net/qede: add new host ring type option
[dpdk.git] / drivers / net / qede / qede_rxtx.c
1 /*
2  * Copyright (c) 2016 QLogic Corporation.
3  * All rights reserved.
4  * www.qlogic.com
5  *
6  * See LICENSE.qede_pmd for copyright and licensing details.
7  */
8
9 #include "qede_rxtx.h"
10
11 static bool gro_disable = 1;    /* mod_param */
12
13 static inline int qede_alloc_rx_buffer(struct qede_rx_queue *rxq)
14 {
15         struct rte_mbuf *new_mb = NULL;
16         struct eth_rx_bd *rx_bd;
17         dma_addr_t mapping;
18         uint16_t idx = rxq->sw_rx_prod & NUM_RX_BDS(rxq);
19
20         new_mb = rte_mbuf_raw_alloc(rxq->mb_pool);
21         if (unlikely(!new_mb)) {
22                 PMD_RX_LOG(ERR, rxq,
23                            "Failed to allocate rx buffer "
24                            "sw_rx_prod %u sw_rx_cons %u mp entries %u free %u",
25                            idx, rxq->sw_rx_cons & NUM_RX_BDS(rxq),
26                            rte_mempool_avail_count(rxq->mb_pool),
27                            rte_mempool_in_use_count(rxq->mb_pool));
28                 return -ENOMEM;
29         }
30         rxq->sw_rx_ring[idx].mbuf = new_mb;
31         rxq->sw_rx_ring[idx].page_offset = 0;
32         mapping = rte_mbuf_data_dma_addr_default(new_mb);
33         /* Advance PROD and get BD pointer */
34         rx_bd = (struct eth_rx_bd *)ecore_chain_produce(&rxq->rx_bd_ring);
35         rx_bd->addr.hi = rte_cpu_to_le_32(U64_HI(mapping));
36         rx_bd->addr.lo = rte_cpu_to_le_32(U64_LO(mapping));
37         rxq->sw_rx_prod++;
38         return 0;
39 }
40
41 static void qede_rx_queue_release_mbufs(struct qede_rx_queue *rxq)
42 {
43         uint16_t i;
44
45         if (rxq->sw_rx_ring != NULL) {
46                 for (i = 0; i < rxq->nb_rx_desc; i++) {
47                         if (rxq->sw_rx_ring[i].mbuf != NULL) {
48                                 rte_pktmbuf_free(rxq->sw_rx_ring[i].mbuf);
49                                 rxq->sw_rx_ring[i].mbuf = NULL;
50                         }
51                 }
52         }
53 }
54
55 void qede_rx_queue_release(void *rx_queue)
56 {
57         struct qede_rx_queue *rxq = rx_queue;
58
59         if (rxq != NULL) {
60                 qede_rx_queue_release_mbufs(rxq);
61                 rte_free(rxq->sw_rx_ring);
62                 rxq->sw_rx_ring = NULL;
63                 rte_free(rxq);
64                 rxq = NULL;
65         }
66 }
67
68 static void qede_tx_queue_release_mbufs(struct qede_tx_queue *txq)
69 {
70         unsigned int i;
71
72         PMD_TX_LOG(DEBUG, txq, "releasing %u mbufs\n", txq->nb_tx_desc);
73
74         if (txq->sw_tx_ring) {
75                 for (i = 0; i < txq->nb_tx_desc; i++) {
76                         if (txq->sw_tx_ring[i].mbuf) {
77                                 rte_pktmbuf_free(txq->sw_tx_ring[i].mbuf);
78                                 txq->sw_tx_ring[i].mbuf = NULL;
79                         }
80                 }
81         }
82 }
83
84 int
85 qede_rx_queue_setup(struct rte_eth_dev *dev, uint16_t queue_idx,
86                     uint16_t nb_desc, unsigned int socket_id,
87                     const struct rte_eth_rxconf *rx_conf,
88                     struct rte_mempool *mp)
89 {
90         struct qede_dev *qdev = dev->data->dev_private;
91         struct ecore_dev *edev = &qdev->edev;
92         struct rte_eth_dev_data *eth_data = dev->data;
93         struct qede_rx_queue *rxq;
94         uint16_t pkt_len = (uint16_t)dev->data->dev_conf.rxmode.max_rx_pkt_len;
95         size_t size;
96         uint16_t data_size;
97         int rc;
98         int i;
99
100         PMD_INIT_FUNC_TRACE(edev);
101
102         /* Note: Ring size/align is controlled by struct rte_eth_desc_lim */
103         if (!rte_is_power_of_2(nb_desc)) {
104                 DP_ERR(edev, "Ring size %u is not power of 2\n",
105                           nb_desc);
106                 return -EINVAL;
107         }
108
109         /* Free memory prior to re-allocation if needed... */
110         if (dev->data->rx_queues[queue_idx] != NULL) {
111                 qede_rx_queue_release(dev->data->rx_queues[queue_idx]);
112                 dev->data->rx_queues[queue_idx] = NULL;
113         }
114
115         /* First allocate the rx queue data structure */
116         rxq = rte_zmalloc_socket("qede_rx_queue", sizeof(struct qede_rx_queue),
117                                  RTE_CACHE_LINE_SIZE, socket_id);
118
119         if (!rxq) {
120                 DP_ERR(edev, "Unable to allocate memory for rxq on socket %u",
121                           socket_id);
122                 return -ENOMEM;
123         }
124
125         rxq->qdev = qdev;
126         rxq->mb_pool = mp;
127         rxq->nb_rx_desc = nb_desc;
128         rxq->queue_id = queue_idx;
129         rxq->port_id = dev->data->port_id;
130
131         /* Sanity check */
