cpu-clock usage by the async-profiler tool can trigger a system hang,
which got bisected back to the following commit by Octavia Togami:
18dbcbfabf ("perf: Fix the POLL_HUP delivery breakage") causes this issue
The root cause of the hang is that cpu-clock is a special type of SW
event which relies on hrtimers. The __perf_event_overflow() callback
is invoked from the hrtimer handler for cpu-clock events, and
__perf_event_overflow() tries to call cpu_clock_event_stop()
to stop the event, which calls htimer_cancel() to cancel the hrtimer.
But that's a recursion into the hrtimer code from a hrtimer handler,
which (unsurprisingly) deadlocks.
To fix this bug, use hrtimer_try_to_cancel() instead, and set
the PERF_HES_STOPPED flag, which causes perf_swevent_hrtimer()
to stop the event once it sees the PERF_HES_STOPPED flag.
[ mingo: Fixed the comments and improved the changelog. ]
Closes: https://lore.kernel.org/all/CAHPNGSQpXEopYreir+uDDEbtXTBvBvi8c6fYXJvceqtgTPao3Q@mail.gmail.com/
Fixes: 18dbcbfabf ("perf: Fix the POLL_HUP delivery breakage")
Reported-by: Octavia Togami <octavia.togami@gmail.com>
Suggested-by: Peter Zijlstra <peterz@infradead.org>
Signed-off-by: Dapeng Mi <dapeng1.mi@linux.intel.com>
Signed-off-by: Peter Zijlstra (Intel) <peterz@infradead.org>
Signed-off-by: Ingo Molnar <mingo@kernel.org>
Tested-by: Octavia Togami <octavia.togami@gmail.com>
Cc: stable@vger.kernel.org
Link: https://github.com/lucko/spark/issues/530
Link: https://patch.msgid.link/20251015051828.12809-1-dapeng1.mi@linux.intel.com
Some file systems like FUSE-based ones or overlayfs may record the backing
file in struct vm_area_struct vm_file, instead of the user file that the
user mmapped.
That causes perf to misreport the device major/minor numbers of the file
system of the file, and the generation of the file, and potentially other
inode details. There is an existing helper file_user_inode() for that
situation.
Use file_user_inode() instead of file_inode() to get the inode for MMAP2
events.
Example:
Setup:
# cd /root
# mkdir test ; cd test ; mkdir lower upper work merged
# cp `which cat` lower
# mount -t overlay overlay -olowerdir=lower,upperdir=upper,workdir=work merged
# perf record -e cycles:u -- /root/test/merged/cat /proc/self/maps
...
55b2c91d0000-55b2c926b000 r-xp 00018000 00:1a 3419 /root/test/merged/cat
...
[ perf record: Woken up 1 times to write data ]
[ perf record: Captured and wrote 0.004 MB perf.data (5 samples) ]
#
# stat /root/test/merged/cat
File: /root/test/merged/cat
Size: 1127792 Blocks: 2208 IO Block: 4096 regular file
Device: 0,26 Inode: 3419 Links: 1
Access: (0755/-rwxr-xr-x) Uid: ( 0/ root) Gid: ( 0/ root)
Access: 2025-09-08 12:23:59.453309624 +0000
Modify: 2025-09-08 12:23:59.454309624 +0000
Change: 2025-09-08 12:23:59.454309624 +0000
Birth: 2025-09-08 12:23:59.453309624 +0000
Before:
Device reported 00:02 differs from stat output and /proc/self/maps
# perf script --show-mmap-events | grep /root/test/merged/cat
cat 377 [-01] 243.078558: PERF_RECORD_MMAP2 377/377: [0x55b2c91d0000(0x9b000) @ 0x18000 00:02 3419 2068525940]: r-xp /root/test/merged/cat
After:
Device reported 00:1a is the same as stat output and /proc/self/maps
# perf script --show-mmap-events | grep /root/test/merged/cat
cat 362 [-01] 127.755167: PERF_RECORD_MMAP2 362/362: [0x55ba6e781000(0x9b000) @ 0x18000 00:1a 3419 0]: r-xp /root/test/merged/cat
With respect to stable kernels, overlayfs mmap function ovl_mmap() was
added in v4.19 but file_user_inode() was not added until v6.8 and never
back-ported to stable kernels. FMODE_BACKING that it depends on was added
in v6.5. This issue has gone largely unnoticed, so back-porting before
v6.8 is probably not worth it, so put 6.8 as the stable kernel prerequisite
version, although in practice the next long term kernel is 6.12.
Signed-off-by: Adrian Hunter <adrian.hunter@intel.com>
Signed-off-by: Peter Zijlstra (Intel) <peterz@infradead.org>
Acked-by: Amir Goldstein <amir73il@gmail.com>
Cc: stable@vger.kernel.org # 6.8
Some file systems like FUSE-based ones or overlayfs may record the backing
file in struct vm_area_struct vm_file, instead of the user file that the
user mmapped.
Since commit def3ae83da ("fs: store real path instead of fake path in
backing file f_path"), file_path() no longer returns the user file path
when applied to a backing file. There is an existing helper
file_user_path() for that situation.
Use file_user_path() instead of file_path() to get the path for MMAP
and MMAP2 events.
Example:
Setup:
# cd /root
# mkdir test ; cd test ; mkdir lower upper work merged
# cp `which cat` lower
# mount -t overlay overlay -olowerdir=lower,upperdir=upper,workdir=work merged
# perf record -e intel_pt//u -- /root/test/merged/cat /proc/self/maps
...
