On Fri, 25 Sep 2020 12:01:32 +0200
Lorenzo Bianconi <lore...@kernel.org> wrote:
> Try to recycle the xdp tx buffer into the in-irq page_pool cache if
> mvneta_txq_bufs_free is executed in the NAPI context.
NACK - I don't think this is safe. That is also why I named the
function postfix rx_napi. The page pool->alloc.cache is associated
with the drivers RX-queue. The xdp_frame's that gets freed could be
coming from a remote driver that use page_pool. This remote drivers
RX-queue processing can run concurrently on a different CPU, than this
drivers TXQ-cleanup.
If you want to speedup this, I instead suggest that you add a
xdp_return_frame_bulk API.
> Signed-off-by: Lorenzo Bianconi <lore...@kernel.org>
> ---
> drivers/net/ethernet/marvell/mvneta.c | 11 +++++++----
> 1 file changed, 7 insertions(+), 4 deletions(-)
>
> diff --git a/drivers/net/ethernet/marvell/mvneta.c
> b/drivers/net/ethernet/marvell/mvneta.c
> index 14df3aec285d..646fbf4ed638 100644
> --- a/drivers/net/ethernet/marvell/mvneta.c
> +++ b/drivers/net/ethernet/marvell/mvneta.c
> @@ -1831,7 +1831,7 @@ static struct mvneta_tx_queue
> *mvneta_tx_done_policy(struct mvneta_port *pp,
> /* Free tx queue skbuffs */
> static void mvneta_txq_bufs_free(struct mvneta_port *pp,
> struct mvneta_tx_queue *txq, int num,
> - struct netdev_queue *nq)
> + struct netdev_queue *nq, bool napi)
> {
> unsigned int bytes_compl = 0, pkts_compl = 0;
> int i;
> @@ -1854,7 +1854,10 @@ static void mvneta_txq_bufs_free(struct mvneta_port
> *pp,
> dev_kfree_skb_any(buf->skb);
> } else if (buf->type == MVNETA_TYPE_XDP_TX ||
> buf->type == MVNETA_TYPE_XDP_NDO) {
> - xdp_return_frame(buf->xdpf);
> + if (napi)
> + xdp_return_frame_rx_napi(buf->xdpf);
> + else
> + xdp_return_frame(buf->xdpf);
> }
> }
>
> @@ -1872,7 +1875,7 @@ static void mvneta_txq_done(struct mvneta_port *pp,
> if (!tx_done)
> return;
>
> - mvneta_txq_bufs_free(pp, txq, tx_done, nq);
> + mvneta_txq_bufs_free(pp, txq, tx_done, nq, true);
>
> txq->count -= tx_done;
>
> @@ -2859,7 +2862,7 @@ static void mvneta_txq_done_force(struct mvneta_port
> *pp,
> struct netdev_queue *nq = netdev_get_tx_queue(pp->dev, txq->id);
> int tx_done = txq->count;
>
> - mvneta_txq_bufs_free(pp, txq, tx_done, nq);
> + mvneta_txq_bufs_free(pp, txq, tx_done, nq, false);
>
> /* reset txq */
> txq->count = 0;
--
Best regards,
Jesper Dangaard Brouer
MSc.CS, Principal Kernel Engineer at Red Hat
LinkedIn: http://www.linkedin.com/in/brouer
