raw/dpaa2_qdma: support RBP mode
authorHemant Agrawal <hemant.agrawal@nxp.com>
Thu, 4 Apr 2019 11:50:25 +0000 (11:50 +0000)
committerThomas Monjalon <thomas@monjalon.net>
Fri, 5 Apr 2019 08:40:43 +0000 (10:40 +0200)
Add support for route by port mode. The route by port
feature in HW helps in translating the PCI address
of connected device.

Signed-off-by: Minghuan Lian <minghuan.lian@nxp.com>
Signed-off-by: Sachin Saxena <sachin.saxena@nxp.com>
Signed-off-by: Hemant Agrawal <hemant.agrawal@nxp.com>
drivers/raw/dpaa2_qdma/Makefile
drivers/raw/dpaa2_qdma/dpaa2_qdma.c
drivers/raw/dpaa2_qdma/dpaa2_qdma.h
drivers/raw/dpaa2_qdma/rte_pmd_dpaa2_qdma.h

index 1611439..6047eea 100644 (file)
@@ -25,7 +25,7 @@ LDLIBS += -lrte_common_dpaax
 
 EXPORT_MAP := rte_pmd_dpaa2_qdma_version.map
 
-LIBABIVER := 2
+LIBABIVER := 3
 
 #
 # all source are stored in SRCS-y
index a1351e6..cf1a1aa 100644 (file)
 #include <portal/dpaa2_hw_pvt.h>
 #include <portal/dpaa2_hw_dpio.h>
 
+#include "rte_pmd_dpaa2_qdma.h"
 #include "dpaa2_qdma.h"
 #include "dpaa2_qdma_logs.h"
-#include "rte_pmd_dpaa2_qdma.h"
 
 /* Dynamic log type identifier */
 int dpaa2_qdma_logtype;
 
+uint32_t dpaa2_coherent_no_alloc_cache;
+uint32_t dpaa2_coherent_alloc_cache;
+
 /* QDMA device */
 static struct qdma_device qdma_dev;
 
@@ -345,14 +348,29 @@ rte_qdma_vq_create(uint32_t lcore_id, uint32_t flags)
 
        qdma_vqs[i].in_use = 1;
        qdma_vqs[i].lcore_id = lcore_id;
-
+       memset(&qdma_vqs[i].rbp, 0, sizeof(struct rte_qdma_rbp));
        rte_spinlock_unlock(&qdma_dev.lock);
 
        return i;
 }
 
+/*create vq for route-by-port*/
+int
+rte_qdma_vq_create_rbp(uint32_t lcore_id, uint32_t flags,
+                       struct rte_qdma_rbp *rbp)
+{
+       int i;
+
+       i = rte_qdma_vq_create(lcore_id, flags);
+
+       memcpy(&qdma_vqs[i].rbp, rbp, sizeof(struct rte_qdma_rbp));
+
+       return i;
+}
+
 static void
 dpaa2_qdma_populate_fle(struct qbman_fle *fle,
+                       struct rte_qdma_rbp *rbp,
                        uint64_t src, uint64_t dest,
                        size_t len, uint32_t flags)
 {
@@ -368,10 +386,36 @@ dpaa2_qdma_populate_fle(struct qbman_fle *fle,
        DPAA2_SET_FLE_LEN(fle, (2 * (sizeof(struct qdma_sdd))));
 
