[dpdk-dev] [PATCH v1 3/4] ixgbe: Kill ixgbe_recv_scattered_pkts()

[Ananyev, Konstantin]

Hi Vlad,

> -----Original Message-----
> From: dev [mailto:dev-bounces at dpdk.org] On Behalf Of Vlad Zolotarov
> Sent: Sunday, April 26, 2015 3:46 PM
> To: dev at dpdk.org
> Subject: [dpdk-dev] [PATCH v1 3/4] ixgbe: Kill ixgbe_recv_scattered_pkts()
> 
> Kill ixgbe_recv_scattered_pkts() - use ixgbe_recv_pkts_lro_single_alloc()
> instead.
> 
> Work against HW queues in LRO and scattered Rx cases is exactly the same.
> Therefore we may drop the inferior callback.
> 
> Signed-off-by: Vlad Zolotarov <vladz at cloudius-systems.com>
> ---
>  lib/librte_pmd_ixgbe/ixgbe_ethdev.c |   2 +-
>  lib/librte_pmd_ixgbe/ixgbe_ethdev.h |   3 -
>  lib/librte_pmd_ixgbe/ixgbe_rxtx.c   | 243 +-----------------------------------
>  3 files changed, 7 insertions(+), 241 deletions(-)
> 
> diff --git a/lib/librte_pmd_ixgbe/ixgbe_ethdev.c b/lib/librte_pmd_ixgbe/ixgbe_ethdev.c
> index aec1de9..5f9a1cf 100644
> --- a/lib/librte_pmd_ixgbe/ixgbe_ethdev.c
> +++ b/lib/librte_pmd_ixgbe/ixgbe_ethdev.c
> @@ -986,7 +986,7 @@ eth_ixgbevf_dev_init(struct rte_eth_dev *eth_dev)
>  	 * RX function */
>  	if (rte_eal_process_type() != RTE_PROC_PRIMARY){
>  		if (eth_dev->data->scattered_rx)
> -			eth_dev->rx_pkt_burst = ixgbe_recv_scattered_pkts;
> +			eth_dev->rx_pkt_burst = ixgbe_recv_pkts_lro_single_alloc;
>  		return 0;
>  	}
> 
> diff --git a/lib/librte_pmd_ixgbe/ixgbe_ethdev.h b/lib/librte_pmd_ixgbe/ixgbe_ethdev.h
> index 5b90115..419ea5d 100644
> --- a/lib/librte_pmd_ixgbe/ixgbe_ethdev.h
> +++ b/lib/librte_pmd_ixgbe/ixgbe_ethdev.h
> @@ -352,9 +352,6 @@ void ixgbevf_dev_rxtx_start(struct rte_eth_dev *dev);
>  uint16_t ixgbe_recv_pkts(void *rx_queue, struct rte_mbuf **rx_pkts,
>  		uint16_t nb_pkts);
> 
> -uint16_t ixgbe_recv_scattered_pkts(void *rx_queue,
> -		struct rte_mbuf **rx_pkts, uint16_t nb_pkts);
> -
>  uint16_t ixgbe_recv_pkts_lro_single_alloc(void *rx_queue,
>  		struct rte_mbuf **rx_pkts, uint16_t nb_pkts);
>  uint16_t ixgbe_recv_pkts_lro_bulk_alloc(void *rx_queue,
> diff --git a/lib/librte_pmd_ixgbe/ixgbe_rxtx.c b/lib/librte_pmd_ixgbe/ixgbe_rxtx.c
> index a45f51e..c23e20f 100644
> --- a/lib/librte_pmd_ixgbe/ixgbe_rxtx.c
> +++ b/lib/librte_pmd_ixgbe/ixgbe_rxtx.c
> @@ -1722,239 +1722,6 @@ ixgbe_recv_pkts_lro_bulk_alloc(void *rx_queue, struct rte_mbuf **rx_pkts,
>  	return ixgbe_recv_pkts_lro(rx_queue, rx_pkts, nb_pkts, true);
>  }
> 
> -uint16_t
> -ixgbe_recv_scattered_pkts(void *rx_queue, struct rte_mbuf **rx_pkts,
> -			  uint16_t nb_pkts)
> -{
> -	struct ixgbe_rx_queue *rxq;
> -	volatile union ixgbe_adv_rx_desc *rx_ring;
> -	volatile union ixgbe_adv_rx_desc *rxdp;
> -	struct ixgbe_rx_entry *sw_ring;
> -	struct ixgbe_rx_entry *rxe;
> -	struct rte_mbuf *first_seg;
> -	struct rte_mbuf *last_seg;
> -	struct rte_mbuf *rxm;
> -	struct rte_mbuf *nmb;
> -	union ixgbe_adv_rx_desc rxd;
> -	uint64_t dma; /* Physical address of mbuf data buffer */
> -	uint32_t staterr;
> -	uint16_t rx_id;
> -	uint16_t nb_rx;
> -	uint16_t nb_hold;
> -	uint16_t data_len;
> -
> -	nb_rx = 0;
> -	nb_hold = 0;
> -	rxq = rx_queue;
> -	rx_id = rxq->rx_tail;
> -	rx_ring = rxq->rx_ring;
> -	sw_ring = rxq->sw_ring;
> -
> -	/*
> -	 * Retrieve RX context of current packet, if any.
> -	 */
> -	first_seg = rxq->pkt_first_seg;
> -	last_seg = rxq->pkt_last_seg;
> -
> -	while (nb_rx < nb_pkts) {
> -	next_desc:
> -		/*
> -		 * The order of operations here is important as the DD status
> -		 * bit must not be read after any other descriptor fields.
