[PATCH 0/2] Correct the processing for base_frequency

[PATCH 0/2] Correct the processing for base_frequency

[Srinivas Pandruvada]

The base_frequency display in cpufreq sysfs for intel_pstate gets the
guaranteed ratio by reading CPPC guaranteed performance register as a
first preference before falling back to x86 MSR for Hardware P-state
Capabilities. The current code in cppc_acpi.c assumed that "guaranteed
performance register" can be an integer field, which is invalid as per
ACPI spec. So this change explicitly check for INTEGER values for
invalid BIOS/firmware and ignore. 
Also guaranteed performance register field is optional and when not
present, nominal performance can be used as the guaranteed performance.
But spec calls that this is true only in non-autonomous mode. So
no change is made in cppc_acpi.c to make nominal as guaranteed in this
case to avoid dependency on autonomous and non-autonomous mode. Instead
a change is added to intel_pstate driver, which is specific to x86 to
make nominal as guaranteed when guaranteed performance field is absent
or has invalid value.
Also we are working to clarify this non-autonomous mode requirement
through ACPI standard body. 

Srinivas Pandruvada (2):
  ACPI / CPPC: Fix processing for guaranteed performance
  cpufreq: intel_pstate: Also use cppc nominal_perf for base_frequency

 drivers/acpi/cppc_acpi.c       | 10 ++++++++--
 drivers/cpufreq/intel_pstate.c |  3 +++
 2 files changed, 11 insertions(+), 2 deletions(-)

-- 
2.17.2

[PATCH 1/2] ACPI / CPPC: Fix processing for guaranteed performance

[Srinivas Pandruvada]

As per ACPI specification "Guaranteed Performance Register" is a "Buffer"
field. It can't be "Integer" field. So treat "Integer" type as invalid and
ignore "Guaranteed Performance Register".
Also save one cpc_read() call, when "Guaranteed Performance Register" is
not present, which means register defined as:
"Register(SystemMemory, 0, 0, 0, 0)".

Fixes: 29523f095397 ("ACPI / CPPC: Add support for guaranteed performance")
Signed-off-by: Srinivas Pandruvada <srinivas.pandruvada@linux.intel.com>
Cc: 4.20+ <stable@vger.kernel.org> # 4.20+
---
 drivers/acpi/cppc_acpi.c | 10 ++++++++--
 1 file changed, 8 insertions(+), 2 deletions(-)

diff --git a/drivers/acpi/cppc_acpi.c b/drivers/acpi/cppc_acpi.c
index 1b207fca1420..3f6c290e06af 100644
--- a/drivers/acpi/cppc_acpi.c
+++ b/drivers/acpi/cppc_acpi.c
@@ -1150,8 +1150,14 @@ int cppc_get_perf_caps(int cpunum, struct cppc_perf_caps *perf_caps)
 	cpc_read(cpunum, nominal_reg, &nom);
 	perf_caps->nominal_perf = nom;
 
-	cpc_read(cpunum, guaranteed_reg, &guaranteed);
-	perf_caps->guaranteed_perf = guaranteed;
+	if (guaranteed_reg->type == ACPI_TYPE_INTEGER  ||
+	    (guaranteed_reg->type == ACPI_TYPE_BUFFER &&
+	     IS_NULL_REG(&guaranteed_reg->cpc_entry.reg))) {
+		perf_caps->guaranteed_perf = 0;
+	} else {
+		cpc_read(cpunum, guaranteed_reg, &guaranteed);
+		perf_caps->guaranteed_perf = guaranteed;
+	}
 
 	cpc_read(cpunum, lowest_non_linear_reg, &min_nonlinear);
 	perf_caps->lowest_nonlinear_perf = min_nonlinear;
-- 
2.17.2

[PATCH 2/2] cpufreq: intel_pstate: Also use cppc nominal_perf for base_frequency

[Srinivas Pandruvada]

ACPI specifications stat that if the "Guaranteed Performance Register" is
not implemented, OSPM assumes guaranteed performance is always equal to
nominal performance. So for invalid and unimplemented guaranteed
performance register, use nominal performance as guaranteed performance.

This change will fallback to nominal_perf when guranteed_perf is invalid.
If nominal_perf is also invalid, then fallback to existing implementation,
which is to read from HWP Capabilities MSR.

