Re: [PATCH 1/2] x86/resctrl: Add io_alloc_min_cbm_all interface for CBM reset

[Babu Moger <bmoger@amd.com>]

Hi Reinette,

On 11/11/25 09:29, Babu Moger wrote:
> Hi Everyone,
>
> On 11/10/25 20:44, Aaron Tomlin wrote:
>> On Tue, Nov 11, 2025 at 01:40:45AM +0000, Luck, Tony wrote:
>>>> Seems like user space has needed information to perform a sane reset?
>>> Like this:
>>>
>>> 1)     Read info files to figure out what minimum mask is needed for 
>>> this system
>>> 2)    Use Babu's suggested syntax of
>>>         # echo "L3:*={min_mask_from_step_1}" > schemata
>>>
>>> Rather than Aaron's
>>>
>>>     # echo "L3:*=0" > schemata
>> Hi Tony,
>>
>> I am satisfied with either approach.
>>
>> However, to clarify, the primary function of the "io_alloc_min_cbm_all"
>> interface is to efficiently set the lowest architecturally valid CBM 
>> across
>> all shared L3 domains. This operation relies entirely on
>> r->cache.min_cbm_bits. Since this value is guaranteed by the hardware 
>> to be
>> valid, this approach is robust and will not return an error (e.g., 
>> -EINVAL)
>> to userspace.
>>
>>
>
> Here’s my understanding of the discussion from this thread:
>
>
> We plan to support the following operation:
>
>    #echo "*=value" > /sys/fs/resctrl/info/L3/io_alloc_cbm
>
Looks like we are going with the above approach where "*" represents all 
the domain.

Couple of things to consider.

a. Send a separate patch to add this feature after [1] is merged.

b. Add this change as part of [1] series in patch 9. The series is 
mostly ready for merge. I can add the changes. Only concern is, it will 
might delay the series merge little bit.  My expectation is to have this 
series ready for next merge window.

[1] https://lore.kernel.org/lkml/cover.1761844489.git.babu.moger@amd.com/

Any thoughts?

Thanks

Babu