Besides fsverity, fscrypt, and the data=journal mode, ext4 now supports
large folios for regular files. Enable this feature by default. However,
since we cannot change the folio order limitation of mappings on active
inodes, setting the journal=data mode via ioctl on an active inode will
not take immediate effect in non-delalloc mode.
Signed-off-by: Zhang Yi <yi.zhang@huawei.com>
Link: https://patch.msgid.link/20250512063319.3539411-9-yi.zhang@huaweicloud.com
Signed-off-by: Theodore Ts'o <tytso@mit.edu>
move_extent_per_page() currently assumes that each folio is the size of
PAGE_SIZE and only copies data for one page. ext4_move_extents() should
call move_extent_per_page() for each page. To support larger folios,
simply modify the calculations for the block start and end offsets
within the folio based on the provided range of 'data_offset_in_page'
and 'block_len_in_page'. This function will continue to handle PAGE_SIZE
of data at a time and will not convert this function to manage an entire
folio. Additionally, we use the source folio to copy data, so it doesn't
matter if the source and dest folios are different in size.
Signed-off-by: Zhang Yi <yi.zhang@huawei.com>
Link: https://patch.msgid.link/20250512063319.3539411-8-yi.zhang@huaweicloud.com
Signed-off-by: Theodore Ts'o <tytso@mit.edu>
In mpage_map_and_submit_buffers(), the 'lblk' is now aligned to
PAGE_SIZE. Convert it to be aligned to folio size. Additionally, modify
the wbc->nr_to_write update to reduce the number of pages in a single
folio, ensuring that the entire writeback path can support large folios.
Signed-off-by: Zhang Yi <yi.zhang@huawei.com>
Link: https://patch.msgid.link/20250512063319.3539411-7-yi.zhang@huaweicloud.com
Signed-off-by: Theodore Ts'o <tytso@mit.edu>
The journal credits calculation in ext4_ext_index_trans_blocks() is
currently inadequate. It only multiplies the depth of the extents tree
and doesn't account for the blocks that may be required for adding the
leaf extents themselves.
After enabling large folios, we can easily run out of handle credits,
triggering a warning in jbd2_journal_dirty_metadata() on filesystems
with a 1KB block size. This occurs because we may need more extents when
iterating through each large folio in
ext4_do_writepages()->mpage_map_and_submit_extent(). Therefore, we
should modify ext4_ext_index_trans_blocks() to include a count of the
leaf extents in the worst case as well.
Signed-off-by: Zhang Yi <yi.zhang@huawei.com>
Link: https://patch.msgid.link/20250512063319.3539411-6-yi.zhang@huaweicloud.com
Signed-off-by: Theodore Ts'o <tytso@mit.edu>
jbd2_journal_blocks_per_page() returns the number of blocks in a single
page. Rename it to jbd2_journal_blocks_per_folio() and make it returns
the number of blocks in the largest folio, preparing for the calculation
of journal credits blocks when allocating blocks within a large folio in
the writeback path.
Signed-off-by: Zhang Yi <yi.zhang@huawei.com>
Link: https://patch.msgid.link/20250512063319.3539411-5-yi.zhang@huaweicloud.com
Signed-off-by: Theodore Ts'o <tytso@mit.edu>
The partial block zero range helper __ext4_block_zero_page_range()
currently only supports folios of PAGE_SIZE in size. The calculations
for the start block and the offset within a folio for the given range
are incorrect. Modify the implementation to use offset_in_folio()
instead of directly masking PAGE_SIZE - 1, which will be able to support
for large folios.
Signed-off-by: Zhang Yi <yi.zhang@huawei.com>
Link: https://patch.msgid.link/20250512063319.3539411-4-yi.zhang@huaweicloud.com
Signed-off-by: Theodore Ts'o <tytso@mit.edu>
The current buffered write path in ext4 can only allocate and handle
folios of PAGE_SIZE size. To support larger folios, modify
ext4_da_write_begin() and ext4_write_begin() to allocate higher-order
folios, and trim the write length if it exceeds the folio size.
Additionally, in ext4_da_do_write_end(), use offset_in_folio() instead
of PAGE_SIZE.
