| Message ID | cover.1493203884.git.viresh.kumar@linaro.org |
|---|---|
| Headers | show |
| Series | PM / Domains: Implement domain performance states | expand |
Sudeep Holla <sudeep.holla@arm.com> writes: > On 28/04/17 21:48, Rob Herring wrote: >> On Wed, Apr 26, 2017 at 04:27:05PM +0530, Viresh Kumar wrote: >>> Power-domains need to express their active states in DT and the devices >>> within the power-domain need to express their dependency on those active >>> states. The power-domains can use the OPP tables without any >>> modifications to the bindings. >>> >>> Add a new property "power-domain-opp", which will contain phandle to the >>> OPP node of the parent power domain. This is required for devices which >>> have dependency on the configured active state of the power domain for >>> their working. >>> >>> For some platforms the actual frequency and voltages of the power >>> domains are managed by the firmware and are so hidden from the high >>> level operating system. The "opp-hz" property is relaxed a bit to >>> contain indexes instead of actual frequency values to support such >>> platforms. >>> >>> Signed-off-by: Viresh Kumar <viresh.kumar@linaro.org> >>> --- >>> Documentation/devicetree/bindings/opp/opp.txt | 74 ++++++++++++++++++++++++++- >>> 1 file changed, 73 insertions(+), 1 deletion(-) >>> >>> diff --git a/Documentation/devicetree/bindings/opp/opp.txt b/Documentation/devicetree/bindings/opp/opp.txt >>> index 63725498bd20..6e30cae2a936 100644 >>> --- a/Documentation/devicetree/bindings/opp/opp.txt >>> +++ b/Documentation/devicetree/bindings/opp/opp.txt >>> @@ -77,7 +77,10 @@ This defines voltage-current-frequency combinations along with other related >>> properties. >>> >>> Required properties: >>> -- opp-hz: Frequency in Hz, expressed as a 64-bit big-endian integer. >>> +- opp-hz: Frequency in Hz, expressed as a 64-bit big-endian integer. In some >>> + cases the exact frequency in Hz may be hidden from the OS by the firmware and >>> + this field may contain values that represent the frequency in a firmware >>> + dependent way, for example an index of an array in the firmware. >> >> Not really sure OPP binding makes sense here. What about all the other >> properties. We expose voltage, but not freq? >> > > I completely agree with that and I have been pushing this to be > represented as just regulators[0]. Mark B seem to dislike that > idea [1] And Mark is right, because what's being described is not (simply) a voltage regultor. While it might be "just" voltage on some SoCs (for now), it is clearly about performance (a.k.a. OPP) on others. Kevin
On 06-05-17, 11:58, Kevin Hilman wrote: > Rob Herring <robh@kernel.org> writes: > > > On Wed, Apr 26, 2017 at 04:27:05PM +0530, Viresh Kumar wrote: > >> Power-domains need to express their active states in DT and the devices > >> within the power-domain need to express their dependency on those active > >> states. The power-domains can use the OPP tables without any > >> modifications to the bindings. > >> > >> Add a new property "power-domain-opp", which will contain phandle to the > >> OPP node of the parent power domain. This is required for devices which > >> have dependency on the configured active state of the power domain for > >> their working. > >> > >> For some platforms the actual frequency and voltages of the power > >> domains are managed by the firmware and are so hidden from the high > >> level operating system. The "opp-hz" property is relaxed a bit to > >> contain indexes instead of actual frequency values to support such > >> platforms. > >> > >> Signed-off-by: Viresh Kumar <viresh.kumar@linaro.org> > >> --- > >> Documentation/devicetree/bindings/opp/opp.txt | 74 ++++++++++++++++++++++++++- > >> 1 file changed, 73 insertions(+), 1 deletion(-) > >> > >> diff --git a/Documentation/devicetree/bindings/opp/opp.txt b/Documentation/devicetree/bindings/opp/opp.txt > >> index 63725498bd20..6e30cae2a936 100644 > >> --- a/Documentation/devicetree/bindings/opp/opp.txt > >> +++ b/Documentation/devicetree/bindings/opp/opp.txt > >> @@ -77,7 +77,10 @@ This defines voltage-current-frequency combinations