> -		 * rx_ring and rxdp are pointing to volatile data so the order
> -		 * of accesses cannot be reordered by the compiler. If they were
> -		 * not volatile, they could be reordered which could lead to
> -		 * using invalid descriptor fields when read from rxd.
> -		 */
> -		rxdp = &rx_ring[rx_id];
> -		staterr = rxdp->wb.upper.status_error;
> -		if (! (staterr & rte_cpu_to_le_32(IXGBE_RXDADV_STAT_DD)))
> -			break;
> -		rxd = *rxdp;
> -
> -		/*
> -		 * Descriptor done.
> -		 *
> -		 * Allocate a new mbuf to replenish the RX ring descriptor.
> -		 * If the allocation fails:
> -		 *    - arrange for that RX descriptor to be the first one
> -		 *      being parsed the next time the receive function is
> -		 *      invoked [on the same queue].
> -		 *
> -		 *    - Stop parsing the RX ring and return immediately.
> -		 *
> -		 * This policy does not drop the packet received in the RX
> -		 * descriptor for which the allocation of a new mbuf failed.
> -		 * Thus, it allows that packet to be later retrieved if
> -		 * mbuf have been freed in the mean time.
> -		 * As a side effect, holding RX descriptors instead of
> -		 * systematically giving them back to the NIC may lead to
> -		 * RX ring exhaustion situations.
> -		 * However, the NIC can gracefully prevent such situations
> -		 * to happen by sending specific "back-pressure" flow control
> -		 * frames to its peer(s).
> -		 */
> -		PMD_RX_LOG(DEBUG, "port_id=%u queue_id=%u rx_id=%u "
> -			   "staterr=0x%x data_len=%u",
> -			   (unsigned) rxq->port_id, (unsigned) rxq->queue_id,
> -			   (unsigned) rx_id, (unsigned) staterr,
> -			   (unsigned) rte_le_to_cpu_16(rxd.wb.upper.length));
> -
> -		nmb = rte_rxmbuf_alloc(rxq->mb_pool);
> -		if (nmb == NULL) {
> -			PMD_RX_LOG(DEBUG, "RX mbuf alloc failed port_id=%u "
> -				   "queue_id=%u", (unsigned) rxq->port_id,
> -				   (unsigned) rxq->queue_id);
> -			rte_eth_devices[rxq->port_id].data->rx_mbuf_alloc_failed++;
> -			break;
> -		}
> -
> -		nb_hold++;
> -		rxe = &sw_ring[rx_id];
> -		rx_id++;
> -		if (rx_id == rxq->nb_rx_desc)
> -			rx_id = 0;
> -
> -		/* Prefetch next mbuf while processing current one. */
> -		rte_ixgbe_prefetch(sw_ring[rx_id].mbuf);
> -
> -		/*
> -		 * When next RX descriptor is on a cache-line boundary,
> -		 * prefetch the next 4 RX descriptors and the next 8 pointers
> -		 * to mbufs.
> -		 */
> -		if ((rx_id & 0x3) == 0) {
> -			rte_ixgbe_prefetch(&rx_ring[rx_id]);
> -			rte_ixgbe_prefetch(&sw_ring[rx_id]);
> -		}
> -
> -		/*
> -		 * Update RX descriptor with the physical address of the new
> -		 * data buffer of the new allocated mbuf.
> -		 */
> -		rxm = rxe->mbuf;
> -		rxe->mbuf = nmb;
> -		dma = rte_cpu_to_le_64(RTE_MBUF_DATA_DMA_ADDR_DEFAULT(nmb));
> -		rxdp->read.hdr_addr = dma;
> -		rxdp->read.pkt_addr = dma;
> -
> -		/*
> -		 * Set data length & data buffer address of mbuf.
> -		 */
> -		data_len = rte_le_to_cpu_16(rxd.wb.upper.length);
> -		rxm->data_len = data_len;
> -		rxm->data_off = RTE_PKTMBUF_HEADROOM;
> -
> -		/*
> -		 * If this is the first buffer of the received packet,
> -		 * set the pointer to the first mbuf of the packet and
> -		 * initialize its context.
> -		 * Otherwise, update the total length and the number of segments
> -		 * of the current scattered packet, and update the pointer to
> -		 * the last mbuf of the current packet.
> -		 */
> -		if (first_seg == NULL) {
> -			first_seg = rxm;
> -			first_seg->pkt_len = data_len;
> -			first_seg->nb_segs = 1;
> -		} else {
