Regulators marked with "regulator-always-on" or "regulator-boot-on"
as well as an "off-on-delay-us", may run into cycling issues that are
hard to detect.

This is caused by the "last_off" state not being initialized in this
case.

Fix the "last_off" initialization by setting it to the current kernel
time upon initialization, regardless of always_on/boot_on state.

Signed-off-by: Christian Kohlschütter <christian@kohlschutter.com>
---
 drivers/regulator/core.c | 5 +++--
 1 file changed, 3 insertions(+), 2 deletions(-)

diff --git a/drivers/regulator/core.c b/drivers/regulator/core.c
index 1e54a833f2c..398c8d6afd4 100644
--- a/drivers/regulator/core.c
+++ b/drivers/regulator/core.c
@@ -1565,6 +1565,9 @@ static int set_machine_constraints(struct regulator_dev *rdev)
 			rdev->constraints->always_on = true;
 	}
 
+	if (rdev->desc->off_on_delay)
+		rdev->last_off = ktime_get();
+
 	/* If the constraints say the regulator should be on at this point
 	 * and we have control then make sure it is enabled.
 	 */
@@ -1592,8 +1595,6 @@ static int set_machine_constraints(struct regulator_dev *rdev)
 
 		if (rdev->constraints->always_on)
 			rdev->use_count++;
-	} else if (rdev->desc->off_on_delay) {
-		rdev->last_off = ktime_get();
 	}
 
 	print_constraints(rdev);
-- 
2.36.2

On Tue, Jul 19, 2022 at 06:49:44PM +0000, Christian Kohlschütter wrote:
> Regulators marked with "regulator-always-on" or "regulator-boot-on"
> as well as an "off-on-delay-us", may run into cycling issues that are
> hard to detect.

I think this already got applied, I had another go at persuading things
to cope which seemed to work - not 100% sure what was going on, git
seemed less forgiving than raw patch here.  It didn't look like a
rebasing issue, it looked like the umlaut was upsetting git.  Should be
commit 218320fec29430438016f88dd4fbebfa1b95ad8d in my tree.

> Am 19.07.2022 um 21:38 schrieb Mark Brown <broonie@kernel.org>:
> 
> On Tue, Jul 19, 2022 at 06:49:44PM +0000, Christian Kohlschütter wrote:
>> Regulators marked with "regulator-always-on" or "regulator-boot-on"
>> as well as an "off-on-delay-us", may run into cycling issues that are
>> hard to detect.
> 
> I think this already got applied, I had another go at persuading things
> to cope which seemed to work - not 100% sure what was going on, git
> seemed less forgiving than raw patch here.  It didn't look like a
> rebasing issue, it looked like the umlaut was upsetting git.  Should be
> commit 218320fec29430438016f88dd4fbebfa1b95ad8d in my tree.

Yes, I just got your emails after I resubmitted. I had thought my email client had messed things up, and I had resent it with a proper git send-email setup.
Thanks again!