Re: [PATCH v2 3/4] cpufreq: mediatek: add Mediatek cpufreq driver
On 12 March 2015 at 16:45, Pi-Cheng Chen wrote: > The voltages of the two regulators need to be always under a limitation: > 100mV < Vsram - Vproc < 200mV > For now, I just calculate the OPPs of Vsram from OPPs of Vproc. > > Another thing I should mention, if the voltage difference of two adjacent > OPPs is greater than 100mV, we need to set the regulator to some > voltages which are not in OPP table considering the limitation above. > I think that will make it more difficult to model such flow in a generic > framework. Actually the problem is that OPP tables are insufficient for such cases, and we are adding work-arounds to fix that. Though it will get fixed with the new bindings we are adding. I am confused on what we should be doing here. Even if you write your own driver, you will be abusing DT with incorrect information. Even if you go around adding a new driver, I would like you to fallback to cpufreq-dt ones the new OPP bindings are in place. -- To unsubscribe from this list: send the line "unsubscribe linux-kernel" in the body of a message to majord...@vger.kernel.org More majordomo info at http://vger.kernel.org/majordomo-info.html Please read the FAQ at http://www.tux.org/lkml/
Re: [PATCH v2 3/4] cpufreq: mediatek: add Mediatek cpufreq driver
On 12 March 2015 at 16:45, Pi-Cheng Chen pi-cheng.c...@linaro.org wrote: The voltages of the two regulators need to be always under a limitation: 100mV Vsram - Vproc 200mV For now, I just calculate the OPPs of Vsram from OPPs of Vproc. Another thing I should mention, if the voltage difference of two adjacent OPPs is greater than 100mV, we need to set the regulator to some voltages which are not in OPP table considering the limitation above. I think that will make it more difficult to model such flow in a generic framework. Actually the problem is that OPP tables are insufficient for such cases, and we are adding work-arounds to fix that. Though it will get fixed with the new bindings we are adding. I am confused on what we should be doing here. Even if you write your own driver, you will be abusing DT with incorrect information. Even if you go around adding a new driver, I would like you to fallback to cpufreq-dt ones the new OPP bindings are in place. -- To unsubscribe from this list: send the line unsubscribe linux-kernel in the body of a message to majord...@vger.kernel.org More majordomo info at http://vger.kernel.org/majordomo-info.html Please read the FAQ at http://www.tux.org/lkml/
Re: [PATCH v2 3/4] cpufreq: mediatek: add Mediatek cpufreq driver
On Thu, Mar 12, 2015 at 5:28 PM, Viresh Kumar wrote: > On 11 March 2015 at 18:15, Mark Brown wrote: >> Ugh, no - that's a hideous bodge which is only going to create trouble >> later. Remember, DT is an ABI and should describe the hardware so if >> we're doing bodges that are visible there to shoehorn things onto our >> implementation that's bad. The concerns that Pi-Cheng had about what >> happens if the PMIC gets changed definitely seem relevant here too. > > Hmm.. > >> Why not just write a custom cpufreq driver if it's too hard to abstract? > > Hmm, probably all that can be solved with the new OPP bindings where > we can have support for multiple regulator or clock sources to the CPU. > > @Pi-cheng: How are you going to pass voltages for both the regulators > as OPPs today only support a single regulator, in case you have to write > your own driver. The voltages of the two regulators need to be always under a limitation: 100mV < Vsram - Vproc < 200mV For now, I just calculate the OPPs of Vsram from OPPs of Vproc. Another thing I should mention, if the voltage difference of two adjacent OPPs is greater than 100mV, we need to set the regulator to some voltages which are not in OPP table considering the limitation above. I think that will make it more difficult to model such flow in a generic framework. Best Regards, Pi-Cheng > > ___ > Linux-mediatek mailing list > linux-media...@lists.infradead.org > http://lists.infradead.org/mailman/listinfo/linux-mediatek -- To unsubscribe from this list: send the line "unsubscribe linux-kernel" in the body of a message to majord...@vger.kernel.org More majordomo info at http://vger.kernel.org/majordomo-info.html Please read the FAQ at http://www.tux.org/lkml/
Re: [PATCH v2 3/4] cpufreq: mediatek: add Mediatek cpufreq driver
On 11 March 2015 at 18:15, Mark Brown wrote: > Ugh, no - that's a hideous bodge which is only going to create trouble > later. Remember, DT is an ABI and should describe the hardware so if > we're doing bodges that are visible there to shoehorn things onto our > implementation that's bad. The concerns that Pi-Cheng had about what > happens if the PMIC gets changed definitely seem relevant here too. Hmm.. > Why not just write a custom cpufreq driver if it's too hard to abstract? Hmm, probably all that can be solved with the new OPP bindings where we can have support for multiple regulator or clock sources to the CPU. @Pi-cheng: How are you going to pass voltages for both the regulators as OPPs today only support a single regulator, in case you have to write your own driver. -- To unsubscribe from this list: send the line "unsubscribe linux-kernel" in the body of a message to majord...@vger.kernel.org More majordomo info at http://vger.kernel.org/majordomo-info.html Please read the FAQ at http://www.tux.org/lkml/
Re: [PATCH v2 3/4] cpufreq: mediatek: add Mediatek cpufreq driver
On 11 March 2015 at 18:15, Mark Brown broo...@kernel.org wrote: Ugh, no - that's a hideous bodge which is only going to create trouble later. Remember, DT is an ABI and should describe the hardware so if we're doing bodges that are visible there to shoehorn things onto our implementation that's bad. The concerns that Pi-Cheng had about what happens if the PMIC gets changed definitely seem relevant here too. Hmm.. Why not just write a custom cpufreq driver if it's too hard to abstract? Hmm, probably all that can be solved with the new OPP bindings where we can have support for multiple regulator or clock sources to the CPU. @Pi-cheng: How are you going to pass voltages for both the regulators as OPPs today only support a single regulator, in case you have to write your own driver. -- To unsubscribe from this list: send the line unsubscribe linux-kernel in the body of a message to majord...@vger.kernel.org More majordomo info at http://vger.kernel.org/majordomo-info.html Please read the FAQ at http://www.tux.org/lkml/
Re: [PATCH v2 3/4] cpufreq: mediatek: add Mediatek cpufreq driver
On Thu, Mar 12, 2015 at 5:28 PM, Viresh Kumar viresh.ku...@linaro.org wrote: On 11 March 2015 at 18:15, Mark Brown broo...@kernel.org wrote: Ugh, no - that's a hideous bodge which is only going to create trouble later. Remember, DT is an ABI and should describe the hardware so if we're doing bodges that are visible there to shoehorn things onto our implementation that's bad. The concerns that Pi-Cheng had about what happens if the PMIC gets changed definitely seem relevant here too. Hmm.. Why not just write a custom cpufreq driver if it's too hard to abstract? Hmm, probably all that can be solved with the new OPP bindings where we can have support for multiple regulator or clock sources to the CPU. @Pi-cheng: How are you going to pass voltages for both the regulators as OPPs today only support a single regulator, in case you have to write your own driver. The voltages of the two regulators need to be always under a limitation: 100mV Vsram - Vproc 200mV For now, I just calculate the OPPs of Vsram from OPPs of Vproc. Another thing I should mention, if the voltage difference of two adjacent OPPs is greater than 100mV, we need to set the regulator to some voltages which are not in OPP table considering the limitation above. I think that will make it more difficult to model such flow in a generic framework. Best Regards, Pi-Cheng ___ Linux-mediatek mailing list linux-media...@lists.infradead.org http://lists.infradead.org/mailman/listinfo/linux-mediatek -- To unsubscribe from this list: send the line unsubscribe linux-kernel in the body of a message to majord...@vger.kernel.org More majordomo info at http://vger.kernel.org/majordomo-info.html Please read the FAQ at http://www.tux.org/lkml/
Re: [PATCH v2 3/4] cpufreq: mediatek: add Mediatek cpufreq driver
On Wed, Mar 11, 2015 at 04:33:49PM +0530, Viresh Kumar wrote: > On mediatek platform, they need to configure two regulators in order to change > DVFS state of the big cluster. The generic cpufreq-dt driver and earlier OPP > bindings have support for a single regulator only and so what Pi-cheng tried > to do is, > - Configure one of the regulators using cpufreq-dt > - And other one using cpufreq frequency change notifiers > This looks awkward.. > What I suggested was to create another virtual regulator for CPU which will > eventually configure both the regulators. And so the question that such > virtual regulators are allowed or not. Ugh, no - that's a hideous bodge which is only going to create trouble later. Remember, DT is an ABI and should describe the hardware so if we're doing bodges that are visible there to shoehorn things onto our implementation that's bad. The concerns that Pi-Cheng had about what happens if the PMIC gets changed definitely seem relevant here too. signature.asc Description: Digital signature
Re: [PATCH v2 3/4] cpufreq: mediatek: add Mediatek cpufreq driver
On Wed, Mar 11, 2015 at 05:16:42PM +0530, Viresh Kumar wrote: > I think until the time we revive the voltage-domain stuff we need to support > mediatek's driver. And probably a virtual regulator is the best approach > unless someone else comes up with another idea. Why not just write a custom cpufreq driver if it's too hard to abstract? signature.asc Description: Digital signature
Re: [PATCH v2 3/4] cpufreq: mediatek: add Mediatek cpufreq driver
On 11 March 2015 at 17:12, Lucas Stach wrote: > Instead of creating virtual regulators I would be strongly in favor of > reviving the voltage-domain work. That would allow us to push all those > voltage dependencies we have seen on various SoCs into the domain > handling code and don't care about it in the drivers. > > In that case cpufreq-dt wouldn't control a regulator directly, but > request a specific voltage from the domain the CPUs are located in and > those in turn would control the regulators supplying them. I agree that it would be the right approach but who is going to do that stuff ? I think until the time we revive the voltage-domain stuff we need to support mediatek's driver. And probably a virtual regulator is the best approach unless someone else comes up with another idea. -- To unsubscribe from this list: send the line "unsubscribe linux-kernel" in the body of a message to majord...@vger.kernel.org More majordomo info at http://vger.kernel.org/majordomo-info.html Please read the FAQ at http://www.tux.org/lkml/
Re: [PATCH v2 3/4] cpufreq: mediatek: add Mediatek cpufreq driver
Am Mittwoch, den 11.03.2015, 16:33 +0530 schrieb Viresh Kumar: > On 11 March 2015 at 16:23, Mark Brown wrote: > > On Tue, Mar 10, 2015 at 08:20:43AM +0530, Viresh Kumar wrote: > > > > Please don't send upstream e-mail to my work account, I use this address > > pretty consistently for upstream. Upstream mail to my work account > > frequently ends up unread. > > Sorry about that, I did exactly opposite of this earlier :( > > >> On 6 March 2015 at 11:19, Pi-Cheng Chen wrote: > >> > On 5 March 2015 at 17:55, Viresh Kumar wrote: > > > >> > About putting > >> > those stuff into regulator driver, I think you mean creating a > >> > "virtual regulator > >> > device" and put all the voltage controlling complex into the driver, > >> > right? > >> > Maybe it's a good idea in this case, but I am sure if this kind of > >> > virtual regulator is acceptable. > > > >> @Mark: Is this allowed to create virtual regulator for a CPU ? > > > > I don't really know what the above means or what problem it's supposed > > to solve. > > On mediatek platform, they need to configure two regulators in order to change > DVFS state of the big cluster. The generic cpufreq-dt driver and earlier OPP > bindings have support for a single regulator only and so what Pi-cheng tried > to do is, > - Configure one of the regulators using cpufreq-dt > - And other one using cpufreq frequency change notifiers > > This looks awkward.. > > What I suggested was to create another virtual regulator for CPU which will > eventually configure both the regulators. And so the question that such > virtual regulators are allowed or not. Instead of creating virtual regulators I would be strongly in favor of reviving the voltage-domain work. That would allow us to push all those voltage dependencies we have seen on various SoCs into the domain handling code and don't care about it in the drivers. In that case cpufreq-dt wouldn't control a regulator directly, but request a specific voltage from the domain the CPUs are located in and those in turn would control the regulators supplying them. Regards, Lucas -- Pengutronix e.K. | Lucas Stach | Industrial Linux Solutions | http://www.pengutronix.de/ | -- To unsubscribe from this list: send the line "unsubscribe linux-kernel" in the body of a message to majord...@vger.kernel.org More majordomo info at http://vger.kernel.org/majordomo-info.html Please read the FAQ at http://www.tux.org/lkml/
