tkernel: irqlatency: hold the cpu lock when removing the hotplug state · Pull Request !974 · OpenCloudOS/OpenCloudOS-Kernel

## Problem

trace_latency_exit() (kernel/tkernel/irqlatency/irqlatency.c:752) calls
cpuhp_remove_state() without holding the cpu read lock. On a kernel
built with CONFIG_PROVE_RCU/lockdep, unloading the module triggers:

WARNING: CPU: ... lockdep_assert_cpus_held()
    in __cpuhp_remove_state_cpuslocked() (kernel/cpu.c:2587)

## Root Cause

cpuhp_remove_state() documents "The caller needs to hold cpus read
locked" and asserts it: its internals may wait for the per-cpu hotplug
threads that run the teardown callback (irqlatency_cpu_offline), which
stops the per-cpu timers before free_percpu().

The other control paths in the same file, enable_write() and
trace_stack_write(), already take cpus_read_lock() around their cpuhp
operations; the module exit path added in the hotplug rework missed it.

## Fix

Take cpus_read_lock() around cpuhp_remove_state() in
trace_latency_exit(), matching the locking order
(control_lock -> cpus_read_lock) already used by the write paths in the
same file.

## Testing

- `make kernel/tkernel/irqlatency/`: compiles clean; irqlatency.ko
  generated.
- `make W=1 kernel/tkernel/irqlatency/`: 0 warnings.
- `make -j16`: `Kernel: arch/x86/boot/bzImage is ready`.
- `make dist-check-commit`: commit looks OK.
- `scripts/checkpatch.pl --terse --no-tree`: 0 errors.

## Risk

low: adds a read lock around an existing call in the module exit path
only; no functional change, no ABI change.

类型	指派人员	状态
审查	zoucao Huang Cun Jianping Liu	已完成 (0/0人)
测试	Jianping Liu	已完成 (0/0人)

GVPOpenCloudOS / OpenCloudOS-Kernel

搜索帮助

GVP OpenCloudOS / OpenCloudOS-Kernel