Signed-off-by: Zhang Yi <yi.zhang@huawei.com>
Link: https://patch.msgid.link/20250512063319.3539411-3-yi.zhang@huaweicloud.com
Signed-off-by: Theodore Ts'o <tytso@mit.edu>
For the extents based inodes, the maxbytes should be sb->s_maxbytes
instead of sbi->s_bitmap_maxbytes. Additionally, for the calculation of
max_end, the -sb->s_blocksize operation is necessary only for
indirect-block based inodes. Correct the maxbytes and max_end value to
correct the behavior of punch hole.
Fixes: 2da376228a ("ext4: limit length to bitmap_maxbytes - blocksize in punch_hole")
Signed-off-by: Zhang Yi <yi.zhang@huawei.com>
Reviewed-by: Jan Kara <jack@suse.cz>
Reviewed-by: Baokun Li <libaokun1@huawei.com>
Link: https://patch.msgid.link/20250506012009.3896990-2-yi.zhang@huaweicloud.com
Signed-off-by: Theodore Ts'o <tytso@mit.edu>
Cc: stable@kernel.org
Since handle->h_transaction may be a NULL pointer, so we should change it
to call is_handle_aborted(handle) first before dereferencing it.
And the following data-race was reported in my fuzzer:
==================================================================
BUG: KCSAN: data-race in jbd2_journal_dirty_metadata / jbd2_journal_dirty_metadata
write to 0xffff888011024104 of 4 bytes by task 10881 on cpu 1:
jbd2_journal_dirty_metadata+0x2a5/0x770 fs/jbd2/transaction.c:1556
__ext4_handle_dirty_metadata+0xe7/0x4b0 fs/ext4/ext4_jbd2.c:358
ext4_do_update_inode fs/ext4/inode.c:5220 [inline]
ext4_mark_iloc_dirty+0x32c/0xd50 fs/ext4/inode.c:5869
__ext4_mark_inode_dirty+0xe1/0x450 fs/ext4/inode.c:6074
ext4_dirty_inode+0x98/0xc0 fs/ext4/inode.c:6103
....
read to 0xffff888011024104 of 4 bytes by task 10880 on cpu 0:
jbd2_journal_dirty_metadata+0xf2/0x770 fs/jbd2/transaction.c:1512
__ext4_handle_dirty_metadata+0xe7/0x4b0 fs/ext4/ext4_jbd2.c:358
ext4_do_update_inode fs/ext4/inode.c:5220 [inline]
ext4_mark_iloc_dirty+0x32c/0xd50 fs/ext4/inode.c:5869
__ext4_mark_inode_dirty+0xe1/0x450 fs/ext4/inode.c:6074
ext4_dirty_inode+0x98/0xc0 fs/ext4/inode.c:6103
....
value changed: 0x00000000 -> 0x00000001
==================================================================
This issue is caused by missing data-race annotation for jh->b_modified.
Therefore, the missing annotation needs to be added.
Reported-by: syzbot+de24c3fe3c4091051710@syzkaller.appspotmail.com
Closes: https://syzkaller.appspot.com/bug?extid=de24c3fe3c4091051710
Fixes: 6e06ae88ed ("jbd2: speedup jbd2_journal_dirty_metadata()")
Signed-off-by: Jeongjun Park <aha310510@gmail.com>
Reviewed-by: Jan Kara <jack@suse.cz>
Link: https://patch.msgid.link/20250514130855.99010-1-aha310510@gmail.com
Signed-off-by: Theodore Ts'o <tytso@mit.edu>
Cc: stable@kernel.org
Luis and David are reporting that after running generic/750 test for 90+
hours on 2k ext4 filesystem, they are able to trigger a warning in
jbd2_journal_dirty_metadata() complaining that there are not enough
credits in the running transaction started in ext4_do_writepages().
Indeed the code in ext4_do_writepages() is racy and the extent tree can
change between the time we compute credits necessary for extent tree
computation and the time we actually modify the extent tree. Thus it may
happen that the number of credits actually needed is higher. Modify
ext4_ext_index_trans_blocks() to count with the worst case of maximum
tree depth. This can reduce the possible number of writers that can
operate in the system in parallel (because the credit estimates now won't
fit in one transaction) but for reasonably sized journals this shouldn't
really be an issue. So just go with a safe and simple fix.