Re: [PATCH v2 3/4] cpufreq: mediatek: add Mediatek cpufreq driver
On 11 March 2015 at 16:23, Mark Brown wrote: > On Tue, Mar 10, 2015 at 08:20:43AM +0530, Viresh Kumar wrote: > > Please don't send upstream e-mail to my work account, I use this address > pretty consistently for upstream. Upstream mail to my work account > frequently ends up unread. Sorry about that, I did exactly opposite of this earlier :( >> On 6 March 2015 at 11:19, Pi-Cheng Chen wrote: >> > On 5 March 2015 at 17:55, Viresh Kumar wrote: > >> > About putting >> > those stuff into regulator driver, I think you mean creating a >> > "virtual regulator >> > device" and put all the voltage controlling complex into the driver, right? >> > Maybe it's a good idea in this case, but I am sure if this kind of >> > virtual regulator is acceptable. > >> @Mark: Is this allowed to create virtual regulator for a CPU ? > > I don't really know what the above means or what problem it's supposed > to solve. On mediatek platform, they need to configure two regulators in order to change DVFS state of the big cluster. The generic cpufreq-dt driver and earlier OPP bindings have support for a single regulator only and so what Pi-cheng tried to do is, - Configure one of the regulators using cpufreq-dt - And other one using cpufreq frequency change notifiers This looks awkward.. What I suggested was to create another virtual regulator for CPU which will eventually configure both the regulators. And so the question that such virtual regulators are allowed or not. -- To unsubscribe from this list: send the line "unsubscribe linux-kernel" in the body of a message to majord...@vger.kernel.org More majordomo info at http://vger.kernel.org/majordomo-info.html Please read the FAQ at http://www.tux.org/lkml/
Re: [PATCH v2 3/4] cpufreq: mediatek: add Mediatek cpufreq driver
On Tue, Mar 10, 2015 at 08:20:43AM +0530, Viresh Kumar wrote: Please don't send upstream e-mail to my work account, I use this address pretty consistently for upstream. Upstream mail to my work account frequently ends up unread. > On 6 March 2015 at 11:19, Pi-Cheng Chen wrote: > > On 5 March 2015 at 17:55, Viresh Kumar wrote: > > About putting > > those stuff into regulator driver, I think you mean creating a > > "virtual regulator > > device" and put all the voltage controlling complex into the driver, right? > > Maybe it's a good idea in this case, but I am sure if this kind of > > virtual regulator is acceptable. > @Mark: Is this allowed to create virtual regulator for a CPU ? I don't really know what the above means or what problem it's supposed to solve. signature.asc Description: Digital signature
Re: [PATCH v2 3/4] cpufreq: mediatek: add Mediatek cpufreq driver
On 11 March 2015 at 16:23, Mark Brown broo...@kernel.org wrote: On Tue, Mar 10, 2015 at 08:20:43AM +0530, Viresh Kumar wrote: Please don't send upstream e-mail to my work account, I use this address pretty consistently for upstream. Upstream mail to my work account frequently ends up unread. Sorry about that, I did exactly opposite of this earlier :( On 6 March 2015 at 11:19, Pi-Cheng Chen pi-cheng.c...@linaro.org wrote: On 5 March 2015 at 17:55, Viresh Kumar viresh.ku...@linaro.org wrote: About putting those stuff into regulator driver, I think you mean creating a virtual regulator device and put all the voltage controlling complex into the driver, right? Maybe it's a good idea in this case, but I am sure if this kind of virtual regulator is acceptable. @Mark: Is this allowed to create virtual regulator for a CPU ? I don't really know what the above means or what problem it's supposed to solve. On mediatek platform, they need to configure two regulators in order to change DVFS state of the big cluster. The generic cpufreq-dt driver and earlier OPP bindings have support for a single regulator only and so what Pi-cheng tried to do is, - Configure one of the regulators using cpufreq-dt - And other one using cpufreq frequency change notifiers This looks awkward.. What I suggested was to create another virtual regulator for CPU which will eventually configure both the regulators. And so the question that such virtual regulators are allowed or not. -- To unsubscribe from this list: send the line unsubscribe linux-kernel in the body of a message to majord...@vger.kernel.org More majordomo info at http://vger.kernel.org/majordomo-info.html Please read the FAQ at http://www.tux.org/lkml/
Re: [PATCH v2 3/4] cpufreq: mediatek: add Mediatek cpufreq driver
On Tue, Mar 10, 2015 at 08:20:43AM +0530, Viresh Kumar wrote: Please don't send upstream e-mail to my work account, I use this address pretty consistently for upstream. Upstream mail to my work account frequently ends up unread. On 6 March 2015 at 11:19, Pi-Cheng Chen pi-cheng.c...@linaro.org wrote: On 5 March 2015 at 17:55, Viresh Kumar viresh.ku...@linaro.org wrote: About putting those stuff into regulator driver, I think you mean creating a virtual regulator device and put all the voltage controlling complex into the driver, right? Maybe it's a good idea in this case, but I am sure if this kind of virtual regulator is acceptable. @Mark: Is this allowed to create virtual regulator for a CPU ? I don't really know what the above means or what problem it's supposed to solve. signature.asc Description: Digital signature
Re: [PATCH v2 3/4] cpufreq: mediatek: add Mediatek cpufreq driver
On 11 March 2015 at 17:12, Lucas Stach l.st...@pengutronix.de wrote: Instead of creating virtual regulators I would be strongly in favor of reviving the voltage-domain work. That would allow us to push all those voltage dependencies we have seen on various SoCs into the domain handling code and don't care about it in the drivers. In that case cpufreq-dt wouldn't control a regulator directly, but request a specific voltage from the domain the CPUs are located in and those in turn would control the regulators supplying them. I agree that it would be the right approach but who is going to do that stuff ? I think until the time we revive the voltage-domain stuff we need to support mediatek's driver. And probably a virtual regulator is the best approach unless someone else comes up with another idea. -- To unsubscribe from this list: send the line unsubscribe linux-kernel in the body of a message to majord...@vger.kernel.org More majordomo info at http://vger.kernel.org/majordomo-info.html Please read the FAQ at http://www.tux.org/lkml/
Re: [PATCH v2 3/4] cpufreq: mediatek: add Mediatek cpufreq driver
On Wed, Mar 11, 2015 at 05:16:42PM +0530, Viresh Kumar wrote: I think until the time we revive the voltage-domain stuff we need to support mediatek's driver. And probably a virtual regulator is the best approach unless someone else comes up with another idea. Why not just write a custom cpufreq driver if it's too hard to abstract? signature.asc Description: Digital signature
Re: [PATCH v2 3/4] cpufreq: mediatek: add Mediatek cpufreq driver
On Wed, Mar 11, 2015 at 04:33:49PM +0530, Viresh Kumar wrote: On mediatek platform, they need to configure two regulators in order to change DVFS state of the big cluster. The generic cpufreq-dt driver and earlier OPP bindings have support for a single regulator only and so what Pi-cheng tried to do is, - Configure one of the regulators using cpufreq-dt - And other one using cpufreq frequency change notifiers This looks awkward.. What I suggested was to create another virtual regulator for CPU which will eventually configure both the regulators. And so the question that such virtual regulators are allowed or not. Ugh, no - that's a hideous bodge which is only going to create trouble later. Remember, DT is an ABI and should describe the hardware so if we're doing bodges that are visible there to shoehorn things onto our implementation that's bad. The concerns that Pi-Cheng had about what happens if the PMIC gets changed definitely seem relevant here too. signature.asc Description: Digital signature
Re: [PATCH v2 3/4] cpufreq: mediatek: add Mediatek cpufreq driver
Am Mittwoch, den 11.03.2015, 16:33 +0530 schrieb Viresh Kumar: On 11 March 2015 at 16:23, Mark Brown broo...@kernel.org wrote: On Tue, Mar 10, 2015 at 08:20:43AM +0530, Viresh Kumar wrote: Please don't send upstream e-mail to my work account, I use this address pretty consistently for upstream. Upstream mail to my work account frequently ends up unread. Sorry about that, I did exactly opposite of this earlier :( On 6 March 2015 at 11:19, Pi-Cheng Chen pi-cheng.c...@linaro.org wrote: On 5 March 2015 at 17:55, Viresh Kumar viresh.ku...@linaro.org wrote: About putting those stuff into regulator driver, I think you mean creating a virtual regulator device and put all the voltage controlling complex into the driver, right? Maybe it's a good idea in this case, but I am sure if this kind of virtual regulator is acceptable. @Mark: Is this allowed to create virtual regulator for a CPU ? I don't really know what the above means or what problem it's supposed to solve. On mediatek platform, they need to configure two regulators in order to change DVFS state of the big cluster. The generic cpufreq-dt driver and earlier OPP bindings have support for a single regulator only and so what Pi-cheng tried to do is, - Configure one of the regulators using cpufreq-dt - And other one using cpufreq frequency change notifiers This looks awkward.. What I suggested was to create another virtual regulator for CPU which will eventually configure both the regulators. And so the question that such virtual regulators are allowed or not. Instead of creating virtual regulators I would be strongly in favor of reviving the voltage-domain work. That would allow us to push all those voltage dependencies we have seen on various SoCs into the domain handling code and don't care about it in the drivers. In that case cpufreq-dt wouldn't control a regulator directly, but request a specific voltage from the domain the CPUs are located in and those in turn would control the regulators supplying them. Regards, Lucas -- Pengutronix e.K. | Lucas Stach | Industrial Linux Solutions | http://www.pengutronix.de/ | -- To unsubscribe from this list: send the line unsubscribe linux-kernel in the body of a message to majord...@vger.kernel.org More majordomo info at http://vger.kernel.org/majordomo-info.html Please read the FAQ at http://www.tux.org/lkml/
Re: [PATCH v2 3/4] cpufreq: mediatek: add Mediatek cpufreq driver
On 6 March 2015 at 11:19, Pi-Cheng Chen wrote: > On 5 March 2015 at 17:55, Viresh Kumar wrote: > About putting > those stuff into regulator driver, I think you mean creating a > "virtual regulator > device" and put all the voltage controlling complex into the driver, right? > Maybe it's a good idea in this case, but I am sure if this kind of > virtual regulator is acceptable. @Mark: Is this allowed to create virtual regulator for a CPU ? > And the flexibility might be an issue, since we might > use different > PMIC for same SoC on different board. We can talk about that separately once Mark replies to my query. > Combining comments and suggestions from you and Sascha[1], I conclude some > architectural changes are going to be made in the next version: > > 1. Use set_rate hook instead of determine_rate in clk driver, and > switch to intermeidate > PLL parent and back to original CPU PLL parent explicitly in set_rate Lets wait for Russell's answer to the query I posted before making any progress here. -- To unsubscribe from this list: send the line "unsubscribe linux-kernel" in the body of a message to majord...@vger.kernel.org More majordomo info at http://vger.kernel.org/majordomo-info.html Please read the FAQ at http://www.tux.org/lkml/
Re: [PATCH v2 3/4] cpufreq: mediatek: add Mediatek cpufreq driver
On 10 March 2015 at 00:28, Russell King - ARM Linux
wrote:
> On Wed, Mar 04, 2015 at 04:49:15PM +0800, pi-cheng.chen wrote:
>> +static int cpu_opp_table_get_freq_index(unsigned int freq)
>> +{
>> + struct cpu_opp_table *opp_tbl = dvfs_info->opp_tbl;
>> + int i;
>> +
>> + for (i = 0; opp_tbl[i].freq != 0; i++) {
>> + if (opp_tbl[i].freq >= freq)
>> + return i;
>> + }
>> +
>> + return -1;
>
> My "return -1" detector fired on this...
