct cachestat_range *cstat_range,
struct cachestat *cstat, unsigned int flags);
DESCRIPTION
cachestat() queries the number of cached pages, number of dirty
pages, number of pages marked for writeback, number of evicted
pages, number of recently evicted pages, in the bytes range given by
`off` and `len`.
An evicted page is a page that is previously in the page cache but
has been evicted since. A page is recently evicted if its last
eviction was recent enough that its reentry to the cache would
indicate that it is actively being used by the system, and that
there is memory pressure on the system.
These values are returned in a cachestat struct, whose address is
given by the `cstat` argument.
The `off` and `len` arguments must be non-negative integers. If
`len` > 0, the queried range is [`off`, `off` + `len`]. If `len` ==
0, we will query in the range from `off` to the end of the file.
The `flags` argument is unused for now, but is included for future
extensibility. User should pass 0 (i.e no flag specified).
Currently, hugetlbfs is not supported.
Because the status of a page can change after cachestat() checks it
but before it returns to the application, the returned values may
contain stale information.
RETURN VALUE
On success, cachestat returns 0. On error, -1 is returned, and errno
is set to indicate the error.
ERRORS
EFAULT cstat or cstat_args points to an invalid address.
Nhat Pham [Wed, 3 May 2023 01:36:06 +0000 (18:36 -0700)]
workingset: refactor LRU refault to expose refault recency check
Patch series "cachestat: a new syscall for page cache state of files",
v13.
There is currently no good way to query the page cache statistics of large
files and directory trees. There is mincore(), but it scales poorly: the
kernel writes out a lot of bitmap data that userspace has to aggregate,
when the user really does not care about per-page information in that
case. The user also needs to mmap and unmap each file as it goes along,
which can be quite slow as well.
Some use cases where this information could come in handy:
* Allowing database to decide whether to perform an index scan or direct
table queries based on the in-memory cache state of the index.
* Visibility into the writeback algorithm, for performance issues
diagnostic.
* Workload-aware writeback pacing: estimating IO fulfilled by page cache
(and IO to be done) within a range of a file, allowing for more
frequent syncing when and where there is IO capacity, and batching
when there is not.
* Computing memory usage of large files/directory trees, analogous to
the du tool for disk usage.
More information about these use cases could be found in this thread:
https://lore.kernel.org/lkml/20230315170934[email protected]/
This series of patches introduces a new system call, cachestat, that
summarizes the page cache statistics (number of cached pages, dirty pages,
pages marked for writeback, evicted pages etc.) of a file, in a specified
range of bytes. It also include a selftest suite that tests some typical
usage. Currently, the syscall is only wired in for x86 architecture.
This interface is inspired by past discussion and concerns with fincore,
which has a similar design (and as a result, issues) as mincore. Relevant
links:
I have also developed a small tool that computes the memory usage of files
and directories, analogous to the du utility. User can choose between
mincore or cachestat (with cachestat exporting more information than
mincore). To compare the performance of these two options, I benchmarked
the tool on the root directory of a Meta's server machine, each for five
runs:
Using cachestat
real -- Median: 33.377s, Average: 33.475s, Standard Deviation: 0.3602
user -- Median: 4.08s, Average: 4.1078s, Standard Deviation: 0.0742
sys -- Median: 28.823s, Average: 28.8866s, Standard Deviation: 0.2689
Using mincore:
real -- Median: 102.352s, Average: 102.3442s, Standard Deviation: 0.2059
user -- Median: 10.149s, Average: 10.1482s, Standard Deviation: 0.0162
sys -- Median: 91.186s, Average: 91.2084s, Standard Deviation: 0.2046
I also ran both syscalls on a 2TB sparse file:
Using cachestat:
real 0m0.009s
user 0m0.000s
sys 0m0.009s
Using mincore:
real 0m37.510s
user 0m2.934s
sys 0m34.558s
Very large files like this are the pathological case for mincore. In
fact, to compute the stats for a single 2TB file, mincore takes as long as
cachestat takes to compute the stats for the entire tree! This could
easily happen inadvertently when we run it on subdirectories. Mincore is
clearly not suitable for a general-purpose command line tool.
