fm10k: reset and release mbuf for vector Tx

Vector TX use different way to manage TX queue, it's necessary
to use different functions to reset TX queue and release mbuf
in TX queue. So, introduce 2 function pointers to do such ops.

Signed-off-by: Chen Jing D(Mark) <jing.d.chen@intel.com>
Acked-by: Cunming Liang <cunming.liang@intel.com>

diff --git a/drivers/net/fm10k/fm10k.h b/drivers/net/fm10k/fm10k.h
index 7c8d737..aafdab7 100644 (file)
--- a/drivers/net/fm10k/fm10k.h
+++ b/drivers/net/fm10k/fm10k.h
@@ -209,11 +209,14 @@ struct fifo {
        uint16_t *endp;
 };
 
+struct fm10k_txq_ops;
+
 struct fm10k_tx_queue {
        struct rte_mbuf **sw_ring;
        struct fm10k_tx_desc *hw_ring;
        uint64_t hw_ring_phys_addr;
        struct fifo rs_tracker;
+       const struct fm10k_txq_ops *ops; /* txq ops */
        uint16_t last_free;
        uint16_t next_free;
        uint16_t nb_free;
@@ -230,6 +233,11 @@ struct fm10k_tx_queue {
        uint16_t queue_id;
 };
 
+struct fm10k_txq_ops {
+       void (*release_mbufs)(struct fm10k_tx_queue *txq);
+       void (*reset)(struct fm10k_tx_queue *txq);
+};
+
 #define MBUF_DMA_ADDR(mb) \
        ((uint64_t) ((mb)->buf_physaddr + (mb)->data_off))
 
@@ -343,4 +351,5 @@ uint16_t fm10k_recv_scattered_pkts_vec(void *, struct rte_mbuf **,
                                        uint16_t);
 uint16_t fm10k_xmit_pkts_vec(void *tx_queue, struct rte_mbuf **tx_pkts,
                uint16_t nb_pkts);
+void fm10k_txq_vec_setup(struct fm10k_tx_queue *txq);
 #endif

diff --git a/drivers/net/fm10k/fm10k_ethdev.c b/drivers/net/fm10k/fm10k_ethdev.c
index 247edd5..d55b532 100644 (file)
--- a/drivers/net/fm10k/fm10k_ethdev.c
+++ b/drivers/net/fm10k/fm10k_ethdev.c
@@ -327,6 +327,11 @@ fm10k_check_mq_mode(struct rte_eth_dev *dev)
        return 0;
 }
 
+static const struct fm10k_txq_ops def_txq_ops = {
+       .release_mbufs = tx_queue_free,
+       .reset = tx_queue_reset,
+};
+
 static int
 fm10k_dev_configure(struct rte_eth_dev *dev)
 {
@@ -721,7 +726,9 @@ fm10k_dev_tx_queue_start(struct rte_eth_dev *dev, uint16_t tx_queue_id)
        PMD_INIT_FUNC_TRACE();
 
        if (tx_queue_id < dev->data->nb_tx_queues) {
-               tx_queue_reset(dev->data->tx_queues[tx_queue_id]);
+               struct fm10k_tx_queue *q = dev->data->tx_queues[tx_queue_id];
+
+               q->ops->reset(q);
 
                /* reset head and tail pointers */
                FM10K_WRITE_REG(hw, FM10K_TDH(tx_queue_id), 0);
@@ -997,8 +1004,11 @@ fm10k_dev_queue_release(struct rte_eth_dev *dev)
        PMD_INIT_FUNC_TRACE();
 
        if (dev->data->tx_queues) {
-               for (i = 0; i < dev->data->nb_tx_queues; i++)
-                       fm10k_tx_queue_release(dev->data->tx_queues[i]);
+               for (i = 0; i < dev->data->nb_tx_queues; i++) {
+                       struct fm10k_tx_queue *txq = dev->data->tx_queues[i];
+
+                       txq->ops->release_mbufs(txq);
+               }
        }
 
        if (dev->data->rx_queues) {
@@ -1670,7 +1680,9 @@ fm10k_tx_queue_setup(struct rte_eth_dev *dev, uint16_t queue_id,
         * different socket than was previously used.
         */
        if (dev->data->tx_queues[queue_id] != NULL) {
-               tx_queue_free(dev->data->tx_queues[queue_id]);
+               struct fm10k_tx_queue *txq = dev->data->tx_queues[queue_id];
+
+               txq->ops->release_mbufs(txq);
                dev->data->tx_queues[queue_id] = NULL;
        }
 
