net/ngbe: support Rx queue setup/release
authorJiawen Wu <jiawenwu@trustnetic.com>
Thu, 8 Jul 2021 09:32:32 +0000 (17:32 +0800)
committerAndrew Rybchenko <Andrew.Rybchenko@oktetlabs.ru>
Mon, 12 Jul 2021 15:55:21 +0000 (17:55 +0200)
Setup device Rx queue and release Rx queue.

Signed-off-by: Jiawen Wu <jiawenwu@trustnetic.com>
drivers/net/ngbe/meson.build
drivers/net/ngbe/ngbe_ethdev.c
drivers/net/ngbe/ngbe_ethdev.h
drivers/net/ngbe/ngbe_rxtx.c [new file with mode: 0644]
drivers/net/ngbe/ngbe_rxtx.h [new file with mode: 0644]

index db7d0eb..815ef4d 100644 (file)
@@ -12,6 +12,7 @@ objs = [base_objs]
 
 sources = files(
         'ngbe_ethdev.c',
+        'ngbe_rxtx.c',
 )
 
 includes += include_directories('base')
index decdbd4..015b179 100644 (file)
@@ -12,6 +12,7 @@
 #include "ngbe_logs.h"
 #include "ngbe.h"
 #include "ngbe_ethdev.h"
+#include "ngbe_rxtx.h"
 
 static int ngbe_dev_close(struct rte_eth_dev *dev);
 
@@ -37,6 +38,12 @@ static const struct rte_pci_id pci_id_ngbe_map[] = {
        { .vendor_id = 0, /* sentinel */ },
 };
 
+static const struct rte_eth_desc_lim rx_desc_lim = {
+       .nb_max = NGBE_RING_DESC_MAX,
+       .nb_min = NGBE_RING_DESC_MIN,
+       .nb_align = NGBE_RXD_ALIGN,
+};
+
 static const struct eth_dev_ops ngbe_eth_dev_ops;
 
 static inline void
@@ -241,12 +248,19 @@ static int
 ngbe_dev_configure(struct rte_eth_dev *dev)
 {
        struct ngbe_interrupt *intr = ngbe_dev_intr(dev);
+       struct ngbe_adapter *adapter = ngbe_dev_adapter(dev);
 
        PMD_INIT_FUNC_TRACE();
 
        /* set flag to update link status after init */
        intr->flags |= NGBE_FLAG_NEED_LINK_UPDATE;
 
+       /*
+        * Initialize to TRUE. If any of Rx queues doesn't meet the bulk
+        * allocation Rx preconditions we will reset it.
+        */
+       adapter->rx_bulk_alloc_allowed = true;
+
        return 0;
 }
 
@@ -266,11 +280,30 @@ ngbe_dev_close(struct rte_eth_dev *dev)
 static int
 ngbe_dev_info_get(struct rte_eth_dev *dev, struct rte_eth_dev_info *dev_info)
 {
-       RTE_SET_USED(dev);
+       struct ngbe_hw *hw = ngbe_dev_hw(dev);
+
+       dev_info->max_rx_queues = (uint16_t)hw->mac.max_rx_queues;
+
+       dev_info->default_rxconf = (struct rte_eth_rxconf) {
+               .rx_thresh = {
+                       .pthresh = NGBE_DEFAULT_RX_PTHRESH,
+                       .hthresh = NGBE_DEFAULT_RX_HTHRESH,
+                       .wthresh = NGBE_DEFAULT_RX_WTHRESH,
+               },
+               .rx_free_thresh = NGBE_DEFAULT_RX_FREE_THRESH,
+               .rx_drop_en = 0,
+               .offloads = 0,
+       };
+
+       dev_info->rx_desc_lim = rx_desc_lim;
 
        dev_info->speed_capa = ETH_LINK_SPEED_1G | ETH_LINK_SPEED_100M |
                                ETH_LINK_SPEED_10M;
 
+       /* Driver-preferred Rx/Tx parameters */
+       dev_info->default_rxportconf.nb_queues = 1;
+       dev_info->default_rxportconf.ring_size = 256;
+
        return 0;
 }
 
@@ -559,6 +592,8 @@ static const struct eth_dev_ops ngbe_eth_dev_ops = {
        .dev_configure              = ngbe_dev_configure,
        .dev_infos_get              = ngbe_dev_info_get,
        .link_update                = ngbe_dev_link_update,
+       .rx_queue_setup             = ngbe_dev_rx_queue_setup,
+       .rx_queue_release           = ngbe_dev_rx_queue_release,
 };
 