132         data_size = (uint16_t)rte_pktmbuf_data_room_size(mp) -
133                                 RTE_PKTMBUF_HEADROOM;
134
135         if (pkt_len > data_size && !dev->data->scattered_rx) {
136                 DP_ERR(edev, "MTU %u should not exceed dataroom %u\n",
137                        pkt_len, data_size);
138                 rte_free(rxq);
139                 return -EINVAL;
140         }
141
142         if (dev->data->scattered_rx)
143                 rxq->rx_buf_size = data_size;
144         else
145                 rxq->rx_buf_size = pkt_len + QEDE_ETH_OVERHEAD;
146
147         qdev->mtu = pkt_len;
148
149         DP_INFO(edev, "MTU = %u ; RX buffer = %u\n",
150                 qdev->mtu, rxq->rx_buf_size);
151
152         if (pkt_len > ETHER_MAX_LEN) {
153                 dev->data->dev_conf.rxmode.jumbo_frame = 1;
154                 DP_NOTICE(edev, false, "jumbo frame enabled\n");
155         } else {
156                 dev->data->dev_conf.rxmode.jumbo_frame = 0;
157         }
158
159         /* Allocate the parallel driver ring for Rx buffers */
160         size = sizeof(*rxq->sw_rx_ring) * rxq->nb_rx_desc;
161         rxq->sw_rx_ring = rte_zmalloc_socket("sw_rx_ring", size,
162                                              RTE_CACHE_LINE_SIZE, socket_id);
163         if (!rxq->sw_rx_ring) {
164                 DP_NOTICE(edev, false,
165                           "Unable to alloc memory for sw_rx_ring on socket %u\n",
166                           socket_id);
167                 rte_free(rxq);
168                 rxq = NULL;
169                 return -ENOMEM;
170         }
171
172         /* Allocate FW Rx ring  */
173         rc = qdev->ops->common->chain_alloc(edev,
174                                             ECORE_CHAIN_USE_TO_CONSUME_PRODUCE,
175                                             ECORE_CHAIN_MODE_NEXT_PTR,
176                                             ECORE_CHAIN_CNT_TYPE_U16,
177                                             rxq->nb_rx_desc,
178                                             sizeof(struct eth_rx_bd),
179                                             &rxq->rx_bd_ring,
180                                             NULL);
181
182         if (rc != ECORE_SUCCESS) {
183                 DP_NOTICE(edev, false,
184                           "Unable to alloc memory for rxbd ring on socket %u\n",
185                           socket_id);
186                 rte_free(rxq->sw_rx_ring);
187                 rxq->sw_rx_ring = NULL;
188                 rte_free(rxq);
189                 rxq = NULL;
190                 return -ENOMEM;
191         }
192
193         /* Allocate FW completion ring */
194         rc = qdev->ops->common->chain_alloc(edev,
195                                             ECORE_CHAIN_USE_TO_CONSUME,
196                                             ECORE_CHAIN_MODE_PBL,
197                                             ECORE_CHAIN_CNT_TYPE_U16,
198                                             rxq->nb_rx_desc,
199                                             sizeof(union eth_rx_cqe),
200                                             &rxq->rx_comp_ring,
201                                             NULL);
202
203         if (rc != ECORE_SUCCESS) {
204                 DP_NOTICE(edev, false,
205                           "Unable to alloc memory for cqe ring on socket %u\n",
206                           socket_id);
207                 /* TBD: Freeing RX BD ring */
208                 rte_free(rxq->sw_rx_ring);
209                 rxq->sw_rx_ring = NULL;
210                 rte_free(rxq);
211                 return -ENOMEM;
212         }
213
214         /* Allocate buffers for the Rx ring */
215         for (i = 0; i < rxq->nb_rx_desc; i++) {
216                 rc = qede_alloc_rx_buffer(rxq);
217                 if (rc) {
218                         DP_NOTICE(edev, false,
219                                   "RX buffer allocation failed at idx=%d\n", i);
220                         goto err4;
221                 }
222         }
223
224         dev->data->rx_queues[queue_idx] = rxq;
225
226         DP_INFO(edev, "rxq %d num_desc %u rx_buf_size=%u socket %u\n",
227                   queue_idx, nb_desc, qdev->mtu, socket_id);
228
229         return 0;
230 err4:
231         qede_rx_queue_release(rxq);
232         return -ENOMEM;
233 }
234
235 void qede_tx_queue_release(void *tx_queue)
236 {
237         struct qede_tx_queue *txq = tx_queue;
238
239         if (txq != NULL) {
240                 qede_tx_queue_release_mbufs(txq);
241                 if (txq->sw_tx_ring) {
242                         rte_free(txq->sw_tx_ring);
243                         txq->sw_tx_ring = NULL;
244                 }
245                 rte_free(txq);
246         }
247         txq = NULL;
248 }
249
250 int
251 qede_tx_queue_setup(struct rte_eth_dev *dev,