        /* source and destination descriptor */
-       DPAA2_SET_SDD_RD_COHERENT(sdd); /* source descriptor CMD */
-       sdd++;
-       DPAA2_SET_SDD_WR_COHERENT(sdd); /* dest descriptor CMD */
+       if (rbp && rbp->enable) {
+               /* source */
+               sdd->read_cmd.portid = rbp->sportid;
+               sdd->rbpcmd_simple.pfid = rbp->spfid;
+               sdd->rbpcmd_simple.vfid = rbp->svfid;
+
+               if (rbp->srbp) {
+                       sdd->read_cmd.rbp = rbp->srbp;
+                       sdd->read_cmd.rdtype = DPAA2_RBP_MEM_RW;
+               } else {
+                       sdd->read_cmd.rdtype = dpaa2_coherent_no_alloc_cache;
+               }
+               sdd++;
+               /* destination */
+               sdd->write_cmd.portid = rbp->dportid;
+               sdd->rbpcmd_simple.pfid = rbp->dpfid;
+               sdd->rbpcmd_simple.vfid = rbp->dvfid;
+
+               if (rbp->drbp) {
+                       sdd->write_cmd.rbp = rbp->drbp;
+                       sdd->write_cmd.wrttype = DPAA2_RBP_MEM_RW;
+               } else {
+                       sdd->write_cmd.wrttype = dpaa2_coherent_alloc_cache;
+               }
 
+       } else {
+               sdd->read_cmd.rdtype = dpaa2_coherent_no_alloc_cache;
+               sdd++;
+               sdd->write_cmd.wrttype = dpaa2_coherent_alloc_cache;
+       }
        fle++;
        /* source frame list to source buffer */
        if (flags & RTE_QDMA_JOB_SRC_PHY) {
@@ -396,31 +440,57 @@ dpaa2_qdma_populate_fle(struct qbman_fle *fle,
        DPAA2_SET_FLE_FIN(fle);
 }
 
-int
-rte_qdma_vq_enqueue_multi(uint16_t vq_id,
-                         struct rte_qdma_job **job,
-                         uint16_t nb_jobs)
+static inline uint16_t dpdmai_dev_set_fd(struct qbman_fd *fd,
+                                       struct rte_qdma_job *job,
+                                       struct rte_qdma_rbp *rbp,
+                                       uint16_t vq_id)
 {
-       struct qdma_virt_queue *qdma_vq = &qdma_vqs[vq_id];
-       struct qdma_hw_queue *qdma_pq = qdma_vq->hw_queue;
-       struct dpaa2_dpdmai_dev *dpdmai_dev = qdma_pq->dpdmai_dev;
        struct qdma_io_meta *io_meta;
-       struct qbman_fd fd_arr[MAX_TX_RING_SLOTS];
-       struct dpaa2_queue *txq;
        struct qbman_fle *fle;
+       int ret = 0;
+       /*
+        * Get an FLE/SDD from FLE pool.
+        * Note: IO metadata is before the FLE and SDD memory.
+        */
+       ret = rte_mempool_get(qdma_dev.fle_pool, (void **)(&io_meta));
+       if (ret) {
+               DPAA2_QDMA_DP_DEBUG("Memory alloc failed for FLE");
+               return ret;
+       }
+
+       /* Set the metadata */
+       io_meta->cnxt = (size_t)job;
+       io_meta->id = vq_id;
+
+       fle = (struct qbman_fle *)(io_meta + 1);
+
+       DPAA2_SET_FD_ADDR(fd, DPAA2_VADDR_TO_IOVA(fle));
+       DPAA2_SET_FD_COMPOUND_FMT(fd);
+       DPAA2_SET_FD_FRC(fd, QDMA_SER_CTX);
+
+       /* Populate FLE */
+       memset(fle, 0, QDMA_FLE_POOL_SIZE);
+       dpaa2_qdma_populate_fle(fle, rbp, job->src, job->dest,
+                               job->len, job->flags);
+
+       return 0;
+}
+
+static int
+dpdmai_dev_enqueue_multi(struct dpaa2_dpdmai_dev *dpdmai_dev,
+                       uint16_t txq_id,
+                       uint16_t vq_id,
+                       struct rte_qdma_rbp *rbp,
+                       struct rte_qdma_job **job,
+                       uint16_t nb_jobs)
+{
+       struct qbman_fd fd[RTE_QDMA_BURST_NB_MAX];
+       struct dpaa2_queue *txq;
        struct qbman_eq_desc eqdesc;
        struct qbman_swp *swp;
        int ret;
        uint32_t num_to_send = 0;
        uint16_t num_tx = 0;
-       uint16_t num_txed = 0;
-
-       /* Return error in case of wrong lcore_id */
-       if (rte_lcore_id() != qdma_vq->lcore_id) {
-               DPAA2_QDMA_ERR("QDMA enqueue for vqid %d on wrong core",
-                               vq_id);
-               return -1;
-       }
 