>
>> +static int cpu_opp_table_get_volt_index(unsigned int volt)
>> +{
>> + struct cpu_opp_table *opp_tbl = dvfs_info->opp_tbl;
>> + int i;
>> +
>> + for (i = 0; opp_tbl[i].vproc != -1; i++)
>> + if (opp_tbl[i].vproc >= volt)
>> + return i;
>> +
>> + return -1;
>
> And this.
>
>> +static int mtk_cpufreq_notify(struct notifier_block *nb,
>> + unsigned long action, void *data)
>> +{
>> + struct cpufreq_freqs *freqs = data;
>> + struct cpu_opp_table *opp_tbl = dvfs_info->opp_tbl;
>> + int old_vproc, new_vproc, old_index, new_index;
>> +
>> + if (!cpumask_test_cpu(freqs->cpu, _info->cpus))
>> + return NOTIFY_DONE;
>> +
>> + old_vproc = regulator_get_voltage(dvfs_info->proc_reg);
>> + old_index = cpu_opp_table_get_volt_index(old_vproc);
>> + new_index = cpu_opp_table_get_freq_index(freqs->new * 1000);
>> + new_vproc = opp_tbl[new_index].vproc;
>
> Let's say that cpu_opp_table_get_freq_index() returns -1. We then
> do no error checking on this, and access the memory immediately
> preceding opp_tbl[0].
>
> Since we'll be loading garbage from opp_tbl[-1], this probably means
> that mtk_cpufreq_voltage_trace() will go wrong. Your method of using
> the vproc values to work out which direction we should walk between
> old_index...new_index means that we could end up walking through
> almost the whole UINT_MAX range to wrap around to the new index.
>
> Yet again, "return -1" proves to be a sure sign of a bug.
Hi Russell,
Thanks for your reviewing. I'll fix it in next version.
Best Regards,
Pi-Cheng
>
> --
> FTTC broadband for 0.8mile line: currently at 10.5Mbps down 400kbps up
> according to speedtest.net.
--
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majord...@vger.kernel.org
More majordomo info at http://vger.kernel.org/majordomo-info.html
Please read the FAQ at http://www.tux.org/lkml/
Re: [PATCH v2 3/4] cpufreq: mediatek: add Mediatek cpufreq driver
On Wed, Mar 04, 2015 at 04:49:15PM +0800, pi-cheng.chen wrote:
> +static int cpu_opp_table_get_freq_index(unsigned int freq)
> +{
> + struct cpu_opp_table *opp_tbl = dvfs_info->opp_tbl;
> + int i;
> +
> + for (i = 0; opp_tbl[i].freq != 0; i++) {
> + if (opp_tbl[i].freq >= freq)
> + return i;
> + }
> +
> + return -1;
My "return -1" detector fired on this...
> +static int cpu_opp_table_get_volt_index(unsigned int volt)
> +{
> + struct cpu_opp_table *opp_tbl = dvfs_info->opp_tbl;
> + int i;
> +
> + for (i = 0; opp_tbl[i].vproc != -1; i++)
> + if (opp_tbl[i].vproc >= volt)
> + return i;
> +
> + return -1;
And this.
> +static int mtk_cpufreq_notify(struct notifier_block *nb,
> + unsigned long action, void *data)
> +{
> + struct cpufreq_freqs *freqs = data;
> + struct cpu_opp_table *opp_tbl = dvfs_info->opp_tbl;
> + int old_vproc, new_vproc, old_index, new_index;
> +
> + if (!cpumask_test_cpu(freqs->cpu, _info->cpus))
> + return NOTIFY_DONE;
> +
> + old_vproc = regulator_get_voltage(dvfs_info->proc_reg);
> + old_index = cpu_opp_table_get_volt_index(old_vproc);
> + new_index = cpu_opp_table_get_freq_index(freqs->new * 1000);
> + new_vproc = opp_tbl[new_index].vproc;
Let's say that cpu_opp_table_get_freq_index() returns -1. We then
do no error checking on this, and access the memory immediately
preceding opp_tbl[0].
Since we'll be loading garbage from opp_tbl[-1], this probably means
that mtk_cpufreq_voltage_trace() will go wrong. Your method of using
the vproc values to work out which direction we should walk between
old_index...new_index means that we could end up walking through
almost the whole UINT_MAX range to wrap around to the new index.
Yet again, "return -1" proves to be a sure sign of a bug.
--
FTTC broadband for 0.8mile line: currently at 10.5Mbps down 400kbps up
according to speedtest.net.
--
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majord...@vger.kernel.org
More majordomo info at http://vger.kernel.org/majordomo-info.html
Please read the FAQ at http://www.tux.org/lkml/
Re: [PATCH v2 3/4] cpufreq: mediatek: add Mediatek cpufreq driver
On Wed, Mar 04, 2015 at 04:49:15PM +0800, pi-cheng.chen wrote:
+static int cpu_opp_table_get_freq_index(unsigned int freq)
+{
+ struct cpu_opp_table *opp_tbl = dvfs_info-opp_tbl;
+ int i;
+
+ for (i = 0; opp_tbl[i].freq != 0; i++) {
+ if (opp_tbl[i].freq = freq)
+ return i;
+ }
+
+ return -1;
My return -1 detector fired on this...
+static int cpu_opp_table_get_volt_index(unsigned int volt)
+{
+ struct cpu_opp_table *opp_tbl = dvfs_info-opp_tbl;
+ int i;
+
+ for (i = 0; opp_tbl[i].vproc != -1; i++)
+ if (opp_tbl[i].vproc = volt)
+ return i;
+
+ return -1;
And this.
+static int mtk_cpufreq_notify(struct notifier_block *nb,
+ unsigned long action, void *data)
+{
+ struct cpufreq_freqs *freqs = data;
+ struct cpu_opp_table *opp_tbl = dvfs_info-opp_tbl;
+ int old_vproc, new_vproc, old_index, new_index;
+
+ if (!cpumask_test_cpu(freqs-cpu, dvfs_info-cpus))
+ return NOTIFY_DONE;
+
+ old_vproc = regulator_get_voltage(dvfs_info-proc_reg);
+ old_index = cpu_opp_table_get_volt_index(old_vproc);
+ new_index = cpu_opp_table_get_freq_index(freqs-new * 1000);
+ new_vproc = opp_tbl[new_index].vproc;
Let's say that cpu_opp_table_get_freq_index() returns -1. We then
do no error checking on this, and access the memory immediately
preceding opp_tbl[0].
Since we'll be loading garbage from opp_tbl[-1], this probably means
that mtk_cpufreq_voltage_trace() will go wrong. Your method of using
the vproc values to work out which direction we should walk between
old_index...new_index means that we could end up walking through
almost the whole UINT_MAX range to wrap around to the new index.
Yet again, return -1 proves to be a sure sign of a bug.
--
FTTC broadband for 0.8mile line: currently at 10.5Mbps down 400kbps up
according to speedtest.net.
--
To unsubscribe from this list: send the line unsubscribe linux-kernel in
the body of a message to majord...@vger.kernel.org
More majordomo info at http://vger.kernel.org/majordomo-info.html
Please read the FAQ at http://www.tux.org/lkml/
Re: [PATCH v2 3/4] cpufreq: mediatek: add Mediatek cpufreq driver
On 10 March 2015 at 00:28, Russell King - ARM Linux
li...@arm.linux.org.uk wrote:
On Wed, Mar 04, 2015 at 04:49:15PM +0800, pi-cheng.chen wrote:
+static int cpu_opp_table_get_freq_index(unsigned int freq)
+{
+ struct cpu_opp_table *opp_tbl = dvfs_info-opp_tbl;
+ int i;
+
+ for (i = 0; opp_tbl[i].freq != 0; i++) {
+ if (opp_tbl[i].freq = freq)
+ return i;
+ }
+
+ return -1;
My return -1 detector fired on this...
