net/ice: support device and queue ops
authorWenzhuo Lu <wenzhuo.lu@intel.com>
Tue, 18 Dec 2018 08:46:25 +0000 (16:46 +0800)
committerFerruh Yigit <ferruh.yigit@intel.com>
Fri, 21 Dec 2018 15:22:41 +0000 (16:22 +0100)
Normally when starting/stopping the device the queue
should be started and stopped. Support them both in
this patch.

Below ops are added,
dev_configure
dev_start
dev_stop
dev_close
dev_reset
rx_queue_start
rx_queue_stop
tx_queue_start
tx_queue_stop
rx_queue_setup
rx_queue_release
tx_queue_setup
tx_queue_release

Signed-off-by: Wenzhuo Lu <wenzhuo.lu@intel.com>
Signed-off-by: Qiming Yang <qiming.yang@intel.com>
Signed-off-by: Xiaoyun Li <xiaoyun.li@intel.com>
Signed-off-by: Jingjing Wu <jingjing.wu@intel.com>
Reviewed-by: Ferruh Yigit <ferruh.yigit@intel.com>
Reviewed-by: Qi Zhang <qi.z.zhang@intel.com>
config/common_base
doc/guides/nics/features/ice.ini
doc/guides/nics/ice.rst
drivers/net/ice/Makefile
drivers/net/ice/ice_ethdev.c
drivers/net/ice/ice_rxtx.c [new file with mode: 0644]
drivers/net/ice/ice_rxtx.h [new file with mode: 0644]
drivers/net/ice/meson.build

index f23cef8..0e3f900 100644 (file)
@@ -303,6 +303,8 @@ CONFIG_RTE_LIBRTE_ICE_PMD=y
 CONFIG_RTE_LIBRTE_ICE_DEBUG_RX=n
 CONFIG_RTE_LIBRTE_ICE_DEBUG_TX=n
 CONFIG_RTE_LIBRTE_ICE_DEBUG_TX_FREE=n
+CONFIG_RTE_LIBRTE_ICE_RX_ALLOW_BULK_ALLOC=y
+CONFIG_RTE_LIBRTE_ICE_16BYTE_RX_DESC=n
 
 # Compile burst-oriented AVF PMD driver
 #
index 085e848..a43a9cd 100644 (file)
@@ -4,6 +4,7 @@
 ; Refer to default.ini for the full list of available PMD features.
 ;
 [Features]
+Queue start/stop     = Y
 BSD nic_uio          = Y
 Linux UIO            = Y
 Linux VFIO           = Y
index 946ed04..96a594f 100644 (file)
@@ -38,6 +38,14 @@ Please note that enabling debugging options may affect system performance.
 
   Toggle display of generic debugging messages.
 
+- ``CONFIG_RTE_LIBRTE_ICE_RX_ALLOW_BULK_ALLOC`` (default ``y``)
+
+  Toggle bulk allocation for RX.
+
+- ``CONFIG_RTE_LIBRTE_ICE_16BYTE_RX_DESC`` (default ``n``)
+
+  Toggle to use a 16-byte RX descriptor, by default the RX descriptor is 32 byte.
+
 Runtime Config Options
 ~~~~~~~~~~~~~~~~~~~~~~
 
index 70f23e3..bc24444 100644 (file)
@@ -11,7 +11,7 @@ LIB = librte_pmd_ice.a
 CFLAGS += -O3
 CFLAGS += $(WERROR_FLAGS)
 
-LDLIBS += -lrte_eal -lrte_ethdev -lrte_kvargs -lrte_bus_pci
+LDLIBS += -lrte_eal -lrte_ethdev -lrte_kvargs -lrte_bus_pci -lrte_mempool
 
 EXPORT_MAP := rte_pmd_ice_version.map
 
@@ -50,5 +50,6 @@ SRCS-$(CONFIG_RTE_LIBRTE_ICE_PMD) += ice_switch.c
 SRCS-$(CONFIG_RTE_LIBRTE_ICE_PMD) += ice_nvm.c
 
 SRCS-$(CONFIG_RTE_LIBRTE_ICE_PMD) += ice_ethdev.c
+SRCS-$(CONFIG_RTE_LIBRTE_ICE_PMD) += ice_rxtx.c
 
 include $(RTE_SDK)/mk/rte.lib.mk
index a514755..0274d9e 100644 (file)
@@ -6,6 +6,7 @@
 
 #include "base/ice_sched.h"
 #include "ice_ethdev.h"
+#include "ice_rxtx.h"
 
 #define ICE_MAX_QP_NUM "max_queue_pair_num"
 #define ICE_DFLT_OUTER_TAG_TYPE ICE_AQ_VSI_OUTER_TAG_VLAN_9100
 int ice_logtype_init;
 int ice_logtype_driver;
 
+static int ice_dev_configure(struct rte_eth_dev *dev);
+static int ice_dev_start(struct rte_eth_dev *dev);
+static void ice_dev_stop(struct rte_eth_dev *dev);
+static void ice_dev_close(struct rte_eth_dev *dev);
+static int ice_dev_reset(struct rte_eth_dev *dev);
+
 static const struct rte_pci_id pci_id_ice_map[] = {
        { RTE_PCI_DEVICE(ICE_INTEL_VENDOR_ID, ICE_DEV_ID_E810C_BACKPLANE) },
        { RTE_PCI_DEVICE(ICE_INTEL_VENDOR_ID, ICE_DEV_ID_E810C_QSFP) },
@@ -21,7 +28,19 @@ static const struct rte_pci_id pci_id_ice_map[] = {
 };
 
 static const struct eth_dev_ops ice_eth_dev_ops = {
-       .dev_configure                = NULL,
+       .dev_configure                = ice_dev_configure,
+       .dev_start                    = ice_dev_start,
+       .dev_stop                     = ice_dev_stop,
+       .dev_close                    = ice_dev_close,
+       .dev_reset                    = ice_dev_reset,
+       .rx_queue_start               = ice_rx_queue_start,
+       .rx_queue_stop                = ice_rx_queue_stop,
+       .tx_queue_start               = ice_tx_queue_start,
+       .tx_queue_stop                = ice_tx_queue_stop,
+       .rx_queue_setup               = ice_rx_queue_setup,
+       .rx_queue_release             = ice_rx_queue_release,
+       .tx_queue_setup               = ice_tx_queue_setup,
+       .tx_queue_release             = ice_tx_queue_release,
 };
 
 static void
@@ -558,12 +577,42 @@ ice_release_vsi(struct ice_vsi *vsi)
        return 0;
 }
 
+static void
+ice_dev_stop(struct rte_eth_dev *dev)
+{
+       struct rte_eth_dev_data *data = dev->data;
+       struct ice_pf *pf = ICE_DEV_PRIVATE_TO_PF(dev->data->dev_private);
+       uint16_t i;
+
+       /* avoid stopping again */
+       if (pf->adapter_stopped)
+               return;
+
+       /* stop and clear all Rx queues */
+       for (i = 0; i < data->nb_rx_queues; i++)
+               ice_rx_queue_stop(dev, i);
+
+       /* stop and clear all Tx queues */
+       for (i = 0; i < data->nb_tx_queues; i++)
+               ice_tx_queue_stop(dev, i);
+
+       /* Clear all queues and release mbufs */
+       ice_clear_queues(dev);
+
+       pf->adapter_stopped = true;
+}
+
 static void
 ice_dev_close(struct rte_eth_dev *dev)
 {
        struct ice_pf *pf = ICE_DEV_PRIVATE_TO_PF(dev->data->dev_private);
        struct ice_hw *hw = ICE_DEV_PRIVATE_TO_HW(dev->data->dev_private);
 
+       ice_dev_stop(dev);
+
+       /* release all queue resource */
+       ice_free_queues(dev);
+
        ice_res_pool_destroy(&pf->msix_pool);
        ice_release_vsi(pf->main_vsi);
 
@@ -593,6 +642,154 @@ ice_dev_uninit(struct rte_eth_dev *dev)
        return 0;
 }
 
+static int
+ice_dev_configure(__rte_unused struct rte_eth_dev *dev)
+{
+       struct ice_adapter *ad =
+               ICE_DEV_PRIVATE_TO_ADAPTER(dev->data->dev_private);
+
+       /* Initialize to TRUE. If any of Rx queues doesn't meet the
+        * bulk allocation or vector Rx preconditions we will reset it.
