Hugh Dickins [Thu, 3 Nov 2022 01:48:45 +0000 (18:48 -0700)]
mm,hugetlb: use folio fields in second tail page
Patch series "mm,huge,rmap: unify and speed up compound mapcounts".
This patch (of 3):
We want to declare one more int in the first tail of a compound page: that
first tail page being valuable property, since every compound page has a
first tail, but perhaps no more than that.
No problem on 64-bit: there is already space for it. No problem with
32-bit THPs: 5.18 commit 5232c63f46fd ("mm: Make compound_pincount always
available") kindly cleared the space for it, apparently not realizing that
only 64-bit architectures enable CONFIG_THP_SWAP (whose use of tail
page->private might conflict) - but make sure of that in its Kconfig.
But hugetlb pages use tail page->private of the first tail page for a
subpool pointer, which will conflict; and they also use page->private of
the 2nd, 3rd and 4th tails.
Undo "mm: add private field of first tail to struct page and struct
folio"'s recent addition of private_1 to the folio tail: instead add
hugetlb_subpool, hugetlb_cgroup, hugetlb_cgroup_rsvd, hugetlb_hwpoison to
a second tail page of the folio: THP has long been using several fields of
that tail, so make better use of it for hugetlb too. This is not how a
generic folio should be declared in future, but it is an effective
transitional way to make use of it.
Delete the SUBPAGE_INDEX stuff, but keep __NR_USED_SUBPAGE: now 3.
Mike Kravetz [Fri, 4 Nov 2022 01:16:32 +0000 (18:16 -0700)]
selftests/vm: update hugetlb madvise
Commit 8ebe0a5eaaeb ("mm,madvise,hugetlb: fix unexpected data loss with
MADV_DONTNEED on hugetlbfs") changed how the passed length was interpreted
for hugetlb mappings. It was changed from align up to align down. The
hugetlb-madvise test explicitly tests this behavior. Change test to
expect new behavior.
Peter Xu [Sun, 30 Oct 2022 21:41:51 +0000 (17:41 -0400)]
mm: use pte markers for swap errors
PTE markers are ideal mechanism for things like SWP_SWAPIN_ERROR. Using a
whole swap entry type for this purpose can be an overkill, especially if
we already have PTE markers. Define a new bit for swapin error and
replace it with pte markers. Then we can safely drop SWP_SWAPIN_ERROR and
give one device slot back to swap.
We used to have SWP_SWAPIN_ERROR taking the page pfn as part of the swap
entry, but it's never used. Neither do I see how it can be useful because
normally the swapin failure should not be caused by a bad page but bad
swap device. Drop it alongside.
Peter Xu [Sun, 30 Oct 2022 21:41:50 +0000 (17:41 -0400)]
mm: always compile in pte markers
Patch series "mm: Use pte marker for swapin errors".
This series uses the pte marker to replace the swapin error swap entry,
then we save one more swap entry slot for swap devices. A new pte marker
bit is defined.
This patch (of 2):
The PTE markers code is tiny and now it's enabled for most of the
distributions. It's fine to keep it as-is, but to make a broader use of
it (e.g. replacing read error swap entry) it needs to be there always
otherwise we need special code path to take care of !PTE_MARKER case.
It'll be easier just make pte marker always exist. Use this chance to
extend its usage to anonymous too by simply touching up some of the old
comments, because it'll be used for anonymous pages in the follow up
patches.
SeongJae Park [Tue, 1 Nov 2022 22:03:25 +0000 (22:03 +0000)]
mm/damon/sysfs-schemes: implement DAMOS-tried regions clear command
When there are huge number of DAMON regions that specific scheme actions
are tried to be applied, directories and files under 'tried_regions'
scheme directory could waste some memory. Add another special input
keyword ('clear_schemes_tried_regions') for 'state' file of each kdamond
sysfs directory that can be used for cleanup of the 'tried_regions'
sub-directories.
SeongJae Park [Tue, 1 Nov 2022 22:03:24 +0000 (22:03 +0000)]
mm/damon/sysfs: implement DAMOS tried regions update command
Implement the code for filling the data of 'tried_regions' DAMON sysfs
directory. With this commit, DAMON sysfs interface users can write a
special keyword, 'update_schemes_tried_regions' to the corresponding
'state' file of the kdamond. Then, DAMON sysfs interface will collect the
tried regions information using the 'before_damos_apply()' callback for
one aggregation interval and populate scheme region directories with the
values.
SeongJae Park [Tue, 1 Nov 2022 22:03:23 +0000 (22:03 +0000)]
mm/damon/sysfs-schemes: implement scheme region directory
Implement region directories under 'tried_regions' directory of each
scheme DAMON sysfs directory. This directory will provide the address
range, the monitored access frequency ('nr_accesses'), and the age of each
DAMON region that corresponding DAMON-based operation scheme has tried to
be applied. Note that this commit doesn't implement the code for filling
the data but only the sysfs directory.
For efficient and simple query-like DAMON monitoring results readings and
deep level investigations of DAMOS, DAMON kernel API
(include/linux/damon.h) users can use 'before_damos_apply' DAMON callback.
However, DAMON sysfs interface users don't have such option.
Add a directory, namely 'tried_regions', under each scheme directory to
use it as the interface for the purpose. Note that this commit is
implementing only the directory but the data filling.
After the data filling change is made, users will be able to signal DAMON
to fill the directory with the regions that corresponding scheme has tried
to be applied. By setting the access pattern of the scheme, users could
do the efficient query-like monitoring.
SeongJae Park [Tue, 1 Nov 2022 22:03:21 +0000 (22:03 +0000)]
mm/damon/core: add a callback for scheme target regions check
Patch series "efficiently expose damos action tried regions information".
DAMON users can retrieve the monitoring results via 'after_aggregation'
callbacks if the user is using the kernel API, or 'damon_aggregated'
tracepoint if the user is in the user space. Those are useful if full
monitoring results are necessary. However, if the user has interest in
only a snapshot of the results for some regions having specific access
pattern, the interfaces could be inefficient. For example, some users
only want to know which memory regions are not accessed for more than a
specific time at the moment.
Also, some DAMOS users would want to know exactly to what memory regions
the schemes' actions tried to be applied, for a debugging or a tuning. As
DAMOS has its internal mechanism for quota and regions prioritization, the
users would need to simulate DAMOS' mechanism against the monitoring
results. That's unnecessarily complex.
This patchset implements DAMON kernel API callbacks and sysfs directory
for efficient exposure of the information for the use cases. The new
callback will be called for each region when a DAMOS action is gonna tried
to be applied to it. The sysfs directory will be called 'tried_regions'
and placed under each scheme sysfs directory. Users can write a special
keyworkd, 'update_schemes_regions', to the 'state' file of a kdamond sysfs
directory. Then, DAMON sysfs interface will fill the directory with the
information of regions that corresponding scheme action was tried to be
applied for next one aggregation interval.
Patches Sequence
----------------
The first one (patch 1) implements the callback for the kernel space
users. Following two patches (patches 2 and 3) implements sysfs
directories for the information and its sub directories. Two patches
(patches 4 and 5) for implementing the special keywords for filling the
data to and cleaning up the directories follow. Patch 6 adds a selftest
for the new sysfs directory. Finally, two patches (patches 7 and 8)
document the new feature in the administrator guide and the ABI document.