55b0ba399000-55b0ba434000 r-xp 00018000 00:1a 3419 /root/test/merged/cat
...
[ perf record: Woken up 1 times to write data ]
[ perf record: Captured and wrote 0.060 MB perf.data ]
#
Before:
File name is wrong (/cat), so decoding fails:
# perf script --no-itrace --show-mmap-events
cat 367 [016] 100.491492: PERF_RECORD_MMAP2 367/367: [0x55b0ba399000(0x9b000) @ 0x18000 00:02 3419 489959280]: r-xp /cat
...
# perf script --itrace=e | wc -l
Warning:
19 instruction trace errors
19
#
After:
File name is correct (/root/test/merged/cat), so decoding is ok:
# perf script --no-itrace --show-mmap-events
cat 364 [016] 72.153006: PERF_RECORD_MMAP2 364/364: [0x55ce4003d000(0x9b000) @ 0x18000 00:02 3419 3132534314]: r-xp /root/test/merged/cat
# perf script --itrace=e
# perf script --itrace=e | wc -l
0
#
Fixes: def3ae83da ("fs: store real path instead of fake path in backing file f_path")
Signed-off-by: Adrian Hunter <adrian.hunter@intel.com>
Signed-off-by: Peter Zijlstra (Intel) <peterz@infradead.org>
Acked-by: Amir Goldstein <amir73il@gmail.com>
Cc: stable@vger.kernel.org
It was reported that Intel PT address filters do not work in Docker
containers. That relates to the use of overlayfs.
overlayfs records the backing file in struct vm_area_struct vm_file,
instead of the user file that the user mmapped. In order for an address
filter to match, it must compare to the user file inode. There is an
existing helper file_user_inode() for that situation.
Use file_user_inode() instead of file_inode() to get the inode for address
filter matching.
Example:
Setup:
# cd /root
# mkdir test ; cd test ; mkdir lower upper work merged
# cp `which cat` lower
# mount -t overlay overlay -olowerdir=lower,upperdir=upper,workdir=work merged
# perf record --buildid-mmap -e intel_pt//u --filter 'filter * @ /root/test/merged/cat' -- /root/test/merged/cat /proc/self/maps
...
55d61d246000-55d61d2e1000 r-xp 00018000 00:1a 3418 /root/test/merged/cat
...
[ perf record: Woken up 1 times to write data ]
[ perf record: Captured and wrote 0.015 MB perf.data ]
# perf buildid-cache --add /root/test/merged/cat
Before:
Address filter does not match so there are no control flow packets
# perf script --itrace=e
# perf script --itrace=b | wc -l
0
# perf script -D | grep 'TIP.PGE' | wc -l
0
#
After:
Address filter does match so there are control flow packets
# perf script --itrace=e
# perf script --itrace=b | wc -l
235
# perf script -D | grep 'TIP.PGE' | wc -l
57
#
With respect to stable kernels, overlayfs mmap function ovl_mmap() was
added in v4.19 but file_user_inode() was not added until v6.8 and never
back-ported to stable kernels. FMODE_BACKING that it depends on was added
in v6.5. This issue has gone largely unnoticed, so back-porting before
v6.8 is probably not worth it, so put 6.8 as the stable kernel prerequisite
version, although in practice the next long term kernel is 6.12.
Closes: https://lore.kernel.org/linux-perf-users/aBCwoq7w8ohBRQCh@fremen.lan
Reported-by: Edd Barrett <edd@theunixzoo.co.uk>
Signed-off-by: Adrian Hunter <adrian.hunter@intel.com>
Signed-off-by: Peter Zijlstra (Intel) <peterz@infradead.org>
Acked-by: Amir Goldstein <amir73il@gmail.com>
Cc: stable@vger.kernel.org # 6.8
Pull driver core updates from Danilo Krummrich:
"Auxiliary:
- Drop call to dev_pm_domain_detach() in auxiliary_bus_probe()
- Optimize logic of auxiliary_match_id()
Rust:
- Auxiliary:
- Use primitive C types from prelude
- DebugFs:
- Add debugfs support for simple read/write files and custom
callbacks through a File-type-based and directory-scope-based
API
- Sample driver code for the File-type-based API
- Sample module code for the directory-scope-based API
- I/O:
- Add io::poll module and implement Rust specific
read_poll_timeout() helper
- IRQ:
- Implement support for threaded and non-threaded device IRQs
based on (&Device<Bound>, IRQ number) tuples (IrqRequest)
- Provide &Device<Bound> cookie in IRQ handlers
- PCI:
- Support IRQ requests from IRQ vectors for a specific
pci::Device<Bound>
- Implement accessors for subsystem IDs, revision, devid and
resource start
- Provide dedicated pci::Vendor and pci::Class types for vendor
and class ID numbers
- Implement Display to print actual vendor and class names; Debug
to print the raw ID numbers
- Add pci::DeviceId::from_class_and_vendor() helper
- Use primitive C types from prelude
- Various minor inline and (safety) comment improvements
- Platform:
- Support IRQ requests from IRQ vectors for a specific
platform::Device<Bound>
- Nova:
- Use pci::DeviceId::from_class_and_vendor() to avoid probing
non-display/compute PCI functions
- Misc:
- Add helper for cpu_relax()
- Update ARef import from sync::aref
sysfs:
- Remove bin_attrs_new field from struct attribute_group
- Remove read_new() and write_new() from struct bin_attribute
Misc:
- Document potential race condition in get_dev_from_fwnode()
- Constify node_group argument in software node registration
functions
- Fix order of kernel-doc parameters in various functions
- Set power.no_pm flag for faux devices
- Set power.no_callbacks flag along with the power.no_pm flag
- Constify the pmu_bus bus type
- Minor spelling fixes"
* tag 'driver-core-6.18-rc1' of git://git.kernel.org/pub/scm/linux/kernel/git/driver-core/driver-core: (43 commits)
rust: pci: display symbolic PCI vendor names
rust: pci: display symbolic PCI class names
rust: pci: fix incorrect platform reference in PCI driver probe doc comment
rust: pci: fix incorrect platform reference in PCI driver unbind doc comment
perf: make pmu_bus const
samples: rust: Add scoped debugfs sample driver
rust: debugfs: Add support for scoped directories
samples: rust: Add debugfs sample driver
rust: debugfs: Add support for callback-based files
rust: debugfs: Add support for writable files
rust: debugfs: Add support for read-only files
rust: debugfs: Add initial support for directories
driver core: auxiliary bus: Optimize logic of auxiliary_match_id()
driver core: auxiliary bus: Drop dev_pm_domain_detach() call
driver core: Fix order of the kernel-doc parameters
driver core: get_dev_from_fwnode(): document potential race
drivers: base: fix "publically"->"publicly"
driver core/PM: Set power.no_callbacks along with power.no_pm
driver core: faux: Set power.no_pm for faux devices
rust: pci: inline several tiny functions
...