252                     uint16_t queue_idx,
253                     uint16_t nb_desc,
254                     unsigned int socket_id,
255                     const struct rte_eth_txconf *tx_conf)
256 {
257         struct qede_dev *qdev = dev->data->dev_private;
258         struct ecore_dev *edev = &qdev->edev;
259         struct qede_tx_queue *txq;
260         int rc;
261
262         PMD_INIT_FUNC_TRACE(edev);
263
264         if (!rte_is_power_of_2(nb_desc)) {
265                 DP_ERR(edev, "Ring size %u is not power of 2\n",
266                        nb_desc);
267                 return -EINVAL;
268         }
269
270         /* Free memory prior to re-allocation if needed... */
271         if (dev->data->tx_queues[queue_idx] != NULL) {
272                 qede_tx_queue_release(dev->data->tx_queues[queue_idx]);
273                 dev->data->tx_queues[queue_idx] = NULL;
274         }
275
276         txq = rte_zmalloc_socket("qede_tx_queue", sizeof(struct qede_tx_queue),
277                                  RTE_CACHE_LINE_SIZE, socket_id);
278
279         if (txq == NULL) {
280                 DP_ERR(edev,
281                        "Unable to allocate memory for txq on socket %u",
282                        socket_id);
283                 return -ENOMEM;
284         }
285
286         txq->nb_tx_desc = nb_desc;
287         txq->qdev = qdev;
288         txq->port_id = dev->data->port_id;
289
290         rc = qdev->ops->common->chain_alloc(edev,
291                                             ECORE_CHAIN_USE_TO_CONSUME_PRODUCE,
292                                             ECORE_CHAIN_MODE_PBL,
293                                             ECORE_CHAIN_CNT_TYPE_U16,
294                                             txq->nb_tx_desc,
295                                             sizeof(union eth_tx_bd_types),
296                                             &txq->tx_pbl,
297                                             NULL);
298         if (rc != ECORE_SUCCESS) {
299                 DP_ERR(edev,
300                        "Unable to allocate memory for txbd ring on socket %u",
301                        socket_id);
302                 qede_tx_queue_release(txq);
303                 return -ENOMEM;
304         }
305
306         /* Allocate software ring */
307         txq->sw_tx_ring = rte_zmalloc_socket("txq->sw_tx_ring",
308                                              (sizeof(struct qede_tx_entry) *
309                                               txq->nb_tx_desc),
310                                              RTE_CACHE_LINE_SIZE, socket_id);
311
312         if (!txq->sw_tx_ring) {
313                 DP_ERR(edev,
314                        "Unable to allocate memory for txbd ring on socket %u",
315                        socket_id);
316                 qede_tx_queue_release(txq);
317                 return -ENOMEM;
318         }
319
320         txq->queue_id = queue_idx;
321
322         txq->nb_tx_avail = txq->nb_tx_desc;
323
324         txq->tx_free_thresh =
325             tx_conf->tx_free_thresh ? tx_conf->tx_free_thresh :
326             (txq->nb_tx_desc - QEDE_DEFAULT_TX_FREE_THRESH);
327
328         dev->data->tx_queues[queue_idx] = txq;
329
330         DP_INFO(edev,
331                   "txq %u num_desc %u tx_free_thresh %u socket %u\n",
332                   queue_idx, nb_desc, txq->tx_free_thresh, socket_id);
333
334         return 0;
335 }
336
337 /* This function inits fp content and resets the SB, RXQ and TXQ arrays */
338 static void qede_init_fp(struct qede_dev *qdev)
339 {
340         struct qede_fastpath *fp;
341         uint8_t i, rss_id, tc;
342         int fp_rx = qdev->fp_num_rx, rxq = 0, txq = 0;
343
344         memset((void *)qdev->fp_array, 0, (QEDE_QUEUE_CNT(qdev) *
345                                            sizeof(*qdev->fp_array)));
346         memset((void *)qdev->sb_array, 0, (QEDE_QUEUE_CNT(qdev) *
347                                            sizeof(*qdev->sb_array)));
348         for_each_queue(i) {
349                 fp = &qdev->fp_array[i];
350                 if (fp_rx) {
351                         fp->type = QEDE_FASTPATH_RX;
352                         fp_rx--;
353                 } else{
354                         fp->type = QEDE_FASTPATH_TX;
355                 }
356                 fp->qdev = qdev;
357                 fp->id = i;
358                 fp->sb_info = &qdev->sb_array[i];
359                 snprintf(fp->name, sizeof(fp->name), "%s-fp-%d", "qdev", i);
360         }
361
362         qdev->gro_disable = gro_disable;
363 }
364
365 void qede_free_fp_arrays(struct qede_dev *qdev)
366 {
367         /* It asseumes qede_free_mem_load() is called before */
368         if (qdev->fp_array != NULL) {
369                 rte_free(qdev->fp_array);
370                 qdev->fp_array = NULL;
371         }
