On 12/02/21 15:17, Vincent Guittot wrote: > Start to update last_blocked_load_update_tick to reduce the possibility > of another cpu starting the update one more time >
IIUC this can happen if e.g. a CPU is busy updating load in update_blocked_averages() while another enters update_nohz_stats() for the same rq. They'll be serialized by the RQ lock, but both can still enter update_blocked_averages() at roughly the same time. Shouldn't then the rq->last_blocked_load_update_tick check be either deferred to or re-done within update_blocked_averages(), with the rq lock held?
