return 0;
}
+#define QEDE_MAX_BULK_ALLOC_COUNT 512
+
+static inline int qede_alloc_rx_bulk_mbufs(struct qede_rx_queue *rxq, int count)
+{
+ void *obj_p[QEDE_MAX_BULK_ALLOC_COUNT] __rte_cache_aligned;
+ struct rte_mbuf *mbuf = NULL;
+ struct eth_rx_bd *rx_bd;
+ dma_addr_t mapping;
+ int i, ret = 0;
+ uint16_t idx;
+
+ idx = rxq->sw_rx_prod & NUM_RX_BDS(rxq);
+
+ if (count > QEDE_MAX_BULK_ALLOC_COUNT)
+ count = QEDE_MAX_BULK_ALLOC_COUNT;
+
+ ret = rte_mempool_get_bulk(rxq->mb_pool, obj_p, count);
+ if (unlikely(ret)) {
+ PMD_RX_LOG(ERR, rxq,
+ "Failed to allocate %d rx buffers "
+ "sw_rx_prod %u sw_rx_cons %u mp entries %u free %u",
+ count, idx, rxq->sw_rx_cons & NUM_RX_BDS(rxq),
+ rte_mempool_avail_count(rxq->mb_pool),
+ rte_mempool_in_use_count(rxq->mb_pool));
+ return -ENOMEM;
+ }
+
+ for (i = 0; i < count; i++) {
+ mbuf = obj_p[i];
+ if (likely(i < count - 1))
+ rte_prefetch0(obj_p[i + 1]);
+
+ idx = rxq->sw_rx_prod & NUM_RX_BDS(rxq);
+ rxq->sw_rx_ring[idx].mbuf = mbuf;
+ rxq->sw_rx_ring[idx].page_offset = 0;
+ mapping = rte_mbuf_data_iova_default(mbuf);
+ rx_bd = (struct eth_rx_bd *)
+ ecore_chain_produce(&rxq->rx_bd_ring);
+ rx_bd->addr.hi = rte_cpu_to_le_32(U64_HI(mapping));
+ rx_bd->addr.lo = rte_cpu_to_le_32(U64_LO(mapping));
+ rxq->sw_rx_prod++;
+ }
+
+ return 0;
+}
+
+/* Criterias for calculating Rx buffer size -
+ * 1) rx_buf_size should not exceed the size of mbuf
+ * 2) In scattered_rx mode - minimum rx_buf_size should be
+ * (MTU + Maximum L2 Header Size + 2) / ETH_RX_MAX_BUFF_PER_PKT
+ * 3) In regular mode - minimum rx_buf_size should be
+ * (MTU + Maximum L2 Header Size + 2)
+ * In above cases +2 corrosponds to 2 bytes padding in front of L2
+ * header.
+ * 4) rx_buf_size should be cacheline-size aligned. So considering
+ * criteria 1, we need to adjust the size to floor instead of ceil,
+ * so that we don't exceed mbuf size while ceiling rx_buf_size.
+ */
+int
+qede_calc_rx_buf_size(struct rte_eth_dev *dev, uint16_t mbufsz,
+ uint16_t max_frame_size)
+{
+ struct qede_dev *qdev = QEDE_INIT_QDEV(dev);
+ struct ecore_dev *edev = QEDE_INIT_EDEV(qdev);
+ int rx_buf_size;
+
+ if (dev->data->scattered_rx) {
+ /* per HW limitation, only ETH_RX_MAX_BUFF_PER_PKT number of
+ * bufferes can be used for single packet. So need to make sure
+ * mbuf size is sufficient enough for this.
