[ Upstream commit 4363264111 ]
If uprobe handler changes instruction pointer we still execute single
step) or emulate the original instruction and increment the (new) ip
with its length.
This makes the new instruction pointer bogus and application will
likely crash on illegal instruction execution.
If user decided to take execution elsewhere, it makes little sense
to execute the original instruction, so let's skip it.
Acked-by: Oleg Nesterov <oleg@redhat.com>
Acked-by: Andrii Nakryiko <andrii@kernel.org>
Signed-off-by: Jiri Olsa <jolsa@kernel.org>
Link: https://lore.kernel.org/r/20250916215301.664963-3-jolsa@kernel.org
Signed-off-by: Alexei Starovoitov <ast@kernel.org>
Signed-off-by: Sasha Levin <sashal@kernel.org>
commit b024d7b56c upstream.
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/core: 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
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
commit 07091aade3 upstream.
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/core: 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
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
[ Upstream commit ba677dbe77 ]
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
Signed-off-by: Sasha Levin <sashal@kernel.org>
[ Upstream commit 3172fb9866 ]
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
Signed-off-by: Sasha Levin <sashal@kernel.org>
[ Upstream commit 61988e36dc ]
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
Signed-off-by: Sasha Levin <sashal@kernel.org>
[ Upstream commit 4f6fc78212 ]
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
Signed-off-by: Sasha Levin <sashal@kernel.org>
[ Upstream commit f51972e6f8 ]
According to the throttling mechanism, the pmu interrupts number can not
exceed the max_samples_per_tick in one tick. But this mechanism is
ineffective when max_samples_per_tick=1, because the throttling check is
skipped during the first interrupt and only performed when the second
interrupt arrives.
Perhaps this bug may cause little influence in one tick, but if in a
larger time scale, the problem can not be underestimated.
When max_samples_per_tick = 1:
Allowed-interrupts-per-second max-samples-per-second default-HZ ARCH
200 100 100 X86
500 250 250 ARM64
...
Obviously, the pmu interrupt number far exceed the user's expect.
Fixes: e050e3f0a7 ("perf: Fix broken interrupt rate throttling")
Signed-off-by: Qing Wang <wangqing7171@gmail.com>
Signed-off-by: Peter Zijlstra (Intel) <peterz@infradead.org>
Link: https://lkml.kernel.org/r/20250405141635.243786-3-wangqing7171@gmail.com
Signed-off-by: Sasha Levin <sashal@kernel.org>
[ Upstream commit 8ce939a0fa ]
The event may have been updated in the PMU-specific implementation,
e.g., Intel PEBS counters snapshotting. The common code should not
read and overwrite the value.
The PERF_SAMPLE_READ in the data->sample_type can be used to detect
whether the PMU-specific value is available. If yes, avoid the
pmu->read() in the common code. Add a new flag, skip_read, to track the
case.
Factor out a perf_pmu_read() to clean up the code.
Signed-off-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>
Link: https://lkml.kernel.org/r/20250121152303.3128733-3-kan.liang@linux.intel.com
Signed-off-by: Sasha Levin <sashal@kernel.org>
[ Upstream commit 061c991697 ]
Currently, __reserve_bp_slot() returns -ENOSPC for unsupported
breakpoint types on the architecture. For example, powerpc
does not support hardware instruction breakpoints. This causes
the perf_skip BPF selftest to fail, as neither ENOENT nor
EOPNOTSUPP is returned by perf_event_open for unsupported
breakpoint types. As a result, the test that should be skipped
for this arch is not correctly identified.
To resolve this, hw_breakpoint_event_init() should exit early by
checking for unsupported breakpoint types using
hw_breakpoint_slots_cached() and return the appropriate error
(-EOPNOTSUPP).
Signed-off-by: Saket Kumar Bhaskar <skb99@linux.ibm.com>
Signed-off-by: Ingo Molnar <mingo@kernel.org>
Cc: Marco Elver <elver@google.com>
Cc: Dmitry Vyukov <dvyukov@google.com>
Cc: Ian Rogers <irogers@google.com>
Cc: Frederic Weisbecker <fweisbec@gmail.com>
Link: https://lore.kernel.org/r/20250303092451.1862862-1-skb99@linux.ibm.com
Signed-off-by: Sasha Levin <sashal@kernel.org>
[ Upstream commit 0ba3a4ab76 ]
Move the get_ctx(child_ctx) call and the child_event->ctx assignment to
occur immediately after the child event is allocated. Ensure that
child_event->ctx is non-NULL before any subsequent error path within
inherit_event calls free_event(), satisfying the assumptions of the
cleanup code.
Details:
There's no clear Fixes tag, because this bug is a side-effect of
multiple interacting commits over time (up to 15 years old), not
a single regression.
The code initially incremented refcount then assigned context
immediately after the child_event was created. Later, an early
validity check for child_event was added before the
refcount/assignment. Even later, a WARN_ON_ONCE() cleanup check was
added, assuming event->ctx is valid if the pmu_ctx is valid.
The problem is that the WARN_ON_ONCE() could trigger after the initial
check passed but before child_event->ctx was assigned, violating its
precondition. The solution is to assign child_event->ctx right after
its initial validation. This ensures the context exists for any
subsequent checks or cleanup routines, resolving the WARN_ON_ONCE().
To resolve it, defer the refcount update and child_event->ctx assignment
directly after child_event->pmu_ctx is set but before checking if the
parent event is orphaned. The cleanup routine depends on
event->pmu_ctx being non-NULL before it verifies event->ctx is
non-NULL. This also maintains the author's original intent of passing
in child_ctx to find_get_pmu_context before its refcount/assignment.