        if (unlikely(!DPAA2_PER_LCORE_DPIO)) {
                ret = dpaa2_affine_qbman_swp();
@@ -431,7 +501,7 @@ rte_qdma_vq_enqueue_multi(uint16_t vq_id,
        }
        swp = DPAA2_PER_LCORE_PORTAL;
 
-       txq = &(dpdmai_dev->tx_queue[qdma_pq->queue_id]);
+       txq = &(dpdmai_dev->tx_queue[txq_id]);
 
        /* Prepare enqueue descriptor */
        qbman_eq_desc_clear(&eqdesc);
@@ -439,6 +509,8 @@ rte_qdma_vq_enqueue_multi(uint16_t vq_id,
        qbman_eq_desc_set_no_orp(&eqdesc, 0);
        qbman_eq_desc_set_response(&eqdesc, 0, 0);
 
+       memset(fd, 0, RTE_QDMA_BURST_NB_MAX * sizeof(struct qbman_fd));
+
        while (nb_jobs > 0) {
                uint32_t loop;
 
@@ -446,73 +518,100 @@ rte_qdma_vq_enqueue_multi(uint16_t vq_id,
                        dpaa2_eqcr_size : nb_jobs;
 
                for (loop = 0; loop < num_to_send; loop++) {
-                       /*
-                        * Get an FLE/SDD from FLE pool.
-                        * Note: IO metadata is before the FLE and SDD memory.
-                        */
-                       ret = rte_mempool_get(qdma_dev.fle_pool,
-                                       (void **)(&io_meta));
-                       if (ret) {
-                               DPAA2_QDMA_DP_WARN("Me alloc failed for FLE");
-                               return ret;
+                       ret = dpdmai_dev_set_fd(&fd[loop],
+                                               job[num_tx], rbp, vq_id);
+                       if (ret < 0) {
+                               /* Set nb_jobs to loop, so outer while loop
+                                * breaks out.
+                                */
+                               nb_jobs = loop;
+                               break;
                        }
 
-                       /* Set the metadata */
-                       io_meta->cnxt = (size_t)job[num_tx];
-                       io_meta->id = vq_id;
-
-                       fle = (struct qbman_fle *)(io_meta + 1);
-
-                       /* populate Frame descriptor */
-                       memset(&fd_arr[loop], 0, sizeof(struct qbman_fd));
-                       DPAA2_SET_FD_ADDR(&fd_arr[loop],
-                                       DPAA2_VADDR_TO_IOVA(fle));
-                       DPAA2_SET_FD_COMPOUND_FMT(&fd_arr[loop]);
-                       DPAA2_SET_FD_FRC(&fd_arr[loop], QDMA_SER_CTX);
-
-                       /* Populate FLE */
-                       memset(fle, 0, QDMA_FLE_POOL_SIZE);
-                       dpaa2_qdma_populate_fle(fle, job[num_tx]->src,
-                                               job[num_tx]->dest,
-                                               job[num_tx]->len,
-                                               job[num_tx]->flags);
-
                        num_tx++;
                }
 
                /* Enqueue the packet to the QBMAN */
                uint32_t enqueue_loop = 0;
-               while (enqueue_loop < num_to_send) {
+               while (enqueue_loop < loop) {
                        enqueue_loop += qbman_swp_enqueue_multiple(swp,
                                                &eqdesc,
-                                               &fd_arr[enqueue_loop],
+                                               &fd[enqueue_loop],
                                                NULL,
-                                               num_to_send - enqueue_loop);
+                                               loop - enqueue_loop);
                }
-
-               num_txed += num_to_send;
-               nb_jobs -= num_to_send;
+               nb_jobs -= loop;
        }
-       qdma_vq->num_enqueues += num_txed;
-       return num_txed;
+       return num_tx;
 }
 
