1 /* SPDX-License-Identifier: BSD-3-Clause
2 * Copyright 2018-2019 NXP
9 #include <rte_atomic.h>
10 #include <rte_lcore.h>
11 #include <rte_rawdev.h>
12 #include <rte_rawdev_pmd.h>
13 #include <rte_malloc.h>
15 #include <rte_mempool.h>
16 #include <rte_prefetch.h>
17 #include <rte_kvargs.h>
19 #include <mc/fsl_dpdmai.h>
20 #include <portal/dpaa2_hw_pvt.h>
21 #include <portal/dpaa2_hw_dpio.h>
23 #include "rte_pmd_dpaa2_qdma.h"
24 #include "dpaa2_qdma.h"
25 #include "dpaa2_qdma_logs.h"
27 #define DPAA2_QDMA_NO_PREFETCH "no_prefetch"
29 /* Dynamic log type identifier */
30 int dpaa2_qdma_logtype;
32 uint32_t dpaa2_coherent_no_alloc_cache;
33 uint32_t dpaa2_coherent_alloc_cache;
36 static struct qdma_device qdma_dev;
38 /* QDMA H/W queues list */
39 TAILQ_HEAD(qdma_hw_queue_list, qdma_hw_queue);
40 static struct qdma_hw_queue_list qdma_queue_list
41 = TAILQ_HEAD_INITIALIZER(qdma_queue_list);
43 /* QDMA Virtual Queues */
44 static struct qdma_virt_queue *qdma_vqs;
46 /* QDMA per core data */
47 static struct qdma_per_core_info qdma_core_info[RTE_MAX_LCORE];
49 typedef int (dpdmai_dev_dequeue_multijob_t)(struct dpaa2_dpdmai_dev *dpdmai_dev,
52 struct rte_qdma_job **job,
55 dpdmai_dev_dequeue_multijob_t *dpdmai_dev_dequeue_multijob;
57 static struct qdma_hw_queue *
58 alloc_hw_queue(uint32_t lcore_id)
60 struct qdma_hw_queue *queue = NULL;
62 DPAA2_QDMA_FUNC_TRACE();
64 /* Get a free queue from the list */
65 TAILQ_FOREACH(queue, &qdma_queue_list, next) {
66 if (queue->num_users == 0) {
67 queue->lcore_id = lcore_id;
77 free_hw_queue(struct qdma_hw_queue *queue)
79 DPAA2_QDMA_FUNC_TRACE();
85 static struct qdma_hw_queue *
86 get_hw_queue(uint32_t lcore_id)
88 struct qdma_per_core_info *core_info;
89 struct qdma_hw_queue *queue, *temp;
90 uint32_t least_num_users;
93 DPAA2_QDMA_FUNC_TRACE();
95 core_info = &qdma_core_info[lcore_id];
96 num_hw_queues = core_info->num_hw_queues;
99 * Allocate a HW queue if there are less queues
100 * than maximum per core queues configured
102 if (num_hw_queues < qdma_dev.max_hw_queues_per_core) {
103 queue = alloc_hw_queue(lcore_id);
105 core_info->hw_queues[num_hw_queues] = queue;
106 core_info->num_hw_queues++;
111 queue = core_info->hw_queues[0];
112 /* In case there is no queue associated with the core return NULL */
116 /* Fetch the least loaded H/W queue */
117 least_num_users = core_info->hw_queues[0]->num_users;
118 for (i = 0; i < num_hw_queues; i++) {
119 temp = core_info->hw_queues[i];
120 if (temp->num_users < least_num_users)
131 put_hw_queue(struct qdma_hw_queue *queue)
133 struct qdma_per_core_info *core_info;
134 int lcore_id, num_hw_queues, i;
136 DPAA2_QDMA_FUNC_TRACE();
139 * If this is the last user of the queue free it.
140 * Also remove it from QDMA core info.