Link: https://lore.kernel.org/all/20250415013641.f2ppw6wov4kn4wq2@offworld
Reported-by: Davidlohr Bueso <dave@stgolabs.net>
Reported-by: Luis Chamberlain <mcgrof@kernel.org>
Tested-by: kdevops@lists.linux.dev
Signed-off-by: Jan Kara <jack@suse.cz>
Reviewed-by: Zhang Yi <yi.zhang@huawei.com>
Link: https://patch.msgid.link/20250429175535.23125-2-jack@suse.cz
Signed-off-by: Theodore Ts'o <tytso@mit.edu>
Cc: stable@kernel.org
check_igot_inode() prints a variable string, which causes a harmless
warning with 'make W=1':
fs/ext4/inode.c:4763:45: error: format string is not a string literal (potentially insecure) [-Werror,-Wformat-security]
4763 | ext4_error_inode(inode, function, line, 0, err_str);
Use a trivial "%s" format string instead.
Signed-off-by: Arnd Bergmann <arnd@arndb.de>
Reviewed-by: Jan Kara <jack@suse.cz>
Link: https://patch.msgid.link/20250423164354.2780635-1-arnd@kernel.org
Signed-off-by: Theodore Ts'o <tytso@mit.edu>
Loading and modifying the extents tree and extent status tree without
holding the inode's i_rwsem or the mapping's invalidate_lock is not
permitted, except during the I/O writeback. Add a new debug helper
ext4_check_map_extents_env(), it will verify whether the extent
loading/modifying context is safe.
Signed-off-by: Zhang Yi <yi.zhang@huawei.com>
Link: https://patch.msgid.link/20250423085257.122685-8-yi.zhang@huaweicloud.com
Signed-off-by: Theodore Ts'o <tytso@mit.edu>
The EXT4_IOC_GET_ES_CACHE and EXT4_IOC_PRECACHE_EXTENTS currently
invokes ext4_ext_precache() to preload the extent cache without holding
the inode's i_rwsem. This can result in stale extent cache entries when
competing with operations such as ext4_collapse_range() which calls
ext4_ext_remove_space() or ext4_ext_shift_extents().
The problem arises when ext4_ext_remove_space() temporarily releases
i_data_sem due to insufficient journal credits. During this interval, a
concurrent EXT4_IOC_GET_ES_CACHE or EXT4_IOC_PRECACHE_EXTENTS may cache
extent entries that are about to be deleted. As a result, these cached
entries become stale and inconsistent with the actual extents.
Loading the extents cache without holding the inode's i_rwsem or the
mapping's invalidate_lock is not permitted besides during the writeback.
Fix this by holding the i_rwsem during EXT4_IOC_GET_ES_CACHE and
EXT4_IOC_PRECACHE_EXTENTS.
Signed-off-by: Zhang Yi <yi.zhang@huawei.com>
Link: https://patch.msgid.link/20250423085257.122685-6-yi.zhang@huaweicloud.com
Signed-off-by: Theodore Ts'o <tytso@mit.edu>
The ext4_fiemap() currently invokes ext4_ext_precache() and
iomap_fiemap() to preload the extent cache and query mapping information
without holding the inode's i_rwsem. This can result in stale extent
cache entries when competing with operations such as
ext4_collapse_range() which calls ext4_ext_remove_space() or
ext4_ext_shift_extents().
The problem arises when ext4_ext_remove_space() temporarily releases
i_data_sem due to insufficient journal credits. During this interval, a
concurrent ext4_fiemap() may cache extent entries that are about to be
deleted. As a result, these cached entries become stale and inconsistent
with the actual extents.
Loading the extents cache without holding the inode's i_rwsem or the
mapping's invalidate_lock is not permitted besides during the writeback.
Fix this by holding the i_rwsem in ext4_fiemap().
Signed-off-by: Zhang Yi <yi.zhang@huawei.com>
Link: https://patch.msgid.link/20250423085257.122685-5-yi.zhang@huaweicloud.com
Signed-off-by: Theodore Ts'o <tytso@mit.edu>
Currently, in the I/O writeback path, ext4_map_blocks() may attempt to
cache additional unrelated extents in the extent status tree without
holding the inode's i_rwsem and the mapping's invalidate_lock. This can
lead to stale extent status entries remaining in certain scenarios,
potentially causing data corruption.