 int
-rte_qdma_vq_enqueue(uint16_t vq_id,
-                   struct rte_qdma_job *job)
+rte_qdma_vq_enqueue_multi(uint16_t vq_id,
+                         struct rte_qdma_job **job,
+                         uint16_t nb_jobs)
 {
+       struct qdma_virt_queue *qdma_vq = &qdma_vqs[vq_id];
+       struct qdma_hw_queue *qdma_pq = qdma_vq->hw_queue;
+       struct dpaa2_dpdmai_dev *dpdmai_dev = qdma_pq->dpdmai_dev;
        int ret;
 
-       ret = rte_qdma_vq_enqueue_multi(vq_id, &job, 1);
+       DPAA2_QDMA_FUNC_TRACE();
+
+       /* Return error in case of wrong lcore_id */
+       if (rte_lcore_id() != qdma_vq->lcore_id) {
+               DPAA2_QDMA_ERR("QDMA enqueue for vqid %d on wrong core",
+                               vq_id);
+               return -EINVAL;
+       }
+
+       ret = dpdmai_dev_enqueue_multi(dpdmai_dev,
+                                qdma_pq->queue_id,
+                                vq_id,
+                                &qdma_vq->rbp,
+                                job,
+                                nb_jobs);
        if (ret < 0) {
                DPAA2_QDMA_ERR("DPDMAI device enqueue failed: %d", ret);
                return ret;
        }
 
-       return 1;
+       qdma_vq->num_enqueues += ret;
+
+       return ret;
+}
+
+int
+rte_qdma_vq_enqueue(uint16_t vq_id,
+                   struct rte_qdma_job *job)
+{
+       return rte_qdma_vq_enqueue_multi(vq_id, &job, 1);
+}
+
+static inline uint16_t dpdmai_dev_get_job(const struct qbman_fd *fd,
+                                       struct rte_qdma_job **job)
+{
+       struct qbman_fle *fle;
+       struct qdma_io_meta *io_meta;
+       uint16_t vqid;
+       /*
+        * Fetch metadata from FLE. job and vq_id were set
+        * in metadata in the enqueue operation.
+        */
+       fle = (struct qbman_fle *)DPAA2_IOVA_TO_VADDR(DPAA2_GET_FD_ADDR(fd));
+       io_meta = (struct qdma_io_meta *)(fle) - 1;
+
+       *job = (struct rte_qdma_job *)(size_t)io_meta->cnxt;
+       (*job)->status = (DPAA2_GET_FD_ERR(fd) << 8) |
+                        (DPAA2_GET_FD_FRC(fd) & 0xFF);
+
+       vqid = io_meta->id;
+
+       /* Free FLE to the pool */
+       rte_mempool_put(qdma_dev.fle_pool, io_meta);
+
+       return vqid;
 }
 
-/* Function to receive a QDMA job for a given device and queue*/
 static int
 dpdmai_dev_dequeue_multijob(struct dpaa2_dpdmai_dev *dpdmai_dev,
                   uint16_t rxq_id,
@@ -520,16 +619,18 @@ dpdmai_dev_dequeue_multijob(struct dpaa2_dpdmai_dev *dpdmai_dev,
                   struct rte_qdma_job **job,
                   uint16_t nb_jobs)
 {
-       struct qdma_io_meta *io_meta;
        struct dpaa2_queue *rxq;
        struct qbman_result *dq_storage;
        struct qbman_pull_desc pulldesc;
-       const struct qbman_fd *fd;
        struct qbman_swp *swp;
-       struct qbman_fle *fle;
        uint32_t fqid;
-       uint8_t status;
-       int ret;
+       uint8_t status, pending;
+       uint8_t num_rx = 0;
+       const struct qbman_fd *fd;
+       uint16_t vqid;
+       int ret, next_pull = nb_jobs, num_pulled = 0;
+
+       DPAA2_QDMA_FUNC_TRACE();
 
        if (unlikely(!DPAA2_PER_LCORE_DPIO)) {
                ret = dpaa2_affine_qbman_swp();
@@ -539,77 +640,75 @@ dpdmai_dev_dequeue_multijob(struct dpaa2_dpdmai_dev *dpdmai_dev,
                }
        }
        swp = DPAA2_PER_LCORE_PORTAL;
+
        rxq = &(dpdmai_dev->rx_queue[rxq_id]);
-       dq_storage = rxq->q_storage->dq_storage[0];
        fqid = rxq->fqid;
 