 RTE_PMD_REGISTER_PCI(net_ngbe, rte_ngbe_pmd);
index cc84132..6d9aa5e 100644 (file)
@@ -30,6 +30,7 @@ struct ngbe_interrupt {
 struct ngbe_adapter {
        struct ngbe_hw             hw;
        struct ngbe_interrupt      intr;
+       bool                       rx_bulk_alloc_allowed;
 };
 
 static inline struct ngbe_adapter *
@@ -58,6 +59,13 @@ ngbe_dev_intr(struct rte_eth_dev *dev)
        return intr;
 }
 
+void ngbe_dev_rx_queue_release(void *rxq);
+
+int  ngbe_dev_rx_queue_setup(struct rte_eth_dev *dev, uint16_t rx_queue_id,
+               uint16_t nb_rx_desc, unsigned int socket_id,
+               const struct rte_eth_rxconf *rx_conf,
+               struct rte_mempool *mb_pool);
+
 int
 ngbe_dev_link_update_share(struct rte_eth_dev *dev,
                int wait_to_complete);
@@ -66,4 +74,12 @@ ngbe_dev_link_update_share(struct rte_eth_dev *dev,
 #define NGBE_LINK_UP_CHECK_TIMEOUT   1000 /* ms */
 #define NGBE_VMDQ_NUM_UC_MAC         4096 /* Maximum nb. of UC MAC addr. */
 