+static int cpu_opp_table_get_volt_index(unsigned int volt)
+{
+ struct cpu_opp_table *opp_tbl = dvfs_info-opp_tbl;
+ int i;
+
+ for (i = 0; opp_tbl[i].vproc != -1; i++)
+ if (opp_tbl[i].vproc = volt)
+ return i;
+
+ return -1;
And this.
+static int mtk_cpufreq_notify(struct notifier_block *nb,
+ unsigned long action, void *data)
+{
+ struct cpufreq_freqs *freqs = data;
+ struct cpu_opp_table *opp_tbl = dvfs_info-opp_tbl;
+ int old_vproc, new_vproc, old_index, new_index;
+
+ if (!cpumask_test_cpu(freqs-cpu, dvfs_info-cpus))
+ return NOTIFY_DONE;
+
+ old_vproc = regulator_get_voltage(dvfs_info-proc_reg);
+ old_index = cpu_opp_table_get_volt_index(old_vproc);
+ new_index = cpu_opp_table_get_freq_index(freqs-new * 1000);
+ new_vproc = opp_tbl[new_index].vproc;
Let's say that cpu_opp_table_get_freq_index() returns -1. We then
do no error checking on this, and access the memory immediately
preceding opp_tbl[0].
Since we'll be loading garbage from opp_tbl[-1], this probably means
that mtk_cpufreq_voltage_trace() will go wrong. Your method of using
the vproc values to work out which direction we should walk between
old_index...new_index means that we could end up walking through
almost the whole UINT_MAX range to wrap around to the new index.
Yet again, return -1 proves to be a sure sign of a bug.
Hi Russell,
Thanks for your reviewing. I'll fix it in next version.
Best Regards,
Pi-Cheng
--
FTTC broadband for 0.8mile line: currently at 10.5Mbps down 400kbps up
according to speedtest.net.
--
To unsubscribe from this list: send the line unsubscribe linux-kernel in
the body of a message to majord...@vger.kernel.org
More majordomo info at http://vger.kernel.org/majordomo-info.html
Please read the FAQ at http://www.tux.org/lkml/
Re: [PATCH v2 3/4] cpufreq: mediatek: add Mediatek cpufreq driver
On 6 March 2015 at 11:19, Pi-Cheng Chen pi-cheng.c...@linaro.org wrote: On 5 March 2015 at 17:55, Viresh Kumar viresh.ku...@linaro.org wrote: About putting those stuff into regulator driver, I think you mean creating a virtual regulator device and put all the voltage controlling complex into the driver, right? Maybe it's a good idea in this case, but I am sure if this kind of virtual regulator is acceptable. @Mark: Is this allowed to create virtual regulator for a CPU ? And the flexibility might be an issue, since we might use different PMIC for same SoC on different board. We can talk about that separately once Mark replies to my query. Combining comments and suggestions from you and Sascha[1], I conclude some architectural changes are going to be made in the next version: 1. Use set_rate hook instead of determine_rate in clk driver, and switch to intermeidate PLL parent and back to original CPU PLL parent explicitly in set_rate Lets wait for Russell's answer to the query I posted before making any progress here. -- To unsubscribe from this list: send the line unsubscribe linux-kernel in the body of a message to majord...@vger.kernel.org More majordomo info at http://vger.kernel.org/majordomo-info.html Please read the FAQ at http://www.tux.org/lkml/
Re: [PATCH v2 3/4] cpufreq: mediatek: add Mediatek cpufreq driver
+cc Sascha On 5 March 2015 at 17:55, Viresh Kumar wrote: > On 5 March 2015 at 12:57, Pi-Cheng Chen wrote: > >> On 4 March 2015 at 19:09, Viresh Kumar wrote: >> There are 2 clusters, but only the big cluster need to do voltage scaling in >> the >> notifier, since the voltage controlling is done by cpufreq-dt driver >> in this version. >> Therefore only one dvfs_info struct here. > > Do you really think its readable enough that way? You must have added some > comments on how this is working. Also, what about putting this stuff in your > regulator driver, so that you don't really have to do this in PRE/POST > notifiers. Okay. I will add comments to describe some details about this. About putting those stuff into regulator driver, I think you mean creating a "virtual regulator device" and put all the voltage controlling complex into the driver, right? Maybe it's a good idea in this case, but I am sure if this kind of virtual regulator is acceptable. And the flexibility might be an issue, since we might use different PMIC for same SoC on different board. > + inter_clk = clk_get(>dev, NULL); >>> >>> How is this supposed to work ? How will pdev->dev give intermediate >>> clock ? >> >> It works with the the device tree binding in the 2nd patch of this series, >> too. >> Since the cpufreq node is not allowed, would you have some suggestions on >> how to get the intermediate clock source in this case? > > How exactly? I am not doubting your work, just that I don't know how that DT > binding will reflect here with clock_get for pdev->dev.. Please correct me if I was wrong. IIUC, It does: clk_get() -> __of_clk_get_by_name() -> __of_clk_get() The "mtk-cpufreq" device tree node specified the intermediate clock source in "clocks" property. And the pdev here came from the "mtk-cpufreq" device tree node, so we can get the "clock specifier" by calling of_parse_phandle_with_args() to find "clocks" property in __of_clk_get(). > + pd->independent_clocks = 1, >>> >>> s/,/; ?? >> >> It's strange that I didn't get a compiling error here. >> Will fix it. > > Its a perfectly valid statement :) and so no errors. Both will execute as they > will in case of ';', just that output of the later one will be > returned. But there > in no variable on LHS (left-hand-side) and so the value doesn't matter. Thanks for your explanation. :) > >>> Don't want to free OPP table here on error ? >> >> Please correct me if I was wrong. Since the OPP table in the dvfs_info is >> allocated by devm_kzalloc(), it is supposed to be freed if the probe function >> failed, isn't it? >> >> And the OPP table initialized by of_init_opp_table() in cpu_opp_table_init() >> was freed right before the function return since it will be initialized >> again in >> the cpufreq-dt driver. > > Okay, I was talking about this only and I missed it. We probably need to fix > this in OPP library so that multiple callers are allowed. > + dev = platform_device_register_data(NULL, "cpufreq-dt", -1, pd, + sizeof(*pd)); >>> >>> So this routine is going to be called only once. Then how are you >>> initializing stuff >>> for both the clusters in the upper for loop ? It looked very very confusing. >> >> Please let me clarify this here. >> We have two clusters, one for big and another for little cores. For >> the little cores' >> cluster, only one voltage source needs to be controlled when doing CPU DVFS. >> Therefore the voltage scaling of little cores' cluster is done in the >> cpufreq-dt. >> But for the big cores' cluster, there are two voltage sources here to >> be controlled >> and these two voltage source need to be scaled up and down in a SoC specific >> manner which is implemented in the mtk_cpufreq_voltage_trace() function. >> Hence, we put the voltage scaling of big cores' cluster in the cpufreq >> notifier and >> that's also why we need a mtk-cpufreq driver in addition to cpufreq-dt. >> >> In the confusing loop above, I am trying to solve two problems: >> 1. to find out which CPUs shares the same clock / power domains among all >> CPUs >> 2. to initialize the dvfs_info which is only needed by big cores' cluster >> >> I think that's why the loop looks so confusing. Maybe doing it in two >> separate loops >> will make the code more readable? I'll try it in next version. > > Yes. Combining comments and suggestions from you and Sascha[1], I conclude some architectural changes are going to be made in the next version: 1. Use set_rate hook instead of determine_rate in clk driver, and switch to intermeidate PLL parent and back to original CPU PLL parent explicitly in set_rate 2. Therefore we don't need intermediate frequency support in cpufreq-dt to implement cpufreq support for Mediatek SoC 3. Use clk notifier to handle voltage controlling corresponding to intermediate clock rate 4. Due to 3. we need to move all voltage controlling part back into the notifier in mtk-cpufreq
Re: [PATCH v2 3/4] cpufreq: mediatek: add Mediatek cpufreq driver
On 5 March 2015 at 12:57, Pi-Cheng Chen wrote: > On 4 March 2015 at 19:09, Viresh Kumar wrote: > There are 2 clusters, but only the big cluster need to do voltage scaling in > the > notifier, since the voltage controlling is done by cpufreq-dt driver > in this version. > Therefore only one dvfs_info struct here. Do you really think its readable enough that way? You must have added some comments on how this is working. Also, what about putting this stuff in your regulator driver, so that you don't really have to do this in PRE/POST notifiers. >>> + inter_clk = clk_get(>dev, NULL); >> >> How is this supposed to work ? How will pdev->dev give intermediate >> clock ? > > It works with the the device tree binding in the 2nd patch of this series, > too. > Since the cpufreq node is not allowed, would you have some suggestions on > how to get the intermediate clock source in this case? How exactly? I am not doubting your work, just that I don't know how that DT binding will reflect here with clock_get for pdev->dev.. >>> + pd->independent_clocks = 1, >> >> s/,/; ?? > > It's strange that I didn't get a compiling error here. > Will fix it. Its a perfectly valid statement :) and so no errors. Both will execute as they will in case of ';', just that output of the later one will be returned. But there in no variable on LHS (left-hand-side) and so the value doesn't matter. >> Don't want to free OPP table here on error ? > > Please correct me if I was wrong. Since the OPP table in the dvfs_info is > allocated by devm_kzalloc(), it is supposed to be freed if the probe function > failed, isn't it? > > And the OPP table initialized by of_init_opp_table() in cpu_opp_table_init() > was freed right before the function return since it will be initialized again > in > the cpufreq-dt driver. Okay, I was talking about this only and I missed it. We probably need to fix this in OPP library so that multiple callers are allowed. >>> + dev = platform_device_register_data(NULL, "cpufreq-dt", -1, pd, >>> + sizeof(*pd)); >> >> So this routine is going to be called only once. Then how are you >> initializing stuff >> for both the clusters in the upper for loop ? It looked very very confusing. > > Please let me clarify this here. > We have two clusters, one for big and another for little cores. For > the little cores' > cluster, only one voltage source needs to be controlled when doing CPU DVFS. > Therefore the voltage scaling of little cores' cluster is done in the > cpufreq-dt. > But for the big cores' cluster, there are two voltage sources here to > be controlled > and these two voltage source need to be scaled up and down in a SoC specific > manner which is implemented in the mtk_cpufreq_voltage_trace() function. > Hence, we put the voltage scaling of big cores' cluster in the cpufreq > notifier and > that's also why we need a mtk-cpufreq driver in addition to cpufreq-dt. > > In the confusing loop above, I am trying to solve two problems: > 1. to find out which CPUs shares the same clock / power domains among all CPUs > 2. to initialize the dvfs_info which is only needed by big cores' cluster > > I think that's why the loop looks so confusing. Maybe doing it in two > separate loops > will make the code more readable? I'll try it in next version. Yes. -- To unsubscribe from this list: send the line "unsubscribe linux-kernel" in the body of a message to majord...@vger.kernel.org More majordomo info at http://vger.kernel.org/majordomo-info.html Please read the FAQ at http://www.tux.org/lkml/