142 if (queue->num_users == 1) {
143 free_hw_queue(queue);
145 /* Remove the physical queue from core info */
146 lcore_id = queue->lcore_id;
147 core_info = &qdma_core_info[lcore_id];
148 num_hw_queues = core_info->num_hw_queues;
149 for (i = 0; i < num_hw_queues; i++) {
150 if (queue == core_info->hw_queues[i])
153 for (; i < num_hw_queues - 1; i++)
154 core_info->hw_queues[i] = core_info->hw_queues[i + 1];
155 core_info->hw_queues[i] = NULL;
164 DPAA2_QDMA_FUNC_TRACE();
166 rte_spinlock_init(&qdma_dev.lock);
172 rte_qdma_attr_get(struct rte_qdma_attr *qdma_attr)
174 DPAA2_QDMA_FUNC_TRACE();
176 qdma_attr->num_hw_queues = qdma_dev.num_hw_queues;
182 struct qdma_hw_queue *queue;
185 DPAA2_QDMA_FUNC_TRACE();
187 /* In case QDMA device is not in stopped state, return -EBUSY */
188 if (qdma_dev.state == 1) {
190 "Device is in running state. Stop before reset.");
194 /* In case there are pending jobs on any VQ, return -EBUSY */
195 for (i = 0; i < qdma_dev.max_vqs; i++) {
196 if (qdma_vqs[i].in_use && (qdma_vqs[i].num_enqueues !=
197 qdma_vqs[i].num_dequeues))
198 DPAA2_QDMA_ERR("Jobs are still pending on VQ: %d", i);
202 /* Reset HW queues */
203 TAILQ_FOREACH(queue, &qdma_queue_list, next)
204 queue->num_users = 0;
206 /* Reset and free virtual queues */
207 for (i = 0; i < qdma_dev.max_vqs; i++) {
208 if (qdma_vqs[i].status_ring)
209 rte_ring_free(qdma_vqs[i].status_ring);
215 /* Reset per core info */
216 memset(&qdma_core_info, 0,
217 sizeof(struct qdma_per_core_info) * RTE_MAX_LCORE);
219 /* Free the FLE pool */
220 if (qdma_dev.fle_pool)
221 rte_mempool_free(qdma_dev.fle_pool);
223 /* Reset QDMA device structure */
224 qdma_dev.mode = RTE_QDMA_MODE_HW;
225 qdma_dev.max_hw_queues_per_core = 0;
226 qdma_dev.fle_pool = NULL;
227 qdma_dev.fle_pool_count = 0;
228 qdma_dev.max_vqs = 0;
234 rte_qdma_configure(struct rte_qdma_config *qdma_config)
237 char fle_pool_name[32]; /* RTE_MEMZONE_NAMESIZE = 32 */
239 DPAA2_QDMA_FUNC_TRACE();
241 /* In case QDMA device is not in stopped state, return -EBUSY */
242 if (qdma_dev.state == 1) {
244 "Device is in running state. Stop before config.");
248 /* Reset the QDMA device */
249 ret = rte_qdma_reset();
251 DPAA2_QDMA_ERR("Resetting QDMA failed");
256 qdma_dev.mode = qdma_config->mode;
258 /* Set max HW queue per core */
259 if (qdma_config->max_hw_queues_per_core > MAX_HW_QUEUE_PER_CORE) {
260 DPAA2_QDMA_ERR("H/W queues per core is more than: %d",
261 MAX_HW_QUEUE_PER_CORE);
264 qdma_dev.max_hw_queues_per_core =
265 qdma_config->max_hw_queues_per_core;
267 /* Allocate Virtual Queues */
268 qdma_vqs = rte_malloc("qdma_virtual_queues",
269 (sizeof(struct qdma_virt_queue) * qdma_config->max_vqs),
270 RTE_CACHE_LINE_SIZE);
272 DPAA2_QDMA_ERR("qdma_virtual_queues allocation failed");
275 qdma_dev.max_vqs = qdma_config->max_vqs;
277 /* Allocate FLE pool; just append PID so that in case of
278 * multiprocess, the pool's don't collide.
280 snprintf(fle_pool_name, sizeof(fle_pool_name), "qdma_fle_pool%u",
282 qdma_dev.fle_pool = rte_mempool_create(fle_pool_name,
283 qdma_config->fle_pool_count, QDMA_FLE_POOL_SIZE,
284 QDMA_FLE_CACHE_SIZE(qdma_config->fle_pool_count), 0,
285 NULL, NULL, NULL, NULL, SOCKET_ID_ANY, 0);
286 if (!qdma_dev.fle_pool) {
287 DPAA2_QDMA_ERR("qdma_fle_pool create failed");
292 qdma_dev.fle_pool_count = qdma_config->fle_pool_count;
300 DPAA2_QDMA_FUNC_TRACE();
308 rte_qdma_vq_create(uint32_t lcore_id, uint32_t flags)
313 DPAA2_QDMA_FUNC_TRACE();
315 rte_spinlock_lock(&qdma_dev.lock);
317 /* Get a free Virtual Queue */
318 for (i = 0; i < qdma_dev.max_vqs; i++) {
319 if (qdma_vqs[i].in_use == 0)
323 /* Return in case no VQ is free */
324 if (i == qdma_dev.max_vqs) {
325 rte_spinlock_unlock(&qdma_dev.lock);
326 DPAA2_QDMA_ERR("Unable to get lock on QDMA device");
330 if (qdma_dev.mode == RTE_QDMA_MODE_HW ||
331 (flags & RTE_QDMA_VQ_EXCLUSIVE_PQ)) {
332 /* Allocate HW queue for a VQ */
333 qdma_vqs[i].hw_queue = alloc_hw_queue(lcore_id);
334 qdma_vqs[i].exclusive_hw_queue = 1;
336 /* Allocate a Ring for Virutal Queue in VQ mode */
337 snprintf(ring_name, sizeof(ring_name), "status ring %d", i);
338 qdma_vqs[i].status_ring = rte_ring_create(ring_name,
339 qdma_dev.fle_pool_count, rte_socket_id(), 0);
