nohz: Fix ondemand governor not always ramping up on 100% load.
PATCH 1/1 nohz: Fix ondemand governor not always ramping up on 100% load.
On systems that perform the context switch with interrupts enabled (e.g. ARM) get_cpu_idle_time_us could get stuck returning the last idle time when the cpu is busy. If an interrupt occurs while switching to the idle thread, and this interrupt wakes up a thread, need-resched will be set on the idle thread, but since we have not finished switching to the idle thread tick_nohz_stop_sched_tick will be called. (idle_cpu is true since the head of the runqueue is the idle thread, but need_resched returns false since we are still running in the context of the old thread)
Fix this by not calling tick_nohz_start_idle when returning from an interrupt while the idle thread is in its schedule phase. (!ts->inidle)
--- kernel/time/tick-sched.c | 2 +- 1 files changed, 1 insertions(+), 1 deletions(-)
kernel/time/tick-sched.c +1 -1
@@ -222,7 +222,6 @@ void tick_nohz_stop_sched_tick(int inidle) cpu =smp_processor_id (); ts = &per_cpu (tick_cpu_sched, cpu); - now =tick_nohz_start_idle (ts);/* * If this cpu is offline and it is the one which updates
@@ -242,6 +241,7 @@ void tick_nohz_stop_sched_tick(int inidle)if (!inidle && !ts->inidle) goto end; + now =tick_nohz_start_idle (ts); ts->inidle =1 ;if (need_resched ())
Signature · 1 line
1.6.1