1. Patchew
  2. linux
  3. [PATCH v6 00/40] arm_mpam: Add KVM/arm64 and resctrl glue code
  4. View patch

[PATCH v6 21/40] arm_mpam: resctrl: Hide CDP emulation behind CONFIG_EXPERT

Ben Horgan posted 40 patches 2 weeks, 5 days ago
[PATCH v6 21/40] arm_mpam: resctrl: Hide CDP emulation behind CONFIG_EXPERT
Posted by Ben Horgan 2 weeks, 5 days ago 
When CDP is not enabled, the 'rmid_entry's in the limbo list,
rmid_busy_llc, map directly to a (PARTID,PMG) pair and when CDP is enabled
the mapping is to two different pairs. As the limbo list is reused between
mounts and CDP disabled on unmount this can lead to stale mapping and the
limbo handler will then make monitor reads with potentially out of range
PARTID. This may then cause an MPAM error interrupt and the driver will
disable MPAM.

No problems are expected if you just mount the resctrl file system
once with CDP enabled and never unmount it. Hide CDP emulation behind
CONFIG_EXPERT to protect the unwary.

Signed-off-by: Ben Horgan <ben.horgan@arm.com>
---
Adding this ugliness in the hope of avoiding patch churn and extra
reviewer work. I am looking into the resctrl changes needed to fix this.
---
 drivers/resctrl/mpam_resctrl.c | 12 ++++++++++++
 1 file changed, 12 insertions(+)

diff --git a/drivers/resctrl/mpam_resctrl.c b/drivers/resctrl/mpam_resctrl.c
index 903d1a0f564f..cab3e9ccb5c7 100644
--- a/drivers/resctrl/mpam_resctrl.c
+++ b/drivers/resctrl/mpam_resctrl.c
@@ -82,6 +82,18 @@ int resctrl_arch_set_cdp_enabled(enum resctrl_res_level rid, bool enable)
 	u32 partid_i = RESCTRL_RESERVED_CLOSID, partid_d = RESCTRL_RESERVED_CLOSID;
 	int cpu;
 
+	if (!IS_ENABLED(CONFIG_EXPERT) && enable) {
+		/*
+		 * If the resctrl fs is mounted more than once, sequentially,
+		 * then CDP can lead to the use of out of range PARTIDs.
+		 */
+		pr_warn("CDP not supported\n");
+		return -EOPNOTSUPP;
+	}
+
+	if (enable)
+		pr_warn("CDP is an expert feature and may cause MPAM to malfunction.\n");
+
 	/*
 	 * resctrl_arch_set_cdp_enabled() is only called with enable set to
 	 * false on error and unmount.
-- 
2.43.0
Re: [PATCH v6 21/40] arm_mpam: resctrl: Hide CDP emulation behind CONFIG_EXPERT
Posted by James Morse 5 days, 14 hours ago 
Hi Ben,

On 13/03/2026 14:45, Ben Horgan wrote:
> When CDP is not enabled, the 'rmid_entry's in the limbo list,
> rmid_busy_llc, map directly to a (PARTID,PMG) pair and when CDP is enabled
> the mapping is to two different pairs.

> As the limbo list is reused between
> mounts and CDP disabled on unmount this can lead to stale mapping and the
> limbo handler will then make monitor reads with potentially out of range
> PARTID.

Bother - I missed that!


> This may then cause an MPAM error interrupt and the driver will
> disable MPAM.

... and that's why it's not a problem on x86 because the RMID range is unaffected by CDP,
whereas MPAM works on a combined value.


> No problems are expected if you just mount the resctrl file system
> once with CDP enabled and never unmount it.

(guess how it was tested!)


> Hide CDP emulation behind CONFIG_EXPERT to protect the unwary.
> 
> Signed-off-by: Ben Horgan <ben.horgan@arm.com>
> ---
> Adding this ugliness in the hope of avoiding patch churn and extra
> reviewer work. I am looking into the resctrl changes needed to fix this.

Makes sense - people can still use this if they're aware of the limitation, and it sounds
like you've got a plan to fix it properly. We just don't want it enabled in distros until
then.


> diff --git a/drivers/resctrl/mpam_resctrl.c b/drivers/resctrl/mpam_resctrl.c
> index 903d1a0f564f..cab3e9ccb5c7 100644
> --- a/drivers/resctrl/mpam_resctrl.c
> +++ b/drivers/resctrl/mpam_resctrl.c
> @@ -82,6 +82,18 @@ int resctrl_arch_set_cdp_enabled(enum resctrl_res_level rid, bool enable)
>  	u32 partid_i = RESCTRL_RESERVED_CLOSID, partid_d = RESCTRL_RESERVED_CLOSID;
>  	int cpu;
>  
> +	if (!IS_ENABLED(CONFIG_EXPERT) && enable) {
> +		/*
> +		 * If the resctrl fs is mounted more than once, sequentially,
> +		 * then CDP can lead to the use of out of range PARTIDs.
> +		 */
> +		pr_warn("CDP not supported\n");
> +		return -EOPNOTSUPP;
> +	}
> +
> +	if (enable)
> +		pr_warn("CDP is an expert feature and may cause MPAM to malfunction.\n");
> +
>  	/*
>  	 * resctrl_arch_set_cdp_enabled() is only called with enable set to
>  	 * false on error and unmount.

