sched_ext: Make scx_rq_online() also test cpu_active() in addition to SCX_RQ_ONLINE
authorTejun Heo <tj@kernel.org>
Wed, 7 Aug 2024 22:13:38 +0000 (12:13 -1000)
committerTejun Heo <tj@kernel.org>
Thu, 8 Aug 2024 23:38:19 +0000 (13:38 -1000)
commit991ef53a4832941c8130008ef35c66ec88c7fa0f
tree41b595da506dce8877f3c5e4106f26575fc3c23d
parent72763ea3d45c7f9fd69b825468afbf4d11c5ffc2
sched_ext: Make scx_rq_online() also test cpu_active() in addition to SCX_RQ_ONLINE

scx_rq_online() currently only tests SCX_RQ_ONLINE. This isn't fully correct
- e.g. consume_dispatch_q() uses task_run_on_remote_rq() which tests
scx_rq_online() to see whether the current rq can run the task, and, if so,
calls consume_remote_task() to migrate the task to @rq. While the test
itself was done while locking @rq, @rq can be temporarily unlocked by
consume_remote_task() and nothing prevents SCX_RQ_ONLINE from going offline
before the migration takes place.

To address the issue, add cpu_active() test to scx_rq_online(). There is a
synchronize_rcu() between cpu_active() being cleared and the rq going
offline, so if an on-going scheduling operation sees cpu_active(), the
associated rq is guaranteed to not go offline until the scheduling operation
is complete.

Signed-off-by: Tejun Heo <tj@kernel.org>
Fixes: 60c27fb59f6c ("sched_ext: Implement sched_ext_ops.cpu_online/offline()")
Acked-by: David Vernet <void@manifault.com>
kernel/sched/ext.c