Pull bpf updates from Alexei Starovoitov:
- Support pulling non-linear xdp data with bpf_xdp_pull_data() kfunc
(Amery Hung)
Applied as a stable branch in bpf-next and net-next trees.
- Support reading skb metadata via bpf_dynptr (Jakub Sitnicki)
Also a stable branch in bpf-next and net-next trees.
- Enforce expected_attach_type for tailcall compatibility (Daniel
Borkmann)
- Replace path-sensitive with path-insensitive live stack analysis in
the verifier (Eduard Zingerman)
This is a significant change in the verification logic. More details,
motivation, long term plans are in the cover letter/merge commit.
- Support signed BPF programs (KP Singh)
This is another major feature that took years to materialize.
Algorithm details are in the cover letter/marge commit
- Add support for may_goto instruction to s390 JIT (Ilya Leoshkevich)
- Add support for may_goto instruction to arm64 JIT (Puranjay Mohan)
- Fix USDT SIB argument handling in libbpf (Jiawei Zhao)
- Allow uprobe-bpf program to change context registers (Jiri Olsa)
- Support signed loads from BPF arena (Kumar Kartikeya Dwivedi and
Puranjay Mohan)
- Allow access to union arguments in tracing programs (Leon Hwang)
- Optimize rcu_read_lock() + migrate_disable() combination where it's
used in BPF subsystem (Menglong Dong)
- Introduce bpf_task_work_schedule*() kfuncs to schedule deferred
execution of BPF callback in the context of a specific task using the
kernel’s task_work infrastructure (Mykyta Yatsenko)
- Enforce RCU protection for KF_RCU_PROTECTED kfuncs (Kumar Kartikeya
Dwivedi)
- Add stress test for rqspinlock in NMI (Kumar Kartikeya Dwivedi)
- Improve the precision of tnum multiplier verifier operation
(Nandakumar Edamana)
- Use tnums to improve is_branch_taken() logic (Paul Chaignon)
- Add support for atomic operations in arena in riscv JIT (Pu Lehui)
- Report arena faults to BPF error stream (Puranjay Mohan)
- Search for tracefs at /sys/kernel/tracing first in bpftool (Quentin
Monnet)
- Add bpf_strcasecmp() kfunc (Rong Tao)
- Support lookup_and_delete_elem command in BPF_MAP_STACK_TRACE (Tao
Chen)
* tag 'bpf-next-6.18' of git://git.kernel.org/pub/scm/linux/kernel/git/bpf/bpf-next: (197 commits)
libbpf: Replace AF_ALG with open coded SHA-256
selftests/bpf: Add stress test for rqspinlock in NMI
selftests/bpf: Add test case for different expected_attach_type
bpf: Enforce expected_attach_type for tailcall compatibility
bpftool: Remove duplicate string.h header
bpf: Remove duplicate crypto/sha2.h header
libbpf: Fix error when st-prefix_ops and ops from differ btf
selftests/bpf: Test changing packet data from kfunc
selftests/bpf: Add stacktrace map lookup_and_delete_elem test case
selftests/bpf: Refactor stacktrace_map case with skeleton
bpf: Add lookup_and_delete_elem for BPF_MAP_STACK_TRACE
selftests/bpf: Fix flaky bpf_cookie selftest
selftests/bpf: Test changing packet data from global functions with a kfunc
bpf: Emit struct bpf_xdp_sock type in vmlinux BTF
selftests/bpf: Task_work selftest cleanup fixes
MAINTAINERS: Delete inactive maintainers from AF_XDP
bpf: Mark kfuncs as __noclone
selftests/bpf: Add kprobe multi write ctx attach test
selftests/bpf: Add kprobe write ctx attach test
selftests/bpf: Add uprobe context ip register change test
...