Fixes: 86d333a8cc7f ("cpufreq: intel_pstate: Add base_frequency attribute")
Signed-off-by: Srinivas Pandruvada <srinivas.pandruvada@linux.intel.com>
Cc: 4.20+ <stable@vger.kernel.org> # 4.20+
---
 drivers/cpufreq/intel_pstate.c | 3 +++
 1 file changed, 3 insertions(+)

diff --git a/drivers/cpufreq/intel_pstate.c b/drivers/cpufreq/intel_pstate.c
index 7b4b0a7ac68b..e16dea241c55 100644
--- a/drivers/cpufreq/intel_pstate.c
+++ b/drivers/cpufreq/intel_pstate.c
@@ -385,6 +385,9 @@ static int intel_pstate_get_cppc_guranteed(int cpu)
 	if (ret)
 		return ret;
 
+	if (!cppc_perf.guaranteed_perf)
+		return cppc_perf.nominal_perf;
+
 	return cppc_perf.guaranteed_perf;
 }
 
-- 
2.17.2

Re: [PATCH 1/2] ACPI / CPPC: Fix processing for guaranteed performance

[Rafael J. Wysocki]

On Fri, Mar 22, 2019 at 11:45 PM Srinivas Pandruvada
<srinivas.pandruvada@linux.intel.com> wrote:
>
> As per ACPI specification "Guaranteed Performance Register" is a "Buffer"
> field. It can't be "Integer" field. So treat "Integer" type as invalid and
> ignore "Guaranteed Performance Register".
> Also save one cpc_read() call, when "Guaranteed Performance Register" is
> not present, which means register defined as:
> "Register(SystemMemory, 0, 0, 0, 0)".
>
> Fixes: 29523f095397 ("ACPI / CPPC: Add support for guaranteed performance")
> Signed-off-by: Srinivas Pandruvada <srinivas.pandruvada@linux.intel.com>
> Cc: 4.20+ <stable@vger.kernel.org> # 4.20+
> ---
>  drivers/acpi/cppc_acpi.c | 10 ++++++++--
>  1 file changed, 8 insertions(+), 2 deletions(-)
>
> diff --git a/drivers/acpi/cppc_acpi.c b/drivers/acpi/cppc_acpi.c
> index 1b207fca1420..3f6c290e06af 100644
> --- a/drivers/acpi/cppc_acpi.c
> +++ b/drivers/acpi/cppc_acpi.c
> @@ -1150,8 +1150,14 @@ int cppc_get_perf_caps(int cpunum, struct cppc_perf_caps *perf_caps)
>         cpc_read(cpunum, nominal_reg, &nom);
>         perf_caps->nominal_perf = nom;
>
> -       cpc_read(cpunum, guaranteed_reg, &guaranteed);
> -       perf_caps->guaranteed_perf = guaranteed;
> +       if (guaranteed_reg->type == ACPI_TYPE_INTEGER  ||
> +           (guaranteed_reg->type == ACPI_TYPE_BUFFER &&
> +            IS_NULL_REG(&guaranteed_reg->cpc_entry.reg))) {

AFAICS anything different from a buffer should be rejected here, so
why don't you do

if (guaranteed_reg->type != ACPI_TYPE_BUFFER  ||
IS_NULL_REG(&guaranteed_reg->cpc_entry.reg))

> +               perf_caps->guaranteed_perf = 0;
> +       } else {
> +               cpc_read(cpunum, guaranteed_reg, &guaranteed);
> +               perf_caps->guaranteed_perf = guaranteed;
> +       }
>
>         cpc_read(cpunum, lowest_non_linear_reg, &min_nonlinear);
>         perf_caps->lowest_nonlinear_perf = min_nonlinear;
> --
> 2.17.2
>

Re: [PATCH 2/2] cpufreq: intel_pstate: Also use cppc nominal_perf for base_frequency

[Rafael J. Wysocki]

On Fri, Mar 22, 2019 at 11:45 PM Srinivas Pandruvada
<srinivas.pandruvada@linux.intel.com> wrote:
>
> ACPI specifications stat that if the "Guaranteed Performance Register" is
> not implemented, OSPM assumes guaranteed performance is always equal to
> nominal performance. So for invalid and unimplemented guaranteed
> performance register, use nominal performance as guaranteed performance.
>
> This change will fallback to nominal_perf when guranteed_perf is invalid.
> If nominal_perf is also invalid, then fallback to existing implementation,
> which is to read from HWP Capabilities MSR.
>
> Fixes: 86d333a8cc7f ("cpufreq: intel_pstate: Add base_frequency attribute")
> Signed-off-by: Srinivas Pandruvada <srinivas.pandruvada@linux.intel.com>
> Cc: 4.20+ <stable@vger.kernel.org> # 4.20+
> ---
>  drivers/cpufreq/intel_pstate.c | 3 +++
>  1 file changed, 3 insertions(+)
>
> diff --git a/drivers/cpufreq/intel_pstate.c b/drivers/cpufreq/intel_pstate.c
> index 7b4b0a7ac68b..e16dea241c55 100644
> --- a/drivers/cpufreq/intel_pstate.c
> +++ b/drivers/cpufreq/intel_pstate.c
> @@ -385,6 +385,9 @@ static int intel_pstate_get_cppc_guranteed(int cpu)
>         if (ret)
>                 return ret;
>
> +       if (!cppc_perf.guaranteed_perf)
> +               return cppc_perf.nominal_perf;
> +
>         return cppc_perf.guaranteed_perf;
>  }

I would do this the other way around, that is

if (cppc_perf.guaranteed_perf)
        return cppc_perf.guaranteed_perf;

return cppc_perf.nominal_perf;

That is slightly easier to follow IMO.