Re: [PATCH v2 04/11] drm/msm: adreno: add GMU_BW_VOTE feature flag

[neil.armstrong@linaro.org]

On 27/11/2024 17:00, Akhil P Oommen wrote:
> On 11/25/2024 1:46 PM, Neil Armstrong wrote:
>> On 23/11/2024 20:43, Akhil P Oommen wrote:
>>> On Tue, Nov 19, 2024 at 06:56:39PM +0100, Neil Armstrong wrote:
>>>> The Adreno GMU Management Unit (GNU) can also scale the DDR Bandwidth
>>>> along the Frequency and Power Domain level, but by default we leave the
>>>> OPP core vote for the interconnect ddr path.
>>>>
>>>> While scaling via the interconnect path was sufficient, newer GPUs
>>>> like the A750 requires specific vote paremeters and bandwidth to
>>>> achieve full functionality.
>>>>
>>>> While the feature will require some data in a6xx_info, it's safer
>>>> to only enable tested platforms with this flag first.
>>>>
>>>> Add a new feature enabling DDR Bandwidth vote via GMU.
>>>>
>>>> Signed-off-by: Neil Armstrong <neil.armstrong@linaro.org>
>>>> ---
>>>>    drivers/gpu/drm/msm/adreno/adreno_gpu.h | 1 +
>>>>    1 file changed, 1 insertion(+)
>>>>
>>>> diff --git a/drivers/gpu/drm/msm/adreno/adreno_gpu.h b/drivers/gpu/
>>>> drm/msm/adreno/adreno_gpu.h
>>>> index
>>>> 4702d4cfca3b58fb3cbb25cb6805f1c19be2ebcb..394b96eb6c83354ae008b15b562bedb96cd391dd 100644
>>>> --- a/drivers/gpu/drm/msm/adreno/adreno_gpu.h
>>>> +++ b/drivers/gpu/drm/msm/adreno/adreno_gpu.h
>>>> @@ -58,6 +58,7 @@ enum adreno_family {
>>>>    #define ADRENO_FEAT_HAS_HW_APRIV        BIT(0)
>>>>    #define ADRENO_FEAT_HAS_CACHED_COHERENT        BIT(1)
>>>>    #define ADRENO_FEAT_PREEMPTION            BIT(2)
>>>> +#define ADRENO_FEAT_GMU_BW_VOTE            BIT(3)
>>>
>>> Do we really need a feature flag for this? We have to carry this for
>>> every
>>> GPU going forward. IB voting is supported on all GMUs from A6xx GEN2 and
>>> newer. So we can just check that along with whether the bw table is
>>> dynamically generated or not.
>>
>> It depends on the bw table _and_ the a6xx_info.gmu table, I don't want to
>> check both in all parts on the driver.
>>
> Thats fine then.

Finally I converted the a6xx_info to a pointer, and it's fine checking
this pointer instead of the quirk, since anyway we already check num_bws.

Neil

> 
> -Akhil.
> 
>> Neil
>>
>>>
>>> -Akhil
>>>
>>>>      /* Helper for formating the chip_id in the way that userspace
>>>> tools like
>>>>     * crashdec expect.
>>>>
>>>> -- 
>>>> 2.34.1
>>>>
>>
>>
>