On Tue 29 Jun 01:34 CDT 2021, Stephen Boyd wrote:

Quoting Bjorn Andersson (2021-06-25 15:54:14)

quoted

As GDSCs are registered and found to be already enabled
gdsc_toggle_logic() will be invoked for votable GDSCs and ensure that
the vote is matching the hardware state. Part of this the related
regulator will be enabled.

But for non-votable GDSCs the regulator and GDSC status will be out of
sync and as the GDSC is later disabled regulator_disable() will face an
unbalanced enable-count, or something might turn off the supply under
the feet of the GDSC.

So ensure that the regulator is enabled even for non-votable GDSCs.

Cc: stable@vger.kernel.org
Fixes: 37416e554961 ("clk: qcom: gdsc: Handle GDSC regulator supplies")
Signed-off-by: Bjorn Andersson <redacted>
---
 drivers/clk/qcom/gdsc.c | 11 +++++++++--
 1 file changed, 9 insertions(+), 2 deletions(-)

diff --git a/drivers/clk/qcom/gdsc.c b/drivers/clk/qcom/gdsc.c
index 51ed640e527b..f7e7759cdb90 100644
--- a/drivers/clk/qcom/gdsc.c
+++ b/drivers/clk/qcom/gdsc.c

@@ -359,10 +359,17 @@ static int gdsc_init(struct gdsc *sc)
 
        /*
         * Votable GDSCs can be ON due to Vote from other masters.
-        * If a Votable GDSC is ON, make sure we have a Vote.
+        * If a Votable GDSC is ON, make sure we have a Vote. If
+        * non-votable, ensure that the supply is kept enabled (as
+        * is done by gdsc_enable).
         */
-       if ((sc->flags & VOTABLE) && on)
+       if ((sc->flags & VOTABLE) && on) {
                gdsc_enable(&sc->pd);
+       } else if (on) {
+               ret = regulator_enable(sc->rsupply);
+               if (ret < 0)
+                       return ret;

Looking at this makes me think we've messed something up with
gdsc_enable() being called or cherry-picking the regulator enable (and
other stuff in this gdsc_init()) out of the enable path. Maybe we should
have a followup patch that replaces the gdsc_enable() with
gdsc_toggle_logic(sc, GDSC_ON) so that we know it isn't doing anything
else during init like asserting a reset when presumably all we want to
do is toggle the enable bit to assert our vote.

And I notice that we already call gdsc_toggle_logic() in gdsc_init(), so
then we'll have a double regulator_enable() in the case of PWRSTS_ON?
And then if the flag is ALWAYS_ON we'll call regulator_enable() yet
again, but luckily only if it isn't on initially, phew! This code is
quite twisted.

It would be super nice to make it more like

	if (on) {
		/* It was on in hardware, sync kernel state */
		regulator_enable();

		if (votable)
			write bit, why do any wait?

		if (retain ff)
			write bit
	} else if (always_on) {
		/* Force on */
		gdsc_enable();
		on = true;
	}

	if (on || ...)

This does look cleaner, I will dig through the logic here once more
before figuring out what to do for v2.

Thanks,
Bjorn

quoted

+       }
 
        /*
         * Make sure the retain bit is set if the GDSC is already on, otherwise