340 if (!qdma_vqs[i].status_ring) {
341 DPAA2_QDMA_ERR("Status ring creation failed for vq");
342 rte_spinlock_unlock(&qdma_dev.lock);
346 /* Get a HW queue (shared) for a VQ */
347 qdma_vqs[i].hw_queue = get_hw_queue(lcore_id);
348 qdma_vqs[i].exclusive_hw_queue = 0;
351 if (qdma_vqs[i].hw_queue == NULL) {
352 DPAA2_QDMA_ERR("No H/W queue available for VQ");
353 if (qdma_vqs[i].status_ring)
354 rte_ring_free(qdma_vqs[i].status_ring);
355 qdma_vqs[i].status_ring = NULL;
356 rte_spinlock_unlock(&qdma_dev.lock);
360 qdma_vqs[i].in_use = 1;
361 qdma_vqs[i].lcore_id = lcore_id;
362 memset(&qdma_vqs[i].rbp, 0, sizeof(struct rte_qdma_rbp));
363 rte_spinlock_unlock(&qdma_dev.lock);
368 /*create vq for route-by-port*/
370 rte_qdma_vq_create_rbp(uint32_t lcore_id, uint32_t flags,
371 struct rte_qdma_rbp *rbp)
375 i = rte_qdma_vq_create(lcore_id, flags);
377 memcpy(&qdma_vqs[i].rbp, rbp, sizeof(struct rte_qdma_rbp));
383 dpaa2_qdma_populate_fle(struct qbman_fle *fle,
384 struct rte_qdma_rbp *rbp,
385 uint64_t src, uint64_t dest,
386 size_t len, uint32_t flags)
388 struct qdma_sdd *sdd;
390 sdd = (struct qdma_sdd *)((uint8_t *)(fle) +
391 (DPAA2_QDMA_MAX_FLE * sizeof(struct qbman_fle)));
393 /* first frame list to source descriptor */
394 DPAA2_SET_FLE_ADDR(fle, DPAA2_VADDR_TO_IOVA(sdd));
395 DPAA2_SET_FLE_LEN(fle, (2 * (sizeof(struct qdma_sdd))));
397 /* source and destination descriptor */
398 if (rbp && rbp->enable) {
400 sdd->read_cmd.portid = rbp->sportid;
401 sdd->rbpcmd_simple.pfid = rbp->spfid;
402 sdd->rbpcmd_simple.vfid = rbp->svfid;
405 sdd->read_cmd.rbp = rbp->srbp;
406 sdd->read_cmd.rdtype = DPAA2_RBP_MEM_RW;
408 sdd->read_cmd.rdtype = dpaa2_coherent_no_alloc_cache;
412 sdd->write_cmd.portid = rbp->dportid;
413 sdd->rbpcmd_simple.pfid = rbp->dpfid;
414 sdd->rbpcmd_simple.vfid = rbp->dvfid;
417 sdd->write_cmd.rbp = rbp->drbp;
418 sdd->write_cmd.wrttype = DPAA2_RBP_MEM_RW;
420 sdd->write_cmd.wrttype = dpaa2_coherent_alloc_cache;
424 sdd->read_cmd.rdtype = dpaa2_coherent_no_alloc_cache;
426 sdd->write_cmd.wrttype = dpaa2_coherent_alloc_cache;
429 /* source frame list to source buffer */
430 if (flags & RTE_QDMA_JOB_SRC_PHY) {
431 DPAA2_SET_FLE_ADDR(fle, src);
432 DPAA2_SET_FLE_BMT(fle);
434 DPAA2_SET_FLE_ADDR(fle, DPAA2_VADDR_TO_IOVA(src));
436 DPAA2_SET_FLE_LEN(fle, len);
439 /* destination frame list to destination buffer */
440 if (flags & RTE_QDMA_JOB_DEST_PHY) {
441 DPAA2_SET_FLE_BMT(fle);
442 DPAA2_SET_FLE_ADDR(fle, dest);
444 DPAA2_SET_FLE_ADDR(fle, DPAA2_VADDR_TO_IOVA(dest));
446 DPAA2_SET_FLE_LEN(fle, len);
448 /* Final bit: 1, for last frame list */
449 DPAA2_SET_FLE_FIN(fle);
452 static inline uint16_t dpdmai_dev_set_fd(struct qbman_fd *fd,
453 struct rte_qdma_job *job,
454 struct rte_qdma_rbp *rbp,
457 struct qdma_io_meta *io_meta;
458 struct qbman_fle *fle;
461 * Get an FLE/SDD from FLE pool.
462 * Note: IO metadata is before the FLE and SDD memory.
464 ret = rte_mempool_get(qdma_dev.fle_pool, (void **)(&io_meta));
466 DPAA2_QDMA_DP_DEBUG("Memory alloc failed for FLE");
470 /* Set the metadata */
471 io_meta->cnxt = (size_t)job;
474 fle = (struct qbman_fle *)(io_meta + 1);
476 DPAA2_SET_FD_ADDR(fd, DPAA2_VADDR_TO_IOVA(fle));
477 DPAA2_SET_FD_COMPOUND_FMT(fd);
478 DPAA2_SET_FD_FRC(fd, QDMA_SER_CTX);
481 memset(fle, 0, QDMA_FLE_POOL_SIZE);
482 dpaa2_qdma_populate_fle(fle, rbp, job->src, job->dest,
483 job->len, job->flags);
489 dpdmai_dev_enqueue_multi(struct dpaa2_dpdmai_dev *dpdmai_dev,
492 struct rte_qdma_rbp *rbp,
493 struct rte_qdma_job **job,
496 struct qbman_fd fd[RTE_QDMA_BURST_NB_MAX];
497 struct dpaa2_queue *txq;
498 struct qbman_eq_desc eqdesc;
499 struct qbman_swp *swp;
501 uint32_t num_to_send = 0;
504 if (unlikely(!DPAA2_PER_LCORE_DPIO)) {
505 ret = dpaa2_affine_qbman_swp();
507 DPAA2_QDMA_ERR("Failure in affining portal");
511 swp = DPAA2_PER_LCORE_PORTAL;
513 txq = &(dpdmai_dev->tx_queue[txq_id]);
515 /* Prepare enqueue descriptor */
516 qbman_eq_desc_clear(&eqdesc);
517 qbman_eq_desc_set_fq(&eqdesc, txq->fqid);
518 qbman_eq_desc_set_no_orp(&eqdesc, 0);
519 qbman_eq_desc_set_response(&eqdesc, 0, 0);
521 memset(fd, 0, RTE_QDMA_BURST_NB_MAX * sizeof(struct qbman_fd));
523 while (nb_jobs > 0) {
526 num_to_send = (nb_jobs > dpaa2_eqcr_size) ?