-       /* Prepare dequeue descriptor */
-       qbman_pull_desc_clear(&pulldesc);
-       qbman_pull_desc_set_fq(&pulldesc, fqid);
-       qbman_pull_desc_set_storage(&pulldesc, dq_storage,
-               (uint64_t)(DPAA2_VADDR_TO_IOVA(dq_storage)), 1);
-       if (nb_jobs > dpaa2_dqrr_size)
-               qbman_pull_desc_set_numframes(&pulldesc, dpaa2_dqrr_size);
-       else
-               qbman_pull_desc_set_numframes(&pulldesc, nb_jobs);
-
-       while (1) {
-               if (qbman_swp_pull(swp, &pulldesc)) {
-                       DPAA2_QDMA_DP_WARN("VDQ command not issued. QBMAN busy");
-                       continue;
+       do {
+               dq_storage = rxq->q_storage->dq_storage[0];
+               /* Prepare dequeue descriptor */
+               qbman_pull_desc_clear(&pulldesc);
+               qbman_pull_desc_set_fq(&pulldesc, fqid);
+               qbman_pull_desc_set_storage(&pulldesc, dq_storage,
+                       (uint64_t)(DPAA2_VADDR_TO_IOVA(dq_storage)), 1);
+
+               if (next_pull > dpaa2_dqrr_size) {
+                       qbman_pull_desc_set_numframes(&pulldesc,
+                                       dpaa2_dqrr_size);
+                       next_pull -= dpaa2_dqrr_size;
+               } else {
+                       qbman_pull_desc_set_numframes(&pulldesc, next_pull);
+                       next_pull = 0;
                }
-               break;
-       }
 
-       rte_prefetch0((void *)((size_t)(dq_storage + 1)));
-       /* Check if the previous issued command is completed. */
-       while (!qbman_check_command_complete(dq_storage))
-               ;
+               while (1) {
+                       if (qbman_swp_pull(swp, &pulldesc)) {
+                               DPAA2_QDMA_DP_WARN("VDQ command not issued. QBMAN busy");
+                               /* Portal was busy, try again */
+                               continue;
+                       }
+                       break;
+               }
 
-       int num_pulled = 0;
-       int pending = 1;
-       do {
-               /* Loop until the dq_storage is updated with
-                * new token by QBMAN
-                */
-               while (!qbman_check_new_result(dq_storage))
+               rte_prefetch0((void *)((size_t)(dq_storage + 1)));
+               /* Check if the previous issued command is completed. */
+               while (!qbman_check_command_complete(dq_storage))
                        ;
 
-               rte_prefetch0((void *)((size_t)(dq_storage + 2)));
-               /* Check whether Last Pull command is Expired and
-                * setting Condition for Loop termination
-                */
-               if (qbman_result_DQ_is_pull_complete(dq_storage)) {
-                       pending = 0;
-                       /* Check for valid frame. */
-                       status = qbman_result_DQ_flags(dq_storage);
-                       if (unlikely((status &
-                               QBMAN_DQ_STAT_VALIDFRAME) == 0))
-                               continue;
-               }
-               fd = qbman_result_DQ_fd(dq_storage);
+               num_pulled = 0;
+               pending = 1;
 
