raw/dpaa2_qdma: refactor the code
authorJun Yang <jun.yang@nxp.com>
Thu, 15 Oct 2020 09:47:04 +0000 (17:47 +0800)
committerThomas Monjalon <thomas@monjalon.net>
Mon, 19 Oct 2020 11:57:41 +0000 (13:57 +0200)
This patch moves qdma queue specific configurations from driver
global configuration to per-queue setup. This is required
as each queue can be configured differently.

Signed-off-by: Jun Yang <jun.yang@nxp.com>
drivers/raw/dpaa2_qdma/dpaa2_qdma.c
drivers/raw/dpaa2_qdma/dpaa2_qdma.h
drivers/raw/dpaa2_qdma/rte_pmd_dpaa2_qdma.h

index 09e295f..6eef7a5 100644 (file)
@@ -26,6 +26,9 @@
 
 #define DPAA2_QDMA_NO_PREFETCH "no_prefetch"
 
+/* Dynamic log type identifier */
+int dpaa2_qdma_logtype;
+
 uint32_t dpaa2_coherent_no_alloc_cache;
 uint32_t dpaa2_coherent_alloc_cache;
 
@@ -37,31 +40,9 @@ TAILQ_HEAD(qdma_hw_queue_list, qdma_hw_queue);
 static struct qdma_hw_queue_list qdma_queue_list
        = TAILQ_HEAD_INITIALIZER(qdma_queue_list);
 
-/* QDMA Virtual Queues */
-static struct qdma_virt_queue *qdma_vqs;
-
 /* QDMA per core data */
 static struct qdma_per_core_info qdma_core_info[RTE_MAX_LCORE];
 
-typedef int (dpdmai_dev_dequeue_multijob_t)(struct dpaa2_dpdmai_dev *dpdmai_dev,
-                                           uint16_t rxq_id,
-                                           uint16_t *vq_id,
-                                           struct rte_qdma_job **job,
-                                           uint16_t nb_jobs);
-
-dpdmai_dev_dequeue_multijob_t *dpdmai_dev_dequeue_multijob;
-
-typedef uint16_t (dpdmai_dev_get_job_t)(struct qdma_device *qdma_dev,
-                                       const struct qbman_fd *fd,
-                                       struct rte_qdma_job **job);
-typedef int (dpdmai_dev_set_fd_t)(struct qdma_device *qdma_dev,
-                                 struct qbman_fd *fd,
-                                 struct rte_qdma_job *job,
-                                 struct rte_qdma_rbp *rbp,
-                                 uint16_t vq_id);
-dpdmai_dev_get_job_t *dpdmai_dev_get_job;
-dpdmai_dev_set_fd_t *dpdmai_dev_set_fd;
-
 static inline int
 qdma_populate_fd_pci(phys_addr_t src, phys_addr_t dest,
                        uint32_t len, struct qbman_fd *fd,
@@ -114,7 +95,7 @@ qdma_populate_fd_ddr(phys_addr_t src, phys_addr_t dest,
        /**
         * src If RBP=0 {NS,RDTTYPE[3:0]}: 0_1011
         * Coherent copy of cacheable memory,
-        * lookup in downstream cache, no allocate
+       * lookup in downstream cache, no allocate
         * on miss
         */
        fd->simple_ddr.rns = 0;
@@ -204,12 +185,11 @@ dpaa2_qdma_populate_fle(struct qbman_fle *fle,
 }
 
 static inline int dpdmai_dev_set_fd_us(
-                               struct qdma_device *qdma_dev __rte_unused,
-                               struct qbman_fd *fd,
-                               struct rte_qdma_job *job,
-                               struct rte_qdma_rbp *rbp,
-                               uint16_t vq_id)
+               struct qdma_virt_queue *qdma_vq,
+               struct qbman_fd *fd,
+               struct rte_qdma_job *job)
 {
+       struct rte_qdma_rbp *rbp = &qdma_vq->rbp;
        struct rte_qdma_job **ppjob;
        size_t iova;
        int ret = 0;
@@ -220,7 +200,7 @@ static inline int dpdmai_dev_set_fd_us(
                iova = (size_t)job->src;
 
        /* Set the metadata */
-       job->vq_id = vq_id;
+       job->vq_id = qdma_vq->vq_id;
        ppjob = (struct rte_qdma_job **)DPAA2_IOVA_TO_VADDR(iova) - 1;
        *ppjob = job;
 
@@ -234,15 +214,17 @@ static inline int dpdmai_dev_set_fd_us(
                                           job->len, fd);
        return ret;
 }
-static inline int dpdmai_dev_set_fd_lf(struct qdma_device *qdma_dev,
-                                       struct qbman_fd *fd,
-                                       struct rte_qdma_job *job,
-                                       struct rte_qdma_rbp *rbp,
-                                       uint16_t vq_id)
+static inline int dpdmai_dev_set_fd_lf(
+               struct qdma_virt_queue *qdma_vq,
+               struct qbman_fd *fd,
+               struct rte_qdma_job *job)
 {
+       struct rte_qdma_rbp *rbp = &qdma_vq->rbp;
        struct rte_qdma_job **ppjob;
        struct qbman_fle *fle;
        int ret = 0;
+       struct qdma_device *qdma_dev = QDMA_DEV_OF_VQ(qdma_vq);
+
        /*
         * Get an FLE/SDD from FLE pool.
         * Note: IO metadata is before the FLE and SDD memory.
@@ -254,7 +236,7 @@ static inline int dpdmai_dev_set_fd_lf(struct qdma_device *qdma_dev,
        }
 
        /* Set the metadata */
-       job->vq_id = vq_id;
+       job->vq_id = qdma_vq->vq_id;
        *ppjob = job;
 
        fle = (struct qbman_fle *)(ppjob + 1);
@@ -272,7 +254,7 @@ static inline int dpdmai_dev_set_fd_lf(struct qdma_device *qdma_dev,
 }
 
 static inline uint16_t dpdmai_dev_get_job_us(
-                               struct qdma_device *qdma_dev __rte_unused,
+                               struct qdma_virt_queue *qdma_vq __rte_unused,
                                const struct qbman_fd *fd,
                                struct rte_qdma_job **job)
 {
@@ -281,7 +263,7 @@ static inline uint16_t dpdmai_dev_get_job_us(
        struct rte_qdma_job **ppjob;
 
        if (fd->simple_pci.saddr_hi & (QDMA_RBP_UPPER_ADDRESS_MASK >> 32))
-               iova = (size_t) (((uint64_t)fd->simple_pci.daddr_hi) << 32
+               iova = (size_t)(((uint64_t)fd->simple_pci.daddr_hi) << 32
                                | (uint64_t)fd->simple_pci.daddr_lo);
        else
                iova = (size_t)(((uint64_t)fd->simple_pci.saddr_hi) << 32
@@ -289,18 +271,22 @@ static inline uint16_t dpdmai_dev_get_job_us(
 
        ppjob = (struct rte_qdma_job **)DPAA2_IOVA_TO_VADDR(iova) - 1;
        *job = (struct rte_qdma_job *)*ppjob;
-       (*job)->status = (fd->simple_pci.acc_err << 8) | (fd->simple_pci.error);
+       (*job)->status = (fd->simple_pci.acc_err << 8) |
+                                       (fd->simple_pci.error);
        vqid = (*job)->vq_id;
 
        return vqid;
 }
 
-static inline uint16_t dpdmai_dev_get_job_lf(struct qdma_device *qdma_dev,
-                                            const struct qbman_fd *fd,
-                                            struct rte_qdma_job **job)
+static inline uint16_t dpdmai_dev_get_job_lf(
+                                               struct qdma_virt_queue *qdma_vq,
+                                               const struct qbman_fd *fd,
+                                               struct rte_qdma_job **job)
 {
        struct rte_qdma_job **ppjob;
        uint16_t vqid;
+       struct qdma_device *qdma_dev = QDMA_DEV_OF_VQ(qdma_vq);
+
        /*
         * Fetch metadata from FLE. job and vq_id were set
         * in metadata in the enqueue operation.
@@ -320,351 +306,268 @@ static inline uint16_t dpdmai_dev_get_job_lf(struct qdma_device *qdma_dev,
        return vqid;
 }
 
