Alexander Lobakin wrote: > I was highly asked to send this WIP before the holidays to trigger > some discussions at least for the generic parts. > > This all depends on libie[0] and WB-on-ITR fix[1]. The RFC does not > guarantee to work perfectly, but at least regular XDP seems to work > for me... > > In fact, here are 3 separate series: > * 01-08: convert idpf to libie and make it more sane; > * 09-25: add XDP to idpf; > * 26-34: add XSk to idpf. > > Most people may want to be interested only in the following generic > changes: > * 11: allow attaching already registered memory models to XDP RxQ info; > * 12-13: generic helpers for adding a frag to &xdp_buff and converting > it to an skb; > * 14: get rid of xdp_frame::mem.id, allow mixing pages from different > page_pools within one &xdp_buff/&xdp_frame; > * 15: some Page Pool helper; > * 18: it's for libie, but I wanted to talk about XDP_TX bulking; > * 26: same as 13, but for converting XSK &xdp_buff to skb. > > The rest is up to you, driver-specific stuff is pretty boring sometimes. > > I'll be polishing and finishing this all starting January 3rd and then > preparing and sending sane series, some early feedback never hurts tho. > > Merry Yule! > > [0] > https://lore.kernel.org/netdev/20231213112835.2262651-1-aleksander.loba...@intel.com > [1] > https://lore.kernel.org/netdev/20231215193721.425087-1-michal.kub...@intel.com
This is great. Thanks for sharing the entire series. Which SHA1 should we apply this to? I'm having a hard time applying cleanly. The libie v7 series applied cleanly on bc044ae9d64b. Which I chose only based on the follow-on page pool patch. But that base commit causes too many conflicts when applying this. Patch 6 had a trivial one in idpf_rx_singleq_clean (`skb = rx_q->skb`). But patch 14 has so many conflicts in page_pool.c, that I'm clearly on the wrong track trying to fix up manually.
