[PATCH sched_ext/for-7.3] sched_ext: skip ops.set_weight() for disabled tasks

[PATCH sched_ext/for-7.3] sched_ext: skip ops.set_weight() for disabled tasks

[Kuba Piecuch]

When switching a task's sched_class away from sched_ext, we get the
following sequence of events in __sched_setscheduler():

sched_change_begin()
  switched_from_scx()
    scx_disable_task(p)
      ops.disable(p)
__setscheduler_params()
  set_load_weight()
    reweight_task_scx(p)
      ops.set_weight(p)
p->sched_class = next_class;
sched_change_end()
  ...

Notably, ops.set_weight() is called _after_ ops.disable().
This violates the expected semantics of the callbacks, the expectation
being that ops.disable() can only be followed by ops.exit_task() or
ops.enable().

Skipping the weight adjustment for disabled tasks should be harmless
since the weight will be recalculated in scx_enable_task() if the task
ever rejoins SCX.

Signed-off-by: Kuba Piecuch <jpiecuch@google.com>
---
 kernel/sched/ext/ext.c | 11 +++++++++++
 1 file changed, 11 insertions(+)

diff --git a/kernel/sched/ext/ext.c b/kernel/sched/ext/ext.c
index 5241b55a58ec..b75d8d71ef4d 100644
--- a/kernel/sched/ext/ext.c
+++ b/kernel/sched/ext/ext.c
@@ -3673,6 +3673,17 @@ static void reweight_task_scx(struct rq *rq, struct task_struct *p,
 	if (task_dead_and_done(p))
 		return;
 
+	/*
+	 * When switching sched_class away from SCX, reweight_task_scx()
+	 * is called _after_ scx_disable_task(). Skip calling ops.set_weight()
+	 * since the BPF scheduler may have already forgotten the task in
+	 * ops.disable().
+	 * p->scx.weight will be recalculated in scx_enable_task() if the task
+	 * ever returns to SCX class.
+	 */
+	if (scx_get_task_state(p) != SCX_TASK_ENABLED)
+		return;
+
 	p->scx.weight = sched_weight_to_cgroup(scale_load_down(lw->weight));
 	if (SCX_HAS_OP(sch, set_weight))
 		SCX_CALL_OP_TASK(sch, set_weight, rq, p, p->scx.weight);
-- 
2.55.0.795.g602f6c329a-goog

Re: [PATCH sched_ext/for-7.3] sched_ext: skip ops.set_weight() for disabled tasks

[Tejun Heo]

Hello,

Applied to sched_ext/for-7.2-fixes, thanks.

Added a Fixes: tag and stable Cc. The reversed set_weight()/disable()
ordering is a regression from the commit that moved ops.disable() ahead
of reweight_task_scx(), so it goes back to v6.19:

  Fixes: 637b0682821b ("sched: Fold sched_class::switch{ing,ed}_{to,from}() into the change pattern")
  Cc: stable@vger.kernel.org # v6.19+

Thanks.

--
tejun