+/*
+ *  Default values for Rx/Tx configuration
+ */
+#define NGBE_DEFAULT_RX_FREE_THRESH  32
+#define NGBE_DEFAULT_RX_PTHRESH      8
+#define NGBE_DEFAULT_RX_HTHRESH      8
+#define NGBE_DEFAULT_RX_WTHRESH      0
+
 #endif /* _NGBE_ETHDEV_H_ */
diff --git a/drivers/net/ngbe/ngbe_rxtx.c b/drivers/net/ngbe/ngbe_rxtx.c
new file mode 100644 (file)
index 0000000..7b37394
--- /dev/null
@@ -0,0 +1,291 @@
+/* SPDX-License-Identifier: BSD-3-Clause
+ * Copyright(c) 2018-2021 Beijing WangXun Technology Co., Ltd.
+ * Copyright(c) 2010-2017 Intel Corporation
+ */
+
+#include <sys/queue.h>
+
+#include <stdint.h>
+#include <rte_ethdev.h>
+#include <ethdev_driver.h>
+#include <rte_malloc.h>
+
+#include "ngbe_logs.h"
+#include "base/ngbe.h"
+#include "ngbe_ethdev.h"
+#include "ngbe_rxtx.h"
+
+/**
+ * ngbe_free_sc_cluster - free the not-yet-completed scattered cluster
+ *
+ * The "next" pointer of the last segment of (not-yet-completed) RSC clusters
+ * in the sw_sc_ring is not set to NULL but rather points to the next
+ * mbuf of this RSC aggregation (that has not been completed yet and still
+ * resides on the HW ring). So, instead of calling for rte_pktmbuf_free() we
+ * will just free first "nb_segs" segments of the cluster explicitly by calling
+ * an rte_pktmbuf_free_seg().
+ *
+ * @m scattered cluster head
+ */
+static void
+ngbe_free_sc_cluster(struct rte_mbuf *m)
+{
+       uint16_t i, nb_segs = m->nb_segs;
+       struct rte_mbuf *next_seg;
+
+       for (i = 0; i < nb_segs; i++) {
+               next_seg = m->next;
+               rte_pktmbuf_free_seg(m);
+               m = next_seg;
+       }
+}
+
+static void
+ngbe_rx_queue_release_mbufs(struct ngbe_rx_queue *rxq)
+{
+       unsigned int i;
+
+       if (rxq->sw_ring != NULL) {
+               for (i = 0; i < rxq->nb_rx_desc; i++) {
+                       if (rxq->sw_ring[i].mbuf != NULL) {
+                               rte_pktmbuf_free_seg(rxq->sw_ring[i].mbuf);
+                               rxq->sw_ring[i].mbuf = NULL;
+                       }
+               }
+               for (i = 0; i < rxq->rx_nb_avail; ++i) {
+                       struct rte_mbuf *mb;
+
+                       mb = rxq->rx_stage[rxq->rx_next_avail + i];
+                       rte_pktmbuf_free_seg(mb);
+               }
+               rxq->rx_nb_avail = 0;
+       }
+
+       if (rxq->sw_sc_ring != NULL)
+               for (i = 0; i < rxq->nb_rx_desc; i++)
+                       if (rxq->sw_sc_ring[i].fbuf != NULL) {
+                               ngbe_free_sc_cluster(rxq->sw_sc_ring[i].fbuf);
+                               rxq->sw_sc_ring[i].fbuf = NULL;
+                       }
+}
+
+static void
+ngbe_rx_queue_release(struct ngbe_rx_queue *rxq)
+{
+       if (rxq != NULL) {
+               ngbe_rx_queue_release_mbufs(rxq);
+               rte_free(rxq->sw_ring);
+               rte_free(rxq->sw_sc_ring);
+               rte_free(rxq);
+       }
+}
+
+void
+ngbe_dev_rx_queue_release(void *rxq)
+{
+       ngbe_rx_queue_release(rxq);
+}
+
+/*
+ * Check if Rx Burst Bulk Alloc function can be used.
+ * Return
+ *        0: the preconditions are satisfied and the bulk allocation function
+ *           can be used.
+ *  -EINVAL: the preconditions are NOT satisfied and the default Rx burst
+ *           function must be used.
+ */
+static inline int
+check_rx_burst_bulk_alloc_preconditions(struct ngbe_rx_queue *rxq)
+{
+       int ret = 0;
+
+       /*
+        * Make sure the following pre-conditions are satisfied:
+        *   rxq->rx_free_thresh >= RTE_PMD_NGBE_RX_MAX_BURST
+        *   rxq->rx_free_thresh < rxq->nb_rx_desc
+        *   (rxq->nb_rx_desc % rxq->rx_free_thresh) == 0
+        * Scattered packets are not supported.  This should be checked
+        * outside of this function.
+        */
+       if (rxq->rx_free_thresh < RTE_PMD_NGBE_RX_MAX_BURST) {
+               PMD_INIT_LOG(DEBUG,
+                            "Rx Burst Bulk Alloc Preconditions: rxq->rx_free_thresh=%d, RTE_PMD_NGBE_RX_MAX_BURST=%d",
+                            rxq->rx_free_thresh, RTE_PMD_NGBE_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;
+       }
+
+       return ret;
+}
+
+/* Reset dynamic ngbe_rx_queue fields back to defaults */
+static void
+ngbe_reset_rx_queue(struct ngbe_adapter *adapter, struct ngbe_rx_queue *rxq)
+{
+       static const struct ngbe_rx_desc zeroed_desc = {
+                                               {{0}, {0} }, {{0}, {0} } };
+       unsigned int i;
+       uint16_t len = rxq->nb_rx_desc;
+
+       /*
+        * By default, the Rx queue setup function allocates enough memory for
+        * NGBE_RING_DESC_MAX.  The Rx Burst bulk allocation function requires