-static struct qdma_hw_queue *
-alloc_hw_queue(uint32_t lcore_id)
+/* Function to receive a QDMA job for a given device and queue*/
+static int
+dpdmai_dev_dequeue_multijob_prefetch(
+                       struct qdma_virt_queue *qdma_vq,
+                       uint16_t *vq_id,
+                       struct rte_qdma_job **job,
+                       uint16_t nb_jobs)
 {
-       struct qdma_hw_queue *queue = NULL;
+       struct qdma_hw_queue *qdma_pq = qdma_vq->hw_queue;
+       struct dpaa2_dpdmai_dev *dpdmai_dev = qdma_pq->dpdmai_dev;
+       uint16_t rxq_id = qdma_pq->queue_id;
 
-       DPAA2_QDMA_FUNC_TRACE();
+       struct dpaa2_queue *rxq;
+       struct qbman_result *dq_storage, *dq_storage1 = NULL;
+       struct qbman_pull_desc pulldesc;
+       struct qbman_swp *swp;
+       struct queue_storage_info_t *q_storage;
+       uint32_t fqid;
+       uint8_t status, pending;
+       uint8_t num_rx = 0;
+       const struct qbman_fd *fd;
+       uint16_t vqid;
+       int ret, pull_size;
 
-       /* Get a free queue from the list */
-       TAILQ_FOREACH(queue, &qdma_queue_list, next) {
-               if (queue->num_users == 0) {
-                       queue->lcore_id = lcore_id;
-                       queue->num_users++;
-                       break;
+       if (unlikely(!DPAA2_PER_LCORE_DPIO)) {
+               ret = dpaa2_affine_qbman_swp();
+               if (ret) {
+                       DPAA2_QDMA_ERR(
+                               "Failed to allocate IO portal, tid: %d\n",
+                               rte_gettid());
+                       return 0;
                }
        }
+       swp = DPAA2_PER_LCORE_PORTAL;
 
-       return queue;
-}
-
-static void
-free_hw_queue(struct qdma_hw_queue *queue)
-{
-       DPAA2_QDMA_FUNC_TRACE();
-
-       queue->num_users--;
-}
-
-
-static struct qdma_hw_queue *
-get_hw_queue(struct qdma_device *qdma_dev, uint32_t lcore_id)
-{
-       struct qdma_per_core_info *core_info;
-       struct qdma_hw_queue *queue, *temp;
-       uint32_t least_num_users;
-       int num_hw_queues, i;
-
-       DPAA2_QDMA_FUNC_TRACE();
-
-       core_info = &qdma_core_info[lcore_id];
-       num_hw_queues = core_info->num_hw_queues;
+       pull_size = (nb_jobs > dpaa2_dqrr_size) ? dpaa2_dqrr_size : nb_jobs;
+       rxq = &(dpdmai_dev->rx_queue[rxq_id]);
+       fqid = rxq->fqid;
+       q_storage = rxq->q_storage;
 
-       /*
-        * Allocate a HW queue if there are less queues
-        * than maximum per core queues configured
-        */
-       if (num_hw_queues < qdma_dev->max_hw_queues_per_core) {
-               queue = alloc_hw_queue(lcore_id);
-               if (queue) {
-                       core_info->hw_queues[num_hw_queues] = queue;
-                       core_info->num_hw_queues++;
-                       return queue;
+       if (unlikely(!q_storage->active_dqs)) {
+               q_storage->toggle = 0;
+               dq_storage = q_storage->dq_storage[q_storage->toggle];
+               q_storage->last_num_pkts = pull_size;
+               qbman_pull_desc_clear(&pulldesc);
+               qbman_pull_desc_set_numframes(&pulldesc,
+                                             q_storage->last_num_pkts);
+               qbman_pull_desc_set_fq(&pulldesc, fqid);
+               qbman_pull_desc_set_storage(&pulldesc, dq_storage,
+                               (size_t)(DPAA2_VADDR_TO_IOVA(dq_storage)), 1);
+               if (check_swp_active_dqs(DPAA2_PER_LCORE_DPIO->index)) {
+                       while (!qbman_check_command_complete(
+                               get_swp_active_dqs(
+                               DPAA2_PER_LCORE_DPIO->index)))
+                               ;
+                       clear_swp_active_dqs(DPAA2_PER_LCORE_DPIO->index);
+               }
+               while (1) {
+                       if (qbman_swp_pull(swp, &pulldesc)) {
+                               DPAA2_QDMA_DP_WARN(
+                                       "VDQ command not issued.QBMAN busy\n");
+                                       /* Portal was busy, try again */
+                               continue;
+                       }
+                       break;
                }
+               q_storage->active_dqs = dq_storage;
+               q_storage->active_dpio_id = DPAA2_PER_LCORE_DPIO->index;
+               set_swp_active_dqs(DPAA2_PER_LCORE_DPIO->index,
+                                  dq_storage);
        }
 
-       queue = core_info->hw_queues[0];
-       /* In case there is no queue associated with the core return NULL */
-       if (!queue)
-               return NULL;
+       dq_storage = q_storage->active_dqs;
+       rte_prefetch0((void *)(size_t)(dq_storage));
+       rte_prefetch0((void *)(size_t)(dq_storage + 1));
 
-       /* Fetch the least loaded H/W queue */
-       least_num_users = core_info->hw_queues[0]->num_users;
-       for (i = 0; i < num_hw_queues; i++) {
-               temp = core_info->hw_queues[i];
-               if (temp->num_users < least_num_users)
-                       queue = temp;
-       }
+       /* Prepare next pull descriptor. This will give space for the
+        * prefething done on DQRR entries
+        */
+       q_storage->toggle ^= 1;
+       dq_storage1 = q_storage->dq_storage[q_storage->toggle];
+       qbman_pull_desc_clear(&pulldesc);
+       qbman_pull_desc_set_numframes(&pulldesc, pull_size);
+       qbman_pull_desc_set_fq(&pulldesc, fqid);
+       qbman_pull_desc_set_storage(&pulldesc, dq_storage1,
+               (size_t)(DPAA2_VADDR_TO_IOVA(dq_storage1)), 1);
 
-       if (queue)
-               queue->num_users++;
+       /* Check if the previous issued command is completed.
+        * Also seems like the SWP is shared between the Ethernet Driver
+        * and the SEC driver.
+        */
+       while (!qbman_check_command_complete(dq_storage))
+               ;
+       if (dq_storage == get_swp_active_dqs(q_storage->active_dpio_id))
+               clear_swp_active_dqs(q_storage->active_dpio_id);
 
-       return queue;
-}
+       pending = 1;
 
-static void
-put_hw_queue(struct qdma_hw_queue *queue)
-{
-       struct qdma_per_core_info *core_info;
-       int lcore_id, num_hw_queues, i;
+       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 + 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);
 
-       DPAA2_QDMA_FUNC_TRACE();
+               vqid = qdma_vq->get_job(qdma_vq, fd, &job[num_rx]);
+               if (vq_id)
+                       vq_id[num_rx] = vqid;
 
-       /*
-        * If this is the last user of the queue free it.
-        * Also remove it from QDMA core info.
-        */
-       if (queue->num_users == 1) {
-               free_hw_queue(queue);
+               dq_storage++;
+               num_rx++;
+       } while (pending);
 
-               /* Remove the physical queue from core info */
-               lcore_id = queue->lcore_id;
-               core_info = &qdma_core_info[lcore_id];
-               num_hw_queues = core_info->num_hw_queues;
-               for (i = 0; i < num_hw_queues; i++) {
-                       if (queue == core_info->hw_queues[i])
-                               break;
+       if (check_swp_active_dqs(DPAA2_PER_LCORE_DPIO->index)) {
+               while (!qbman_check_command_complete(
+                       get_swp_active_dqs(DPAA2_PER_LCORE_DPIO->index)))
+                       ;
+               clear_swp_active_dqs(DPAA2_PER_LCORE_DPIO->index);
+       }
+       /* issue a volatile dequeue command for next pull */
+       while (1) {
+               if (qbman_swp_pull(swp, &pulldesc)) {
+                       DPAA2_QDMA_DP_WARN(
+                               "VDQ command is not issued. QBMAN is busy (2)\n");
+                       continue;
                }
-               for (; i < num_hw_queues - 1; i++)
-                       core_info->hw_queues[i] = core_info->hw_queues[i + 1];
-               core_info->hw_queues[i] = NULL;
-       } else {
-               queue->num_users--;
+               break;
        }
-}
-
-static int
-dpaa2_qdma_attr_get(struct rte_rawdev *rawdev,
-                   __rte_unused const char *attr_name,
-                   uint64_t *attr_value)
-{
-       struct dpaa2_dpdmai_dev *dpdmai_dev = rawdev->dev_private;
-       struct qdma_device *qdma_dev = dpdmai_dev->qdma_dev;
-       struct rte_qdma_attr *qdma_attr = (struct rte_qdma_attr *)attr_value;
-
-       DPAA2_QDMA_FUNC_TRACE();
 