This patch (of 8):
Getting DAMON monitoring results of only specific access pattern (e.g.,
getting address ranges of memory that not accessed at all for two minutes)
can be useful for efficient monitoring of the system. The information can
also be helpful for deep level investigation of DAMON-based operation
schemes.
For that, users need to record (in case of the user space users) or
iterate (in case of the kernel space users) full monitoring results and
filter it out for the specific access pattern. In case of the DAMOS
investigation, users will even need to simulate DAMOS' quota and
prioritization mechanisms. It's inefficient and complex.
Add a new DAMON callback that will be called before each scheme is applied
to each region. DAMON kernel API users will be able to do the query-like
monitoring results collection, or DAMOS investigation in an efficient and
simple way using it.
Commits for providing the capability to the user space users will follow.
Sidhartha Kumar [Tue, 1 Nov 2022 22:30:51 +0000 (15:30 -0700)]
mm/hugetlb_cgroup: convert __set_hugetlb_cgroup() to folios
Patch series "convert hugetlb_cgroup helper functions to folios", v2.
This patch series continues the conversion of hugetlb code from being
managed in pages to folios by converting many of the hugetlb_cgroup helper
functions to use folios. This allows the core hugetlb functions to pass
in a folio to these helper functions.
This patch (of 9);
Change __set_hugetlb_cgroup() to use folios so it is explicit that the
function operates on a head page.
Kees Cook [Fri, 28 Oct 2022 15:53:01 +0000 (08:53 -0700)]
mempool: do not use ksize() for poisoning
Nothing appears to be using ksize() within the kmalloc-backed mempools
except the mempool poisoning logic. Use the actual pool size instead of
the ksize() to avoid needing any special handling of the memory as needed
by KASAN, UBSAN_BOUNDS, nor FORTIFY_SOURCE.
Liam Howlett [Fri, 28 Oct 2022 14:45:34 +0000 (14:45 +0000)]
maple_tree: mte_set_full() and mte_clear_full() clang-analyzer clean up
mte_set_full() and mte_clear_full() were incorrectly setting a pointer to
a value without returning a result. Fix this by returning the modified
pointer to be use as necessary. Also add a third function to return if
the bit is set or not.
Johannes Weiner [Wed, 26 Oct 2022 18:01:33 +0000 (14:01 -0400)]
mm: vmscan: split khugepaged stats from direct reclaim stats
Direct reclaim stats are useful for identifying a potential source for
application latency, as well as spotting issues with kswapd. However,
khugepaged currently distorts the picture: as a kernel thread it doesn't
impose allocation latencies on userspace, and it explicitly opts out of
kswapd reclaim. Its activity showing up in the direct reclaim stats is
misleading. Counting it as kswapd reclaim could also cause confusion when
trying to understand actual kswapd behavior.
Break out khugepaged from the direct reclaim counters into new
pgsteal_khugepaged, pgdemote_khugepaged, pgscan_khugepaged counters.
Test with a huge executable (CONFIG_READ_ONLY_THP_FOR_FS):
SeongJae Park [Mon, 24 Oct 2022 17:46:19 +0000 (17:46 +0000)]
Docs/admin-guide/mm/damon/usage: fix wrong usage example of init_regions file
DAMON debugfs interface assumes the users will write all inputs at once.
However, redirecting a string of multiple lines sometimes end up writing
line by line. Therefore, the example usage of 'init_regions' file, which
writes input as a string of multiple lines can fail. Fix it to use a
single line string instead. Also update the description of the usage to
not assume users will write inputs in multiple lines.
SeongJae Park [Mon, 24 Oct 2022 17:46:18 +0000 (17:46 +0000)]
Docs/admin-guide/mm/damon/usage: describe the rules of sysfs region directories
Patch series "Docs/admin-buide/mm/damon/usage: minor fixes".
DAMON usage document contains an unclear description and a wrong usage
example. This patchset fixes the two minor problems.
This patch (of 2):
Target region directories of DAMON sysfs interface should contain no
overlap and sorted by the address, but not clearly documented. Actually,
a user had an issue[1] due to the poor documentation. Add clear
description of it on the usage document.
Muchun Song [Thu, 27 Oct 2022 03:36:41 +0000 (11:36 +0800)]
mm: hugetlb_vmemmap: remove redundant list_del()
The ->lru field will be assigned to a new value in __free_page(). So it
is unnecessary to delete it from the @list. Just remove it to simplify
the code.
Tony Luck [Fri, 21 Oct 2022 20:01:19 +0000 (13:01 -0700)]
mm, hwpoison: try to recover from copy-on write faults
Patch series "Copy-on-write poison recovery", v3.
Part 1 deals with the process that triggered the copy on write fault with
a store to a shared read-only page. That process is send a SIGBUS with
the usual machine check decoration to specify the virtual address of the
lost page, together with the scope.
Part 2 sets up to asynchronously take the page with the uncorrected error
offline to prevent additional machine check faults. H/t to Miaohe Lin
<[email protected]> and Shuai Xue <[email protected]> for
pointing me to the existing function to queue a call to memory_failure().
On x86 there is some duplicate reporting (because the error is also
signalled by the memory controller as well as by the core that triggered
the machine check). Console logs look like this:
This patch (of 2):
If the kernel is copying a page as the result of a copy-on-write
fault and runs into an uncorrectable error, Linux will crash because
it does not have recovery code for this case where poison is consumed
by the kernel.
It is easy to set up a test case. Just inject an error into a private
page, fork(2), and have the child process write to the page.
Add a new copy_user_highpage_mc() function that uses copy_mc_to_kernel()
on architectures where that is available (currently x86 and powerpc).
When an error is detected during the page copy, return VM_FAULT_HWPOISON
to caller of wp_page_copy(). This propagates up the call stack. Both x86
and powerpc have code in their fault handler to deal with this code by
sending a SIGBUS to the application.
Note that this patch avoids a system crash and signals the process that
triggered the copy-on-write action. It does not take any action for the
memory error that is still in the shared page. To handle that a call to
memory_failure() is needed. But this cannot be done from wp_page_copy()
because it holds mmap_lock(). Perhaps the architecture fault handlers
can deal with this loose end in a subsequent patch?
On Intel/x86 this loose end will often be handled automatically because
the memory controller provides an additional notification of the h/w
poison in memory, the handler for this will call memory_failure(). This
isn't a 100% solution. If there are multiple errors, not all may be
logged in this way.
The percpu_counter is used for scenarios where performance is more
important than the accuracy. For percpu_counter users, who want more
accurate information in their slowpath, percpu_counter_sum is provided
which traverses all the online CPUs to accumulate the data. The reason it
only needs to traverse online CPUs is because percpu_counter does
implement CPU offline callback which syncs the local data of the offlined
CPU.
However there is a small race window between the online CPUs traversal of
percpu_counter_sum and the CPU offline callback. The offline callback has
to traverse all the percpu_counters on the system to flush the CPU local
data which can be a lot. During that time, the CPU which is going offline
has already been published as offline to all the readers. So, as the
offline callback is running, percpu_counter_sum can be called for one
counter which has some state on the CPU going offline. Since
percpu_counter_sum only traverses online CPUs, it will skip that specific
CPU and the offline callback might not have flushed the state for that
specific percpu_counter on that offlined CPU.
Normally this is not an issue because percpu_counter users can deal with
some inaccuracy for small time window. However a new user i.e. mm_struct
on the cleanup path wants to check the exact state of the percpu_counter
through check_mm(). For such users, this patch introduces
percpu_counter_sum_all() which traverses all possible CPUs and it is used
in fork.c:check_mm() to avoid the potential race.