> -			first_seg->pkt_len = (uint16_t)(first_seg->pkt_len
> -					+ data_len);
> -			first_seg->nb_segs++;
> -			last_seg->next = rxm;
> -		}
> -
> -		/*
> -		 * If this is not the last buffer of the received packet,
> -		 * update the pointer to the last mbuf of the current scattered
> -		 * packet and continue to parse the RX ring.
> -		 */
> -		if (! (staterr & IXGBE_RXDADV_STAT_EOP)) {
> -			last_seg = rxm;
> -			goto next_desc;
> -		}
> -
> -		/*
> -		 * This is the last buffer of the received packet.
> -		 * If the CRC is not stripped by the hardware:
> -		 *   - Subtract the CRC	length from the total packet length.
> -		 *   - If the last buffer only contains the whole CRC or a part
> -		 *     of it, free the mbuf associated to the last buffer.
> -		 *     If part of the CRC is also contained in the previous
> -		 *     mbuf, subtract the length of that CRC part from the
> -		 *     data length of the previous mbuf.
> -		 */
> -		rxm->next = NULL;
> -		if (unlikely(rxq->crc_len > 0)) {
> -			first_seg->pkt_len -= ETHER_CRC_LEN;
> -			if (data_len <= ETHER_CRC_LEN) {
> -				rte_pktmbuf_free_seg(rxm);
> -				first_seg->nb_segs--;
> -				last_seg->data_len = (uint16_t)
> -					(last_seg->data_len -
> -					 (ETHER_CRC_LEN - data_len));
> -				last_seg->next = NULL;
> -			} else
> -				rxm->data_len =
> -					(uint16_t) (data_len - ETHER_CRC_LEN);
> -		}
> -
> -		/* Initialize the first mbuf of the returned packet */
> -		ixgbe_fill_cluster_head_buf(first_seg, &rxd, rxq->port_id,
> -					    staterr);
> -
> -		/* Prefetch data of first segment, if configured to do so. */
> -		rte_packet_prefetch((char *)first_seg->buf_addr +
> -			first_seg->data_off);
> -
> -		/*
> -		 * Store the mbuf address into the next entry of the array
> -		 * of returned packets.
> -		 */
> -		rx_pkts[nb_rx++] = first_seg;
> -
> -		/*
> -		 * Setup receipt context for a new packet.
> -		 */
> -		first_seg = NULL;
> -	}
> -
> -	/*
> -	 * Record index of the next RX descriptor to probe.
> -	 */
> -	rxq->rx_tail = rx_id;
> -
> -	/*
> -	 * Save receive context.
> -	 */
> -	rxq->pkt_first_seg = first_seg;
> -	rxq->pkt_last_seg = last_seg;
> -
> -	/*
> -	 * If the number of free RX descriptors is greater than the RX free
> -	 * threshold of the queue, advance the Receive Descriptor Tail (RDT)
> -	 * register.
> -	 * Update the RDT with the value of the last processed RX descriptor
> -	 * minus 1, to guarantee that the RDT register is never equal to the
> -	 * RDH register, which creates a "full" ring situtation from the
> -	 * hardware point of view...
> -	 */
> -	nb_hold = (uint16_t) (nb_hold + rxq->nb_rx_hold);
> -	if (nb_hold > rxq->rx_free_thresh) {
> -		PMD_RX_LOG(DEBUG, "port_id=%u queue_id=%u rx_tail=%u "
> -			   "nb_hold=%u nb_rx=%u",
> -			   (unsigned) rxq->port_id, (unsigned) rxq->queue_id,
> -			   (unsigned) rx_id, (unsigned) nb_hold,
> -			   (unsigned) nb_rx);
> -		rx_id = (uint16_t) ((rx_id == 0) ?
> -				     (rxq->nb_rx_desc - 1) : (rx_id - 1));
> -		IXGBE_PCI_REG_WRITE(rxq->rdt_reg_addr, rx_id);
> -		nb_hold = 0;
> -	}
> -	rxq->nb_rx_hold = nb_hold;
> -	return (nb_rx);
> -}
> -
>  /*********************************************************************
>   *
>   *  Queue management functions
> @@ -2623,7 +2390,7 @@ ixgbe_dev_rx_queue_setup(struct rte_eth_dev *dev,
>  		return (-ENOMEM);
>  	}
> 
> -	if (rsc_requested) {
> +	if (rsc_requested || dev_rx_mode->enable_scatter) {
>  		rxq->sw_rsc_ring =
>  			rte_zmalloc_socket("rxq->sw_rsc_ring",
>  					   sizeof(struct ixgbe_rsc_entry) * len,

