Re: [PATCH bpf] xsk: Do not free ZC Rx buffer when the Rx ring is full

[Maciej Fijalkowski]

On Wed, Oct 28, 2026 at 08:40:06AM +0000, Marcin Knap wrote:
> > Marcin, FWIW, we'll discuss this topic on the office hours today [1].
> > Feel free to join! I added a link with suggested wording for a future
> > doc [2], and asked an LLM to audit the drivers [3].
> 
> > Björn
> 
> > [1] https://lore.kernel.org/netdev/aqPxxX9MeGmAqGy6@boxer/
> > [2] https://gist.github.com/fb-bjorn/d6835c49650bc3433b427a9c9708fbe2
> > [3] https://gist.github.com/fb-bjorn/181786547c4dd8182ba09e23cf394dbe
> 
> Sorry, for late answer, I couldn't join last week, I can join today if needed.
> I saw your audit which is very descriptive, that helps with understanding it.
> Thanks! As I understand you would like to fix the drivers ,which broke 
> the R2 handling considering RX Ring Full rule.
> I can send fix proposal for mlx5 driver. if so should I create another patch
> from scratch and link that conversation only?

I believe we agreed on making drivers responsible for releasing buff, so
full audit requires addressing existing -ENOBUFS-aware drivers (intel),
removing release from core and then teach rest of the drivers to handle
-ENOBUFS properly.

Basically this is what you initially proposed but with bigger driver
scope, but maybe for now stick to addressing drivers that Sashiko pointed
out (gve?).

Bjorn might be OoO today but this is what I had in my mind after the
meeting, without revisiting the links.

> 
> Regards,
> Marcin