Pull performance events updates from Ingo Molnar:
"Core perf code updates:
- Convert mmap() related reference counts to refcount_t. This is in
reaction to the recently fixed refcount bugs, which could have been
detected earlier and could have mitigated the bug somewhat (Thomas
Gleixner, Peter Zijlstra)
- Clean up and simplify the callchain code, in preparation for
sframes (Steven Rostedt, Josh Poimboeuf)
Uprobes updates:
- Add support to optimize usdt probes on x86-64, which gives a
substantial speedup (Jiri Olsa)
- Cleanups and fixes on x86 (Peter Zijlstra)
PMU driver updates:
- Various optimizations and fixes to the Intel PMU driver (Dapeng Mi)
Misc cleanups and fixes:
- Remove redundant __GFP_NOWARN (Qianfeng Rong)"
* tag 'perf-core-2025-09-26' of git://git.kernel.org/pub/scm/linux/kernel/git/tip/tip: (57 commits)
selftests/bpf: Fix uprobe_sigill test for uprobe syscall error value
uprobes/x86: Return error from uprobe syscall when not called from trampoline
perf: Skip user unwind if the task is a kernel thread
perf: Simplify get_perf_callchain() user logic
perf: Use current->flags & PF_KTHREAD|PF_USER_WORKER instead of current->mm == NULL
perf: Have get_perf_callchain() return NULL if crosstask and user are set
perf: Remove get_perf_callchain() init_nr argument
perf/x86: Print PMU counters bitmap in x86_pmu_show_pmu_cap()
perf/x86/intel: Add ICL_FIXED_0_ADAPTIVE bit into INTEL_FIXED_BITS_MASK
perf/x86/intel: Change macro GLOBAL_CTRL_EN_PERF_METRICS to BIT_ULL(48)
perf/x86: Add PERF_CAP_PEBS_TIMING_INFO flag
perf/x86/intel: Fix IA32_PMC_x_CFG_B MSRs access error
perf/x86/intel: Use early_initcall() to hook bts_init()
uprobes: Remove redundant __GFP_NOWARN
selftests/seccomp: validate uprobe syscall passes through seccomp
seccomp: passthrough uprobe systemcall without filtering
selftests/bpf: Fix uprobe syscall shadow stack test
selftests/bpf: Change test_uretprobe_regs_change for uprobe and uretprobe
selftests/bpf: Add uprobe_regs_equal test
selftests/bpf: Add optimized usdt variant for basic usdt test
...
Currently uprobe (BPF_PROG_TYPE_KPROBE) program can't write to the
context registers data. While this makes sense for kprobe attachments,
for uprobe attachment it might make sense to be able to change user
space registers to alter application execution.
Since uprobe and kprobe programs share the same type (BPF_PROG_TYPE_KPROBE),
we can't deny write access to context during the program load. We need
to check on it during program attachment to see if it's going to be
kprobe or uprobe.
Storing the program's write attempt to context and checking on it
during the attachment.
Acked-by: Andrii Nakryiko <andrii@kernel.org>
Signed-off-by: Jiri Olsa <jolsa@kernel.org>
Link: https://lore.kernel.org/r/20250916215301.664963-2-jolsa@kernel.org
Signed-off-by: Alexei Starovoitov <ast@kernel.org>
To determine if a task is a kernel thread or not, it is more reliable to
use (current->flags & (PF_KTHREAD|PF_USER_WORKERi)) than to rely on
current->mm being NULL. That is because some kernel tasks (io_uring
helpers) may have a mm field.
Signed-off-by: Steven Rostedt (Google) <rostedt@goodmis.org>
Signed-off-by: Peter Zijlstra (Intel) <peterz@infradead.org>
Link: https://lore.kernel.org/r/20250820180428.592367294@kernel.org
The recently fixed reference count leaks could have been detected by using
refcount_t and refcount_t would have mitigated the potential overflow at
least.
Now that the code is properly structured, convert the mmap() related
mmap_count variants over to refcount_t.
No functional change intended.
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
Signed-off-by: Peter Zijlstra (Intel) <peterz@infradead.org>
Reviewed-by: Lorenzo Stoakes <lorenzo.stoakes@oracle.com>
Link: https://lore.kernel.org/r/20250812104020.071507932@infradead.org
Needed because refcount_inc() doesn't allow the 0->1 transition.
Specifically, this is the case where we've created the RB, this means
there was no RB, and as such there could not have been an mmap.
Additionally we hold mmap_mutex to serialize everything.
This must be the first.
Suggested-by: Thomas Gleixner <tglx@linutronix.de>
Signed-off-by: Peter Zijlstra (Intel) <peterz@infradead.org>
Link: https://lore.kernel.org/r/20250812104019.956479989@infradead.org
if (cond) {
A;
} else {
B;
}
C;
into
if (cond) {
A;
C;
} else {
B;
C;
}
Notably C has a success branch and both A and B have two places for
success. For A (rb case), duplicate the success case because later
patches will result in them no longer being identical. For B (aux
case), share using goto (cleaned up later).
Suggested-by: Thomas Gleixner <tglx@linutronix.de>
Signed-off-by: Peter Zijlstra (Intel) <peterz@infradead.org>
Reviewed-by: Lorenzo Stoakes <lorenzo.stoakes@oracle.com>
Link: https://lore.kernel.org/r/20250812104019.016252852@infradead.org
It is already checked whether the VMA size is the same as
nr_pages * PAGE_SIZE, so later checking both:
aux_size == vma_size && aux_size == nr_pages * PAGE_SIZE
is redundant. Remove the vma_size check as nr_pages is what is actually
used in the allocation function. That prepares for splitting out the buffer
allocation into separate functions, so that only nr_pages needs to be
handed in.