+        */
+       ad->rx_bulk_alloc_allowed = true;
+       ad->tx_simple_allowed = true;
+
+       return 0;
+}
+
+static int ice_init_rss(struct ice_pf *pf)
+{
+       struct ice_hw *hw = ICE_PF_TO_HW(pf);
+       struct ice_vsi *vsi = pf->main_vsi;
+       struct rte_eth_dev *dev = pf->adapter->eth_dev;
+       struct rte_eth_rss_conf *rss_conf;
+       struct ice_aqc_get_set_rss_keys key;
+       uint16_t i, nb_q;
+       int ret = 0;
+
+       rss_conf = &dev->data->dev_conf.rx_adv_conf.rss_conf;
+       nb_q = dev->data->nb_rx_queues;
+       vsi->rss_key_size = ICE_AQC_GET_SET_RSS_KEY_DATA_RSS_KEY_SIZE;
+       vsi->rss_lut_size = hw->func_caps.common_cap.rss_table_size;
+
+       if (!vsi->rss_key)
+               vsi->rss_key = rte_zmalloc(NULL,
+                                          vsi->rss_key_size, 0);
+       if (!vsi->rss_lut)
+               vsi->rss_lut = rte_zmalloc(NULL,
+                                          vsi->rss_lut_size, 0);
+
+       /* configure RSS key */
+       if (!rss_conf->rss_key) {
+               /* Calculate the default hash key */
+               for (i = 0; i <= vsi->rss_key_size; i++)
+                       vsi->rss_key[i] = (uint8_t)rte_rand();
+       } else {
+               rte_memcpy(vsi->rss_key, rss_conf->rss_key,
+                          RTE_MIN(rss_conf->rss_key_len,
+                                  vsi->rss_key_size));
+       }
+       rte_memcpy(key.standard_rss_key, vsi->rss_key, vsi->rss_key_size);
+       ret = ice_aq_set_rss_key(hw, vsi->idx, &key);
+       if (ret)
+               return -EINVAL;
+
+       /* init RSS LUT table */
+       for (i = 0; i < vsi->rss_lut_size; i++)
+               vsi->rss_lut[i] = i % nb_q;
+
+       ret = ice_aq_set_rss_lut(hw, vsi->idx,
+                                ICE_AQC_GSET_RSS_LUT_TABLE_TYPE_PF,
+                                vsi->rss_lut, vsi->rss_lut_size);
+       if (ret)
+               return -EINVAL;
+
+       return 0;
+}
+
+static int
+ice_dev_start(struct rte_eth_dev *dev)
+{
+       struct rte_eth_dev_data *data = dev->data;
+       struct ice_hw *hw = ICE_DEV_PRIVATE_TO_HW(dev->data->dev_private);
+       struct ice_pf *pf = ICE_DEV_PRIVATE_TO_PF(dev->data->dev_private);
+       uint16_t nb_rxq = 0;
+       uint16_t nb_txq, i;
+       int ret;
+
+       /* program Tx queues' context in hardware */
+       for (nb_txq = 0; nb_txq < data->nb_tx_queues; nb_txq++) {
+               ret = ice_tx_queue_start(dev, nb_txq);
+               if (ret) {
+                       PMD_DRV_LOG(ERR, "fail to start Tx queue %u", nb_txq);
+                       goto tx_err;
+               }
+       }
+
+       /* program Rx queues' context in hardware*/
+       for (nb_rxq = 0; nb_rxq < data->nb_rx_queues; nb_rxq++) {
+               ret = ice_rx_queue_start(dev, nb_rxq);
+               if (ret) {
+                       PMD_DRV_LOG(ERR, "fail to start Rx queue %u", nb_rxq);
+                       goto rx_err;
+               }
+       }
+
+       ret = ice_init_rss(pf);
+       if (ret) {
+               PMD_DRV_LOG(ERR, "Failed to enable rss for PF");
+               goto rx_err;
+       }
+
+       ret = ice_aq_set_event_mask(hw, hw->port_info->lport,
+                                   ((u16)(ICE_AQ_LINK_EVENT_LINK_FAULT |
+                                    ICE_AQ_LINK_EVENT_PHY_TEMP_ALARM |
+                                    ICE_AQ_LINK_EVENT_EXCESSIVE_ERRORS |
+                                    ICE_AQ_LINK_EVENT_SIGNAL_DETECT |
+                                    ICE_AQ_LINK_EVENT_AN_COMPLETED |
+                                    ICE_AQ_LINK_EVENT_PORT_TX_SUSPENDED)),
+                                    NULL);
+       if (ret != ICE_SUCCESS)
+               PMD_DRV_LOG(WARNING, "Fail to set phy mask");
+
+       pf->adapter_stopped = false;
+
+       return 0;
+
+       /* stop the started queues if failed to start all queues */
+rx_err:
+       for (i = 0; i < nb_rxq; i++)
+               ice_rx_queue_stop(dev, i);
+tx_err:
+       for (i = 0; i < nb_txq; i++)
+               ice_tx_queue_stop(dev, i);
+
+       return -EIO;
+}
+
+static int
+ice_dev_reset(struct rte_eth_dev *dev)
+{
+       int ret;
+
+       if (dev->data->sriov.active)
+               return -ENOTSUP;
+
+       ret = ice_dev_uninit(dev);
+       if (ret) {
+               PMD_INIT_LOG(ERR, "failed to uninit device, status = %d", ret);
+               return -ENXIO;
+       }
+
+       ret = ice_dev_init(dev);
+       if (ret) {
+               PMD_INIT_LOG(ERR, "failed to init device, status = %d", ret);
+               return -ENXIO;
+       }
+
+       return 0;
+}
+
 static int
 ice_pci_probe(struct rte_pci_driver *pci_drv __rte_unused,
              struct rte_pci_device *pci_dev)
diff --git a/drivers/net/ice/ice_rxtx.c b/drivers/net/ice/ice_rxtx.c
new file mode 100644 (file)
index 0000000..9c5eee1
--- /dev/null
@@ -0,0 +1,923 @@
+/* SPDX-License-Identifier: BSD-3-Clause
+ * Copyright(c) 2018 Intel Corporation
+ */
+
+#include <rte_ethdev_driver.h>
+#include <rte_net.h>
+
+#include "ice_rxtx.h"
+
+#define ICE_TD_CMD ICE_TX_DESC_CMD_EOP
+
+#define ICE_TX_CKSUM_OFFLOAD_MASK (             \
+               PKT_TX_IP_CKSUM |                \
+               PKT_TX_L4_MASK |                 \
+               PKT_TX_TCP_SEG |                 \
+               PKT_TX_OUTER_IP_CKSUM)
+
+#define ICE_RX_ERR_BITS 0x3f
+
+static enum ice_status
+ice_program_hw_rx_queue(struct ice_rx_queue *rxq)
+{
+       struct ice_vsi *vsi = rxq->vsi;
+       struct ice_hw *hw = ICE_VSI_TO_HW(vsi);
+       struct rte_eth_dev *dev = ICE_VSI_TO_ETH_DEV(rxq->vsi);
+       struct ice_rlan_ctx rx_ctx;
+       enum ice_status err;
+       uint16_t buf_size, len;
+       struct rte_eth_rxmode *rxmode = &dev->data->dev_conf.rxmode;
+       uint32_t regval;
+
+       /**
+        * The kernel driver uses flex descriptor. It sets the register
+        * to flex descriptor mode.
