sched/fair: Document the slow path and fast path in select_task_rq_fair
authorBarry Song <song.bao.hua@hisilicon.com>
Sat, 16 Oct 2021 11:11:09 +0000 (19:11 +0800)
committerPeter Zijlstra <peterz@infradead.org>
Tue, 7 Dec 2021 14:14:10 +0000 (15:14 +0100)
All People I know including myself took a long time to figure out that
typical wakeup will always go to fast path and never go to slow path
except WF_FORK and WF_EXEC.

Vincent reminded me once in a linaro meeting and made me understand
slow path won't happen for WF_TTWU. But my other friends repeatedly
wasted a lot of time on testing this path like me before I reminded
them.

So obviously the code needs some document.

Signed-off-by: Barry Song <song.bao.hua@hisilicon.com>
Signed-off-by: Peter Zijlstra (Intel) <peterz@infradead.org>
Link: https://lkml.kernel.org/r/20211016111109.5559-1-21cnbao@gmail.com
kernel/sched/fair.c

index 06722188df49a8136e70e53e8c810af0d43224f1..f34f2f344fe9fecccd981a709b2e23511660b7be 100644 (file)
@@ -6917,6 +6917,11 @@ select_task_rq_fair(struct task_struct *p, int prev_cpu, int wake_flags)
                        break;
                }
 
+               /*
+                * Usually only true for WF_EXEC and WF_FORK, as sched_domains
+                * usually do not have SD_BALANCE_WAKE set. That means wakeup
+                * will usually go to the fast path.
+                */
                if (tmp->flags & sd_flag)
                        sd = tmp;
                else if (!want_affine)