+ */
+ if ((mbufsz * ETH_RX_MAX_BUFF_PER_PKT) <
+ (max_frame_size + QEDE_ETH_OVERHEAD)) {
+ DP_ERR(edev, "mbuf %d size is not enough to hold max fragments (%d) for max rx packet length (%d)\n",
+ mbufsz, ETH_RX_MAX_BUFF_PER_PKT, max_frame_size);
+ return -EINVAL;
+ }
+
+ rx_buf_size = RTE_MAX(mbufsz,
+ (max_frame_size + QEDE_ETH_OVERHEAD) /
+ ETH_RX_MAX_BUFF_PER_PKT);
+ } else {
+ rx_buf_size = max_frame_size + QEDE_ETH_OVERHEAD;
+ }
+
+ /* Align to cache-line size if needed */
+ return QEDE_FLOOR_TO_CACHE_LINE_SIZE(rx_buf_size);
+}
+
int
qede_rx_queue_setup(struct rte_eth_dev *dev, uint16_t queue_idx,
uint16_t nb_desc, unsigned int socket_id,
/* Fix up RX buffer size */
bufsz = (uint16_t)rte_pktmbuf_data_room_size(mp) - RTE_PKTMBUF_HEADROOM;
+ /* cache align the mbuf size to simplfy rx_buf_size calculation */
+ bufsz = QEDE_FLOOR_TO_CACHE_LINE_SIZE(bufsz);
if ((rxmode->offloads & DEV_RX_OFFLOAD_SCATTER) ||
(max_rx_pkt_len + QEDE_ETH_OVERHEAD) > bufsz) {
if (!dev->data->scattered_rx) {
}
}
- if (dev->data->scattered_rx)
- rxq->rx_buf_size = bufsz + ETHER_HDR_LEN +
- ETHER_CRC_LEN + QEDE_ETH_OVERHEAD;
- else
- rxq->rx_buf_size = max_rx_pkt_len + QEDE_ETH_OVERHEAD;
- /* Align to cache-line size if needed */
- rxq->rx_buf_size = QEDE_CEIL_TO_CACHE_LINE_SIZE(rxq->rx_buf_size);
+ rc = qede_calc_rx_buf_size(dev, bufsz, max_rx_pkt_len);
+ if (rc < 0) {
+ rte_free(rxq);
+ return rc;
+ }
+
+ rxq->rx_buf_size = rc;
DP_INFO(edev, "mtu %u mbufsz %u bd_max_bytes %u scatter_mode %d\n",
qdev->mtu, bufsz, rxq->rx_buf_size, dev->data->scattered_rx);
void qede_rx_queue_release(void *rx_queue)
{
struct qede_rx_queue *rxq = rx_queue;
- struct qede_dev *qdev = rxq->qdev;
- struct ecore_dev *edev = QEDE_INIT_EDEV(qdev);
-
- PMD_INIT_FUNC_TRACE(edev);
+ struct qede_dev *qdev;
+ struct ecore_dev *edev;
if (rxq) {
+ qdev = rxq->qdev;
+ edev = QEDE_INIT_EDEV(qdev);
+ PMD_INIT_FUNC_TRACE(edev);
qede_rx_queue_release_mbufs(rxq);
qdev->ops->common->chain_free(edev, &rxq->rx_bd_ring);
qdev->ops->common->chain_free(edev, &rxq->rx_comp_ring);
void qede_tx_queue_release(void *tx_queue)
{
struct qede_tx_queue *txq = tx_queue;
- struct qede_dev *qdev = txq->qdev;
- struct ecore_dev *edev = QEDE_INIT_EDEV(qdev);
-
- PMD_INIT_FUNC_TRACE(edev);
+ struct qede_dev *qdev;
+ struct ecore_dev *edev;
if (txq) {
+ qdev = txq->qdev;
+ edev = QEDE_INIT_EDEV(qdev);
+ PMD_INIT_FUNC_TRACE(edev);
qede_tx_queue_release_mbufs(txq);
qdev->ops->common->chain_free(edev, &txq->tx_pbl);
rte_free(txq->sw_tx_ring);
static inline uint32_t qede_rx_cqe_to_pkt_type_outer(struct rte_mbuf *m)
{
uint32_t packet_type = RTE_PTYPE_UNKNOWN;
- struct ether_hdr *eth_hdr;
+ struct rte_ether_hdr *eth_hdr;
struct ipv4_hdr *ipv4_hdr;
struct ipv6_hdr *ipv6_hdr;
- struct vlan_hdr *vlan_hdr;