[ mingo: Expanded the changelog from another email by Gabriel Shahrouzi. ]
Reported-by: syzbot+ff3aa851d46ab82953a3@syzkaller.appspotmail.com
Signed-off-by: Gabriel Shahrouzi <gshahrouzi@gmail.com>
Signed-off-by: Ingo Molnar <mingo@kernel.org>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Ravi Bangoria <ravi.bangoria@amd.com>
Cc: Kan Liang <kan.liang@linux.intel.com>
Cc: Oleg Nesterov <oleg@redhat.com>
Cc: Alexander Shishkin <alexander.shishkin@linux.intel.com>
Link: https://lore.kernel.org/r/20250405203036.582721-1-gshahrouzi@gmail.com
Closes: https://syzkaller.appspot.com/bug?extid=ff3aa851d46ab82953a3
Signed-off-by: Sasha Levin <sashal@kernel.org>
[ Upstream commit a3c3c66670 ]
The perf events code fails to account for total_time_enabled of
inactive events.
Here is a failure case for accounting total_time_enabled for
CPU PMU events:
sudo ./perf stat -vvv -e armv8_pmuv3_0/event=0x08/ -e armv8_pmuv3_1/event=0x08/ -- stress-ng --pthread=2 -t 2s
...
armv8_pmuv3_0/event=0x08/: 1138698008 2289429840 2174835740
armv8_pmuv3_1/event=0x08/: 1826791390 1950025700 847648440
` ` `
` ` > total_time_running with child
` > total_time_enabled with child
> count with child
Performance counter stats for 'stress-ng --pthread=2 -t 2s':
1,138,698,008 armv8_pmuv3_0/event=0x08/ (94.99%)
1,826,791,390 armv8_pmuv3_1/event=0x08/ (43.47%)
The two events above are opened on two different CPU PMUs, for example,
each event is opened for a cluster in an Arm big.LITTLE system, they
will never run on the same CPU. In theory, the total enabled time should
be same for both events, as two events are opened and closed together.
As the result show, the two events' total enabled time including
child event is different (2289429840 vs 1950025700).
This is because child events are not accounted properly
if a event is INACTIVE state when the task exits:
perf_event_exit_event()
`> perf_remove_from_context()
`> __perf_remove_from_context()
`> perf_child_detach() -> Accumulate child_total_time_enabled
`> list_del_event() -> Update child event's time
The problem is the time accumulation happens prior to child event's
time updating. Thus, it misses to account the last period's time when
the event exits.
The perf core layer follows the rule that timekeeping is tied to state
change. To address the issue, make __perf_remove_from_context()
handle the task exit case by passing 'DETACH_EXIT' to it and
invoke perf_event_state() for state alongside with accounting the time.
Then, perf_child_detach() populates the time into the parent's time metrics.
After this patch, the bug is fixed:
sudo ./perf stat -vvv -e armv8_pmuv3_0/event=0x08/ -e armv8_pmuv3_1/event=0x08/ -- stress-ng --pthread=2 -t 10s
...
armv8_pmuv3_0/event=0x08/: 15396770398 32157963940 21898169000
armv8_pmuv3_1/event=0x08/: 22428964974 32157963940 10259794940
Performance counter stats for 'stress-ng --pthread=2 -t 10s':
15,396,770,398 armv8_pmuv3_0/event=0x08/ (68.10%)
22,428,964,974 armv8_pmuv3_1/event=0x08/ (31.90%)
[ mingo: Clarified the changelog. ]
Fixes: ef54c1a476 ("perf: Rework perf_event_exit_event()")
Suggested-by: Peter Zijlstra <peterz@infradead.org>
Signed-off-by: Yeoreum Yun <yeoreum.yun@arm.com>
Signed-off-by: Ingo Molnar <mingo@kernel.org>
Tested-by: Leo Yan <leo.yan@arm.com>
Link: https://lore.kernel.org/r/20250326082003.1630986-1-yeoreum.yun@arm.com
Signed-off-by: Sasha Levin <sashal@kernel.org>
[ Upstream commit 003659fec9 ]
There is a fairly obvious race between perf_init_event() doing
idr_find() and perf_pmu_register() doing idr_alloc() with an
incompletely initialized PMU pointer.
Avoid by doing idr_alloc() on a NULL pointer to register the id, and
swizzling the real struct pmu pointer at the end using idr_replace().
Also making sure to not set struct pmu members after publishing
the struct pmu, duh.
[ introduce idr_cmpxchg() in order to better handle the idr_replace()
error case -- if it were to return an unexpected pointer, it will
already have replaced the value and there is no going back. ]
Signed-off-by: Peter Zijlstra (Intel) <peterz@infradead.org>
Signed-off-by: Ingo Molnar <mingo@kernel.org>
Link: https://lore.kernel.org/r/20241104135517.858805880@infradead.org
Signed-off-by: Sasha Levin <sashal@kernel.org>
commit b583ef82b6 upstream.
Max Makarov reported kernel panic [1] in perf user callchain code.
The reason for that is the race between uprobe_free_utask and bpf
profiler code doing the perf user stack unwind and is triggered
within uprobe_free_utask function:
- after current->utask is freed and
- before current->utask is set to NULL
general protection fault, probably for non-canonical address 0x9e759c37ee555c76: 0000 [#1] SMP PTI
RIP: 0010:is_uprobe_at_func_entry+0x28/0x80
...