+        * DPDK uses legacy descriptor. It should set the register back
+        * to the default value, then uses legacy descriptor mode.
+        */
+       regval = (0x01 << QRXFLXP_CNTXT_RXDID_PRIO_S) &
+                QRXFLXP_CNTXT_RXDID_PRIO_M;
+       ICE_WRITE_REG(hw, QRXFLXP_CNTXT(rxq->reg_idx), regval);
+
+       /* Set buffer size as the head split is disabled. */
+       buf_size = (uint16_t)(rte_pktmbuf_data_room_size(rxq->mp) -
+                             RTE_PKTMBUF_HEADROOM);
+       rxq->rx_hdr_len = 0;
+       rxq->rx_buf_len = RTE_ALIGN(buf_size, (1 << ICE_RLAN_CTX_DBUF_S));
+       len = ICE_SUPPORT_CHAIN_NUM * rxq->rx_buf_len;
+       rxq->max_pkt_len = RTE_MIN(len,
+                                  dev->data->dev_conf.rxmode.max_rx_pkt_len);
+
+       if (rxmode->offloads & DEV_RX_OFFLOAD_JUMBO_FRAME) {
+               if (rxq->max_pkt_len <= ETHER_MAX_LEN ||
+                   rxq->max_pkt_len > ICE_FRAME_SIZE_MAX) {
+                       PMD_DRV_LOG(ERR, "maximum packet length must "
+                                   "be larger than %u and smaller than %u,"
+                                   "as jumbo frame is enabled",
+                                   (uint32_t)ETHER_MAX_LEN,
+                                   (uint32_t)ICE_FRAME_SIZE_MAX);
+                       return -EINVAL;
+               }
+       } else {
+               if (rxq->max_pkt_len < ETHER_MIN_LEN ||
+                   rxq->max_pkt_len > ETHER_MAX_LEN) {
+                       PMD_DRV_LOG(ERR, "maximum packet length must be "
+                                   "larger than %u and smaller than %u, "
+                                   "as jumbo frame is disabled",
+                                   (uint32_t)ETHER_MIN_LEN,
+                                   (uint32_t)ETHER_MAX_LEN);
+                       return -EINVAL;
+               }
+       }
+
+       memset(&rx_ctx, 0, sizeof(rx_ctx));
+
+       rx_ctx.base = rxq->rx_ring_phys_addr / ICE_QUEUE_BASE_ADDR_UNIT;
+       rx_ctx.qlen = rxq->nb_rx_desc;
+       rx_ctx.dbuf = rxq->rx_buf_len >> ICE_RLAN_CTX_DBUF_S;
+       rx_ctx.hbuf = rxq->rx_hdr_len >> ICE_RLAN_CTX_HBUF_S;
+       rx_ctx.dtype = 0; /* No Header Split mode */
+#ifndef RTE_LIBRTE_ICE_16BYTE_RX_DESC
+       rx_ctx.dsize = 1; /* 32B descriptors */
+#endif
+       rx_ctx.rxmax = rxq->max_pkt_len;
+       /* TPH: Transaction Layer Packet (TLP) processing hints */
+       rx_ctx.tphrdesc_ena = 1;
+       rx_ctx.tphwdesc_ena = 1;
+       rx_ctx.tphdata_ena = 1;
+       rx_ctx.tphhead_ena = 1;
+       /* Low Receive Queue Threshold defined in 64 descriptors units.
+        * When the number of free descriptors goes below the lrxqthresh,
+        * an immediate interrupt is triggered.
+        */
+       rx_ctx.lrxqthresh = 2;
+       /*default use 32 byte descriptor, vlan tag extract to L2TAG2(1st)*/
+       rx_ctx.l2tsel = 1;
+       rx_ctx.showiv = 0;
+
+       err = ice_clear_rxq_ctx(hw, rxq->reg_idx);
+       if (err) {
+               PMD_DRV_LOG(ERR, "Failed to clear Lan Rx queue (%u) context",
+                           rxq->queue_id);
+               return -EINVAL;
+       }
+       err = ice_write_rxq_ctx(hw, &rx_ctx, rxq->reg_idx);
+       if (err) {
+               PMD_DRV_LOG(ERR, "Failed to write Lan Rx queue (%u) context",
+                           rxq->queue_id);
+               return -EINVAL;
+       }
+
+       buf_size = (uint16_t)(rte_pktmbuf_data_room_size(rxq->mp) -
+                             RTE_PKTMBUF_HEADROOM);
+
+       rxq->qrx_tail = hw->hw_addr + QRX_TAIL(rxq->reg_idx);
+
+       /* Init the Rx tail register*/
+       ICE_PCI_REG_WRITE(rxq->qrx_tail, rxq->nb_rx_desc - 1);
+
+       return 0;
+}
+
+/* Allocate mbufs for all descriptors in rx queue */
+static int
+ice_alloc_rx_queue_mbufs(struct ice_rx_queue *rxq)
+{
+       struct ice_rx_entry *rxe = rxq->sw_ring;
+       uint64_t dma_addr;
+       uint16_t i;
+
+       for (i = 0; i < rxq->nb_rx_desc; i++) {
+               volatile union ice_rx_desc *rxd;
+               struct rte_mbuf *mbuf = rte_mbuf_raw_alloc(rxq->mp);
+
+               if (unlikely(!mbuf)) {
+                       PMD_DRV_LOG(ERR, "Failed to allocate mbuf for RX");
+                       return -ENOMEM;
+               }
+
+               rte_mbuf_refcnt_set(mbuf, 1);
+               mbuf->next = NULL;
+               mbuf->data_off = RTE_PKTMBUF_HEADROOM;
+               mbuf->nb_segs = 1;
+               mbuf->port = rxq->port_id;
+
+               dma_addr =
+                       rte_cpu_to_le_64(rte_mbuf_data_iova_default(mbuf));
+
+               rxd = &rxq->rx_ring[i];
+               rxd->read.pkt_addr = dma_addr;
+               rxd->read.hdr_addr = 0;
+#ifndef RTE_LIBRTE_ICE_16BYTE_RX_DESC
+               rxd->read.rsvd1 = 0;
+               rxd->read.rsvd2 = 0;
+#endif
+               rxe[i].mbuf = mbuf;
+       }
+
+       return 0;
+}
+
+/* Free all mbufs for descriptors in rx queue */
+static void
+ice_rx_queue_release_mbufs(struct ice_rx_queue *rxq)
+{
+       uint16_t i;
+
+       if (!rxq || !rxq->sw_ring) {
+               PMD_DRV_LOG(DEBUG, "Pointer to sw_ring is NULL");
+               return;
+       }
+
+       for (i = 0; i < rxq->nb_rx_desc; i++) {
+               if (rxq->sw_ring[i].mbuf) {
+                       rte_pktmbuf_free_seg(rxq->sw_ring[i].mbuf);
+                       rxq->sw_ring[i].mbuf = NULL;
+               }
+       }
+#ifdef RTE_LIBRTE_ICE_RX_ALLOW_BULK_ALLOC
+               if (rxq->rx_nb_avail == 0)
+                       return;
+               for (i = 0; i < rxq->rx_nb_avail; i++) {
+                       struct rte_mbuf *mbuf;
+
+                       mbuf = rxq->rx_stage[rxq->rx_next_avail + i];
+                       rte_pktmbuf_free_seg(mbuf);
+               }
+               rxq->rx_nb_avail = 0;
+#endif /* RTE_LIBRTE_ICE_RX_ALLOW_BULK_ALLOC */
+}
+
+/* turn on or off rx queue
+ * @q_idx: queue index in pf scope
+ * @on: turn on or off the queue
+ */
+static int
+ice_switch_rx_queue(struct ice_hw *hw, uint16_t q_idx, bool on)
+{
+       uint32_t reg;
+       uint16_t j;
+
+       /* QRX_CTRL = QRX_ENA */
+       reg = ICE_READ_REG(hw, QRX_CTRL(q_idx));
+
+       if (on) {
+               if (reg & QRX_CTRL_QENA_STAT_M)
+                       return 0; /* Already on, skip */
+               reg |= QRX_CTRL_QENA_REQ_M;
+       } else {
+               if (!(reg & QRX_CTRL_QENA_STAT_M))
+                       return 0; /* Already off, skip */
+               reg &= ~QRX_CTRL_QENA_REQ_M;
+       }
+
+       /* Write the register */
+       ICE_WRITE_REG(hw, QRX_CTRL(q_idx), reg);
+       /* Check the result. It is said that QENA_STAT
+        * follows the QENA_REQ not more than 10 use.