For example, when performing a collapse range in ext4_collapse_range(),
it clears the extent cache and dirty pages before removing blocks and
shifting extents. It also holds the i_data_sem during these two
operations. However, both ext4_ext_remove_space() and
ext4_ext_shift_extents() may briefly release the i_data_sem if journal
credits are insufficient (ext4_datasem_ensure_credits()). If another
writeback process writes dirty pages from other regions during this
interval, it may cache extents that are about to be modified. Unless
ext4_collapse_range() explicitly clears the extent cache again, these
cached entries can become stale and inconsistent with the actual
extents.
0 a n b c m
| | | | | |
[www][wwwwww][wwwwwwww]...[wwwww][wwww]...
| |
N M
Assume that block a is dirty. The collapse range operation is removing
data from n to m and drops i_data_sem immediately after removing the
extent from b to c. At the same time, a concurrent writeback begins to
write back block a; it will reloads the extent from [n, b) into the
extent status tree since it does not hold the i_rwsem or the
invalidate_lock. After the collapse range operation, it left the stale
extent [n, b), which points logical block n to N, but the actual
physical block of n should be M.
Similarly, both ext4_insert_range() and ext4_truncate() have the same
problem. ext4_punch_hole() survived since it re-add a hole extent entry
after removing space since commit 9f1118223a ("ext4: add a hole extent
entry in cache after punch").
In most cases, during dirty page writeback, the block mapping
information is likely to be found in the extent cache, making it less
necessary to search for physical extents. Consequently, loading
unrelated extent caches during writeback appears to be ineffective.
Therefore, fix this by adds EXT4_EX_NOCACHE in the writeback path to
prevent caching of unrelated extents, eliminating this potential source
of corruption.
Signed-off-by: Zhang Yi <yi.zhang@huawei.com>
Link: https://patch.msgid.link/20250423085257.122685-4-yi.zhang@huaweicloud.com
Signed-off-by: Theodore Ts'o <tytso@mit.edu>
Currently, the EXT4_GET_BLOCKS_IO_SUBMIT flag is only used during data
writeback to indicate that in ordered mode, the journal commit thread
should skip re-submitting data and simply wait for I/O completion.
To prepare for later patches that need to detect I/O submission context
in ext4_map_blocks(), generalizes the meaning of
EXT4_GET_BLOCKS_IO_SUBMIT. This flag will be set during:
1) data I/O writeback,
2) I/O completion extents conversion,
3) journal performing commit in fast_commit.
This change doesn't affect current usage of this flag and provides a
clear way to identify I/O submission context.
Signed-off-by: Zhang Yi <yi.zhang@huawei.com>
Link: https://patch.msgid.link/20250423085257.122685-3-yi.zhang@huaweicloud.com
Signed-off-by: Theodore Ts'o <tytso@mit.edu>
When removing space, we should use EXT4_EX_NOCACHE because we don't
need to cache extents, and we should also use EXT4_EX_NOFAIL to prevent
metadata inconsistencies that may arise from memory allocation failures.
While ext4_ext_remove_space() already uses these two flags in most
places, they are missing in ext4_ext_search_right() and
read_extent_tree_block() calls. Unify the flags to ensure consistent
behavior throughout the extent removal process.
Signed-off-by: Zhang Yi <yi.zhang@huawei.com>
Link: https://patch.msgid.link/20250423085257.122685-2-yi.zhang@huaweicloud.com
Signed-off-by: Theodore Ts'o <tytso@mit.edu>
This patch reworks fast commit's commit path to remove locking the
journal for the entire duration of a fast commit. Instead, we only lock
the journal while marking all the eligible inodes as "committing". This
allows handles to make progress in parallel with the fast commit.