372
373         if (qdev->sb_array != NULL) {
374                 rte_free(qdev->sb_array);
375                 qdev->sb_array = NULL;
376         }
377 }
378
379 int qede_alloc_fp_array(struct qede_dev *qdev)
380 {
381         struct qede_fastpath *fp;
382         struct ecore_dev *edev = &qdev->edev;
383         int i;
384
385         qdev->fp_array = rte_calloc("fp", QEDE_QUEUE_CNT(qdev),
386                                     sizeof(*qdev->fp_array),
387                                     RTE_CACHE_LINE_SIZE);
388
389         if (!qdev->fp_array) {
390                 DP_ERR(edev, "fp array allocation failed\n");
391                 return -ENOMEM;
392         }
393
394         qdev->sb_array = rte_calloc("sb", QEDE_QUEUE_CNT(qdev),
395                                     sizeof(*qdev->sb_array),
396                                     RTE_CACHE_LINE_SIZE);
397
398         if (!qdev->sb_array) {
399                 DP_ERR(edev, "sb array allocation failed\n");
400                 rte_free(qdev->fp_array);
401                 return -ENOMEM;
402         }
403
404         return 0;
405 }
406
407 /* This function allocates fast-path status block memory */
408 static int
409 qede_alloc_mem_sb(struct qede_dev *qdev, struct ecore_sb_info *sb_info,
410                   uint16_t sb_id)
411 {
412         struct ecore_dev *edev = &qdev->edev;
413         struct status_block *sb_virt;
414         dma_addr_t sb_phys;
415         int rc;
416
417         sb_virt = OSAL_DMA_ALLOC_COHERENT(edev, &sb_phys, sizeof(*sb_virt));
418
419         if (!sb_virt) {
420                 DP_ERR(edev, "Status block allocation failed\n");
421                 return -ENOMEM;
422         }
423
424         rc = qdev->ops->common->sb_init(edev, sb_info,
425                                         sb_virt, sb_phys, sb_id,
426                                         QED_SB_TYPE_L2_QUEUE);
427         if (rc) {
428                 DP_ERR(edev, "Status block initialization failed\n");
429                 /* TBD: No dma_free_coherent possible */
430                 return rc;
431         }
432
433         return 0;
434 }
435
436 int qede_alloc_fp_resc(struct qede_dev *qdev)
437 {
438         struct ecore_dev *edev = &qdev->edev;
439         struct qede_fastpath *fp;
440         uint32_t num_sbs;
441         int rc, i;
442
443         if (IS_VF(edev))
444                 ecore_vf_get_num_sbs(ECORE_LEADING_HWFN(edev), &num_sbs);
445         else
446                 num_sbs = (ecore_cxt_get_proto_cid_count
447                           (ECORE_LEADING_HWFN(edev), PROTOCOLID_ETH, NULL)) / 2;
448
449         if (num_sbs == 0) {
450                 DP_ERR(edev, "No status blocks available\n");
451                 return -EINVAL;
452         }
453
454         if (qdev->fp_array)
455                 qede_free_fp_arrays(qdev);
456
457         rc = qede_alloc_fp_array(qdev);
458         if (rc != 0)
459                 return rc;
460
461         qede_init_fp(qdev);
462
463         for (i = 0; i < QEDE_QUEUE_CNT(qdev); i++) {
464                 fp = &qdev->fp_array[i];
465                 if (qede_alloc_mem_sb(qdev, fp->sb_info, i % num_sbs)) {
466                         qede_free_fp_arrays(qdev);
467                         return -ENOMEM;
468                 }
469         }
470
471         return 0;
472 }
473
474 void qede_dealloc_fp_resc(struct rte_eth_dev *eth_dev)
475 {
476         struct qede_dev *qdev = QEDE_INIT_QDEV(eth_dev);
477
478         qede_free_mem_load(eth_dev);
479         qede_free_fp_arrays(qdev);
480 }
481
482 static inline void
483 qede_update_rx_prod(struct qede_dev *edev, struct qede_rx_queue *rxq)
484 {
485         uint16_t bd_prod = ecore_chain_get_prod_idx(&rxq->rx_bd_ring);
486         uint16_t cqe_prod = ecore_chain_get_prod_idx(&rxq->rx_comp_ring);
487         struct eth_rx_prod_data rx_prods = { 0 };
488
489         /* Update producers */
490         rx_prods.bd_prod = rte_cpu_to_le_16(bd_prod);
491         rx_prods.cqe_prod = rte_cpu_to_le_16(cqe_prod);
492
493         /* Make sure that the BD and SGE data is updated before updating the
494          * producers since FW might read the BD/SGE right after the producer
495          * is updated.
496          */
497         rte_wmb();
498
499         internal_ram_wr(rxq->hw_rxq_prod_addr, sizeof(rx_prods),
500                         (uint32_t *)&rx_prods);
501
502         /* mmiowb is needed to synchronize doorbell writes from more than one
503          * processor. It guarantees that the write arrives to the device before
504          * the napi lock is released and another qede_poll is called (possibly
505          * on another CPU). Without this barrier, the next doorbell can bypass
506          * this doorbell. This is applicable to IA64/Altix systems.
507          */
508         rte_wmb();
509