This issue is exposed by the later patch "mm: convert mm's rss stats into
percpu_counter".
Shakeel Butt [Mon, 24 Oct 2022 05:28:41 +0000 (05:28 +0000)]
mm: convert mm's rss stats into percpu_counter
Currently mm_struct maintains rss_stats which are updated on page fault
and the unmapping codepaths. For page fault codepath the updates are
cached per thread with the batch of TASK_RSS_EVENTS_THRESH which is 64.
The reason for caching is performance for multithreaded applications
otherwise the rss_stats updates may become hotspot for such applications.
However this optimization comes with the cost of error margin in the rss
stats. The rss_stats for applications with large number of threads can be
very skewed. At worst the error margin is (nr_threads * 64) and we have a
lot of applications with 100s of threads, so the error margin can be very
high. Internally we had to reduce TASK_RSS_EVENTS_THRESH to 32.
Recently we started seeing the unbounded errors for rss_stats for specific
applications which use TCP rx0cp. It seems like vm_insert_pages()
codepath does not sync rss_stats at all.
This patch converts the rss_stats into percpu_counter to convert the error
margin from (nr_threads * 64) to approximately (nr_cpus ^ 2). However
this conversion enable us to get the accurate stats for situations where
accuracy is more important than the cpu cost.
This patch does not make such tradeoffs - we can just use
percpu_counter_add_local() for the updates and percpu_counter_sum() (or
percpu_counter_sync() + percpu_counter_read) for the readers. At the
moment the readers are either procfs interface, oom_killer and memory
reclaim which I think are not performance critical and should be ok with
slow read. However I think we can make that change in a separate patch.
SeongJae Park [Tue, 25 Oct 2022 17:36:50 +0000 (17:36 +0000)]
selftests/damon: add tests for DAMON_LRU_SORT's enabled parameter
Add simple test cases for DAMON_LRU_SORT's 'enabled' parameter. Those
tests are focusing on the synchronous behavior of DAMON_RECLAIM enabling
and disabling.
SeongJae Park [Tue, 25 Oct 2022 17:36:49 +0000 (17:36 +0000)]
mm/damon/lru_sort: enable and disable synchronously
Writing a value to DAMON_RECLAIM's 'enabled' parameter turns on or off
DAMON in an ansychronous way. This means the parameter cannot be used to
read the current status of DAMON_RECLAIM. 'kdamond_pid' parameter should
be used instead for the purpose. The documentation is easy to be read as
it works in a synchronous way, so it is a little bit confusing. It also
makes the user space tooling dirty.
There's no real reason to have the asynchronous behavior, though. Simply
make the parameter works synchronously, rather than updating the document.
SeongJae Park [Tue, 25 Oct 2022 17:36:48 +0000 (17:36 +0000)]
selftests/damon: add tests for DAMON_RECLAIM's enabled parameter
Add simple test cases for DAMON_RECLAIM's 'enabled' parameter. Those
tests are focusing on the synchronous behavior of DAMON_RECLAIM enabling
and disabling.
SeongJae Park [Tue, 25 Oct 2022 17:36:47 +0000 (17:36 +0000)]
mm/damon/reclaim: enable and disable synchronously
Patch series "mm/damon/reclaim,lru_sort: enable/disable synchronously".
Writing a value to DAMON_RECLAIM and DAMON_LRU_SORT's 'enabled' parameters
turns on or off DAMON in an ansychronous way. This means the parameter
cannot be used to read the current status of them. 'kdamond_pid'
parameter should be used instead for the purpose. The documentation is
easy to be read as it works in a synchronous way, so it is a little bit
confusing. It also makes the user space tooling dirty.
There's no real reason to have the asynchronous behavior, though. Simply
make the parameter works synchronously, rather than updating the document.
The first and second patches changes the behavior of the 'enabled'
parameter for DAMON_RECLAIM and adds a selftest for the changed behavior,
respectively. Following two patches make the same changes for
DAMON_LRU_SORT.
This patch (of 4):
Writing a value to DAMON_RECLAIM's 'enabled' parameter turns on or off
DAMON in an ansychronous way. This means the parameter cannot be used to
read the current status of DAMON_RECLAIM. 'kdamond_pid' parameter should
be used instead for the purpose. The documentation is easy to be read as
it works in a synchronous way, so it is a little bit confusing. It also
makes the user space tooling dirty.
There's no real reason to have the asynchronous behavior, though. Simply
make the parameter works synchronously, rather than updating the document.
SeongJae Park [Wed, 26 Oct 2022 22:59:42 +0000 (22:59 +0000)]
mm/damon/modules: deduplicate init steps for DAMON context setup
DAMON_RECLAIM and DAMON_LRU_SORT has duplicated code for DAMON context and
target initializations. Deduplicate the part by implementing a function
for the initialization in 'modules-common.c' and using it.
SeongJae Park [Wed, 26 Oct 2022 22:59:41 +0000 (22:59 +0000)]
mm/damon/sysfs: split out schemes directory implementation to separate file
DAMON sysfs interface for 'schemes' directory is implemented using about
one thousand lines of code. It has no strong dependency with other
parts of its file, so split it out to another file for better code
management.
SeongJae Park [Wed, 26 Oct 2022 22:59:40 +0000 (22:59 +0000)]
mm/damon/sysfs: split out kdamond-independent schemes stats update logic into a new function
'damon_sysfs_schemes_update_stats()' is coupled with both
damon_sysfs_kdamond and damon_sysfs_schemes. It's a wide range of types
dependency. It makes splitting the logics a little bit distracting.
Split the function so that each function is coupled with smaller range of
types.
SeongJae Park [Wed, 26 Oct 2022 22:59:39 +0000 (22:59 +0000)]
mm/damon/sysfs: move unsigned long range directory to common module
The implementation of unsigned long type range directories can be reused
by multiple DAMON sysfs directories including those for DAMON-based
Operation Schemes and the range of number of monitoring regions. Move the
code into the files for DAMON sysfs common logics.
SeongJae Park [Wed, 26 Oct 2022 22:59:38 +0000 (22:59 +0000)]
mm/damon/sysfs: move sysfs_lock to common module
DAMON sysfs interface is implemented in a single file, sysfs.c, which has
about 2,800 lines of code. As the interface is hierarchical and some of
the code can be reused by different hierarchies, it would make more sense
to split out the implementation into common parts and different parts in
multiple files. As the beginning of the work, create files for common
code and move the global mutex for directories modifications protection
into the new file.
SeongJae Park [Wed, 26 Oct 2022 22:59:37 +0000 (22:59 +0000)]
mm/damon/sysfs: remove parameters of damon_sysfs_region_alloc()
'damon_sysfs_region_alloc()' is always called with zero-filled 'struct
damon_addr_range', because the start and end addresses should set by
users. Remove unnecessary parameters of the function and simplify the
body by using 'kzalloc()'.
SeongJae Park [Wed, 26 Oct 2022 22:59:36 +0000 (22:59 +0000)]
mm/damon/sysfs: use damon_addr_range for region's start and end values
DAMON has a struct for each address range but DAMON sysfs interface is
using the low type (unsigned long) for storing the start and end addresses
of regions. Use the dedicated struct for better type safety.
