Re: [PATCH v5 4/5] selftests/resctrl: Use bandwidth values for AMD MBA schemata

[Babu Moger <babu.moger@amd.com>]

Hi Reinette,

On 11/6/26 17:37, Reinette Chatre wrote:
> Hi Babu,
> 
> On 11/6/26 1:25 PM, Babu Moger wrote:
>> On 11/4/26 22:16, Reinette Chatre wrote:
>>> On 11/4/26 8:05 PM, Swapnil Sapkal wrote:
>>>> On 29-10-2026 00:15, Reinette Chatre wrote:
>>>>> On 10/22/26 6:25 AM, Swapnil Sapkal wrote:
>>>>>
>>>>>> diff --git a/tools/testing/selftests/resctrl/mba_test.c b/tools/testing/selftests/resctrl/mba_test.c
>>>>>> index 17ae73d3476c..947735e26099 100644
>>>>>> --- a/tools/testing/selftests/resctrl/mba_test.c
>>>>>> +++ b/tools/testing/selftests/resctrl/mba_test.c
>>>>>> @@ -13,10 +13,51 @@
>>>>>>     #define RESULT_FILE_NAME    "result_mba"
>>>>>>     #define NUM_OF_RUNS        5
>>>>>>     #define MAX_DIFF_PERCENT    15
>>>>>> +
>>>>>> +/* Intel MBA is specified as a percentage: 10%, 20% ... 100%. */
>>>>>
>>>>> The percentage based interface is the generic interface and not Intel specific.
>>>>> You can drop "Intel" from references to this interface.
>>>>>
>>>>
>>>> Sure, will do.
>>>>
>>>>>>     #define ALLOCATION_MAX        100
>>>>>>     #define ALLOCATION_MIN        10
>>>>>>     #define ALLOCATION_STEP        10
>>>>>>     +/*
>>>>>> + * AMD MBA is not a percentage but an absolute memory bandwidth value. The
>>>>>> + * test sweeps the bandwidth from 4 GB/s to 40 GB/s in steps of 4 GB/s,
>>>>>> + * which the hardware represents in units of 1/8 GB/s (i.e. 4 GB/s -> 32,
>>>>>> + * 8 GB/s -> 64, ... 40 GB/s -> 320).
>>>>>> + */
>>>>>> +#define AMD_ALLOCATION_MAX    320
>>>>>
>>>>> The AMD maximum is enumerated from hardware. How can it be ensured this is
>>>>> the appropriate maximum for all AMD systems?
>>>>>
>>>>
>>>> As Babu mentioned, it's difficult for a single-threaded workload to
>>>> reach the hardware-reported value. For that reason, I used these
>>>> values to verify whether throttling takes effect.
>>>
>>> Just to confirm, my question was not about needing this value to be the actual
>>> maximum of a system. Instead, I just wanted to confirm that this "320" value that
>>> is hardcoded here for all AMD systems is a _valid_ value. Since the actual maximum
>>> value is enumerated from  hardware it is not obvious to me that all systems will support
>>> this hardcoded maximum.
>>>
>>
>> The value 320 corresponds to 40 GB/s. I have tested this on several systems and was able to achieve the expected 40 GB/s bandwidth. The tests did not fail.
>>
>> There are some older systems where the test is not expected to run because AMD UMC support is not available.
>>
>> If you are concerned, since we are no longer depending on the maximum value, we can be conservative and start with the following adjusted values:
>>
>> #define AMD_ALLOCATION_MIN    32
>> #define AMD_ALLOCATION_STEP    16
>>
>> This allows the value to reach 32 + 16 × 10(num_allocs) = 192, which corresponds to approximately 24 GB/s.
>>
>> Hope this answers your question.
>>
> I seem to have trouble articulating my question. Let me try to ask a different way.

My bad.
> 
> Consider the scenario where:
> 	#define AMD_ALLOCATION_MAX    320
> 
> If the below (kernel code) snippet is run on all AMD systems on which the
> test could run:
> 	u32 eax, ebx, ecx, edx;
> 	cpuid_count(0x80000020, 1, &eax, &ebx, &ecx, &edx);
> 
> Is it possible for eax to ever contain 8 or less?
> 
No. It has never been below 11 (0xB), so this should be fine.

Thanks
Babu