[PATCH 0/2] opp: core: add regulators enable and disable

[PATCH 0/2] opp: core: add regulators enable and disable

[Viresh Kumar]

Hi,

This series reintroduces the usage of regulator_enable/disable() to the
OPP core after the previous attempt was reverted [1] shortly after getting
applied. This time the regulator is enabled only after it is configured
by the OPP core.

Marek, Kamil and Clément: Can you guys please test this out and report
if this doesn't work as expected ?

--
viresh

[1] https://lore.kernel.org/lkml/20191017102758.8104-1-m.szyprowski@samsung.com/

Kamil Konieczny (1):
  opp: core: add regulators enable and disable

Viresh Kumar (1):
  opp: Reorder the code for !target_freq case

 drivers/opp/core.c | 39 ++++++++++++++++++++++++++++++++++-----
 drivers/opp/opp.h  |  2 ++
 2 files changed, 36 insertions(+), 5 deletions(-)

-- 
2.25.0.rc1.19.g042ed3e048af

[PATCH 1/2] opp: Reorder the code for !target_freq case

[Viresh Kumar]

Reorder the code a bit to make it more readable. Add additional comment
as well.

Signed-off-by: Viresh Kumar <viresh.kumar@linaro.org>
---
 drivers/opp/core.c | 13 +++++++++----
 1 file changed, 9 insertions(+), 4 deletions(-)

diff --git a/drivers/opp/core.c b/drivers/opp/core.c
index e4f01e7771a2..dda8164fad56 100644
--- a/drivers/opp/core.c
+++ b/drivers/opp/core.c
@@ -817,15 +817,20 @@ int dev_pm_opp_set_rate(struct device *dev, unsigned long target_freq)
 	}
 
 	if (unlikely(!target_freq)) {
-		if (opp_table->required_opp_tables) {
-			ret = _set_required_opps(dev, opp_table, NULL);
-		} else if (!_get_opp_count(opp_table)) {
+		/*
+		 * Some drivers need to support cases where some platforms may
+		 * have OPP table for the device, while others don't and
+		 * opp_set_rate() just needs to behave like clk_set_rate().
+		 */
+		if (!_get_opp_count(opp_table))
 			return 0;
-		} else {
+
+		if (!opp_table->required_opp_tables) {
 			dev_err(dev, "target frequency can't be 0\n");
 			ret = -EINVAL;
 		}
 
+		ret = _set_required_opps(dev, opp_table, NULL);
 		goto put_opp_table;
 	}
 
-- 
2.25.0.rc1.19.g042ed3e048af

[PATCH 2/2] opp: core: add regulators enable and disable

[Viresh Kumar]

From: Kamil Konieczny <k.konieczny@samsung.com>

Add enable regulators to dev_pm_opp_set_regulators() and disable
regulators to dev_pm_opp_put_regulators(). Even if bootloader
leaves regulators enabled, they should be enabled in kernel in
order to increase the reference count.

Signed-off-by: Kamil Konieczny <k.konieczny@samsung.com>
[ Viresh: Enable the regulator only after it is programmed and add a
	  flag to track its status. ]
Signed-off-by: Viresh Kumar <viresh.kumar@linaro.org>
---
 drivers/opp/core.c | 28 ++++++++++++++++++++++++++--
 drivers/opp/opp.h  |  2 ++
 2 files changed, 28 insertions(+), 2 deletions(-)

diff --git a/drivers/opp/core.c b/drivers/opp/core.c
index dda8164fad56..da6533d5526f 100644
--- a/drivers/opp/core.c
+++ b/drivers/opp/core.c
@@ -664,7 +664,7 @@ static inline int _generic_set_opp_clk_only(struct device *dev, struct clk *clk,
 	return ret;
 }
 
-static int _generic_set_opp_regulator(const struct opp_table *opp_table,
+static int _generic_set_opp_regulator(struct opp_table *opp_table,
 				      struct device *dev,
 				      unsigned long old_freq,
 				      unsigned long freq,
@@ -699,6 +699,18 @@ static int _generic_set_opp_regulator(const struct opp_table *opp_table,
 			goto restore_freq;
 	}
 
+	/*
+	 * Enable the regulator after setting its voltages, otherwise it breaks
+	 * some boot-enabled regulators.
+	 */
+	if (unlikely(!opp_table->regulator_enabled)) {
+		ret = regulator_enable(reg);
+		if (ret < 0)
+			dev_warn(dev, "Failed to enable regulator: %d", ret);
+		else
+			opp_table->regulator_enabled = true;
+	}
+
 	return 0;
 
 restore_freq:
@@ -825,11 +837,16 @@ int dev_pm_opp_set_rate(struct device *dev, unsigned long target_freq)
 		if (!_get_opp_count(opp_table))
 			return 0;
 
