Re: [PATCH next] arm_mpam: Fix MBA alloc_capable handling

[James Morse]

Hi Zeng, Ben,

On 17/04/2026 10:52, Ben Horgan wrote:
> On 4/13/26 10:00, Zeng Heng wrote:
>> The resctrl_arch_set_cdp_enabled() is never invoked with RDT_RESOURCE_MBA
>> as the rid parameter. Consequently,
>> mpam_resctrl_controls[RDT_RESOURCE_MBA].cdp_enabled always remains false,
>> making it unreliable for conditional checks.
>>
>> The original logic uses this field to determine MBA state, causing a bug
>> where MBA allocation is permanently disabled after the mount with CDP
>> option. Remounting without CDP cannot restore the MBA partition capability.
> 
> Ah, I see. Thanks for spotting this and sending a patch.

This shape of bug is likely - I've not had access to a platform with bandwidth
monitors!


>> Fix by using the cdp_enabled global parameter directly. Enable MBA
>> partition only when CDP is disabled and the MBA resource class is
>> present. Disable MBA partition otherwise.

>> diff --git a/drivers/resctrl/mpam_resctrl.c b/drivers/resctrl/mpam_resctrl.c
>> index a9938006d0e6..161fb8905e28 100644
>> --- a/drivers/resctrl/mpam_resctrl.c
>> +++ b/drivers/resctrl/mpam_resctrl.c
>> @@ -217,12 +217,10 @@ int resctrl_arch_set_cdp_enabled(enum resctrl_res_level rid, bool enable)
>>  		l3->mon.num_rmid = resctrl_arch_system_num_rmid_idx();
>>  
>>  	/* The mbw_max feature can't hide cdp as it's a per-partid maximum. */
>> -	if (cdp_enabled && !mpam_resctrl_controls[RDT_RESOURCE_MBA].cdp_enabled)
>> -		mpam_resctrl_controls[RDT_RESOURCE_MBA].resctrl_res.alloc_capable = false;
>> -
>> -	if (mpam_resctrl_controls[RDT_RESOURCE_MBA].cdp_enabled &&
>> -	    mpam_resctrl_controls[RDT_RESOURCE_MBA].class)
>> +	if (!cdp_enabled && mpam_resctrl_controls[RDT_RESOURCE_MBA].class)
>>  		mpam_resctrl_controls[RDT_RESOURCE_MBA].resctrl_res.alloc_capable = true;
>> +	else
>> +		mpam_resctrl_controls[RDT_RESOURCE_MBA].resctrl_res.alloc_capable = false;

My reading of the original:
|	if (mpam_resctrl_controls[RDT_RESOURCE_MBA].cdp_enabled &&
|	    mpam_resctrl_controls[RDT_RESOURCE_MBA].class)
|		mpam_resctrl_controls[RDT_RESOURCE_MBA].resctrl_res.alloc_capable = true;

is that it was trying to do this reset - but never running. (as Zeng's commit message
describes).

I've spoken to Ben - and the intent here is a bit more complicated...

if resctrl does start calling:
| resctrl_arch_set_cdp_enabled(RDT_RESOURCE_MBA, true)

then we expect to receive two configurations, one for code and one for data, so the
aliasing controls problem goes away. The above original code is re-enabling MBA in this
case. It needs to ignore the global 'cdp_enabled' to be immune to the order of
resctrl_arch_set_cdp_enabled() calls.

While this might seem odd, as resctrl doesn't do this today - it is a valid enum value,
and this is in the woolly "are we Xeon shaped" area.

I think long term we need a list of non-aliasing controls (min and max etc) which need
disabling when CDP enabled. But re-enabling them when called explicitly makes sense as
we have to handle those enum value anyway.


> Isn't it more that the existing checks are correct (if not totally necessary) and guard against
> resctrl_arch_set_cdp_enabled() being called with RDT_RESOURCE_MBA but that re-enable on unmount is missing?
> Does just adding this check make sense? As already commented in the function we assume that
> resctrl_arch_set_cdp_enabled() is only called with enable=false on error or unmount.
> 
> --- a/drivers/resctrl/mpam_resctrl.c
> +++ b/drivers/resctrl/mpam_resctrl.c
> @@ -224,6 +224,9 @@ int resctrl_arch_set_cdp_enabled(enum resctrl_res_level rid, bool enable)
>             mpam_resctrl_controls[RDT_RESOURCE_MBA].class)
>                 mpam_resctrl_controls[RDT_RESOURCE_MBA].resctrl_res.alloc_capable = true;
> 
> +       if (!cdp_enabled && mpam_resctrl_controls[RDT_RESOURCE_MBA].class)
> +               mpam_resctrl_controls[RDT_RESOURCE_MBA].resctrl_res.alloc_capable = true;
> +
>         if (enable) {
>                 if (mpam_partid_max < 1)
>                         return -EINVAL;


I'll add a comment above the original code quoted above describing what it does, and
take your hunk here, (with a comment that its for umount):
-------------%<-------------
arm_mpam: resctrl: Fix MBA CDP alloc_capable handling on unmount

The code to set MBA's alloc_capable to true appears to be trying to
restore alloc_capable on unmount. This can never work because
resctrl_arch_set_cdp_enabled() is never invoked with RDT_RESOURCE_MBA
as the rid parameter. Consequently,
mpam_resctrl_controls[RDT_RESOURCE_MBA].cdp_enabled always remains false.

The alloc_capable setting in resctrl_arch_set_cdp_enabled() is to
re-enable MBA if the caller opts in to separate control values using
CDP for this resource. This doesn't happen today.

Add a comment to describe this.

However a bug remains where MBA allocation is permanently disabled after
the mount with CDP option. Remounting without CDP cannot restore the MBA
partition capability.

Add a check to re-enable MBA when CDP is disabled, which happens on
unmount.
---
@@ -220,10 +220,18 @@ int resctrl_arch_set_cdp_enabled(enum resctrl_res_level rid, bool
enable)
        if (cdp_enabled && !mpam_resctrl_controls[RDT_RESOURCE_MBA].cdp_enabled)
                mpam_resctrl_controls[RDT_RESOURCE_MBA].resctrl_res.alloc_capable = false;

+       /*
+        * if resctrl has attempted to enable CDP on MBA, re-enable MBA as two
+        * configurations will provided so there is no aliasing problem.
+        */
        if (mpam_resctrl_controls[RDT_RESOURCE_MBA].cdp_enabled &&
            mpam_resctrl_controls[RDT_RESOURCE_MBA].class)
                mpam_resctrl_controls[RDT_RESOURCE_MBA].resctrl_res.alloc_capable = true;

+       /* On unmount when CDP is disabled, re-enable MBA */
+       if (!cdp_enabled && mpam_resctrl_controls[RDT_RESOURCE_MBA].class)
+               mpam_resctrl_controls[RDT_RESOURCE_MBA].resctrl_res.alloc_capable = true;
+
        if (enable) {
                if (mpam_partid_max < 1)
                        return -EINVAL;
-------------%<-------------

Ben - do you want a Co-developed by for this?

While I'm at it:
Reviewed-by: James Morse <james.morse@xxxxxxx>


Thanks,

James