Signed-off-by: Harshad Shirwadkar <harshadshirwadkar@gmail.com>
Link: https://patch.msgid.link/20250508175908.1004880-5-harshadshirwadkar@gmail.com
Signed-off-by: Theodore Ts'o <tytso@mit.edu>
If the inode that's being requested to track using ext4_fc_track_inode
is being committed, then wait until the inode finishes the
commit. Also, add calls to ext4_fc_track_inode at the right places.
With this patch, now calling ext4_reserve_inode_write() results in
inode being tracked for next fast commit. This ensures that by the
time ext4_reserve_inode_write() returns, it is ready to be modified
and won't be committed until the corresponding handle is open.
A subtle lock ordering requirement with i_data_sem (which is
documented in the code) requires that ext4_fc_track_inode() be called
before grabbing i_data_sem. So, this patch also adds explicit
ext4_fc_track_inode() calls in places where i_data_sem grabbed.
Signed-off-by: Harshad Shirwadkar <harshadshirwadkar@gmail.com>
Link: https://patch.msgid.link/20250508175908.1004880-3-harshadshirwadkar@gmail.com
Signed-off-by: Theodore Ts'o <tytso@mit.edu>
Pull PCI fixes from Bjorn Helgaas:
- When releasing a start-aligned resource, e.g., a bridge window, save
start/end/flags for the next assignment attempt; fixes a v6.15-rc1
regression (Ilpo Järvinen)
- Move set_pcie_speed.sh from TEST_PROGS to TEST_FILE; fixes a bwctrl
selftest v6.15-rc1 regression (Ilpo Järvinen)
- Add Manivannan Sadhasivam as maintainer of native host bridge and
endpoint drivers (Manivannan Sadhasivam)
- In endpoint test driver, defer IRQ allocation from .probe() until
ioctl() to fix a regression on platforms where the Vendor/Device ID
match doesn't include driver_data (Niklas Cassel)
* tag 'pci-v6.15-fixes-3' of git://git.kernel.org/pub/scm/linux/kernel/git/pci/pci:
misc: pci_endpoint_test: Defer IRQ allocation until ioctl(PCITEST_SET_IRQTYPE)
MAINTAINERS: Move Manivannan Sadhasivam as PCI Native host bridge and endpoint maintainer
selftests/pcie_bwctrl: Fix test progs list
PCI: Restore assigned resources fully after release
Pull nfsd fix from Chuck Lever:
- Revert a v6.15 patch due to a report of SELinux test failures
* tag 'nfsd-6.15-2' of git://git.kernel.org/pub/scm/linux/kernel/git/cel/linux:
Revert "sunrpc: clean cache_detail immediately when flush is written frequently"
Pull misc x86 fixes from Ingo Molnar:
- Fix 32-bit kernel boot crash if passed physical memory with more than
32 address bits
- Fix Xen PV crash
- Work around build bug in certain limited build environments
- Fix CTEST instruction decoding in insn_decoder_test
* tag 'x86-urgent-2025-04-26' of git://git.kernel.org/pub/scm/linux/kernel/git/tip/tip:
x86/insn: Fix CTEST instruction decoding
x86/boot: Work around broken busybox 'truncate' tool
x86/mm: Fix _pgd_alloc() for Xen PV mode
x86/e820: Discard high memory that can't be addressed by 32-bit systems
Pull scheduler fix from Ingo Molnar:
"Fix sporadic crashes in dequeue_entities() due to ... bad math.