along with other related > >> properties. > >> > >> Required properties: > >> -- opp-hz: Frequency in Hz, expressed as a 64-bit big-endian integer. > >> +- opp-hz: Frequency in Hz, expressed as a 64-bit big-endian integer. In some > >> + cases the exact frequency in Hz may be hidden from the OS by the firmware and > >> + this field may contain values that represent the frequency in a firmware > >> + dependent way, for example an index of an array in the firmware. > > > > Not really sure OPP binding makes sense here. > > I think OPP makes perfect sense here, because microcontroller firmware > is managaging OPPs in hardware. We just may not know the exact voltage > and/or frequency (and the firmware/hardware may even be doing AVS for > micro-adjustments.) Yes, AVS is being done for the Qcom SoC as well. > > What about all the other properties. We expose voltage, but not freq? > > I had the same question. Seems the same comment about an abstract > "index" is needed for voltage also. Why should we do that? Here are the cases that I had in mind while writing this: - DT only contains the performance-index and nothing else (i.e. voltages aren't exposed). We wouldn't be required to fill the microvolt property as it is optional. - DT contains both performance-index and voltages. The microvolts property will contain the actual voltages and opp-hz will contain the index. I don't see why would we like to put some index value in the microvolts property. We are setting the index value in the opp-hz property to avoid adding extra fields and making sure opp-hz is still the unique property for the nodes. > >> > >> Optional properties: > >> - opp-microvolt: voltage in micro Volts. > >> @@ -154,6 +157,13 @@ properties. > >> > >> - status: Marks the node enabled/disabled. > >> > >> +- power-domain-opp: Phandle to the OPP node of the parent power-domain. The > >> + parent power-domain should be configured to the OPP whose node is pointed by > >> + the phandle, in order to configure the device for the OPP node that contains > >> + this property. The order in which the device and power domain should be > >> + configured is implementation defined. The OPP table of a device can set this > >> + property only if the device node contains "power-domains" property. > >> + > > I do understand the need to map a device OPP to a parent power-domain > OPP, but I really don't like another phandle. > > First, just because a device OPP changes does not mean that a > power-domain OPP has to change. What really needs to be specified is a > minimum requirement, not an exact OPP. IOW, if a device changes OPP, > the power-domain OPP has to be *at least* an OPP that can guarantee that > level of performance, but could also be a more performant OPP, right? Right and that's how the code is interpreting it right now. Yes, the description above should have been more clear on that though. > Also, the parent power-domain driver will have a list of all its > devices, and be able to get OPPs from those devices. > > IMO, we should do the first (few) implementations of this feature from > the power-domain driver itself, and not try to figure out how to define > this for everyone in DT until we have a better handle on it (pun > intended) ;) Hmm, I am not sure how things are going to work in that case. The opp-hz value read from the phandle is passed to the QoS framework in this series, which makes sure that we select the highest requested performance point for a particular power-domain. The index value is required to be present with the OPP framework to make it all work, at least based on the way I have designed it for now. -- viresh