Re: [PATCH v2 3/4] cpufreq: mediatek: add Mediatek cpufreq driver
On 5 March 2015 at 12:57, Pi-Cheng Chen pi-cheng.c...@linaro.org wrote: On 4 March 2015 at 19:09, Viresh Kumar viresh.ku...@linaro.org wrote: There are 2 clusters, but only the big cluster need to do voltage scaling in the notifier, since the voltage controlling is done by cpufreq-dt driver in this version. Therefore only one dvfs_info struct here. Do you really think its readable enough that way? You must have added some comments on how this is working. Also, what about putting this stuff in your regulator driver, so that you don't really have to do this in PRE/POST notifiers. + inter_clk = clk_get(pdev-dev, NULL); How is this supposed to work ? How will pdev-dev give intermediate clock ? It works with the the device tree binding in the 2nd patch of this series, too. Since the cpufreq node is not allowed, would you have some suggestions on how to get the intermediate clock source in this case? How exactly? I am not doubting your work, just that I don't know how that DT binding will reflect here with clock_get for pdev-dev.. + pd-independent_clocks = 1, s/,/; ?? It's strange that I didn't get a compiling error here. Will fix it. Its a perfectly valid statement :) and so no errors. Both will execute as they will in case of ';', just that output of the later one will be returned. But there in no variable on LHS (left-hand-side) and so the value doesn't matter. Don't want to free OPP table here on error ? Please correct me if I was wrong. Since the OPP table in the dvfs_info is allocated by devm_kzalloc(), it is supposed to be freed if the probe function failed, isn't it? And the OPP table initialized by of_init_opp_table() in cpu_opp_table_init() was freed right before the function return since it will be initialized again in the cpufreq-dt driver. Okay, I was talking about this only and I missed it. We probably need to fix this in OPP library so that multiple callers are allowed. + dev = platform_device_register_data(NULL, cpufreq-dt, -1, pd, + sizeof(*pd)); So this routine is going to be called only once. Then how are you initializing stuff for both the clusters in the upper for loop ? It looked very very confusing. Please let me clarify this here. We have two clusters, one for big and another for little cores. For the little cores' cluster, only one voltage source needs to be controlled when doing CPU DVFS. Therefore the voltage scaling of little cores' cluster is done in the cpufreq-dt. But for the big cores' cluster, there are two voltage sources here to be controlled and these two voltage source need to be scaled up and down in a SoC specific manner which is implemented in the mtk_cpufreq_voltage_trace() function. Hence, we put the voltage scaling of big cores' cluster in the cpufreq notifier and that's also why we need a mtk-cpufreq driver in addition to cpufreq-dt. In the confusing loop above, I am trying to solve two problems: 1. to find out which CPUs shares the same clock / power domains among all CPUs 2. to initialize the dvfs_info which is only needed by big cores' cluster I think that's why the loop looks so confusing. Maybe doing it in two separate loops will make the code more readable? I'll try it in next version. Yes. -- To unsubscribe from this list: send the line unsubscribe linux-kernel in the body of a message to majord...@vger.kernel.org More majordomo info at http://vger.kernel.org/majordomo-info.html Please read the FAQ at http://www.tux.org/lkml/
Re: [PATCH v2 3/4] cpufreq: mediatek: add Mediatek cpufreq driver
+cc Sascha On 5 March 2015 at 17:55, Viresh Kumar viresh.ku...@linaro.org wrote: On 5 March 2015 at 12:57, Pi-Cheng Chen pi-cheng.c...@linaro.org wrote: On 4 March 2015 at 19:09, Viresh Kumar viresh.ku...@linaro.org wrote: There are 2 clusters, but only the big cluster need to do voltage scaling in the notifier, since the voltage controlling is done by cpufreq-dt driver in this version. Therefore only one dvfs_info struct here. Do you really think its readable enough that way? You must have added some comments on how this is working. Also, what about putting this stuff in your regulator driver, so that you don't really have to do this in PRE/POST notifiers. Okay. I will add comments to describe some details about this. About putting those stuff into regulator driver, I think you mean creating a virtual regulator device and put all the voltage controlling complex into the driver, right? Maybe it's a good idea in this case, but I am sure if this kind of virtual regulator is acceptable. And the flexibility might be an issue, since we might use different PMIC for same SoC on different board. + inter_clk = clk_get(pdev-dev, NULL); How is this supposed to work ? How will pdev-dev give intermediate clock ? It works with the the device tree binding in the 2nd patch of this series, too. Since the cpufreq node is not allowed, would you have some suggestions on how to get the intermediate clock source in this case? How exactly? I am not doubting your work, just that I don't know how that DT binding will reflect here with clock_get for pdev-dev.. Please correct me if I was wrong. IIUC, It does: clk_get() - __of_clk_get_by_name() - __of_clk_get() The mtk-cpufreq device tree node specified the intermediate clock source in clocks property. And the pdev here came from the mtk-cpufreq device tree node, so we can get the clock specifier by calling of_parse_phandle_with_args() to find clocks property in __of_clk_get(). + pd-independent_clocks = 1, s/,/; ?? It's strange that I didn't get a compiling error here. Will fix it. Its a perfectly valid statement :) and so no errors. Both will execute as they will in case of ';', just that output of the later one will be returned. But there in no variable on LHS (left-hand-side) and so the value doesn't matter. Thanks for your explanation. :) Don't want to free OPP table here on error ? Please correct me if I was wrong. Since the OPP table in the dvfs_info is allocated by devm_kzalloc(), it is supposed to be freed if the probe function failed, isn't it? And the OPP table initialized by of_init_opp_table() in cpu_opp_table_init() was freed right before the function return since it will be initialized again in the cpufreq-dt driver. Okay, I was talking about this only and I missed it. We probably need to fix this in OPP library so that multiple callers are allowed. + dev = platform_device_register_data(NULL, cpufreq-dt, -1, pd, + sizeof(*pd)); So this routine is going to be called only once. Then how are you initializing stuff for both the clusters in the upper for loop ? It looked very very confusing. Please let me clarify this here. We have two clusters, one for big and another for little cores. For the little cores' cluster, only one voltage source needs to be controlled when doing CPU DVFS. Therefore the voltage scaling of little cores' cluster is done in the cpufreq-dt. But for the big cores' cluster, there are two voltage sources here to be controlled and these two voltage source need to be scaled up and down in a SoC specific manner which is implemented in the mtk_cpufreq_voltage_trace() function. Hence, we put the voltage scaling of big cores' cluster in the cpufreq notifier and that's also why we need a mtk-cpufreq driver in addition to cpufreq-dt. In the confusing loop above, I am trying to solve two problems: 1. to find out which CPUs shares the same clock / power domains among all CPUs 2. to initialize the dvfs_info which is only needed by big cores' cluster I think that's why the loop looks so confusing. Maybe doing it in two separate loops will make the code more readable? I'll try it in next version. Yes. Combining comments and suggestions from you and Sascha[1], I conclude some architectural changes are going to be made in the next version: 1. Use set_rate hook instead of determine_rate in clk driver, and switch to intermeidate PLL parent and back to original CPU PLL parent explicitly in set_rate 2. Therefore we don't need intermediate frequency support in cpufreq-dt to implement cpufreq support for Mediatek SoC 3. Use clk notifier to handle voltage controlling corresponding to intermediate clock rate 4. Due to 3. we need to move all voltage controlling part back into the notifier in mtk-cpufreq (Voltage controlling for little cores' cluster is handled in cpufreq-dt in this version.) And I have
Re: [PATCH v2 3/4] cpufreq: mediatek: add Mediatek cpufreq driver
Hi Viresh,
Thanks for reviewing.
Please see my reply below:
On 4 March 2015 at 19:09, Viresh Kumar wrote:
> Haven't reviewed it completely yet, but this is all I have done.
>
> On 4 March 2015 at 14:19, pi-cheng.chen wrote:
>
>> +static int mtk_cpufreq_notify(struct notifier_block *nb,
>> + unsigned long action, void *data)
>> +{
>> + struct cpufreq_freqs *freqs = data;
>> + struct cpu_opp_table *opp_tbl = dvfs_info->opp_tbl;
>
> There is only one dvfs info ? but there are two clusters, sorry got confused
> a bit..
There are 2 clusters, but only the big cluster need to do voltage scaling in the
notifier, since the voltage controlling is done by cpufreq-dt driver
in this version.
Therefore only one dvfs_info struct here.