+        * extra memory at the end of the descriptor ring to be zero'd out.
+        */
+       if (adapter->rx_bulk_alloc_allowed)
+               /* zero out extra memory */
+               len += RTE_PMD_NGBE_RX_MAX_BURST;
+
+       /*
+        * Zero out HW ring memory. Zero out extra memory at the end of
+        * the H/W ring so look-ahead logic in Rx Burst bulk alloc function
+        * reads extra memory as zeros.
+        */
+       for (i = 0; i < len; i++)
+               rxq->rx_ring[i] = zeroed_desc;
+
+       /*
+        * initialize extra software ring entries. Space for these extra
+        * entries is always allocated
+        */
+       memset(&rxq->fake_mbuf, 0x0, sizeof(rxq->fake_mbuf));
+       for (i = rxq->nb_rx_desc; i < len; ++i)
+               rxq->sw_ring[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);
+       rxq->rx_tail = 0;
+       rxq->nb_rx_hold = 0;
+       rxq->pkt_first_seg = NULL;
+       rxq->pkt_last_seg = NULL;
+}
+
+int
+ngbe_dev_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)
+{
+       const struct rte_memzone *rz;
+       struct ngbe_rx_queue *rxq;
+       struct ngbe_hw     *hw;
+       uint16_t len;
+       struct ngbe_adapter *adapter = ngbe_dev_adapter(dev);
+
+       PMD_INIT_FUNC_TRACE();
+       hw = ngbe_dev_hw(dev);
+
+       /* Free memory prior to re-allocation if needed... */
+       if (dev->data->rx_queues[queue_idx] != NULL) {
+               ngbe_rx_queue_release(dev->data->rx_queues[queue_idx]);
+               dev->data->rx_queues[queue_idx] = NULL;
+       }
+
+       /* First allocate the Rx queue data structure */
+       rxq = rte_zmalloc_socket("ethdev RX queue",
+                                sizeof(struct ngbe_rx_queue),
+                                RTE_CACHE_LINE_SIZE, socket_id);
+       if (rxq == NULL)
+               return -ENOMEM;
+       rxq->mb_pool = mp;
+       rxq->nb_rx_desc = nb_desc;
+       rxq->rx_free_thresh = rx_conf->rx_free_thresh;
+       rxq->queue_id = queue_idx;
+       rxq->reg_idx = queue_idx;
+       rxq->port_id = dev->data->port_id;
+       rxq->drop_en = rx_conf->rx_drop_en;
+       rxq->rx_deferred_start = rx_conf->rx_deferred_start;
+
+       /*
+        * Allocate Rx ring hardware descriptors. A memzone large enough to
+        * handle the maximum ring size is allocated in order to allow for
+        * resizing in later calls to the queue setup function.
+        */
+       rz = rte_eth_dma_zone_reserve(dev, "rx_ring", queue_idx,
+                                     RX_RING_SZ, NGBE_ALIGN, socket_id);
+       if (rz == NULL) {
+               ngbe_rx_queue_release(rxq);
+               return -ENOMEM;
+       }
+
+       /*
+        * Zero init all the descriptors in the ring.
+        */
+       memset(rz->addr, 0, RX_RING_SZ);
+
+       rxq->rdt_reg_addr = NGBE_REG_ADDR(hw, NGBE_RXWP(rxq->reg_idx));
+       rxq->rdh_reg_addr = NGBE_REG_ADDR(hw, NGBE_RXRP(rxq->reg_idx));
+
+       rxq->rx_ring_phys_addr = TMZ_PADDR(rz);
+       rxq->rx_ring = (struct ngbe_rx_desc *)TMZ_VADDR(rz);
+
+       /*
+        * Certain constraints must be met in order to use the bulk buffer
+        * allocation Rx burst function. If any of Rx queues doesn't meet them
+        * the feature should be disabled for the whole port.
+        */
+       if (check_rx_burst_bulk_alloc_preconditions(rxq)) {
+               PMD_INIT_LOG(DEBUG,
+                            "queue[%d] doesn't meet Rx Bulk Alloc preconditions - canceling the feature for the whole port[%d]",
+                            rxq->queue_id, rxq->port_id);
+               adapter->rx_bulk_alloc_allowed = false;
+       }
+
+       /*
+        * Allocate software ring. Allow for space at the end of the
+        * S/W ring to make sure look-ahead logic in bulk alloc Rx burst
+        * function does not access an invalid memory region.
+        */
+       len = nb_desc;
+       if (adapter->rx_bulk_alloc_allowed)
+               len += RTE_PMD_NGBE_RX_MAX_BURST;
+
+       rxq->sw_ring = rte_zmalloc_socket("rxq->sw_ring",
+                                         sizeof(struct ngbe_rx_entry) * len,
+                                         RTE_CACHE_LINE_SIZE, socket_id);
+       if (rxq->sw_ring == NULL) {
+               ngbe_rx_queue_release(rxq);
+               return -ENOMEM;
+       }
+
+       /*
+        * Always allocate even if it's not going to be needed in order to
+        * simplify the code.
+        *
+        * This ring is used in Scattered Rx cases and Scattered Rx may
+        * be requested in ngbe_dev_rx_init(), which is called later from
+        * dev_start() flow.
+        */
+       rxq->sw_sc_ring =