No functional change.
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
Signed-off-by: Peter Zijlstra (Intel) <peterz@infradead.org>
Reviewed-by: Lorenzo Stoakes <lorenzo.stoakes@oracle.com>
Link: https://lore.kernel.org/r/20250812104018.424519320@infradead.org
Calling pmu->start()/stop() on perf events in PERF_EVENT_STATE_OFF can
leave event->hw.idx at -1. When PMU drivers later attempt to use this
negative index as a shift exponent in bitwise operations, it leads to UBSAN
shift-out-of-bounds reports.
The issue is a logical flaw in how event groups handle throttling when some
members are intentionally disabled. Based on the analysis and the
reproducer provided by Mark Rutland (this issue on both arm64 and x86-64).
The scenario unfolds as follows:
1. A group leader event is configured with a very aggressive sampling
period (e.g., sample_period = 1). This causes frequent interrupts and
triggers the throttling mechanism.
2. A child event in the same group is created in a disabled state
(.disabled = 1). This event remains in PERF_EVENT_STATE_OFF.
Since it hasn't been scheduled onto the PMU, its event->hw.idx remains
initialized at -1.
3. When throttling occurs, perf_event_throttle_group() and later
perf_event_unthrottle_group() iterate through all siblings, including
the disabled child event.
4. perf_event_throttle()/unthrottle() are called on this inactive child
event, which then call event->pmu->start()/stop().
5. The PMU driver receives the event with hw.idx == -1 and attempts to
use it as a shift exponent. e.g., in macros like PMCNTENSET(idx),
leading to the UBSAN report.
The throttling mechanism attempts to start/stop events that are not
actively scheduled on the hardware.
Move the state check into perf_event_throttle()/perf_event_unthrottle() so
that inactive events are skipped entirely. This ensures only active events
with a valid hw.idx are processed, preventing undefined behavior and
silencing UBSAN warnings. The corrected check ensures true before
proceeding with PMU operations.
The problem can be reproduced with the syzkaller reproducer:
Fixes: 9734e25fbf ("perf: Fix the throttle logic for a group")
Signed-off-by: Yunseong Kim <ysk@kzalloc.com>
Signed-off-by: Peter Zijlstra (Intel) <peterz@infradead.org>
Reviewed-by: Kan Liang <kan.liang@linux.intel.com>
Link: https://lore.kernel.org/r/20250812181046.292382-2-ysk@kzalloc.com
The perf mmap code is careful about mmap()'ing the user page with the
ringbuffer and additionally the auxiliary buffer, when the event supports
it. Once the first mapping is established, subsequent mapping have to use
the same offset and the same size in both cases. The reference counting for
the ringbuffer and the auxiliary buffer depends on this being correct.
Though perf does not prevent that a related mapping is split via mmap(2),
munmap(2) or mremap(2). A split of a VMA results in perf_mmap_open() calls,
which take reference counts, but then the subsequent perf_mmap_close()
calls are not longer fulfilling the offset and size checks. This leads to
reference count leaks.
As perf already has the requirement for subsequent mappings to match the
initial mapping, the obvious consequence is that VMA splits, caused by
resizing of a mapping or partial unmapping, have to be prevented.
Implement the vm_operations_struct::may_split() callback and return
unconditionally -EINVAL.
That ensures that the mapping offsets and sizes cannot be changed after the
fact. Remapping to a different fixed address with the same size is still
possible as it takes the references for the new mapping and drops those of
the old mapping.
Fixes: 45bfb2e504 ("perf: Add AUX area to ring buffer for raw data streams")
Reported-by: zdi-disclosures@trendmicro.com # ZDI-CAN-27504
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
Reviewed-by: Lorenzo Stoakes <lorenzo.stoakes@oracle.com>
Acked-by: Arnaldo Carvalho de Melo <acme@redhat.com>
Acked-by: Vlastimil Babka <vbabka@suse.cz>
Cc: stable@vger.kernel.org
After successful allocation of a buffer or a successful attachment to an
existing buffer perf_mmap() tries to map the buffer read only into the page
table. If that fails, the already set up page table entries are zapped, but
the other perf specific side effects of that failure are not handled. The
calling code just cleans up the VMA and does not invoke perf_mmap_close().
This leaks reference counts, corrupts user->vm accounting and also results
in an unbalanced invocation of event::event_mapped().
Cure this by moving the event::event_mapped() invocation before the
map_range() call so that on map_range() failure perf_mmap_close() can be
invoked without causing an unbalanced event::event_unmapped() call.
perf_mmap_close() undoes the reference counts and eventually frees buffers.
Fixes: b709eb872e ("perf: map pages in advance")
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
Reviewed-by: Lorenzo Stoakes <lorenzo.stoakes@oracle.com>
Cc: stable@vger.kernel.org
When perf_mmap() fails to allocate a buffer, it still invokes the
event_mapped() callback of the related event. On X86 this might increase
the perf_rdpmc_allowed reference counter. But nothing undoes this as
perf_mmap_close() is never called in this case, which causes another
reference count leak.
Return early on failure to prevent that.
Fixes: 1e0fb9ec67 ("perf: Add pmu callbacks to track event mapping and unmapping")
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
Reviewed-by: Lorenzo Stoakes <lorenzo.stoakes@oracle.com>
Cc: stable@vger.kernel.org
Failure of the AUX buffer allocation leaks the reference count.