510         PMD_RX_LOG(DEBUG, rxq, "bd_prod %u  cqe_prod %u\n", bd_prod, cqe_prod);
511 }
512
513 static int qede_start_queues(struct rte_eth_dev *eth_dev, bool clear_stats)
514 {
515         struct qede_dev *qdev = eth_dev->data->dev_private;
516         struct ecore_dev *edev = &qdev->edev;
517         struct ecore_queue_start_common_params q_params;
518         struct qed_dev_info *qed_info = &qdev->dev_info.common;
519         struct qed_update_vport_params vport_update_params;
520         struct qede_tx_queue *txq;
521         struct qede_fastpath *fp;
522         dma_addr_t p_phys_table;
523         int txq_index;
524         uint16_t page_cnt;
525         int vlan_removal_en = 1;
526         int rc, tc, i;
527
528         for_each_queue(i) {
529                 fp = &qdev->fp_array[i];
530                 if (fp->type & QEDE_FASTPATH_RX) {
531                         p_phys_table = ecore_chain_get_pbl_phys(&fp->rxq->
532                                                                 rx_comp_ring);
533                         page_cnt = ecore_chain_get_page_cnt(&fp->rxq->
534                                                                 rx_comp_ring);
535
536                         memset(&q_params, 0, sizeof(q_params));
537                         q_params.queue_id = i;
538                         q_params.vport_id = 0;
539                         q_params.sb = fp->sb_info->igu_sb_id;
540                         q_params.sb_idx = RX_PI;
541
542                         ecore_sb_ack(fp->sb_info, IGU_INT_DISABLE, 0);
543
544                         rc = qdev->ops->q_rx_start(edev, i, &q_params,
545                                            fp->rxq->rx_buf_size,
546                                            fp->rxq->rx_bd_ring.p_phys_addr,
547                                            p_phys_table,
548                                            page_cnt,
549                                            &fp->rxq->hw_rxq_prod_addr);
550                         if (rc) {
551                                 DP_ERR(edev, "Start rxq #%d failed %d\n",
552                                        fp->rxq->queue_id, rc);
553                                 return rc;
554                         }
555
556                         fp->rxq->hw_cons_ptr =
557                                         &fp->sb_info->sb_virt->pi_array[RX_PI];
558
559                         qede_update_rx_prod(qdev, fp->rxq);
560                 }
561
562                 if (!(fp->type & QEDE_FASTPATH_TX))
563                         continue;
564                 for (tc = 0; tc < qdev->num_tc; tc++) {
565                         txq = fp->txqs[tc];
566                         txq_index = tc * QEDE_RSS_COUNT(qdev) + i;
567
568                         p_phys_table = ecore_chain_get_pbl_phys(&txq->tx_pbl);
569                         page_cnt = ecore_chain_get_page_cnt(&txq->tx_pbl);
570
571                         memset(&q_params, 0, sizeof(q_params));
572                         q_params.queue_id = txq->queue_id;
573                         q_params.vport_id = 0;
574                         q_params.sb = fp->sb_info->igu_sb_id;
575                         q_params.sb_idx = TX_PI(tc);
576
577                         rc = qdev->ops->q_tx_start(edev, i, &q_params,
578                                                    p_phys_table,
579                                                    page_cnt, /* **pp_doorbell */
580                                                    &txq->doorbell_addr);
581                         if (rc) {
582                                 DP_ERR(edev, "Start txq %u failed %d\n",
583                                        txq_index, rc);
584                                 return rc;
585                         }
586
587                         txq->hw_cons_ptr =
588                             &fp->sb_info->sb_virt->pi_array[TX_PI(tc)];
589                         SET_FIELD(txq->tx_db.data.params,
590                                   ETH_DB_DATA_DEST, DB_DEST_XCM);
591                         SET_FIELD(txq->tx_db.data.params, ETH_DB_DATA_AGG_CMD,
592                                   DB_AGG_CMD_SET);
593                         SET_FIELD(txq->tx_db.data.params,
594                                   ETH_DB_DATA_AGG_VAL_SEL,
595                                   DQ_XCM_ETH_TX_BD_PROD_CMD);
596
597                         txq->tx_db.data.agg_flags = DQ_XCM_ETH_DQ_CF_CMD;
598                 }
599         }
600
601         /* Prepare and send the vport enable */
602         memset(&vport_update_params, 0, sizeof(vport_update_params));
603         /* Update MTU via vport update */
604         vport_update_params.mtu = qdev->mtu;
605         vport_update_params.vport_id = 0;
606         vport_update_params.update_vport_active_flg = 1;
607         vport_update_params.vport_active_flg = 1;
608
609         /* @DPDK */
