return;
addr = offset[gl_idx] + queue_id * HNS3_TQP_INTR_REG_SIZE;
- value = HNS3_GL_USEC_TO_REG(gl_value);
+ if (hw->intr.gl_unit == HNS3_INTR_COALESCE_GL_UINT_1US)
+ value = gl_value | HNS3_TQP_INTR_GL_UNIT_1US;
+ else
+ value = HNS3_GL_USEC_TO_REG(gl_value);
hns3_write_dev(hw, addr, value);
}
hns3_write_dev(hw, addr, value);
}
+void
+hns3_set_queue_intr_ql(struct hns3_hw *hw, uint16_t queue_id, uint16_t ql_value)
+{
+ uint32_t addr;
+
+ if (hw->intr.coalesce_mode == HNS3_INTR_COALESCE_NON_QL)
+ return;
+
+ addr = HNS3_TQP_INTR_TX_QL_REG + queue_id * HNS3_TQP_INTR_REG_SIZE;
+ hns3_write_dev(hw, addr, ql_value);
+
+ addr = HNS3_TQP_INTR_RX_QL_REG + queue_id * HNS3_TQP_INTR_REG_SIZE;
+ hns3_write_dev(hw, addr, ql_value);
+}
+
static void
hns3_queue_intr_enable(struct hns3_hw *hw, uint16_t queue_id, bool en)
{
return -EINVAL;
}
+ if (conf->rx_drop_en == 0)
+ hns3_warn(hw, "if there are no available Rx descriptors,"
+ "incoming packets are always dropped. input parameter"
+ " conf->rx_drop_en(%u) is uneffective.",
+ conf->rx_drop_en);
+
if (dev->data->rx_queues[idx]) {
hns3_rx_queue_release(dev->data->rx_queues[idx]);
dev->data->rx_queues[idx] = NULL;
rxq->ol3_csum_erros = 0;
rxq->ol4_csum_erros = 0;
+ /* CRC len set here is used for amending packet length */
+ if (dev->data->dev_conf.rxmode.offloads & DEV_RX_OFFLOAD_KEEP_CRC)
+ rxq->crc_len = RTE_ETHER_CRC_LEN;
+ else
+ rxq->crc_len = 0;
+
rte_spinlock_lock(&hw->lock);
dev->data->rx_queues[idx] = rxq;
rte_spinlock_unlock(&hw->lock);
}
}
+static inline void
+recalculate_data_len(struct rte_mbuf *first_seg, struct rte_mbuf *last_seg,
+ struct rte_mbuf *rxm, struct hns3_rx_queue *rxq,
+ uint16_t data_len)
+{
+ uint8_t crc_len = rxq->crc_len;
+
+ if (data_len <= crc_len) {
+ rte_pktmbuf_free_seg(rxm);
+ first_seg->nb_segs--;
+ last_seg->data_len = (uint16_t)(last_seg->data_len -
+ (crc_len - data_len));
+ last_seg->next = NULL;
+ } else
+ rxm->data_len = (uint16_t)(data_len - crc_len);
+}
+
uint16_t
hns3_recv_pkts(void *rx_queue, struct rte_mbuf **rx_pkts, uint16_t nb_pkts)
{
rxdp->rx.bd_base_info = 0;
rxdp->addr = dma_addr;
- /* Load remained descriptor data and extract necessary fields */
+ /*
+ * Load remained descriptor data and extract necessary fields.
+ * Data size from buffer description may contains CRC len,
+ * packet len should subtract it.
+ */
data_len = (uint16_t)(rte_le_to_cpu_16(rxd.rx.size));
l234_info = rte_le_to_cpu_32(rxd.rx.l234_info);
ol_info = rte_le_to_cpu_32(rxd.rx.ol_info);
continue;
}
- /* The last buffer of the received packet */
+ /*
+ * The last buffer of the received packet. packet len from
+ * buffer description may contains CRC len, packet len should
+ * subtract it, same as data len.
