On 1/16/26 6:15 PM, Numan Siddique wrote:
> On Thu, Jan 15, 2026 at 3:41 PM Eelco Chaudron <[email protected]> wrote:
>>
>>
>>
>> On 15 Jan 2026, at 20:06, [email protected] wrote:
>>
>>> From: Numan Siddique <[email protected]>
>>>
>>> The netdev-offload-tc module uses a flow's recirc_id (a uint32_t)
>>> as the TC flower chain ID during offloading.  However, the kernel's
>>> TC implementation reserves the upper 4 bits of the chain index for
>>> extended action opcodes.  If a recirc_id exceeds (2^28 - 1), the
>>> kernel rejects the flow offload with EINVAL.  This patch caps the
>>> maximum recirc_id to (2^28 - 1) to prevent these failures.
>>>
>>> This limit is sufficiently high that it is unlikely to impact the
>>> number of active datapath flows in practice.
>>>
>>> Reported-at: 
>>> https://mail.openvswitch.org/pipermail/ovs-dev/2025-November/427485.html
>>> Suggested-by: Ilya Maximets <[email protected]>
>>> Suggested-by: Eelco Chaudron <[email protected]>
>>> Signed-off-by: Numan Siddique <[email protected]>
>>> ---
>>
>> Thanks for following this through! It looks good to me.
> 
> Thanks Eelco for the review and Acks.
> I rebased and submitted v4 of both the patches with your acks even though 
> there
> were conflicts so that the CI can run with the latest main.

FWIW, 'git am -3' is smart enough to correctly apply the v3 and I was
about to push it. :)  But no worries.

Best regards, Ilya Maximets.

> 
> Thanks
> Numan
> 
>>
>> Acked-by: Eelco Chaudron <[email protected]>
>>

_______________________________________________
dev mailing list
[email protected]
https://mail.openvswitch.org/mailman/listinfo/ovs-dev

Reply via email to