-               /*
-                * Fetch metadata from FLE. job and vq_id were set
-                * in metadata in the enqueue operation.
-                */
-               fle = (struct qbman_fle *)
-                               DPAA2_IOVA_TO_VADDR(DPAA2_GET_FD_ADDR(fd));
-               io_meta = (struct qdma_io_meta *)(fle) - 1;
-               if (vq_id)
-                       vq_id[num_pulled] = io_meta->id;
+               do {
+                       /* Loop until dq_storage is updated
+                        * with new token by QBMAN
+                        */
+                       while (!qbman_check_new_result(dq_storage))
+                               ;
+                       rte_prefetch0((void *)((size_t)(dq_storage + 2)));
+
+                       if (qbman_result_DQ_is_pull_complete(dq_storage)) {
+                               pending = 0;
+                               /* Check for valid frame. */
+                               status = qbman_result_DQ_flags(dq_storage);
+                               if (unlikely((status &
+                                       QBMAN_DQ_STAT_VALIDFRAME) == 0))
+                                       continue;
+                       }
+                       fd = qbman_result_DQ_fd(dq_storage);
 
-               job[num_pulled] = (struct rte_qdma_job *)(size_t)io_meta->cnxt;
-               job[num_pulled]->status = DPAA2_GET_FD_ERR(fd);
+                       vqid = dpdmai_dev_get_job(fd, &job[num_rx]);
+                       if (vq_id)
+                               vq_id[num_rx] = vqid;
 
-               /* Free FLE to the pool */
-               rte_mempool_put(qdma_dev.fle_pool, io_meta);
+                       dq_storage++;
+                       num_rx++;
+                       num_pulled++;
 
-               dq_storage++;
-               num_pulled++;
-       } while (pending && (num_pulled <= dpaa2_dqrr_size));
+               } while (pending);
+       /* Last VDQ provided all packets and more packets are requested */
+       } while (next_pull && num_pulled == dpaa2_dqrr_size);
 
-       return num_pulled;
+       return num_rx;
 }
 
 int
@@ -664,9 +763,9 @@ rte_qdma_vq_dequeue_multi(uint16_t vq_id,
                                temp_qdma_vq = &qdma_vqs[temp_vq_id[i]];
                                rte_ring_enqueue(temp_qdma_vq->status_ring,
                                        (void *)(job[i]));
-                               ring_count = rte_ring_count(
-                                       qdma_vq->status_ring);
                        }
+                       ring_count = rte_ring_count(
+                                       qdma_vq->status_ring);
                }
 
                if (ring_count) {
@@ -743,6 +842,35 @@ rte_qdma_vq_destroy(uint16_t vq_id)
        return 0;
 }
 
+int
+rte_qdma_vq_destroy_rbp(uint16_t vq_id)
+{
+       struct qdma_virt_queue *qdma_vq = &qdma_vqs[vq_id];
+
+       DPAA2_QDMA_FUNC_TRACE();
+
+       /* In case there are pending jobs on any VQ, return -EBUSY */
+       if (qdma_vq->num_enqueues != qdma_vq->num_dequeues)
+               return -EBUSY;
+
+       rte_spinlock_lock(&qdma_dev.lock);
+
+       if (qdma_vq->exclusive_hw_queue) {
+               free_hw_queue(qdma_vq->hw_queue);
+       } else {
+               if (qdma_vqs->status_ring)
+                       rte_ring_free(qdma_vqs->status_ring);
+
+               put_hw_queue(qdma_vq->hw_queue);
+       }
+
+       memset(qdma_vq, 0, sizeof(struct qdma_virt_queue));
+
+       rte_spinlock_lock(&qdma_dev.lock);
+
+       return 0;
+}
+
 void
 rte_qdma_stop(void)
 {
@@ -939,6 +1067,21 @@ dpaa2_dpdmai_dev_init(struct rte_rawdev *rawdev, int dpdmai_id)
                DPAA2_QDMA_ERR("Adding H/W queue to list failed");
                goto init_err;
        }
+
+       if (!dpaa2_coherent_no_alloc_cache) {
+               if (dpaa2_svr_family == SVR_LX2160A) {
+                       dpaa2_coherent_no_alloc_cache =
+                               DPAA2_LX2_COHERENT_NO_ALLOCATE_CACHE;
+                       dpaa2_coherent_alloc_cache =
+                               DPAA2_LX2_COHERENT_ALLOCATE_CACHE;
+               } else {
+                       dpaa2_coherent_no_alloc_cache =
+                               DPAA2_COHERENT_NO_ALLOCATE_CACHE;
+                       dpaa2_coherent_alloc_cache =
+                               DPAA2_COHERENT_ALLOCATE_CACHE;
+               }
+       }
+
        DPAA2_QDMA_DEBUG("Initialized dpdmai object successfully");
 