? die_addr+0x36/0x90
? exc_general_protection+0x217/0x420
? asm_exc_general_protection+0x26/0x30
? is_uprobe_at_func_entry+0x28/0x80
perf_callchain_user+0x20a/0x360
get_perf_callchain+0x147/0x1d0
bpf_get_stackid+0x60/0x90
bpf_prog_9aac297fb833e2f5_do_perf_event+0x434/0x53b
? __smp_call_single_queue+0xad/0x120
bpf_overflow_handler+0x75/0x110
...
asm_sysvec_apic_timer_interrupt+0x1a/0x20
RIP: 0010:__kmem_cache_free+0x1cb/0x350
...
? uprobe_free_utask+0x62/0x80
? acct_collect+0x4c/0x220
uprobe_free_utask+0x62/0x80
mm_release+0x12/0xb0
do_exit+0x26b/0xaa0
__x64_sys_exit+0x1b/0x20
do_syscall_64+0x5a/0x80
It can be easily reproduced by running following commands in
separate terminals:
# while :; do bpftrace -e 'uprobe:/bin/ls:_start { printf("hit\n"); }' -c ls; done
# bpftrace -e 'profile:hz:100000 { @[ustack()] = count(); }'
Fixing this by making sure current->utask pointer is set to NULL
before we start to release the utask object.
[1] https://github.com/grafana/pyroscope/issues/3673
Fixes: cfa7f3d2c5 ("perf,x86: avoid missing caller address in stack traces captured in uprobe")
Reported-by: Max Makarov <maxpain@linux.com>
Signed-off-by: Jiri Olsa <jolsa@kernel.org>
Signed-off-by: Peter Zijlstra (Intel) <peterz@infradead.org>
Acked-by: Oleg Nesterov <oleg@redhat.com>
Acked-by: Andrii Nakryiko <andrii@kernel.org>
Link: https://lore.kernel.org/r/20250109141440.2692173-1-jolsa@kernel.org
[Christian Simon: Rebased for 6.12.y, due to mainline change https://lore.kernel.org/all/20240929144239.GA9475@redhat.com/]
Signed-off-by: Christian Simon <simon@swine.de>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
[ Upstream commit 2565e42539 ]
Commit a63fbed776 ("perf/tracing/cpuhotplug: Fix locking order")
placed pmus_lock inside pmus_srcu, this makes perf_pmu_unregister()
trip lockdep.
Move the locking about such that only pmu_idr and pmus (list) are
modified while holding pmus_lock. This avoids doing synchronize_srcu()
while holding pmus_lock and all is well again.
Fixes: a63fbed776 ("perf/tracing/cpuhotplug: Fix locking order")
Signed-off-by: Peter Zijlstra (Intel) <peterz@infradead.org>
Signed-off-by: Ingo Molnar <mingo@kernel.org>
Link: https://lore.kernel.org/r/20241104135517.679556858@infradead.org
Signed-off-by: Sasha Levin <sashal@kernel.org>
[ Upstream commit bddf10d26e ]
We triggered the following crash in syzkaller tests:
BUG: Bad page state in process syz.7.38 pfn:1eff3
page: refcount:0 mapcount:0 mapping:0000000000000000 index:0x0 pfn:0x1eff3
flags: 0x3fffff00004004(referenced|reserved|node=0|zone=1|lastcpupid=0x1fffff)
raw: 003fffff00004004 ffffe6c6c07bfcc8 ffffe6c6c07bfcc8 0000000000000000
raw: 0000000000000000 0000000000000000 00000000fffffffe 0000000000000000
page dumped because: PAGE_FLAGS_CHECK_AT_FREE flag(s) set
Hardware name: QEMU Standard PC (i440FX + PIIX, 1996), BIOS 1.13.0-1ubuntu1.1 04/01/2014
Call Trace:
<TASK>
dump_stack_lvl+0x32/0x50
bad_page+0x69/0xf0
free_unref_page_prepare+0x401/0x500
free_unref_page+0x6d/0x1b0
uprobe_write_opcode+0x460/0x8e0
install_breakpoint.part.0+0x51/0x80
register_for_each_vma+0x1d9/0x2b0
__uprobe_register+0x245/0x300
bpf_uprobe_multi_link_attach+0x29b/0x4f0
link_create+0x1e2/0x280
__sys_bpf+0x75f/0xac0
__x64_sys_bpf+0x1a/0x30
do_syscall_64+0x56/0x100
entry_SYSCALL_64_after_hwframe+0x78/0xe2
BUG: Bad rss-counter state mm:00000000452453e0 type:MM_FILEPAGES val:-1
The following syzkaller test case can be used to reproduce:
r2 = creat(&(0x7f0000000000)='./file0\x00', 0x8)
write$nbd(r2, &(0x7f0000000580)=ANY=[], 0x10)
r4 = openat(0xffffffffffffff9c, &(0x7f0000000040)='./file0\x00', 0x42, 0x0)
mmap$IORING_OFF_SQ_RING(&(0x7f0000ffd000/0x3000)=nil, 0x3000, 0x0, 0x12, r4, 0x0)
r5 = userfaultfd(0x80801)
ioctl$UFFDIO_API(r5, 0xc018aa3f, &(0x7f0000000040)={0xaa, 0x20})
r6 = userfaultfd(0x80801)
ioctl$UFFDIO_API(r6, 0xc018aa3f, &(0x7f0000000140))
ioctl$UFFDIO_REGISTER(r6, 0xc020aa00, &(0x7f0000000100)={{&(0x7f0000ffc000/0x4000)=nil, 0x4000}, 0x2})
ioctl$UFFDIO_ZEROPAGE(r5, 0xc020aa04, &(0x7f0000000000)={{&(0x7f0000ffd000/0x1000)=nil, 0x1000}})
r7 = bpf$PROG_LOAD(0x5, &(0x7f0000000140)={0x2, 0x3, &(0x7f0000000200)=ANY=[@ANYBLOB="1800000000120000000000000000000095"], &(0x7f0000000000)='GPL\x00', 0x7, 0x0, 0x0, 0x0, 0x0, '\x00', 0x0, @fallback=0x30, 0xffffffffffffffff, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x10, 0x0, @void, @value}, 0x94)
bpf$BPF_LINK_CREATE_XDP(0x1c, &(0x7f0000000040)={r7, 0x0, 0x30, 0x1e, @val=@uprobe_multi={&(0x7f0000000080)='./file0\x00', &(0x7f0000000100)=[0x2], 0x0, 0x0, 0x1}}, 0x40)
The cause is that zero pfn is set to the PTE without increasing the RSS
count in mfill_atomic_pte_zeropage() and the refcount of zero folio does
not increase accordingly. Then, the operation on the same pfn is performed
in uprobe_write_opcode()->__replace_page() to unconditional decrease the
RSS count and old_folio's refcount.