527 dpaa2_eqcr_size : nb_jobs;
529 for (loop = 0; loop < num_to_send; loop++) {
530 ret = dpdmai_dev_set_fd(&fd[loop],
531 job[num_tx], rbp, vq_id);
533 /* Set nb_jobs to loop, so outer while loop
543 /* Enqueue the packet to the QBMAN */
544 uint32_t enqueue_loop = 0;
545 while (enqueue_loop < loop) {
546 enqueue_loop += qbman_swp_enqueue_multiple(swp,
550 loop - enqueue_loop);
558 rte_qdma_vq_enqueue_multi(uint16_t vq_id,
559 struct rte_qdma_job **job,
562 struct qdma_virt_queue *qdma_vq = &qdma_vqs[vq_id];
563 struct qdma_hw_queue *qdma_pq = qdma_vq->hw_queue;
564 struct dpaa2_dpdmai_dev *dpdmai_dev = qdma_pq->dpdmai_dev;
567 /* Return error in case of wrong lcore_id */
568 if (rte_lcore_id() != qdma_vq->lcore_id) {
569 DPAA2_QDMA_ERR("QDMA enqueue for vqid %d on wrong core",
574 ret = dpdmai_dev_enqueue_multi(dpdmai_dev,
581 DPAA2_QDMA_ERR("DPDMAI device enqueue failed: %d", ret);
585 qdma_vq->num_enqueues += ret;
591 rte_qdma_vq_enqueue(uint16_t vq_id,
592 struct rte_qdma_job *job)
594 return rte_qdma_vq_enqueue_multi(vq_id, &job, 1);
597 static inline uint16_t dpdmai_dev_get_job(const struct qbman_fd *fd,
598 struct rte_qdma_job **job)
600 struct qbman_fle *fle;
601 struct qdma_io_meta *io_meta;
604 * Fetch metadata from FLE. job and vq_id were set
605 * in metadata in the enqueue operation.
607 fle = (struct qbman_fle *)DPAA2_IOVA_TO_VADDR(DPAA2_GET_FD_ADDR(fd));
608 io_meta = (struct qdma_io_meta *)(fle) - 1;
610 *job = (struct rte_qdma_job *)(size_t)io_meta->cnxt;
611 (*job)->status = (DPAA2_GET_FD_ERR(fd) << 8) |
612 (DPAA2_GET_FD_FRC(fd) & 0xFF);
616 /* Free FLE to the pool */
617 rte_mempool_put(qdma_dev.fle_pool, io_meta);
622 /* Function to receive a QDMA job for a given device and queue*/
624 dpdmai_dev_dequeue_multijob_prefetch(
625 struct dpaa2_dpdmai_dev *dpdmai_dev,
628 struct rte_qdma_job **job,
631 struct dpaa2_queue *rxq;
632 struct qbman_result *dq_storage, *dq_storage1 = NULL;
633 struct qbman_pull_desc pulldesc;
634 struct qbman_swp *swp;
635 struct queue_storage_info_t *q_storage;
637 uint8_t status, pending;
639 const struct qbman_fd *fd;
643 if (unlikely(!DPAA2_PER_LCORE_DPIO)) {
644 ret = dpaa2_affine_qbman_swp();
646 DPAA2_QDMA_ERR("Failure in affining portal");
650 swp = DPAA2_PER_LCORE_PORTAL;
652 pull_size = (nb_jobs > dpaa2_dqrr_size) ? dpaa2_dqrr_size : nb_jobs;
653 rxq = &(dpdmai_dev->rx_queue[rxq_id]);
655 q_storage = rxq->q_storage;
657 if (unlikely(!q_storage->active_dqs)) {
658 q_storage->toggle = 0;
659 dq_storage = q_storage->dq_storage[q_storage->toggle];
660 q_storage->last_num_pkts = pull_size;
661 qbman_pull_desc_clear(&pulldesc);
662 qbman_pull_desc_set_numframes(&pulldesc,
663 q_storage->last_num_pkts);
664 qbman_pull_desc_set_fq(&pulldesc, fqid);
665 qbman_pull_desc_set_storage(&pulldesc, dq_storage,
666 (size_t)(DPAA2_VADDR_TO_IOVA(dq_storage)), 1);
667 if (check_swp_active_dqs(DPAA2_PER_LCORE_DPIO->index)) {
668 while (!qbman_check_command_complete(
670 DPAA2_PER_LCORE_DPIO->index)))
672 clear_swp_active_dqs(DPAA2_PER_LCORE_DPIO->index);
675 if (qbman_swp_pull(swp, &pulldesc)) {
677 "VDQ command not issued.QBMAN busy\n");
678 /* Portal was busy, try again */
683 q_storage->active_dqs = dq_storage;
684 q_storage->active_dpio_id = DPAA2_PER_LCORE_DPIO->index;
685 set_swp_active_dqs(DPAA2_PER_LCORE_DPIO->index,
689 dq_storage = q_storage->active_dqs;
690 rte_prefetch0((void *)(size_t)(dq_storage));
691 rte_prefetch0((void *)(size_t)(dq_storage + 1));
693 /* Prepare next pull descriptor. This will give space for the
694 * prefething done on DQRR entries
696 q_storage->toggle ^= 1;
697 dq_storage1 = q_storage->dq_storage[q_storage->toggle];
698 qbman_pull_desc_clear(&pulldesc);
699 qbman_pull_desc_set_numframes(&pulldesc, pull_size);
700 qbman_pull_desc_set_fq(&pulldesc, fqid);
701 qbman_pull_desc_set_storage(&pulldesc, dq_storage1,
702 (size_t)(DPAA2_VADDR_TO_IOVA(dq_storage1)), 1);
704 /* Check if the previous issued command is completed.
