Hi Vincent, On 11/27/23 14:32, Vincent Guittot wrote:
Following comment in [1], I prepared a patch to remove UTIL_EST_FASTUP. This enables us to simplify util_est behavior as proposed in patch 2. [1] https://lore.kernel.org/lkml/CAKfTPtCAZWp7tRgTpwJmyEAkyN65acmYrfu9naEUpBZVWNTcQA@xxxxxxxxxxxxxx/ Vincent Guittot (2): sched/fair: Remove SCHED_FEAT(UTIL_EST_FASTUP, true) sched/fair: Simplify util_est Documentation/scheduler/schedutil.rst | 7 +-- include/linux/sched.h | 35 ++---------- kernel/sched/debug.c | 7 +-- kernel/sched/fair.c | 81 ++++++++++----------------- kernel/sched/features.h | 1 - kernel/sched/pelt.h | 4 +- 6 files changed, 43 insertions(+), 92 deletions(-)
I recovered my pixel6 and applied these changes. No power regression in Jankbench. No performance regression in GB5. Better score in Chrome running Speedometer +3..5%. The code looks much more clean, without the 'struct util_est' (we will have to adjust our trace events to that change but it's worth). Also, I was a bit surprised that the UTIL_EST_FASTUP wasn't helping that much comparing to that new 'runnable' signal for the underestimation corner case... Reviewed-and-tested-by: Lukasz Luba <lukasz.luba@xxxxxxx>