Therefore, two bugs are introduced:
1. The RSS count is incorrect, when process exit, the check_mm() report
error "Bad rss-count".
2. The reserved folio (zero folio) is freed when folio->refcount is zero,
then free_pages_prepare->free_page_is_bad() report error
"Bad page state".
There is more, the following warning could also theoretically be triggered:
__replace_page()
-> ...
-> folio_remove_rmap_pte()
-> VM_WARN_ON_FOLIO(is_zero_folio(folio), folio)
Considering that uprobe hit on the zero folio is a very rare case, just
reject zero old folio immediately after get_user_page_vma_remote().
[ mingo: Cleaned up the changelog ]
Fixes: 7396fa818d ("uprobes/core: Make background page replacement logic account for rss_stat counters")
Fixes: 2b14449835 ("uprobes, mm, x86: Add the ability to install and remove uprobes breakpoints")
Signed-off-by: Tong Tiangen <tongtiangen@huawei.com>
Signed-off-by: Ingo Molnar <mingo@kernel.org>
Reviewed-by: David Hildenbrand <david@redhat.com>
Reviewed-by: Oleg Nesterov <oleg@redhat.com>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Masami Hiramatsu <mhiramat@kernel.org>
Link: https://lore.kernel.org/r/20250224031149.1598949-1-tongtiangen@huawei.com
Signed-off-by: Sasha Levin <sashal@kernel.org>
[ Upstream commit 2016066c66 ]
Syskaller triggers a warning due to prev_epc->pmu != next_epc->pmu in
perf_event_swap_task_ctx_data(). vmcore shows that two lists have the same
perf_event_pmu_context, but not in the same order.
The problem is that the order of pmu_ctx_list for the parent is impacted by
the time when an event/PMU is added. While the order for a child is
impacted by the event order in the pinned_groups and flexible_groups. So
the order of pmu_ctx_list in the parent and child may be different.
To fix this problem, insert the perf_event_pmu_context to its proper place
after iteration of the pmu_ctx_list.
The follow testcase can trigger above warning:
# perf record -e cycles --call-graph lbr -- taskset -c 3 ./a.out &
# perf stat -e cpu-clock,cs -p xxx // xxx is the pid of a.out
test.c
void main() {
int count = 0;
pid_t pid;
printf("%d running\n", getpid());
sleep(30);
printf("running\n");
pid = fork();
if (pid == -1) {
printf("fork error\n");
return;
}
if (pid == 0) {
while (1) {
count++;
}
} else {
while (1) {
count++;
}
}
}
The testcase first opens an LBR event, so it will allocate task_ctx_data,
and then open tracepoint and software events, so the parent context will
have 3 different perf_event_pmu_contexts. On inheritance, child ctx will
insert the perf_event_pmu_context in another order and the warning will
trigger.
[ mingo: Tidied up the changelog. ]
Fixes: bd27568117 ("perf: Rewrite core context handling")
Signed-off-by: Luo Gengkun <luogengkun@huaweicloud.com>
Signed-off-by: Ingo Molnar <mingo@kernel.org>
Reviewed-by: Kan Liang <kan.liang@linux.intel.com>
Link: https://lore.kernel.org/r/20250122073356.1824736-1-luogengkun@huaweicloud.com
Signed-off-by: Sasha Levin <sashal@kernel.org>
[ Upstream commit b9c44b9147 ]
Currently, space for raw sample data is always allocated within sample
records for both BPF output and tracepoint events. This leads to unused
space in sample records when raw sample data is not requested.
This patch enforces checking sample type of an event in
perf_sample_save_raw_data(). So raw sample data will only be saved if
explicitly requested, reducing overhead when it is not needed.
Fixes: 0a9081cf0a ("perf/core: Add perf_sample_save_raw_data() helper")
Signed-off-by: Yabin Cui <yabinc@google.com>
Signed-off-by: Ingo Molnar <mingo@kernel.org>
Reviewed-by: Ian Rogers <irogers@google.com>
Acked-by: Namhyung Kim <namhyung@kernel.org>
Link: https://lore.kernel.org/r/20240515193610.2350456-2-yabinc@google.com
Signed-off-by: Sasha Levin <sashal@kernel.org>
commit 62c0b10615 upstream.
In perf_adjust_period, we will first calculate period, and then use
this period to calculate delta. However, when delta is less than 0,
there will be a deviation compared to when delta is greater than or
equal to 0. For example, when delta is in the range of [-14,-1], the
range of delta = delta + 7 is between [-7,6], so the final value of
delta/8 is 0. Therefore, the impact of -1 and -2 will be ignored.
This is unacceptable when the target period is very short, because
we will lose a lot of samples.