On 05/08/2017 09:45 AM, Viresh Kumar wrote: > On 06-05-17, 11:58, Kevin Hilman wrote: >> Rob Herring <robh@kernel.org> writes: >> >>> On Wed, Apr 26, 2017 at 04:27:05PM +0530, Viresh Kumar wrote: >>>> Power-domains need to express their active states in DT and the devices >>>> within the power-domain need to express their dependency on those active >>>> states. The power-domains can use the OPP tables without any >>>> modifications to the bindings. >>>> >>>> Add a new property "power-domain-opp", which will contain phandle to the >>>> OPP node of the parent power domain. This is required for devices which >>>> have dependency on the configured active state of the power domain for >>>> their working. >>>> >>>> For some platforms the actual frequency and voltages of the power >>>> domains are managed by the firmware and are so hidden from the high >>>> level operating system. The "opp-hz" property is relaxed a bit to >>>> contain indexes instead of actual frequency values to support such >>>> platforms. >>>> >>>> Signed-off-by: Viresh Kumar <viresh.kumar@linaro.org> >>>> --- >>>> Documentation/devicetree/bindings/opp/opp.txt | 74 ++++++++++++++++++++++++++- >>>> 1 file changed, 73 insertions(+), 1 deletion(-) >>>> >>>> diff --git a/Documentation/devicetree/bindings/opp/opp.txt b/Documentation/devicetree/bindings/opp/opp.txt >>>> index 63725498bd20..6e30cae2a936 100644 >>>> --- a/Documentation/devicetree/bindings/opp/opp.txt >>>> +++ b/Documentation/devicetree/bindings/opp/opp.txt >>>> @@ -77,7 +77,10 @@ This defines voltage-current-frequency combinations along with other related >>>> properties. >>>> >>>> Required properties: >>>> -- opp-hz: Frequency in Hz, expressed as a 64-bit big-endian integer. >>>> +- opp-hz: Frequency in Hz, expressed as a 64-bit big-endian integer. In some >>>> + cases the exact frequency in Hz may be hidden from the OS by the firmware and >>>> + this field may contain values that represent the frequency in a firmware >>>> + dependent way, for example an index of an array in the firmware. >>> >>> Not really sure OPP binding makes sense here. >> >> I think OPP makes perfect sense here, because microcontroller firmware >> is managaging OPPs in hardware. We just may not know the exact voltage >> and/or frequency (and the firmware/hardware may even be doing AVS for >> micro-adjustments.) > > Yes, AVS is being done for the Qcom SoC as well. > >>> What about all the other properties. We expose voltage, but not freq? >> >> I had the same question. Seems the same comment about an abstract >> "index" is needed for voltage also. > > Why should we do that? Here are the cases that I had in mind while writing this: > > - DT only contains the performance-index and nothing else (i.e. voltages aren't > exposed). > > We wouldn't be required to fill the microvolt property as it is optional. So the performance-index is specified in opp-hz property? What if the microcontroller firmware maps the performance-index to voltage but expects linux to scale the frequency? There is no way to specify a performance-index *and* a frequency for a OPP now I guess? > > - DT contains both performance-index and voltages. > > The microvolts property will contain the actual voltages and opp-hz will > contain the index. So this is for cases where the performance-index maps to a freq managed by the microcontroller and voltages managed by linux? I have a case of exact opposite and I don't see now how to handle it now with these bindings. > > I don't see why would we like to put some index value in the microvolts > property. We are setting the index value in the opp-hz property to avoid adding > extra fields and making sure opp-hz is still the unique property for the nodes. Maybe to handle the case like what I described above? -- Qualcomm Innovation Center, Inc. is a member of Code Aurora Forum, hosted by The Linux Foundation
On 08/05/17 08:13, Viresh Kumar wrote: > On 03-05-17, 12:29, Sudeep Holla wrote: >> >> >> On 28/04/17 21:48, Rob Herring wrote: >>> On Wed, Apr 26, 2017 at 04:27:05PM +0530, Viresh Kumar wrote: >>>> Power-domains need to express their active states in DT and the devices >>>> within the power-domain need to express their dependency on those active >>>> states. The power-domains can use the OPP tables without any >>>> modifications to the bindings. >>>> >>>> Add a new property "power-domain-opp", which will contain phandle to the >>>> OPP node of the parent power domain. This is required for devices which >>>> have dependency on the configured active state of the power domain for >>>> their working. >>>> >>>> For some platforms the