-       qdma_attr->num_hw_queues = qdma_dev->num_hw_queues;
+       q_storage->active_dqs = dq_storage1;
+       q_storage->active_dpio_id = DPAA2_PER_LCORE_DPIO->index;
+       set_swp_active_dqs(DPAA2_PER_LCORE_DPIO->index, dq_storage1);
 
-       return 0;
+       return num_rx;
 }
 
 static int
-dpaa2_qdma_reset(struct rte_rawdev *rawdev)
+dpdmai_dev_dequeue_multijob_no_prefetch(
+               struct qdma_virt_queue *qdma_vq,
+               uint16_t *vq_id,
+               struct rte_qdma_job **job,
+               uint16_t nb_jobs)
 {
-       struct qdma_hw_queue *queue;
-       struct dpaa2_dpdmai_dev *dpdmai_dev = rawdev->dev_private;
-       struct qdma_device *qdma_dev = dpdmai_dev->qdma_dev;
-       int i;
+       struct qdma_hw_queue *qdma_pq = qdma_vq->hw_queue;
+       struct dpaa2_dpdmai_dev *dpdmai_dev = qdma_pq->dpdmai_dev;
+       uint16_t rxq_id = qdma_pq->queue_id;
 
-       DPAA2_QDMA_FUNC_TRACE();
+       struct dpaa2_queue *rxq;
+       struct qbman_result *dq_storage;
+       struct qbman_pull_desc pulldesc;
+       struct qbman_swp *swp;
+       uint32_t fqid;
+       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;
 
-       /* In case QDMA device is not in stopped state, return -EBUSY */
-       if (qdma_dev->state == 1) {
-               DPAA2_QDMA_ERR(
-                       "Device is in running state. Stop before reset.");
-               return -EBUSY;
+       if (unlikely(!DPAA2_PER_LCORE_DPIO)) {
+               ret = dpaa2_affine_qbman_swp();
+               if (ret) {
+                       DPAA2_QDMA_ERR(
+                               "Failed to allocate IO portal, tid: %d\n",
+                               rte_gettid());
+                       return 0;
+               }
        }
+       swp = DPAA2_PER_LCORE_PORTAL;
 
-       /* In case there are pending jobs on any VQ, return -EBUSY */
-       for (i = 0; i < qdma_dev->max_vqs; i++) {
-               if (qdma_vqs[i].in_use && (qdma_vqs[i].num_enqueues !=
-                   qdma_vqs[i].num_dequeues)) {
-                       DPAA2_QDMA_ERR("Jobs are still pending on VQ: %d", i);
-                       return -EBUSY;
-               }
-       }
-
-       /* Reset HW queues */
-       TAILQ_FOREACH(queue, &qdma_queue_list, next)
-               queue->num_users = 0;
-
-       /* Reset and free virtual queues */
-       for (i = 0; i < qdma_dev->max_vqs; i++) {
-               if (qdma_vqs[i].status_ring)
-                       rte_ring_free(qdma_vqs[i].status_ring);
-       }
-       if (qdma_vqs)
-               rte_free(qdma_vqs);
-       qdma_vqs = NULL;
-
-       /* Reset per core info */
-       memset(&qdma_core_info, 0,
-               sizeof(struct qdma_per_core_info) * RTE_MAX_LCORE);
-
-       /* Free the FLE pool */
-       if (qdma_dev->fle_pool)
-               rte_mempool_free(qdma_dev->fle_pool);
-
-       /* Reset QDMA device structure */
-       qdma_dev->mode = RTE_QDMA_MODE_HW;
-       qdma_dev->max_hw_queues_per_core = 0;
-       qdma_dev->fle_pool = NULL;
-       qdma_dev->fle_pool_count = 0;
-       qdma_dev->max_vqs = 0;
-
-       return 0;
-}
-
-static int
-dpaa2_qdma_configure(const struct rte_rawdev *rawdev,
-                        rte_rawdev_obj_t config,
-                        size_t config_size)
-{
-       char fle_pool_name[32]; /* RTE_MEMZONE_NAMESIZE = 32 */
-       struct rte_qdma_config *qdma_config = (struct rte_qdma_config *)config;
-       struct dpaa2_dpdmai_dev *dpdmai_dev = rawdev->dev_private;
-       struct qdma_device *qdma_dev = dpdmai_dev->qdma_dev;
-
-       DPAA2_QDMA_FUNC_TRACE();
-
-       if (config_size != sizeof(*qdma_config))
-               return -EINVAL;
-
-       /* In case QDMA device is not in stopped state, return -EBUSY */
-       if (qdma_dev->state == 1) {
-               DPAA2_QDMA_ERR(
-                       "Device is in running state. Stop before config.");
-               return -1;
-       }
-
-       /* Set mode */
-       qdma_dev->mode = qdma_config->mode;
-
-       /* Set max HW queue per core */
-       if (qdma_config->max_hw_queues_per_core > MAX_HW_QUEUE_PER_CORE) {
-               DPAA2_QDMA_ERR("H/W queues per core is more than: %d",
-                              MAX_HW_QUEUE_PER_CORE);
-               return -EINVAL;
-       }
-       qdma_dev->max_hw_queues_per_core =
-               qdma_config->max_hw_queues_per_core;
-
-       /* Allocate Virtual Queues */
-       qdma_vqs = rte_malloc("qdma_virtual_queues",
-                       (sizeof(struct qdma_virt_queue) * qdma_config->max_vqs),
-                       RTE_CACHE_LINE_SIZE);
-       if (!qdma_vqs) {
-               DPAA2_QDMA_ERR("qdma_virtual_queues allocation failed");
-               return -ENOMEM;
-       }
-       qdma_dev->max_vqs = qdma_config->max_vqs;
-
-       /* Allocate FLE pool; just append PID so that in case of
-        * multiprocess, the pool's don't collide.
-        */
-       snprintf(fle_pool_name, sizeof(fle_pool_name), "qdma_fle_pool%u",
-                getpid());
-       qdma_dev->fle_pool = rte_mempool_create(fle_pool_name,
-                       qdma_config->fle_pool_count, QDMA_FLE_POOL_SIZE,
-                       QDMA_FLE_CACHE_SIZE(qdma_config->fle_pool_count), 0,
-                       NULL, NULL, NULL, NULL, SOCKET_ID_ANY, 0);
-       if (!qdma_dev->fle_pool) {
-               DPAA2_QDMA_ERR("qdma_fle_pool create failed");
-               rte_free(qdma_vqs);
-               qdma_vqs = NULL;
-               return -ENOMEM;
-       }
-       qdma_dev->fle_pool_count = qdma_config->fle_pool_count;
-
-       if (qdma_config->format == RTE_QDMA_ULTRASHORT_FORMAT) {
-               dpdmai_dev_get_job = dpdmai_dev_get_job_us;
-               dpdmai_dev_set_fd = dpdmai_dev_set_fd_us;
-       } else {
-               dpdmai_dev_get_job = dpdmai_dev_get_job_lf;
-               dpdmai_dev_set_fd = dpdmai_dev_set_fd_lf;
-       }
-       return 0;
-}
-
-static int
-dpaa2_qdma_start(struct rte_rawdev *rawdev)
-{
-       struct dpaa2_dpdmai_dev *dpdmai_dev = rawdev->dev_private;
-       struct qdma_device *qdma_dev = dpdmai_dev->qdma_dev;
-
-       DPAA2_QDMA_FUNC_TRACE();
-
-       qdma_dev->state = 1;
-
-       return 0;
-}
-
-static int
-dpaa2_qdma_queue_setup(struct rte_rawdev *rawdev,
-                         __rte_unused uint16_t queue_id,
-                         rte_rawdev_obj_t queue_conf,
-                         size_t conf_size)
-{
-       char ring_name[32];
-       int i;
-       struct dpaa2_dpdmai_dev *dpdmai_dev = rawdev->dev_private;
-       struct qdma_device *qdma_dev = dpdmai_dev->qdma_dev;
-       struct rte_qdma_queue_config *q_config =
-               (struct rte_qdma_queue_config *)queue_conf;
-
-       DPAA2_QDMA_FUNC_TRACE();
+       rxq = &(dpdmai_dev->rx_queue[rxq_id]);
+       fqid = rxq->fqid;
 