[ Arguably if pick_eevdf()/pick_next_entity() was less trusting of
complex math being correct it could have de-escalated a crash into
a warning, but that's for a different patch ]"
* tag 'sched-urgent-2025-04-26' of git://git.kernel.org/pub/scm/linux/kernel/git/tip/tip:
sched/eevdf: Fix se->slice being set to U64_MAX and resulting crash
Pull misc perf events fixes from Ingo Molnar:
- Use POLLERR for events in error state, instead of the ambiguous
POLLHUP error value
- Fix non-sampling (counting) events on certain x86 platforms
* tag 'perf-urgent-2025-04-26' of git://git.kernel.org/pub/scm/linux/kernel/git/tip/tip:
perf/x86: Fix non-sampling (counting) events on certain x86 platforms
perf/core: Change to POLLERR for pinned events with error
Pull irq fix from Ingo Molnar:
"Fix crashes in the gic-v2m irqchip driver, caused by an incorrect
__init annotation"
* tag 'irq-urgent-2025-04-26' of git://git.kernel.org/pub/scm/linux/kernel/git/tip/tip:
irqchip/gic-v2m: Prevent use after free of gicv2m_get_fwnode()
Pull LoongArch fixes from Huacai Chen:
"Add a missing Kconfig option, fix some bugs in exception handlers,
memory management and KVM"
* tag 'loongarch-fixes-6.15-1' of git://git.kernel.org/pub/scm/linux/kernel/git/chenhuacai/linux-loongson:
LoongArch: KVM: Fix PMU pass-through issue if VM exits to host finally
LoongArch: KVM: Fully clear some CSRs when VM reboot
LoongArch: KVM: Fix multiple typos of KVM code
LoongArch: Return NULL from huge_pte_offset() for invalid PMD
LoongArch: Remove a bogus reference to ZONE_DMA
LoongArch: Handle fp, lsx, lasx and lbt assembly symbols
LoongArch: Make do_xyz() exception handlers more robust
LoongArch: Make regs_irqs_disabled() more clear
LoongArch: Select ARCH_USE_MEMTEST
Pull OpenRISC updates from Stafford Horne:
- Support for cacheinfo API to expose OpenRISC cache info via sysfs,
this also translated to some cleanups to OpenRISC cache flush and
invalidate API's
- Documentation updates for new mailing list and toolchain binaries
* tag 'for-linus' of https://github.com/openrisc/linux:
Documentation: openrisc: Update toolchain binaries URL
Documentation: openrisc: Update mailing list
openrisc: Add cacheinfo support
openrisc: Introduce new utility functions to flush and invalidate caches
openrisc: Refactor struct cpuinfo_or1k to reduce duplication
Ondrej reports that certain SELinux tests are failing after commit
fc2a169c56 ("sunrpc: clean cache_detail immediately when flush is
written frequently"), merged during the v6.15 merge window.
Reported-by: Ondrej Mosnacek <omosnace@redhat.com>
Fixes: fc2a169c56 ("sunrpc: clean cache_detail immediately when flush is written frequently")
Signed-off-by: Chuck Lever <chuck.lever@oracle.com>
Pull kunit fix from Kees Cook:
"A single fix for the kunit lib/tests/ relocation:
- Ensure prime numbers tests are included in KUnit test runs (Mark Brown)"
* tag 'move-lib-kunit-v6.15-rc4' of git://git.kernel.org/pub/scm/linux/kernel/git/kees/linux:
lib: Ensure prime numbers tests are included in KUnit test runs
Pull drm fixes from Dave Airlie:
"Weekly drm fixes, mostly amdgpu, with some exynos cleanups and a
couple of minor fixes, seems a bit quiet, but probably some lag from
Easter holidays.
amdgpu:
- P2P DMA fixes
- Display reset fixes
- DCN 3.5 fixes
- ACPI EDID fix
- LTTPR fix
- mode_valid() fix
exynos:
- fix spelling error
- remove redundant error handling in exynos_drm_vidi.c module
- marks struct decon_data as const in the exynos7_drm_decon driver
since it is only read
- Remove unnecessary checking in exynos_drm_drv.c module
meson:
- Fix VCLK calculation
panel:
- jd9365a: Fix reset polarity"
* tag 'drm-fixes-2025-04-26' of https://gitlab.freedesktop.org/drm/kernel:
drm/exynos: Fix spelling mistake "enqueu" -> "enqueue"
drm/exynos: exynos7_drm_decon: Consstify struct decon_data
drm/exynos: fixed a spelling error
drm/exynos/vidi: Remove redundant error handling in vidi_get_modes()
drm/exynos: Remove unnecessary checking
drm/amd/display: do not copy invalid CRTC timing info
drm/amd/display: Default IPS to RCG_IN_ACTIVE_IPS2_IN_OFF
drm/amd/display: Use 16ms AUX read interval for LTTPR with old sinks
drm/amd/display: Fix ACPI edid parsing on some Lenovo systems
drm/amdgpu: Allow P2P access through XGMI
drm/amd/display: Enable urgent latency adjustment on DCN35
drm/amd/display: Force full update in gpu reset
drm/amd/display: Fix gpu reset in multidisplay config
drm/amdgpu: Don't pin VRAM without DMABUF_MOVE_NOTIFY
drm/amdgpu: Use allowed_domains for pinning dmabufs
drm: panel: jd9365da: fix reset signal polarity in unprepare
drm/meson: use unsigned long long / Hz for frequency types
Revert "drm/meson: vclk: fix calculation of 59.94 fractional rates"
There is a code path in dequeue_entities() that can set the slice of a
sched_entity to U64_MAX, which sometimes results in a crash.
