summaryrefslogtreecommitdiff
path: root/kernel
AgeCommit message (Collapse)Author
2016-11-12Bump MuQSS version to 0.1404.7-muqss-140Con Kolivas
2016-11-12Revert all non-scheduler related timeout changes in the kernel.ckolivas
Too many assumptions are made in the kernel about timeout behaviour and many calls are not properly made freezer safe. The scheduler changes should not own this issue.
2016-11-11Fix build failure.Con Kolivas
2016-11-11vtime_init_idle is missing from init_idle in muqss.Con Kolivas
2016-11-11Only force a direct schedule in __set_cpus_allowed_ptr on kthreads.ckolivas
2016-11-11Remove unnecessary calls to get/put online_cpus in sched_setaffinity.ckolivas
2016-11-05Bump MuQSS version to 0.1354.7-muqss-135Con Kolivas
2016-11-05Kick out of the idle loop if there are softirqs pending, avoiding trying to ↵Con Kolivas
call nohz idle and service them on exit from schedule.
2016-11-05Remove check for codepath which is no longer hit.Con Kolivas
2016-11-05Don't reschedule balance across nodes unless a CPU is idle.Con Kolivas
2016-11-05Remove unnecessary check in edt()Con Kolivas
2016-11-04Revert "Unset base is idle in next_timer_event based on hrtimers as well."Con Kolivas
This reverts commit ab4b15e59636f6af12136d71cc96950a9048c222. Not relevant for 4.7 and causes build failure.
2016-11-04Convert msleep to use hrtimers when active.Con Kolivas
2016-11-04Special case only schedule_timeouts of 1 to use highres timeouts of 1ms.Con Kolivas
2016-11-04Add functions for min and msec hrtimeout un/interruptible.Con Kolivas
2016-11-01Use high resolution timers in place of regular ones whenever possible on ↵Con Kolivas
muqss since they are mandatory and low Hz is recommended.
2016-10-31Unset base is idle in next_timer_event based on hrtimers as well.Con Kolivas
2016-10-31Add a dodgy workaround to do pending softirqs if we detect them upon a ↵Con Kolivas
runqueue going idle.
2016-10-29Bump MuQSS version to 0.1204.7-muqss-120Con Kolivas
2016-10-28Don't warn on calling hrexpiry from another CPU.Con Kolivas
2016-10-28When one CPU has all the load, make it report its load ramping up quickly to ↵Con Kolivas
benefit more from turbo modes.
2016-10-28Start ticks on a CPU going offlineCon Kolivas
2016-10-28Fix various combinations of irq accounting and dynticks config by relying on ↵Con Kolivas
the existing code in cputime.c and removing it from MuQSS.c.
2016-10-28QNR count is redundant now on MuQSS.Con Kolivas
2016-10-28Use niffies in update load avg and include iowait in load.Con Kolivas
2016-10-28Set default HZ to 100Con Kolivas
2016-10-27Create full nohz support regardless of load when highres timers are active ↵ckolivas
since we are not dependent on the timer tick at all.
2016-10-27Add high res timeslice expiry when available, increasing resolution of ↵ckolivas
highres timers to 100us to make low latency possible with low Hz values (i.e. 100Hz), retaining the benefit of the increased throughput low Hz values provide.
2016-10-25Account for unaccounted CPU time in wunt before distributing time_slice to ↵Con Kolivas
child.
2016-10-25Lock rq in do_set_cpus_allowed before resched_task.Con Kolivas
2016-10-25Revert "There is no need to expire timeslice in sched yield now that the ↵Con Kolivas
latency/accounting bug has been fixed." This reverts commit 04a890aaf847e7eceb65b568958379d35dc7ad3c. Wrong again.
2016-10-25Revert task_cpu test against wake_cpu in set_task_cpu.Con Kolivas
2016-10-24Bump MuQSS version to 0.1164.7-muqss-116Con Kolivas
2016-10-24There is no need to expire timeslice in sched yield now that the ↵Con Kolivas
latency/accounting bug has been fixed.
2016-10-24Fix niffies not being updated in scheduler_tick leading to unfair CPU ↵Con Kolivas
distribution and latency outliers. Bug tracked down with the extensive help of Serge Belyshev.
2016-10-23Set deadline and time slice on idle for first forked tasks (worker threads) ↵Con Kolivas
to inherit sensible values.
2016-10-23Revert the damage caused by 6b45f1f363d7f6959b648cf49252f378022d11c6 and ↵Con Kolivas
dependent patches.
2016-10-23Check_siblings was accidentally removed from schedule() leading to ↵Con Kolivas
misbehaving niced/idleprio tasks on SMT.
2016-10-22Bump MuQSS version to 0.1154.7-muqss-115Con Kolivas
2016-10-22Fix incorrect cpu being returned in select_best_cpu which would lead to boot ↵Con Kolivas
failures.
2016-10-22Make the cpu_idle_map cacheline aligned.Con Kolivas
2016-10-22Resched when needed after calling do_set_cpus_allowed from outside the ↵Con Kolivas
scheduler code, consolidating smp_send_reschedule to only be called if the remote CPU is online.
2016-10-22We have nr_running now for rq load.Con Kolivas
2016-10-22Provide unique low overhead edt function for UP.Con Kolivas
2016-10-22Optimise randomlevel in skip list to be simple ffb and rarer for higher levelsCon Kolivas
2016-10-22Remove the last remnants of the global runqueue, moving all variables to ↵Con Kolivas
per-runqueue except for the cpu idle map.
2016-10-21Bump MuQSS version to 0.1144.7-muqss-114Con Kolivas
2016-10-21Convert sub-tick accounting to nanosecond precision to cope with tasks that ↵Con Kolivas
run for less than 100us.
2016-10-21Rely on niffies being monotonic and use jiffies as sanity check for minimum ↵Con Kolivas
forward progress without scattered use of niffy_diff.
2016-10-21Use niffies for accounting wherever possibleCon Kolivas
generated by cgit v1.2.3 (git 2.39.1) at 2024-07-02 16:00:50 +0000