-       if (conf_size != sizeof(*q_config))
-               return -EINVAL;
+       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);
 
-       rte_spinlock_lock(&qdma_dev->lock);
+               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;
+               }
 
-       /* Get a free Virtual Queue */
-       for (i = 0; i < qdma_dev->max_vqs; i++) {
-               if (qdma_vqs[i].in_use == 0)
+               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;
-       }
+               }
 
-       /* Return in case no VQ is free */
-       if (i == qdma_dev->max_vqs) {
-               rte_spinlock_unlock(&qdma_dev->lock);
-               DPAA2_QDMA_ERR("Unable to get lock on QDMA device");
-               return -ENODEV;
-       }
+               rte_prefetch0((void *)((size_t)(dq_storage + 1)));
+               /* Check if the previous issued command is completed. */
+               while (!qbman_check_command_complete(dq_storage))
+                       ;
 
-       if (qdma_dev->mode == RTE_QDMA_MODE_HW ||
-                       (q_config->flags & RTE_QDMA_VQ_EXCLUSIVE_PQ)) {
-               /* Allocate HW queue for a VQ */
-               qdma_vqs[i].hw_queue = alloc_hw_queue(q_config->lcore_id);
-               qdma_vqs[i].exclusive_hw_queue = 1;
-       } else {
-               /* Allocate a Ring for Virutal Queue in VQ mode */
-               snprintf(ring_name, sizeof(ring_name), "status ring %d", i);
-               qdma_vqs[i].status_ring = rte_ring_create(ring_name,
-                       qdma_dev->fle_pool_count, rte_socket_id(), 0);
-               if (!qdma_vqs[i].status_ring) {
-                       DPAA2_QDMA_ERR("Status ring creation failed for vq");
-                       rte_spinlock_unlock(&qdma_dev->lock);
-                       return rte_errno;
-               }
+               num_pulled = 0;
+               pending = 1;
 
-               /* Get a HW queue (shared) for a VQ */
-               qdma_vqs[i].hw_queue = get_hw_queue(qdma_dev,
-                                                   q_config->lcore_id);
-               qdma_vqs[i].exclusive_hw_queue = 0;
-       }
+               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 (qdma_vqs[i].hw_queue == NULL) {
-               DPAA2_QDMA_ERR("No H/W queue available for VQ");
-               if (qdma_vqs[i].status_ring)
-                       rte_ring_free(qdma_vqs[i].status_ring);
-               qdma_vqs[i].status_ring = NULL;
-               rte_spinlock_unlock(&qdma_dev->lock);
-               return -ENODEV;
-       }
+                       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);
 
-       qdma_vqs[i].in_use = 1;
-       qdma_vqs[i].lcore_id = q_config->lcore_id;
-       memset(&qdma_vqs[i].rbp, 0, sizeof(struct rte_qdma_rbp));
-       rte_spinlock_unlock(&qdma_dev->lock);
+                       vqid = qdma_vq->get_job(qdma_vq, fd, &job[num_rx]);
+                       if (vq_id)
+                               vq_id[num_rx] = vqid;
 
-       if (q_config->rbp != NULL)
-               memcpy(&qdma_vqs[i].rbp, q_config->rbp,
-                      sizeof(struct rte_qdma_rbp));
+                       dq_storage++;
+                       num_rx++;
+                       num_pulled++;
 
-       return i;
+               } while (pending);
+       /* Last VDQ provided all packets and more packets are requested */
+       } while (next_pull && num_pulled == dpaa2_dqrr_size);
+
+       return num_rx;
 }
 
 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,
+dpdmai_dev_enqueue_multi(
+                       struct qdma_virt_queue *qdma_vq,
                        struct rte_qdma_job **job,
                        uint16_t nb_jobs)
 {
+       struct qdma_hw_queue *qdma_pq = qdma_vq->hw_queue;
+       struct dpaa2_dpdmai_dev *dpdmai_dev = qdma_pq->dpdmai_dev;
+       uint16_t txq_id = qdma_pq->queue_id;
+
        struct qbman_fd fd[RTE_QDMA_BURST_NB_MAX];
        struct dpaa2_queue *txq;
        struct qbman_eq_desc eqdesc;
@@ -701,8 +604,7 @@ dpdmai_dev_enqueue_multi(struct dpaa2_dpdmai_dev *dpdmai_dev,
                        dpaa2_eqcr_size : nb_jobs;
 
                for (loop = 0; loop < num_to_send; loop++) {
-                       ret = dpdmai_dev_set_fd(dpdmai_dev->qdma_dev, &fd[loop],
-                                               job[num_tx], rbp, vq_id);
+                       ret = qdma_vq->set_fd(qdma_vq, &fd[loop], job[num_tx]);
                        if (ret < 0) {
                                /* Set nb_jobs to loop, so outer while loop
                                 * breaks out.
@@ -716,6 +618,7 @@ dpdmai_dev_enqueue_multi(struct dpaa2_dpdmai_dev *dpdmai_dev,
 
                /* Enqueue the packet to the QBMAN */
                uint32_t enqueue_loop = 0, retry_count = 0;
+
                while (enqueue_loop < loop) {
                        ret = qbman_swp_enqueue_multiple(swp,
                                                &eqdesc,
@@ -736,299 +639,434 @@ dpdmai_dev_enqueue_multi(struct dpaa2_dpdmai_dev *dpdmai_dev,
        return num_tx;
 }
 
-static int
-dpaa2_qdma_enqueue(__rte_unused struct rte_rawdev *rawdev,
-                 __rte_unused struct rte_rawdev_buf **buffers,
-                 unsigned int nb_jobs,
-                 rte_rawdev_obj_t context)
+static struct qdma_hw_queue *
+alloc_hw_queue(uint32_t lcore_id)
 {
-       struct rte_qdma_enqdeq *e_context = (struct rte_qdma_enqdeq *)context;
-       struct qdma_virt_queue *qdma_vq = &qdma_vqs[e_context->vq_id];
-       struct qdma_hw_queue *qdma_pq = qdma_vq->hw_queue;
-       struct dpaa2_dpdmai_dev *dpdmai_dev = qdma_pq->dpdmai_dev;
-       int ret;
+       struct qdma_hw_queue *queue = NULL;
 
-       /* 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",
-                               e_context->vq_id);
-               return -EINVAL;
-       }
+       DPAA2_QDMA_FUNC_TRACE();
 
-       ret = dpdmai_dev_enqueue_multi(dpdmai_dev,
-                                qdma_pq->queue_id,
-                                e_context->vq_id,
-                                &qdma_vq->rbp,
-                                e_context->job,
-                                nb_jobs);
-       if (ret < 0) {
-               DPAA2_QDMA_ERR("DPDMAI device enqueue failed: %d", ret);
-               return ret;
+       /* Get a free queue from the list */
+       TAILQ_FOREACH(queue, &qdma_queue_list, next) {
+               if (queue->num_users == 0) {
+                       queue->lcore_id = lcore_id;
+                       queue->num_users++;
+                       break;
+               }
        }
 
-       qdma_vq->num_enqueues += ret;
+       return queue;
+}
 
-       return ret;
+static void
+free_hw_queue(struct qdma_hw_queue *queue)
+{
+       DPAA2_QDMA_FUNC_TRACE();
+
+       queue->num_users--;
 }
 