+ */
pkt_len = (uint16_t)(rte_le_to_cpu_16(rxd.rx.pkt_len));
first_seg->pkt_len = pkt_len;
+
+ /*
+ * 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 -= rxq->crc_len;
+ recalculate_data_len(first_seg, last_seg, rxm, rxq,
+ data_len);
+ }
+
first_seg->port = rxq->port_id;
first_seg->hash.rss = rte_le_to_cpu_32(rxd.rx.rss_hash);
first_seg->ol_flags = PKT_RX_RSS_HASH;
rte_le_to_cpu_32(rxd.rx.fd_id);
first_seg->ol_flags |= PKT_RX_FDIR | PKT_RX_FDIR_ID;
}
- rxm->next = NULL;
gro_size = hns3_get_field(bd_base_info, HNS3_RXD_GRO_SIZE_M,
HNS3_RXD_GRO_SIZE_S);
txq->configured = true;
txq->io_base = (void *)((char *)hw->io_base + HNS3_TQP_REG_OFFSET +
idx * HNS3_TQP_REG_SIZE);
+ txq->min_tx_pkt_len = hw->min_tx_pkt_len;
txq->over_length_pkt_cnt = 0;
txq->exceed_limit_bd_pkt_cnt = 0;
txq->exceed_limit_bd_reassem_fail = 0;
}
static void
-hns3_set_tso(struct hns3_desc *desc,
- uint64_t ol_flags, struct rte_mbuf *rxm)
+hns3_set_tso(struct hns3_desc *desc, uint64_t ol_flags,
+ uint32_t paylen, struct rte_mbuf *rxm)
{
- uint32_t paylen, hdr_len;
- uint32_t tmp;
uint8_t l2_len = rxm->l2_len;
+ uint32_t tmp;
if (!hns3_pkt_is_tso(rxm))
return;
if (hns3_tso_proc_tunnel(desc, ol_flags, rxm, &l2_len))
return;
- hdr_len = rxm->l2_len + rxm->l3_len + rxm->l4_len;
- hdr_len += (ol_flags & PKT_TX_TUNNEL_MASK) ?
- rxm->outer_l2_len + rxm->outer_l3_len : 0;
- paylen = rxm->pkt_len - hdr_len;
if (paylen <= rxm->tso_segsz)
return;
desc->tx.mss = rte_cpu_to_le_16(rxm->tso_segsz);
}
+static inline void
+hns3_fill_per_desc(struct hns3_desc *desc, struct rte_mbuf *rxm)
+{
+ desc->addr = rte_mbuf_data_iova(rxm);
+ desc->tx.send_size = rte_cpu_to_le_16(rte_pktmbuf_data_len(rxm));
+ desc->tx.tp_fe_sc_vld_ra_ri = rte_cpu_to_le_16(BIT(HNS3_TXD_VLD_B));
+}
+
static void
-fill_desc(struct hns3_tx_queue *txq, uint16_t tx_desc_id, struct rte_mbuf *rxm,
- bool first, int offset)
+hns3_fill_first_desc(struct hns3_tx_queue *txq, struct hns3_desc *desc,
+ struct rte_mbuf *rxm)
{
- struct hns3_desc *tx_ring = txq->tx_ring;
- struct hns3_desc *desc = &tx_ring[tx_desc_id];
- uint8_t frag_end = rxm->next == NULL ? 1 : 0;
uint64_t ol_flags = rxm->ol_flags;
- uint16_t size = rxm->data_len;
- uint16_t rrcfv = 0;
uint32_t hdr_len;
uint32_t paylen;
- uint32_t tmp;
- desc->addr = rte_mbuf_data_iova(rxm) + offset;
- desc->tx.send_size = rte_cpu_to_le_16(size);
- hns3_set_bit(rrcfv, HNS3_TXD_VLD_B, 1);
-
- if (first) {
- hdr_len = rxm->l2_len + rxm->l3_len + rxm->l4_len;
- hdr_len += (ol_flags & PKT_TX_TUNNEL_MASK) ?
+ hdr_len = rxm->l2_len + rxm->l3_len + rxm->l4_len;
+ hdr_len += (ol_flags & PKT_TX_TUNNEL_MASK) ?
