[dpdk-dev] [PATCH v8 3/4] net/ixgbe: cleanup Tx buffers

[Ananyev, Konstantin]

> 
> Add support to the ixgbe driver for the API rte_eth_tx_done_cleanup
> to force free consumed buffers on Tx ring.
> 
> Signed-off-by: Chenxu Di <chenxux.di at intel.com>
> ---
>  drivers/net/ixgbe/ixgbe_ethdev.c        |   2 +
>  drivers/net/ixgbe/ixgbe_rxtx.c          | 109 ++++++++++++++++++++++++
>  drivers/net/ixgbe/ixgbe_rxtx.h          |   8 +-
>  drivers/net/ixgbe/ixgbe_rxtx_vec_neon.c |   1 +
>  drivers/net/ixgbe/ixgbe_rxtx_vec_sse.c  |   1 +
>  5 files changed, 120 insertions(+), 1 deletion(-)
> 
> diff --git a/drivers/net/ixgbe/ixgbe_ethdev.c b/drivers/net/ixgbe/ixgbe_ethdev.c
> index 2c6fd0f13..75bdd391a 100644
> --- a/drivers/net/ixgbe/ixgbe_ethdev.c
> +++ b/drivers/net/ixgbe/ixgbe_ethdev.c
> @@ -601,6 +601,7 @@ static const struct eth_dev_ops ixgbe_eth_dev_ops = {
>  	.udp_tunnel_port_add  = ixgbe_dev_udp_tunnel_port_add,
>  	.udp_tunnel_port_del  = ixgbe_dev_udp_tunnel_port_del,
>  	.tm_ops_get           = ixgbe_tm_ops_get,
> +	.tx_done_cleanup      = ixgbe_dev_tx_done_cleanup,
>  };
> 
>  /*
> @@ -649,6 +650,7 @@ static const struct eth_dev_ops ixgbevf_eth_dev_ops = {
>  	.reta_query           = ixgbe_dev_rss_reta_query,
>  	.rss_hash_update      = ixgbe_dev_rss_hash_update,
>  	.rss_hash_conf_get    = ixgbe_dev_rss_hash_conf_get,
> +	.tx_done_cleanup      = ixgbe_dev_tx_done_cleanup,
>  };
> 
>  /* store statistics names and its offset in stats structure */
> diff --git a/drivers/net/ixgbe/ixgbe_rxtx.c b/drivers/net/ixgbe/ixgbe_rxtx.c
> index fa572d184..23c897d3a 100644
> --- a/drivers/net/ixgbe/ixgbe_rxtx.c
> +++ b/drivers/net/ixgbe/ixgbe_rxtx.c
> @@ -2306,6 +2306,114 @@ ixgbe_tx_queue_release_mbufs(struct ixgbe_tx_queue *txq)
>  	}
>  }
> 
> +int
> +ixgbe_tx_done_cleanup_full(struct ixgbe_tx_queue *txq, uint32_t free_cnt)
> +{
> +	uint32_t pkt_cnt;
> +	uint16_t i;
> +	uint16_t tx_last;
> +	uint16_t tx_id;
> +	uint16_t nb_tx_to_clean;
> +	uint16_t nb_tx_free_last;
> +	struct ixgbe_tx_entry *swr_ring = txq->sw_ring;
> +
> +	/* Start free mbuf from the next of tx_tail */
> +	tx_last = txq->tx_tail;
> +	tx_id  = swr_ring[tx_last].next_id;
> +
> +	if (txq->nb_tx_free == 0 && ixgbe_xmit_cleanup(txq))
> +		return 0;
> +
> +	nb_tx_to_clean = txq->nb_tx_free;
> +	nb_tx_free_last = txq->nb_tx_free;
> +	if (!free_cnt)
> +		free_cnt = txq->nb_tx_desc;
> +
> +	/* Loop through swr_ring to count the amount of
> +	 * freeable mubfs and packets.
> +	 */
> +	for (pkt_cnt = 0; pkt_cnt < free_cnt; ) {
> +		for (i = 0; i < nb_tx_to_clean &&
> +			pkt_cnt < free_cnt &&
> +			tx_id != tx_last; i++) {
> +			if (swr_ring[tx_id].mbuf != NULL) {
> +				rte_pktmbuf_free_seg(swr_ring[tx_id].mbuf);
> +				swr_ring[tx_id].mbuf = NULL;
> +
> +				/*
> +				 * last segment in the packet,
> +				 * increment packet count
> +				 */
> +				pkt_cnt += (swr_ring[tx_id].last_id == tx_id);
> +			}
> +
> +			tx_id = swr_ring[tx_id].next_id;
> +		}
> +
> +		if (txq->tx_rs_thresh > txq->nb_tx_desc -