610         if (qed_info->mf_mode == MF_NPAR && qed_info->tx_switching) {
611                 /* TBD: Check SRIOV enabled for VF */
612                 vport_update_params.update_tx_switching_flg = 1;
613                 vport_update_params.tx_switching_flg = 1;
614         }
615
616         rc = qdev->ops->vport_update(edev, &vport_update_params);
617         if (rc) {
618                 DP_ERR(edev, "Update V-PORT failed %d\n", rc);
619                 return rc;
620         }
621
622         return 0;
623 }
624
625 static bool qede_tunn_exist(uint16_t flag)
626 {
627         return !!((PARSING_AND_ERR_FLAGS_TUNNELEXIST_MASK <<
628                     PARSING_AND_ERR_FLAGS_TUNNELEXIST_SHIFT) & flag);
629 }
630
631 /*
632  * qede_check_tunn_csum_l4:
633  * Returns:
634  * 1 : If L4 csum is enabled AND if the validation has failed.
635  * 0 : Otherwise
636  */
637 static inline uint8_t qede_check_tunn_csum_l4(uint16_t flag)
638 {
639         if ((PARSING_AND_ERR_FLAGS_TUNNELL4CHKSMWASCALCULATED_MASK <<
640              PARSING_AND_ERR_FLAGS_TUNNELL4CHKSMWASCALCULATED_SHIFT) & flag)
641                 return !!((PARSING_AND_ERR_FLAGS_TUNNELL4CHKSMERROR_MASK <<
642                         PARSING_AND_ERR_FLAGS_TUNNELL4CHKSMERROR_SHIFT) & flag);
643
644         return 0;
645 }
646
647 static inline uint8_t qede_check_notunn_csum_l4(uint16_t flag)
648 {
649         if ((PARSING_AND_ERR_FLAGS_L4CHKSMWASCALCULATED_MASK <<
650              PARSING_AND_ERR_FLAGS_L4CHKSMWASCALCULATED_SHIFT) & flag)
651                 return !!((PARSING_AND_ERR_FLAGS_L4CHKSMERROR_MASK <<
652                            PARSING_AND_ERR_FLAGS_L4CHKSMERROR_SHIFT) & flag);
653
654         return 0;
655 }
656
657 static inline uint8_t
658 qede_check_notunn_csum_l3(struct rte_mbuf *m, uint16_t flag)
659 {
660         struct ipv4_hdr *ip;
661         uint16_t pkt_csum;
662         uint16_t calc_csum;
663         uint16_t val;
664
665         val = ((PARSING_AND_ERR_FLAGS_IPHDRERROR_MASK <<
666                 PARSING_AND_ERR_FLAGS_IPHDRERROR_SHIFT) & flag);
667
668         if (unlikely(val)) {
669                 m->packet_type = qede_rx_cqe_to_pkt_type(flag);
670                 if (RTE_ETH_IS_IPV4_HDR(m->packet_type)) {
671                         ip = rte_pktmbuf_mtod_offset(m, struct ipv4_hdr *,
672                                            sizeof(struct ether_hdr));
673                         pkt_csum = ip->hdr_checksum;
674                         ip->hdr_checksum = 0;
675                         calc_csum = rte_ipv4_cksum(ip);
676                         ip->hdr_checksum = pkt_csum;
677                         return (calc_csum != pkt_csum);
678                 } else if (RTE_ETH_IS_IPV6_HDR(m->packet_type)) {
679                         return 1;
680                 }
681         }
682         return 0;
683 }
684
685 static inline void qede_rx_bd_ring_consume(struct qede_rx_queue *rxq)
686 {
687         ecore_chain_consume(&rxq->rx_bd_ring);
688         rxq->sw_rx_cons++;
689 }
690
691 static inline void
692 qede_reuse_page(struct qede_dev *qdev,
693                 struct qede_rx_queue *rxq, struct qede_rx_entry *curr_cons)
694 {
695         struct eth_rx_bd *rx_bd_prod = ecore_chain_produce(&rxq->rx_bd_ring);
696         uint16_t idx = rxq->sw_rx_cons & NUM_RX_BDS(rxq);
697         struct qede_rx_entry *curr_prod;
698         dma_addr_t new_mapping;
699
700         curr_prod = &rxq->sw_rx_ring[idx];
701         *curr_prod = *curr_cons;
702
703         new_mapping = rte_mbuf_data_dma_addr_default(curr_prod->mbuf) +
704                       curr_prod->page_offset;
705
706         rx_bd_prod->addr.hi = rte_cpu_to_le_32(U64_HI(new_mapping));
707         rx_bd_prod->addr.lo = rte_cpu_to_le_32(U64_LO(new_mapping));
708
709         rxq->sw_rx_prod++;
710 }
711
712 static inline void
713 qede_recycle_rx_bd_ring(struct qede_rx_queue *rxq,
714                         struct qede_dev *qdev, uint8_t count)
715 {
716         struct qede_rx_entry *curr_cons;
717
718         for (; count > 0; count--) {
719                 curr_cons = &rxq->sw_rx_ring[rxq->sw_rx_cons & NUM_RX_BDS(rxq)];
720                 qede_reuse_page(qdev, rxq, curr_cons);
721                 qede_rx_bd_ring_consume(rxq);
722         }
723 }
724
725 static inline uint32_t qede_rx_cqe_to_pkt_type(uint16_t flags)
726 {
727         uint16_t val;
728
729         /* Lookup table */
730         static const uint32_t
731         ptype_lkup_tbl[QEDE_PKT_TYPE_MAX] __rte_cache_aligned = {
732                 [QEDE_PKT_TYPE_IPV4] = RTE_PTYPE_L3_IPV4,
733                 [QEDE_PKT_TYPE_IPV6] = RTE_PTYPE_L3_IPV6,
734                 [QEDE_PKT_TYPE_IPV4_TCP] = RTE_PTYPE_L3_IPV4 | RTE_PTYPE_L4_TCP,
735                 [QEDE_PKT_TYPE_IPV6_TCP] = RTE_PTYPE_L3_IPV6 | RTE_PTYPE_L4_TCP,
736                 [QEDE_PKT_TYPE_IPV4_UDP] = RTE_PTYPE_L3_IPV4 | RTE_PTYPE_L4_UDP,