+        * TODO: need to change the wait counter later
+        */
+       for (j = 0; j < ICE_CHK_Q_ENA_COUNT; j++) {
+               rte_delay_us(ICE_CHK_Q_ENA_INTERVAL_US);
+               reg = ICE_READ_REG(hw, QRX_CTRL(q_idx));
+               if (on) {
+                       if ((reg & QRX_CTRL_QENA_REQ_M) &&
+                           (reg & QRX_CTRL_QENA_STAT_M))
+                               break;
+               } else {
+                       if (!(reg & QRX_CTRL_QENA_REQ_M) &&
+                           !(reg & QRX_CTRL_QENA_STAT_M))
+                               break;
+               }
+       }
+
+       /* Check if it is timeout */
+       if (j >= ICE_CHK_Q_ENA_COUNT) {
+               PMD_DRV_LOG(ERR, "Failed to %s rx queue[%u]",
+                           (on ? "enable" : "disable"), q_idx);
+               return -ETIMEDOUT;
+       }
+
+       return 0;
+}
+
+static inline int
+#ifdef RTE_LIBRTE_ICE_RX_ALLOW_BULK_ALLOC
+ice_check_rx_burst_bulk_alloc_preconditions(struct ice_rx_queue *rxq)
+#else
+ice_check_rx_burst_bulk_alloc_preconditions
+       (__rte_unused struct ice_rx_queue *rxq)
+#endif
+{
+       int ret = 0;
+
+#ifdef RTE_LIBRTE_ICE_RX_ALLOW_BULK_ALLOC
+       if (!(rxq->rx_free_thresh >= ICE_RX_MAX_BURST)) {
+               PMD_INIT_LOG(DEBUG, "Rx Burst Bulk Alloc Preconditions: "
+                            "rxq->rx_free_thresh=%d, "
+                            "ICE_RX_MAX_BURST=%d",
+                            rxq->rx_free_thresh, ICE_RX_MAX_BURST);
+               ret = -EINVAL;
+       } else if (!(rxq->rx_free_thresh < rxq->nb_rx_desc)) {
+               PMD_INIT_LOG(DEBUG, "Rx Burst Bulk Alloc Preconditions: "
+                            "rxq->rx_free_thresh=%d, "
+                            "rxq->nb_rx_desc=%d",
+                            rxq->rx_free_thresh, rxq->nb_rx_desc);
+               ret = -EINVAL;
+       } else if (rxq->nb_rx_desc % rxq->rx_free_thresh != 0) {
+               PMD_INIT_LOG(DEBUG, "Rx Burst Bulk Alloc Preconditions: "
+                            "rxq->nb_rx_desc=%d, "
+                            "rxq->rx_free_thresh=%d",
+                            rxq->nb_rx_desc, rxq->rx_free_thresh);
+               ret = -EINVAL;
+       }
+#else
+       ret = -EINVAL;
+#endif
+
+       return ret;
+}
+
+/* reset fields in ice_rx_queue back to default */
+static void
+ice_reset_rx_queue(struct ice_rx_queue *rxq)
+{
+       unsigned int i;
+       uint16_t len;
+
+       if (!rxq) {
+               PMD_DRV_LOG(DEBUG, "Pointer to rxq is NULL");
+               return;
+       }
+
+#ifdef RTE_LIBRTE_ICE_RX_ALLOW_BULK_ALLOC
+       if (ice_check_rx_burst_bulk_alloc_preconditions(rxq) == 0)
+               len = (uint16_t)(rxq->nb_rx_desc + ICE_RX_MAX_BURST);
+       else
+#endif /* RTE_LIBRTE_ICE_RX_ALLOW_BULK_ALLOC */
+               len = rxq->nb_rx_desc;
+
+       for (i = 0; i < len * sizeof(union ice_rx_desc); i++)
+               ((volatile char *)rxq->rx_ring)[i] = 0;
+
+#ifdef RTE_LIBRTE_ICE_RX_ALLOW_BULK_ALLOC
+       memset(&rxq->fake_mbuf, 0x0, sizeof(rxq->fake_mbuf));
+       for (i = 0; i < ICE_RX_MAX_BURST; ++i)
+               rxq->sw_ring[rxq->nb_rx_desc + i].mbuf = &rxq->fake_mbuf;
+
+       rxq->rx_nb_avail = 0;
+       rxq->rx_next_avail = 0;
+       rxq->rx_free_trigger = (uint16_t)(rxq->rx_free_thresh - 1);
+#endif /* RTE_LIBRTE_ICE_RX_ALLOW_BULK_ALLOC */
+
+       rxq->rx_tail = 0;
+       rxq->nb_rx_hold = 0;
+       rxq->pkt_first_seg = NULL;
+       rxq->pkt_last_seg = NULL;
+}
+
+int
+ice_rx_queue_start(struct rte_eth_dev *dev, uint16_t rx_queue_id)
+{
+       struct ice_rx_queue *rxq;
+       int err;
+       struct ice_hw *hw = ICE_DEV_PRIVATE_TO_HW(dev->data->dev_private);
+
+       PMD_INIT_FUNC_TRACE();
+
+       if (rx_queue_id >= dev->data->nb_rx_queues) {
+               PMD_DRV_LOG(ERR, "RX queue %u is out of range %u",
+                           rx_queue_id, dev->data->nb_rx_queues);
+               return -EINVAL;
+       }
+
+       rxq = dev->data->rx_queues[rx_queue_id];
+       if (!rxq || !rxq->q_set) {
+               PMD_DRV_LOG(ERR, "RX queue %u not available or setup",
+                           rx_queue_id);
+               return -EINVAL;
+       }
+
+       err = ice_program_hw_rx_queue(rxq);
+       if (err) {
+               PMD_DRV_LOG(ERR, "fail to program RX queue %u",
+                           rx_queue_id);
+               return -EIO;
+       }
+
+       err = ice_alloc_rx_queue_mbufs(rxq);
+       if (err) {
+               PMD_DRV_LOG(ERR, "Failed to allocate RX queue mbuf");
+               return -ENOMEM;
+       }
+
+       rte_wmb();
+
+       /* Init the RX tail register. */
+       ICE_PCI_REG_WRITE(rxq->qrx_tail, rxq->nb_rx_desc - 1);
+
+       err = ice_switch_rx_queue(hw, rxq->reg_idx, TRUE);
+       if (err) {
+               PMD_DRV_LOG(ERR, "Failed to switch RX queue %u on",
+                           rx_queue_id);
+
+               ice_rx_queue_release_mbufs(rxq);
+               ice_reset_rx_queue(rxq);
+               return -EINVAL;
+       }
+
+       dev->data->rx_queue_state[rx_queue_id] =
+               RTE_ETH_QUEUE_STATE_STARTED;
+
+       return 0;
+}
+
+int
+ice_rx_queue_stop(struct rte_eth_dev *dev, uint16_t rx_queue_id)
+{
+       struct ice_rx_queue *rxq;
+       int err;
+       struct ice_hw *hw = ICE_DEV_PRIVATE_TO_HW(dev->data->dev_private);
+
+       if (rx_queue_id < dev->data->nb_rx_queues) {
+               rxq = dev->data->rx_queues[rx_queue_id];
+
+               err = ice_switch_rx_queue(hw, rxq->reg_idx, FALSE);
+               if (err) {
+                       PMD_DRV_LOG(ERR, "Failed to switch RX queue %u off",
+                                   rx_queue_id);
+                       return -EINVAL;
+               }
+               ice_rx_queue_release_mbufs(rxq);
+               ice_reset_rx_queue(rxq);
+               dev->data->rx_queue_state[rx_queue_id] =
