[V2,0/8] PM / OPP: Multiple regulator support

Message ID 20161024042659.GD24749@vireshk-i7
State New
Headers show

Commit Message

Viresh Kumar Oct. 24, 2016, 4:26 a.m.
On 23-10-16, 20:08, Dave Gerlach wrote:
> Overall this series looks good to me apart from a few small things. Most

> importantly I was able to get a working implementation using two regulators

> on ti dra7xx platform with proper sequencing built on top of this series. We

> have cpu regulator and Adaptive body bias (abb) regulator that must be

> scaled in a certain order before or after clock scaling and I was able to

> implement a rough custom set_rate to perform this and ran some dvfs stress

> tests that all worked fine.


Thanks for testing it buddy.

> First comment, I think the platform specific set_rate is a good place to

> hook in for adaptive voltage scaling as well. I was able to implement TI

> Class0 AVS in the same code by using the requested transition voltage as a

> reference and programming AVS voltage using that, along with scaling the

> additional regulators in sequence (the original multi regulator

> functionality). 


Hmm, interesting..

> I would think some people would want to use this even with

> single regulator platforms, no?


Maybe, but I would like to see such user code first. It may be possible to
handle much of AVS stuff in core so that everyone isn't required to do it.

> This raises some concerns about dependencies/probe sequencing. Right now we

> just need to make sure the cpufreq-dt driver probes after we have called

> _set_regulators, but if our platform code fails cpufreq-dt currently will

> treat this as no regulator needed for the platform and operate without one,

> which will likely hang the system. Is there a good way to to guarantee this

> doesn't happen? My main concern is that if we plan to provide a platform

> specific set-rate function, we should have a way to indicate this and

> prevent things from progressing if it isn't yet ready.

> 

> Again, overall I think it solves the multi regulator problem, and it works

> well for AVS as well. For the series:

> 

> Tested-by: Dave Gerlach <d-gerlach@ti.com>


Thanks.

For the concern you shared about, does the below patch fix it ? I will include
that in V3 then.

-------------------------8<-------------------------

From: Viresh Kumar <viresh.kumar@linaro.org>

Date: Mon, 24 Oct 2016 09:45:30 +0530
Subject: [PATCH] PM / OPP: Don't assume platform doesn't have regulators

If the regulators aren't set explicitly by the platform, the OPP core
assumes that the platform doesn't have any regulator and uses the
clk-only callback.

If the platform failed to register a regulator with the core, then this
can turn out to be a dangerous assumption as the OPP core will try to
change clk without changing regulators.

Handle that properly by making sure that the DT didn't had any entries
for supply voltages as well.

Signed-off-by: Viresh Kumar <viresh.kumar@linaro.org>

---
 drivers/base/power/opp/core.c | 12 +++++++++++-
 1 file changed, 11 insertions(+), 1 deletion(-)

-- 
viresh
--
To unsubscribe from this list: send the line "unsubscribe linux-pm" in
the body of a message to majordomo@vger.kernel.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html

Comments

Dave Gerlach Oct. 25, 2016, 9:13 p.m. | #1
Hi,
On 10/23/2016 11:26 PM, Viresh Kumar wrote:
> On 23-10-16, 20:08, Dave Gerlach wrote:

>> Overall this series looks good to me apart from a few small things. Most

>> importantly I was able to get a working implementation using two regulators

>> on ti dra7xx platform with proper sequencing built on top of this series. We

>> have cpu regulator and Adaptive body bias (abb) regulator that must be

>> scaled in a certain order before or after clock scaling and I was able to

>> implement a rough custom set_rate to perform this and ran some dvfs stress

>> tests that all worked fine.

>

> Thanks for testing it buddy.

>

>> First comment, I think the platform specific set_rate is a good place to

>> hook in for adaptive voltage scaling as well. I was able to implement TI

>> Class0 AVS in the same code by using the requested transition voltage as a

>> reference and programming AVS voltage using that, along with scaling the

>> additional regulators in sequence (the original multi regulator

>> functionality).