> +			txq->nb_tx_free || tx_id == tx_last)
> +			break;
> +
> +		if (pkt_cnt < free_cnt) {
> +			if (ixgbe_xmit_cleanup(txq))
> +				break;
> +
> +			nb_tx_to_clean = txq->nb_tx_free - nb_tx_free_last;
> +			nb_tx_free_last = txq->nb_tx_free;
> +		}
> +	}
> +
> +	return (int)pkt_cnt;
> +}
> +
> +int
> +ixgbe_tx_done_cleanup_vec(struct ixgbe_tx_queue *txq __rte_unused,
> +			uint32_t free_cnt __rte_unused)
> +{
> +	return -ENOTSUP;
> +}
> +
> +int
> +ixgbe_tx_done_cleanup_simple(struct ixgbe_tx_queue *txq,
> +			uint32_t free_cnt)
> +{
> +	int i, n, cnt;
> +
> +	if (free_cnt == 0 || free_cnt > txq->nb_tx_desc)
> +		free_cnt = txq->nb_tx_desc;
> +
> +	cnt = free_cnt - free_cnt % txq->tx_rs_thresh;
> +
> +	for (i = 0; i < cnt; i += n) {
> +		if (txq->nb_tx_desc - txq->nb_tx_free < txq->tx_rs_thresh)
> +			break;
> +
> +		n = ixgbe_tx_free_bufs(txq);
> +
> +		if (n == 0)
> +			break;
> +	}
> +
> +	return i;
> +}
> +
> +int
> +ixgbe_dev_tx_done_cleanup(void *tx_queue, uint32_t free_cnt)
> +{
> +	struct ixgbe_tx_queue *txq = (struct ixgbe_tx_queue *)tx_queue;
> +	return txq->ops->txq_done_cleanup(txq, free_cnt);
> +}
> +
> +int
> +ixgbe_tx_done_cleanup(struct ixgbe_tx_queue *txq, uint32_t free_cnt)
> +{
> +	/* Use a simple Tx queue (no offloads, no multi segs) if possible */
> +	if (txq->offloads == 0)
> +		return ixgbe_tx_done_cleanup_simple(txq, free_cnt);
> +	else
> +		return ixgbe_tx_done_cleanup_full(txq, free_cnt);
> +}
> +
>  static void __attribute__((cold))
>  ixgbe_tx_free_swring(struct ixgbe_tx_queue *txq)
>  {
> @@ -2375,6 +2483,7 @@ static const struct ixgbe_txq_ops def_txq_ops = {
>  	.release_mbufs = ixgbe_tx_queue_release_mbufs,
>  	.free_swring = ixgbe_tx_free_swring,
>  	.reset = ixgbe_reset_tx_queue,
> +	.txq_done_cleanup = ixgbe_tx_done_cleanup,
>  };
> 
>  /* Takes an ethdev and a queue and sets up the tx function to be used based on
> diff --git a/drivers/net/ixgbe/ixgbe_rxtx.h b/drivers/net/ixgbe/ixgbe_rxtx.h
> index 505d344b9..41a3738ce 100644
> --- a/drivers/net/ixgbe/ixgbe_rxtx.h
> +++ b/drivers/net/ixgbe/ixgbe_rxtx.h
> @@ -238,6 +238,7 @@ struct ixgbe_txq_ops {
>  	void (*release_mbufs)(struct ixgbe_tx_queue *txq);
>  	void (*free_swring)(struct ixgbe_tx_queue *txq);
>  	void (*reset)(struct ixgbe_tx_queue *txq);
> +	int (*txq_done_cleanup)(struct ixgbe_tx_queue *txq, uint32_t free_cnt);
>  };
> 
>  /*
> @@ -253,7 +254,6 @@ struct ixgbe_txq_ops {
>  			 IXGBE_ADVTXD_DCMD_DEXT |\
>  			 IXGBE_ADVTXD_DCMD_EOP)
> 
> -
>  /* Takes an ethdev and a queue and sets up the tx function to be used based on
>   * the queue parameters. Used in tx_queue_setup by primary process and then
>   * in dev_init by secondary process when attaching to an existing ethdev.
> @@ -285,6 +285,12 @@ int ixgbe_rx_vec_dev_conf_condition_check(struct rte_eth_dev *dev);
>  int ixgbe_rxq_vec_setup(struct ixgbe_rx_queue *rxq);
>  void ixgbe_rx_queue_release_mbufs_vec(struct ixgbe_rx_queue *rxq);
> 
> +int ixgbe_dev_tx_done_cleanup(void *tx_queue, uint32_t free_cnt);