+                       RTE_ETH_QUEUE_STATE_STOPPED;
+       }
+
+       return 0;
+}
+
+int
+ice_tx_queue_start(struct rte_eth_dev *dev, uint16_t tx_queue_id)
+{
+       struct ice_tx_queue *txq;
+       int err;
+       struct ice_vsi *vsi;
+       struct ice_hw *hw;
+       struct ice_aqc_add_tx_qgrp txq_elem;
+       struct ice_tlan_ctx tx_ctx;
+
+       PMD_INIT_FUNC_TRACE();
+
+       if (tx_queue_id >= dev->data->nb_tx_queues) {
+               PMD_DRV_LOG(ERR, "TX queue %u is out of range %u",
+                           tx_queue_id, dev->data->nb_tx_queues);
+               return -EINVAL;
+       }
+
+       txq = dev->data->tx_queues[tx_queue_id];
+       if (!txq || !txq->q_set) {
+               PMD_DRV_LOG(ERR, "TX queue %u is not available or setup",
+                           tx_queue_id);
+               return -EINVAL;
+       }
+
+       vsi = txq->vsi;
+       hw = ICE_VSI_TO_HW(vsi);
+
+       memset(&txq_elem, 0, sizeof(txq_elem));
+       memset(&tx_ctx, 0, sizeof(tx_ctx));
+       txq_elem.num_txqs = 1;
+       txq_elem.txqs[0].txq_id = rte_cpu_to_le_16(txq->reg_idx);
+
+       tx_ctx.base = txq->tx_ring_phys_addr / ICE_QUEUE_BASE_ADDR_UNIT;
+       tx_ctx.qlen = txq->nb_tx_desc;
+       tx_ctx.pf_num = hw->pf_id;
+       tx_ctx.vmvf_type = ICE_TLAN_CTX_VMVF_TYPE_PF;
+       tx_ctx.src_vsi = vsi->vsi_id;
+       tx_ctx.port_num = hw->port_info->lport;
+       tx_ctx.tso_ena = 1; /* tso enable */
+       tx_ctx.tso_qnum = txq->reg_idx; /* index for tso state structure */
+       tx_ctx.legacy_int = 1; /* Legacy or Advanced Host Interface */
+
+       ice_set_ctx((uint8_t *)&tx_ctx, txq_elem.txqs[0].txq_ctx,
+                   ice_tlan_ctx_info);
+
+       txq->qtx_tail = hw->hw_addr + QTX_COMM_DBELL(txq->reg_idx);
+
+       /* Init the Tx tail register*/
+       ICE_PCI_REG_WRITE(txq->qtx_tail, 0);
+
+       err = ice_ena_vsi_txq(hw->port_info, vsi->idx, 0, 1, &txq_elem,
+                             sizeof(txq_elem), NULL);
+       if (err) {
+               PMD_DRV_LOG(ERR, "Failed to add lan txq");
+               return -EIO;
+       }
+       /* store the schedule node id */
+       txq->q_teid = txq_elem.txqs[0].q_teid;
+
+       dev->data->tx_queue_state[tx_queue_id] = RTE_ETH_QUEUE_STATE_STARTED;
+       return 0;
+}
+
+/* Free all mbufs for descriptors in tx queue */
+static void
+ice_tx_queue_release_mbufs(struct ice_tx_queue *txq)
+{
+       uint16_t i;
+
+       if (!txq || !txq->sw_ring) {
+               PMD_DRV_LOG(DEBUG, "Pointer to txq or sw_ring is NULL");
+               return;
+       }
+
+       for (i = 0; i < txq->nb_tx_desc; i++) {
+               if (txq->sw_ring[i].mbuf) {
+                       rte_pktmbuf_free_seg(txq->sw_ring[i].mbuf);
+                       txq->sw_ring[i].mbuf = NULL;
+               }
+       }
+}
+
+static void
+ice_reset_tx_queue(struct ice_tx_queue *txq)
+{
+       struct ice_tx_entry *txe;
+       uint16_t i, prev, size;
+
+       if (!txq) {
+               PMD_DRV_LOG(DEBUG, "Pointer to txq is NULL");
+               return;
+       }
+
+       txe = txq->sw_ring;
+       size = sizeof(struct ice_tx_desc) * txq->nb_tx_desc;
+       for (i = 0; i < size; i++)
+               ((volatile char *)txq->tx_ring)[i] = 0;
+
+       prev = (uint16_t)(txq->nb_tx_desc - 1);
+       for (i = 0; i < txq->nb_tx_desc; i++) {
+               volatile struct ice_tx_desc *txd = &txq->tx_ring[i];
+
+               txd->cmd_type_offset_bsz =
+                       rte_cpu_to_le_64(ICE_TX_DESC_DTYPE_DESC_DONE);
+               txe[i].mbuf =  NULL;
+               txe[i].last_id = i;
+               txe[prev].next_id = i;
+               prev = i;
+       }
+
+       txq->tx_next_dd = (uint16_t)(txq->tx_rs_thresh - 1);
+       txq->tx_next_rs = (uint16_t)(txq->tx_rs_thresh - 1);
+
+       txq->tx_tail = 0;
+       txq->nb_tx_used = 0;
+
+       txq->last_desc_cleaned = (uint16_t)(txq->nb_tx_desc - 1);
+       txq->nb_tx_free = (uint16_t)(txq->nb_tx_desc - 1);
+}
+
+int
+ice_tx_queue_stop(struct rte_eth_dev *dev, uint16_t tx_queue_id)
+{
+       struct ice_tx_queue *txq;
+       struct ice_hw *hw = ICE_DEV_PRIVATE_TO_HW(dev->data->dev_private);
+       enum ice_status status;
+       uint16_t q_ids[1];
+       uint32_t q_teids[1];
+
+       if (tx_queue_id >= dev->data->nb_tx_queues) {
+               PMD_DRV_LOG(ERR, "TX queue %u is out of range %u",
+                           tx_queue_id, dev->data->nb_tx_queues);
+               return -EINVAL;
+       }
+
+       txq = dev->data->tx_queues[tx_queue_id];
+       if (!txq) {
+               PMD_DRV_LOG(ERR, "TX queue %u is not available",
+                           tx_queue_id);
+               return -EINVAL;
+       }
+
+       q_ids[0] = txq->reg_idx;
+       q_teids[0] = txq->q_teid;
+
+       status = ice_dis_vsi_txq(hw->port_info, 1, q_ids, q_teids,
+                                ICE_NO_RESET, 0, NULL);
+       if (status != ICE_SUCCESS) {
+               PMD_DRV_LOG(DEBUG, "Failed to disable Lan Tx queue");
+               return -EINVAL;
+       }
+
+       ice_tx_queue_release_mbufs(txq);
+       ice_reset_tx_queue(txq);
+       dev->data->tx_queue_state[tx_queue_id] = RTE_ETH_QUEUE_STATE_STOPPED;
+
+       return 0;
+}
+
+int
+ice_rx_queue_setup(struct rte_eth_dev *dev,
+                  uint16_t queue_idx,
+                  uint16_t nb_desc,
+                  unsigned int socket_id,
+                  const struct rte_eth_rxconf *rx_conf,
+                  struct rte_mempool *mp)
+{
+       struct ice_pf *pf = ICE_DEV_PRIVATE_TO_PF(dev->data->dev_private);
+       struct ice_adapter *ad =
+               ICE_DEV_PRIVATE_TO_ADAPTER(dev->data->dev_private);
+       struct ice_vsi *vsi = pf->main_vsi;
+       struct ice_rx_queue *rxq;
+       const struct rte_memzone *rz;
+       uint32_t ring_size;
+       uint16_t len;
+       int use_def_burst_func = 1;
+