actual frequency and voltages of the power >>>> domains are managed by the firmware and are so hidden from the high >>>> level operating system. The "opp-hz" property is relaxed a bit to >>>> contain indexes instead of actual frequency values to support such >>>> platforms. >>>> >>>> Signed-off-by: Viresh Kumar <viresh.kumar@linaro.org> >>>> --- >>>> Documentation/devicetree/bindings/opp/opp.txt | 74 ++++++++++++++++++++++++++- >>>> 1 file changed, 73 insertions(+), 1 deletion(-) >>>> >>>> diff --git a/Documentation/devicetree/bindings/opp/opp.txt b/Documentation/devicetree/bindings/opp/opp.txt >>>> index 63725498bd20..6e30cae2a936 100644 >>>> --- a/Documentation/devicetree/bindings/opp/opp.txt >>>> +++ b/Documentation/devicetree/bindings/opp/opp.txt >>>> @@ -77,7 +77,10 @@ This defines voltage-current-frequency combinations along with other related >>>> properties. >>>> >>>> Required properties: >>>> -- opp-hz: Frequency in Hz, expressed as a 64-bit big-endian integer. >>>> +- opp-hz: Frequency in Hz, expressed as a 64-bit big-endian integer. In some >>>> + cases the exact frequency in Hz may be hidden from the OS by the firmware and >>>> + this field may contain values that represent the frequency in a firmware >>>> + dependent way, for example an index of an array in the firmware. >>> >>> Not really sure OPP binding makes sense here. What about all the other >>> properties. We expose voltage, but not freq? >>> >> >> I completely agree with that and I have been pushing this to be >> represented as just regulators[0]. Mark B seem to dislike that >> idea [1] > > Just as an update, Rajendra confirmed (offline) that for some of the > implementations, the microcontroller handles both frequency and > voltages of a device. So it isn't just a regulator anymore and as me > and Kevin were saying, we need a complete OPP here. > Yes, I followed the thread and figured that out. But Rajendra also raised "What if the microcontroller firmware maps the performance-index to voltage but expects linux to scale the frequency? There is no way to specify a performance-index *and* a frequency for a OPP now I guess? So this needs to be addressd now IIUC. So as Kevin pointed out, we need to experiment and look at all possibilities before finalizing the bindings. Better to have examples for all these and describe how bindings are be used including how to distinguish between these use-case from the bindings if it's not implicit. -- Regards, Sudeep
On 08-05-17, 14:57, Sudeep Holla wrote: > Yes, I followed the thread and figured that out. But Rajendra also > raised "What if the microcontroller firmware maps the performance-index > to voltage but expects linux to scale the frequency? There is no way to > specify a performance-index *and* a frequency for a OPP now I guess? So > this needs to be addressd now IIUC. No, he misunderstood it. He was saying that the domain needs a performance-index and the device needs freq-scaling, how do we do that? He thought that there will be just one OPP table for the device here, but we will actually have two and that would work. > So as Kevin pointed out, we need to experiment and look at all > possibilities before finalizing the bindings. Better to have examples > for all these and describe how bindings are be used including how to > distinguish between these use-case from the bindings if it's not implicit. Yeah, I have some doubts on how we are going to implement that and looking for more input from him. -- viresh
On 12 May 2017 at 20:29, Kevin Hilman <khilman@baylibre.com> wrote: > Viresh Kumar <viresh.kumar@linaro.org> writes: >> Why should we do that? > > For starters, because the lack of it looks very strange upon first read > (notice that both Rob and I pointed that out), and because you didn't > explain why in the first place, it draws attention. :) >> I don't see why would we like to put some index value in the microvolts >> property. We are setting the index value in the opp-hz property to avoid adding >> extra fields and making sure opp-hz is still the unique property for the nodes. > > What about the case where firmware wants exact frequencies, and > microvolts property is just an index? > > The point is, you have a very specific SoC and use-case in mind, but the > goal of a binding change like this is to make something that could be > generically useful. I agree, but I am not sure of having such a case in very near future at least. Wouldn't it be wise to not touch opp-microvolt for now and update it only when needed? Its not a big change anyway.. >> Hmm, I am not sure how things are going to work in that case. The opp-hz value >> read from the phandle is passed to the QoS framework in this series, which makes >> sure that we select the highest requested performance point for a particular >> power-domain. The index value is required to be present with the OPP framework >> to make it all work, at least based on the way I have designed it for now. > > IMO, this kind of dependency isn't the job of the OPP framework, it's > the job of the power-domain governor. Okay. So the way it will work with the current suggestions is: - OPP framework gets DVFS update request for device X - OPP framework finds that the device has a power-domain and so it asks the power-domain framework to set the device in a particular state corresponding to the OPP (if we are going to a higher OPP). - If the power-domain supports state selection, it does that or returns error. (Actually we can optimize this by asking the genpd initially if state selection is possible, only then OPP core calls the genpd API). - The genpd API will manage a list of all devices in the domain (which it already does) and also the states selected for them. It finds the max of the requested states and selects that. - Note that the QoS framework isn't there in the picture anymore. Will that be fine ? -- viresh
Hi, Here is the 6th version of the series, which incorporates feedback from Kevin and Sudeep: - Use freq/voltage in OPP table as it is for power domain and don't create "domain-performance-level" property - Take care of domain providers that provide multiple domains Here is a brief summary of the problem I am trying to solve. Some platforms have the capability to configure the performance state of their power domains. The process of configuring the performance state is pretty much platform dependent and we may need to work with a wide range of configurables. For some platforms, like Qcom, it can be a positive integer value alone, while in other cases it can be voltage levels, etc. The power-domain framework until now was only designed for the idle state management of the device and this needs to change in order to reuse the power-domain framework for active state management of the devices. This series adapts the genpd and OPP frameworks to allow OPP tables to be used for the genpd devices as well. The first 2 patches update the DT bindings of the power-domains and OPP tables. And the other 7 patches implement the details in QoS, genpd and OPP frameworks. This is tested currently by hacking the kernel a bit with virtual power-domains for the dual A15 exynos platform. The earlier version of patches was also tested by Rajendra Nayak (Qcom) on *real* Qualcomm hardware for which this work is getting done. Hope this version should work as well. Here is sample DT and C code we need to write for platforms: DT: --- / { domain_opp_table: opp_table0 { compatible = "operating-points-v2"; domain_opp_1: opp-1 { opp-hz = /bits/ 64 <1>; opp-microvolt = <975000 970000 985000>; }; domain_opp_2: opp-2 { opp-hz = /bits/ 64 <2>; opp-microvolt = <1075000 1000000 1085000>; }; }; foo_domain: power-controller@12340000 { compatible = "foo,power-controller"; reg = <0x12340000 0x1000>; #power-domain-cells = <0>; operating-points-v2 = <&domain_opp_table>; } cpu0_opp_table: opp_table1 { compatible = "operating-points-v2"; opp-shared; opp-1000000000 { opp-hz = /bits/ 64 <1000000000>; power-domain-opp = <&domain_opp_1>; }; opp-1100000000 { opp-hz = /bits/ 64 <1100000000>; power-domain-opp = <&domain_opp_2>; }; opp-1200000000 { opp-hz = /bits/ 64 <1200000000>; power-domain-opp = <&domain_opp_2>; }; }; cpus { #address-cells = <1>; #size-cells = <0>; cpu@0 { compatible = "arm,cortex-a9"; reg = <0>; clocks = <&clk_controller 0>; clock-names = "cpu"; operating-points-v2 = <&cpu0_opp_table>; power-domains = <&foo_domain>; }; }; }; Driver code: ------------ static int pd_performance(struct generic_pm_domain *domain, unsigned int state) { struct dev_pm_opp *opp; opp = dev_pm_opp_find_freq_exact(&domain->dev, state, true); /* Use OPP and state in platform specific way */ return 0; } static const struct of_device_id pm_domain_of_match[] __initconst = { { .compatible = "foo,genpd", }, { }, }; static int __init genpd_test_init(void) { struct device *dev = get_cpu_device(0); struct device_node *np; const struct of_device_id *match; int n; int ret; for_each_matching_node_and_match(np, pm_domain_of_match, &match) { pd.name = kstrdup_const(strrchr(np->full_name, '/') + 1, GFP_KERNEL); if (!pd.name) { of_node_put(np); return -ENOMEM; } pd.set_performance_state = pd_performance; pm_genpd_init(&pd, NULL, false); of_genpd_add_provider_simple(np, &pd); } ret = dev_pm_domain_attach(dev, false); return ret; } Pushed here as well: https://git.linaro.org/people/viresh.kumar/linux.git/log/?h=opp/genpd-performance-state V5->V6: - Use freq/voltage in OPP table as it is for power domain and don't create "domain-performance-level" property - Create new "power-domain-opp" property for the devices. - Take care of domain providers that provide multiple domains and extend "operating-points-v2" property to contain a list of phandles - Update code according to those bindings. V4->V5: - Only 3 patches were resent and 2 of them are Acked from Ulf. V3->V4: - Use OPP table for genpd devices as well. - Add struct device to genpd, in order to reuse OPP infrastructure. - Based over: https://marc.info/?l=linux-kernel&m=148972988002317&w=2 - Fixed examples in DT document to have voltage in target,min,max order. V2->V3: - Based over latest pm/linux-next - Bindings and code are merged together - Lots of updates in bindings - the performance-states node is present within the power-domain now, instead of its phandle. - performance-level property is replaced by "reg". - domain-performance-state property of the consumers contain an integer value now instead of phandle. - Lots of updates to the code as well - Patch "PM / QOS: Add default case to the switch" is merged with other patches and the code is changed a bit as well. - Don't pass 'type' to dev_pm_qos_add_notifier(), rather handle all notifiers with a single list. A new patch is added for that. - The OPP framework patch can be applied now and has proper SoB from me. - Dropped "PM / domain: Save/restore performance state at runtime suspend/resume". - Drop all WARN(). - Tested-by Rajendra nayak. V1->V2: - Based over latest pm/linux-next - It is mostly a resend of what is sent earlier as this series hasn't got any reviews so far and Rafael suggested that its better I resend it. - Only the 4/6 patch got an update, which was shared earlier as reply to V1 as well. It has got several fixes for taking care of power domain hierarchy, etc. -- viresh Viresh Kumar (9): PM / OPP: Introduce "power-domain-opp" property PM / Domains: Allow OPP table to be used for power-domains PM / QOS: Keep common notifier list for genpd constraints PM / QOS: Add DEV_PM_QOS_PERFORMANCE request PM / OPP: Add support to parse "power-domain-opp" property PM / OPP: Implement dev_pm_opp_of_add_table_indexed() PM / domain: Register PM QOS performance notifier PM / Domain: Add struct device to genpd PM / Domain: Add support to parse domain's OPP table Documentation/devicetree/bindings/opp/opp.txt | 74 ++++++- .../devicetree/bindings/power/power_domain.txt | 106 ++++++++++ Documentation/power/pm_qos_interface.txt | 2 +- drivers/base/power/domain.c | 222 +++++++++++++++++++-- drivers/base/power/opp/core.c | 72 +++++++ drivers/base/power/opp/debugfs.c | 3 + drivers/base/power/opp/of.c | 123 +++++++++++- drivers/base/power/opp/opp.h | 12 ++ drivers/base/power/qos.c | 36 +++- include/linux/pm_domain.h | 6 + include/linux/pm_opp.h | 6 + include/linux/pm_qos.h | 16 ++ kernel/power/qos.c | 2 +- 13 files changed, 641 insertions(+), 39 deletions(-) -- 2.12.0.432.g71c3a4f4ba37