Regarding security concerns, cachestat() should not pose any additional
issues. The caller already has read permission to the file itself (since
they need an fd to that file to call cachestat). This means that the
caller can access the underlying data in its entirety, which is a much
greater source of information (and as a result, a much greater security
risk) than the cache status itself.
The latest API change (in v13 of the patch series) is suggested by Jens
Axboe. It allows for 64-bit length argument, even on 32-bit architecture
(which is previously not possible due to the limit on the number of
syscall arguments). Furthermore, it eliminates the need for compatibility
handling - every user can use the same ABI.
This patch (of 4):
In preparation for computing recently evicted pages in cachestat, refactor
workingset_refault and lru_gen_refault to expose a helper function that
would test if an evicted page is recently evicted.
memcg, oom: remove explicit wakeup in mem_cgroup_oom_synchronize()
Before commit 29ef680ae7c2 ("memcg, oom: move out_of_memory back to the
charge path"), all memcg oom killers were delayed to page fault path. And
the explicit wakeup is used in this case:
thread A:
...
if (locked) { // complete oom-kill, hold the lock
mem_cgroup_oom_unlock(memcg);
...
}
...
thread B:
...
if (locked && !memcg->oom_kill_disable) {
...
} else {
schedule(); // can't acquire the lock
...
}
...
The reason is that thread A kicks off the OOM-killer, which leads to
wakeups from the uncharges of the exiting task. But thread B is not
guaranteed to see them if it enters the OOM path after the OOM kills but
before thread A releases the lock.
Now only oom_kill_disable case is handled from the #PF path. In that case
it is userspace to trigger the wake up not the #PF path itself. All
potential paths to free some charges are responsible to call
memcg_oom_recover() , so the explicit wakeup is not needed in the
mem_cgroup_oom_synchronize() path which doesn't release any memory itself.
memcg, oom: remove unnecessary check in mem_cgroup_oom_synchronize()
mem_cgroup_oom_synchronize() is only used when the memcg oom handling is
handed over to the edge of the #PF path. Since commit 29ef680ae7c2
("memcg, oom: move out_of_memory back to the charge path") this is the
case only when the kernel memcg oom killer is disabled
(current->memcg_in_oom is only set if memcg->oom_kill_disable). Therefore
a check for oom_kill_disable in mem_cgroup_oom_synchronize() is not
required.
Yosry Ahmed [Fri, 21 Apr 2023 17:40:18 +0000 (17:40 +0000)]
memcg: calculate root usage from global state
Currently, we approximate the root usage by adding the memcg stats for
anon, file, and conditionally swap (for memsw). To read the memcg stats
we need to invoke an rstat flush. rstat flushes can be expensive, they
scale with the number of cpus and cgroups on the system.
mem_cgroup_usage() is called by memcg_events()->mem_cgroup_threshold()
with irqs disabled, so such an expensive operation with irqs disabled can
cause problems.
Instead, approximate the root usage from global state. This is not 100%
accurate, but the root usage has always been ill-defined anyway.
Yosry Ahmed [Fri, 21 Apr 2023 17:40:17 +0000 (17:40 +0000)]
memcg: flush stats non-atomically in mem_cgroup_wb_stats()
The previous patch moved the wb_over_bg_thresh()->mem_cgroup_wb_stats()
code path in wb_writeback() outside the lock section. We no longer need
to flush the stats atomically. Flush the stats non-atomically.
Patch series "cgroup: eliminate atomic rstat flushing", v5.
A previous patch series [1] changed most atomic rstat flushing contexts to
become non-atomic. This was done to avoid an expensive operation that
scales with # cgroups and # cpus to happen with irqs disabled and
scheduling not permitted. There were two remaining atomic flushing
contexts after that series. This series tries to eliminate them as well,
eliminating atomic rstat flushing completely.
The two remaining atomic flushing contexts are:
(a) wb_over_bg_thresh()->mem_cgroup_wb_stats()
(b) mem_cgroup_threshold()->mem_cgroup_usage()
For (a), flushing needs to be atomic as wb_writeback() calls
wb_over_bg_thresh() with a spinlock held. However, it seems like the call
to wb_over_bg_thresh() doesn't need to be protected by that spinlock, so
this series proposes a refactoring that moves the call outside the lock
criticial section and makes the stats flushing in mem_cgroup_wb_stats()
non-atomic.