705 * Also seems like the SWP is shared between the Ethernet Driver
706 * and the SEC driver.
708 while (!qbman_check_command_complete(dq_storage))
710 if (dq_storage == get_swp_active_dqs(q_storage->active_dpio_id))
711 clear_swp_active_dqs(q_storage->active_dpio_id);
716 /* Loop until the dq_storage is updated with
719 while (!qbman_check_new_result(dq_storage))
721 rte_prefetch0((void *)((size_t)(dq_storage + 2)));
722 /* Check whether Last Pull command is Expired and
723 * setting Condition for Loop termination
725 if (qbman_result_DQ_is_pull_complete(dq_storage)) {
727 /* Check for valid frame. */
728 status = qbman_result_DQ_flags(dq_storage);
729 if (unlikely((status & QBMAN_DQ_STAT_VALIDFRAME) == 0))
732 fd = qbman_result_DQ_fd(dq_storage);
734 vqid = dpdmai_dev_get_job(fd, &job[num_rx]);
736 vq_id[num_rx] = vqid;
742 if (check_swp_active_dqs(DPAA2_PER_LCORE_DPIO->index)) {
743 while (!qbman_check_command_complete(
744 get_swp_active_dqs(DPAA2_PER_LCORE_DPIO->index)))
746 clear_swp_active_dqs(DPAA2_PER_LCORE_DPIO->index);
748 /* issue a volatile dequeue command for next pull */
750 if (qbman_swp_pull(swp, &pulldesc)) {
751 DPAA2_QDMA_DP_WARN("VDQ command is not issued."
752 "QBMAN is busy (2)\n");
758 q_storage->active_dqs = dq_storage1;
759 q_storage->active_dpio_id = DPAA2_PER_LCORE_DPIO->index;
760 set_swp_active_dqs(DPAA2_PER_LCORE_DPIO->index, dq_storage1);
766 dpdmai_dev_dequeue_multijob_no_prefetch(
767 struct dpaa2_dpdmai_dev *dpdmai_dev,
770 struct rte_qdma_job **job,
773 struct dpaa2_queue *rxq;
774 struct qbman_result *dq_storage;
775 struct qbman_pull_desc pulldesc;
776 struct qbman_swp *swp;
778 uint8_t status, pending;
780 const struct qbman_fd *fd;
782 int ret, next_pull = nb_jobs, num_pulled = 0;
784 if (unlikely(!DPAA2_PER_LCORE_DPIO)) {
785 ret = dpaa2_affine_qbman_swp();
787 DPAA2_QDMA_ERR("Failure in affining portal");
791 swp = DPAA2_PER_LCORE_PORTAL;
793 rxq = &(dpdmai_dev->rx_queue[rxq_id]);
797 dq_storage = rxq->q_storage->dq_storage[0];
798 /* Prepare dequeue descriptor */
799 qbman_pull_desc_clear(&pulldesc);
800 qbman_pull_desc_set_fq(&pulldesc, fqid);
801 qbman_pull_desc_set_storage(&pulldesc, dq_storage,
802 (uint64_t)(DPAA2_VADDR_TO_IOVA(dq_storage)), 1);
804 if (next_pull > dpaa2_dqrr_size) {
805 qbman_pull_desc_set_numframes(&pulldesc,
807 next_pull -= dpaa2_dqrr_size;
809 qbman_pull_desc_set_numframes(&pulldesc, next_pull);
814 if (qbman_swp_pull(swp, &pulldesc)) {
815 DPAA2_QDMA_DP_WARN("VDQ command not issued. QBMAN busy");
816 /* Portal was busy, try again */
822 rte_prefetch0((void *)((size_t)(dq_storage + 1)));
823 /* Check if the previous issued command is completed. */
824 while (!qbman_check_command_complete(dq_storage))
831 /* Loop until dq_storage is updated
832 * with new token by QBMAN
834 while (!qbman_check_new_result(dq_storage))
836 rte_prefetch0((void *)((size_t)(dq_storage + 2)));
838 if (qbman_result_DQ_is_pull_complete(dq_storage)) {
840 /* Check for valid frame. */
841 status = qbman_result_DQ_flags(dq_storage);
842 if (unlikely((status &
843 QBMAN_DQ_STAT_VALIDFRAME) == 0))
846 fd = qbman_result_DQ_fd(dq_storage);
848 vqid = dpdmai_dev_get_job(fd, &job[num_rx]);
850 vq_id[num_rx] = vqid;
857 /* Last VDQ provided all packets and more packets are requested */
858 } while (next_pull && num_pulled == dpaa2_dqrr_size);
864 rte_qdma_vq_dequeue_multi(uint16_t vq_id,
865 struct rte_qdma_job **job,
868 struct qdma_virt_queue *qdma_vq = &qdma_vqs[vq_id];
869 struct qdma_hw_queue *qdma_pq = qdma_vq->hw_queue;
870 struct qdma_virt_queue *temp_qdma_vq;
871 struct dpaa2_dpdmai_dev *dpdmai_dev = qdma_pq->dpdmai_dev;
872 int ring_count, ret = 0, i;
874 /* Return error in case of wrong lcore_id */
875 if (rte_lcore_id() != (unsigned int)(qdma_vq->lcore_id)) {
876 DPAA2_QDMA_WARN("QDMA dequeue for vqid %d on wrong core",
881 /* Only dequeue when there are pending jobs on VQ */
882 if (qdma_vq->num_enqueues == qdma_vq->num_dequeues)
885 if (qdma_vq->num_enqueues < (qdma_vq->num_dequeues + nb_jobs))
886 nb_jobs = (qdma_vq->num_enqueues - qdma_vq->num_dequeues);
888 if (qdma_vq->exclusive_hw_queue) {
889 /* In case of exclusive queue directly fetch from HW queue */
890 ret = dpdmai_dev_dequeue_multijob(dpdmai_dev, qdma_pq->queue_id,
894 "Dequeue from DPDMAI device failed: %d", ret);
897 qdma_vq->num_dequeues += ret;
899 uint16_t temp_vq_id[RTE_QDMA_BURST_NB_MAX];
901 * Get the QDMA completed jobs from the software ring.