+       if (nb_desc % ICE_ALIGN_RING_DESC != 0 ||
+           nb_desc > ICE_MAX_RING_DESC ||
+           nb_desc < ICE_MIN_RING_DESC) {
+               PMD_INIT_LOG(ERR, "Number (%u) of receive descriptors is "
+                            "invalid", nb_desc);
+               return -EINVAL;
+       }
+
+       /* Free memory if needed */
+       if (dev->data->rx_queues[queue_idx]) {
+               ice_rx_queue_release(dev->data->rx_queues[queue_idx]);
+               dev->data->rx_queues[queue_idx] = NULL;
+       }
+
+       /* Allocate the rx queue data structure */
+       rxq = rte_zmalloc_socket(NULL,
+                                sizeof(struct ice_rx_queue),
+                                RTE_CACHE_LINE_SIZE,
+                                socket_id);
+       if (!rxq) {
+               PMD_INIT_LOG(ERR, "Failed to allocate memory for "
+                            "rx queue data structure");
+               return -ENOMEM;
+       }
+       rxq->mp = mp;
+       rxq->nb_rx_desc = nb_desc;
+       rxq->rx_free_thresh = rx_conf->rx_free_thresh;
+       rxq->queue_id = queue_idx;
+
+       rxq->reg_idx = vsi->base_queue + queue_idx;
+       rxq->port_id = dev->data->port_id;
+       if (dev->data->dev_conf.rxmode.offloads & DEV_RX_OFFLOAD_KEEP_CRC)
+               rxq->crc_len = ETHER_CRC_LEN;
+       else
+               rxq->crc_len = 0;
+
+       rxq->drop_en = rx_conf->rx_drop_en;
+       rxq->vsi = vsi;
+       rxq->rx_deferred_start = rx_conf->rx_deferred_start;
+
+       /* Allocate the maximun number of RX ring hardware descriptor. */
+       len = ICE_MAX_RING_DESC;
+
+#ifdef RTE_LIBRTE_ICE_RX_ALLOW_BULK_ALLOC
+       /**
+        * Allocating a little more memory because vectorized/bulk_alloc Rx
+        * functions doesn't check boundaries each time.
+        */
+       len += ICE_RX_MAX_BURST;
+#endif
+
+       /* Allocate the maximum number of RX ring hardware descriptor. */
+       ring_size = sizeof(union ice_rx_desc) * len;
+       ring_size = RTE_ALIGN(ring_size, ICE_DMA_MEM_ALIGN);
+       rz = rte_eth_dma_zone_reserve(dev, "rx_ring", queue_idx,
+                                     ring_size, ICE_RING_BASE_ALIGN,
+                                     socket_id);
+       if (!rz) {
+               ice_rx_queue_release(rxq);
+               PMD_INIT_LOG(ERR, "Failed to reserve DMA memory for RX");
+               return -ENOMEM;
+       }
+
+       /* Zero all the descriptors in the ring. */
+       memset(rz->addr, 0, ring_size);
+
+       rxq->rx_ring_phys_addr = rz->phys_addr;
+       rxq->rx_ring = (union ice_rx_desc *)rz->addr;
+
+#ifdef RTE_LIBRTE_ICE_RX_ALLOW_BULK_ALLOC
+       len = (uint16_t)(nb_desc + ICE_RX_MAX_BURST);
+#else
+       len = nb_desc;
+#endif
+
+       /* Allocate the software ring. */
+       rxq->sw_ring = rte_zmalloc_socket(NULL,
+                                         sizeof(struct ice_rx_entry) * len,
+                                         RTE_CACHE_LINE_SIZE,
+                                         socket_id);
+       if (!rxq->sw_ring) {
+               ice_rx_queue_release(rxq);
+               PMD_INIT_LOG(ERR, "Failed to allocate memory for SW ring");
+               return -ENOMEM;
+       }
+
+       ice_reset_rx_queue(rxq);
+       rxq->q_set = TRUE;
+       dev->data->rx_queues[queue_idx] = rxq;
+
+       use_def_burst_func = ice_check_rx_burst_bulk_alloc_preconditions(rxq);
+
+       if (!use_def_burst_func) {
+#ifdef RTE_LIBRTE_ICE_RX_ALLOW_BULK_ALLOC
+               PMD_INIT_LOG(DEBUG, "Rx Burst Bulk Alloc Preconditions are "
+                            "satisfied. Rx Burst Bulk Alloc function will be "
+                            "used on port=%d, queue=%d.",
+                            rxq->port_id, rxq->queue_id);
+#endif /* RTE_LIBRTE_ICE_RX_ALLOW_BULK_ALLOC */
+       } else {
+               PMD_INIT_LOG(DEBUG, "Rx Burst Bulk Alloc Preconditions are "
+                            "not satisfied, Scattered Rx is requested, "
+                            "or RTE_LIBRTE_ICE_RX_ALLOW_BULK_ALLOC is "
+                            "not enabled on port=%d, queue=%d.",
+                            rxq->port_id, rxq->queue_id);
+               ad->rx_bulk_alloc_allowed = false;
+       }
+
+       return 0;
+}
+
+void
+ice_rx_queue_release(void *rxq)
+{
+       struct ice_rx_queue *q = (struct ice_rx_queue *)rxq;
+
+       if (!q) {
+               PMD_DRV_LOG(DEBUG, "Pointer to rxq is NULL");
+               return;
+       }
+
+       ice_rx_queue_release_mbufs(q);
+       rte_free(q->sw_ring);
+       rte_free(q);
+}
+
+int
+ice_tx_queue_setup(struct rte_eth_dev *dev,
+                  uint16_t queue_idx,
+                  uint16_t nb_desc,
+                  unsigned int socket_id,
+                  const struct rte_eth_txconf *tx_conf)
+{
+       struct ice_pf *pf = ICE_DEV_PRIVATE_TO_PF(dev->data->dev_private);
+       struct ice_vsi *vsi = pf->main_vsi;
+       struct ice_tx_queue *txq;
+       const struct rte_memzone *tz;
+       uint32_t ring_size;
+       uint16_t tx_rs_thresh, tx_free_thresh;
+       uint64_t offloads;
+
+       offloads = tx_conf->offloads | dev->data->dev_conf.txmode.offloads;
+
+       if (nb_desc % ICE_ALIGN_RING_DESC != 0 ||
+           nb_desc > ICE_MAX_RING_DESC ||
+           nb_desc < ICE_MIN_RING_DESC) {
+               PMD_INIT_LOG(ERR, "Number (%u) of transmit descriptors is "
+                            "invalid", nb_desc);
+               return -EINVAL;
+       }
+
+       /**
+        * The following two parameters control the setting of the RS bit on
+        * transmit descriptors. TX descriptors will have their RS bit set
+        * after txq->tx_rs_thresh descriptors have been used. The TX
+        * descriptor ring will be cleaned after txq->tx_free_thresh
+        * descriptors are used or if the number of descriptors required to
+        * transmit a packet is greater than the number of free TX descriptors.
