On 2023/10/10 Peter Zijlstra wrote:
On Sat, Oct 07, 2023 at 04:44:46PM +0800, Hao Jia wrote:
That is, would not something like the below make more sense?
If we understand correctly, this may not work.
After applying this patch, the following situation will trigger the rq->clock_update_flags < RQCF_ACT_SKIP warning.
If rq_clock_skip_update() is called before __schedule(), so RQCF_REQ_SKIP of rq->clock_update_flags is set.
__schedule() { rq_lock(rq, &rf); [rq->clock_update_flags is RQCF_REQ_SKIP] rq->clock_update_flags <<= 1; update_rq_clock(rq); [rq->clock_update_flags is RQCF_ACT_SKIP]
- rq->clock_update_flags &= ~(RQCF_ACT_SKIP|RQCF_REQ_SKIP);
- At this time, rq->clock_update_flags = 0; *
Fixed easily enough, just change to:
rq->clock_updated_flags = RQCF_UPDATED;
Thanks for your suggestions and help, I revised the commit message and sent patch v2.
https://lore.kernel.org/all/20231012090003.11450-1-jiahao.os@bytedance.com/
Please review again.
Thanks, Hao
pick_next_task_fair set_next_entity update_load_avg assert_clock_updated() <---}