SeongJae Park [Wed, 26 Oct 2022 22:59:35 +0000 (22:59 +0000)]
mm/damon/core: split out scheme quota adjustment logic into a new function
DAMOS quota adjustment logic in 'kdamond_apply_schemes()', has some amount
of code, and the logic is not so straightforward. Split it out to a new
function for better readability.
SeongJae Park [Wed, 26 Oct 2022 22:59:34 +0000 (22:59 +0000)]
mm/damon/core: split out scheme stat update logic into a new function
The function for applying a given DAMON scheme action to a given DAMON
region, 'damos_apply_scheme()' is not quite short. Make it better to read
by splitting out the stat update logic into a new function.
SeongJae Park [Wed, 26 Oct 2022 22:59:33 +0000 (22:59 +0000)]
mm/damon/core: split damos application logic into a new function
The DAMOS action applying function, 'damon_do_apply_schemes()', is still
long and not easy to read. Split out the code for applying a single
action to a single region into a new function for better readability.
SeongJae Park [Wed, 26 Oct 2022 22:59:32 +0000 (22:59 +0000)]
mm/damon/core: split out DAMOS-charged region skip logic into a new function
Patch series "mm/damon: cleanup and refactoring code", v2.
This patchset cleans up and refactors a range of DAMON code including the
core, DAMON sysfs interface, and DAMON modules, for better readability and
convenient future feature implementations.
In detail, this patchset splits unnecessarily long and complex functions
in core into smaller functions (patches 1-4). Then, it cleans up the
DAMON sysfs interface by using more type-safe code (patch 5) and removing
unnecessary function parameters (patch 6). Further, it refactor the code
by distributing the code into multiple files (patches 7-10). Last two
patches (patches 11 and 12) deduplicates and remove unnecessary header
inclusion in DAMON modules (reclaim and lru_sort).
This patch (of 12):
The DAMOS action applying function, 'damon_do_apply_schemes()', is quite
long and not so simple. Split out the already quota-charged region skip
code, which is not a small amount of simple code, into a new function with
some comments for better readability.
HOSTCC scripts/mod/modpost.o - due to target missing
In file included from include/linux/string.h:5,
from scripts/mod/../../include/linux/license.h:5,
from scripts/mod/modpost.c:24:
include/linux/compiler.h:246:10: fatal error: asm/rwonce.h: No such file or directory
246 | #include <asm/rwonce.h>
| ^~~~~~~~~~~~~~
compilation terminated.
...
The problem is that HOSTCC is not necessarily the same compiler or even
architecture as CC and pulling in <linux/compiler.h> or <asm/rwonce.h>
files indirectly isn't a good idea then.
My toolchain is providing HOSTCC = gcc (MacPorts) and CC = arm-linux-gnueabihf
(built from gcc source) and all running on Darwin.
If I change the include to <string.h> I can then "HOSTCC scripts/mod/modpost.c"
but then it fails for "CC kernel/module/main.c" not finding <string.h>:
CC kernel/module/main.o - due to target missing
In file included from kernel/module/main.c:43:0:
./include/linux/license.h:5:20: fatal error: string.h: No such file or directory
#include <string.h>
^
compilation terminated.
Lee Jones [Fri, 25 Nov 2022 12:07:50 +0000 (12:07 +0000)]
Kconfig.debug: provide a little extra FRAME_WARN leeway when KASAN is enabled
When enabled, KASAN enlarges function's stack-frames. Pushing quite a few
over the current threshold. This can mainly be seen on 32-bit
architectures where the present limit (when !GCC) is a lowly 1024-Bytes.
Lee Jones [Fri, 25 Nov 2022 12:07:49 +0000 (12:07 +0000)]
drm/amdgpu: temporarily disable broken Clang builds due to blown stack-frame
Patch series "Fix a bunch of allmodconfig errors", v2.
Since b339ec9c229aa ("kbuild: Only default to -Werror if COMPILE_TEST")
WERROR now defaults to COMPILE_TEST meaning that it's enabled for
allmodconfig builds. This leads to some interesting build failures when
using Clang, each resolved in this set.
With this set applied, I am able to obtain a successful allmodconfig Arm
build.
This patch (of 2):
calculate_bandwidth() is presently broken on all !(X86_64 || SPARC64 ||
ARM64) architectures built with Clang (all released versions), whereby the
stack frame gets blown up to well over 5k. This would cause an immediate
kernel panic on most architectures. We'll revert this when the following
bug report has been resolved:
https://github.com/llvm/llvm-project/issues/41896.
Jann Horn [Fri, 25 Nov 2022 21:37:14 +0000 (22:37 +0100)]
mm/khugepaged: invoke MMU notifiers in shmem/file collapse paths
Any codepath that zaps page table entries must invoke MMU notifiers to
ensure that secondary MMUs (like KVM) don't keep accessing pages which
aren't mapped anymore. Secondary MMUs don't hold their own references to
pages that are mirrored over, so failing to notify them can lead to page
use-after-free.
I'm marking this as addressing an issue introduced in commit f3f0e1d2150b
("khugepaged: add support of collapse for tmpfs/shmem pages"), but most of
the security impact of this only came in commit 27e1f8273113 ("khugepaged:
enable collapse pmd for pte-mapped THP"), which actually omitted flushes
for the removal of present PTEs, not just for the removal of empty page
tables.
Jann Horn [Fri, 25 Nov 2022 21:37:13 +0000 (22:37 +0100)]
mm/khugepaged: fix GUP-fast interaction by sending IPI
Since commit 70cbc3cc78a99 ("mm: gup: fix the fast GUP race against THP
collapse"), the lockless_pages_from_mm() fastpath rechecks the pmd_t to
ensure that the page table was not removed by khugepaged in between.
However, lockless_pages_from_mm() still requires that the page table is
not concurrently freed. Fix it by sending IPIs (if the architecture uses
semi-RCU-style page table freeing) before freeing/reusing page tables.
Jann Horn [Fri, 25 Nov 2022 21:37:12 +0000 (22:37 +0100)]
mm/khugepaged: take the right locks for page table retraction
pagetable walks on address ranges mapped by VMAs can be done under the
mmap lock, the lock of an anon_vma attached to the VMA, or the lock of the
VMA's address_space. Only one of these needs to be held, and it does not
need to be held in exclusive mode.
Under those circumstances, the rules for concurrent access to page table
entries are:
- Terminal page table entries (entries that don't point to another page
table) can be arbitrarily changed under the page table lock, with the
exception that they always need to be consistent for
hardware page table walks and lockless_pages_from_mm().
This includes that they can be changed into non-terminal entries.
- Non-terminal page table entries (which point to another page table)
can not be modified; readers are allowed to READ_ONCE() an entry, verify
that it is non-terminal, and then assume that its value will stay as-is.
Retracting a page table involves modifying a non-terminal entry, so
page-table-level locks are insufficient to protect against concurrent page
table traversal; it requires taking all the higher-level locks under which
it is possible to start a page walk in the relevant range in exclusive
mode.
The collapse_huge_page() path for anonymous THP already follows this rule,
but the shmem/file THP path was getting it wrong, making it possible for
concurrent rmap-based operations to cause corruption.
Gavin Shan [Thu, 24 Nov 2022 09:55:23 +0000 (17:55 +0800)]
mm: migrate: fix THP's mapcount on isolation
The issue is reported when removing memory through virtio_mem device. The
transparent huge page, experienced copy-on-write fault, is wrongly
regarded as pinned. The transparent huge page is escaped from being
isolated in isolate_migratepages_block(). The transparent huge page can't
be migrated and the corresponding memory block can't be put into offline
state.
