| Message ID | 1517944987-343-1-git-send-email-vincent.guittot@linaro.org |
|---|---|
| Headers | show |
| Series | Update blocked load | expand |
On Tue, Feb 06, 2018 at 08:23:04PM +0100, Vincent Guittot wrote: > This patchset applies on top of Peter's sched/esting branch minus the last 2 commit: > 56eb46798b33 ("sched: Clean up nohz enter/exit") So what will update the blocked load after we wake from a total idle? I don't see anything here that would do that (yet?).
On 6 February 2018 at 22:25, Peter Zijlstra <peterz@infradead.org> wrote: > On Tue, Feb 06, 2018 at 08:23:04PM +0100, Vincent Guittot wrote: >> This patchset applies on top of Peter's sched/esting branch minus the last 2 commit: >> 56eb46798b33 ("sched: Clean up nohz enter/exit") > > So what will update the blocked load after we wake from a total idle? > > I don't see anything here that would do that (yet?). The fast wake up path uses only idle state and runnable load average. The later is normally up to date and null ( thanks to the propagation of runnable) because of the fact that we were totally idle. For the other "slow" path, we need to update the utilization of the CPUs when we use them (and the load as well even if we don't use it). This will be part of another patch as the overhead need to be evaluated That should be enough to take the right decision at wake up from a long total idle state Then for a complete update of the load metric, we can either wait for the next tick or kick an ilb at wake up (what you don't like).
This patchset applies on top of Peter's sched/esting branch minus the last 2 commit: 56eb46798b33 ("sched: Clean up nohz enter/exit") v2: - minor naming updates Vincent Guittot (3): sched: Stop nohz stats when decayed sched: reduce the periodic update duration sched: update blocked load when newly idle kernel/sched/fair.c | 203 ++++++++++++++++++++++++++++++++++++++++++--------- kernel/sched/sched.h | 1 + 2 files changed, 169 insertions(+), 35 deletions(-) -- 2.7.4