-		if (!opp_table->required_opp_tables) {
+		if (!opp_table->required_opp_tables && !opp_table->regulators) {
 			dev_err(dev, "target frequency can't be 0\n");
 			ret = -EINVAL;
 		}
 
+		if (opp_table->regulator_enabled) {
+			regulator_disable(opp_table->regulators[0]);
+			opp_table->regulator_enabled = false;
+		}
+
 		ret = _set_required_opps(dev, opp_table, NULL);
 		goto put_opp_table;
 	}
@@ -1675,6 +1692,13 @@ void dev_pm_opp_put_regulators(struct opp_table *opp_table)
 	/* Make sure there are no concurrent readers while updating opp_table */
 	WARN_ON(!list_empty(&opp_table->opp_list));
 
+	if (opp_table->regulator_enabled) {
+		for (i = opp_table->regulator_count - 1; i >= 0; i--)
+			regulator_disable(opp_table->regulators[i]);
+
+		opp_table->regulator_enabled = false;
+	}
+
 	for (i = opp_table->regulator_count - 1; i >= 0; i--)
 		regulator_put(opp_table->regulators[i]);
 
diff --git a/drivers/opp/opp.h b/drivers/opp/opp.h
index d14e27102730..b4b2713d84f1 100644
--- a/drivers/opp/opp.h
+++ b/drivers/opp/opp.h
@@ -144,6 +144,7 @@ enum opp_table_access {
  * @clk: Device's clock handle
  * @regulators: Supply regulators
  * @regulator_count: Number of power supply regulators. Its value can be -1
+ * @regulator_enabled: Set to true if regulators were previously enabled.
  * (uninitialized), 0 (no opp-microvolt property) or > 0 (has opp-microvolt
  * property).
  * @genpd_performance_state: Device's power domain support performance state.
@@ -189,6 +190,7 @@ struct opp_table {
 	struct clk *clk;
 	struct regulator **regulators;
 	int regulator_count;
+	bool regulator_enabled;
 	bool genpd_performance_state;
 	bool is_genpd;
 
-- 
2.25.0.rc1.19.g042ed3e048af

Re: [PATCH 0/2] opp: core: add regulators enable and disable

[Marek Szyprowski]

Hi Viresh,

On 15.05.2020 09:57, Viresh Kumar wrote:
> This series reintroduces the usage of regulator_enable/disable() to the
> OPP core after the previous attempt was reverted [1] shortly after getting
> applied. This time the regulator is enabled only after it is configured
> by the OPP core.
>
> Marek, Kamil and Clément: Can you guys please test this out and report
> if this doesn't work as expected ?

Works fine for my test cases, especially Samsung Chromebook Peach-Pit/Pi 
still boots fine. Feel free to add to the both patches:

Tested-by: Marek Szyprowski <m.szyprowski@samsung.com>

Thanks!

Best regards
-- 
Marek Szyprowski, PhD
Samsung R&D Institute Poland

Re: [PATCH 0/2] opp: core: add regulators enable and disable

[Clément Péron]

Hi Viresh,

On Fri, 15 May 2020 at 09:57, Viresh Kumar <viresh.kumar@linaro.org> wrote:
>
> Hi,
>
> This series reintroduces the usage of regulator_enable/disable() to the
> OPP core after the previous attempt was reverted [1] shortly after getting
> applied. This time the regulator is enabled only after it is configured
> by the OPP core.
>
> Marek, Kamil and Clément: Can you guys please test this out and report
> if this doesn't work as expected ?

I have reviewed the patch and it seems fine for my use case.
Unfortunately I can't test it until next week.