-/* Function to receive a QDMA job for a given device and queue*/
-static int
-dpdmai_dev_dequeue_multijob_prefetch(
-                       struct dpaa2_dpdmai_dev *dpdmai_dev,
-                       uint16_t rxq_id,
-                       uint16_t *vq_id,
-                       struct rte_qdma_job **job,
-                       uint16_t nb_jobs)
+
+static struct qdma_hw_queue *
+get_hw_queue(struct qdma_device *qdma_dev, uint32_t lcore_id)
 {
-       struct dpaa2_queue *rxq;
-       struct qbman_result *dq_storage, *dq_storage1 = NULL;
-       struct qbman_pull_desc pulldesc;
-       struct qbman_swp *swp;
-       struct queue_storage_info_t *q_storage;
-       uint32_t fqid;
-       uint8_t status, pending;
-       uint8_t num_rx = 0;
-       const struct qbman_fd *fd;
-       uint16_t vqid;
-       int ret, pull_size;
+       struct qdma_per_core_info *core_info;
+       struct qdma_hw_queue *queue, *temp;
+       uint32_t least_num_users;
+       int num_hw_queues, i;
 
-       if (unlikely(!DPAA2_PER_LCORE_DPIO)) {
-               ret = dpaa2_affine_qbman_swp();
-               if (ret) {
-                       DPAA2_QDMA_ERR(
-                               "Failed to allocate IO portal, tid: %d\n",
-                               rte_gettid());
-                       return 0;
+       DPAA2_QDMA_FUNC_TRACE();
+
+       core_info = &qdma_core_info[lcore_id];
+       num_hw_queues = core_info->num_hw_queues;
+
+       /*
+        * Allocate a HW queue if there are less queues
+        * than maximum per core queues configured
+        */
+       if (num_hw_queues < qdma_dev->max_hw_queues_per_core) {
+               queue = alloc_hw_queue(lcore_id);
+               if (queue) {
+                       core_info->hw_queues[num_hw_queues] = queue;
+                       core_info->num_hw_queues++;
+                       return queue;
                }
        }
-       swp = DPAA2_PER_LCORE_PORTAL;
 
-       pull_size = (nb_jobs > dpaa2_dqrr_size) ? dpaa2_dqrr_size : nb_jobs;
-       rxq = &(dpdmai_dev->rx_queue[rxq_id]);
-       fqid = rxq->fqid;
-       q_storage = rxq->q_storage;
+       queue = core_info->hw_queues[0];
+       /* In case there is no queue associated with the core return NULL */
+       if (!queue)
+               return NULL;
 
-       if (unlikely(!q_storage->active_dqs)) {
-               q_storage->toggle = 0;
-               dq_storage = q_storage->dq_storage[q_storage->toggle];
-               q_storage->last_num_pkts = pull_size;
-               qbman_pull_desc_clear(&pulldesc);
-               qbman_pull_desc_set_numframes(&pulldesc,
-                                             q_storage->last_num_pkts);
-               qbman_pull_desc_set_fq(&pulldesc, fqid);
-               qbman_pull_desc_set_storage(&pulldesc, dq_storage,
-                               (size_t)(DPAA2_VADDR_TO_IOVA(dq_storage)), 1);
-               if (check_swp_active_dqs(DPAA2_PER_LCORE_DPIO->index)) {
-                       while (!qbman_check_command_complete(
-                              get_swp_active_dqs(
-                              DPAA2_PER_LCORE_DPIO->index)))
-                               ;
-                       clear_swp_active_dqs(DPAA2_PER_LCORE_DPIO->index);
+       /* Fetch the least loaded H/W queue */
+       least_num_users = core_info->hw_queues[0]->num_users;
+       for (i = 0; i < num_hw_queues; i++) {
+               temp = core_info->hw_queues[i];
+               if (temp->num_users < least_num_users)
+                       queue = temp;
+       }
+
+       if (queue)
+               queue->num_users++;
+
+       return queue;
+}
+
+static void
+put_hw_queue(struct qdma_hw_queue *queue)
+{
+       struct qdma_per_core_info *core_info;
+       int lcore_id, num_hw_queues, i;
+
+       DPAA2_QDMA_FUNC_TRACE();
+
+       /*
+        * If this is the last user of the queue free it.
+        * Also remove it from QDMA core info.
+        */
+       if (queue->num_users == 1) {
+               free_hw_queue(queue);
+
+               /* Remove the physical queue from core info */
+               lcore_id = queue->lcore_id;
+               core_info = &qdma_core_info[lcore_id];
+               num_hw_queues = core_info->num_hw_queues;
+               for (i = 0; i < num_hw_queues; i++) {
+                       if (queue == core_info->hw_queues[i])
+                               break;
                }
-               while (1) {
-                       if (qbman_swp_pull(swp, &pulldesc)) {
-                               DPAA2_QDMA_DP_WARN(
-                                       "VDQ command not issued.QBMAN busy\n");
-                                       /* Portal was busy, try again */
-                               continue;
-                       }
-                       break;
+               for (; i < num_hw_queues - 1; i++)
+                       core_info->hw_queues[i] = core_info->hw_queues[i + 1];
+               core_info->hw_queues[i] = NULL;
+       } else {
+               queue->num_users--;
+       }
+}
+
+static int
+dpaa2_qdma_attr_get(struct rte_rawdev *rawdev,
+                   __rte_unused const char *attr_name,
+                   uint64_t *attr_value)
+{
+       struct dpaa2_dpdmai_dev *dpdmai_dev = rawdev->dev_private;
+       struct qdma_device *qdma_dev = dpdmai_dev->qdma_dev;
+       struct rte_qdma_attr *qdma_attr = (struct rte_qdma_attr *)attr_value;
+
+       DPAA2_QDMA_FUNC_TRACE();
+
+       qdma_attr->num_hw_queues = qdma_dev->num_hw_queues;
+
+       return 0;
+}
+
+static int
+dpaa2_qdma_reset(struct rte_rawdev *rawdev)
+{
+       struct qdma_hw_queue *queue;
+       struct dpaa2_dpdmai_dev *dpdmai_dev = rawdev->dev_private;
+       struct qdma_device *qdma_dev = dpdmai_dev->qdma_dev;
+       int i;
+
+       DPAA2_QDMA_FUNC_TRACE();
+
+       /* In case QDMA device is not in stopped state, return -EBUSY */
+       if (qdma_dev->state == 1) {
+               DPAA2_QDMA_ERR(
+                       "Device is in running state. Stop before reset.");
+               return -EBUSY;
+       }
+
+       /* In case there are pending jobs on any VQ, return -EBUSY */
+       for (i = 0; i < qdma_dev->max_vqs; i++) {
+               if (qdma_dev->vqs[i].in_use && (qdma_dev->vqs[i].num_enqueues !=
+                   qdma_dev->vqs[i].num_dequeues)) {
+                       DPAA2_QDMA_ERR("Jobs are still pending on VQ: %d", i);
+                       return -EBUSY;
                }
-               q_storage->active_dqs = dq_storage;
-               q_storage->active_dpio_id = DPAA2_PER_LCORE_DPIO->index;
-               set_swp_active_dqs(DPAA2_PER_LCORE_DPIO->index,
-                                  dq_storage);
        }
 
-       dq_storage = q_storage->active_dqs;
-       rte_prefetch0((void *)(size_t)(dq_storage));
-       rte_prefetch0((void *)(size_t)(dq_storage + 1));
+       /* Reset HW queues */
+       TAILQ_FOREACH(queue, &qdma_queue_list, next)
+               queue->num_users = 0;
 
-       /* Prepare next pull descriptor. This will give space for the
-        * prefething done on DQRR entries
-        */
-       q_storage->toggle ^= 1;
-       dq_storage1 = q_storage->dq_storage[q_storage->toggle];
-       qbman_pull_desc_clear(&pulldesc);
-       qbman_pull_desc_set_numframes(&pulldesc, pull_size);
-       qbman_pull_desc_set_fq(&pulldesc, fqid);
-       qbman_pull_desc_set_storage(&pulldesc, dq_storage1,
-               (size_t)(DPAA2_VADDR_TO_IOVA(dq_storage1)), 1);
+       /* Reset and free virtual queues */
+       for (i = 0; i < qdma_dev->max_vqs; i++) {
+               if (qdma_dev->vqs[i].status_ring)
+                       rte_ring_free(qdma_dev->vqs[i].status_ring);
+       }
+       if (qdma_dev->vqs)
+               rte_free(qdma_dev->vqs);
+       qdma_dev->vqs = NULL;
 