Here are some tests and analyzes:
before:
# perf record -e cs -F 1000 ./a.out
[ perf record: Woken up 1 times to write data ]
[ perf record: Captured and wrote 0.022 MB perf.data (518 samples) ]
# perf script
...
a.out 396 257.956048: 23 cs: ffffffff81f4eeec schedul>
a.out 396 257.957891: 23 cs: ffffffff81f4eeec schedul>
a.out 396 257.959730: 23 cs: ffffffff81f4eeec schedul>
a.out 396 257.961545: 23 cs: ffffffff81f4eeec schedul>
a.out 396 257.963355: 23 cs: ffffffff81f4eeec schedul>
a.out 396 257.965163: 23 cs: ffffffff81f4eeec schedul>
a.out 396 257.966973: 23 cs: ffffffff81f4eeec schedul>
a.out 396 257.968785: 23 cs: ffffffff81f4eeec schedul>
a.out 396 257.970593: 23 cs: ffffffff81f4eeec schedul>
...
after:
# perf record -e cs -F 1000 ./a.out
[ perf record: Woken up 1 times to write data ]
[ perf record: Captured and wrote 0.058 MB perf.data (1466 samples) ]
# perf script
...
a.out 395 59.338813: 11 cs: ffffffff81f4eeec schedul>
a.out 395 59.339707: 12 cs: ffffffff81f4eeec schedul>
a.out 395 59.340682: 13 cs: ffffffff81f4eeec schedul>
a.out 395 59.341751: 13 cs: ffffffff81f4eeec schedul>
a.out 395 59.342799: 12 cs: ffffffff81f4eeec schedul>
a.out 395 59.343765: 11 cs: ffffffff81f4eeec schedul>
a.out 395 59.344651: 11 cs: ffffffff81f4eeec schedul>
a.out 395 59.345539: 12 cs: ffffffff81f4eeec schedul>
a.out 395 59.346502: 13 cs: ffffffff81f4eeec schedul>
...
test.c
int main() {
for (int i = 0; i < 20000; i++)
usleep(10);
return 0;
}
# time ./a.out
real 0m1.583s
user 0m0.040s
sys 0m0.298s
The above results were tested on x86-64 qemu with KVM enabled using
test.c as test program. Ideally, we should have around 1500 samples,
but the previous algorithm had only about 500, whereas the modified
algorithm now has about 1400. Further more, the new version shows 1
sample per 0.001s, while the previous one is 1 sample per 0.002s.This
indicates that the new algorithm is more sensitive to small negative
values compared to old algorithm.
Fixes: bd2b5b1284 ("perf_counter: More aggressive frequency adjustment")
Signed-off-by: Luo Gengkun <luogengkun@huaweicloud.com>
Signed-off-by: Peter Zijlstra (Intel) <peterz@infradead.org>
Reviewed-by: Adrian Hunter <adrian.hunter@intel.com>
Reviewed-by: Kan Liang <kan.liang@linux.intel.com>
Cc: stable@vger.kernel.org
Link: https://lkml.kernel.org/r/20240831074316.2106159-2-luogengkun@huaweicloud.com
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
[ Upstream commit cfa7f3d2c5 ]
When tracing user functions with uprobe functionality, it's common to
install the probe (e.g., a BPF program) at the first instruction of the
function. This is often going to be `push %rbp` instruction in function
preamble, which means that within that function frame pointer hasn't
been established yet. This leads to consistently missing an actual
caller of the traced function, because perf_callchain_user() only
records current IP (capturing traced function) and then following frame
pointer chain (which would be caller's frame, containing the address of
caller's caller).
So when we have target_1 -> target_2 -> target_3 call chain and we are
tracing an entry to target_3, captured stack trace will report
target_1 -> target_3 call chain, which is wrong and confusing.
This patch proposes a x86-64-specific heuristic to detect `push %rbp`
(`push %ebp` on 32-bit architecture) instruction being traced. Given
entire kernel implementation of user space stack trace capturing works
under assumption that user space code was compiled with frame pointer
register (%rbp/%ebp) preservation, it seems pretty reasonable to use
this instruction as a strong indicator that this is the entry to the
function. In that case, return address is still pointed to by %rsp/%esp,
so we fetch it and add to stack trace before proceeding to unwind the
rest using frame pointer-based logic.
We also check for `endbr64` (for 64-bit modes) as another common pattern
for function entry, as suggested by Josh Poimboeuf. Even if we get this
wrong sometimes for uprobes attached not at the function entry, it's OK
because stack trace will still be overall meaningful, just with one
extra bogus entry. If we don't detect this, we end up with guaranteed to
be missing caller function entry in the stack trace, which is worse
overall.
Signed-off-by: Andrii Nakryiko <andrii@kernel.org>
Signed-off-by: Peter Zijlstra (Intel) <peterz@infradead.org>
Link: https://lkml.kernel.org/r/20240729175223.23914-1-andrii@kernel.org
Signed-off-by: Sasha Levin <sashal@kernel.org>
[ Upstream commit 5e9629d0ae ]
Use perf_allow_kernel() for 'pa_enable' (physical addresses),
'pct_enable' (physical timestamps) and context IDs. This means that
perf_event_paranoid is now taken into account and LSM hooks can be used,
which is more consistent with other perf_event_open calls. For example
PERF_SAMPLE_PHYS_ADDR uses perf_allow_kernel() rather than just
perfmon_capable().
This also indirectly fixes the following error message which is
misleading because perf_event_paranoid is not taken into account by
perfmon_capable():
$ perf record -e arm_spe/pa_enable/
Error:
Access to performance monitoring and observability operations is
limited. Consider adjusting /proc/sys/kernel/perf_event_paranoid
setting ...