Fix it by replacing page_mapcount() with total_mapcount(). With this, the
transparent huge page can be isolated and migrated, and the memory block
can be put into offline state. Besides, The page's refcount is increased
a bit earlier to avoid the page is released when the check is executed.
Juergen Gross [Wed, 23 Nov 2022 06:45:10 +0000 (07:45 +0100)]
mm: introduce arch_has_hw_nonleaf_pmd_young()
When running as a Xen PV guests commit eed9a328aa1a ("mm: x86: add
CONFIG_ARCH_HAS_NONLEAF_PMD_YOUNG") can cause a protection violation in
pmdp_test_and_clear_young():
This happens because the Xen hypervisor can't emulate direct writes to
page table entries other than PTEs.
This can easily be fixed by introducing arch_has_hw_nonleaf_pmd_young()
similar to arch_has_hw_pte_young() and test that instead of
CONFIG_ARCH_HAS_NONLEAF_PMD_YOUNG.
Juergen Gross [Wed, 30 Nov 2022 22:49:41 +0000 (14:49 -0800)]
mm: add dummy pmd_young() for architectures not having it
In order to avoid #ifdeffery add a dummy pmd_young() implementation as a
fallback. This is required for the later patch "mm: introduce
arch_has_hw_nonleaf_pmd_young()".
SeongJae Park [Tue, 22 Nov 2022 19:48:31 +0000 (19:48 +0000)]
mm/damon/sysfs: fix wrong empty schemes assumption under online tuning in damon_sysfs_set_schemes()
Commit da87878010e5 ("mm/damon/sysfs: support online inputs update") made
'damon_sysfs_set_schemes()' to be called for running DAMON context, which
could have schemes. In the case, DAMON sysfs interface is supposed to
update, remove, or add schemes to reflect the sysfs files. However, the
code is assuming the DAMON context wouldn't have schemes at all, and
therefore creates and adds new schemes. As a result, the code doesn't
work as intended for online schemes tuning and could have more than
expected memory footprint. The schemes are all in the DAMON context, so
it doesn't leak the memory, though.
Remove the wrong asssumption (the DAMON context wouldn't have schemes) in
'damon_sysfs_set_schemes()' to fix the bug.
Tiezhu Yang [Sat, 19 Nov 2022 02:36:59 +0000 (10:36 +0800)]
tools/vm/slabinfo-gnuplot: use "grep -E" instead of "egrep"
The latest version of grep claims the egrep is now obsolete so the build
now contains warnings that look like:
egrep: warning: egrep is obsolescent; using grep -E
fix this up by moving the related file to use "grep -E" instead.
sed -i "s/egrep/grep -E/g" `grep egrep -rwl tools/vm`
Here are the steps to install the latest grep:
wget http://ftp.gnu.org/gnu/grep/grep-3.8.tar.gz
tar xf grep-3.8.tar.gz
cd grep-3.8 && ./configure && make
sudo make install
export PATH=/usr/local/bin:$PATH
If DAT metadata file is corrupted on disk, there is a case where
req->pr_desc_bh is NULL and blocknr is 0 at nilfs_dat_commit_end() during
a b-tree operation that cascadingly updates ancestor nodes of the b-tree,
because nilfs_dat_commit_alloc() for a lower level block can initialize
the blocknr on the same DAT entry between nilfs_dat_prepare_end() and
nilfs_dat_commit_end().
If this happens, nilfs_dat_commit_end() calls nilfs_dat_commit_free()
without valid buffer heads in req->pr_desc_bh and req->pr_bitmap_bh, and
causes the NULL pointer dereference above in
nilfs_palloc_commit_free_entry() function, which leads to a crash.
Fix this by adding a NULL check on req->pr_desc_bh and req->pr_bitmap_bh
before nilfs_palloc_commit_free_entry() in nilfs_dat_commit_free().
This also calls nilfs_error() in that case to notify that there is a fatal
flaw in the filesystem metadata and prevent further operations.
Mike Kravetz [Mon, 14 Nov 2022 23:55:06 +0000 (15:55 -0800)]
hugetlb: don't delete vma_lock in hugetlb MADV_DONTNEED processing
madvise(MADV_DONTNEED) ends up calling zap_page_range() to clear page
tables associated with the address range. For hugetlb vmas,
zap_page_range will call __unmap_hugepage_range_final. However,
__unmap_hugepage_range_final assumes the passed vma is about to be removed
and deletes the vma_lock to prevent pmd sharing as the vma is on the way
out. In the case of madvise(MADV_DONTNEED) the vma remains, but the
missing vma_lock prevents pmd sharing and could potentially lead to issues
with truncation/fault races.
This issue was originally reported here [1] as a BUG triggered in
page_try_dup_anon_rmap. Prior to the introduction of the hugetlb
vma_lock, __unmap_hugepage_range_final cleared the VM_MAYSHARE flag to
prevent pmd sharing. Subsequent faults on this vma were confused as
VM_MAYSHARE indicates a sharable vma, but was not set so page_mapping was
not set in new pages added to the page table. This resulted in pages that
appeared anonymous in a VM_SHARED vma and triggered the BUG.
Address issue by adding a new zap flag ZAP_FLAG_UNMAP to indicate an unmap
call from unmap_vmas(). This is used to indicate the 'final' unmapping of
a hugetlb vma. When called via MADV_DONTNEED, this flag is not set and
the vm_lock is not deleted.
Mike Kravetz [Mon, 14 Nov 2022 23:55:05 +0000 (15:55 -0800)]
madvise: use zap_page_range_single for madvise dontneed
This series addresses the issue first reported in [1], and fully described
in patch 2. Patches 1 and 2 address the user visible issue and are tagged
for stable backports.
While exploring solutions to this issue, related problems with mmu
notification calls were discovered. This is addressed in the patch
"hugetlb: remove duplicate mmu notifications:". Since there are no user
visible effects, this third is not tagged for stable backports.
Previous discussions suggested further cleanup by removing the
routine zap_page_range. This is possible because zap_page_range_single
is now exported, and all callers of zap_page_range pass ranges entirely
within a single vma. This work will be done in a later patch so as not
to distract from this bug fix.
Expose the routine zap_page_range_single to zap a range within a single
vma. The madvise routine madvise_dontneed_single_vma can use this routine
as it explicitly operates on a single vma. Also, update the mmu
notification range in zap_page_range_single to take hugetlb pmd sharing
into account. This is required as MADV_DONTNEED supports hugetlb vmas.
It is because khugepaged allocates pages with __GFP_THISNODE, but the
preferred node is bogus. The previous patch fixed the khugepaged code to
avoid allocating page from non-existing node. But it is still racy
against memory hotremove. There is no synchronization with the memory
hotplug so it is possible that memory gets offline during a longer taking
scanning.
So this warning still seems not quite helpful because:
* There is no guarantee the node is online for __GFP_THISNODE context
for all the callsites.
* Kernel just fails the allocation regardless the warning, and it looks
all callsites handle the allocation failure gracefully.
Although while the warning has helped to identify a buggy code, it is not
safe in general and this warning could panic the system with panic-on-warn
configuration which tends to be used surprisingly often. So replace
VM_WARN_ON to pr_warn(). And the warning will be triggered if
__GFP_NOWARN is set since the allocator would print out warning for such
case if __GFP_NOWARN is not set.