>

> Hmm, interesting..

>

>> I would think some people would want to use this even with

>> single regulator platforms, no?

>

> Maybe, but I would like to see such user code first. It may be possible to

> handle much of AVS stuff in core so that everyone isn't required to do it.


Ok, I think it would be a logical next step to look at once this series 
gets accepted. For now, the particular implementation I did just looks 
up an optimized value for the requested voltage from a register and 
programs the optimal value instead of the requested voltage.

>

>> This raises some concerns about dependencies/probe sequencing. Right now we

>> just need to make sure the cpufreq-dt driver probes after we have called

>> _set_regulators, but if our platform code fails cpufreq-dt currently will

>> treat this as no regulator needed for the platform and operate without one,

>> which will likely hang the system. Is there a good way to to guarantee this

>> doesn't happen? My main concern is that if we plan to provide a platform

>> specific set-rate function, we should have a way to indicate this and

>> prevent things from progressing if it isn't yet ready.

>>

>> Again, overall I think it solves the multi regulator problem, and it works

>> well for AVS as well. For the series:

>>

>> Tested-by: Dave Gerlach <d-gerlach@ti.com>

>

> Thanks.

>

> For the concern you shared about, does the below patch fix it ? I will include

> that in V3 then.


I think what you have shared below is a good safety check but if I 
rename the regulator properties in the DT for the cpu (to vdd and vbb, 
meaning cpufreq detects no regulator) and do *not* call 
dev_pm_opp_set_regulators before cpufreq-dt probes we fail before we 
even get to that point:

[16.946] cpu cpu0: opp_parse_supplies: Invalid number of elements in 
opp-microvolt property (6) with supplies (1)
[16.967] cpu cpu0: _of_add_opp_table_v2: Failed to add OPP, -22
[16.982] cpu cpu0: dev_pm_opp_get_opp_count: OPP table not found (-19)
[16.982] cpu cpu0: OPP table is not ready, deferring probe

This failure is because opp_parse_supplies assumes a count of 1 
regulator if no regulators at all are present and then hard fails if too 
many voltages have been passed for each OPP. It seems we need a check 
much earlier similar to what you suggested below to allow us to defer if 
an OPP has supplied voltages but no regulator has been registered with 
the system. I think this is reasonable even for the 1 regulator case, 
no? If we have passed voltages then we presumably are hoping to use them 
with a regulator, and if no regulators are present, OPP framework should 
defer.

cpufreq-dt won't handle this properly as is, but now that the opp core 
is evolving perhaps it makes sense to modify the resources_available 
check slightly to rely on the OPP core rather than just a dummy 
regulator_get_optional to see if the regulator is ready.

Regards,
Dave

>

> -------------------------8<-------------------------

>

> From: Viresh Kumar <viresh.kumar@linaro.org>

> Date: Mon, 24 Oct 2016 09:45:30 +0530

> Subject: [PATCH] PM / OPP: Don't assume platform doesn't have regulators

>

> If the regulators aren't set explicitly by the platform, the OPP core

> assumes that the platform doesn't have any regulator and uses the

> clk-only callback.

>

> If the platform failed to register a regulator with the core, then this

> can turn out to be a dangerous assumption as the OPP core will try to

> change clk without changing regulators.

>

> Handle that properly by making sure that the DT didn't had any entries

> for supply voltages as well.

>

> Signed-off-by: Viresh Kumar <viresh.kumar@linaro.org>

> ---

>  drivers/base/power/opp/core.c | 12 +++++++++++-

>  1 file changed, 11 insertions(+), 1 deletion(-)

>

> diff --git a/drivers/base/power/opp/core.c b/drivers/base/power/opp/core.c

> index b69908b74ed6..fb4250532180 100644

> --- a/drivers/base/power/opp/core.c

> +++ b/drivers/base/power/opp/core.c

> @@ -737,7 +737,17 @@ int dev_pm_opp_set_rate(struct device *dev, unsigned long target_freq)

>

>  	/* Only frequency scaling */

>  	if (!regulators) {

> -		rcu_read_unlock();

> +		/*

> +		 * DT contained supply ratings? Consider platform failed to set

> +		 * regulators.