For (b), flushing needs to be atomic as mem_cgroup_threshold() is called
with irqs disabled. We only flush the stats when calculating the root
usage, as it is approximated as the sum of some memcg stats (file, anon,
and optionally swap) instead of the conventional page counter. This
series proposes changing this calculation to use the global stats instead,
eliminating the need for a memcg stat flush.
After these 2 contexts are eliminated, we no longer need
mem_cgroup_flush_stats_atomic() or cgroup_rstat_flush_atomic(). We can
remove them and simplify the code.
wb_over_bg_thresh() calls mem_cgroup_wb_stats() which invokes an rstat
flush, which can be expensive on large systems. Currently,
wb_writeback() calls wb_over_bg_thresh() within a lock section, so we
have to do the rstat flush atomically. On systems with a lot of
cpus and/or cgroups, this can cause us to disable irqs for a long time,
potentially causing problems.
Move the call to wb_over_bg_thresh() outside the lock section in
preparation to make the rstat flush in mem_cgroup_wb_stats() non-atomic.
The list_empty(&wb->work_list) check should be okay outside the lock
section of wb->list_lock as it is protected by a separate lock
(wb->work_lock), and wb_over_bg_thresh() doesn't seem like it is
modifying any of wb->b_* lists the wb->list_lock is protecting.
Also, the loop seems to be already releasing and reacquring the
lock, so this refactoring looks safe.
Baolin Wang [Mon, 24 Apr 2023 13:45:39 +0000 (21:45 +0800)]
mm/page_alloc: drop the unnecessary pfn_valid() for start pfn
__pageblock_pfn_to_page() currently performs both pfn_valid check and
pfn_to_online_page(). The former one is redundant because the latter is a
stronger check. Drop pfn_valid().
Wen Yang [Tue, 25 Apr 2023 15:52:35 +0000 (23:52 +0800)]
mm: compaction: optimize compact_memory to comply with the admin-guide
For the /proc/sys/vm/compact_memory file, the admin-guide states: When 1
is written to the file, all zones are compacted such that free memory is
available in contiguous blocks where possible. This can be important for
example in the allocation of huge pages although processes will also
directly compact memory as required
But it was not strictly followed, writing any value would cause all zones
to be compacted.
It has been slightly optimized to comply with the admin-guide. Enforce
the 1 on the unlikely chance that the sysctl handler is ever extended to
do something different.
Commit ef4984384172 ("mm/compaction: remove unused variable
sysctl_compact_memory") has also been optimized a bit here, as the
declaration in the external header file has been eliminated, and
sysctl_compact_memory also needs to be verified.
Yosry Ahmed [Fri, 28 Apr 2023 13:24:06 +0000 (13:24 +0000)]
memcg: dump memory.stat during cgroup OOM for v1
Patch series "memcg: OOM log improvements", v2.
This short patch series brings back some cgroup v1 stats in OOM logs
that were unnecessarily changed before. It also makes memcg OOM logs
less reliant on printk() internals.
This patch (of 2):
Commit c8713d0b2312 ("mm: memcontrol: dump memory.stat during cgroup OOM")
made sure we dump all the stats in memory.stat during a cgroup OOM, but it
also introduced a slight behavioral change. The code used to print the
non-hierarchical v1 cgroup stats for the entire cgroup subtree, now it
only prints the v2 cgroup stats for the cgroup under OOM.
For cgroup v1 users, this introduces a few problems:
(a) The non-hierarchical stats of the memcg under OOM are no longer
shown.
(b) A couple of v1-only stats (e.g. pgpgin, pgpgout) are no longer
shown.
(c) We show the list of cgroup v2 stats, even in cgroup v1. This list
of stats is not tracked with v1 in mind. While most of the stats seem
to be working on v1, there may be some stats that are not fully or
correctly tracked.
Although OOM log is not set in stone, we should not change it for no
reason. When upgrading the kernel version to a version including commit c8713d0b2312 ("mm: memcontrol: dump memory.stat during cgroup OOM"), these
behavioral changes are noticed in cgroup v1.
The fix is simple. Commit c8713d0b2312 ("mm: memcontrol: dump memory.stat
during cgroup OOM") separated stats formatting from stats display for v2,
to reuse the stats formatting in the OOM logs. Do the same for v1.