902 * In case they are not available on the ring poke the HW
903 * to fetch completed jobs from corresponding HW queues
905 ring_count = rte_ring_count(qdma_vq->status_ring);
906 if (ring_count < nb_jobs) {
907 /* TODO - How to have right budget */
908 ret = dpdmai_dev_dequeue_multijob(dpdmai_dev,
910 temp_vq_id, job, nb_jobs);
911 for (i = 0; i < ret; i++) {
912 temp_qdma_vq = &qdma_vqs[temp_vq_id[i]];
913 rte_ring_enqueue(temp_qdma_vq->status_ring,
916 ring_count = rte_ring_count(
917 qdma_vq->status_ring);
921 /* Dequeue job from the software ring
922 * to provide to the user
924 ret = rte_ring_dequeue_bulk(qdma_vq->status_ring,
925 (void **)job, ring_count, NULL);
927 qdma_vq->num_dequeues += ret;
934 struct rte_qdma_job *
935 rte_qdma_vq_dequeue(uint16_t vq_id)
938 struct rte_qdma_job *job = NULL;
940 ret = rte_qdma_vq_dequeue_multi(vq_id, &job, 1);
942 DPAA2_QDMA_DP_WARN("DPDMAI device dequeue failed: %d", ret);
948 rte_qdma_vq_stats(uint16_t vq_id,
949 struct rte_qdma_vq_stats *vq_status)
951 struct qdma_virt_queue *qdma_vq = &qdma_vqs[vq_id];
953 if (qdma_vq->in_use) {
954 vq_status->exclusive_hw_queue = qdma_vq->exclusive_hw_queue;
955 vq_status->lcore_id = qdma_vq->lcore_id;
956 vq_status->num_enqueues = qdma_vq->num_enqueues;
957 vq_status->num_dequeues = qdma_vq->num_dequeues;
958 vq_status->num_pending_jobs = vq_status->num_enqueues -
959 vq_status->num_dequeues;
964 rte_qdma_vq_destroy(uint16_t vq_id)
966 struct qdma_virt_queue *qdma_vq = &qdma_vqs[vq_id];
968 DPAA2_QDMA_FUNC_TRACE();
970 /* In case there are pending jobs on any VQ, return -EBUSY */
971 if (qdma_vq->num_enqueues != qdma_vq->num_dequeues)
974 rte_spinlock_lock(&qdma_dev.lock);
976 if (qdma_vq->exclusive_hw_queue)
977 free_hw_queue(qdma_vq->hw_queue);
979 if (qdma_vqs->status_ring)
980 rte_ring_free(qdma_vqs->status_ring);
982 put_hw_queue(qdma_vq->hw_queue);
985 memset(qdma_vq, 0, sizeof(struct qdma_virt_queue));
987 rte_spinlock_unlock(&qdma_dev.lock);
993 rte_qdma_vq_destroy_rbp(uint16_t vq_id)
995 struct qdma_virt_queue *qdma_vq = &qdma_vqs[vq_id];
997 DPAA2_QDMA_FUNC_TRACE();
999 /* In case there are pending jobs on any VQ, return -EBUSY */
1000 if (qdma_vq->num_enqueues != qdma_vq->num_dequeues)
1003 rte_spinlock_lock(&qdma_dev.lock);
1005 if (qdma_vq->exclusive_hw_queue) {
1006 free_hw_queue(qdma_vq->hw_queue);
1008 if (qdma_vqs->status_ring)
1009 rte_ring_free(qdma_vqs->status_ring);
1011 put_hw_queue(qdma_vq->hw_queue);
1014 memset(qdma_vq, 0, sizeof(struct qdma_virt_queue));
1016 rte_spinlock_unlock(&qdma_dev.lock);
1024 DPAA2_QDMA_FUNC_TRACE();
1030 rte_qdma_destroy(void)
1032 DPAA2_QDMA_FUNC_TRACE();
1037 static const struct rte_rawdev_ops dpaa2_qdma_ops;
1040 add_hw_queues_to_list(struct dpaa2_dpdmai_dev *dpdmai_dev)
1042 struct qdma_hw_queue *queue;
1045 DPAA2_QDMA_FUNC_TRACE();
1047 for (i = 0; i < dpdmai_dev->num_queues; i++) {
1048 queue = rte_zmalloc(NULL, sizeof(struct qdma_hw_queue), 0);
1051 "Memory allocation failed for QDMA queue");
1055 queue->dpdmai_dev = dpdmai_dev;
1056 queue->queue_id = i;
1058 TAILQ_INSERT_TAIL(&qdma_queue_list, queue, next);
1059 qdma_dev.num_hw_queues++;