+        *
+        * The following constraints must be satisfied:
+        *  - tx_rs_thresh must be greater than 0.
+        *  - tx_rs_thresh must be less than the size of the ring minus 2.
+        *  - tx_rs_thresh must be less than or equal to tx_free_thresh.
+        *  - tx_rs_thresh must be a divisor of the ring size.
+        *  - tx_free_thresh must be greater than 0.
+        *  - tx_free_thresh must be less than the size of the ring minus 3.
+        *
+        * One descriptor in the TX ring is used as a sentinel to avoid a H/W
+        * race condition, hence the maximum threshold constraints. When set
+        * to zero use default values.
+        */
+       tx_rs_thresh = (uint16_t)(tx_conf->tx_rs_thresh ?
+                                 tx_conf->tx_rs_thresh :
+                                 ICE_DEFAULT_TX_RSBIT_THRESH);
+       tx_free_thresh = (uint16_t)(tx_conf->tx_free_thresh ?
+                                   tx_conf->tx_free_thresh :
+                                   ICE_DEFAULT_TX_FREE_THRESH);
+       if (tx_rs_thresh >= (nb_desc - 2)) {
+               PMD_INIT_LOG(ERR, "tx_rs_thresh must be less than the "
+                            "number of TX descriptors minus 2. "
+                            "(tx_rs_thresh=%u port=%d queue=%d)",
+                            (unsigned int)tx_rs_thresh,
+                            (int)dev->data->port_id,
+                            (int)queue_idx);
+               return -EINVAL;
+       }
+       if (tx_free_thresh >= (nb_desc - 3)) {
+               PMD_INIT_LOG(ERR, "tx_rs_thresh must be less than the "
+                            "tx_free_thresh must be less than the "
+                            "number of TX descriptors minus 3. "
+                            "(tx_free_thresh=%u port=%d queue=%d)",
+                            (unsigned int)tx_free_thresh,
+                            (int)dev->data->port_id,
+                            (int)queue_idx);
+               return -EINVAL;
+       }
+       if (tx_rs_thresh > tx_free_thresh) {
+               PMD_INIT_LOG(ERR, "tx_rs_thresh must be less than or "
+                            "equal to tx_free_thresh. (tx_free_thresh=%u"
+                            " tx_rs_thresh=%u port=%d queue=%d)",
+                            (unsigned int)tx_free_thresh,
+                            (unsigned int)tx_rs_thresh,
+                            (int)dev->data->port_id,
+                            (int)queue_idx);
+               return -EINVAL;
+       }
+       if ((nb_desc % tx_rs_thresh) != 0) {
+               PMD_INIT_LOG(ERR, "tx_rs_thresh must be a divisor of the "
+                            "number of TX descriptors. (tx_rs_thresh=%u"
+                            " port=%d queue=%d)",
+                            (unsigned int)tx_rs_thresh,
+                            (int)dev->data->port_id,
+                            (int)queue_idx);
+               return -EINVAL;
+       }
+       if (tx_rs_thresh > 1 && tx_conf->tx_thresh.wthresh != 0) {
+               PMD_INIT_LOG(ERR, "TX WTHRESH must be set to 0 if "
+                            "tx_rs_thresh is greater than 1. "
+                            "(tx_rs_thresh=%u port=%d queue=%d)",
+                            (unsigned int)tx_rs_thresh,
+                            (int)dev->data->port_id,
+                            (int)queue_idx);
+               return -EINVAL;
+       }
+
+       /* Free memory if needed. */
+       if (dev->data->tx_queues[queue_idx]) {
+               ice_tx_queue_release(dev->data->tx_queues[queue_idx]);
+               dev->data->tx_queues[queue_idx] = NULL;
+       }
+
+       /* Allocate the TX queue data structure. */
+       txq = rte_zmalloc_socket(NULL,
+                                sizeof(struct ice_tx_queue),
+                                RTE_CACHE_LINE_SIZE,
+                                socket_id);
+       if (!txq) {
+               PMD_INIT_LOG(ERR, "Failed to allocate memory for "
+                            "tx queue structure");
+               return -ENOMEM;
+       }
+
+       /* Allocate TX hardware ring descriptors. */
+       ring_size = sizeof(struct ice_tx_desc) * ICE_MAX_RING_DESC;
+       ring_size = RTE_ALIGN(ring_size, ICE_DMA_MEM_ALIGN);
+       tz = rte_eth_dma_zone_reserve(dev, "tx_ring", queue_idx,
+                                     ring_size, ICE_RING_BASE_ALIGN,
+                                     socket_id);
+       if (!tz) {
+               ice_tx_queue_release(txq);
+               PMD_INIT_LOG(ERR, "Failed to reserve DMA memory for TX");
+               return -ENOMEM;
+       }
+
+       txq->nb_tx_desc = nb_desc;
+       txq->tx_rs_thresh = tx_rs_thresh;
+       txq->tx_free_thresh = tx_free_thresh;
+       txq->pthresh = tx_conf->tx_thresh.pthresh;
+       txq->hthresh = tx_conf->tx_thresh.hthresh;
+       txq->wthresh = tx_conf->tx_thresh.wthresh;
+       txq->queue_id = queue_idx;
+
+       txq->reg_idx = vsi->base_queue + queue_idx;
+       txq->port_id = dev->data->port_id;
+       txq->offloads = offloads;
+       txq->vsi = vsi;
+       txq->tx_deferred_start = tx_conf->tx_deferred_start;
+
+       txq->tx_ring_phys_addr = tz->phys_addr;
+       txq->tx_ring = (struct ice_tx_desc *)tz->addr;
+
+       /* Allocate software ring */
+       txq->sw_ring =
+               rte_zmalloc_socket(NULL,
+                                  sizeof(struct ice_tx_entry) * nb_desc,
+                                  RTE_CACHE_LINE_SIZE,
+                                  socket_id);
+       if (!txq->sw_ring) {
+               ice_tx_queue_release(txq);
+               PMD_INIT_LOG(ERR, "Failed to allocate memory for SW TX ring");
+               return -ENOMEM;
+       }
+
+       ice_reset_tx_queue(txq);
+       txq->q_set = TRUE;
+       dev->data->tx_queues[queue_idx] = txq;
+
+       return 0;
+}
+
+void
+ice_tx_queue_release(void *txq)
+{
+       struct ice_tx_queue *q = (struct ice_tx_queue *)txq;
+
+       if (!q) {
+               PMD_DRV_LOG(DEBUG, "Pointer to TX queue is NULL");
+               return;
+       }
+
+       ice_tx_queue_release_mbufs(q);
+       rte_free(q->sw_ring);
+       rte_free(q);
+}
+
+void
+ice_clear_queues(struct rte_eth_dev *dev)
+{
+       uint16_t i;
+
+       PMD_INIT_FUNC_TRACE();
+
+       for (i = 0; i < dev->data->nb_tx_queues; i++) {
+               ice_tx_queue_release_mbufs(dev->data->tx_queues[i]);
+               ice_reset_tx_queue(dev->data->tx_queues[i]);
+       }
+
+       for (i = 0; i < dev->data->nb_rx_queues; i++) {
+               ice_rx_queue_release_mbufs(dev->data->rx_queues[i]);
+               ice_reset_rx_queue(dev->data->rx_queues[i]);
+       }
+}
+
+void
+ice_free_queues(struct rte_eth_dev *dev)
