On Tue, Apr 16, 2019 at 5:10 PM Xiaolong Ye <xiaolong...@intel.com> wrote:
> As David pointed out, if we reserve N slots for Tx, but only submit n
> slots, we would end up with an incorrect opinion of the number of available
> slots later, we also would get wrong idx when we call
> xsk_ring_prod__reserve next time. It also applies to
> xsk_ring_cons__peek()/xsk_ring_cons__release().
>
> This patch ensures that both reserve/submit and peek/release are
> consistent.
>
> Fixes: f1debd77efaf ("net/af_xdp: introduce AF_XDP PMD")
>
> Suggested-by: David Marchand <david.march...@redhat.com>
> Signed-off-by: Xiaolong Ye <xiaolong...@intel.com>
> ---
> drivers/net/af_xdp/rte_eth_af_xdp.c | 77 +++++++++++++++--------------
> 1 file changed, 39 insertions(+), 38 deletions(-)
>
> diff --git a/drivers/net/af_xdp/rte_eth_af_xdp.c
> b/drivers/net/af_xdp/rte_eth_af_xdp.c
> index 5cc643ce2..b00cd6e03 100644
> --- a/drivers/net/af_xdp/rte_eth_af_xdp.c
> +++ b/drivers/net/af_xdp/rte_eth_af_xdp.c
> @@ -134,30 +134,34 @@ static const struct rte_eth_link pmd_link = {
> };
>
> static inline int
> -reserve_fill_queue(struct xsk_umem_info *umem, int reserve_size)
> +reserve_fill_queue(struct xsk_umem_info *umem, uint16_t reserve_size)
> {
> struct xsk_ring_prod *fq = &umem->fq;
> + void *addrs[reserve_size];
> uint32_t idx;
> - int i, ret;
> + uint16_t i;
> +
> + if (rte_ring_dequeue_bulk(umem->buf_ring, addrs, reserve_size,
> NULL)
> + != reserve_size) {
> + AF_XDP_LOG(DEBUG, "Failed to get enough buffers for
> fq.\n");
> + return -1;
> + }
>
> - ret = xsk_ring_prod__reserve(fq, reserve_size, &idx);
> - if (unlikely(!ret)) {
> + if (unlikely(!xsk_ring_prod__reserve(fq, reserve_size, &idx))) {
> AF_XDP_LOG(ERR, "Failed to reserve enough fq descs.\n");
>
Is this situation happening a lot ?
If this is the case, I would prefer see this as a DEBUG message rather than
ERR.
- return ret;
> + rte_ring_enqueue_bulk(umem->buf_ring, addrs,
> + reserve_size, NULL);
> + return -1;
> }
>
> for (i = 0; i < reserve_size; i++) {
> __u64 *fq_addr;
> - void *addr = NULL;
> - if (rte_ring_dequeue(umem->buf_ring, &addr)) {
> - i--;
> - break;
> - }
> +
> fq_addr = xsk_ring_prod__fill_addr(fq, idx++);
> - *fq_addr = (uint64_t)addr;
> + *fq_addr = (uint64_t)addrs[i];
> }
>
> - xsk_ring_prod__submit(fq, i);
> + xsk_ring_prod__submit(fq, reserve_size);
>
> return 0;
> }
> @@ -174,21 +178,20 @@ eth_af_xdp_rx(void *queue, struct rte_mbuf **bufs,
> uint16_t nb_pkts)
> struct rte_mbuf *mbufs[ETH_AF_XDP_TX_BATCH_SIZE];
> unsigned long dropped = 0;
> unsigned long rx_bytes = 0;
> - uint16_t count = 0;
> int rcvd, i;
>
> nb_pkts = RTE_MIN(nb_pkts, ETH_AF_XDP_TX_BATCH_SIZE);
>
Nothing to do with your patch, but it should be ETH_AF_XDP_ "R"
X_BATCH_SIZE (we have two occurences of this in eth_af_xdp_rx).
I can send a follow up patch after yours.