1066 remove_hw_queues_from_list(struct dpaa2_dpdmai_dev *dpdmai_dev)
1068 struct qdma_hw_queue *queue = NULL;
1069 struct qdma_hw_queue *tqueue = NULL;
1071 DPAA2_QDMA_FUNC_TRACE();
1073 TAILQ_FOREACH_SAFE(queue, &qdma_queue_list, next, tqueue) {
1074 if (queue->dpdmai_dev == dpdmai_dev) {
1075 TAILQ_REMOVE(&qdma_queue_list, queue, next);
1083 dpaa2_dpdmai_dev_uninit(struct rte_rawdev *rawdev)
1085 struct dpaa2_dpdmai_dev *dpdmai_dev = rawdev->dev_private;
1088 DPAA2_QDMA_FUNC_TRACE();
1090 /* Remove HW queues from global list */
1091 remove_hw_queues_from_list(dpdmai_dev);
1093 ret = dpdmai_disable(&dpdmai_dev->dpdmai, CMD_PRI_LOW,
1096 DPAA2_QDMA_ERR("dmdmai disable failed");
1098 /* Set up the DQRR storage for Rx */
1099 for (i = 0; i < dpdmai_dev->num_queues; i++) {
1100 struct dpaa2_queue *rxq = &(dpdmai_dev->rx_queue[i]);
1102 if (rxq->q_storage) {
1103 dpaa2_free_dq_storage(rxq->q_storage);
1104 rte_free(rxq->q_storage);
1108 /* Close the device at underlying layer*/
1109 ret = dpdmai_close(&dpdmai_dev->dpdmai, CMD_PRI_LOW, dpdmai_dev->token);
1111 DPAA2_QDMA_ERR("Failure closing dpdmai device");
1117 check_devargs_handler(__rte_unused const char *key, const char *value,
1118 __rte_unused void *opaque)
1120 if (strcmp(value, "1"))
1127 dpaa2_get_devargs(struct rte_devargs *devargs, const char *key)
1129 struct rte_kvargs *kvlist;
1134 kvlist = rte_kvargs_parse(devargs->args, NULL);
1138 if (!rte_kvargs_count(kvlist, key)) {
1139 rte_kvargs_free(kvlist);
1143 if (rte_kvargs_process(kvlist, key,
1144 check_devargs_handler, NULL) < 0) {
1145 rte_kvargs_free(kvlist);
1148 rte_kvargs_free(kvlist);
1154 dpaa2_dpdmai_dev_init(struct rte_rawdev *rawdev, int dpdmai_id)
1156 struct dpaa2_dpdmai_dev *dpdmai_dev = rawdev->dev_private;
1157 struct dpdmai_rx_queue_cfg rx_queue_cfg;
1158 struct dpdmai_attr attr;
1159 struct dpdmai_rx_queue_attr rx_attr;
1160 struct dpdmai_tx_queue_attr tx_attr;
1163 DPAA2_QDMA_FUNC_TRACE();
1165 /* Open DPDMAI device */
1166 dpdmai_dev->dpdmai_id = dpdmai_id;
1167 dpdmai_dev->dpdmai.regs = rte_mcp_ptr_list[MC_PORTAL_INDEX];
1168 ret = dpdmai_open(&dpdmai_dev->dpdmai, CMD_PRI_LOW,
1169 dpdmai_dev->dpdmai_id, &dpdmai_dev->token);
1171 DPAA2_QDMA_ERR("dpdmai_open() failed with err: %d", ret);
1175 /* Get DPDMAI attributes */
1176 ret = dpdmai_get_attributes(&dpdmai_dev->dpdmai, CMD_PRI_LOW,
1177 dpdmai_dev->token, &attr);
1179 DPAA2_QDMA_ERR("dpdmai get attributes failed with err: %d",
1183 dpdmai_dev->num_queues = attr.num_of_queues;
1185 /* Set up Rx Queues */
1186 for (i = 0; i < dpdmai_dev->num_queues; i++) {
1187 struct dpaa2_queue *rxq;
1189 memset(&rx_queue_cfg, 0, sizeof(struct dpdmai_rx_queue_cfg));
1190 ret = dpdmai_set_rx_queue(&dpdmai_dev->dpdmai,
1193 i, 0, &rx_queue_cfg);
1195 DPAA2_QDMA_ERR("Setting Rx queue failed with err: %d",
1200 /* Allocate DQ storage for the DPDMAI Rx queues */
1201 rxq = &(dpdmai_dev->rx_queue[i]);
1202 rxq->q_storage = rte_malloc("dq_storage",
1203 sizeof(struct queue_storage_info_t),
1204 RTE_CACHE_LINE_SIZE);
1205 if (!rxq->q_storage) {
1206 DPAA2_QDMA_ERR("q_storage allocation failed");
1211 memset(rxq->q_storage, 0, sizeof(struct queue_storage_info_t));
1212 ret = dpaa2_alloc_dq_storage(rxq->q_storage);
