On Mon, 4 Jun 2018 14:06:00 +0200 Björn Töpel <bjorn.to...@gmail.com> wrote:
> Here, ndo_xsk_async_xmit is implemented. As a shortcut, the existing > XDP Tx rings are used for zero-copy. This will result in other devices > doing XDP_REDIRECT to an AF_XDP enabled queue will have its packets > dropped. This behavior is problematic, because XDP Tx rings are smp_processor_id based, and several RX queues can (via proc smp_affinity settings) be assigned to the same CPU. Thus, other RX-queues (than the AF_XDP enabled queue) can experience packet drops. And other devices doing redirect through i40e, which happen to run on a CPU which XDP Tx queue is "hijacked" will see dropped packets. Any plans to allocate/create a dedicated TX ring per AF_XDP socket? -- Best regards, Jesper Dangaard Brouer MSc.CS, Principal Kernel Engineer at Red Hat LinkedIn: http://www.linkedin.com/in/brouer