Suggested-by: Al Grant <al.grant@arm.com>
Signed-off-by: James Clark <james.clark@linaro.org>
Link: https://lore.kernel.org/r/20240827145113.1224604-1-james.clark@linaro.org
Link: https://lore.kernel.org/all/20240807120039.GD37996@noisy.programming.kicks-ass.net/
Signed-off-by: Will Deacon <will@kernel.org>
Signed-off-by: Sasha Levin <sashal@kernel.org>
commit 2ab9d83026 upstream.
Ole reported that event->mmap_mutex is strictly insufficient to
serialize the AUX buffer, add a per RB mutex to fully serialize it.
Note that in the lock order comment the perf_event::mmap_mutex order
was already wrong, that is, it nesting under mmap_lock is not new with
this patch.
Fixes: 45bfb2e504 ("perf: Add AUX area to ring buffer for raw data streams")
Reported-by: Ole <ole@binarygecko.com>
Signed-off-by: Peter Zijlstra (Intel) <peterz@infradead.org>
Signed-off-by: Ingo Molnar <mingo@kernel.org>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
[ Upstream commit 0be9ae5486 ]
Since commit 0108a4e9f3 ("bpf: ensure main program has an extable"),
prog->aux->func[0]->kallsyms is left as uninitialized. For BPF programs
with subprogs, the symbol for the main program is missing just as shown
in the output of perf script below:
ffffffff81284b69 qp_trie_lookup_elem+0xb9 ([kernel.kallsyms])
ffffffffc0011125 bpf_prog_a4a0eb0651e6af8b_lookup_qp_trie+0x5d (bpf...)
ffffffff8127bc2b bpf_for_each_array_elem+0x7b ([kernel.kallsyms])
ffffffffc00110a1 +0x25 ()
ffffffff8121a89a trace_call_bpf+0xca ([kernel.kallsyms])
Fix it by always using prog instead prog->aux->func[0] to emit ksymbol
event for the main program. After the fix, the output of perf script
will be correct:
ffffffff81284b96 qp_trie_lookup_elem+0xe6 ([kernel.kallsyms])
ffffffffc001382d bpf_prog_a4a0eb0651e6af8b_lookup_qp_trie+0x5d (bpf...)
ffffffff8127bc2b bpf_for_each_array_elem+0x7b ([kernel.kallsyms])
ffffffffc0013779 bpf_prog_245c55ab25cfcf40_qp_trie_lookup+0x25 (bpf...)
ffffffff8121a89a trace_call_bpf+0xca ([kernel.kallsyms])
Fixes: 0108a4e9f3 ("bpf: ensure main program has an extable")
Signed-off-by: Hou Tao <houtao1@huawei.com>
Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
Tested-by: Yonghong Song <yonghong.song@linux.dev>
Reviewed-by: Krister Johansen <kjlx@templeofstupid.com>
Reviewed-by: Jiri Olsa <jolsa@kernel.org>
Link: https://lore.kernel.org/bpf/20240714065533.1112616-1-houtao@huaweicloud.com
Signed-off-by: Sasha Levin <sashal@kernel.org>
commit 3a5465418f upstream.
The perf pending task work is never waited upon the matching event
release. In the case of a child event, released via free_event()
directly, this can potentially result in a leaked event, such as in the
following scenario that doesn't even require a weak IRQ work
implementation to trigger:
schedule()
prepare_task_switch()
=======> <NMI>
perf_event_overflow()
event->pending_sigtrap = ...
irq_work_queue(&event->pending_irq)
<======= </NMI>
perf_event_task_sched_out()
event_sched_out()
event->pending_sigtrap = 0;
atomic_long_inc_not_zero(&event->refcount)
task_work_add(&event->pending_task)
finish_lock_switch()
=======> <IRQ>
perf_pending_irq()
//do nothing, rely on pending task work
<======= </IRQ>
begin_new_exec()
perf_event_exit_task()
perf_event_exit_event()
// If is child event
free_event()
WARN(atomic_long_cmpxchg(&event->refcount, 1, 0) != 1)
// event is leaked
Similar scenarios can also happen with perf_event_remove_on_exec() or
simply against concurrent perf_event_release().
Fix this with synchonizing against the possibly remaining pending task
work while freeing the event, just like is done with remaining pending
IRQ work. This means that the pending task callback neither need nor
should hold a reference to the event, preventing it from ever beeing
freed.
Fixes: 517e6a301f ("perf: Fix perf_pending_task() UaF")
Signed-off-by: Frederic Weisbecker <frederic@kernel.org>
Signed-off-by: Peter Zijlstra (Intel) <peterz@infradead.org>
Cc: stable@vger.kernel.org
Link: https://lore.kernel.org/r/20240621091601.18227-5-frederic@kernel.org
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
commit 2fd5ad3f31 upstream.
When a task is scheduled out, pending sigtrap deliveries are deferred
to the target task upon resume to userspace via task_work.
However failures while adding an event's callback to the task_work
engine are ignored. And since the last call for events exit happen
after task work is eventually closed, there is a small window during
which pending sigtrap can be queued though ignored, leaking the event
refcount addition such as in the following scenario:
TASK A
-----
do_exit()
exit_task_work(tsk);
<IRQ>
perf_event_overflow()
event->pending_sigtrap = pending_id;
irq_work_queue(&event->pending_irq);
</IRQ>
=========> PREEMPTION: TASK A -> TASK B
event_sched_out()
event->pending_sigtrap = 0;
atomic_long_inc_not_zero(&event->refcount)
// FAILS: task work has exited
task_work_add(&event->pending_task)
[...]
