On Sat, 2023-04-15 at 16:42 +0200, Gerhard Engleder wrote:
> Send and complete XSK pool frames within TX NAPI context. NAPI context
> is triggered by ndo_xsk_wakeup.
>
> Test results with A53 1.2GHz:
>
> xdpsock txonly copy mode:
> pps pkts 1.00
> tx 284,409 11,398,144
> Two CPUs with 100% and 10% utilization.
>
> xdpsock txonly zero-copy mode:
> pps pkts 1.00
> tx 511,929 5,890,368
> Two CPUs with 100% and 1% utilization.
>
> Packet rate increases and CPU utilization is reduced.
>
> 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(-)
>
> diff --git a/drivers/net/ethernet/engleder/tsnep.h b/drivers/net/ethernet/engleder/tsnep.h
> index d0bea605a1d1..11b29f56aaf9 100644
> --- a/drivers/net/ethernet/engleder/tsnep.h
> +++ b/drivers/net/ethernet/engleder/tsnep.h
> @@ -70,6 +70,7 @@ struct tsnep_tx_entry {
> union {
> struct sk_buff *skb;
> struct xdp_frame *xdpf;
> + bool zc;
> };
> size_t len;
> DEFINE_DMA_UNMAP_ADDR(dma);
> @@ -88,6 +89,7 @@ struct tsnep_tx {
> int read;
> u32 owner_counter;
> int increment_owner_counter;
> + struct xsk_buff_pool *xsk_pool;
>
> u32 packets;
> u32 bytes;
> diff --git a/drivers/net/ethernet/engleder/tsnep_main.c b/drivers/net/ethernet/engleder/tsnep_main.c
> index 13e5d4438082..de51d0cc8935 100644
> --- a/drivers/net/ethernet/engleder/tsnep_main.c
> +++ b/drivers/net/ethernet/engleder/tsnep_main.c
> @@ -54,6 +54,8 @@
> #define TSNEP_TX_TYPE_SKB_FRAG BIT(1)
> #define TSNEP_TX_TYPE_XDP_TX BIT(2)
> #define TSNEP_TX_TYPE_XDP_NDO BIT(3)
> +#define TSNEP_TX_TYPE_XDP (TSNEP_TX_TYPE_XDP_TX | TSNEP_TX_TYPE_XDP_NDO)
> +#define TSNEP_TX_TYPE_XSK BIT(4)
>
> #define TSNEP_XDP_TX BIT(0)
> #define TSNEP_XDP_REDIRECT BIT(1)
> @@ -322,13 +324,51 @@ static void tsnep_tx_init(struct tsnep_tx *tx)
> tx->increment_owner_counter = TSNEP_RING_SIZE - 1;
> }
>
> +static void tsnep_tx_enable(struct tsnep_tx *tx)
> +{
> + struct netdev_queue *nq;
> +
> + nq = netdev_get_tx_queue(tx->adapter->netdev, tx->queue_index);
> +
> + local_bh_disable();
> + __netif_tx_lock(nq, smp_processor_id());
The above 2 statements could be replaced with:
__netif_tx_lock_bh()
> + netif_tx_wake_queue(nq);
> + __netif_tx_unlock(nq);
> + local_bh_enable();
__netif_tx_unlock_bh()
> +}
> +
> +static void tsnep_tx_disable(struct tsnep_tx *tx, struct napi_struct *napi)
> +{
> + struct netdev_queue *nq;
> + u32 val;
> +
> + nq = netdev_get_tx_queue(tx->adapter->netdev, tx->queue_index);
> +
> + local_bh_disable();
> + __netif_tx_lock(nq, smp_processor_id());
Same here.
Thanks!
Paolo
