[PATCH] arm_mpam: Warn when firmware leaves PARTIDs disabled

[Ben Horgan]

Hi Ping,

On 01/11/2026 02:55, Ping Li wrote:
> Hi Ben,
> 
> Thanks for sharing the direction. I asked our hardware team why they
> implemented HAS_ENDIS: the motivation is power. Enabling PARTIDs that
> are never used does not affect performance, but costs leakage in the
> registers and combinational logic behind them - with e.g. 32 PARTIDs
> implemented and only a handful ever used, the unused portion is pure
> waste. Their expectation was that software would enable only the
> PARTIDs it needs at initialization time.

Thanks for this information. It may also be useful for getting data associated out of the cache
quickly once the associated PARTID is unused.

> 
> For a general-purpose driver that cannot predict how many PARTIDs
> userspace will actually use, that is not something we can do. So what
> I plan to do is either enable all the PARTIDs from the driver, or ask

Enabling all the PARTIDs in the driver is a good first step to support hardware with ENDIS. However,
going forward I think we could actually make use of the feature by enabling PARTIDs when a control
group is created in resctrl and disabling them when a control group is deleted.

Thanks,

Ben

> the hardware team to drop the HAS_ENDIS feature.
> 
> I am happy to drop this warning patch in favour of your driver-side
> enabling patch.
> 
> Thanks,
> Ping