[PATCH] cpuidle: teo: Do not return a disabled idle state

[Xueqin Luo]

If idle state 0 is disabled and none of the enabled idle states satisfies
the PM QoS latency constraint, constraint_idx in teo_select() stays at its
initial value of 0, because the loop that updates it starts at state 1,
and so the candidate state index is capped at state 0, which is disabled,
and that state is returned to the caller.

The cpuidle core does not validate the returned state index against the
per-state disable flags, so the CPU ends up entering the disabled state 0.

Address this by falling back to the shallowest enabled idle state, found
by the same loop that computes the candidate, whenever the state selected
by the latency constraint check turns out to be disabled.

Note that if none of the enabled idle states meets the latency constraint,
choosing the shallowest enabled one is the least intrusive option, and it
is also consistent with the behavior of the menu governor.

Fixes: c410a9a142f1 ("cpuidle: teo: Change the main idle state selection logic")
Signed-off-by: Xueqin Luo <luoxueqin@xxxxxxxxxx>
---
 drivers/cpuidle/governors/teo.c | 8 ++++++++
 1 file changed, 8 insertions(+)

diff --git a/drivers/cpuidle/governors/teo.c b/drivers/cpuidle/governors/teo.c
index ac43b9b013b3..b4b14c86b2a9 100644
--- a/drivers/cpuidle/governors/teo.c
+++ b/drivers/cpuidle/governors/teo.c
@@ -433,6 +433,14 @@ static int teo_select(struct cpuidle_driver *drv, struct cpuidle_device *dev,
 	if (idx > constraint_idx)
 		idx = constraint_idx;
 
+	/*
+	 * If the state selected above is disabled, which is only possible if
+	 * it is state 0 (all of the enabled states violate the latency
+	 * constraint in that case), fall back to the shallowest enabled one.
+	 */
+	if (dev->states_usage[idx].disable)
+		idx = idx0;
+
 	/*
 	 * If the tick has not been stopped and either the candidate state is
 	 * state 0 or its target residency is low enough, there is basically
-- 
2.43.0