<IRQ WORK>
perf_pending_irq()
// early return: event->oncpu = -1
</IRQ WORK>
[...]
=========> TASK B -> TASK A
perf_event_exit_task(tsk)
perf_event_exit_event()
free_event()
WARN(atomic_long_cmpxchg(&event->refcount, 1, 0) != 1)
// leak event due to unexpected refcount == 2
As a result the event is never released while the task exits.
Fix this with appropriate task_work_add()'s error handling.
Fixes: 517e6a301f ("perf: Fix perf_pending_task() UaF")
Signed-off-by: Frederic Weisbecker <frederic@kernel.org>
Signed-off-by: Peter Zijlstra (Intel) <peterz@infradead.org>
Cc: stable@vger.kernel.org
Link: https://lore.kernel.org/r/20240621091601.18227-4-frederic@kernel.org
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
[ Upstream commit 43deb76b19 ]
The default aux_watermark is half the AUX area buffer size. In general,
on a 64-bit architecture, the AUX area buffer size could be a bigger than
fits in a 32-bit type, but the calculation does not allow for that
possibility.
However the aux_watermark value is recorded in a u32, so should not be
more than U32_MAX either.
Fix by doing the calculation in a correctly sized type, and limiting the
result to U32_MAX.
Fixes: d68e6799a5 ("perf: Cap allocation order at aux_watermark")
Signed-off-by: Adrian Hunter <adrian.hunter@intel.com>
Signed-off-by: Peter Zijlstra (Intel) <peterz@infradead.org>
Link: https://lore.kernel.org/r/20240624201101.60186-7-adrian.hunter@intel.com
Signed-off-by: Sasha Levin <sashal@kernel.org>
[ Upstream commit dbc48c8f41 ]
nr_pages is unsigned long but gets passed to rb_alloc_aux() as an int,
and is stored as an int.
Only power-of-2 values are accepted, so if nr_pages is a 64_bit value, it
will be passed to rb_alloc_aux() as zero.
That is not ideal because:
1. the value is incorrect
2. rb_alloc_aux() is at risk of misbehaving, although it manages to
return -ENOMEM in that case, it is a result of passing zero to get_order()
even though the get_order() result is documented to be undefined in that
case.
Fix by simply validating the maximum supported value in the first place.
Use -ENOMEM error code for consistency with the current error code that
is returned in that case.
Fixes: 45bfb2e504 ("perf: Add AUX area to ring buffer for raw data streams")
Signed-off-by: Adrian Hunter <adrian.hunter@intel.com>
Signed-off-by: Peter Zijlstra (Intel) <peterz@infradead.org>
Link: https://lore.kernel.org/r/20240624201101.60186-6-adrian.hunter@intel.com
Signed-off-by: Sasha Levin <sashal@kernel.org>
commit 74751ef5c1 upstream.
In our production environment, we found many hung tasks which are
blocked for more than 18 hours. Their call traces are like this:
[346278.191038] __schedule+0x2d8/0x890
[346278.191046] schedule+0x4e/0xb0
[346278.191049] perf_event_free_task+0x220/0x270
[346278.191056] ? init_wait_var_entry+0x50/0x50
[346278.191060] copy_process+0x663/0x18d0
[346278.191068] kernel_clone+0x9d/0x3d0
[346278.191072] __do_sys_clone+0x5d/0x80
[346278.191076] __x64_sys_clone+0x25/0x30
[346278.191079] do_syscall_64+0x5c/0xc0
[346278.191083] ? syscall_exit_to_user_mode+0x27/0x50
[346278.191086] ? do_syscall_64+0x69/0xc0
[346278.191088] ? irqentry_exit_to_user_mode+0x9/0x20
[346278.191092] ? irqentry_exit+0x19/0x30
[346278.191095] ? exc_page_fault+0x89/0x160
[346278.191097] ? asm_exc_page_fault+0x8/0x30
[346278.191102] entry_SYSCALL_64_after_hwframe+0x44/0xae
The task was waiting for the refcount become to 1, but from the vmcore,
we found the refcount has already been 1. It seems that the task didn't
get woken up by perf_event_release_kernel() and got stuck forever. The
below scenario may cause the problem.
Thread A Thread B
... ...
perf_event_free_task perf_event_release_kernel
...
acquire event->child_mutex
...
get_ctx
... release event->child_mutex
acquire ctx->mutex
...
perf_free_event (acquire/release event->child_mutex)
...
release ctx->mutex
wait_var_event
acquire ctx->mutex
acquire event->child_mutex
# move existing events to free_list
release event->child_mutex
release ctx->mutex
put_ctx
... ...
In this case, all events of the ctx have been freed, so we couldn't
find the ctx in free_list and Thread A will miss the wakeup. It's thus
necessary to add a wakeup after dropping the reference.
Fixes: 1cf8dfe8a6 ("perf/core: Fix race between close() and fork()")
Signed-off-by: Haifeng Xu <haifeng.xu@shopee.com>
Signed-off-by: Peter Zijlstra (Intel) <peterz@infradead.org>
Reviewed-by: Frederic Weisbecker <frederic@kernel.org>
Acked-by: Mark Rutland <mark.rutland@arm.com>
Cc: stable@vger.kernel.org
Link: https://lkml.kernel.org/r/20240513103948.33570-1-haifeng.xu@shopee.com
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
commit 7e2c1e4b34 upstream.