Set the reference count to 1 only when the allocation succeeds.
Fixes: 45bfb2e504 ("perf: Add AUX area to ring buffer for raw data streams")
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
Reviewed-by: Lorenzo Stoakes <lorenzo.stoakes@oracle.com>
Cc: stable@vger.kernel.org
A recent overhaul sets the return value to 0 unconditionally after the
allocations, which causes reference count leaks and corrupts the user->vm
accounting.
Preserve the AUX buffer allocation failure return value, so that the
subsequent code works correctly.
Fixes: 0983593f32 ("perf/core: Lift event->mmap_mutex in perf_mmap()")
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
Reviewed-by: Lorenzo Stoakes <lorenzo.stoakes@oracle.com>
Cc: stable@vger.kernel.org
Jann reports that uprobes can be used destructively when used in the
middle of an instruction. The kernel only verifies there is a valid
instruction at the requested offset, but due to variable instruction
length cannot determine if this is an instruction as seen by the
intended execution stream.
Additionally, Mark Rutland notes that on architectures that mix data
in the text segment (like arm64), a similar things can be done if the
data word is 'mistaken' for an instruction.
As such, require CAP_SYS_ADMIN for uprobes.
Fixes: c9e0924e5c ("perf/core: open access to probes for CAP_PERFMON privileged process")
Reported-by: Jann Horn <jannh@google.com>
Signed-off-by: Peter Zijlstra (Intel) <peterz@infradead.org>
Link: https://lkml.kernel.org/r/CAG48ez1n4520sq0XrWYDHKiKxE_+WCfAK+qt9qkY4ZiBGmL-5g@mail.gmail.com
If an AUX event overruns, the event core layer intends to disable the
event by setting the 'pending_disable' flag. Unfortunately, the event
is not actually disabled afterwards.
In commit:
ca6c21327c ("perf: Fix missing SIGTRAPs")
the 'pending_disable' flag was changed to a boolean. However, the
AUX event code was not updated accordingly. The flag ends up holding a
CPU number. If this number is zero, the flag is taken as false and the
IRQ work is never triggered.
Later, with commit:
2b84def990 ("perf: Split __perf_pending_irq() out of perf_pending_irq()")
a new IRQ work 'pending_disable_irq' was introduced to handle event
disabling. The AUX event path was not updated to kick off the work queue.
To fix this bug, when an AUX ring buffer overrun is detected, call
perf_event_disable_inatomic() to initiate the pending disable flow.
Also update the outdated comment for setting the flag, to reflect the
boolean values (0 or 1).
Fixes: 2b84def990 ("perf: Split __perf_pending_irq() out of perf_pending_irq()")
Fixes: ca6c21327c ("perf: Fix missing SIGTRAPs")
Signed-off-by: Leo Yan <leo.yan@arm.com>
Signed-off-by: Ingo Molnar <mingo@kernel.org>
Reviewed-by: James Clark <james.clark@linaro.org>
Reviewed-by: Yeoreum Yun <yeoreum.yun@arm.com>
Cc: Adrian Hunter <adrian.hunter@intel.com>
Cc: Alexander Shishkin <alexander.shishkin@linux.intel.com>
Cc: Arnaldo Carvalho de Melo <acme@kernel.org>
Cc: Ian Rogers <irogers@google.com>
Cc: Jiri Olsa <jolsa@redhat.com>
Cc: Liang Kan <kan.liang@linux.intel.com>
Cc: Marco Elver <elver@google.com>
Cc: Mark Rutland <mark.rutland@arm.com>
Cc: Namhyung Kim <namhyung@kernel.org>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Sebastian Andrzej Siewior <bigeasy@linutronix.de>
Cc: linux-perf-users@vger.kernel.org
Link: https://lore.kernel.org/r/20250625170737.2918295-1-leo.yan@arm.com
There may be concurrency between perf_cgroup_switch and
perf_cgroup_event_disable. Consider the following scenario: after a new
perf cgroup event is created on CPU0, the new event may not trigger
a reprogramming, causing ctx->is_active to be 0. In this case, when CPU1
disables this perf event, it executes __perf_remove_from_context->
list _del_event->perf_cgroup_event_disable on CPU1, which causes a race
with perf_cgroup_switch running on CPU0.
The following describes the details of this concurrency scenario:
CPU0 CPU1
perf_cgroup_switch:
...
# cpuctx->cgrp is not NULL here
if (READ_ONCE(cpuctx->cgrp) == NULL)
return;
perf_remove_from_context:
...
raw_spin_lock_irq(&ctx->lock);
...
# ctx->is_active == 0 because reprogramm is not
# tigger, so CPU1 can do __perf_remove_from_context
# for CPU0
__perf_remove_from_context:
perf_cgroup_event_disable:
...
if (--ctx->nr_cgroups)
...
# this warning will happened because CPU1 changed
# ctx.nr_cgroups to 0.
WARN_ON_ONCE(cpuctx->ctx.nr_cgroups == 0);
[peterz: use guard instead of goto unlock]
Fixes: db4a835601 ("perf/core: Set cgroup in CPU contexts for new cgroup events")
Signed-off-by: Luo Gengkun <luogengkun@huaweicloud.com>
Signed-off-by: Peter Zijlstra (Intel) <peterz@infradead.org>
Link: https://lkml.kernel.org/r/20250604033924.3914647-3-luogengkun@huaweicloud.com
Commit a3c3c6667("perf/core: Fix child_total_time_enabled accounting
bug at task exit") moves the event->state update to before
list_del_event(). This makes the event->state test in list_del_event()
always false; never calling perf_cgroup_event_disable().