I think here is a problem:
We allocate sw_rsc_ring only if user explicitly requested LRO or scattered rx. 
Though later, ixgbe_dev_rx_init() might implicitly enable scattered rx, if the provided mbufs are too small:

buf_size = (uint16_t) ((srrctl & IXGBE_SRRCTL_BSIZEPKT_MASK) <<
                                       IXGBE_SRRCTL_BSIZEPKT_SHIFT);

 /* It adds dual VLAN length for supporting dual VLAN */
 if (dev->data->dev_conf.rxmode.max_rx_pkt_len +
                                           2 * IXGBE_VLAN_TAG_SIZE > buf_size)
dev->data->scattered_rx = 1;

So, ixgbe_recv_pkts_lro_*_alloc() will be selected, but ixgbe_recv_pkts_lro will be 0.

Probably the easiest and safest fix, is to always allocate sw_rsc_ring for the queue.
After all, it would consume at max a bit more than 32KB - doesn't seem that much to me.
Konstantin

> @@ -4017,12 +3784,13 @@ void ixgbe_set_rx_function(struct rte_eth_dev *dev)
> 
>  			dev->rx_pkt_burst = ixgbe_recv_scattered_pkts_vec;
>  		} else {
> -			PMD_INIT_LOG(DEBUG, "Using Regualr (non-vector) "
> +			PMD_INIT_LOG(DEBUG, "Using Regualr (non-vector, "
> +					    "single allocation) "
>  					    "Scattered Rx callback "
>  					    "(port=%d).",
>  				     dev->data->port_id);
> 
> -			dev->rx_pkt_burst = ixgbe_recv_scattered_pkts;
> +			dev->rx_pkt_burst = ixgbe_recv_pkts_lro_single_alloc;
>  		}
>  	/*
>  	 * Below we set "simple" callbacks according to port/queues parameters.
> @@ -4855,7 +4623,8 @@ ixgbevf_dev_rx_init(struct rte_eth_dev *dev)
>  					ixgbe_recv_scattered_pkts_vec;
>  			else
>  #endif
> -				dev->rx_pkt_burst = ixgbe_recv_scattered_pkts;
> +				dev->rx_pkt_burst =
> +					ixgbe_recv_pkts_lro_single_alloc;
>  		}
>  	}
> 
> --
> 2.1.0