rxm->outer_l2_len + rxm->outer_l3_len : 0;
- paylen = rxm->pkt_len - hdr_len;
- desc->tx.paylen = rte_cpu_to_le_32(paylen);
- hns3_set_tso(desc, ol_flags, rxm);
- }
-
- hns3_set_bit(rrcfv, HNS3_TXD_FE_B, frag_end);
- desc->tx.tp_fe_sc_vld_ra_ri = rte_cpu_to_le_16(rrcfv);
-
- if (frag_end) {
- if (ol_flags & (PKT_TX_VLAN_PKT | PKT_TX_QINQ_PKT)) {
- tmp = rte_le_to_cpu_32(desc->tx.type_cs_vlan_tso_len);
- hns3_set_bit(tmp, HNS3_TXD_VLAN_B, 1);
- desc->tx.type_cs_vlan_tso_len = rte_cpu_to_le_32(tmp);
- desc->tx.vlan_tag = rte_cpu_to_le_16(rxm->vlan_tci);
- }
+ paylen = rxm->pkt_len - hdr_len;
+ desc->tx.paylen = rte_cpu_to_le_32(paylen);
+ hns3_set_tso(desc, ol_flags, paylen, rxm);
- if (ol_flags & PKT_TX_QINQ_PKT) {
- tmp = rte_le_to_cpu_32(desc->tx.ol_type_vlan_len_msec);
- hns3_set_bit(tmp, HNS3_TXD_OVLAN_B, 1);
- desc->tx.ol_type_vlan_len_msec = rte_cpu_to_le_32(tmp);
+ /*
+ * Currently, hardware doesn't support more than two layers VLAN offload
+ * in Tx direction based on hns3 network engine. So when the number of
+ * VLANs in the packets represented by rxm plus the number of VLAN
+ * offload by hardware such as PVID etc, exceeds two, the packets will
+ * be discarded or the original VLAN of the packets will be overwitted
+ * by hardware. When the PF PVID is enabled by calling the API function
+ * named rte_eth_dev_set_vlan_pvid or the VF PVID is enabled by the hns3
+ * PF kernel ether driver, the outer VLAN tag will always be the PVID.
+ * To avoid the VLAN of Tx descriptor is overwritten by PVID, it should
+ * be added to the position close to the IP header when PVID is enabled.
+ */
+ if (!txq->pvid_state && ol_flags & (PKT_TX_VLAN_PKT |
+ PKT_TX_QINQ_PKT)) {
+ desc->tx.ol_type_vlan_len_msec |=
+ rte_cpu_to_le_32(BIT(HNS3_TXD_OVLAN_B));
+ if (ol_flags & PKT_TX_QINQ_PKT)
desc->tx.outer_vlan_tag =
- rte_cpu_to_le_16(rxm->vlan_tci_outer);
- }
+ rte_cpu_to_le_16(rxm->vlan_tci_outer);
+ else
+ desc->tx.outer_vlan_tag =
+ rte_cpu_to_le_16(rxm->vlan_tci);
+ }
+
+ if (ol_flags & PKT_TX_QINQ_PKT ||
+ ((ol_flags & PKT_TX_VLAN_PKT) && txq->pvid_state)) {
+ desc->tx.type_cs_vlan_tso_len |=
+ rte_cpu_to_le_32(BIT(HNS3_TXD_VLAN_B));
+ desc->tx.vlan_tag = rte_cpu_to_le_16(rxm->vlan_tci);
}
}
return 0;
}
+static inline void
+hns3_pktmbuf_copy_hdr(struct rte_mbuf *new_pkt, struct rte_mbuf *old_pkt)
+{
+ new_pkt->ol_flags = old_pkt->ol_flags;
+ new_pkt->pkt_len = rte_pktmbuf_pkt_len(old_pkt);
+ new_pkt->outer_l2_len = old_pkt->outer_l2_len;
+ new_pkt->outer_l3_len = old_pkt->outer_l3_len;
+ new_pkt->l2_len = old_pkt->l2_len;
+ new_pkt->l3_len = old_pkt->l3_len;
+ new_pkt->l4_len = old_pkt->l4_len;
+ new_pkt->vlan_tci_outer = old_pkt->vlan_tci_outer;
+ new_pkt->vlan_tci = old_pkt->vlan_tci;
+}
+
static int
hns3_reassemble_tx_pkts(void *tx_queue, struct rte_mbuf *tx_pkt,
struct rte_mbuf **new_pkt)
mb_pool = tx_pkt->pool;
buf_size = tx_pkt->buf_len - RTE_PKTMBUF_HEADROOM;
- nb_new_buf = (tx_pkt->pkt_len - 1) / buf_size + 1;
+ nb_new_buf = (rte_pktmbuf_pkt_len(tx_pkt) - 1) / buf_size + 1;
+ if (nb_new_buf > HNS3_MAX_NON_TSO_BD_PER_PKT)
+ return -EINVAL;
- last_buf_len = tx_pkt->pkt_len % buf_size;
+ last_buf_len = rte_pktmbuf_pkt_len(tx_pkt) % buf_size;
if (last_buf_len == 0)
last_buf_len = buf_size;
/* Copy the original packet content to the new mbufs */
temp = tx_pkt;
s = rte_pktmbuf_mtod(temp, char *);
- len_s = temp->data_len;
+ len_s = rte_pktmbuf_data_len(temp);
temp_new = new_mbuf;
for (i = 0; i < nb_new_buf; i++) {
d = rte_pktmbuf_mtod(temp_new, char *);
if (temp == NULL)
break;
s = rte_pktmbuf_mtod(temp, char *);
- len_s = temp->data_len;
+ len_s = rte_pktmbuf_data_len(temp);
}
}
temp_new->data_len = buf_len;
temp_new = temp_new->next;
}
+ hns3_pktmbuf_copy_hdr(new_mbuf, tx_pkt);
/* free original mbufs */
rte_pktmbuf_free(tx_pkt);
struct rte_net_hdr_lens hdr_lens = {0};
struct hns3_tx_queue *txq = tx_queue;
struct hns3_entry *tx_bak_pkt;
+ struct hns3_desc *tx_ring;
struct rte_mbuf *tx_pkt;
struct rte_mbuf *m_seg;
+ struct hns3_desc *desc;
uint32_t nb_hold = 0;
uint16_t tx_next_use;
uint16_t tx_pkt_num;
tx_next_use = txq->next_to_use;
tx_bd_max = txq->nb_tx_desc;
tx_pkt_num = nb_pkts;
+ tx_ring = txq->tx_ring;
/* send packets */
tx_bak_pkt = &txq->sw_ring[tx_next_use];
}
/*
- * If packet length is less than minimum packet size, driver
- * need to pad it.