+{
+       uint16_t i;
+
+       PMD_INIT_FUNC_TRACE();
+
+       for (i = 0; i < dev->data->nb_rx_queues; i++) {
+               if (!dev->data->rx_queues[i])
+                       continue;
+               ice_rx_queue_release(dev->data->rx_queues[i]);
+               dev->data->rx_queues[i] = NULL;
+       }
+       dev->data->nb_rx_queues = 0;
+
+       for (i = 0; i < dev->data->nb_tx_queues; i++) {
+               if (!dev->data->tx_queues[i])
+                       continue;
+               ice_tx_queue_release(dev->data->tx_queues[i]);
+               dev->data->tx_queues[i] = NULL;
+       }
+       dev->data->nb_tx_queues = 0;
+}
diff --git a/drivers/net/ice/ice_rxtx.h b/drivers/net/ice/ice_rxtx.h
new file mode 100644 (file)
index 0000000..088a206
--- /dev/null
@@ -0,0 +1,137 @@
+/* SPDX-License-Identifier: BSD-3-Clause
+ * Copyright(c) 2018 Intel Corporation
+ */
+
+#ifndef _ICE_RXTX_H_
+#define _ICE_RXTX_H_
+
+#include "ice_ethdev.h"
+
+#define ICE_ALIGN_RING_DESC  32
+#define ICE_MIN_RING_DESC    64
+#define ICE_MAX_RING_DESC    4096
+#define ICE_DMA_MEM_ALIGN    4096
+#define ICE_RING_BASE_ALIGN  128
+
+#define ICE_RX_MAX_BURST 32
+#define ICE_TX_MAX_BURST 32
+
+#define ICE_CHK_Q_ENA_COUNT        100
+#define ICE_CHK_Q_ENA_INTERVAL_US  100
+
+#ifdef RTE_LIBRTE_ICE_16BYTE_RX_DESC
+#define ice_rx_desc ice_16byte_rx_desc
+#else
+#define ice_rx_desc ice_32byte_rx_desc
+#endif
+
+#define ICE_SUPPORT_CHAIN_NUM 5
+
+struct ice_rx_entry {
+       struct rte_mbuf *mbuf;
+};
+
+struct ice_rx_queue {
+       struct rte_mempool *mp; /* mbuf pool to populate RX ring */
+       volatile union ice_rx_desc *rx_ring;/* RX ring virtual address */
+       uint64_t rx_ring_phys_addr; /* RX ring DMA address */
+       struct ice_rx_entry *sw_ring; /* address of RX soft ring */
+       uint16_t nb_rx_desc; /* number of RX descriptors */
+       uint16_t rx_free_thresh; /* max free RX desc to hold */
+       uint16_t rx_tail; /* current value of tail */
+       uint16_t nb_rx_hold; /* number of held free RX desc */
+       struct rte_mbuf *pkt_first_seg; /**< first segment of current packet */
+       struct rte_mbuf *pkt_last_seg; /**< last segment of current packet */
+#ifdef RTE_LIBRTE_ICE_RX_ALLOW_BULK_ALLOC
+       uint16_t rx_nb_avail; /**< number of staged packets ready */
+       uint16_t rx_next_avail; /**< index of next staged packets */
+       uint16_t rx_free_trigger; /**< triggers rx buffer allocation */
+       struct rte_mbuf fake_mbuf; /**< dummy mbuf */
+       struct rte_mbuf *rx_stage[ICE_RX_MAX_BURST * 2];
+#endif
+       uint8_t port_id; /* device port ID */
+       uint8_t crc_len; /* 0 if CRC stripped, 4 otherwise */
+       uint16_t queue_id; /* RX queue index */
+       uint16_t reg_idx; /* RX queue register index */
+       uint8_t drop_en; /* if not 0, set register bit */
+       volatile uint8_t *qrx_tail; /* register address of tail */
+       struct ice_vsi *vsi; /* the VSI this queue belongs to */
+       uint16_t rx_buf_len; /* The packet buffer size */
+       uint16_t rx_hdr_len; /* The header buffer size */
+       uint16_t max_pkt_len; /* Maximum packet length */
+       bool q_set; /* indicate if rx queue has been configured */
+       bool rx_deferred_start; /* don't start this queue in dev start */
+};
+
+struct ice_tx_entry {
+       struct rte_mbuf *mbuf;
+       uint16_t next_id;
+       uint16_t last_id;
+};
+
+struct ice_tx_queue {
+       uint16_t nb_tx_desc; /* number of TX descriptors */
+       uint64_t tx_ring_phys_addr; /* TX ring DMA address */
+       volatile struct ice_tx_desc *tx_ring; /* TX ring virtual address */
+       struct ice_tx_entry *sw_ring; /* virtual address of SW ring */
+       uint16_t tx_tail; /* current value of tail register */
+       volatile uint8_t *qtx_tail; /* register address of tail */
+       uint16_t nb_tx_used; /* number of TX desc used since RS bit set */
+       /* index to last TX descriptor to have been cleaned */
+       uint16_t last_desc_cleaned;
+       /* Total number of TX descriptors ready to be allocated. */
+       uint16_t nb_tx_free;
+       /* Start freeing TX buffers if there are less free descriptors than
+        * this value.
+        */
+       uint16_t tx_free_thresh;
+       /* Number of TX descriptors to use before RS bit is set. */
+       uint16_t tx_rs_thresh;
+       uint8_t pthresh; /**< Prefetch threshold register. */
+       uint8_t hthresh; /**< Host threshold register. */
+       uint8_t wthresh; /**< Write-back threshold reg. */
+       uint8_t port_id; /* Device port identifier. */
+       uint16_t queue_id; /* TX queue index. */
+       uint32_t q_teid; /* TX schedule node id. */
+       uint16_t reg_idx;
+       uint64_t offloads;
+       struct ice_vsi *vsi; /* the VSI this queue belongs to */
+       uint16_t tx_next_dd;
+       uint16_t tx_next_rs;
+       bool tx_deferred_start; /* don't start this queue in dev start */
+       bool q_set; /* indicate if tx queue has been configured */
+};
+
+/* Offload features */
+union ice_tx_offload {
+       uint64_t data;
+       struct {
+               uint64_t l2_len:7; /* L2 (MAC) Header Length. */
+               uint64_t l3_len:9; /* L3 (IP) Header Length. */
+               uint64_t l4_len:8; /* L4 Header Length. */
+               uint64_t tso_segsz:16; /* TCP TSO segment size */
+               uint64_t outer_l2_len:8; /* outer L2 Header Length */
+               uint64_t outer_l3_len:16; /* outer L3 Header Length */
+       };
+};
+
+int ice_rx_queue_setup(struct rte_eth_dev *dev,
+                      uint16_t queue_idx,
+                      uint16_t nb_desc,
+                      unsigned int socket_id,
+                      const struct rte_eth_rxconf *rx_conf,
+                      struct rte_mempool *mp);
+int ice_tx_queue_setup(struct rte_eth_dev *dev,
+                      uint16_t queue_idx,
+                      uint16_t nb_desc,
+                      unsigned int socket_id,
+                      const struct rte_eth_txconf *tx_conf);
+int ice_rx_queue_start(struct rte_eth_dev *dev, uint16_t rx_queue_id);
+int ice_rx_queue_stop(struct rte_eth_dev *dev, uint16_t rx_queue_id);
+int ice_tx_queue_start(struct rte_eth_dev *dev, uint16_t tx_queue_id);
+int ice_tx_queue_stop(struct rte_eth_dev *dev, uint16_t tx_queue_id);
+void ice_rx_queue_release(void *rxq);
+void ice_tx_queue_release(void *txq);
+void ice_clear_queues(struct rte_eth_dev *dev);
+void ice_free_queues(struct rte_eth_dev *dev);
+#endif /* _ICE_RXTX_H_ */
index 9ed7b27..857dc0e 100644 (file)
@@ -5,7 +5,8 @@ subdir('base')
 objs = [base_objs]
 
 sources = files(
-       'ice_ethdev.c'
+       'ice_ethdev.c',
+       'ice_rxtx.c'
        )
 
 deps += ['hash']