On Tue, 14 Feb 2017 11:02:01 -0800 Eric Dumazet <eduma...@google.com> wrote:
> On Tue, Feb 14, 2017 at 10:46 AM, Jesper Dangaard Brouer > <bro...@redhat.com> wrote: > > > > > > > With this Intel driver page count based recycle approach, the recycle > > size is tied to the size of the RX ring. As Eric and Tariq discovered. > > And for other performance reasons (memory footprint of walking RX ring > > data-structures), don't want to increase the RX ring sizes. Thus, it > > create two opposite performance needs. That is why I think a more > > explicit approach with a pool is more attractive. > > > > How is this approach doing to work for XDP? > > (XDP doesn't "share" the page, and in-general we don't want the extra > > atomic.) > > > > We absolutely need recycling with XDP, when transmitting out another > > device, and the other devices DMA-TX completion need some way of > > returning this page. > > What is basically needed is a standardized callback to allow the remote > > driver to return the page to the originating driver. As we don't have > > a NDP for XDP-forward/transmit yet, we could pass this callback as a > > parameter along with the packet-page to send? > > > > > > > mlx4 already has a cache for XDP. > I believe I did not change this part, it still should work. > > commit d576acf0a22890cf3f8f7a9b035f1558077f6770 > Author: Brenden Blanco <bbla...@plumgrid.com> > Date: Tue Jul 19 12:16:52 2016 -0700 > > net/mlx4_en: add page recycle to prepare rx ring for tx support This obviously does not work for the case I'm talking about (transmitting out another device with XDP). -- Best regards, Jesper Dangaard Brouer MSc.CS, Principal Kernel Engineer at Red Hat LinkedIn: http://www.linkedin.com/in/brouer