+               rte_zmalloc_socket("rxq->sw_sc_ring",
+                                 sizeof(struct ngbe_scattered_rx_entry) * len,
+                                 RTE_CACHE_LINE_SIZE, socket_id);
+       if (rxq->sw_sc_ring == NULL) {
+               ngbe_rx_queue_release(rxq);
+               return -ENOMEM;
+       }
+
+       PMD_INIT_LOG(DEBUG,
+                    "sw_ring=%p sw_sc_ring=%p hw_ring=%p dma_addr=0x%" PRIx64,
+                    rxq->sw_ring, rxq->sw_sc_ring, rxq->rx_ring,
+                    rxq->rx_ring_phys_addr);
+
+       dev->data->rx_queues[queue_idx] = rxq;
+
+       ngbe_reset_rx_queue(adapter, rxq);
+
+       return 0;
+}
+
diff --git a/drivers/net/ngbe/ngbe_rxtx.h b/drivers/net/ngbe/ngbe_rxtx.h
new file mode 100644 (file)
index 0000000..2ad6890
--- /dev/null
@@ -0,0 +1,100 @@
+/* SPDX-License-Identifier: BSD-3-Clause
+ * Copyright(c) 2018-2021 Beijing WangXun Technology Co., Ltd.
+ * Copyright(c) 2010-2017 Intel Corporation
+ */
+
+#ifndef _NGBE_RXTX_H_
+#define _NGBE_RXTX_H_
+
+/*****************************************************************************
+ * Receive Descriptor
+ *****************************************************************************/
+struct ngbe_rx_desc {
+       struct {
+               union {
+                       rte_le32_t dw0;
+                       struct {
+                               rte_le16_t pkt;
+                               rte_le16_t hdr;
+                       } lo;
+               };
+               union {
+                       rte_le32_t dw1;
+                       struct {
+                               rte_le16_t ipid;
+                               rte_le16_t csum;
+                       } hi;
+               };
+       } qw0; /* also as r.pkt_addr */
+       struct {
+               union {
+                       rte_le32_t dw2;
+                       struct {
+                               rte_le32_t status;
+                       } lo;
+               };
+               union {
+                       rte_le32_t dw3;
+                       struct {
+                               rte_le16_t len;
+                               rte_le16_t tag;
+                       } hi;
+               };
+       } qw1; /* also as r.hdr_addr */
+};
+
+#define RTE_PMD_NGBE_RX_MAX_BURST 32
+
+#define RX_RING_SZ ((NGBE_RING_DESC_MAX + RTE_PMD_NGBE_RX_MAX_BURST) * \
+                   sizeof(struct ngbe_rx_desc))
+
+
+/**
+ * Structure associated with each descriptor of the Rx ring of a Rx queue.
+ */
+struct ngbe_rx_entry {
+       struct rte_mbuf *mbuf; /**< mbuf associated with Rx descriptor. */
+};
+
+struct ngbe_scattered_rx_entry {
+       struct rte_mbuf *fbuf; /**< First segment of the fragmented packet. */
+};
+
+/**
+ * Structure associated with each Rx queue.
+ */
+struct ngbe_rx_queue {
+       struct rte_mempool   *mb_pool; /**< mbuf pool to populate Rx ring */
+       uint64_t             rx_ring_phys_addr; /**< Rx ring DMA address */
+       volatile uint32_t    *rdt_reg_addr; /**< RDT register address */
+       volatile uint32_t    *rdh_reg_addr; /**< RDH register address */
+
+       volatile struct ngbe_rx_desc   *rx_ring; /**< Rx ring virtual address */
+       /** address of Rx software ring */
+       struct ngbe_rx_entry           *sw_ring;
+       /** address of scattered Rx software ring */
+       struct ngbe_scattered_rx_entry *sw_sc_ring;
+
+       struct rte_mbuf *pkt_first_seg; /**< First segment of current packet */
+       struct rte_mbuf *pkt_last_seg; /**< Last segment of current packet */
+       uint16_t        nb_rx_desc; /**< number of Rx descriptors */
+       uint16_t        rx_tail;  /**< current value of RDT register */
+       uint16_t        nb_rx_hold; /**< number of held free Rx desc */
+
+       uint16_t rx_nb_avail; /**< nr of staged pkts ready to ret to app */
+       uint16_t rx_next_avail; /**< idx of next staged pkt to ret to app */
+       uint16_t rx_free_trigger; /**< triggers rx buffer allocation */
+
+       uint16_t        rx_free_thresh; /**< max free Rx desc to hold */
+       uint16_t        queue_id; /**< RX queue index */
+       uint16_t        reg_idx;  /**< RX queue register index */
+       uint16_t        port_id;  /**< Device port identifier */
+       uint8_t         drop_en;  /**< If not 0, set SRRCTL.Drop_En */
+       uint8_t         rx_deferred_start; /**< not in global dev start */
+       /** need to alloc dummy mbuf, for wraparound when scanning hw ring */
+       struct rte_mbuf fake_mbuf;
+       /** hold packets to return to application */
+       struct rte_mbuf *rx_stage[RTE_PMD_NGBE_RX_MAX_BURST * 2];
+};
+
+#endif /* _NGBE_RXTX_H_ */