Acked-by: Clément Péron <peron.clem@gmail.com>

Regards,
Clement


>
> --
> viresh
>
> [1] https://lore.kernel.org/lkml/20191017102758.8104-1-m.szyprowski@samsung.com/
>
> Kamil Konieczny (1):
>   opp: core: add regulators enable and disable
>
> Viresh Kumar (1):
>   opp: Reorder the code for !target_freq case
>
>  drivers/opp/core.c | 39 ++++++++++++++++++++++++++++++++++-----
>  drivers/opp/opp.h  |  2 ++
>  2 files changed, 36 insertions(+), 5 deletions(-)
>
> --
> 2.25.0.rc1.19.g042ed3e048af
>

Re: [PATCH 0/2] opp: core: add regulators enable and disable

[Clément Péron]

Hi,

On Fri, 15 May 2020 at 14:00, Clément Péron <peron.clem@gmail.com> wrote:
>
> Hi Viresh,
>
> On Fri, 15 May 2020 at 09:57, Viresh Kumar <viresh.kumar@linaro.org> wrote:
> >
> > Hi,
> >
> > This series reintroduces the usage of regulator_enable/disable() to the
> > OPP core after the previous attempt was reverted [1] shortly after getting
> > applied. This time the regulator is enabled only after it is configured
> > by the OPP core.
> >
> > Marek, Kamil and Clément: Can you guys please test this out and report
> > if this doesn't work as expected ?
>
> I have reviewed the patch and it seems fine for my use case.
> Unfortunately I can't test it until next week.

Ok, before the patch the regulator was released by regulator_late_cleanup() :
[   33.756849] vdd-gpu: disabling

Now it works fine and the vdd-gpu is no more disabled.

Tested-by: Clément Péron <peron.clem@gmail.com>

Regards,
Clement

>
> Acked-by: Clément Péron <peron.clem@gmail.com>
>
> Regards,
> Clement
>
>
> >
> > --
> > viresh
> >
> > [1] https://lore.kernel.org/lkml/20191017102758.8104-1-m.szyprowski@samsung.com/
> >
> > Kamil Konieczny (1):
> >   opp: core: add regulators enable and disable
> >
> > Viresh Kumar (1):
> >   opp: Reorder the code for !target_freq case
> >
> >  drivers/opp/core.c | 39 ++++++++++++++++++++++++++++++++++-----
> >  drivers/opp/opp.h  |  2 ++
> >  2 files changed, 36 insertions(+), 5 deletions(-)
> >
> > --
> > 2.25.0.rc1.19.g042ed3e048af
> >

Re: [PATCH 0/2] opp: core: add regulators enable and disable

[Clément Péron]

Hi Viresh,

On Thu, 21 May 2020 at 14:23, Clément Péron <peron.clem@gmail.com> wrote:
>
> Hi,
>
> On Fri, 15 May 2020 at 14:00, Clément Péron <peron.clem@gmail.com> wrote:
> >
> > Hi Viresh,
> >
> > On Fri, 15 May 2020 at 09:57, Viresh Kumar <viresh.kumar@linaro.org> wrote:
> > >
> > > Hi,
> > >
> > > This series reintroduces the usage of regulator_enable/disable() to the
> > > OPP core after the previous attempt was reverted [1] shortly after getting
> > > applied. This time the regulator is enabled only after it is configured
> > > by the OPP core.
> > >
> > > Marek, Kamil and Clément: Can you guys please test this out and report
> > > if this doesn't work as expected ?
> >
> > I have reviewed the patch and it seems fine for my use case.
> > Unfortunately I can't test it until next week.
>
> Ok, before the patch the regulator was released by regulator_late_cleanup() :
> [   33.756849] vdd-gpu: disabling
>
> Now it works fine and the vdd-gpu is no more disabled.

Today, I compiled my kernel without any program requiring GPU
computing at boot. This makes the dev_pm_opp_set_rate() to never be
called and so the regulator is not enabled before the regulator
framework switches off all the regulators that haven't been enabled.

Unfortunately switching off the GPU regulator makes my board hang..

I'm not sure what is the best approach to fix this.

Is it required that the dev_pm_opp_set_rate() must be called one time
at the GPU driver init?

Panfost already calls devfreq_recommended_opp() and dev_pm_opp_put()
but that doesn't trigger dev_pm_opp_set_rate().