The offending case is when dequeue_entities() is called to dequeue a
delayed group entity, and then the entity's parent's dequeue is delayed.
In that case:
1. In the if (entity_is_task(se)) else block at the beginning of
dequeue_entities(), slice is set to
cfs_rq_min_slice(group_cfs_rq(se)). If the entity was delayed, then
it has no queued tasks, so cfs_rq_min_slice() returns U64_MAX.
2. The first for_each_sched_entity() loop dequeues the entity.
3. If the entity was its parent's only child, then the next iteration
tries to dequeue the parent.
4. If the parent's dequeue needs to be delayed, then it breaks from the
first for_each_sched_entity() loop _without updating slice_.
5. The second for_each_sched_entity() loop sets the parent's ->slice to
the saved slice, which is still U64_MAX.
This throws off subsequent calculations with potentially catastrophic
results. A manifestation we saw in production was:
6. In update_entity_lag(), se->slice is used to calculate limit, which
ends up as a huge negative number.
7. limit is used in se->vlag = clamp(vlag, -limit, limit). Because limit
is negative, vlag > limit, so se->vlag is set to the same huge
negative number.
8. In place_entity(), se->vlag is scaled, which overflows and results in
another huge (positive or negative) number.
9. The adjusted lag is subtracted from se->vruntime, which increases or
decreases se->vruntime by a huge number.
10. pick_eevdf() calls entity_eligible()/vruntime_eligible(), which
incorrectly returns false because the vruntime is so far from the
other vruntimes on the queue, causing the
(vruntime - cfs_rq->min_vruntime) * load calulation to overflow.
11. Nothing appears to be eligible, so pick_eevdf() returns NULL.
12. pick_next_entity() tries to dereference the return value of
pick_eevdf() and crashes.
Dumping the cfs_rq states from the core dumps with drgn showed tell-tale
huge vruntime ranges and bogus vlag values, and I also traced se->slice
being set to U64_MAX on live systems (which was usually "benign" since
the rest of the runqueue needed to be in a particular state to crash).
Fix it in dequeue_entities() by always setting slice from the first
non-empty cfs_rq.
Fixes: aef6987d89 ("sched/eevdf: Propagate min_slice up the cgroup hierarchy")
Signed-off-by: Omar Sandoval <osandov@fb.com>
Signed-off-by: Peter Zijlstra (Intel) <peterz@infradead.org>
Signed-off-by: Ingo Molnar <mingo@kernel.org>
Link: https://lkml.kernel.org/r/f0c2d1072be229e1bdddc73c0703919a8b00c652.1745570998.git.osandov@fb.com
With ACPI in place, gicv2m_get_fwnode() is registered with the pci
subsystem as pci_msi_get_fwnode_cb(), which may get invoked at runtime
during a PCI host bridge probe. But, the call back is wrongly marked as
__init, causing it to be freed, while being registered with the PCI
subsystem and could trigger:
Unable to handle kernel paging request at virtual address ffff8000816c0400
gicv2m_get_fwnode+0x0/0x58 (P)
pci_set_bus_msi_domain+0x74/0x88
pci_register_host_bridge+0x194/0x548
This is easily reproducible on a Juno board with ACPI boot.
Retain the function for later use.
Fixes: 0644b3daca ("irqchip/gic-v2m: acpi: Introducing GICv2m ACPI support")
Signed-off-by: Suzuki K Poulose <suzuki.poulose@arm.com>
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
Signed-off-by: Ingo Molnar <mingo@kernel.org>
Reviewed-by: Marc Zyngier <maz@kernel.org>
Cc: stable@vger.kernel.org
