Re: [PATCH v9 net-next 6/6] octeontx2-af: npc: Support for custom KPU profile from filesystem

[Paolo Abeni <pabeni@redhat.com>]

On 3/30/26 7:31 AM, Ratheesh Kannoth wrote:
> @@ -1635,9 +1651,9 @@ int npc_cn20k_apply_custom_kpu(struct rvu *rvu,
>  	}
>  
>  	/* Verify if profile fits the HW */
> -	if (fw->kpus > profile->kpus) {
> -		dev_warn(rvu->dev, "Not enough KPUs: %d > %ld\n", fw->kpus,
> -			 profile->kpus);
> +	if (fw->kpus > rvu->hw->npc_kpus) {
> +		dev_warn(rvu->dev, "Not enough KPUs: %d > %d\n", fw->kpus,
> +			 rvu->hw->npc_kpus);
>  		return -EINVAL;
>  	}

AI review says:

---
The bounds check was changed from profile->kpus (which equals
ARRAY_SIZE(npc_kpu_profiles) = 16) to rvu->hw->npc_kpus (a 5-bit
hardware field, max value 31). Can this cause an out-of-bounds write?

In the subsequent loop at lines 1658-1685 (visible earlier in the
function), the code writes to profile->kpu[kpu].cam[entry] and
profile->kpu[kpu].action[entry], where profile->kpu points to the static
global npc_kpu_profiles[] array that has exactly 16 elements.

If cn20k hardware reports npc_kpus > 16 and the firmware provides
fw->kpus > 16, the new check passes but the loop iterates beyond the
array bounds, corrupting adjacent memory.
---

Also there are several remarks for sashiko, some of them looks valid:

https://sashiko.dev/#/patchset/20260330053105.2722453-1-rkannoth%40marvell.com

please a look.

Thanks,

Paolo