+ struct rte_vlan_hdr *vlan_hdr;
uint16_t ethertype;
bool vlan_tagged = 0;
uint16_t len;
- eth_hdr = rte_pktmbuf_mtod(m, struct ether_hdr *);
- len = sizeof(struct ether_hdr);
+ eth_hdr = rte_pktmbuf_mtod(m, struct rte_ether_hdr *);
+ len = sizeof(struct rte_ether_hdr);
ethertype = rte_cpu_to_be_16(eth_hdr->ether_type);
/* Note: Valid only if VLAN stripping is disabled */
if (ethertype == ETHER_TYPE_VLAN) {
vlan_tagged = 1;
- vlan_hdr = (struct vlan_hdr *)(eth_hdr + 1);
- len += sizeof(struct vlan_hdr);
+ vlan_hdr = (struct rte_vlan_hdr *)(eth_hdr + 1);
+ len += sizeof(struct rte_vlan_hdr);
ethertype = rte_cpu_to_be_16(vlan_hdr->eth_proto);
}
m->packet_type = qede_rx_cqe_to_pkt_type(flag);
if (RTE_ETH_IS_IPV4_HDR(m->packet_type)) {
ip = rte_pktmbuf_mtod_offset(m, struct ipv4_hdr *,
- sizeof(struct ether_hdr));
+ sizeof(struct rte_ether_hdr));
pkt_csum = ip->hdr_checksum;
ip->hdr_checksum = 0;
calc_csum = rte_ipv4_cksum(ip);
struct qede_rx_queue *rxq, struct qede_rx_entry *curr_cons)
{
struct eth_rx_bd *rx_bd_prod = ecore_chain_produce(&rxq->rx_bd_ring);
- uint16_t idx = rxq->sw_rx_cons & NUM_RX_BDS(rxq);
+ uint16_t idx = rxq->sw_rx_prod & NUM_RX_BDS(rxq);
struct qede_rx_entry *curr_prod;
dma_addr_t new_mapping;
uint8_t bitfield_val;
#endif
uint8_t tunn_parse_flag;
- uint8_t j;
struct eth_fast_path_rx_tpa_start_cqe *cqe_start_tpa;
uint64_t ol_flags;
uint32_t packet_type;
uint8_t offset, tpa_agg_idx, flags;
struct qede_agg_info *tpa_info = NULL;
uint32_t rss_hash;
+ int rx_alloc_count = 0;
+
+
+ /* Allocate buffers that we used in previous loop */
+ if (rxq->rx_alloc_count) {
+ if (unlikely(qede_alloc_rx_bulk_mbufs(rxq,
+ rxq->rx_alloc_count))) {
+ struct rte_eth_dev *dev;
+
+ PMD_RX_LOG(ERR, rxq,
+ "New buffer allocation failed,"
+ "dropping incoming packetn");
+ dev = &rte_eth_devices[rxq->port_id];
+ dev->data->rx_mbuf_alloc_failed +=
+ rxq->rx_alloc_count;
+ rxq->rx_alloc_errors += rxq->rx_alloc_count;
+ return 0;
+ }
+ qede_update_rx_prod(qdev, rxq);
+ rxq->rx_alloc_count = 0;
+ }
hw_comp_cons = rte_le_to_cpu_16(*rxq->hw_cons_ptr);
sw_comp_cons = ecore_chain_get_cons_idx(&rxq->rx_comp_ring);
rx_mb->hash.rss = rss_hash;
}
- if (unlikely(qede_alloc_rx_buffer(rxq) != 0)) {
- PMD_RX_LOG(ERR, rxq,
- "New buffer allocation failed,"
- "dropping incoming packet\n");
- qede_recycle_rx_bd_ring(rxq, qdev, fp_cqe->bd_num);
- rte_eth_devices[rxq->port_id].
- data->rx_mbuf_alloc_failed++;
- rxq->rx_alloc_errors++;
- break;
- }
+ rx_alloc_count++;
qede_rx_bd_ring_consume(rxq);
if (!tpa_start_flg && fp_cqe->bd_num > 1) {
if (qede_process_sg_pkts(p_rxq, seg1, num_segs,
pkt_len - len))
goto next_cqe;
- for (j = 0; j < num_segs; j++) {
- if (qede_alloc_rx_buffer(rxq)) {
- PMD_RX_LOG(ERR, rxq,
- "Buffer allocation failed");
- rte_eth_devices[rxq->port_id].
