Re: [PATCH v3] selftests/cgroup: add PSI pressure trigger and validation tests

在 2026/8/13 21:54, [email protected] 写道:
> Thank you for your contribution! Sashiko AI review found 1 potential issue(s) to consider:
> - [Medium] The child processes spawned to generate CPU pressure execute an unconditional infinite loop (`for (;;);`). If the test is interrupted (e.g., via SIGINT or a CI timeout sending SIGTERM), the parent process will exit immediately without executing the cleanup block (`cg_killall()`). This leaves `ncpus + 1` child processes running forever, consuming 100% CPU on all cores and effectively locking up the system.
> --
> 
> --- Patch [1]: [PATCH v3] selftests/cgroup: add PSI pressure trigger and validation tests ---
> commit a3c72bd52fb11f36beba6ee125a9c7100e986a5a
> Author: Tao Cui <[email protected]>
> selftests/cgroup: add PSI pressure trigger and validation tests
> 
> This commit adds per-resource trigger smoke tests, a cgroup.pressure
> hide/show toggle test, and a CPU-pressure trigger test using
> over-subscription to the cgroup selftests.
> 
> Link: https://lore.kernel.org/all/[email protected]/
> Link: https://lore.kernel.org/all/[email protected]/
> 
>> diff --git a/tools/testing/selftests/cgroup/test_psi.c b/tools/testing/selftests/cgroup/test_psi.c
>> new file mode 100644
>> index 0000000000000..cfb412f614588
>> --- /dev/null
>> +++ b/tools/testing/selftests/cgroup/test_psi.c
> [ ... ]
>> +/* A child that burns CPU forever; stopped by cg_killall() in the parent. */
>> +static int hog_cpu(const char *cgroup, void *arg)
>> +{
>> +	for (;;)
>> +		;
> 
> [Severity: Medium]
> Will this leave runaway CPU hogs consuming 100% CPU on all cores if the test
> is interrupted?
> 
> If the parent process is interrupted by a signal (like SIGINT or a CI runner
> timeout sending SIGTERM) while waiting for poll(), it will exit immediately
> without executing the cg_killall() cleanup block.
> 
> Could a fallback termination condition be added here, such as
> prctl(PR_SET_PDEATHSIG, SIGKILL) or checking getppid(), to prevent orphaned
> processes from locking up the system?
>
Agreed. The hogs run forever, so interrupting the parent during poll()
skips cg_killall() and orphans them. I'll add prctl(PR_SET_PDEATHSIG,
SIGKILL) at the start of each hog so the kernel reaps them if the
parent dies -- for v4.
 
>> +	return 0;
>> +}
>