1214 DPAA2_QDMA_ERR("dpaa2_alloc_dq_storage failed");
1219 /* Get Rx and Tx queues FQID's */
1220 for (i = 0; i < dpdmai_dev->num_queues; i++) {
1221 ret = dpdmai_get_rx_queue(&dpdmai_dev->dpdmai, CMD_PRI_LOW,
1222 dpdmai_dev->token, i, 0, &rx_attr);
1224 DPAA2_QDMA_ERR("Reading device failed with err: %d",
1228 dpdmai_dev->rx_queue[i].fqid = rx_attr.fqid;
1230 ret = dpdmai_get_tx_queue(&dpdmai_dev->dpdmai, CMD_PRI_LOW,
1231 dpdmai_dev->token, i, 0, &tx_attr);
1233 DPAA2_QDMA_ERR("Reading device failed with err: %d",
1237 dpdmai_dev->tx_queue[i].fqid = tx_attr.fqid;
1240 /* Enable the device */
1241 ret = dpdmai_enable(&dpdmai_dev->dpdmai, CMD_PRI_LOW,
1244 DPAA2_QDMA_ERR("Enabling device failed with err: %d", ret);
1248 /* Add the HW queue to the global list */
1249 ret = add_hw_queues_to_list(dpdmai_dev);
1251 DPAA2_QDMA_ERR("Adding H/W queue to list failed");
1255 if (dpaa2_get_devargs(rawdev->device->devargs,
1256 DPAA2_QDMA_NO_PREFETCH)) {
1257 /* If no prefetch is configured. */
1258 dpdmai_dev_dequeue_multijob =
1259 dpdmai_dev_dequeue_multijob_no_prefetch;
1260 DPAA2_QDMA_INFO("No Prefetch RX Mode enabled");
1262 dpdmai_dev_dequeue_multijob =
1263 dpdmai_dev_dequeue_multijob_prefetch;
1266 if (!dpaa2_coherent_no_alloc_cache) {
1267 if (dpaa2_svr_family == SVR_LX2160A) {
1268 dpaa2_coherent_no_alloc_cache =
1269 DPAA2_LX2_COHERENT_NO_ALLOCATE_CACHE;
1270 dpaa2_coherent_alloc_cache =
1271 DPAA2_LX2_COHERENT_ALLOCATE_CACHE;
1273 dpaa2_coherent_no_alloc_cache =
1274 DPAA2_COHERENT_NO_ALLOCATE_CACHE;
1275 dpaa2_coherent_alloc_cache =
1276 DPAA2_COHERENT_ALLOCATE_CACHE;
1280 DPAA2_QDMA_DEBUG("Initialized dpdmai object successfully");
1284 dpaa2_dpdmai_dev_uninit(rawdev);
1289 rte_dpaa2_qdma_probe(struct rte_dpaa2_driver *dpaa2_drv,
1290 struct rte_dpaa2_device *dpaa2_dev)
1292 struct rte_rawdev *rawdev;
1295 DPAA2_QDMA_FUNC_TRACE();
1297 rawdev = rte_rawdev_pmd_allocate(dpaa2_dev->device.name,
1298 sizeof(struct dpaa2_dpdmai_dev),
1301 DPAA2_QDMA_ERR("Unable to allocate rawdevice");
1305 dpaa2_dev->rawdev = rawdev;
1306 rawdev->dev_ops = &dpaa2_qdma_ops;
1307 rawdev->device = &dpaa2_dev->device;
1308 rawdev->driver_name = dpaa2_drv->driver.name;
1310 /* Invoke PMD device initialization function */
1311 ret = dpaa2_dpdmai_dev_init(rawdev, dpaa2_dev->object_id);
1313 rte_rawdev_pmd_release(rawdev);
1321 rte_dpaa2_qdma_remove(struct rte_dpaa2_device *dpaa2_dev)
1323 struct rte_rawdev *rawdev = dpaa2_dev->rawdev;
1326 DPAA2_QDMA_FUNC_TRACE();
1328 dpaa2_dpdmai_dev_uninit(rawdev);
1330 ret = rte_rawdev_pmd_release(rawdev);
1332 DPAA2_QDMA_ERR("Device cleanup failed");
1337 static struct rte_dpaa2_driver rte_dpaa2_qdma_pmd = {
1338 .drv_flags = RTE_DPAA2_DRV_IOVA_AS_VA,
1339 .drv_type = DPAA2_QDMA,
1340 .probe = rte_dpaa2_qdma_probe,
1341 .remove = rte_dpaa2_qdma_remove,
1344 RTE_PMD_REGISTER_DPAA2(dpaa2_qdma, rte_dpaa2_qdma_pmd);
1345 RTE_PMD_REGISTER_PARAM_STRING(dpaa2_qdma,
1346 "no_prefetch=<int> ");
1348 RTE_INIT(dpaa2_qdma_init_log)
1350 dpaa2_qdma_logtype = rte_log_register("pmd.raw.dpaa2.qdma");
1351 if (dpaa2_qdma_logtype >= 0)
1352 rte_log_set_level(dpaa2_qdma_logtype, RTE_LOG_INFO);