d/numam-dpdk
1
0
Fork 0
Code Issues Pull Requests Projects Releases Wiki Activity

mpipe: optimize buffer return mechanism

Browse Source 
This submit has changes to optimize the mpipe buffer return. When
a packet is received, instead of allocating and refilling the
buffer stack right away, it tracks the number of pending buffers,
and use HW buffer return as an optimization when the pending
number is below certain threshold, thus save two MMIO writes and
improves performance especially for bidirectional traffic case.

Signed-off-by: Liming Sun <lsun@ezchip.com>
Acked-by: Zhigang Lu <zlu@ezchip.com>
This commit is contained in:
Liming Sun 2016-01-08 09:30:37 -05:00 committed by Thomas Monjalon
parent 23f58cd012
commit ca7e25e632
1 changed files with 39 additions and 10 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

49
drivers/net/mpipe/mpipe_tilegx.c
View File

@ -78,6 +78,13 @@ struct mpipe_context {
struct mpipe_channel_config channels[MPIPE_MAX_CHANNELS];
};
/* Per-core local data. */
struct mpipe_local {
int mbuf_push_debt[RTE_MAX_ETHPORTS]; /* Buffer push debt. */
} __rte_cache_aligned;
#define MPIPE_BUF_DEBT_THRESHOLD 32
static __thread struct mpipe_local mpipe_local;
static struct mpipe_context mpipe_contexts[GXIO_MPIPE_INSTANCE_MAX];
static int mpipe_instances;
static const char *drivername = "MPIPE PMD";
@ -137,7 +144,7 @@ struct mpipe_dev_priv {
int first_bucket; /* mPIPE bucket start index. */
int first_ring; /* mPIPE notif ring start index. */
int notif_group; /* mPIPE notif group. */
rte_atomic32_t dp_count; /* Active datapath thread count. */
rte_atomic32_t dp_count __rte_cache_aligned; /* DP Entry count. */
int tx_stat_mapping[RTE_ETHDEV_QUEUE_STAT_CNTRS];
int rx_stat_mapping[RTE_ETHDEV_QUEUE_STAT_CNTRS];
};
@ -461,6 +468,14 @@ mpipe_dp_wait(struct mpipe_dev_priv *priv)
}
}
static inline int
mpipe_mbuf_stack_index(struct mpipe_dev_priv *priv, struct rte_mbuf *mbuf)
{
return (mbuf->port < RTE_MAX_ETHPORTS) ?
mpipe_priv(&rte_eth_devices[mbuf->port])->stack :
priv->stack;
}
static inline struct rte_mbuf *
mpipe_recv_mbuf(struct mpipe_dev_priv *priv, gxio_mpipe_idesc_t *idesc,
int in_port)
@ -1267,6 +1282,7 @@ mpipe_do_xmit(struct mpipe_tx_queue *tx_queue, struct rte_mbuf **tx_pkts,
unsigned nb_bytes = 0;
unsigned nb_sent = 0;
int nb_slots, i;
uint8_t port_id;
PMD_DEBUG_TX("Trying to transmit %d packets on %s:%d.\n",
nb_pkts, mpipe_name(tx_queue->q.priv),
@ -1315,14 +1331,22 @@ mpipe_do_xmit(struct mpipe_tx_queue *tx_queue, struct rte_mbuf **tx_pkts,
if (priv->tx_comps[idx])
rte_pktmbuf_free_seg(priv->tx_comps[idx]);
port_id = (mbuf->port < RTE_MAX_ETHPORTS) ?
mbuf->port : priv->port_id;
desc = (gxio_mpipe_edesc_t) { {
.va = rte_pktmbuf_mtod(mbuf, uintptr_t),
.xfer_size = rte_pktmbuf_data_len(mbuf),
.bound = next ? 0 : 1,
.stack_idx = mpipe_mbuf_stack_index(priv, mbuf),
} };
if (mpipe_local.mbuf_push_debt[port_id] > 0) {
mpipe_local.mbuf_push_debt[port_id]--;
desc.hwb = 1;
priv->tx_comps[idx] = NULL;
} else
priv->tx_comps[idx] = mbuf;
nb_bytes += mbuf->data_len;
priv->tx_comps[idx] = mbuf;
gxio_mpipe_equeue_put_at(equeue, desc, slot + i);
PMD_DEBUG_TX("%s:%d: Sending packet %p, len %d\n",
@ -1443,17 +1467,22 @@ mpipe_do_recv(struct mpipe_rx_queue *rx_queue, struct rte_mbuf **rx_pkts,
continue;
}
mbuf = __rte_mbuf_raw_alloc(priv->rx_mpool);
if (unlikely(!mbuf)) {
nb_nomem++;
gxio_mpipe_iqueue_drop(iqueue, idesc);
PMD_DEBUG_RX("%s:%d: RX alloc failure\n",
if (mpipe_local.mbuf_push_debt[in_port] <
MPIPE_BUF_DEBT_THRESHOLD)
mpipe_local.mbuf_push_debt[in_port]++;
else {
mbuf = __rte_mbuf_raw_alloc(priv->rx_mpool);
if (unlikely(!mbuf)) {
nb_nomem++;
gxio_mpipe_iqueue_drop(iqueue, idesc);
PMD_DEBUG_RX("%s:%d: alloc failure\n",
mpipe_name(rx_queue->q.priv),
rx_queue->q.queue_idx);
continue;
}
continue;
}
mpipe_recv_push(priv, mbuf);
mpipe_recv_push(priv, mbuf);
}
/* Get and setup the mbuf for the received packet. */
mbuf = mpipe_recv_mbuf(priv, idesc, in_port);