737                 [QEDE_PKT_TYPE_IPV6_UDP] = RTE_PTYPE_L3_IPV6 | RTE_PTYPE_L4_UDP,
738         };
739
740         /* Bits (0..3) provides L3/L4 protocol type */
741         val = ((PARSING_AND_ERR_FLAGS_L3TYPE_MASK <<
742                PARSING_AND_ERR_FLAGS_L3TYPE_SHIFT) |
743                (PARSING_AND_ERR_FLAGS_L4PROTOCOL_MASK <<
744                 PARSING_AND_ERR_FLAGS_L4PROTOCOL_SHIFT)) & flags;
745
746         if (val < QEDE_PKT_TYPE_MAX)
747                 return ptype_lkup_tbl[val] | RTE_PTYPE_L2_ETHER;
748         else
749                 return RTE_PTYPE_UNKNOWN;
750 }
751
752 static inline uint32_t qede_rx_cqe_to_tunn_pkt_type(uint16_t flags)
753 {
754         uint32_t val;
755
756         /* Lookup table */
757         static const uint32_t
758         ptype_tunn_lkup_tbl[QEDE_PKT_TYPE_TUNN_MAX_TYPE] __rte_cache_aligned = {
759                 [QEDE_PKT_TYPE_UNKNOWN] = RTE_PTYPE_UNKNOWN,
760                 [QEDE_PKT_TYPE_TUNN_GENEVE] = RTE_PTYPE_TUNNEL_GENEVE,
761                 [QEDE_PKT_TYPE_TUNN_GRE] = RTE_PTYPE_TUNNEL_GRE,
762                 [QEDE_PKT_TYPE_TUNN_VXLAN] = RTE_PTYPE_TUNNEL_VXLAN,
763                 [QEDE_PKT_TYPE_TUNN_L2_TENID_NOEXIST_GENEVE] =
764                                 RTE_PTYPE_TUNNEL_GENEVE | RTE_PTYPE_L2_ETHER,
765                 [QEDE_PKT_TYPE_TUNN_L2_TENID_NOEXIST_GRE] =
766                                 RTE_PTYPE_TUNNEL_GRE | RTE_PTYPE_L2_ETHER,
767                 [QEDE_PKT_TYPE_TUNN_L2_TENID_NOEXIST_VXLAN] =
768                                 RTE_PTYPE_TUNNEL_VXLAN | RTE_PTYPE_L2_ETHER,
769                 [QEDE_PKT_TYPE_TUNN_L2_TENID_EXIST_GENEVE] =
770                                 RTE_PTYPE_TUNNEL_GENEVE | RTE_PTYPE_L2_ETHER,
771                 [QEDE_PKT_TYPE_TUNN_L2_TENID_EXIST_GRE] =
772                                 RTE_PTYPE_TUNNEL_GRE | RTE_PTYPE_L2_ETHER,
773                 [QEDE_PKT_TYPE_TUNN_L2_TENID_EXIST_VXLAN] =
774                                 RTE_PTYPE_TUNNEL_VXLAN | RTE_PTYPE_L2_ETHER,
775                 [QEDE_PKT_TYPE_TUNN_IPV4_TENID_NOEXIST_GENEVE] =
776                                 RTE_PTYPE_TUNNEL_GENEVE | RTE_PTYPE_L3_IPV4,
777                 [QEDE_PKT_TYPE_TUNN_IPV4_TENID_NOEXIST_GRE] =
778                                 RTE_PTYPE_TUNNEL_GRE | RTE_PTYPE_L3_IPV4,
779                 [QEDE_PKT_TYPE_TUNN_IPV4_TENID_NOEXIST_VXLAN] =
780                                 RTE_PTYPE_TUNNEL_VXLAN | RTE_PTYPE_L3_IPV4,
781                 [QEDE_PKT_TYPE_TUNN_IPV4_TENID_EXIST_GENEVE] =
782                                 RTE_PTYPE_TUNNEL_GENEVE | RTE_PTYPE_L3_IPV4,
783                 [QEDE_PKT_TYPE_TUNN_IPV4_TENID_EXIST_GRE] =
784                                 RTE_PTYPE_TUNNEL_GRE | RTE_PTYPE_L3_IPV4,
785                 [QEDE_PKT_TYPE_TUNN_IPV4_TENID_EXIST_VXLAN] =
786                                 RTE_PTYPE_TUNNEL_VXLAN | RTE_PTYPE_L3_IPV4,
787                 [QEDE_PKT_TYPE_TUNN_IPV6_TENID_NOEXIST_GENEVE] =
788                                 RTE_PTYPE_TUNNEL_GENEVE | RTE_PTYPE_L3_IPV6,
789                 [QEDE_PKT_TYPE_TUNN_IPV6_TENID_NOEXIST_GRE] =
790                                 RTE_PTYPE_TUNNEL_GRE | RTE_PTYPE_L3_IPV6,
791                 [QEDE_PKT_TYPE_TUNN_IPV6_TENID_NOEXIST_VXLAN] =
792                                 RTE_PTYPE_TUNNEL_VXLAN | RTE_PTYPE_L3_IPV6,
793                 [QEDE_PKT_TYPE_TUNN_IPV6_TENID_EXIST_GENEVE] =
794                                 RTE_PTYPE_TUNNEL_GENEVE | RTE_PTYPE_L3_IPV6,
795                 [QEDE_PKT_TYPE_TUNN_IPV6_TENID_EXIST_GRE] =
796                                 RTE_PTYPE_TUNNEL_GRE | RTE_PTYPE_L3_IPV6,
797                 [QEDE_PKT_TYPE_TUNN_IPV6_TENID_EXIST_VXLAN] =
798                                 RTE_PTYPE_TUNNEL_VXLAN | RTE_PTYPE_L3_IPV6,
799         };
800
801         /* Cover bits[4-0] to include tunn_type and next protocol */
802         val = ((ETH_TUNNEL_PARSING_FLAGS_TYPE_MASK <<
803                 ETH_TUNNEL_PARSING_FLAGS_TYPE_SHIFT) |
804                 (ETH_TUNNEL_PARSING_FLAGS_NEXT_PROTOCOL_MASK <<
805                 ETH_TUNNEL_PARSING_FLAGS_NEXT_PROTOCOL_SHIFT)) & flags;
806
807         if (val < QEDE_PKT_TYPE_TUNN_MAX_TYPE)
808                 return ptype_tunn_lkup_tbl[val];
809         else
810                 return RTE_PTYPE_UNKNOWN;
811 }
812
813
814 int qede_process_sg_pkts(void *p_rxq,  struct rte_mbuf *rx_mb,
815                          int num_segs, uint16_t pkt_len)
816 {
817         struct qede_rx_queue *rxq = p_rxq;
818         struct qede_dev *qdev = rxq->qdev;
819         struct ecore_dev *edev = &qdev->edev;
820         uint16_t sw_rx_index, cur_size;
821
822         register struct rte_mbuf *seg1 = NULL;
823         register struct rte_mbuf *seg2 = NULL;
824
825         seg1 = rx_mb;
826         while (num_segs) {
827                 cur_size = pkt_len > rxq->rx_buf_size ?