-       /* Check if the previous issued command is completed.
-        * Also seems like the SWP is shared between the Ethernet Driver
-        * and the SEC driver.
+       /* Reset per core info */
+       memset(&qdma_core_info, 0,
+               sizeof(struct qdma_per_core_info) * RTE_MAX_LCORE);
+
+       /* Free the FLE pool */
+       if (qdma_dev->fle_pool)
+               rte_mempool_free(qdma_dev->fle_pool);
+
+       /* Reset QDMA device structure */
+       qdma_dev->max_hw_queues_per_core = 0;
+       qdma_dev->fle_pool = NULL;
+       qdma_dev->fle_pool_count = 0;
+       qdma_dev->max_vqs = 0;
+
+       return 0;
+}
+
+static int
+dpaa2_qdma_configure(const struct rte_rawdev *rawdev,
+                        rte_rawdev_obj_t config,
+                        size_t config_size)
+{
+       char name[32]; /* RTE_MEMZONE_NAMESIZE = 32 */
+       struct rte_qdma_config *qdma_config = (struct rte_qdma_config *)config;
+       struct dpaa2_dpdmai_dev *dpdmai_dev = rawdev->dev_private;
+       struct qdma_device *qdma_dev = dpdmai_dev->qdma_dev;
+
+       DPAA2_QDMA_FUNC_TRACE();
+
+       if (config_size != sizeof(*qdma_config))
+               return -EINVAL;
+
+       /* In case QDMA device is not in stopped state, return -EBUSY */
+       if (qdma_dev->state == 1) {
+               DPAA2_QDMA_ERR(
+                       "Device is in running state. Stop before config.");
+               return -1;
+       }
+
+       /* Set max HW queue per core */
+       if (qdma_config->max_hw_queues_per_core > MAX_HW_QUEUE_PER_CORE) {
+               DPAA2_QDMA_ERR("H/W queues per core is more than: %d",
+                              MAX_HW_QUEUE_PER_CORE);
+               return -EINVAL;
+       }
+       qdma_dev->max_hw_queues_per_core =
+               qdma_config->max_hw_queues_per_core;
+
+       /* Allocate Virtual Queues */
+       sprintf(name, "qdma_%d_vq", rawdev->dev_id);
+       qdma_dev->vqs = rte_malloc(name,
+                       (sizeof(struct qdma_virt_queue) * qdma_config->max_vqs),
+                       RTE_CACHE_LINE_SIZE);
+       if (!qdma_dev->vqs) {
+               DPAA2_QDMA_ERR("qdma_virtual_queues allocation failed");
+               return -ENOMEM;
+       }
+       qdma_dev->max_vqs = qdma_config->max_vqs;
+
+       /* Allocate FLE pool; just append PID so that in case of
+        * multiprocess, the pool's don't collide.
         */
-       while (!qbman_check_command_complete(dq_storage))
-               ;
-       if (dq_storage == get_swp_active_dqs(q_storage->active_dpio_id))
-               clear_swp_active_dqs(q_storage->active_dpio_id);
+       snprintf(name, sizeof(name), "qdma_fle_pool%u",
+                getpid());
+       qdma_dev->fle_pool = rte_mempool_create(name,
+                       qdma_config->fle_pool_count, QDMA_FLE_POOL_SIZE,
+                       QDMA_FLE_CACHE_SIZE(qdma_config->fle_pool_count), 0,
+                       NULL, NULL, NULL, NULL, SOCKET_ID_ANY, 0);
+       if (!qdma_dev->fle_pool) {
+               DPAA2_QDMA_ERR("qdma_fle_pool create failed");
+               rte_free(qdma_dev->vqs);
+               qdma_dev->vqs = NULL;
+               return -ENOMEM;
+       }
+       qdma_dev->fle_pool_count = qdma_config->fle_pool_count;
 
-       pending = 1;
+       return 0;
+}
 
-       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 + 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);
+static int
+dpaa2_qdma_start(struct rte_rawdev *rawdev)
+{
+       struct dpaa2_dpdmai_dev *dpdmai_dev = rawdev->dev_private;
+       struct qdma_device *qdma_dev = dpdmai_dev->qdma_dev;
+
+       DPAA2_QDMA_FUNC_TRACE();
+
+       qdma_dev->state = 1;
 
-               vqid = dpdmai_dev_get_job(dpdmai_dev->qdma_dev, fd,
-                                         &job[num_rx]);
-               if (vq_id)
-                       vq_id[num_rx] = vqid;
+       return 0;
+}
 
-               dq_storage++;
-               num_rx++;
-       } while (pending);
+static int
+check_devargs_handler(__rte_unused const char *key, const char *value,
+                     __rte_unused void *opaque)
+{
+       if (strcmp(value, "1"))
+               return -1;
 
-       if (check_swp_active_dqs(DPAA2_PER_LCORE_DPIO->index)) {
-               while (!qbman_check_command_complete(
-                      get_swp_active_dqs(DPAA2_PER_LCORE_DPIO->index)))
-                       ;
-               clear_swp_active_dqs(DPAA2_PER_LCORE_DPIO->index);
-       }
-       /* issue a volatile dequeue command for next pull */
-       while (1) {
-               if (qbman_swp_pull(swp, &pulldesc)) {
-                       DPAA2_QDMA_DP_WARN("VDQ command is not issued."
-                                         "QBMAN is busy (2)\n");
-                       continue;
-               }
-               break;
+       return 0;
+}
+
+static int
+dpaa2_get_devargs(struct rte_devargs *devargs, const char *key)
+{
+       struct rte_kvargs *kvlist;
+
+       if (!devargs)
+               return 0;
+
+       kvlist = rte_kvargs_parse(devargs->args, NULL);
+       if (!kvlist)
+               return 0;
+
+       if (!rte_kvargs_count(kvlist, key)) {
+               rte_kvargs_free(kvlist);
+               return 0;
        }
 
-       q_storage->active_dqs = dq_storage1;
-       q_storage->active_dpio_id = DPAA2_PER_LCORE_DPIO->index;
-       set_swp_active_dqs(DPAA2_PER_LCORE_DPIO->index, dq_storage1);
+       if (rte_kvargs_process(kvlist, key,
+                              check_devargs_handler, NULL) < 0) {
+               rte_kvargs_free(kvlist);
+               return 0;
+       }
+       rte_kvargs_free(kvlist);
 
-       return num_rx;
+       return 1;
 }
 
 static int
-dpdmai_dev_dequeue_multijob_no_prefetch(
-               struct dpaa2_dpdmai_dev *dpdmai_dev,
-               uint16_t rxq_id,
-               uint16_t *vq_id,
-               struct rte_qdma_job **job,
-               uint16_t nb_jobs)
+dpaa2_qdma_queue_setup(struct rte_rawdev *rawdev,
+                         __rte_unused uint16_t queue_id,
+                         rte_rawdev_obj_t queue_conf,
+                         size_t conf_size)
 {
-       struct dpaa2_queue *rxq;
-       struct qbman_result *dq_storage;
-       struct qbman_pull_desc pulldesc;
-       struct qbman_swp *swp;
-       uint32_t fqid;
-       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;
+       char ring_name[32];
+       int i;
+       struct dpaa2_dpdmai_dev *dpdmai_dev = rawdev->dev_private;
+       struct qdma_device *qdma_dev = dpdmai_dev->qdma_dev;
+       struct rte_qdma_queue_config *q_config =
+               (struct rte_qdma_queue_config *)queue_conf;
 
