On Tuesday, February 4, 2025 8:36 PM, Fijalkowski, Maciej maciej.fijalkowski@intel.com wrote:
On Tue, Feb 04, 2025 at 12:07:21PM +0100, Song, Yoong Siang wrote:
[...]
"insert an empty packet" is a launch time trick to send a packet in next Qbv cycle. The design is, the driver will still sending the packet, even the empty packet insertion trick is fail (unable to allocate). The intention of this patch set is to enable launch time on XDP zero-copy data path, so I try not to change the original behavior of launch time.
btw, do you think driver should drop the packet if something went wrong with the launch time, like launch time offload not enabled, launch time over horizon, empty packet insertion fail, etc? If yes, then maybe i can submit another patch to change the behavior of launch time and we can continue to discuss there.
That's rather a question to you since I am no TSN expert here :P the alloc skbs failures would rather be a minor thing but anyways it didn't look correct from a first glance to silently ignore this behavior if rest of the logic relies on this. I won't be insisting on any changes here but it's something you could consider to change maybe.
I got plan to refactor the launch time configuration, but that requires more discussion, so I prefer to submit another separate patch for it. I will keep the launch time configuration the original way, so that this patch set have least impact to non XDP path.
The real question is in 5/5, regarding the cleaning of these empty descs from ZC path.
Sure, I replied to your comments in 5/5. Let's continue the discussion there.
Thanks & Regards Siang