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

[Huang Rui]

On Wed, Apr 21, 2021 at 09:42:15PM +0800, Fontenot, Nathan wrote:
> On 4/20/2021 9:38 PM, Huang Rui 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;
> > +
> > +	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;
> > +	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;
> > +}
> > +EXPORT_SYMBOL_GPL(amd_get_highest_perf);
> 
> Should this be an update to cpp_get_perf_caps()?
> 
> This approach would ensure that all callers have the correct value
> and remove the need to fix up individual callers to use this new
> routine to get the correct value.
> 

It's a good idea to modify cppc_get_perf_caps() to correct the right
highest perf. I would like to keep amd_get_highest_perf() function in amd.c
as well because it can be called in other spaces without querying an ACPI
register in case only wants to get highest perf value.

---

diff --git a/drivers/acpi/cppc_acpi.c b/drivers/acpi/cppc_acpi.c
index 69057fcd2c04..58e72b6e222f 100644
--- a/drivers/acpi/cppc_acpi.c
+++ b/drivers/acpi/cppc_acpi.c
@@ -1107,8 +1107,12 @@ int cppc_get_perf_caps(int cpunum, struct cppc_perf_caps *perf_caps)
                }
        }

-       cpc_read(cpunum, highest_reg, &high);
-       perf_caps->highest_perf = high;
+       if (boot_cpu_data.x86_vendor == X86_VENDOR_AMD) {
+               perf_caps->highest_perf = amd_get_highest_perf();
+       } else {
+               cpc_read(cpunum, highest_reg, &high);
+               perf_caps->highest_perf = high;
+       }

        cpc_read(cpunum, lowest_reg, &low);
        perf_caps->lowest_perf = low;

---

Thanks,
Ray