-       if (unlikely(!DPAA2_PER_LCORE_DPIO)) {
-               ret = dpaa2_affine_qbman_swp();
-               if (ret) {
-                       DPAA2_QDMA_ERR(
-                               "Failed to allocate IO portal, tid: %d\n",
-                               rte_gettid());
-                       return 0;
+       DPAA2_QDMA_FUNC_TRACE();
+
+       if (conf_size != sizeof(*q_config))
+               return -EINVAL;
+
+       rte_spinlock_lock(&qdma_dev->lock);
+
+       /* Get a free Virtual Queue */
+       for (i = 0; i < qdma_dev->max_vqs; i++) {
+               if (qdma_dev->vqs[i].in_use == 0)
+                       break;
+       }
+
+       /* Return in case no VQ is free */
+       if (i == qdma_dev->max_vqs) {
+               rte_spinlock_unlock(&qdma_dev->lock);
+               DPAA2_QDMA_ERR("Unable to get lock on QDMA device");
+               return -ENODEV;
+       }
+
+       if (q_config->flags & RTE_QDMA_VQ_EXCLUSIVE_PQ) {
+               /* Allocate HW queue for a VQ */
+               qdma_dev->vqs[i].hw_queue = alloc_hw_queue(q_config->lcore_id);
+               qdma_dev->vqs[i].exclusive_hw_queue = 1;
+       } else {
+               /* Allocate a Ring for Virtual Queue in VQ mode */
+               snprintf(ring_name, sizeof(ring_name), "status ring %d", i);
+               qdma_dev->vqs[i].status_ring = rte_ring_create(ring_name,
+                       qdma_dev->fle_pool_count, rte_socket_id(), 0);
+               if (!qdma_dev->vqs[i].status_ring) {
+                       DPAA2_QDMA_ERR("Status ring creation failed for vq");
+                       rte_spinlock_unlock(&qdma_dev->lock);
+                       return rte_errno;
                }
+
+               /* Get a HW queue (shared) for a VQ */
+               qdma_dev->vqs[i].hw_queue = get_hw_queue(qdma_dev,
+                                                   q_config->lcore_id);
+               qdma_dev->vqs[i].exclusive_hw_queue = 0;
        }
-       swp = DPAA2_PER_LCORE_PORTAL;
 
-       rxq = &(dpdmai_dev->rx_queue[rxq_id]);
-       fqid = rxq->fqid;
+       if (qdma_dev->vqs[i].hw_queue == NULL) {
+               DPAA2_QDMA_ERR("No H/W queue available for VQ");
+               if (qdma_dev->vqs[i].status_ring)
+                       rte_ring_free(qdma_dev->vqs[i].status_ring);
+               qdma_dev->vqs[i].status_ring = NULL;
+               rte_spinlock_unlock(&qdma_dev->lock);
+               return -ENODEV;
+       }
 
-       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);
+       qdma_dev->vqs[i].in_use = 1;
+       qdma_dev->vqs[i].lcore_id = q_config->lcore_id;
+       memset(&qdma_dev->vqs[i].rbp, 0, sizeof(struct rte_qdma_rbp));
 
-               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;
-               }
+       if (q_config->flags & RTE_QDMA_VQ_FD_LONG_FORMAT) {
+               qdma_dev->vqs[i].set_fd = dpdmai_dev_set_fd_lf;
+               qdma_dev->vqs[i].get_job = dpdmai_dev_get_job_lf;
+       } else {
+               qdma_dev->vqs[i].set_fd = dpdmai_dev_set_fd_us;
+               qdma_dev->vqs[i].get_job = dpdmai_dev_get_job_us;
+       }
+       if (dpaa2_get_devargs(rawdev->device->devargs,
+                       DPAA2_QDMA_NO_PREFETCH) ||
+                       (getenv("DPAA2_NO_QDMA_PREFETCH_RX"))) {
+               /* If no prefetch is configured. */
+               qdma_dev->vqs[i].dequeue_job =
+                               dpdmai_dev_dequeue_multijob_no_prefetch;
+               DPAA2_QDMA_INFO("No Prefetch RX Mode enabled");
+       } else {
+               qdma_dev->vqs[i].dequeue_job =
+                       dpdmai_dev_dequeue_multijob_prefetch;
+       }
 
-               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;
-               }
+       qdma_dev->vqs[i].enqueue_job = dpdmai_dev_enqueue_multi;
 
-               rte_prefetch0((void *)((size_t)(dq_storage + 1)));
-               /* Check if the previous issued command is completed. */
-               while (!qbman_check_command_complete(dq_storage))
-                       ;
+       if (q_config->rbp != NULL)
+               memcpy(&qdma_dev->vqs[i].rbp, q_config->rbp,
+                               sizeof(struct rte_qdma_rbp));
 
-               num_pulled = 0;
-               pending = 1;
+       rte_spinlock_unlock(&qdma_dev->lock);
 
-               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)));
+       return i;
+}
 
-                       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);
+static int
+dpaa2_qdma_enqueue(struct rte_rawdev *rawdev,
+                 __rte_unused struct rte_rawdev_buf **buffers,
+                 unsigned int nb_jobs,
+                 rte_rawdev_obj_t context)
+{
+       struct dpaa2_dpdmai_dev *dpdmai_dev = rawdev->dev_private;
+       struct rte_qdma_enqdeq *e_context =
+               (struct rte_qdma_enqdeq *)context;
+       struct qdma_virt_queue *qdma_vq =
+               &dpdmai_dev->qdma_dev->vqs[e_context->vq_id];
+       int ret;
 
-                       vqid = dpdmai_dev_get_job(dpdmai_dev->qdma_dev, fd,
-                                                 &job[num_rx]);
-                       if (vq_id)
-                               vq_id[num_rx] = vqid;
+       /* 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",
+                               e_context->vq_id);
+               return -EINVAL;
+       }
 
-                       dq_storage++;
-                       num_rx++;
-                       num_pulled++;
+       ret = qdma_vq->enqueue_job(qdma_vq, e_context->job, nb_jobs);
+       if (ret < 0) {
+               DPAA2_QDMA_ERR("DPDMAI device enqueue failed: %d", ret);
+               return ret;
+       }
 
-               } while (pending);
-       /* Last VDQ provided all packets and more packets are requested */
-       } while (next_pull && num_pulled == dpaa2_dqrr_size);
+       qdma_vq->num_enqueues += ret;
 
-       return num_rx;
+       return ret;
 }
 
 static int
-dpaa2_qdma_dequeue(__rte_unused struct rte_rawdev *rawdev,
+dpaa2_qdma_dequeue(struct rte_rawdev *rawdev,
                   __rte_unused struct rte_rawdev_buf **buffers,
                   unsigned int nb_jobs,
                   rte_rawdev_obj_t cntxt)
 {
-       struct rte_qdma_enqdeq *context = (struct rte_qdma_enqdeq *)cntxt;
-       struct qdma_virt_queue *qdma_vq = &qdma_vqs[context->vq_id];
-       struct qdma_hw_queue *qdma_pq = qdma_vq->hw_queue;
+       struct dpaa2_dpdmai_dev *dpdmai_dev = rawdev->dev_private;
+       struct qdma_device *qdma_dev = dpdmai_dev->qdma_dev;
+       struct rte_qdma_enqdeq *context =
+               (struct rte_qdma_enqdeq *)cntxt;
+       struct qdma_virt_queue *qdma_vq = &qdma_dev->vqs[context->vq_id];
        struct qdma_virt_queue *temp_qdma_vq;
-       struct dpaa2_dpdmai_dev *dpdmai_dev = qdma_pq->dpdmai_dev;
        int ret = 0, i;
        unsigned int ring_count;
 
@@ -1044,12 +1082,12 @@ dpaa2_qdma_dequeue(__rte_unused struct rte_rawdev *rawdev,
                return 0;
 
        if (qdma_vq->num_enqueues < (qdma_vq->num_dequeues + nb_jobs))
-               nb_jobs = (qdma_vq->num_enqueues -  qdma_vq->num_dequeues);
+               nb_jobs = (qdma_vq->num_enqueues - qdma_vq->num_dequeues);
 
        if (qdma_vq->exclusive_hw_queue) {
                /* In case of exclusive queue directly fetch from HW queue */
-               ret = dpdmai_dev_dequeue_multijob(dpdmai_dev, qdma_pq->queue_id,
-                                        NULL, context->job, nb_jobs);
+               ret = qdma_vq->dequeue_job(qdma_vq, NULL,
+                                       context->job, nb_jobs);
                if (ret < 0) {
                        DPAA2_QDMA_ERR(
                                "Dequeue from DPDMAI device failed: %d", ret);
@@ -1066,11 +1104,10 @@ dpaa2_qdma_dequeue(__rte_unused struct rte_rawdev *rawdev,
                ring_count = rte_ring_count(qdma_vq->status_ring);
                if (ring_count < nb_jobs) {
                        /* TODO - How to have right budget */
-                       ret = dpdmai_dev_dequeue_multijob(dpdmai_dev,
-                                       qdma_pq->queue_id,
+                       ret = qdma_vq->dequeue_job(qdma_vq,
                                        temp_vq_id, context->job, nb_jobs);
                        for (i = 0; i < ret; i++) {
-                               temp_qdma_vq = &qdma_vqs[temp_vq_id[i]];
+                               temp_qdma_vq = &qdma_dev->vqs[temp_vq_id[i]];
                                rte_ring_enqueue(temp_qdma_vq->status_ring,
                                        (void *)(context->job[i]));
                        }
@@ -1094,10 +1131,13 @@ dpaa2_qdma_dequeue(__rte_unused struct rte_rawdev *rawdev,
 }
 