>
>> + int old_vproc, new_vproc, old_index, new_index;
>> +
>> + if (!cpumask_test_cpu(freqs->cpu, _info->cpus))
>> + return NOTIFY_DONE;
>> +
>> + old_vproc = regulator_get_voltage(dvfs_info->proc_reg);
>> + old_index = cpu_opp_table_get_volt_index(old_vproc);
>> + new_index = cpu_opp_table_get_freq_index(freqs->new * 1000);
>> + new_vproc = opp_tbl[new_index].vproc;
>> +
>> + if (old_vproc == new_vproc)
>> + return 0;
>> +
>> + if ((action == CPUFREQ_PRECHANGE && old_vproc < new_vproc) ||
>> + (action == CPUFREQ_POSTCHANGE && old_vproc > new_vproc))
>> + mtk_cpufreq_voltage_trace(old_index, new_index);
>> +
>> + return NOTIFY_OK;
>> +}
>> +
>> +static struct notifier_block mtk_cpufreq_nb = {
>> + .notifier_call = mtk_cpufreq_notify,
>> +};
>> +
>> +static int cpu_opp_table_init(struct device *dev)
>> +{
>> + struct device *cpu_dev = dvfs_info->cpu_dev;
>> + struct cpu_opp_table *opp_tbl;
>> + struct dev_pm_opp *opp;
>> + int ret, cnt, i;
>> + unsigned long rate, vproc, vsram;
>> +
>> + ret = of_init_opp_table(cpu_dev);
>> + if (ret) {
>> + dev_err(dev, "Failed to init mtk_opp_table: %d\n", ret);
>> + return ret;
>> + }
>> +
>> + rcu_read_lock();
>> +
>> + cnt = dev_pm_opp_get_opp_count(cpu_dev);
>> + if (cnt < 0) {
>> + dev_err(cpu_dev, "No OPP table is found: %d", cnt);
>> + ret = cnt;
>> + goto out_free_opp_tbl;
>> + }
>> +
>> + opp_tbl = devm_kcalloc(dev, (cnt + 1), sizeof(struct cpu_opp_table),
>> + GFP_ATOMIC);
>> + if (!opp_tbl) {
>> + ret = -ENOMEM;
>> + goto out_free_opp_tbl;
>> + }
>> +
>> + for (i = 0, rate = 0; i < cnt; i++, rate++) {
>> + opp = dev_pm_opp_find_freq_ceil(cpu_dev, );
>> + if (IS_ERR(opp)) {
>> + ret = PTR_ERR(opp);
>> + goto out_free_opp_tbl;
>> + }
>> +
>> + vproc = dev_pm_opp_get_voltage(opp);
>> + vproc = get_regulator_voltage_ceil(dvfs_info->proc_reg,
>> vproc);
>> + vsram = vproc + VOLT_SHIFT_LOWER_LIMIT;
>> + vsram = get_regulator_voltage_ceil(dvfs_info->sram_reg,
>> vsram);
>> +
>> + if (vproc < 0 || vsram < 0) {
>> + ret = -EINVAL;
>> + goto out_free_opp_tbl;
>> + }
>> +
>> + opp_tbl[i].freq = rate;
>> + opp_tbl[i].vproc = vproc;
>> + opp_tbl[i].vsram = vsram;
>> + }
>> +
>> + opp_tbl[i].freq = 0;
>> + opp_tbl[i].vproc = -1;
>> + opp_tbl[i].vsram = -1;
>> + dvfs_info->opp_tbl = opp_tbl;
>> +
>> +out_free_opp_tbl:
>> + rcu_read_unlock();
>> + of_free_opp_table(cpu_dev);
>> +
>> + return ret;
>> +}
>> +
>> +static struct cpufreq_cpu_domain *get_cpu_domain(struct list_head
>> *domain_list,
>> +int cpu)
>> +{
>> + struct list_head *node;
>> +
>> + list_for_each(node, domain_list) {
>> + struct cpufreq_cpu_domain *domain;
>> +
>> + domain = container_of(node, struct cpufreq_cpu_domain, node);
>> + if (cpumask_test_cpu(cpu, >cpus))
>> + return domain;
>> + }
>> +
>> + return NULL;
>> +}
>> +
>> +static int mtk_cpufreq_probe(struct platform_device *pdev)
>
> On a dual cluster big LITTLE (your system), how many times is probe
> getting called ? Once or twice, i.e. for each cluster ??
The probe function will be called only once since it's triggered by the device
tree node in the 2nd patch of this series. Though it's not acceptable obviously.
>
>> +{
>> + struct clk *inter_clk;
>> + struct cpufreq_dt_platform_data *pd;
>> + struct platform_device *dev;
>> + unsigned long inter_freq;
>> + int cpu, ret;
>> +
>> + inter_clk = clk_get(>dev, NULL);
>
> How is this supposed to work ? How will pdev->dev give intermediate
>
Re: [PATCH v2 3/4] cpufreq: mediatek: add Mediatek cpufreq driver
Haven't reviewed it completely yet, but this is all I have done.
On 4 March 2015 at 14:19, pi-cheng.chen wrote:
> +static int mtk_cpufreq_notify(struct notifier_block *nb,
> + unsigned long action, void *data)
> +{
> + struct cpufreq_freqs *freqs = data;
> + struct cpu_opp_table *opp_tbl = dvfs_info->opp_tbl;
There is only one dvfs info ? but there are two clusters, sorry got confused
a bit..
> + int old_vproc, new_vproc, old_index, new_index;
> +
> + if (!cpumask_test_cpu(freqs->cpu, _info->cpus))
> + return NOTIFY_DONE;
> +
> + old_vproc = regulator_get_voltage(dvfs_info->proc_reg);
> + old_index = cpu_opp_table_get_volt_index(old_vproc);
> + new_index = cpu_opp_table_get_freq_index(freqs->new * 1000);
> + new_vproc = opp_tbl[new_index].vproc;
> +
> + if (old_vproc == new_vproc)
> + return 0;
> +
> + if ((action == CPUFREQ_PRECHANGE && old_vproc < new_vproc) ||
> + (action == CPUFREQ_POSTCHANGE && old_vproc > new_vproc))
> + mtk_cpufreq_voltage_trace(old_index, new_index);
> +
> + return NOTIFY_OK;
> +}
> +
> +static struct notifier_block mtk_cpufreq_nb = {
> + .notifier_call = mtk_cpufreq_notify,
> +};
> +
> +static int cpu_opp_table_init(struct device *dev)
> +{
> + struct device *cpu_dev = dvfs_info->cpu_dev;
> + struct cpu_opp_table *opp_tbl;
> + struct dev_pm_opp *opp;
> + int ret, cnt, i;
> + unsigned long rate, vproc, vsram;
> +
> + ret = of_init_opp_table(cpu_dev);
> + if (ret) {
> + dev_err(dev, "Failed to init mtk_opp_table: %d\n", ret);
> + return ret;
> + }
> +
> + rcu_read_lock();
> +
> + cnt = dev_pm_opp_get_opp_count(cpu_dev);
> + if (cnt < 0) {
> + dev_err(cpu_dev, "No OPP table is found: %d", cnt);
> + ret = cnt;
> + goto out_free_opp_tbl;
> + }
> +
> + opp_tbl = devm_kcalloc(dev, (cnt + 1), sizeof(struct cpu_opp_table),
> + GFP_ATOMIC);
> + if (!opp_tbl) {
> + ret = -ENOMEM;
> + goto out_free_opp_tbl;
> + }
> +
> + for (i = 0, rate = 0; i < cnt; i++, rate++) {
> + opp = dev_pm_opp_find_freq_ceil(cpu_dev, );
> + if (IS_ERR(opp)) {
> + ret = PTR_ERR(opp);
> + goto out_free_opp_tbl;
> + }
> +
> + vproc = dev_pm_opp_get_voltage(opp);
> + vproc = get_regulator_voltage_ceil(dvfs_info->proc_reg,
> vproc);
> + vsram = vproc + VOLT_SHIFT_LOWER_LIMIT;
> + vsram = get_regulator_voltage_ceil(dvfs_info->sram_reg,
> vsram);
> +
> + if (vproc < 0 || vsram < 0) {
> + ret = -EINVAL;
> + goto out_free_opp_tbl;
> + }
> +
> + opp_tbl[i].freq = rate;
> + opp_tbl[i].vproc = vproc;
> + opp_tbl[i].vsram = vsram;
> + }
> +
> + opp_tbl[i].freq = 0;
> + opp_tbl[i].vproc = -1;
> + opp_tbl[i].vsram = -1;
> + dvfs_info->opp_tbl = opp_tbl;
> +
> +out_free_opp_tbl:
> + rcu_read_unlock();
> + of_free_opp_table(cpu_dev);
> +
> + return ret;
> +}
> +
> +static struct cpufreq_cpu_domain *get_cpu_domain(struct list_head
> *domain_list,
> +int cpu)
> +{
> + struct list_head *node;
> +
> + list_for_each(node, domain_list) {
> + struct cpufreq_cpu_domain *domain;
> +
> + domain = container_of(node, struct cpufreq_cpu_domain, node);
> + if (cpumask_test_cpu(cpu, >cpus))
> + return domain;
> + }
> +
> + return NULL;
> +}
> +
> +static int mtk_cpufreq_probe(struct platform_device *pdev)
On a dual cluster big LITTLE (your system), how many times is probe
getting called ? Once or twice, i.e. for each cluster ??
> +{
> + struct clk *inter_clk;
> + struct cpufreq_dt_platform_data *pd;
> + struct platform_device *dev;
> + unsigned long inter_freq;
> + int cpu, ret;
> +
> + inter_clk = clk_get(>dev, NULL);
How is this supposed to work ? How will pdev->dev give intermediate
clock ?
> + if (IS_ERR(inter_clk)) {
> + if (PTR_ERR(inter_clk) == -EPROBE_DEFER) {
> + dev_warn(>dev, "clock not ready. defer
> probeing.\n");
> + return -EPROBE_DEFER;
> + }
> +
> + dev_err(>dev, "Failed to get intermediate clock\n");
> + return -ENODEV;
> + }
> + inter_freq = clk_get_rate(inter_clk);
> +
> + pd = devm_kzalloc(>dev, sizeof(*pd), GFP_KERNEL);
> + if (!pd)
> + return -ENOMEM;
> +
> + dvfs_info = devm_kzalloc(>dev, sizeof(*dvfs_info), GFP_KERNEL);
> +
[PATCH v2 3/4] cpufreq: mediatek: add Mediatek cpufreq driver
In this patch, some SoC specific voltage scaling flow is implemented in the
cpufreq notifier of mtk-cpufreq driver.
Signed-off-by: pi-cheng.chen
---
drivers/cpufreq/Kconfig.arm | 6 +
drivers/cpufreq/Makefile | 1 +
drivers/cpufreq/mtk-cpufreq.c | 346 ++
3 files changed, 353 insertions(+)
create mode 100644 drivers/cpufreq/mtk-cpufreq.c
diff --git a/drivers/cpufreq/Kconfig.arm b/drivers/cpufreq/Kconfig.arm
index 1b06fc4..f421653 100644
--- a/drivers/cpufreq/Kconfig.arm
+++ b/drivers/cpufreq/Kconfig.arm
@@ -263,3 +263,9 @@ config ARM_PXA2xx_CPUFREQ
This add the CPUFreq driver support for Intel PXA2xx SOCs.
If in doubt, say N.
+
+config ARM_MTK_CPUFREQ
+ bool "Mediatek CPUFreq support"
+ depends on ARCH_MEDIATEK && CPUFREQ_DT && REGULATOR
+ help
+ This adds the CPUFreq driver support for Mediatek SoCs.
diff --git a/drivers/cpufreq/Makefile b/drivers/cpufreq/Makefile
index 82a1821..05cb596 100644
--- a/drivers/cpufreq/Makefile
+++ b/drivers/cpufreq/Makefile
@@ -62,6 +62,7 @@ obj-$(CONFIG_ARM_HIGHBANK_CPUFREQ)+= highbank-cpufreq.o
obj-$(CONFIG_ARM_IMX6Q_CPUFREQ)+= imx6q-cpufreq.o
obj-$(CONFIG_ARM_INTEGRATOR) += integrator-cpufreq.o
obj-$(CONFIG_ARM_KIRKWOOD_CPUFREQ) += kirkwood-cpufreq.o
+obj-$(CONFIG_ARM_MTK_CPUFREQ) += mtk-cpufreq.o
obj-$(CONFIG_ARM_OMAP2PLUS_CPUFREQ)+= omap-cpufreq.o
obj-$(CONFIG_ARM_PXA2xx_CPUFREQ) += pxa2xx-cpufreq.o
obj-$(CONFIG_PXA3xx) += pxa3xx-cpufreq.o
diff --git a/drivers/cpufreq/mtk-cpufreq.c b/drivers/cpufreq/mtk-cpufreq.c
new file mode 100644
index 000..344d588
--- /dev/null
+++ b/drivers/cpufreq/mtk-cpufreq.c
@@ -0,0 +1,346 @@
+/*
+* Copyright (c) 2015 Linaro Ltd.