Li Hua [Mon, 21 Nov 2022 03:06:20 +0000 (11:06 +0800)]
test_kprobes: fix implicit declaration error of test_kprobes
If KPROBES_SANITY_TEST and ARCH_CORRECT_STACKTRACE_ON_KRETPROBE is enabled, but
STACKTRACE is not set. Build failed as below:
lib/test_kprobes.c: In function `stacktrace_return_handler':
lib/test_kprobes.c:228:8: error: implicit declaration of function `stack_trace_save'; did you mean `stacktrace_driver'? [-Werror=implicit-function-declaration]
ret = stack_trace_save(stack_buf, STACK_BUF_SIZE, 0);
^~~~~~~~~~~~~~~~
stacktrace_driver
cc1: all warnings being treated as errors
scripts/Makefile.build:250: recipe for target 'lib/test_kprobes.o' failed
make[2]: *** [lib/test_kprobes.o] Error 1
To fix this error, Select STACKTRACE if ARCH_CORRECT_STACKTRACE_ON_KRETPROBE is enabled.
Chen Zhongjin [Fri, 18 Nov 2022 06:33:04 +0000 (14:33 +0800)]
nilfs2: fix nilfs_sufile_mark_dirty() not set segment usage as dirty
When extending segments, nilfs_sufile_alloc() is called to get an
unassigned segment, then mark it as dirty to avoid accidentally allocating
the same segment in the future.
But for some special cases such as a corrupted image it can be unreliable.
If such corruption of the dirty state of the segment occurs, nilfs2 may
reallocate a segment that is in use and pick the same segment for writing
twice at the same time.
This case started with segbuf1.segnum = 3, nextnum = 4 when constructed.
It supposed segment 4 has already been allocated and marked as dirty.
However the dirty state was corrupted and segment 4 usage was not dirty.
For the first time nilfs_segctor_extend_segments() segment 4 was allocated
again, which made segbuf2 and next segbuf3 had same segment 4.
sb_getblk() will get same bh for segbuf2 and segbuf3, and this bh is added
to both buffer lists of two segbuf. It makes the lists broken which
causes NULL pointer dereference.
Fix the problem by setting usage as dirty every time in
nilfs_sufile_mark_dirty(), which is called during constructing current
segment to be written out and before allocating next segment.
mm/cgroup/reclaim: fix dirty pages throttling on cgroup v1
balance_dirty_pages doesn't do the required dirty throttling on cgroupv1.
See commit 9badce000e2c ("cgroup, writeback: don't enable cgroup writeback
on traditional hierarchies"). Instead, the kernel depends on writeback
throttling in shrink_folio_list to achieve the same goal. With large
memory systems, the flusher may not be able to writeback quickly enough
such that we will start finding pages in the shrink_folio_list already in
writeback. Hence for cgroupv1 let's do a reclaim throttle after waking up
the flusher.
The below test which used to fail on a 256GB system completes till the the
file system is full with this change.
Qi Zheng [Fri, 18 Nov 2022 10:00:11 +0000 (18:00 +0800)]
mm: fix unexpected changes to {failslab|fail_page_alloc}.attr
When we specify __GFP_NOWARN, we only expect that no warnings will be
issued for current caller. But in the __should_failslab() and
__should_fail_alloc_page(), the local GFP flags alter the global
{failslab|fail_page_alloc}.attr, which is persistent and shared by all
tasks. This is not what we expected, let's fix it.
Chen Wandun [Fri, 18 Nov 2022 13:38:50 +0000 (21:38 +0800)]
swapfile: fix soft lockup in scan_swap_map_slots
A softlockup occurs in scan free swap slot under huge memory pressure.
The test scenario is: 64 CPU cores, 64GB memory, and 28 zram devices, the
disksize of each zram device is 50MB.
LATENCY_LIMIT is used to prevent softlockups in scan_swap_map_slots(), but
the real loop number would more than LATENCY_LIMIT because of "goto checks
and goto scan" repeatly without decreasing latency limit.
In order to fix it, decrease latency_ration in advance.
There is also a suspicious place that will cause softlockups in
get_swap_pages(). In this function, the "goto start_over" may result in
continuous scanning of the swap partition. If there is no cond_sched in
scan_swap_map_slots(), it would cause a softlockup (I am not sure about
this).
Mike Kravetz [Fri, 18 Nov 2022 19:52:49 +0000 (11:52 -0800)]
hugetlb: fix __prep_compound_gigantic_page page flag setting
Commit 2b21624fc232 ("hugetlb: freeze allocated pages before creating
hugetlb pages") changed the order page flags were cleared and set in the
head page. It moved the __ClearPageReserved after __SetPageHead.
However, there is a check to make sure __ClearPageReserved is never done
on a head page. If CONFIG_DEBUG_VM_PGFLAGS is enabled, the following BUG
will be hit when creating a hugetlb gigantic page:
page dumped because: VM_BUG_ON_PAGE(1 && PageCompound(page))
------------[ cut here ]------------
kernel BUG at include/linux/page-flags.h:500!
Call Trace will differ depending on whether hugetlb page is created
at boot time or run time.
Make sure to __ClearPageReserved BEFORE __SetPageHead.
Marco Elver [Fri, 18 Nov 2022 15:22:16 +0000 (16:22 +0100)]
kfence: fix stack trace pruning
Commit b14051352465 ("mm/sl[au]b: generalize kmalloc subsystem")
refactored large parts of the kmalloc subsystem, resulting in the stack
trace pruning logic done by KFENCE to no longer work.
While b14051352465 attempted to fix the situation by including
'__kmem_cache_free' in the list of functions KFENCE should skip through,
this only works when the compiler actually optimized the tail call from
kfree() to __kmem_cache_free() into a jump (and thus kfree() _not_
appearing in the full stack trace to begin with).
In some configurations, the compiler no longer optimizes the tail call
into a jump, and __kmem_cache_free() appears in the stack trace. This
means that the pruned stack trace shown by KFENCE would include kfree()
which is not intended - for example:
| BUG: KFENCE: invalid free in kfree+0x7c/0x120
|
| Invalid free of 0xffff8883ed8fefe0 (in kfence-#126):
| kfree+0x7c/0x120
| test_double_free+0x116/0x1a9
| kunit_try_run_case+0x90/0xd0
| [...]
Fix it by moving __kmem_cache_free() to the list of functions that may be
tail called by an allocator entry function, making the pruning logic work
in both the optimized and unoptimized tail call cases.
Yu Zhao [Wed, 16 Nov 2022 01:38:07 +0000 (18:38 -0700)]
mm: multi-gen LRU: retry folios written back while isolated
The page reclaim isolates a batch of folios from the tail of one of the
LRU lists and works on those folios one by one. For a suitable
swap-backed folio, if the swap device is async, it queues that folio for
writeback. After the page reclaim finishes an entire batch, it puts back
the folios it queued for writeback to the head of the original LRU list.
In the meantime, the page writeback flushes the queued folios also by
batches. Its batching logic is independent from that of the page reclaim.
For each of the folios it writes back, the page writeback calls
folio_rotate_reclaimable() which tries to rotate a folio to the tail.
folio_rotate_reclaimable() only works for a folio after the page reclaim
has put it back. If an async swap device is fast enough, the page
writeback can finish with that folio while the page reclaim is still
working on the rest of the batch containing it. In this case, that folio
will remain at the head and the page reclaim will not retry it before
reaching there.
