On Sun, Apr 02, 2023 at 09:38:38PM +0200, Gerhard Engleder wrote:
> Send and complete XSK pool frames within TX NAPI context. NAPI context
> is triggered by ndo_xsk_wakeup.
>
> Signed-off-by: Gerhard Engleder <gerhard@xxxxxxxxxxxxxxxxxxxxx>
> ---
> drivers/net/ethernet/engleder/tsnep.h | 2 +
> drivers/net/ethernet/engleder/tsnep_main.c | 131 +++++++++++++++++++--
> 2 files changed, 123 insertions(+), 10 deletions(-)
>
(...)
> +static void tsnep_xdp_xmit_zc(struct tsnep_tx *tx)
> +{
> + int desc_available = tsnep_tx_desc_available(tx);
> + struct xdp_desc xdp_desc;
> + bool xmit = false;
> +
> + /* ensure that TX ring is not filled up by XDP, always MAX_SKB_FRAGS
> + * will be available for normal TX path and queue is stopped there if
> + * necessary
> + */
> + if (desc_available <= (MAX_SKB_FRAGS + 1))
> + return;
> + desc_available -= MAX_SKB_FRAGS + 1;
> +
> + while (xsk_tx_peek_desc(tx->xsk_pool, &xdp_desc) && desc_available--) {
Again, I am curious how batch API usage would improve your perf.
> + tsnep_xdp_xmit_frame_ring_zc(&xdp_desc, tx);
> + xmit = true;
> + }
> +
> + if (xmit) {
> + tsnep_xdp_xmit_flush(tx);
> + xsk_tx_release(tx->xsk_pool);
> + }
> +}
> +