+* Author: Pi-Cheng Chen
+*
+* This program is free software; you can redistribute it and/or modify
+* it under the terms of the GNU General Public License version 2 as
+* published by the Free Software Foundation.
+*
+* This program is distributed in the hope that it will be useful,
+* but WITHOUT ANY WARRANTY; without even the implied warranty of
+* MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
+* GNU General Public License for more details.
+*/
+
+#include
+#include
+#include
+#include
+#include
+#include
+#include
+#include
+#include
+#include
+#include
+
+#define VOLT_SHIFT_LOWER_LIMIT 10
+#define VOLT_SHIFT_UPPER_LIMIT 20
+
+struct cpu_opp_table {
+ unsigned int freq;
+ int vproc;
+ int vsram;
+};
+
+static struct dvfs_info {
+ struct cpumask cpus;
+ struct cpu_opp_table *opp_tbl;
+ struct device *cpu_dev;
+ struct regulator *proc_reg;
+ struct regulator *sram_reg;
+} *dvfs_info;
+
+static int cpu_opp_table_get_freq_index(unsigned int freq)
+{
+ struct cpu_opp_table *opp_tbl = dvfs_info->opp_tbl;
+ int i;
+
+ for (i = 0; opp_tbl[i].freq != 0; i++) {
+ if (opp_tbl[i].freq >= freq)
+ return i;
+ }
+
+ return -1;
+}
+
+static int cpu_opp_table_get_volt_index(unsigned int volt)
+{
+ struct cpu_opp_table *opp_tbl = dvfs_info->opp_tbl;
+ int i;
+
+ for (i = 0; opp_tbl[i].vproc != -1; i++)
+ if (opp_tbl[i].vproc >= volt)
+ return i;
+
+ return -1;
+}
+
+static int get_regulator_voltage_ceil(struct regulator *regulator, int voltage)
+{
+ int cnt, i, volt = -1;
+
+ cnt = regulator_count_voltages(regulator);
+
+ for (i = 0; i < cnt && volt < voltage; i++)
+ volt = regulator_list_voltage(regulator, i);
+
+ return volt;
+}
+
+static int mtk_cpufreq_voltage_trace(int old_index, int new_index)
+{
+ struct cpu_opp_table *opp_tbl = dvfs_info->opp_tbl;
+ int old_vproc, new_vproc, i, j;
+
+ old_vproc = regulator_get_voltage(dvfs_info->proc_reg);
+ new_vproc = opp_tbl[new_index].vproc;
+
+ if (old_vproc > new_vproc) {
+ for (i = old_index; i > new_index;) {
+ for (j = i; j >= new_index; j--)
+ if (opp_tbl[i].vsram - opp_tbl[j].vproc
+ > VOLT_SHIFT_UPPER_LIMIT)
+ break;
+ i = j + 1;
+
+ regulator_set_voltage_tol(dvfs_info->proc_reg,
+ opp_tbl[i].vproc, 0);
+ regulator_set_voltage_tol(dvfs_info->sram_reg,
+ opp_tbl[i].vsram, 0);
+ }
+ } else if (old_vproc < new_vproc) {
+ for (i = old_index; i < new_index;) {
+ for (j = i; j <= new_index; j++)
+ if (opp_tbl[j].vsram - opp_tbl[i].vproc
+ > VOLT_SHIFT_UPPER_LIMIT)
+
Re: [PATCH v2 3/4] cpufreq: mediatek: add Mediatek cpufreq driver
Hi Viresh,
Thanks for reviewing.
Please see my reply below:
On 4 March 2015 at 19:09, Viresh Kumar viresh.ku...@linaro.org wrote:
Haven't reviewed it completely yet, but this is all I have done.
On 4 March 2015 at 14:19, pi-cheng.chen pi-cheng.c...@linaro.org wrote:
+static int mtk_cpufreq_notify(struct notifier_block *nb,
+ unsigned long action, void *data)
+{
+ struct cpufreq_freqs *freqs = data;
+ struct cpu_opp_table *opp_tbl = dvfs_info-opp_tbl;
There is only one dvfs info ? but there are two clusters, sorry got confused
a bit..
There are 2 clusters, but only the big cluster need to do voltage scaling in the
notifier, since the voltage controlling is done by cpufreq-dt driver
in this version.
Therefore only one dvfs_info struct here.
+ int old_vproc, new_vproc, old_index, new_index;
+
+ if (!cpumask_test_cpu(freqs-cpu, dvfs_info-cpus))
+ return NOTIFY_DONE;
+
+ old_vproc = regulator_get_voltage(dvfs_info-proc_reg);
+ old_index = cpu_opp_table_get_volt_index(old_vproc);
+ new_index = cpu_opp_table_get_freq_index(freqs-new * 1000);
+ new_vproc = opp_tbl[new_index].vproc;
+
+ if (old_vproc == new_vproc)
+ return 0;
+
+ if ((action == CPUFREQ_PRECHANGE old_vproc new_vproc) ||
+ (action == CPUFREQ_POSTCHANGE old_vproc new_vproc))
+ mtk_cpufreq_voltage_trace(old_index, new_index);
+
+ return NOTIFY_OK;
+}
+
+static struct notifier_block mtk_cpufreq_nb = {
+ .notifier_call = mtk_cpufreq_notify,
+};
+
+static int cpu_opp_table_init(struct device *dev)
+{
+ struct device *cpu_dev = dvfs_info-cpu_dev;
+ struct cpu_opp_table *opp_tbl;
+ struct dev_pm_opp *opp;
+ int ret, cnt, i;
+ unsigned long rate, vproc, vsram;
+
+ ret = of_init_opp_table(cpu_dev);
+ if (ret) {
+ dev_err(dev, Failed to init mtk_opp_table: %d\n, ret);
+ return ret;
+ }
+
+ rcu_read_lock();
+
+ cnt = dev_pm_opp_get_opp_count(cpu_dev);
+ if (cnt 0) {
+ dev_err(cpu_dev, No OPP table is found: %d, cnt);
+ ret = cnt;
+ goto out_free_opp_tbl;
+ }
+
+ opp_tbl = devm_kcalloc(dev, (cnt + 1), sizeof(struct cpu_opp_table),
+ GFP_ATOMIC);
+ if (!opp_tbl) {
+ ret = -ENOMEM;
+ goto out_free_opp_tbl;
+ }
+
+ for (i = 0, rate = 0; i cnt; i++, rate++) {
+ opp = dev_pm_opp_find_freq_ceil(cpu_dev, rate);
+ if (IS_ERR(opp)) {
+ ret = PTR_ERR(opp);
+ goto out_free_opp_tbl;
+ }
+
+ vproc = dev_pm_opp_get_voltage(opp);
+ vproc = get_regulator_voltage_ceil(dvfs_info-proc_reg,
vproc);
+ vsram = vproc + VOLT_SHIFT_LOWER_LIMIT;
+ vsram = get_regulator_voltage_ceil(dvfs_info-sram_reg,
vsram);
+
+ if (vproc 0 || vsram 0) {
+ ret = -EINVAL;
+ goto out_free_opp_tbl;
+ }
+
+ opp_tbl[i].freq = rate;
+ opp_tbl[i].vproc = vproc;
+ opp_tbl[i].vsram = vsram;
+ }
+
+ opp_tbl[i].freq = 0;
+ opp_tbl[i].vproc = -1;
+ opp_tbl[i].vsram = -1;
+ dvfs_info-opp_tbl = opp_tbl;
+
+out_free_opp_tbl:
+ rcu_read_unlock();
+ of_free_opp_table(cpu_dev);
+
+ return ret;
+}
+
+static struct cpufreq_cpu_domain *get_cpu_domain(struct list_head
*domain_list,
+int cpu)
+{
+ struct list_head *node;
+
+ list_for_each(node, domain_list) {
+ struct cpufreq_cpu_domain *domain;
+
+ domain = container_of(node, struct cpufreq_cpu_domain, node);
+ if (cpumask_test_cpu(cpu, domain-cpus))
+ return domain;
+ }
+
+ return NULL;
+}
+
+static int mtk_cpufreq_probe(struct platform_device *pdev)
On a dual cluster big LITTLE (your system), how many times is probe
getting called ? Once or twice, i.e. for each cluster ??
The probe function will be called only once since it's triggered by the device
tree node in the 2nd patch of this series. Though it's not acceptable obviously.
+{
+ struct clk *inter_clk;
+ struct cpufreq_dt_platform_data *pd;
+ struct platform_device *dev;
+ unsigned long inter_freq;
+ int cpu, ret;
+
+ inter_clk = clk_get(pdev-dev, NULL);
How is this supposed to work ? How will pdev-dev give intermediate
clock ?
It works with the the device tree binding in the 2nd patch of this series, too.
Since the cpufreq node is not allowed, would you have some suggestions on
how to get the intermediate clock source in this case?
[PATCH v2 3/4] cpufreq: mediatek: add Mediatek cpufreq driver
In this patch, some SoC specific voltage scaling flow is implemented in the
cpufreq notifier of mtk-cpufreq driver.
Signed-off-by: pi-cheng.chen pi-cheng.c...@linaro.org
---
drivers/cpufreq/Kconfig.arm | 6 +
drivers/cpufreq/Makefile | 1 +
drivers/cpufreq/mtk-cpufreq.c | 346 ++
3 files changed, 353 insertions(+)
create mode 100644 drivers/cpufreq/mtk-cpufreq.c
diff --git a/drivers/cpufreq/Kconfig.arm b/drivers/cpufreq/Kconfig.arm
index 1b06fc4..f421653 100644
--- a/drivers/cpufreq/Kconfig.arm
+++ b/drivers/cpufreq/Kconfig.arm
@@ -263,3 +263,9 @@ config ARM_PXA2xx_CPUFREQ
This add the CPUFreq driver support for Intel PXA2xx SOCs.
If in doubt, say N.