> +		 */

> +		if (unlikely(opp->supplies[0].u_volt)) {

> +			rcu_read_unlock();

> +			dev_err(dev, "%s: Regulator not registered with OPP core\n",

> +				__func__);

> +			return -EINVAL;

> +		}

> +

>  		return _generic_opp_set_rate_clk_only(dev, clk, old_freq, freq);

>  	}

>

>


--
To unsubscribe from this list: send the line "unsubscribe linux-pm" in
the body of a message to majordomo@vger.kernel.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Viresh Kumar Oct. 26, 2016, 3:21 a.m. | #2
On 25-10-16, 16:13, Dave Gerlach wrote:
> I think what you have shared below is a good safety check but if I rename

> the regulator properties in the DT for the cpu (to vdd and vbb, meaning

> cpufreq detects no regulator) and do *not* call dev_pm_opp_set_regulators

> before cpufreq-dt probes we fail before we even get to that point:

> 

> [16.946] cpu cpu0: opp_parse_supplies: Invalid number of elements in

> opp-microvolt property (6) with supplies (1)

> [16.967] cpu cpu0: _of_add_opp_table_v2: Failed to add OPP, -22

> [16.982] cpu cpu0: dev_pm_opp_get_opp_count: OPP table not found (-19)

> [16.982] cpu cpu0: OPP table is not ready, deferring probe

> 

> This failure is because opp_parse_supplies assumes a count of 1 regulator if

> no regulators at all are present and then hard fails if too many voltages

> have been passed for each OPP.


Exactly. And yes this is intentional.

> It seems we need a check much earlier similar

> to what you suggested below to allow us to defer if an OPP has supplied

> voltages but no regulator has been registered with the system. I think this

> is reasonable even for the 1 regulator case, no?


No.

OPP core needs to know about regulators only if the user drivers want it to
manage DVFS. It is still possible for cpufreq drivers to use OPP framework for
managing the tables, but do the real DVFS stuff themselves. That's why it is not
compulsory in the code to set regulator names.

And its only wrong if dev_pm_opp_set_rate() is called without first setting the
regulators..

> cpufreq-dt won't handle this properly as is, but now that the opp core is

> evolving perhaps it makes sense to modify the resources_available check

> slightly to rely on the OPP core rather than just a dummy

> regulator_get_optional to see if the regulator is ready.


I am not sure yet on what to change there. You mean regarding multiple
regulators?

-- 
viresh
--
To unsubscribe from this list: send the line "unsubscribe linux-pm" in
the body of a message to majordomo@vger.kernel.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html

Patch

diff --git a/drivers/base/power/opp/core.c b/drivers/base/power/opp/core.c
index b69908b74ed6..fb4250532180 100644
--- a/drivers/base/power/opp/core.c
+++ b/drivers/base/power/opp/core.c
@@ -737,7 +737,17 @@  int dev_pm_opp_set_rate(struct device *dev, unsigned long target_freq)
 
 	/* Only frequency scaling */
 	if (!regulators) {
-		rcu_read_unlock();
+		/*
+		 * DT contained supply ratings? Consider platform failed to set
+		 * regulators.
+		 */
+		if (unlikely(opp->supplies[0].u_volt)) {
+			rcu_read_unlock();
+			dev_err(dev, "%s: Regulator not registered with OPP core\n",
+				__func__);
+			return -EINVAL;
+		}
+
 		return _generic_opp_set_rate_clk_only(dev, clk, old_freq, freq);
 	}