This patch adds a retry to evict_folios(). After evict_folios() has
finished an entire batch and before it puts back folios it cannot free
immediately, it retries those that may have missed the rotation.
Before this patch, ~60% of folios swapped to an Intel Optane missed
folio_rotate_reclaimable(). After this patch, ~99% of missed folios were
reclaimed upon retry.
This problem affects relatively slow async swap devices like Samsung 980
Pro much less and does not affect sync swap devices like zram or zswap at
all.
Alistair Popple [Fri, 11 Nov 2022 00:51:35 +0000 (11:51 +1100)]
mm/migrate_device: return number of migrating pages in args->cpages
migrate_vma->cpages originally contained a count of the number of pages
migrating including non-present pages which can be populated directly on
the target.
Commit 241f68859656 ("mm/migrate_device.c: refactor migrate_vma and
migrate_device_coherent_page()") inadvertantly changed this to contain
just the number of pages that were unmapped. Usage of migrate_vma->cpages
isn't documented, but most drivers use it to see if all the requested
addresses can be migrated so restore the original behaviour.
Sam James [Wed, 16 Nov 2022 18:26:34 +0000 (18:26 +0000)]
kbuild: fix -Wimplicit-function-declaration in license_is_gpl_compatible
Add missing <linux/string.h> include for strcmp.
Clang 16 makes -Wimplicit-function-declaration an error by default.
Unfortunately, out of tree modules may use this in configure scripts,
which means failure might cause silent miscompilation or misconfiguration.
For more information, see LWN.net [0] or LLVM's Discourse [1], gentoo-dev@ [2],
or the (new) c-std-porting mailing list [3].
[0] https://lwn.net/Articles/913505/
[1] https://discourse.llvm.org/t/configure-script-breakage-with-the-new-werror-implicit-function-declaration/65213
[2] https://archives.gentoo.org/gentoo-dev/message/dd9f2d3082b8b6f8dfbccb0639e6e240
[3] hosted at lists.linux.dev.
Ian Cowan [Mon, 14 Nov 2022 00:33:49 +0000 (19:33 -0500)]
mm: mmap: fix documentation for vma_mas_szero
When the struct_mm input, mm, was changed to a struct ma_state, mas, the
documentation for the function was never updated. This updates that
documentation reference.
SeongJae Park [Mon, 14 Nov 2022 17:55:52 +0000 (17:55 +0000)]
mm/damon/sysfs-schemes: skip stats update if the scheme directory is removed
A DAMON sysfs interface user can start DAMON with a scheme, remove the
sysfs directory for the scheme, and then ask update of the scheme's stats.
Because the schemes stats update logic isn't aware of the situation, it
results in an invalid memory access. Fix the bug by checking if the
scheme sysfs directory exists.
Alistair Popple [Mon, 14 Nov 2022 11:55:37 +0000 (22:55 +1100)]
mm/memory: return vm_fault_t result from migrate_to_ram() callback
The migrate_to_ram() callback should always succeed, but in rare cases can
fail usually returning VM_FAULT_SIGBUS. Commit 16ce101db85d
("mm/memory.c: fix race when faulting a device private page") incorrectly
stopped passing the return code up the stack. Fix this by setting the ret
variable, restoring the previous behaviour on migrate_to_ram() failure.
Li Liguang [Mon, 14 Nov 2022 19:48:28 +0000 (14:48 -0500)]
mm: correctly charge compressed memory to its memcg
Kswapd will reclaim memory when memory pressure is high, the annonymous
memory will be compressed and stored in the zpool if zswap is enabled.
The memcg_kmem_bypass() in get_obj_cgroup_from_page() will bypass the
kernel thread and cause the compressed memory not be charged to its memory
cgroup.
Remove the memcg_kmem_bypass() call and properly charge compressed memory
to its corresponding memory cgroup.
Mike Kravetz [Mon, 14 Nov 2022 21:00:18 +0000 (13:00 -0800)]
ipc/shm: call underlying open/close vm_ops
Shared memory segments can be created that are backed by hugetlb pages.
When this happens, the vmas associated with any mappings (shmat) are
marked VM_HUGETLB, yet the vm_ops for such mappings are provided by
ipc/shm (shm_vm_ops). There is a mechanism to call the underlying hugetlb
vm_ops, and this is done for most operations. However, it is not done for
open and close.
This was not an issue until the introduction of the hugetlb vma_lock.
This lock structure is pointed to by vm_private_data and the open/close
vm_ops help maintain this structure. The special hugetlb routine called
at fork took care of structure updates at fork time. However,
vma_splitting is not properly handled for ipc shared memory mappings
backed by hugetlb pages. This can result in a "kernel NULL pointer
dereference" BUG or use after free as two vmas point to the same lock
structure.
Update the shm open and close routines to always call the underlying open
and close routines.
Mukesh Ojha [Wed, 9 Nov 2022 19:01:37 +0000 (00:31 +0530)]
gcov: clang: fix the buffer overflow issue
Currently, in clang version of gcov code when module is getting removed
gcov_info_add() incorrectly adds the sfn_ptr->counter to all the
dst->functions and it result in the kernel panic in below crash report.
Fix this by properly handling it.
Gautam Menghani [Wed, 26 Oct 2022 04:45:24 +0000 (10:15 +0530)]
mm/khugepaged: refactor mm_khugepaged_scan_file tracepoint to remove filename from function call
Refactor the mm_khugepaged_scan_file tracepoint to move filename
dereference to the tracepoint definition, to maintain consistency with
other tracepoints[1].
mm/page_exit: fix kernel doc warning in page_ext_put()
Fix the below compiler warnings reported with 'make W=1 mm/'.
mm/page_ext.c:178: warning: Function parameter or member 'page_ext' not
described in 'page_ext_put'.
The khugepaged code would pick up the node with the most hit as the preferred
node, and also tries to do some balance if several nodes have the same
hit record. Basically it does conceptually:
* If the target_node <= last_target_node, then iterate from
last_target_node + 1 to MAX_NUMNODES (1024 on default config)
* If the max_value == node_load[nid], then target_node = nid
But there is a corner case, paritucularly for MADV_COLLAPSE, that the
non-existing node may be returned as preferred node.
Assuming the system has 2 nodes, the target_node is 0 and the
last_target_node is 1, if MADV_COLLAPSE path is hit, the max_value may
be 0, then it may return 2 for target_node, but it is actually not
existing (offline), so the warn is triggered.
The node balance was introduced by commit 9f1b868a13ac ("mm: thp:
khugepaged: add policy for finding target node") to satisfy
"numactl --interleave=all". But interleaving is a mere hint rather than
something that has hard requirements.
So use nodemask to record the nodes which have the same hit record, the
hugepage allocation could fallback to those nodes. And remove
__GFP_THISNODE since it does disallow fallback. And if the nodemask
just has one node set, it means there is one single node has the most
hit record, the nodemask approach actually behaves like __GFP_THISNODE.
While he reclaimer requested 4M, vmscan reclaimed close to 16G, most of it
by swapping. These requests take over a minute, during which the write()
to memory.reclaim is unkillably stuck inside the kernel.
Digging into the source, this is caused by the proportional reclaim
bailout logic. This code tries to resolve a fundamental conflict: to
reclaim roughly what was requested, while also aging all LRUs fairly and
in accordance to their size, swappiness, refault rates etc. The way it
attempts fairness is that once the reclaim goal has been reached, it stops
scanning the LRUs with the smaller remaining scan targets, and adjusts the
remainder of the bigger LRUs according to how much of the smaller LRUs was
scanned. It then finishes scanning that remainder regardless of the
reclaim goal.
