net/dpaa2: support atomic queues
authorNipun Gupta <nipun.gupta@nxp.com>
Wed, 17 Jan 2018 11:39:14 +0000 (17:09 +0530)
committerJerin Jacob <jerin.jacob@caviumnetworks.com>
Fri, 19 Jan 2018 15:09:56 +0000 (16:09 +0100)
Signed-off-by: Nipun Gupta <nipun.gupta@nxp.com>
Acked-by: Hemant Agrawal <hemant.agrawal@nxp.com>
drivers/net/dpaa2/Makefile
drivers/net/dpaa2/dpaa2_ethdev.c
drivers/net/dpaa2/dpaa2_ethdev.h
drivers/net/dpaa2/dpaa2_rxtx.c

index 068e9d3..5a93a0b 100644 (file)
@@ -25,6 +25,7 @@ CFLAGS += -I$(RTE_SDK)/drivers/bus/fslmc/qbman/include
 CFLAGS += -I$(RTE_SDK)/drivers/bus/fslmc/mc
 CFLAGS += -I$(RTE_SDK)/drivers/bus/fslmc/portal
 CFLAGS += -I$(RTE_SDK)/drivers/mempool/dpaa2
+CFLAGS += -I$(RTE_SDK)/drivers/event/dpaa2
 CFLAGS += -I$(RTE_SDK)/lib/librte_eal/linuxapp/eal
 
 # versioning export map
index 821c862..5f24759 100644 (file)
@@ -1670,6 +1670,8 @@ int dpaa2_eth_eventq_attach(const struct rte_eth_dev *dev,
 
        if (queue_conf->ev.sched_type == RTE_SCHED_TYPE_PARALLEL)
                dpaa2_ethq->cb = dpaa2_dev_process_parallel_event;
+       else if (queue_conf->ev.sched_type == RTE_SCHED_TYPE_ATOMIC)
+               dpaa2_ethq->cb = dpaa2_dev_process_atomic_event;
        else
                return -EINVAL;
 
@@ -1679,6 +1681,11 @@ int dpaa2_eth_eventq_attach(const struct rte_eth_dev *dev,
        cfg.destination.id = dpcon_id;
        cfg.destination.priority = queue_conf->ev.priority;
 
+       if (queue_conf->ev.sched_type == RTE_SCHED_TYPE_ATOMIC) {
+               options |= DPNI_QUEUE_OPT_HOLD_ACTIVE;
+               cfg.destination.hold_active = 1;
+       }
+
        options |= DPNI_QUEUE_OPT_USER_CTX;
        cfg.user_context = (uint64_t)(dpaa2_ethq);
 
index 8fa1872..ba0856f 100644 (file)
@@ -118,6 +118,11 @@ void dpaa2_dev_process_parallel_event(struct qbman_swp *swp,
                                      const struct qbman_result *dq,
                                      struct dpaa2_queue *rxq,
                                      struct rte_event *ev);
+void dpaa2_dev_process_atomic_event(struct qbman_swp *swp,
+                                   const struct qbman_fd *fd,
+                                   const struct qbman_result *dq,
+                                   struct dpaa2_queue *rxq,
+                                   struct rte_event *ev);
 uint16_t dpaa2_dev_tx(void *queue, struct rte_mbuf **bufs, uint16_t nb_pkts);
 uint16_t dummy_dev_tx(void *queue, struct rte_mbuf **bufs, uint16_t nb_pkts);
 #endif /* _DPAA2_ETHDEV_H */
index c43e3a1..e88b034 100644 (file)
 #include <rte_string_fns.h>
 #include <rte_dev.h>
 
+#include <rte_fslmc.h>
 #include <fslmc_logs.h>
 #include <fslmc_vfio.h>
 #include <dpaa2_hw_pvt.h>
 #include <dpaa2_hw_dpio.h>
 #include <dpaa2_hw_mempool.h>
+#include <dpaa2_eventdev.h>
 
 #include "dpaa2_ethdev.h"
 #include "base/dpaa2_hw_dpni_annot.h"
@@ -641,6 +643,30 @@ dpaa2_dev_process_parallel_event(struct qbman_swp *swp,
        qbman_swp_dqrr_consume(swp, dq);
 }
 