As a result, cpuctx->cgrp won't be cleared properly; causing havoc.
Fixes: a3c3c6667("perf/core: Fix child_total_time_enabled accounting bug at task exit")
Signed-off-by: Yeoreum Yun <yeoreum.yun@arm.com>
Signed-off-by: Peter Zijlstra (Intel) <peterz@infradead.org>
Tested-by: David Wang <00107082@163.com>
Link: https://lore.kernel.org/all/aD2TspKH%2F7yvfYoO@e129823.arm.com/
While chasing down a missing perf_cgroup_event_disable() elsewhere,
Leo Yan found that both perf_put_aux_event() and
perf_remove_sibling_event() were also missing one.
Specifically, the rule is that events that switch to OFF,ERROR need to
call perf_cgroup_event_disable().
Unify the disable paths to ensure this.
Fixes: ab43762ef0 ("perf: Allow normal events to output AUX data")
Fixes: 9f0c4fa111 ("perf/core: Add a new PERF_EV_CAP_SIBLING event capability")
Reported-by: Leo Yan <leo.yan@arm.com>
Signed-off-by: Peter Zijlstra (Intel) <peterz@infradead.org>
Link: https://lkml.kernel.org/r/20250605123343.GD35970@noisy.programming.kicks-ass.net
Baisheng Gao reported an ARM64 crash, which Mark decoded as being a
synchronous external abort -- most likely due to trying to access
MMIO in bad ways.
The crash further shows perf trying to do a user stack sample while in
exit_mmap()'s tlb_finish_mmu() -- i.e. while tearing down the address
space it is trying to access.
It turns out that we stop perf after we tear down the userspace mm; a
receipie for disaster, since perf likes to access userspace for
various reasons.
Flip this order by moving up where we stop perf in do_exit().
Additionally, harden PERF_SAMPLE_CALLCHAIN and PERF_SAMPLE_STACK_USER
to abort when the current task does not have an mm (exit_mm() makes
sure to set current->mm = NULL; before commencing with the actual
teardown). Such that CPU wide events don't trip on this same problem.
Fixes: c5ebcedb56 ("perf: Add ability to attach user stack dump to sample")
Reported-by: Baisheng Gao <baisheng.gao@unisoc.com>
Suggested-by: Mark Rutland <mark.rutland@arm.com>
Signed-off-by: Peter Zijlstra (Intel) <peterz@infradead.org>
Link: https://lkml.kernel.org/r/20250605110815.GQ39944@noisy.programming.kicks-ass.net
The PERF_RECORD_THROTTLE records are dumped for all throttled events.
It's not necessary for group events, which are throttled altogether.
Optimize it by only dump the throttle log for the leader.
The sample right after the THROTTLE record must be generated by the
actual target event. It is good enough for the perf tool to locate the
actual target event.
Suggested-by: Namhyung Kim <namhyung@kernel.org>
Signed-off-by: Kan Liang <kan.liang@linux.intel.com>
Signed-off-by: Peter Zijlstra (Intel) <peterz@infradead.org>
Acked-by: Namhyung Kim <namhyung@kernel.org>
Link: https://lore.kernel.org/r/20250520181644.2673067-3-kan.liang@linux.intel.com
The current throttle logic doesn't work well with a group, e.g., the
following sampling-read case.
$ perf record -e "{cycles,cycles}:S" ...
$ perf report -D | grep THROTTLE | tail -2
THROTTLE events: 426 ( 9.0%)
UNTHROTTLE events: 425 ( 9.0%)
$ perf report -D | grep PERF_RECORD_SAMPLE -a4 | tail -n 5
0 1020120874009167 0x74970 [0x68]: PERF_RECORD_SAMPLE(IP, 0x1):
... sample_read:
.... group nr 2
..... id 0000000000000327, value 000000000cbb993a, lost 0
..... id 0000000000000328, value 00000002211c26df, lost 0
The second cycles event has a much larger value than the first cycles
event in the same group.
The current throttle logic in the generic code only logs the THROTTLE
event. It relies on the specific driver implementation to disable
events. For all ARCHs, the implementation is similar. Only the event is
disabled, rather than the group.
The logic to disable the group should be generic for all ARCHs. Add the
logic in the generic code. The following patch will remove the buggy
driver-specific implementation.
The throttle only happens when an event is overflowed. Stop the entire
group when any event in the group triggers the throttle.
The MAX_INTERRUPTS is set to all throttle events.
The unthrottled could happen in 3 places.
- event/group sched. All events in the group are scheduled one by one.
All of them will be unthrottled eventually. Nothing needs to be
changed.
- The perf_adjust_freq_unthr_events for each tick. Needs to restart the
group altogether.
- The __perf_event_period(). The whole group needs to be restarted
altogether as well.