Or you do it, your choice :-)
> + if (unlikely(rte_pktmbuf_alloc_bulk(rxq->mb_pool, mbufs, nb_pkts)
> != 0))
> + return 0;
> +
> rcvd = xsk_ring_cons__peek(rx, nb_pkts, &idx_rx);
> if (rcvd == 0)
> - return 0;
> + goto out;
>
> if (xsk_prod_nb_free(fq, free_thresh) >= free_thresh)
> (void)reserve_fill_queue(umem, ETH_AF_XDP_RX_BATCH_SIZE);
>
> - if (unlikely(rte_pktmbuf_alloc_bulk(rxq->mb_pool, mbufs, rcvd) !=
> 0))
> - return 0;
> -
> for (i = 0; i < rcvd; i++) {
> const struct xdp_desc *desc;
> uint64_t addr;
> @@ -204,7 +207,7 @@ eth_af_xdp_rx(void *queue, struct rte_mbuf **bufs,
> uint16_t nb_pkts)
> rte_pktmbuf_pkt_len(mbufs[i]) = len;
> rte_pktmbuf_data_len(mbufs[i]) = len;
> rx_bytes += len;
> - bufs[count++] = mbufs[i];
> + bufs[i] = mbufs[i];
>
> rte_ring_enqueue(umem->buf_ring, (void *)addr);
> }
> @@ -215,7 +218,12 @@ eth_af_xdp_rx(void *queue, struct rte_mbuf **bufs,
> uint16_t nb_pkts)
> rxq->stats.rx_pkts += (rcvd - dropped);
> rxq->stats.rx_bytes += rx_bytes;
>
> - return count;
> + out:
>
No space before label.
+ if (rcvd != nb_pkts)
> + rte_mempool_put_bulk(rxq->mb_pool, (void **)&mbufs[rcvd],
> + nb_pkts - rcvd);
> +
> + return rcvd;
> }
>
> static void
> @@ -262,7 +270,7 @@ eth_af_xdp_tx(void *queue, struct rte_mbuf **bufs,
> uint16_t nb_pkts)
> struct rte_mbuf *mbuf;
> void *addrs[ETH_AF_XDP_TX_BATCH_SIZE];
> unsigned long tx_bytes = 0;
> - int i, valid = 0;
> + int i;
> uint32_t idx_tx;
>
> nb_pkts = RTE_MIN(nb_pkts, ETH_AF_XDP_TX_BATCH_SIZE);
> @@ -283,20 +291,18 @@ eth_af_xdp_tx(void *queue, struct rte_mbuf **bufs,
> uint16_t nb_pkts)
> for (i = 0; i < nb_pkts; i++) {
> struct xdp_desc *desc;
> void *pkt;
> - uint32_t buf_len = ETH_AF_XDP_FRAME_SIZE
> - - ETH_AF_XDP_DATA_HEADROOM;
> +
> desc = xsk_ring_prod__tx_desc(&txq->tx, idx_tx + i);
> mbuf = bufs[i];
> - if (mbuf->pkt_len <= buf_len) {
> - desc->addr = (uint64_t)addrs[valid];
> - desc->len = mbuf->pkt_len;
> - pkt = xsk_umem__get_data(umem->mz->addr,
> - desc->addr);
> - rte_memcpy(pkt, rte_pktmbuf_mtod(mbuf, void *),
> - desc->len);
> - valid++;
> - tx_bytes += mbuf->pkt_len;
> - }
> +
> + desc->addr = (uint64_t)addrs[i];
> + desc->len = mbuf->pkt_len;
> + pkt = xsk_umem__get_data(umem->mz->addr,
> + desc->addr);
> + rte_memcpy(pkt, rte_pktmbuf_mtod(mbuf, void *),
> + desc->len);
> + tx_bytes += mbuf->pkt_len;
> +
> rte_pktmbuf_free(mbuf);
> }
>
>
I can see that the buffers in umem can embed ETH_AF_XDP_FRAME_SIZE -
ETH_AF_XDP_DATA_HEADROOM bytes.
And this driver does not support multi segment.
So we are missing a check in eth_dev_mtu_set().
The min_mtu / max_mtu fields are not filled in eth_dev_info().
You should fix this in a preparation patch before this change.
@@ -304,12 +310,7 @@ eth_af_xdp_tx(void *queue, struct rte_mbuf **bufs,
> uint16_t nb_pkts)
>
> kick_tx(txq);
>
> - if (valid < nb_pkts)
> - rte_ring_enqueue_bulk(umem->buf_ring, &addrs[valid],
> - nb_pkts - valid, NULL);
> -
> - txq->stats.err_pkts += nb_pkts - valid;
>
err_pkts stats is not used anymore afaics.
- txq->stats.tx_pkts += valid;
> + txq->stats.tx_pkts += nb_pkts;
> txq->stats.tx_bytes += tx_bytes;
>
> return nb_pkts;
> --
> 2.17.1
>
>
--
David Marchand