Reviewed-by: James Morse <james.morse@arm.com>


Thanks,

James
Re: [PATCH v6 21/40] arm_mpam: resctrl: Hide CDP emulation behind CONFIG_EXPERT
Posted by Gavin Shan 1 week, 2 days ago 
On 3/14/26 12:45 AM, Ben Horgan wrote:
> When CDP is not enabled, the 'rmid_entry's in the limbo list,
> rmid_busy_llc, map directly to a (PARTID,PMG) pair and when CDP is enabled
> the mapping is to two different pairs. As the limbo list is reused between
> mounts and CDP disabled on unmount this can lead to stale mapping and the
> limbo handler will then make monitor reads with potentially out of range
> PARTID. This may then cause an MPAM error interrupt and the driver will
> disable MPAM.
> 
> No problems are expected if you just mount the resctrl file system
> once with CDP enabled and never unmount it. Hide CDP emulation behind
> CONFIG_EXPERT to protect the unwary.
> 
> Signed-off-by: Ben Horgan <ben.horgan@arm.com>
> ---
> Adding this ugliness in the hope of avoiding patch churn and extra
> reviewer work. I am looking into the resctrl changes needed to fix this.
> ---
>   drivers/resctrl/mpam_resctrl.c | 12 ++++++++++++
>   1 file changed, 12 insertions(+)
> 
Reviewed-by: Gavin Shan <gshan@redhat.com>
Re: [PATCH v6 21/40] arm_mpam: resctrl: Hide CDP emulation behind CONFIG_EXPERT
Posted by Zeng Heng 2 weeks ago 
Hi Ben,

On 2026/3/13 22:45, Ben Horgan wrote:
> When CDP is not enabled, the 'rmid_entry's in the limbo list,
> rmid_busy_llc, map directly to a (PARTID,PMG) pair and when CDP is enabled
> the mapping is to two different pairs. As the limbo list is reused between
> mounts and CDP disabled on unmount this can lead to stale mapping and the
> limbo handler will then make monitor reads with potentially out of range
> PARTID. This may then cause an MPAM error interrupt and the driver will
> disable MPAM.
> 
> No problems are expected if you just mount the resctrl file system
> once with CDP enabled and never unmount it. Hide CDP emulation behind
> CONFIG_EXPERT to protect the unwary.
> 
> Signed-off-by: Ben Horgan <ben.horgan@arm.com>
> ---
> Adding this ugliness in the hope of avoiding patch churn and extra
> reviewer work. I am looking into the resctrl changes needed to fix this.
> ---
>   drivers/resctrl/mpam_resctrl.c | 12 ++++++++++++
>   1 file changed, 12 insertions(+)
> 
> diff --git a/drivers/resctrl/mpam_resctrl.c b/drivers/resctrl/mpam_resctrl.c
> index 903d1a0f564f..cab3e9ccb5c7 100644
> --- a/drivers/resctrl/mpam_resctrl.c
> +++ b/drivers/resctrl/mpam_resctrl.c
> @@ -82,6 +82,18 @@ int resctrl_arch_set_cdp_enabled(enum resctrl_res_level rid, bool enable)
>   	u32 partid_i = RESCTRL_RESERVED_CLOSID, partid_d = RESCTRL_RESERVED_CLOSID;
>   	int cpu;
>   
> +	if (!IS_ENABLED(CONFIG_EXPERT) && enable) {
> +		/*
> +		 * If the resctrl fs is mounted more than once, sequentially,
> +		 * then CDP can lead to the use of out of range PARTIDs.
> +		 */
> +		pr_warn("CDP not supported\n");
> +		return -EOPNOTSUPP;
> +	}
> +
> +	if (enable)
> +		pr_warn("CDP is an expert feature and may cause MPAM to malfunction.\n");
> +

Agree that a proper fix will require modifications to resctrl to handle 
the limbo state across mount sessions. Hiding CDP behind CONFIG_EXPERT 
is acceptable as a short-term mitigation to prevent users from 
accidentally hitting the mentioned issue.

Reviewed-by: Zeng Heng <zengheng4@huawei.com>

Patchew 2.1-devel-ea86937

© 2016 - 2026 Red Hat, Inc.