This works fine if priority levels are low and the LRU lists are
comparable in size. However, in this instance, the cgroup that is
targeted by proactive reclaim has almost no files left - they've already
been squeezed out by proactive reclaim earlier - and the remaining anon
pages are hot. Anon rotations cause the priority level to drop to 0,
which results in reclaim targeting all of anon (a lot) and all of file
(almost nothing). By the time reclaim decides to bail, it has scanned
most or all of the file target, and therefor must also scan most or all of
the enormous anon target. This target is thousands of times larger than
the reclaim goal, thus causing the overreclaim.
The bailout code hasn't changed in years, why is this failing now? The
most likely explanations are two other recent changes in anon reclaim:
1. Before the series starting with commit 5df741963d52 ("mm: fix LRU
balancing effect of new transparent huge pages"), the VM was
overall relatively reluctant to swap at all, even if swap was
configured. This means the LRU balancing code didn't come into play
as often as it does now, and mostly in high pressure situations
where pronounced swap activity wouldn't be as surprising.
2. For historic reasons, shrink_lruvec() loops on the scan targets of
all LRU lists except the active anon one, meaning it would bail if
the only remaining pages to scan were active anon - even if there
were a lot of them.
Before the series starting with commit ccc5dc67340c ("mm/vmscan:
make active/inactive ratio as 1:1 for anon lru"), most anon pages
would live on the active LRU; the inactive one would contain only a
handful of preselected reclaim candidates. After the series, anon
gets aged similarly to file, and the inactive list is the default
for new anon pages as well, making it often the much bigger list.
As a result, the VM is now more likely to actually finish large
anon targets than before.
Change the code such that only one SWAP_CLUSTER_MAX-sized nudge toward the
larger LRU lists is made before bailing out on a met reclaim goal.
This fixes the extreme overreclaim problem.
Fairness is more subtle and harder to evaluate. No obvious misbehavior
was observed on the test workload, in any case. Conceptually, fairness
should primarily be a cumulative effect from regular, lower priority
scans. Once the VM is in trouble and needs to escalate scan targets to
make forward progress, fairness needs to take a backseat. This is also
acknowledged by the myriad exceptions in get_scan_count(). This patch
makes fairness decrease gradually, as it keeps fairness work static over
increasing priority levels with growing scan targets. This should make
more sense - although we may have to re-visit the exact values.
Kairui Song [Sun, 23 Oct 2022 16:25:33 +0000 (00:25 +0800)]
swap: add a limit for readahead page-cluster value
Currenty there is no upper limit for /proc/sys/vm/page-cluster, and it's a
bit shift value, so it could result in overflow of the 32-bit integer.
Add a reasonable upper limit for it, read-in at most 2**31 pages, which is
a large enough value for readahead.
Currently PageHWPoison flag does not behave well when experiencing memory
hotremove/hotplug. Any data field in struct page is unreliable when the
associated memory is offlined, and the current mechanism can't tell
whether a memory block is onlined because a new memory devices is
installed or because previous failed offline operations are undone.
Especially if there's a hwpoisoned memory, it's unclear what the best
option is.
So introduce a new mechanism to make struct memory_block remember that a
memory block has hwpoisoned memory inside it. And make any online event
fail if the onlining memory block contains hwpoison. struct memory_block
is freed and reallocated over ACPI-based hotremove/hotplug, but not over
sysfs-based hotremove/hotplug. So the new counter can distinguish these
cases.
Naoya Horiguchi [Mon, 24 Oct 2022 06:20:10 +0000 (15:20 +0900)]
mm/hwpoison: move definitions of num_poisoned_pages_* to memory-failure.c
These interfaces will be used by drivers/base/memory.c by later patch, so
as a preparatory work move them to more common header file visible to the
file.
Naoya Horiguchi [Mon, 24 Oct 2022 06:20:09 +0000 (15:20 +0900)]
mm,hwpoison,hugetlb,memory_hotplug: hotremove memory section with hwpoisoned hugepage
Patch series "mm, hwpoison: improve handling workload related to hugetlb
and memory_hotplug", v7.
This patchset tries to solve the issue among memory_hotplug, hugetlb and hwpoison.
In this patchset, memory hotplug handles hwpoison pages like below:
- hwpoison pages should not prevent memory hotremove,
- memory block with hwpoison pages should not be onlined.
This patch (of 4):
HWPoisoned page is not supposed to be accessed once marked, but currently
such accesses can happen during memory hotremove because
do_migrate_range() can be called before dissolve_free_huge_pages() is
called.
Clear HPageMigratable for hwpoisoned hugepages to prevent them from being
migrated. This should be done in hugetlb_lock to avoid race against
isolate_hugetlb().
get_hwpoison_huge_page() needs to have a flag to show it's called from
unpoison to take refcount of hwpoisoned hugepages, so add it.
Baolin Wang [Mon, 24 Oct 2022 08:34:22 +0000 (16:34 +0800)]
mm: migrate: try again if THP split is failed due to page refcnt
When creating a virtual machine, we will use memfd_create() to get a file
descriptor which can be used to create share memory mappings using the
mmap function, meanwhile the mmap() will set the MAP_POPULATE flag to
allocate physical pages for the virtual machine.
When allocating physical pages for the guest, the host can fallback to
allocate some CMA pages for the guest when over half of the zone's free
memory is in the CMA area.
In guest os, when the application wants to do some data transaction with
DMA, our QEMU will call VFIO_IOMMU_MAP_DMA ioctl to do longterm-pin and
create IOMMU mappings for the DMA pages. However, when calling
VFIO_IOMMU_MAP_DMA ioctl to pin the physical pages, we found it will be
failed to longterm-pin sometimes.
After some invetigation, we found the pages used to do DMA mapping can
contain some CMA pages, and these CMA pages will cause a possible failure
of the longterm-pin, due to failed to migrate the CMA pages. The reason
of migration failure may be temporary reference count or memory allocation
failure. So that will cause the VFIO_IOMMU_MAP_DMA ioctl returns error,
which makes the application failed to start.
I observed one migration failure case (which is not easy to reproduce) is
that, the 'thp_migration_fail' count is 1 and the 'thp_split_page_failed'
count is also 1.
That means when migrating a THP which is in CMA area, but can not allocate
a new THP due to memory fragmentation, so it will split the THP. However
THP split is also failed, probably the reason is temporary reference count
of this THP. And the temporary reference count can be caused by dropping
page caches (I observed the drop caches operation in the system), but we
can not drop the shmem page caches due to they are already dirty at that
time.
Especially for THP split failure, which is caused by temporary reference
count, we can try again to mitigate the failure of migration in this case
according to previous discussion [1].
According to commit "vfs: parse: deal with zero length string value",
kernel will set the param->string to null pointer in vfs_parse_fs_string()
if fs string has zero length.
Yet the problem is that, hugetlbfs_parse_param() will dereference the
param->string, without checking whether it is a null pointer. To be more
specific, if hugetlbfs_parse_param() parses an illegal mount parameter,
such as "size=,", kernel will constructs struct fs_parameter with null
pointer in vfs_parse_fs_string(), then passes this struct fs_parameter to
hugetlbfs_parse_param(), which triggers the above null-ptr-deref bug.
This patch solves it by adding sanity check on param->string
in hugetlbfs_parse_param().
projects / linux.git / log