On 6/26/25 13:16, Moon Hee Lee wrote:
The step_after_suspend_test verifies that the system successfully suspended and resumed by setting a timerfd and checking whether the timer fully expired. However, this method is unreliable due to timing races.
In practice, the system may take time to enter suspend, during which the timer may expire just before or during the transition. As a result, the remaining time after resume may show non-zero nanoseconds, even if suspend/resume completed successfully. This leads to false test failures.
Replace the timer-based check with a read from /sys/power/suspend_stats/success. This counter is incremented only after a full suspend/resume cycle, providing a reliable and race-free indicator.
Also remove the unused file descriptor for /sys/power/state, which remained after switching to a system() call to trigger suspend [1].
[1] https://lore.kernel.org/all/20240930224025.2858767-1-yifei.l.liu@oracle.com/
Fixes: c66be905cda2 ("selftests: breakpoints: use remaining time to check if suspend succeed") Signed-off-by: Moon Hee Lee moonhee.lee.ca@gmail.com
Applied to linux-kselftest next branch for Linux 6.17-rc1
thanks, -- Shuah