When lockdep is enabled, the for_each_sibling_event(sibling, event)
macro checks that event->ctx->mutex is held. When creating a new group
leader event, we call perf_event_validate_size() on a partially
initialized event where event->ctx is NULL, and so when
for_each_sibling_event() attempts to check event->ctx->mutex, we get a
splat, as reported by Lucas De Marchi:
WARNING: CPU: 8 PID: 1471 at kernel/events/core.c:1950 __do_sys_perf_event_open+0xf37/0x1080
This only happens for a new event which is its own group_leader, and in
this case there cannot be any sibling events. Thus it's safe to skip the
check for siblings, which avoids having to make invasive and ugly
changes to for_each_sibling_event().
Avoid the splat by bailing out early when the new event is its own
group_leader.
Fixes: 382c27f4ed ("perf: Fix perf_event_validate_size()")
Closes: https://lore.kernel.org/lkml/20231214000620.3081018-1-lucas.demarchi@intel.com/
Closes: https://lore.kernel.org/lkml/ZXpm6gQ%2Fd59jGsuW@xpf.sh.intel.com/
Reported-by: Lucas De Marchi <lucas.demarchi@intel.com>
Reported-by: Pengfei Xu <pengfei.xu@intel.com>
Signed-off-by: Mark Rutland <mark.rutland@arm.com>
Signed-off-by: Peter Zijlstra (Intel) <peterz@infradead.org>
Link: https://lkml.kernel.org/r/20231215112450.3972309-1-mark.rutland@arm.com
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
[ Upstream commit 382c27f4ed ]
Budimir noted that perf_event_validate_size() only checks the size of
the newly added event, even though the sizes of all existing events
can also change due to not all events having the same read_format.
When we attach the new event, perf_group_attach(), we do re-compute
the size for all events.
Fixes: a723968c0e ("perf: Fix u16 overflows")
Reported-by: Budimir Markovic <markovicbudimir@gmail.com>
Signed-off-by: Peter Zijlstra (Intel) <peterz@infradead.org>
Signed-off-by: Sasha Levin <sashal@kernel.org>
[ Upstream commit 54aee5f15b ]
When perf-record with a large AUX area, e.g 4GB, it fails with:
#perf record -C 0 -m ,4G -e arm_spe_0// -- sleep 1
failed to mmap with 12 (Cannot allocate memory)
and it reveals a WARNING with __alloc_pages():
------------[ cut here ]------------
WARNING: CPU: 44 PID: 17573 at mm/page_alloc.c:5568 __alloc_pages+0x1ec/0x248
Call trace:
__alloc_pages+0x1ec/0x248
__kmalloc_large_node+0xc0/0x1f8
__kmalloc_node+0x134/0x1e8
rb_alloc_aux+0xe0/0x298
perf_mmap+0x440/0x660
mmap_region+0x308/0x8a8
do_mmap+0x3c0/0x528
vm_mmap_pgoff+0xf4/0x1b8
ksys_mmap_pgoff+0x18c/0x218
__arm64_sys_mmap+0x38/0x58
invoke_syscall+0x50/0x128
el0_svc_common.constprop.0+0x58/0x188
do_el0_svc+0x34/0x50
el0_svc+0x34/0x108
el0t_64_sync_handler+0xb8/0xc0
el0t_64_sync+0x1a4/0x1a8
'rb->aux_pages' allocated by kcalloc() is a pointer array which is used to
maintains AUX trace pages. The allocated page for this array is physically
contiguous (and virtually contiguous) with an order of 0..MAX_ORDER. If the
size of pointer array crosses the limitation set by MAX_ORDER, it reveals a
WARNING.
So bail out early with -ENOMEM if the request AUX area is out of bound,
e.g.:
#perf record -C 0 -m ,4G -e arm_spe_0// -- sleep 1
failed to mmap with 12 (Cannot allocate memory)
Signed-off-by: Shuai Xue <xueshuai@linux.alibaba.com>
Signed-off-by: Peter Zijlstra (Intel) <peterz@infradead.org>
Signed-off-by: Ingo Molnar <mingo@kernel.org>
Signed-off-by: Sasha Levin <sashal@kernel.org>
[ Upstream commit f06cc667f7 ]
Namhyung reported that bd27568117 ("perf: Rewrite core context handling")
regresses context switch overhead when perf-cgroup is in use together
with 'slow' PMUs like uncore.
Specifically, perf_cgroup_switch()'s perf_ctx_disable() /
ctx_sched_out() etc.. all iterate the full list of active PMUs for
that CPU, even if they don't have cgroup events.
Previously there was cgrp_cpuctx_list which linked the relevant PMUs
together, but that got lost in the rework. Instead of re-instruducing
a similar list, let the perf_event_pmu_context iteration skip those
that do not have cgroup events. This avoids growing multiple versions
of the perf_event_pmu_context iteration.
Measured performance (on a slightly different patch):
Before)
$ taskset -c 0 ./perf bench sched pipe -l 10000 -G AAA,BBB
# Running 'sched/pipe' benchmark:
# Executed 10000 pipe operations between two processes
Total time: 0.901 [sec]
90.128700 usecs/op
11095 ops/sec
After)
$ taskset -c 0 ./perf bench sched pipe -l 10000 -G AAA,BBB
# Running 'sched/pipe' benchmark:
# Executed 10000 pipe operations between two processes
Total time: 0.065 [sec]
6.560100 usecs/op
152436 ops/sec
Fixes: bd27568117 ("perf: Rewrite core context handling")
Reported-by: Namhyung Kim <namhyung@kernel.org>
Debugged-by: Namhyung Kim <namhyung@kernel.org>
Signed-off-by: Peter Zijlstra (Intel) <peterz@infradead.org>
Link: https://lkml.kernel.org/r/20231009210425.GC6307@noisy.programming.kicks-ass.net
Signed-off-by: Sasha Levin <sashal@kernel.org>
