Re: [PATCH 1/3] x86/cpufreq: Clean up powernow registration

[Jan Beulich <jbeulich@xxxxxxxx>]

On 15.11.2021 10:27, Roger Pau Monné wrote:
> On Fri, Nov 12, 2021 at 06:28:16PM +0000, Andrew Cooper wrote:
>> @@ -353,25 +349,13 @@ static const struct cpufreq_driver __initconstrel 
>> powernow_cpufreq_driver = {
>>      .update = powernow_cpufreq_update
>>  };
>>  
>> -unsigned int __init powernow_register_driver()
>> +unsigned int __init powernow_register_driver(void)
>>  {
>> -    unsigned int i, ret = 0;
>> +    if ( !(boot_cpu_data.x86_vendor & (X86_VENDOR_AMD | X86_VENDOR_HYGON)) )
>> +        return -ENODEV;
>>  
>> -    for_each_online_cpu(i) {
>> -        struct cpuinfo_x86 *c = &cpu_data[i];
>> -        if (!(c->x86_vendor & (X86_VENDOR_AMD | X86_VENDOR_HYGON)))
>> -            ret = -ENODEV;
>> -        else
>> -        {
>> -            u32 eax, ebx, ecx, edx;
>> -            cpuid(CPUID_FREQ_VOLT_CAPABILITIES, &eax, &ebx, &ecx, &edx);
>> -            if ((edx & USE_HW_PSTATE) != USE_HW_PSTATE)
>> -                ret = -ENODEV;
>> -        }
>> -        if (ret)
>> -            return ret;
>> -    }
>> +    if ( !(cpuid_edx(CPUID_FREQ_VOLT_CAPABILITIES) & USE_HW_PSTATE) )
>> +        return -ENODEV;
> 
> I wonder if we could move this check into cpufreq_driver_init and get
> rid of powernow_register_driver.

That's a vendor specific leaf, so I'd prefer to keep checking it in
vendor-specific code. Dropping the vendor check from here would seem
reasonable, considering it's already done in the caller.

Jan