sched: Avoid throttle_cfs_rq() racing with period_timer stopping
authorBen Segall <bsegall@google.com>
Wed, 16 Oct 2013 18:16:32 +0000 (11:16 -0700)
committerIngo Molnar <mingo@kernel.org>
Tue, 29 Oct 2013 11:02:32 +0000 (12:02 +0100)
commitf9f9ffc237dd924f048204e8799da74f9ecf40cf
tree81ed0c3435dfe54781d0f120d3a5938d571bacd1
parent0ac9b1c21874d2490331233b3242085f8151e166
sched: Avoid throttle_cfs_rq() racing with period_timer stopping

throttle_cfs_rq() doesn't check to make sure that period_timer is running,
and while update_curr/assign_cfs_runtime does, a concurrently running
period_timer on another cpu could cancel itself between this cpu's
update_curr and throttle_cfs_rq(). If there are no other cfs_rqs running
in the tg to restart the timer, this causes the cfs_rq to be stranded
forever.

Fix this by calling __start_cfs_bandwidth() in throttle if the timer is
inactive.

(Also add some sched_debug lines for cfs_bandwidth.)

Tested: make a run/sleep task in a cgroup, loop switching the cgroup
between 1ms/100ms quota and unlimited, checking for timer_active=0 and
throttled=1 as a failure. With the throttle_cfs_rq() change commented out
this fails, with the full patch it passes.

Signed-off-by: Ben Segall <bsegall@google.com>
Signed-off-by: Peter Zijlstra <peterz@infradead.org>
Cc: pjt@google.com
Link: http://lkml.kernel.org/r/20131016181632.22647.84174.stgit@sword-of-the-dawn.mtv.corp.google.com
Signed-off-by: Ingo Molnar <mingo@kernel.org>
kernel/sched/debug.c
kernel/sched/fair.c