[PATCH] arm_mpam: Round the CDP RMID count down to whole CLOSIDs

Ben Horgan ben.horgan at arm.com
Wed Nov 4 06:21:54 PST 2026


Hi Jinjie,

For patches touching the glue between MPAM and resctrl, essentially mpam_resctrl.c, the prefix used
is 'arm_mpam: resctrl' rather than just 'arm_mpam'.

On 04/11/2026 10:16, Jinjie Ruan wrote:
> resctrl_arch_set_cdp_enabled() halves the whole RMID index space, while
> the resctrl core halves the CLOSID count: schemata_list_add() divides a
> CDP resource's schema by two, and closid_init() takes the smallest
> across the schemata.
> 
> When the PMG count and the CLOSID count are both odd, halving the
> product counts RMIDs that belong to a CLOSID resctrl never allocates.
> With 3 PMGs and 5 CLOSIDs MPAM reports 7 RMIDs, but the 2 CLOSIDs that
> resctrl has times 3 PMGs only allow 6. The count is visible in
> /sys/fs/resctrl/info/L3_MON/num_rmids, so userspace reads a value that
> includes RMIDs no CLOSID can use; 

Good spot.

It occurs to me that 'num_rmids' is also too high on a monitor only system too as there is only one
PARTID that can be used.  Can you see a clean way to fix this in one go?

Thanks,

Ben

RMID allocation itself is bounded by
> the free list.
> 
> Derive the number of RMID groups from the index space and halve that, so
> the count follows resctrl_arch_system_num_rmid_idx().
> 
> Fixes: 264c285999fc ("arm_mpam: resctrl: Add monitor initialisation and domain boilerplate")
> Signed-off-by: Jinjie Ruan <ruanjinjie at huawei.com>
> Cc: <stable at vger.kernel.org> # 7.1.x
> Cc: James Morse <james.morse at arm.com>
> Cc: Ben Horgan <ben.horgan at arm.com>
> Cc: Reinette Chatre <reinette.chatre at intel.com>
> Cc: Fenghua Yu <fenghuay at nvidia.com>
> ---
>  drivers/resctrl/mpam_resctrl.c | 12 +++++++++---
>  1 file changed, 9 insertions(+), 3 deletions(-)
> 
> diff --git a/drivers/resctrl/mpam_resctrl.c b/drivers/resctrl/mpam_resctrl.c
> index 5468367776e2..d3f957162b68 100644
> --- a/drivers/resctrl/mpam_resctrl.c
> +++ b/drivers/resctrl/mpam_resctrl.c
> @@ -217,10 +217,16 @@ int resctrl_arch_set_cdp_enabled(enum resctrl_res_level rid, bool enable)
>  	cdp_enabled = enable;
>  	mpam_resctrl_controls[rid].cdp_enabled = enable;
>  
> -	if (enable)
> -		l3->mon.num_rmid = resctrl_arch_system_num_rmid_idx() / 2;
> -	else
> +	if (enable) {
> +		u32 num_group;
> +
> +		/* Halve the number of RMID groups, keeping whole PMG rows. */
> +		num_group = resctrl_arch_system_num_rmid_idx() /
> +			    (mpam_pmg_max + 1);
> +		l3->mon.num_rmid = (mpam_pmg_max + 1) * (num_group / 2);
> +	} else {
>  		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)




More information about the linux-arm-kernel mailing list