Re: [PATCH v2] x86, sched: Fix the AMD CPPC maximum perf on some specific generations

[Rafael J. Wysocki]

On Wed, Apr 21, 2021 at 4:38 AM Huang Rui <ray.huang@xxxxxxx> wrote:
>
> Some AMD Ryzen generations has different calculation method on maximum
> perf. 255 is not for all asics, some specific generations should use 166
> as the maximum perf. Otherwise, it will report incorrect frequency value
> like below:
>
> ~ → lscpu | grep MHz
> CPU MHz:                         3400.000
> CPU max MHz:                     7228.3198
> CPU min MHz:                     2200.0000
>
> Fixes: 41ea667227ba ("x86, sched: Calculate frequency invariance for AMD systems")
> Fixes: 3c55e94c0ade ("cpufreq: ACPI: Extend frequency tables to cover boost frequencies")
>
> Reported-by: Jason Bagavatsingham <jason.bagavatsingham@xxxxxxxxx>
> Tested-by: Jason Bagavatsingham <jason.bagavatsingham@xxxxxxxxx>
> Bugzilla: https://bugzilla.kernel.org/show_bug.cgi?id=211791
> Signed-off-by: Huang Rui <ray.huang@xxxxxxx>
> Cc: Alex Deucher <alexander.deucher@xxxxxxx>
> Cc: Nathan Fontenot <nathan.fontenot@xxxxxxx>
> Cc: Rafael J. Wysocki <rafael.j.wysocki@xxxxxxxxx>
> Cc: Borislav Petkov <bp@xxxxxxx>
> Cc: x86@xxxxxxxxxx
> Cc: stable@xxxxxxxxxxxxxxx
> ---
>
> Changes from V1 -> V2:
> - Enhance the commit message.
> - Move amd_get_highest_perf() into amd.c.
> - Refine the implementation of switch-case.
> - Cc stable mail list.
>
> ---
>  arch/x86/include/asm/processor.h |  2 ++
>  arch/x86/kernel/cpu/amd.c        | 22 ++++++++++++++++++++++
>  arch/x86/kernel/smpboot.c        |  2 +-
>  drivers/cpufreq/acpi-cpufreq.c   | 19 +++++++++++++++++++
>  4 files changed, 44 insertions(+), 1 deletion(-)
>
> diff --git a/arch/x86/include/asm/processor.h b/arch/x86/include/asm/processor.h
> index f1b9ed5efaa9..908bcaea1361 100644
> --- a/arch/x86/include/asm/processor.h
> +++ b/arch/x86/include/asm/processor.h
> @@ -804,8 +804,10 @@ DECLARE_PER_CPU(u64, msr_misc_features_shadow);
>
>  #ifdef CONFIG_CPU_SUP_AMD
>  extern u32 amd_get_nodes_per_socket(void);
> +extern u32 amd_get_highest_perf(void);
>  #else
>  static inline u32 amd_get_nodes_per_socket(void)       { return 0; }
> +static inline u32 amd_get_highest_perf(void)           { return 0; }
>  #endif
>
>  static inline uint32_t hypervisor_cpuid_base(const char *sig, uint32_t leaves)
> diff --git a/arch/x86/kernel/cpu/amd.c b/arch/x86/kernel/cpu/amd.c
> index 347a956f71ca..aadb691d9357 100644
> --- a/arch/x86/kernel/cpu/amd.c
> +++ b/arch/x86/kernel/cpu/amd.c
> @@ -1170,3 +1170,25 @@ void set_dr_addr_mask(unsigned long mask, int dr)
>                 break;
>         }
>  }
> +
> +u32 amd_get_highest_perf(void)
> +{
> +       struct cpuinfo_x86 *c = &boot_cpu_data;
> +       u32 cppc_max_perf = 225;

The extra local variable is redundant.

> +
> +       switch (c->x86) {
> +       case 0x17:
> +               if ((c->x86_model >= 0x30 && c->x86_model < 0x40) ||
> +                   (c->x86_model >= 0x70 && c->x86_model < 0x80))
> +                       cppc_max_perf = 166;
> +               break;

Also it would be cleaner to write this as

if (c->x86 == 0x17 && ((c->x86_model >= 0x30 && c->x86_model < 0x40) ||
    (c->x86_model >= 0x70 && c->x86_model < 0x80))
        return 166;

And analogously below.

> +       case 0x19:
> +               if ((c->x86_model >= 0x20 && c->x86_model < 0x30) ||
> +                   (c->x86_model >= 0x40 && c->x86_model < 0x70))
> +                       cppc_max_perf = 166;
> +               break;
> +       }
> +
> +       return cppc_max_perf;

And here

return 225;

> +}
> +EXPORT_SYMBOL_GPL(amd_get_highest_perf);
> diff --git a/arch/x86/kernel/smpboot.c b/arch/x86/kernel/smpboot.c
> index 02813a7f3a7c..7bec57d04a87 100644
> --- a/arch/x86/kernel/smpboot.c
> +++ b/arch/x86/kernel/smpboot.c
> @@ -2046,7 +2046,7 @@ static bool amd_set_max_freq_ratio(void)
>                 return false;
>         }
>
> -       highest_perf = perf_caps.highest_perf;
> +       highest_perf = amd_get_highest_perf();
>         nominal_perf = perf_caps.nominal_perf;
>
>         if (!highest_perf || !nominal_perf) {
> diff --git a/drivers/cpufreq/acpi-cpufreq.c b/drivers/cpufreq/acpi-cpufreq.c
> index d1bbc16fba4b..3f0a19dd658c 100644
> --- a/drivers/cpufreq/acpi-cpufreq.c
> +++ b/drivers/cpufreq/acpi-cpufreq.c
> @@ -630,6 +630,22 @@ static int acpi_cpufreq_blacklist(struct cpuinfo_x86 *c)
>  #endif
>
>  #ifdef CONFIG_ACPI_CPPC_LIB
> +
> +static u64 get_amd_max_boost_ratio(unsigned int cpu, u64 nominal_perf)
> +{
> +       u64 boost_ratio, cppc_max_perf;
> +
> +       if (!nominal_perf)
> +               return 0;
> +
> +       cppc_max_perf = amd_get_highest_perf();
> +
> +       boost_ratio = div_u64(cppc_max_perf << SCHED_CAPACITY_SHIFT,
> +                             nominal_perf);
> +
> +       return boost_ratio;
> +}

The function above is not necessary if I'm not mistaken.

> +
>  static u64 get_max_boost_ratio(unsigned int cpu)
>  {
>         struct cppc_perf_caps perf_caps;
> @@ -646,6 +662,9 @@ static u64 get_max_boost_ratio(unsigned int cpu)
>                 return 0;
>         }
>
> +       if (boot_cpu_data.x86_vendor == X86_VENDOR_AMD)
> +               return get_amd_max_boost_ratio(cpu, perf_caps.nominal_perf);
> +
>         highest_perf = perf_caps.highest_perf;

The above can be written as

if (boot_cpu_data.x86_vendor == X86_VENDOR_AMD)
        highest_perf = amd_get_highest_perf();
else
        highest_perf = perf_caps.highest_perf;

>         nominal_perf = perf_caps.nominal_perf;
>
> --