+
+config ARM_MTK_CPUFREQ
+ bool Mediatek CPUFreq support
+ depends on ARCH_MEDIATEK CPUFREQ_DT REGULATOR
+ help
+ This adds the CPUFreq driver support for Mediatek SoCs.
diff --git a/drivers/cpufreq/Makefile b/drivers/cpufreq/Makefile
index 82a1821..05cb596 100644
--- a/drivers/cpufreq/Makefile
+++ b/drivers/cpufreq/Makefile
@@ -62,6 +62,7 @@ obj-$(CONFIG_ARM_HIGHBANK_CPUFREQ)+= highbank-cpufreq.o
obj-$(CONFIG_ARM_IMX6Q_CPUFREQ)+= imx6q-cpufreq.o
obj-$(CONFIG_ARM_INTEGRATOR) += integrator-cpufreq.o
obj-$(CONFIG_ARM_KIRKWOOD_CPUFREQ) += kirkwood-cpufreq.o
+obj-$(CONFIG_ARM_MTK_CPUFREQ) += mtk-cpufreq.o
obj-$(CONFIG_ARM_OMAP2PLUS_CPUFREQ)+= omap-cpufreq.o
obj-$(CONFIG_ARM_PXA2xx_CPUFREQ) += pxa2xx-cpufreq.o
obj-$(CONFIG_PXA3xx) += pxa3xx-cpufreq.o
diff --git a/drivers/cpufreq/mtk-cpufreq.c b/drivers/cpufreq/mtk-cpufreq.c
new file mode 100644
index 000..344d588
--- /dev/null
+++ b/drivers/cpufreq/mtk-cpufreq.c
@@ -0,0 +1,346 @@
+/*
+* Copyright (c) 2015 Linaro Ltd.
+* Author: Pi-Cheng Chen pi-cheng.c...@linaro.org
+*
+* This program is free software; you can redistribute it and/or modify
+* it under the terms of the GNU General Public License version 2 as
+* published by the Free Software Foundation.
+*
+* This program is distributed in the hope that it will be useful,
+* but WITHOUT ANY WARRANTY; without even the implied warranty of
+* MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
+* GNU General Public License for more details.
+*/
+
+#include linux/clk.h
+#include linux/cpu.h
+#include linux/cpufreq.h
+#include linux/cpufreq-dt.h
+#include linux/cpumask.h
+#include linux/module.h
+#include linux/of.h
+#include linux/platform_device.h
+#include linux/pm_opp.h
+#include linux/regulator/consumer.h
+#include linux/slab.h
+
+#define VOLT_SHIFT_LOWER_LIMIT 10
+#define VOLT_SHIFT_UPPER_LIMIT 20
+
+struct cpu_opp_table {
+ unsigned int freq;
+ int vproc;
+ int vsram;
+};
+
+static struct dvfs_info {
+ struct cpumask cpus;
+ struct cpu_opp_table *opp_tbl;
+ struct device *cpu_dev;
+ struct regulator *proc_reg;
+ struct regulator *sram_reg;
+} *dvfs_info;
+
+static int cpu_opp_table_get_freq_index(unsigned int freq)
+{
+ struct cpu_opp_table *opp_tbl = dvfs_info-opp_tbl;
+ int i;
+
+ for (i = 0; opp_tbl[i].freq != 0; i++) {
+ if (opp_tbl[i].freq = freq)
+ return i;
+ }
+
+ return -1;
+}
+
+static int cpu_opp_table_get_volt_index(unsigned int volt)
+{
+ struct cpu_opp_table *opp_tbl = dvfs_info-opp_tbl;
+ int i;
+
+ for (i = 0; opp_tbl[i].vproc != -1; i++)
+ if (opp_tbl[i].vproc = volt)
+ return i;
+
+ return -1;
+}
+
+static int get_regulator_voltage_ceil(struct regulator *regulator, int voltage)
+{
+ int cnt, i, volt = -1;
+
+ cnt = regulator_count_voltages(regulator);
+
+ for (i = 0; i cnt volt voltage; i++)
+ volt = regulator_list_voltage(regulator, i);
+
+ return volt;
+}
+
+static int mtk_cpufreq_voltage_trace(int old_index, int new_index)
+{
+ struct cpu_opp_table *opp_tbl = dvfs_info-opp_tbl;
+ int old_vproc, new_vproc, i, j;
+
+ old_vproc = regulator_get_voltage(dvfs_info-proc_reg);
+ new_vproc = opp_tbl[new_index].vproc;
+
+ if (old_vproc new_vproc) {
+ for (i = old_index; i new_index;) {
+ for (j = i; j = new_index; j--)
+ if (opp_tbl[i].vsram - opp_tbl[j].vproc
+VOLT_SHIFT_UPPER_LIMIT)
+ break;
+ i = j + 1;
+
+ regulator_set_voltage_tol(dvfs_info-proc_reg,
+ opp_tbl[i].vproc, 0);
+ regulator_set_voltage_tol(dvfs_info-sram_reg,
+ opp_tbl[i].vsram, 0);
+ }
+ } else if (old_vproc new_vproc) {
+ for (i = old_index; i new_index;) {
+ for (j = i;
Re: [PATCH v2 3/4] cpufreq: mediatek: add Mediatek cpufreq driver
Haven't reviewed it completely yet, but this is all I have done.
On 4 March 2015 at 14:19, pi-cheng.chen pi-cheng.c...@linaro.org wrote:
+static int mtk_cpufreq_notify(struct notifier_block *nb,
+ unsigned long action, void *data)
+{
+ struct cpufreq_freqs *freqs = data;
+ struct cpu_opp_table *opp_tbl = dvfs_info-opp_tbl;
There is only one dvfs info ? but there are two clusters, sorry got confused
a bit..
+ int old_vproc, new_vproc, old_index, new_index;
+
+ if (!cpumask_test_cpu(freqs-cpu, dvfs_info-cpus))
+ return NOTIFY_DONE;
+
+ old_vproc = regulator_get_voltage(dvfs_info-proc_reg);
+ old_index = cpu_opp_table_get_volt_index(old_vproc);
+ new_index = cpu_opp_table_get_freq_index(freqs-new * 1000);
+ new_vproc = opp_tbl[new_index].vproc;
+
+ if (old_vproc == new_vproc)
+ return 0;
+
+ if ((action == CPUFREQ_PRECHANGE old_vproc new_vproc) ||
+ (action == CPUFREQ_POSTCHANGE old_vproc new_vproc))
+ mtk_cpufreq_voltage_trace(old_index, new_index);
+
+ return NOTIFY_OK;
+}
+
+static struct notifier_block mtk_cpufreq_nb = {
+ .notifier_call = mtk_cpufreq_notify,
+};
+
+static int cpu_opp_table_init(struct device *dev)
+{
+ struct device *cpu_dev = dvfs_info-cpu_dev;
+ struct cpu_opp_table *opp_tbl;
+ struct dev_pm_opp *opp;
+ int ret, cnt, i;
+ unsigned long rate, vproc, vsram;
+
+ ret = of_init_opp_table(cpu_dev);
+ if (ret) {
+ dev_err(dev, Failed to init mtk_opp_table: %d\n, ret);
+ return ret;
+ }
+
+ rcu_read_lock();
+
+ cnt = dev_pm_opp_get_opp_count(cpu_dev);
+ if (cnt 0) {
+ dev_err(cpu_dev, No OPP table is found: %d, cnt);
+ ret = cnt;
+ goto out_free_opp_tbl;
+ }
+
+ opp_tbl = devm_kcalloc(dev, (cnt + 1), sizeof(struct cpu_opp_table),
+ GFP_ATOMIC);
+ if (!opp_tbl) {
+ ret = -ENOMEM;
+ goto out_free_opp_tbl;
+ }
+
+ for (i = 0, rate = 0; i cnt; i++, rate++) {
+ opp = dev_pm_opp_find_freq_ceil(cpu_dev, rate);
+ if (IS_ERR(opp)) {
+ ret = PTR_ERR(opp);
+ goto out_free_opp_tbl;
+ }
+
+ vproc = dev_pm_opp_get_voltage(opp);
+ vproc = get_regulator_voltage_ceil(dvfs_info-proc_reg,
vproc);
+ vsram = vproc + VOLT_SHIFT_LOWER_LIMIT;
+ vsram = get_regulator_voltage_ceil(dvfs_info-sram_reg,
vsram);
+
+ if (vproc 0 || vsram 0) {
+ ret = -EINVAL;
+ goto out_free_opp_tbl;
+ }
+
+ opp_tbl[i].freq = rate;
+ opp_tbl[i].vproc = vproc;
+ opp_tbl[i].vsram = vsram;
+ }
+
+ opp_tbl[i].freq = 0;
+ opp_tbl[i].vproc = -1;
+ opp_tbl[i].vsram = -1;
+ dvfs_info-opp_tbl = opp_tbl;
+
+out_free_opp_tbl:
+ rcu_read_unlock();
+ of_free_opp_table(cpu_dev);
+
+ return ret;
+}
+
+static struct cpufreq_cpu_domain *get_cpu_domain(struct list_head
*domain_list,
+int cpu)
+{
+ struct list_head *node;
+
+ list_for_each(node, domain_list) {
+ struct cpufreq_cpu_domain *domain;
+
+ domain = container_of(node, struct cpufreq_cpu_domain, node);
+ if (cpumask_test_cpu(cpu, domain-cpus))
+ return domain;
+ }
+
+ return NULL;
+}
+
+static int mtk_cpufreq_probe(struct platform_device *pdev)
On a dual cluster big LITTLE (your system), how many times is probe
getting called ? Once or twice, i.e. for each cluster ??
+{
+ struct clk *inter_clk;
+ struct cpufreq_dt_platform_data *pd;
+ struct platform_device *dev;
+ unsigned long inter_freq;
+ int cpu, ret;
+
+ inter_clk = clk_get(pdev-dev, NULL);
How is this supposed to work ? How will pdev-dev give intermediate
clock ?
+ if (IS_ERR(inter_clk)) {
+ if (PTR_ERR(inter_clk) == -EPROBE_DEFER) {
+ dev_warn(pdev-dev, clock not ready. defer
probeing.\n);
+ return -EPROBE_DEFER;
+ }
+
+ dev_err(pdev-dev, Failed to get intermediate clock\n);
+ return -ENODEV;
+ }
+ inter_freq = clk_get_rate(inter_clk);
+
+ pd = devm_kzalloc(pdev-dev, sizeof(*pd), GFP_KERNEL);
+ if (!pd)
+ return -ENOMEM;
+
+ dvfs_info = devm_kzalloc(pdev-dev, sizeof(*dvfs_info), GFP_KERNEL);
+ if (!dvfs_info)
+ return -ENOMEM;
Instead of two allocations, you could have made pd part
