Re: [PATCH 4/4] cpufreq: qcom-nvmem: make qcom_cpufreq_get_msm_id() return the SoC ID

[Robert Marko]

On Fri, 3 Mar 2023 at 21:46, Konrad Dybcio <konrad.dybcio@xxxxxxxxxx> wrote:
>
>
>
> On 3.03.2023 19:38, Robert Marko wrote:
> > On Sat, 18 Feb 2023 at 21:40, Konrad Dybcio <konrad.dybcio@xxxxxxxxxx> wrote:
> >>
> >>
> >>
> >> On 18.02.2023 21:36, Dmitry Baryshkov wrote:
> >>> On Sat, 18 Feb 2023 at 16:43, Konrad Dybcio <konrad.dybcio@xxxxxxxxxx> wrote:
> >>>>
> >>>>
> >>>>
> >>>> On 21.01.2023 12:29, Robert Marko wrote:
> >>>>> Currently, qcom_cpufreq_get_msm_id() does not simply return the SoC ID
> >>>>> after getting it via SMEM call but instead uses an enum to encode the
> >>>>> matched SMEM ID to 2 variants of MSM8996 which are then used in
> >>>>> qcom_cpufreq_kryo_name_version() to set the supported version.
> >>>>>
> >>>>> This prevents qcom_cpufreq_get_msm_id() from being universal and its doing
> >>>>> more than its name suggests, so lets make it just return the SoC ID
> >>>>> directly which allows matching directly on the SoC ID and removes the need
> >>>>> for msm8996_version enum which simplifies the driver.
> >>>>> It also allows reusing the qcom_cpufreq_get_msm_id() for new SoC-s.
> >>>>>
> >>>>> Signed-off-by: Robert Marko <robimarko@xxxxxxxxx>
> >>>>> ---
> >>>>>  drivers/cpufreq/qcom-cpufreq-nvmem.c | 44 ++++++++--------------------
> >>>>>  1 file changed, 12 insertions(+), 32 deletions(-)
> >>>>>
> >>>>> diff --git a/drivers/cpufreq/qcom-cpufreq-nvmem.c b/drivers/cpufreq/qcom-cpufreq-nvmem.c
> >>>>> index da55d2e1925a..9deaf9521d6d 100644
> >>>>> --- a/drivers/cpufreq/qcom-cpufreq-nvmem.c
> >>>>> +++ b/drivers/cpufreq/qcom-cpufreq-nvmem.c
> >>>>> @@ -32,12 +32,6 @@
> >>>>>
> >>>>>  #include <dt-bindings/arm/qcom,ids.h>
> >>>>>
> >>>>> -enum _msm8996_version {
> >>>>> -     MSM8996_V3,
> >>>>> -     MSM8996_SG,
> >>>>> -     NUM_OF_MSM8996_VERSIONS,
> >>>>> -};
> >>>>> -
> >>>>>  struct qcom_cpufreq_drv;
> >>>>>
> >>>>>  struct qcom_cpufreq_match_data {
> >>>>> @@ -134,30 +128,16 @@ static void get_krait_bin_format_b(struct device *cpu_dev,
> >>>>>       dev_dbg(cpu_dev, "PVS version: %d\n", *pvs_ver);
> >>>>>  }
> >>>>>
> >>>>> -static enum _msm8996_version qcom_cpufreq_get_msm_id(void)
> >>>>> +static int qcom_cpufreq_get_msm_id(void)
> >>>> This should be u32 as info->id is __le32
> >
> > Nice catch.
> >
> >
> >>>>
> >>>> And please export this function from socinfo, it'll come in
> >>>> useful for other drivers!
> >
> > I intentionally did not do that as socinfo is currently fully optional
> > and I dont really like
> > the idea of making it required for anything using SMEM.
> "anything using SMEM"? As in the drivers, or SoCs?
> If the former, I don't see how exporting a function from within
> socid and using it here would make it required for other drivers.
> If the latter, we're talking non-qcom SoCs. SMEM has been with
> us forever.

I feel we have a misunderstanding,
currently, cpufreq-nvmem does not depend on socinfo being built
so I don't want to require it as a dependency in order to get the SMEM ID.

Granted, socinfo is useful on any QCA SoC so if adding new dependecies to
cpufreq-nvmem is acceptable I am not against exporting it there.
>
>
> I'm planning to reuse this for Adreno speedbin matching. It's one
> of those blocks that don't have a revision and/or bin reigster
> within themselves.

I understand the use case, I am sure it will be required in some other places
sooner or later as well.

Regards,
Robert
>
> Konrad
> >
> > Regards,
> > Robert
> >
> >>
> >> Konrad
> >>>