With the fix,
$ sudo perf report -D | grep PERF_RECORD_SAMPLE -a4 | tail -n 5
0 3573470770332 0x12f5f8 [0x70]: PERF_RECORD_SAMPLE(IP, 0x2):
... sample_read:
.... group nr 2
..... id 0000000000000a28, value 00000004fd3dfd8f, lost 0
..... id 0000000000000a29, value 00000004fd3dfd8f, lost 0
Suggested-by: "Peter Zijlstra (Intel)" <peterz@infradead.org>
Signed-off-by: Kan Liang <kan.liang@linux.intel.com>
Signed-off-by: Peter Zijlstra (Intel) <peterz@infradead.org>
Acked-by: Namhyung Kim <namhyung@kernel.org>
Link: https://lore.kernel.org/r/20250520181644.2673067-2-kan.liang@linux.intel.com
While an event tears down all links to it as an aux, the iteration
happens on the event's group leader instead of the group itself.
If the event is a group leader, it has no effect because the event is
also its own group leader. But otherwise there would be a risk to detach
all the siblings events from the wrong group leader.
It just happens to work because each sibling's aux link is tested
against the right event before proceeding. Also the ctx lock is the same
for the events and their group leader so the iteration is safe.
Yet the iteration is confusing. Clarify the actual intent.
Signed-off-by: Frederic Weisbecker <frederic@kernel.org>
Signed-off-by: Peter Zijlstra (Intel) <peterz@infradead.org>
Link: https://lkml.kernel.org/r/20250424161128.29176-5-frederic@kernel.org
The following commit:
da916e96e2 ("perf: Make perf_pmu_unregister() useable")
has introduced two significant event's parent lifecycle changes:
1) An event that has exited now has EVENT_TOMBSTONE as a parent.
This can result in a situation where the delayed wakeup irq_work can
accidentally dereference EVENT_TOMBSTONE on:
CPU 0 CPU 1
----- -----
__schedule()
local_irq_disable()
rq_lock()
<NMI>
perf_event_overflow()
irq_work_queue(&child->pending_irq)
</NMI>
perf_event_task_sched_out()
raw_spin_lock(&ctx->lock)
ctx_sched_out()
ctx->is_active = 0
event_sched_out(child)
raw_spin_unlock(&ctx->lock)
perf_event_release_kernel(parent)
perf_remove_from_context(child)
raw_spin_lock_irq(&ctx->lock)
// Sees !ctx->is_active
// Removes from context inline
__perf_remove_from_context(child)
perf_child_detach(child)
event->parent = EVENT_TOMBSTONE
raw_spin_rq_unlock_irq(rq);
<IRQ>
perf_pending_irq()
perf_event_wakeup(child)
ring_buffer_wakeup(child)
rcu_dereference(child->parent->rb) <--- CRASH
This also concerns the call to kill_fasync() on parent->fasync.
2) The final parent reference count decrement can now happen before the
the final child reference count decrement. ie: the parent can now
be freed before its child. On PREEMPT_RT, this can result in a
situation where the delayed wakeup irq_work can accidentally
dereference a freed parent:
CPU 0 CPU 1 CPU 2
----- ----- ------
perf_pmu_unregister()
pmu_detach_events()
pmu_get_event()
atomic_long_inc_not_zero(&child->refcount)
<NMI>
perf_event_overflow()
irq_work_queue(&child->pending_irq);
</NMI>
<IRQ>
irq_work_run()
wake_irq_workd()
</IRQ>
preempt_schedule_irq()
=========> SWITCH to workd
irq_work_run_list()
perf_pending_irq()
perf_event_wakeup(child)
ring_buffer_wakeup(child)
event = child->parent
perf_event_release_kernel(parent)
// Not last ref, PMU holds it
put_event(child)
// Last ref
put_event(parent)
free_event()
call_rcu(...)
rcu_core()
free_event_rcu()
rcu_dereference(event->rb) <--- CRASH
This also concerns the call to kill_fasync() on parent->fasync.
The "easy" solution to 1) is to check that event->parent is not
EVENT_TOMBSTONE on perf_event_wakeup() (including both ring buffer
and fasync uses).
The "easy" solution to 2) is to turn perf_event_wakeup() to wholefully
run under rcu_read_lock().
However because of 2), sanity would prescribe to make event::parent
an __rcu pointer and annotate each and every users to prove they are
reliable.
Propose an alternate solution and restore the stable pointer to the
parent until all its children have called _free_event() themselves to
avoid any further accident. Also revert the EVENT_TOMBSTONE design
that is mostly here to determine which caller of perf_event_exit_event()
must perform the refcount decrement on a child event matching the
increment in inherit_event().
Arrange instead for checking the attach state of an event prior to its
removal and decrement the refcount of the child accordingly.
Fixes: da916e96e2 ("perf: Make perf_pmu_unregister() useable")
Signed-off-by: Frederic Weisbecker <frederic@kernel.org>
Signed-off-by: Peter Zijlstra (Intel) <peterz@infradead.org>
When inherit_event() fails after the child allocation but before the
parent refcount has been incremented, calling put_event() wrongly
decrements the reference to the parent, risking to free it too early.
Also pmu_get_event() can't be holding a reference to the child
concurrently at this point since it is under pmus_srcu critical section.
Fix it with restoring the deleted free_event() function and call it on
the failing child in order to free it directly under the verified
assumption that its refcount is only 1. The refcount to the parent is
then voluntarily omitted.
Fixes: da916e96e2 ("perf: Make perf_pmu_unregister() useable")
Signed-off-by: Frederic Weisbecker <frederic@kernel.org>
Signed-off-by: Peter Zijlstra (Intel) <peterz@infradead.org>
Link: https://lkml.kernel.org/r/20250424161128.29176-2-frederic@kernel.org