+ * If packet length is less than minimum packet length supported
+ * by hardware in Tx direction, driver need to pad it to avoid
+ * error.
*/
- if (unlikely(rte_pktmbuf_pkt_len(tx_pkt) < HNS3_MIN_PKT_SIZE)) {
+ if (unlikely(rte_pktmbuf_pkt_len(tx_pkt) <
+ txq->min_tx_pkt_len)) {
uint16_t add_len;
char *appended;
- add_len = HNS3_MIN_PKT_SIZE -
+ add_len = txq->min_tx_pkt_len -
rte_pktmbuf_pkt_len(tx_pkt);
appended = rte_pktmbuf_append(tx_pkt, add_len);
if (appended == NULL) {
goto end_of_tx;
i = 0;
+ desc = &tx_ring[tx_next_use];
+
+ /*
+ * If the packet is divided into multiple Tx Buffer Descriptors,
+ * only need to fill vlan, paylen and tso into the first Tx
+ * Buffer Descriptor.
+ */
+ hns3_fill_first_desc(txq, desc, m_seg);
+
do {
- fill_desc(txq, tx_next_use, m_seg, (i == 0), 0);
+ desc = &tx_ring[tx_next_use];
+ /*
+ * Fill valid bits, DMA address and data length for each
+ * Tx Buffer Descriptor.
+ */
+ hns3_fill_per_desc(desc, m_seg);
tx_bak_pkt->mbuf = m_seg;
m_seg = m_seg->next;
tx_next_use++;
i++;
} while (m_seg != NULL);
+ /* Add end flag for the last Tx Buffer Descriptor */
+ desc->tx.tp_fe_sc_vld_ra_ri |=
+ rte_cpu_to_le_16(BIT(HNS3_TXD_FE_B));
+
nb_hold += i;
txq->next_to_use = tx_next_use;
txq->tx_bd_ready -= i;
eth_dev->tx_pkt_prepare = hns3_dummy_rxtx_burst;
}
}
+
+void
+hns3_rxq_info_get(struct rte_eth_dev *dev, uint16_t queue_id,
+ struct rte_eth_rxq_info *qinfo)
+{
+ struct hns3_rx_queue *rxq = dev->data->rx_queues[queue_id];
+
+ qinfo->mp = rxq->mb_pool;
+ qinfo->nb_desc = rxq->nb_rx_desc;
+ qinfo->scattered_rx = dev->data->scattered_rx;
+
+ /*
+ * If there are no available Rx buffer descriptors, incoming packets
+ * are always dropped by hardware based on hns3 network engine.
+ */
+ qinfo->conf.rx_drop_en = 1;
+ qinfo->conf.offloads = dev->data->dev_conf.rxmode.offloads;
+ qinfo->conf.rx_free_thresh = rxq->rx_free_thresh;
+ qinfo->conf.rx_deferred_start = rxq->rx_deferred_start;
+}
+
+void
+hns3_txq_info_get(struct rte_eth_dev *dev, uint16_t queue_id,
+ struct rte_eth_txq_info *qinfo)
+{
+ struct hns3_tx_queue *txq = dev->data->tx_queues[queue_id];
+
+ qinfo->nb_desc = txq->nb_tx_desc;
+ qinfo->conf.offloads = dev->data->dev_conf.txmode.offloads;
+ qinfo->conf.tx_deferred_start = txq->tx_deferred_start;
+}