Re: [PATCH net v3 1/1] i40e: xsk: advance next_to_clean on status descriptors

From: Maciej Fijalkowski

Date: Mon Nov 17 2025 - 11:38:25 EST


On Sat, Nov 15, 2025 at 06:58:41PM +1100, Alessandro Decina wrote:
> On Fri, Nov 14, 2025 at 02:01:14PM +0100, Maciej Fijalkowski wrote:
> > Woah, that's not what I had on mind...I meant to pull whole block that
> > takes care of FDIR descriptors onto common function. That logic should be
> > shared between normal Rx and ZC Rx. The only different action we need to
> > take is how we release the buffer.
> >
> > Could you try pulling whole i40e_rx_is_programming_status() branch onto
> > function within i40e_txrx_common.h and see how much of a work would it
> > take to have this as a common function?
>
> Just before I send another rev, you mean something like this?
> https://github.com/alessandrod/linux/commit/a6fa91d5b5d1cc283a2f1faa378085c44bda8b4a
>
> My rationale for i40e_inc_ntp_ntc was that _that_ is where the bug lies:
> letting ntp and ntc get out of sync. By introducing a function that
> forces you to _have_ to think about ntc and explicitly pass NULL if you
> don't want to sync it, bugs like this become less easy to introduce.
>
> That said I don't mind either way! Let me know if you want me to send v4
> with the i40e_clean_programming_status() change.

This revision is much more clear to me. Only thing that might be bothering
someone is doubled i40e_rx_bi() call in i40e_get_rx_buffer(). Not sure if
we can do about it though as we need to use ntp from before potential
increment.

...maybe pass rx_buffer to i40e_get_rx_buffer() ?

>
> Ciao,
> Alessandro