sched: fix high wake up latencies with FAIR_USER_SCHED
commit296825cbe14d4c95ee9c41ca5824f7487bfb4d9d
authorSrivatsa Vaddagiri <vatsa@linux.vnet.ibm.com>
Thu, 31 Jan 2008 21:45:22 +0000 (31 22:45 +0100)
committerIngo Molnar <mingo@elte.hu>
Thu, 31 Jan 2008 21:45:22 +0000 (31 22:45 +0100)
treec1d608ee1dd774f1d726c12873feff8a29422fff
parent095031052b4c03a7c8ffa51cbab031e442c4f8b7
sched: fix high wake up latencies with FAIR_USER_SCHED

The reason why we are getting better wakeup latencies for
!FAIR_USER_SCHED is because of this snippet of code in place_entity():

if (!initial) {
/* sleeps upto a single latency don't count. */
if (sched_feat(NEW_FAIR_SLEEPERS) && entity_is_task(se))
     ^^^^^^^^^^^^^^^^^^
vruntime -= sysctl_sched_latency;

/* ensure we never gain time by being placed backwards. */
vruntime = max_vruntime(se->vruntime, vruntime);
}

NEW_FAIR_SLEEPERS feature gives credit for sleeping only to tasks and
not group-level entities. With the patch attached, I could see that
wakeup latencies with FAIR_USER_SCHED are restored to the same level as
!FAIR_USER_SCHED.

Signed-off-by: Ingo Molnar <mingo@elte.hu>
kernel/sched_fair.c