        return 0;
index 0cbe902..f15dda6 100644 (file)
@@ -22,28 +22,24 @@ struct qdma_io_meta;
 
 /** Notification by FQD_CTX[fqid] */
 #define QDMA_SER_CTX (1 << 8)
-
+#define DPAA2_RBP_MEM_RW            0x0
 /**
  * Source descriptor command read transaction type for RBP=0:
  * coherent copy of cacheable memory
  */
-#define DPAA2_SET_SDD_RD_COHERENT(sdd) ((sdd)->cmd = (0xb << 28))
+#define DPAA2_COHERENT_NO_ALLOCATE_CACHE       0xb
+#define DPAA2_LX2_COHERENT_NO_ALLOCATE_CACHE   0x7
 /**
  * Destination descriptor command write transaction type for RBP=0:
  * coherent copy of cacheable memory
  */
-#define DPAA2_SET_SDD_WR_COHERENT(sdd) ((sdd)->cmd = (0x6 << 28))
+#define DPAA2_COHERENT_ALLOCATE_CACHE          0x6
+#define DPAA2_LX2_COHERENT_ALLOCATE_CACHE      0xb
 
 /** Maximum possible H/W Queues on each core */
 #define MAX_HW_QUEUE_PER_CORE          64
 
-/**
- * In case of Virtual Queue mode, this specifies the number of
- * dequeue the 'qdma_vq_dequeue/multi' API does from the H/W Queue
- * in case there is no job present on the Virtual Queue ring.
- */
-#define QDMA_DEQUEUE_BUDGET            64
-
+#define QDMA_RBP_UPPER_ADDRESS_MASK (0xfff0000000000)
 /**
  * Represents a QDMA device.
  * A single QDMA device exists which is combination of multiple DPDMAI rawdev's.
@@ -90,6 +86,8 @@ struct qdma_virt_queue {
        struct rte_ring *status_ring;
        /** Associated hw queue */
        struct qdma_hw_queue *hw_queue;
+       /** Route by port */
+       struct rte_qdma_rbp rbp;
        /** Associated lcore id */
        uint32_t lcore_id;
        /** States if this vq is in use or not */
@@ -118,7 +116,7 @@ struct qdma_io_meta {
         */
        uint64_t cnxt;
        /** VQ ID is stored as a part of metadata of the enqueue command */
-        uint64_t id;
+       uint64_t id;
 };
 
 /** Source/Destination Descriptor */
@@ -127,9 +125,48 @@ struct qdma_sdd {
        /** Stride configuration */
        uint32_t stride;
        /** Route-by-port command */
-       uint32_t rbpcmd;
-       uint32_t cmd;
-} __attribute__((__packed__));
+       union {
+               uint32_t rbpcmd;
+               struct rbpcmd_st {
+                       uint32_t vfid:6;
+                       uint32_t rsv4:2;
+                       uint32_t pfid:1;
+                       uint32_t rsv3:7;
+                       uint32_t attr:3;
+                       uint32_t rsv2:1;
+                       uint32_t at:2;
+                       uint32_t vfa:1;
+                       uint32_t ca:1;
+                       uint32_t tc:3;
+                       uint32_t rsv1:5;
+               } rbpcmd_simple;
+       };
+       union {
+               uint32_t cmd;
+               struct rcmd_simple {
+                       uint32_t portid:4;
+                       uint32_t rsv1:14;
+                       uint32_t rbp:1;
+                       uint32_t ssen:1;
+                       uint32_t rthrotl:4;
+                       uint32_t sqos:3;
+                       uint32_t ns:1;
+                       uint32_t rdtype:4;
+               } read_cmd;
+               struct wcmd_simple {
+                       uint32_t portid:4;
+                       uint32_t rsv3:10;
+                       uint32_t rsv2:2;
+                       uint32_t lwc:2;
+                       uint32_t rbp:1;
+                       uint32_t dsen:1;
+                       uint32_t rsv1:4;
+                       uint32_t dqos:3;
+                       uint32_t ns:1;
+                       uint32_t wrttype:4;
+               } write_cmd;
+       };
+} __attribute__ ((__packed__));
 