@@ -1686,6 +1698,7 @@ fm10k_tx_queue_setup(struct rte_eth_dev *dev, uint16_t queue_id,
        q->nb_desc = nb_desc;
        q->port_id = dev->data->port_id;
        q->queue_id = queue_id;
+       q->ops = &def_txq_ops;
        q->tail_ptr = (volatile uint32_t *)
                &((uint32_t *)hw->hw_addr)[FM10K_TDT(queue_id)];
        if (handle_txconf(q, conf))
@@ -1744,9 +1757,10 @@ fm10k_tx_queue_setup(struct rte_eth_dev *dev, uint16_t queue_id,
 static void
 fm10k_tx_queue_release(void *queue)
 {
+       struct fm10k_tx_queue *q = queue;
        PMD_INIT_FUNC_TRACE();
 
-       tx_queue_free(queue);
+       q->ops->release_mbufs(q);
 }
 
 static int

diff --git a/drivers/net/fm10k/fm10k_rxtx_vec.c b/drivers/net/fm10k/fm10k_rxtx_vec.c
index 4515b26..06beca9 100644 (file)
--- a/drivers/net/fm10k/fm10k_rxtx_vec.c
+++ b/drivers/net/fm10k/fm10k_rxtx_vec.c
@@ -44,6 +44,11 @@
 #pragma GCC diagnostic ignored "-Wcast-qual"
 #endif
 
+static void
+fm10k_tx_queue_release_mbufs_vec(struct fm10k_tx_queue *txq);
+static void
+fm10k_reset_tx_queue(struct fm10k_tx_queue *txq);
+
 /* Handling the offload flags (olflags) field takes computation
  * time when receiving packets. Therefore we provide a flag to disable
  * the processing of the olflags field when they are not needed. This
@@ -628,6 +633,17 @@ fm10k_recv_scattered_pkts_vec(void *rx_queue,
                &split_flags[i]);
 }
 
+static const struct fm10k_txq_ops vec_txq_ops = {
+       .release_mbufs = fm10k_tx_queue_release_mbufs_vec,
+       .reset = fm10k_reset_tx_queue,
+};
+
+void __attribute__((cold))
+fm10k_txq_vec_setup(struct fm10k_tx_queue *txq)
+{
+       txq->ops = &vec_txq_ops;
+}
+
 static inline void
 vtx1(volatile struct fm10k_tx_desc *txdp,
                struct rte_mbuf *pkt, uint64_t flags)
@@ -777,3 +793,55 @@ fm10k_xmit_pkts_vec(void *tx_queue, struct rte_mbuf **tx_pkts,
 
        return nb_pkts;
 }
+
+static void __attribute__((cold))
+fm10k_tx_queue_release_mbufs_vec(struct fm10k_tx_queue *txq)
+{
+       unsigned i;
+       const uint16_t max_desc = (uint16_t)(txq->nb_desc - 1);
+
+       if (txq->sw_ring == NULL || txq->nb_free == max_desc)
+               return;
+
+       /* release the used mbufs in sw_ring */
+       for (i = txq->next_dd - (txq->rs_thresh - 1);
+            i != txq->next_free;
+            i = (i + 1) & max_desc)
+               rte_pktmbuf_free_seg(txq->sw_ring[i]);
+
+       txq->nb_free = max_desc;
+
+       /* reset tx_entry */
+       for (i = 0; i < txq->nb_desc; i++)
+               txq->sw_ring[i] = NULL;
+
+       rte_free(txq->sw_ring);
+       txq->sw_ring = NULL;
+}
+
+static void __attribute__((cold))
+fm10k_reset_tx_queue(struct fm10k_tx_queue *txq)
+{
+       static const struct fm10k_tx_desc zeroed_desc = {0};
+       struct rte_mbuf **txe = txq->sw_ring;
+       uint16_t i;
+
+       /* Zero out HW ring memory */
+       for (i = 0; i < txq->nb_desc; i++)
+               txq->hw_ring[i] = zeroed_desc;
+
+       /* Initialize SW ring entries */
+       for (i = 0; i < txq->nb_desc; i++)
+               txe[i] = NULL;
+
+       txq->next_dd = (uint16_t)(txq->rs_thresh - 1);
+       txq->next_rs = (uint16_t)(txq->rs_thresh - 1);
+
+       txq->next_free = 0;
+       txq->nb_used = 0;
+       /* Always allow 1 descriptor to be un-allocated to avoid
+        * a H/W race condition
+        */
+       txq->nb_free = (uint16_t)(txq->nb_desc - 1);
+       FM10K_PCI_REG_WRITE(txq->tail_ptr, 0);
+}