Move the v2 specific formatting from memory_stat_format() to
memcg_stat_format(), add memcg1_stat_format() for v1, and make
memory_stat_format() select between them based on cgroup version. Since
memory_stat_show() now works for both v1 & v2, drop memcg_stat_show().
Yosry Ahmed [Fri, 28 Apr 2023 13:24:05 +0000 (13:24 +0000)]
memcg: use seq_buf_do_printk() with mem_cgroup_print_oom_meminfo()
Currently, we format all the memcg stats into a buffer in
mem_cgroup_print_oom_meminfo() and use pr_info() to dump it to the logs.
However, this buffer is large in size. Although it is currently working
as intended, ther is a dependency between the memcg stats buffer and the
printk record size limit.
If we add more stats in the future and the buffer becomes larger than the
printk record size limit, or if the prink record size limit is reduced,
the logs may be truncated.
It is safer to use seq_buf_do_printk(), which will automatically break up
the buffer at line breaks and issue small printk() calls.
Refactor the code to move the seq_buf from memory_stat_format() to its
callers, and use seq_buf_do_printk() to print the seq_buf in
mem_cgroup_print_oom_meminfo().
Douglas Anderson [Fri, 28 Apr 2023 20:54:38 +0000 (13:54 -0700)]
migrate_pages: avoid blocking for IO in MIGRATE_SYNC_LIGHT
The MIGRATE_SYNC_LIGHT mode is intended to block for things that will
finish quickly but not for things that will take a long time. Exactly how
long is too long is not well defined, but waits of tens of milliseconds is
likely non-ideal.
When putting a Chromebook under memory pressure (opening over 90 tabs on a
4GB machine) it was fairly easy to see delays waiting for some locks in
the kcompactd code path of > 100 ms. While the laptop wasn't amazingly
usable in this state, it was still limping along and this state isn't
something artificial. Sometimes we simply end up with a lot of memory
pressure.
Putting the same Chromebook under memory pressure while it was running
Android apps (though not stressing them) showed a much worse result (NOTE:
this was on a older kernel but the codepaths here are similar). Android
apps on ChromeOS currently run from a 128K-block, zlib-compressed,
loopback-mounted squashfs disk. If we get a page fault from something
backed by the squashfs filesystem we could end up holding a folio lock
while reading enough from disk to decompress 128K (and then decompressing
it using the somewhat slow zlib algorithms). That reading goes through
the ext4 subsystem (because it's a loopback mount) before eventually
ending up in the block subsystem. This extra jaunt adds extra overhead.
Without much work I could see cases where we ended up blocked on a folio
lock for over a second. With more extreme memory pressure I could see up
to 25 seconds.
We considered adding a timeout in the case of MIGRATE_SYNC_LIGHT for the
two locks that were seen to be slow [1] and that generated much
discussion. After discussion, it was decided that we should avoid waiting
for the two locks during MIGRATE_SYNC_LIGHT if they were being held for
IO. We'll continue with the unbounded wait for the more full SYNC modes.
With this change, I couldn't see any slow waits on these locks with my
previous testcases.
NOTE: The reason I stated digging into this originally isn't because some
benchmark had gone awry, but because we've received in-the-field crash
reports where we have a hung task waiting on the page lock (which is the
equivalent code path on old kernels). While the root cause of those
crashes is likely unrelated and won't be fixed by this patch, analyzing
those crash reports did point out these very long waits seemed like
something good to fix. With this patch we should no longer hang waiting
on these locks, but presumably the system will still be in a bad shape and
hang somewhere else.
Roman Gushchin [Tue, 2 May 2023 16:08:39 +0000 (09:08 -0700)]
mm: memcg: use READ_ONCE()/WRITE_ONCE() to access stock->cached
A memcg pointer in the percpu stock can be accessed by drain_all_stock()
from another cpu in a lockless way. In theory it might lead to an issue,
similar to the one which has been discovered with stock->cached_objcg,
where the pointer was zeroed between the check for being NULL and
dereferencing. In this case the issue is unlikely a real problem, but to
make it bulletproof and similar to stock->cached_objcg, let's annotate all
accesses to stock->cached with READ_ONCE()/WTRITE_ONCE().
projects / linux.git / log