- data->rx_mbuf_alloc_failed++;
- rxq->rx_alloc_errors++;
- break;
- }
- rxq->rx_segs++;
- }
+
+ rx_alloc_count += num_segs;
+ rxq->rx_segs += num_segs;
}
rxq->rx_segs++; /* for the first segment */
}
}
- qede_update_rx_prod(qdev, rxq);
+ /* Request number of bufferes to be allocated in next loop */
+ rxq->rx_alloc_count = rx_alloc_count;
rxq->rcv_pkts += rx_pkt;
}
}
if (ol_flags & QEDE_TX_OFFLOAD_NOTSUP_MASK) {
+ /* We support only limited tunnel protocols */
+ if (ol_flags & PKT_TX_TUNNEL_MASK) {
+ uint64_t temp;
+
+ temp = ol_flags & PKT_TX_TUNNEL_MASK;
+ if (temp == PKT_TX_TUNNEL_VXLAN ||
+ temp == PKT_TX_TUNNEL_GENEVE ||
+ temp == PKT_TX_TUNNEL_MPLSINUDP ||
+ temp == PKT_TX_TUNNEL_GRE)
+ continue;
+ }
+
rte_errno = -ENOTSUP;
break;
}
txq->nb_tx_avail -= bd1->data.nbds;
txq->sw_tx_prod++;
- rte_prefetch0(txq->sw_tx_ring[TX_PROD(txq)].mbuf);
bd_prod =
rte_cpu_to_le_16(ecore_chain_get_prod_idx(&txq->tx_pbl));
#ifdef RTE_LIBRTE_QEDE_DEBUG_TX
{
return 0;
}
+
+
+/* this function does a fake walk through over completion queue
+ * to calculate number of BDs used by HW.
+ * At the end, it restores the state of completion queue.
+ */
+static uint16_t
+qede_parse_fp_cqe(struct qede_rx_queue *rxq)
+{
+ uint16_t hw_comp_cons, sw_comp_cons, bd_count = 0;
+ union eth_rx_cqe *cqe, *orig_cqe = NULL;
+
+ hw_comp_cons = rte_le_to_cpu_16(*rxq->hw_cons_ptr);
+ sw_comp_cons = ecore_chain_get_cons_idx(&rxq->rx_comp_ring);
+
+ if (hw_comp_cons == sw_comp_cons)
+ return 0;
+
+ /* Get the CQE from the completion ring */
+ cqe = (union eth_rx_cqe *)ecore_chain_consume(&rxq->rx_comp_ring);
+ orig_cqe = cqe;
+
+ while (sw_comp_cons != hw_comp_cons) {
+ switch (cqe->fast_path_regular.type) {
+ case ETH_RX_CQE_TYPE_REGULAR:
+ bd_count += cqe->fast_path_regular.bd_num;
+ break;
+ case ETH_RX_CQE_TYPE_TPA_END:
+ bd_count += cqe->fast_path_tpa_end.num_of_bds;
+ break;
+ default:
+ break;
+ }
+
+ cqe =
+ (union eth_rx_cqe *)ecore_chain_consume(&rxq->rx_comp_ring);
+ sw_comp_cons = ecore_chain_get_cons_idx(&rxq->rx_comp_ring);
+ }
+
+ /* revert comp_ring to original state */
+ ecore_chain_set_cons(&rxq->rx_comp_ring, sw_comp_cons, orig_cqe);
+
+ return bd_count;
+}
+
+int
+qede_rx_descriptor_status(void *p_rxq, uint16_t offset)
+{
+ uint16_t hw_bd_cons, sw_bd_cons, sw_bd_prod;
+ uint16_t produced, consumed;
+ struct qede_rx_queue *rxq = p_rxq;
+
+ if (offset > rxq->nb_rx_desc)
+ return -EINVAL;
+
+ sw_bd_cons = ecore_chain_get_cons_idx(&rxq->rx_bd_ring);
+ sw_bd_prod = ecore_chain_get_prod_idx(&rxq->rx_bd_ring);
+
+ /* find BDs used by HW from completion queue elements */
+ hw_bd_cons = sw_bd_cons + qede_parse_fp_cqe(rxq);
+
+ if (hw_bd_cons < sw_bd_cons)
+ /* wraparound case */
+ consumed = (0xffff - sw_bd_cons) + hw_bd_cons;
+ else
+ consumed = hw_bd_cons - sw_bd_cons;
+
+ if (offset <= consumed)
+ return RTE_ETH_RX_DESC_DONE;
+
+ if (sw_bd_prod < sw_bd_cons)
+ /* wraparound case */
+ produced = (0xffff - sw_bd_cons) + sw_bd_prod;
+ else
+ produced = sw_bd_prod - sw_bd_cons;
+
+ if (offset <= produced)
+ return RTE_ETH_RX_DESC_AVAIL;
+
+ return RTE_ETH_RX_DESC_UNAVAIL;
+}
git.droids-corp.org / dpdk.git / blobdiff