]> git.ipfire.org Git - thirdparty/kernel/linux.git/commit
perf: Fix sample vs do_exit()
authorPeter Zijlstra <peterz@infradead.org>
Thu, 5 Jun 2025 10:31:45 +0000 (12:31 +0200)
committerPeter Zijlstra <peterz@infradead.org>
Thu, 5 Jun 2025 12:37:51 +0000 (14:37 +0200)
commit4f6fc782128355931527cefe3eb45338abd8ab39
tree26dc5fffde5db0a533a9ee7ef206016c3582cf0a
parent86aa94cd50b138be0dd872b0779fa3036e641881
perf: Fix sample vs do_exit()

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: c5ebcedb566e ("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
kernel/events/core.c
kernel/exit.c