 /** Represents a DPDMAI raw device */
 struct dpaa2_dpdmai_dev {
index e1ccc19..bbc66a2 100644 (file)
@@ -13,7 +13,7 @@
  */
 
 /** Maximum qdma burst size */
-#define RTE_QDMA_BURST_NB_MAX 32
+#define RTE_QDMA_BURST_NB_MAX 256
 
 /** Determines the mode of operation */
 enum {
@@ -73,6 +73,40 @@ struct rte_qdma_config {
        int fle_pool_count;
 };
 
+struct rte_qdma_rbp {
+       uint32_t use_ultrashort:1;
+       uint32_t enable:1;
+       /**
+        * dportid:
+        * 0000 PCI-Express 1
+        * 0001 PCI-Express 2
+        * 0010 PCI-Express 3
+        * 0011 PCI-Express 4
+        * 0100 PCI-Express 5
+        * 0101 PCI-Express 6
+        */
+       uint32_t dportid:4;
+       uint32_t dpfid:2;
+       uint32_t dvfid:6;
+       /*using route by port for destination */
+       uint32_t drbp:1;
+       /**
+        * sportid:
+        * 0000 PCI-Express 1
+        * 0001 PCI-Express 2
+        * 0010 PCI-Express 3
+        * 0011 PCI-Express 4
+        * 0100 PCI-Express 5
+        * 0101 PCI-Express 6
+        */
+       uint32_t sportid:4;
+       uint32_t spfid:2;
+       uint32_t svfid:6;
+       /* using route by port for source */
+       uint32_t srbp:1;
+       uint32_t rsv:4;
+};
+
 /** Provides QDMA device statistics */
 struct rte_qdma_vq_stats {
        /** States if this vq has exclusively associated hw queue */
@@ -105,8 +139,10 @@ struct rte_qdma_job {
        /**
         * Status of the transaction.
         * This is filled in the dequeue operation by the driver.
+        * upper 8bits acc_err for route by port.
+        * lower 8bits fd error
         */
-       uint8_t status;
+       uint16_t status;
 };
 
 /**
@@ -177,6 +213,11 @@ rte_qdma_start(void);
 int
 rte_qdma_vq_create(uint32_t lcore_id, uint32_t flags);
 
+/*create vq for route-by-port*/
+int
+rte_qdma_vq_create_rbp(uint32_t lcore_id, uint32_t flags,
+                       struct rte_qdma_rbp *rbp);
+
 /**
  * Enqueue multiple jobs to a Virtual Queue.
  * If the enqueue is successful, the H/W will perform DMA operations
@@ -275,6 +316,21 @@ rte_qdma_vq_stats(uint16_t vq_id,
 int
 rte_qdma_vq_destroy(uint16_t vq_id);
 
+/**
+ * Destroy the RBP specific Virtual Queue specified by vq_id.
+ * This API can be called from any thread/core. User can create/destroy
+ * VQ's at runtime.
+ *
+ * @param vq_id
+ *   RBP based Virtual Queue ID which needs to be deinialized.
+ *
+ * @returns
+ *   - 0: Success.
+ *   - <0: Error code.
+ */
+
+int __rte_experimental
+rte_qdma_vq_destroy_rbp(uint16_t vq_id);
 /**
  * Stop QDMA device.
  */