Thanks for your help,
BR,
Clement


>
> Tested-by: Clément Péron <peron.clem@gmail.com>
>
> Regards,
> Clement
>
> >
> > Acked-by: Clément Péron <peron.clem@gmail.com>
> >
> > Regards,
> > Clement
> >
> >
> > >
> > > --
> > > viresh
> > >
> > > [1] https://lore.kernel.org/lkml/20191017102758.8104-1-m.szyprowski@samsung.com/
> > >
> > > Kamil Konieczny (1):
> > >   opp: core: add regulators enable and disable
> > >
> > > Viresh Kumar (1):
> > >   opp: Reorder the code for !target_freq case
> > >
> > >  drivers/opp/core.c | 39 ++++++++++++++++++++++++++++++++++-----
> > >  drivers/opp/opp.h  |  2 ++
> > >  2 files changed, 36 insertions(+), 5 deletions(-)
> > >
> > > --
> > > 2.25.0.rc1.19.g042ed3e048af
> > >

Re: [PATCH 0/2] opp: core: add regulators enable and disable

[Viresh Kumar]

On 03-09-22, 22:35, Clément Péron wrote:
> Today, I compiled my kernel without any program requiring GPU
> computing at boot. This makes the dev_pm_opp_set_rate() to never be
> called and so the regulator is not enabled before the regulator
> framework switches off all the regulators that haven't been enabled.
> 
> Unfortunately switching off the GPU regulator makes my board hang..

Why does the board hang? I mean the kernel should boot fine with the
GPU disabled, isn't it ? Or is the regulator shared with some other
critical resource, or something else.

> I'm not sure what is the best approach to fix this.
> 
> Is it required that the dev_pm_opp_set_rate() must be called one time
> at the GPU driver init?

Right now, Yes. And it looks like the right approach as well.

> Panfost already calls devfreq_recommended_opp() and dev_pm_opp_put()
> but that doesn't trigger dev_pm_opp_set_rate().

Can you also point to your code ? Which file are you working on ?

-- 
viresh

Re: [PATCH 0/2] opp: core: add regulators enable and disable

[Clément Péron]

Hi Viresh,

On Mon, 5 Sept 2022 at 06:35, Viresh Kumar <viresh.kumar@linaro.org> wrote:
>
> On 03-09-22, 22:35, Clément Péron wrote:
> > Today, I compiled my kernel without any program requiring GPU
> > computing at boot. This makes the dev_pm_opp_set_rate() to never be
> > called and so the regulator is not enabled before the regulator
> > framework switches off all the regulators that haven't been enabled.
> >
> > Unfortunately switching off the GPU regulator makes my board hang..
>
> Why does the board hang? I mean the kernel should boot fine with the
> GPU disabled, isn't it ? Or is the regulator shared with some other
> critical resource, or something else.

The regulator is dedicated to the GPU and the board could certainly
run without GPU, the issue is that the driver (here panfrost) may do
some regular access to GPU memory (I suppose).

>
> > I'm not sure what is the best approach to fix this.
> >
> > Is it required that the dev_pm_opp_set_rate() must be called one time
> > at the GPU driver init?
>
> Right now, Yes. And it looks like the right approach as well.
>
> > Panfost already calls devfreq_recommended_opp() and dev_pm_opp_put()
> > but that doesn't trigger dev_pm_opp_set_rate().
>
> Can you also point to your code ? Which file are you working on ?

The code I'm pointing is panfrost_devfreq_init() in
drivers/gpu/drm/panfrost/panfrost_devfreq.c

Regards,
Clement

>
> --
> viresh

Re: [PATCH 0/2] opp: core: add regulators enable and disable

[Viresh Kumar]

On 05-09-22, 10:28, Clément Péron wrote:
> Hi Viresh,
> 
> On Mon, 5 Sept 2022 at 06:35, Viresh Kumar <viresh.kumar@linaro.org> wrote:
> >
> > On 03-09-22, 22:35, Clément Péron wrote:
> > > Today, I compiled my kernel without any program requiring GPU
> > > computing at boot.

I thought you disabled most of GPU stuff here and so thought panfrost
devfreq will be gone too :)

> > > This makes the dev_pm_opp_set_rate() to never be
> > > called and so the regulator is not enabled before the regulator
> > > framework switches off all the regulators that haven't been enabled.
> > >
> > > Unfortunately switching off the GPU regulator makes my board hang..
> >
> > Why does the board hang? I mean the kernel should boot fine with the
> > GPU disabled, isn't it ? Or is the regulator shared with some other
> > critical resource, or something else.
> 
> The regulator is dedicated to the GPU and the board could certainly
> run without GPU, the issue is that the driver (here panfrost) may do
> some regular access to GPU memory (I suppose).

So we do need GPU to be functional at this point ? i.e. both
clk/regulators should be enabled.

> The code I'm pointing is panfrost_devfreq_init() in
> drivers/gpu/drm/panfrost/panfrost_devfreq.c

I think it would be better to call dev_pm_opp_set_opp() after calling
devfreq_recommended_opp() in this driver, since you already have the
OPP known to you.

-- 
viresh