As a nit: I don't think you need to make these 4 functions below extrernal.
_cleanup(), cleanup_full and cleanup_simple can be static ones in ixgbe_rxtx.c

_cleanup_vec() can be static in ixgbe_rxtx_vec_common.h
BTW, I think _cleanup_vec() will be identical to cleanup_simple().

Apart, from that:
Acked-by: Konstantin Ananyev <konstantin.ananyev at intel.com>

As a side notice, I think we need to add into test-pmd ability to call/test
tx_cleanup_done (either as a separate command, or new fwd mode, or...).
But that probably subject for separate patch series. 

> +int ixgbe_tx_done_cleanup(struct ixgbe_tx_queue *txq, uint32_t free_cnt);
> +int ixgbe_tx_done_cleanup_full(struct ixgbe_tx_queue *txq, uint32_t free_cnt);
> +int ixgbe_tx_done_cleanup_vec(struct ixgbe_tx_queue *txq, uint32_t free_cnt);
> +int ixgbe_tx_done_cleanup_simple(struct ixgbe_tx_queue *txq, uint32_t free_cnt);
> +
>  extern const uint32_t ptype_table[IXGBE_PACKET_TYPE_MAX];
>  extern const uint32_t ptype_table_tn[IXGBE_PACKET_TYPE_TN_MAX];
> 
> diff --git a/drivers/net/ixgbe/ixgbe_rxtx_vec_neon.c b/drivers/net/ixgbe/ixgbe_rxtx_vec_neon.c
> index feb86c61e..cd9b7dc01 100644
> --- a/drivers/net/ixgbe/ixgbe_rxtx_vec_neon.c
> +++ b/drivers/net/ixgbe/ixgbe_rxtx_vec_neon.c
> @@ -559,6 +559,7 @@ static const struct ixgbe_txq_ops vec_txq_ops = {
>  	.release_mbufs = ixgbe_tx_queue_release_mbufs_vec,
>  	.free_swring = ixgbe_tx_free_swring,
>  	.reset = ixgbe_reset_tx_queue,
> +	.txq_done_cleanup = ixgbe_tx_done_cleanup_vec,
>  };
> 
>  int __attribute__((cold))
> diff --git a/drivers/net/ixgbe/ixgbe_rxtx_vec_sse.c b/drivers/net/ixgbe/ixgbe_rxtx_vec_sse.c
> index 599ba30e5..63bfac9fa 100644
> --- a/drivers/net/ixgbe/ixgbe_rxtx_vec_sse.c
> +++ b/drivers/net/ixgbe/ixgbe_rxtx_vec_sse.c
> @@ -730,6 +730,7 @@ static const struct ixgbe_txq_ops vec_txq_ops = {
>  	.release_mbufs = ixgbe_tx_queue_release_mbufs_vec,
>  	.free_swring = ixgbe_tx_free_swring,
>  	.reset = ixgbe_reset_tx_queue,
> +	.txq_done_cleanup = ixgbe_tx_done_cleanup_vec,
>  };
> 
>  int __attribute__((cold))
> --
> 2.17.1