 void
-rte_qdma_vq_stats(uint16_t vq_id,
-                 struct rte_qdma_vq_stats *vq_status)
+rte_qdma_vq_stats(struct rte_rawdev *rawdev,
+               uint16_t vq_id,
+               struct rte_qdma_vq_stats *vq_status)
 {
-       struct qdma_virt_queue *qdma_vq = &qdma_vqs[vq_id];
+       struct dpaa2_dpdmai_dev *dpdmai_dev = rawdev->dev_private;
+       struct qdma_device *qdma_dev = dpdmai_dev->qdma_dev;
+       struct qdma_virt_queue *qdma_vq = &qdma_dev->vqs[vq_id];
 
        if (qdma_vq->in_use) {
                vq_status->exclusive_hw_queue = qdma_vq->exclusive_hw_queue;
@@ -1116,7 +1156,7 @@ dpaa2_qdma_queue_release(struct rte_rawdev *rawdev,
        struct dpaa2_dpdmai_dev *dpdmai_dev = rawdev->dev_private;
        struct qdma_device *qdma_dev = dpdmai_dev->qdma_dev;
 
-       struct qdma_virt_queue *qdma_vq = &qdma_vqs[vq_id];
+       struct qdma_virt_queue *qdma_vq = &qdma_dev->vqs[vq_id];
 
        DPAA2_QDMA_FUNC_TRACE();
 
@@ -1129,8 +1169,8 @@ dpaa2_qdma_queue_release(struct rte_rawdev *rawdev,
        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);
+               if (qdma_vq->status_ring)
+                       rte_ring_free(qdma_vq->status_ring);
 
                put_hw_queue(qdma_vq->hw_queue);
        }
@@ -1253,43 +1293,6 @@ dpaa2_dpdmai_dev_uninit(struct rte_rawdev *rawdev)
        return 0;
 }
 
-static int
-check_devargs_handler(__rte_unused const char *key, const char *value,
-                     __rte_unused void *opaque)
-{
-       if (strcmp(value, "1"))
-               return -1;
-
-       return 0;
-}
-
-static int
-dpaa2_get_devargs(struct rte_devargs *devargs, const char *key)
-{
-       struct rte_kvargs *kvlist;
-
-       if (!devargs)
-               return 0;
-
-       kvlist = rte_kvargs_parse(devargs->args, NULL);
-       if (!kvlist)
-               return 0;
-
-       if (!rte_kvargs_count(kvlist, key)) {
-               rte_kvargs_free(kvlist);
-               return 0;
-       }
-
-       if (rte_kvargs_process(kvlist, key,
-                              check_devargs_handler, NULL) < 0) {
-               rte_kvargs_free(kvlist);
-               return 0;
-       }
-       rte_kvargs_free(kvlist);
-
-       return 1;
-}
-
 static int
 dpaa2_dpdmai_dev_init(struct rte_rawdev *rawdev, int dpdmai_id)
 {
@@ -1393,17 +1396,6 @@ dpaa2_dpdmai_dev_init(struct rte_rawdev *rawdev, int dpdmai_id)
                goto init_err;
        }
 
-       if (dpaa2_get_devargs(rawdev->device->devargs,
-               DPAA2_QDMA_NO_PREFETCH)) {
-               /* If no prefetch is configured. */
-               dpdmai_dev_dequeue_multijob =
-                               dpdmai_dev_dequeue_multijob_no_prefetch;
-               DPAA2_QDMA_INFO("No Prefetch RX Mode enabled");
-       } else {
-               dpdmai_dev_dequeue_multijob =
-                       dpdmai_dev_dequeue_multijob_prefetch;
-       }
-
        if (!dpaa2_coherent_no_alloc_cache) {
                if (dpaa2_svr_family == SVR_LX2160A) {
                        dpaa2_coherent_no_alloc_cache =
index 3c112d2..4265ee8 100644 (file)
@@ -52,10 +52,11 @@ struct qdma_device {
         * This is limited by MAX_HW_QUEUE_PER_CORE
         */
        uint16_t max_hw_queues_per_core;
+
+       /** VQ's of this device */
+       struct qdma_virt_queue *vqs;
        /** Maximum number of VQ's */
        uint16_t max_vqs;
-       /** mode of operation - physical(h/w) or virtual */
-       uint8_t mode;
        /** Device state - started or stopped */
        uint8_t state;
        /** FLE pool for the device */
@@ -80,6 +81,26 @@ struct qdma_hw_queue {
        uint32_t num_users;
 };
 
+struct qdma_virt_queue;
+
+typedef uint16_t (qdma_get_job_t)(struct qdma_virt_queue *qdma_vq,
+                                       const struct qbman_fd *fd,
+                                       struct rte_qdma_job **job);
+typedef int (qdma_set_fd_t)(struct qdma_virt_queue *qdma_vq,
+                                       struct qbman_fd *fd,
+                                       struct rte_qdma_job *job);
+
+typedef int (qdma_dequeue_multijob_t)(
+                               struct qdma_virt_queue *qdma_vq,
+                               uint16_t *vq_id,
+                               struct rte_qdma_job **job,
+                               uint16_t nb_jobs);
+
+typedef int (qdma_enqueue_multijob_t)(
+                       struct qdma_virt_queue *qdma_vq,
+                       struct rte_qdma_job **job,
+                       uint16_t nb_jobs);
+
 /** Represents a QDMA virtual queue */
 struct qdma_virt_queue {
        /** Status ring of the virtual queue */
@@ -98,6 +119,14 @@ struct qdma_virt_queue {
        uint64_t num_enqueues;
        /* Total number of dequeues from this VQ */
        uint64_t num_dequeues;
+
+       uint16_t vq_id;
+
+       qdma_set_fd_t *set_fd;
+       qdma_get_job_t *get_job;
+
+       qdma_dequeue_multijob_t *dequeue_job;
+       qdma_enqueue_multijob_t *enqueue_job;
 };
 
 /** Represents a QDMA per core hw queues allocation in virtual mode */
@@ -176,4 +205,10 @@ struct dpaa2_dpdmai_dev {
        struct qdma_device *qdma_dev;
 };
 
+static inline struct qdma_device *
+QDMA_DEV_OF_VQ(struct qdma_virt_queue *vq)
+{
+       return vq->hw_queue->dpdmai_dev->qdma_dev;
+}
+
 #endif /* __DPAA2_QDMA_H__ */
index 71894d3..ff4fc1d 100644 (file)
@@ -50,6 +50,8 @@ enum {
  */
 #define RTE_QDMA_VQ_EXCLUSIVE_PQ       (1ULL)
 
+#define RTE_QDMA_VQ_FD_LONG_FORMAT             (1ULL << 1)
+
 /** States if the source addresses is physical. */
 #define RTE_QDMA_JOB_SRC_PHY           (1ULL)
 
@@ -68,10 +70,6 @@ struct rte_qdma_config {
        uint16_t max_hw_queues_per_core;
        /** Maximum number of VQ's to be used. */
        uint16_t max_vqs;
-       /** mode of operation - physical(h/w) or virtual */
-       uint8_t mode;
-       /** FD format */
-       uint8_t format;
        /**
         * User provides this as input to the driver as a size of the FLE pool.
         * FLE's (and corresponding source/destination descriptors) are
@@ -182,13 +180,16 @@ struct rte_qdma_queue_config {
 /**
  * Get a Virtual Queue statistics.
  *
+ * @param rawdev
+ *   Raw Device.
  * @param vq_id
  *   Virtual Queue ID.
  * @param vq_stats
  *   VQ statistics structure which will be filled in by the driver.
  */
 void
-rte_qdma_vq_stats(uint16_t vq_id,
-                 struct rte_qdma_vq_stats *vq_stats);
+rte_qdma_vq_stats(struct rte_rawdev *rawdev,
+               uint16_t vq_id,
+               struct rte_qdma_vq_stats *vq_stats);
 
 #endif /* __RTE_PMD_DPAA2_QDMA_H__*/