+void dpaa2_dev_process_atomic_event(struct qbman_swp *swp __attribute__((unused)),
+                                   const struct qbman_fd *fd,
+                                   const struct qbman_result *dq,
+                                   struct dpaa2_queue *rxq,
+                                   struct rte_event *ev)
+{
+       uint8_t dqrr_index = qbman_get_dqrr_idx(dq);
+
+       ev->mbuf = eth_fd_to_mbuf(fd);
+
+       ev->flow_id = rxq->ev.flow_id;
+       ev->sub_event_type = rxq->ev.sub_event_type;
+       ev->event_type = RTE_EVENT_TYPE_ETHDEV;
+       ev->op = RTE_EVENT_OP_NEW;
+       ev->sched_type = rxq->ev.sched_type;
+       ev->queue_id = rxq->ev.queue_id;
+       ev->priority = rxq->ev.priority;
+
+       ev->mbuf->seqn = dqrr_index + 1;
+       DPAA2_PER_LCORE_DQRR_SIZE++;
+       DPAA2_PER_LCORE_DQRR_HELD |= 1 << dqrr_index;
+       DPAA2_PER_LCORE_DQRR_MBUF(dqrr_index) = ev->mbuf;
+}
+
 /*
  * Callback to handle sending packets through WRIOP based interface
  */
@@ -661,6 +687,7 @@ dpaa2_dev_tx(void *queue, struct rte_mbuf **bufs, uint16_t nb_pkts)
        uint16_t bpid;
        struct rte_eth_dev *dev = dpaa2_q->dev;
        struct dpaa2_dev_priv *priv = dev->data->dev_private;
+       uint32_t flags[MAX_TX_RING_SLOTS] = {0};
 
        if (unlikely(!DPAA2_PER_LCORE_DPIO)) {
                ret = dpaa2_affine_qbman_swp();
@@ -679,7 +706,6 @@ dpaa2_dev_tx(void *queue, struct rte_mbuf **bufs, uint16_t nb_pkts)
        qbman_eq_desc_set_response(&eqdesc, 0, 0);
        qbman_eq_desc_set_qd(&eqdesc, priv->qdid,
                             dpaa2_q->flow_id, dpaa2_q->tc_index);
-
        /*Clear the unused FD fields before sending*/
        while (nb_pkts) {
                /*Check if the queue is congested*/
@@ -694,6 +720,16 @@ dpaa2_dev_tx(void *queue, struct rte_mbuf **bufs, uint16_t nb_pkts)
                frames_to_send = (nb_pkts >> 3) ? MAX_TX_RING_SLOTS : nb_pkts;
 
                for (loop = 0; loop < frames_to_send; loop++) {
+                       if ((*bufs)->seqn) {
+                               uint8_t dqrr_index = (*bufs)->seqn - 1;
+
+                               flags[loop] = QBMAN_ENQUEUE_FLAG_DCA |
+                                               dqrr_index;
+                               DPAA2_PER_LCORE_DQRR_SIZE--;
+                               DPAA2_PER_LCORE_DQRR_HELD &= ~(1 << dqrr_index);
+                               (*bufs)->seqn = DPAA2_INVALID_MBUF_SEQN;
+                       }
+
                        fd_arr[loop].simple.frc = 0;
                        DPAA2_RESET_FD_CTRL((&fd_arr[loop]));
                        DPAA2_SET_FD_FLC((&fd_arr[loop]), NULL);
@@ -761,7 +797,7 @@ dpaa2_dev_tx(void *queue, struct rte_mbuf **bufs, uint16_t nb_pkts)
                loop = 0;
                while (loop < frames_to_send) {
                        loop += qbman_swp_enqueue_multiple(swp, &eqdesc,
-                                       &fd_arr[loop], NULL,
+                                       &fd_arr[loop], &flags[loop],
                                        frames_to_send - loop);
                }
 
@@ -778,7 +814,8 @@ send_n_return:
 
                while (i < loop) {
                        i += qbman_swp_enqueue_multiple(swp, &eqdesc,
-                                                       &fd_arr[i], NULL,
+                                                       &fd_arr[i],
+                                                       &flags[loop],
                                                        loop - i);
                }
                num_tx += loop;