828                                 rxq->rx_buf_size : pkt_len;
829                 if (!cur_size) {
830                         PMD_RX_LOG(DEBUG, rxq,
831                                    "SG packet, len and num BD mismatch\n");
832                         qede_recycle_rx_bd_ring(rxq, qdev, num_segs);
833                         return -EINVAL;
834                 }
835
836                 if (qede_alloc_rx_buffer(rxq)) {
837                         uint8_t index;
838
839                         PMD_RX_LOG(DEBUG, rxq, "Buffer allocation failed\n");
840                         index = rxq->port_id;
841                         rte_eth_devices[index].data->rx_mbuf_alloc_failed++;
842                         rxq->rx_alloc_errors++;
843                         return -ENOMEM;
844                 }
845
846                 sw_rx_index = rxq->sw_rx_cons & NUM_RX_BDS(rxq);
847                 seg2 = rxq->sw_rx_ring[sw_rx_index].mbuf;
848                 qede_rx_bd_ring_consume(rxq);
849                 pkt_len -= cur_size;
850                 seg2->data_len = cur_size;
851                 seg1->next = seg2;
852                 seg1 = seg1->next;
853                 num_segs--;
854                 rxq->rx_segs++;
855                 continue;
856         }
857         seg1 = NULL;
858
859         if (pkt_len)
860                 PMD_RX_LOG(DEBUG, rxq,
861                            "Mapped all BDs of jumbo, but still have %d bytes\n",
862                            pkt_len);
863
864         return ECORE_SUCCESS;
865 }
866
867 uint16_t
868 qede_recv_pkts(void *p_rxq, struct rte_mbuf **rx_pkts, uint16_t nb_pkts)
869 {
870         struct qede_rx_queue *rxq = p_rxq;
871         struct qede_dev *qdev = rxq->qdev;
872         struct ecore_dev *edev = &qdev->edev;
873         struct qede_fastpath *fp = &qdev->fp_array[rxq->queue_id];
874         uint16_t hw_comp_cons, sw_comp_cons, sw_rx_index;
875         uint16_t rx_pkt = 0;
876         union eth_rx_cqe *cqe;
877         struct eth_fast_path_rx_reg_cqe *fp_cqe;
878         register struct rte_mbuf *rx_mb = NULL;
879         register struct rte_mbuf *seg1 = NULL;
880         enum eth_rx_cqe_type cqe_type;
881         uint16_t len, pad, preload_idx, pkt_len, parse_flag;
882         uint8_t csum_flag, num_segs;
883         enum rss_hash_type htype;
884         uint8_t tunn_parse_flag;
885         int ret;
886
887         hw_comp_cons = rte_le_to_cpu_16(*rxq->hw_cons_ptr);
888         sw_comp_cons = ecore_chain_get_cons_idx(&rxq->rx_comp_ring);
889
890         rte_rmb();
891
892         if (hw_comp_cons == sw_comp_cons)
893                 return 0;
894
895         while (sw_comp_cons != hw_comp_cons) {
896                 /* Get the CQE from the completion ring */
897                 cqe =
898                     (union eth_rx_cqe *)ecore_chain_consume(&rxq->rx_comp_ring);
899                 cqe_type = cqe->fast_path_regular.type;
900
901                 if (unlikely(cqe_type == ETH_RX_CQE_TYPE_SLOW_PATH)) {
902                         PMD_RX_LOG(DEBUG, rxq, "Got a slowath CQE\n");
903
904                         qdev->ops->eth_cqe_completion(edev, fp->id,
905                                 (struct eth_slow_path_rx_cqe *)cqe);
906                         goto next_cqe;
907                 }
908
909                 /* Get the data from the SW ring */
910                 sw_rx_index = rxq->sw_rx_cons & NUM_RX_BDS(rxq);
911                 rx_mb = rxq->sw_rx_ring[sw_rx_index].mbuf;
912                 assert(rx_mb != NULL);
913
914                 /* non GRO */
915                 fp_cqe = &cqe->fast_path_regular;
916
917                 len = rte_le_to_cpu_16(fp_cqe->len_on_first_bd);
918                 pad = fp_cqe->placement_offset;
919                 assert((len + pad) <= rx_mb->buf_len);
920
921                 PMD_RX_LOG(DEBUG, rxq,
922                            "CQE type = 0x%x, flags = 0x%x, vlan = 0x%x"
923                            " len = %u, parsing_flags = %d\n",
924                            cqe_type, fp_cqe->bitfields,
925                            rte_le_to_cpu_16(fp_cqe->vlan_tag),
926                            len, rte_le_to_cpu_16(fp_cqe->pars_flags.flags));
927
928                 /* If this is an error packet then drop it */
929                 parse_flag =
930                     rte_le_to_cpu_16(cqe->fast_path_regular.pars_flags.flags);
931
932                 rx_mb->ol_flags = 0;
933
934                 if (qede_tunn_exist(parse_flag)) {
935                         PMD_RX_LOG(DEBUG, rxq, "Rx tunneled packet\n");
936                         if (unlikely(qede_check_tunn_csum_l4(parse_flag))) {
937                                 PMD_RX_LOG(ERR, rxq,
938                                             "L4 csum failed, flags = 0x%x\n",
939                                             parse_flag);
940                                 rxq->rx_hw_errors++;
941                                 rx_mb->ol_flags |= PKT_RX_L4_CKSUM_BAD;
942                         } else {
943                                 tunn_parse_flag =
944                                                 fp_cqe->tunnel_pars_flags.flags;
945                                 rx_mb->packet_type =
946                                         qede_rx_cqe_to_tunn_pkt_type(
947                                                         tunn_parse_flag);
948                         }
949                 } else {
950                         PMD_RX_LOG(DEBUG, rxq, "Rx non-tunneled packet\n");
951                         if (unlikely(qede_check_notunn_csum_l4(parse_flag))) {
952                                 PMD_RX_LOG(ERR, rxq,
953                                             "L4 csum failed, flags = 0x%x\n",
954                                             parse_flag);
955                                 rxq->rx_hw_errors++;
956                                 rx_mb->ol_flags |= PKT_RX_L4_CKSUM_BAD;
957                         } else if (unlikely(qede_check_notunn_csum_l3(rx_mb,
958                                                         parse_flag))) {
959                                 PMD_RX_LOG(ERR, rxq,
960                                            "IP csum failed, flags = 0x%x\n",
961                                            parse_flag);
962                                 rxq->rx_hw_errors++;
963                                 rx_mb->ol_flags |= PKT_RX_IP_CKSUM_BAD;
964                         } else {
965                                 rx_mb->packet_type =
966                                         qede_rx_cqe_to_pkt_type(parse_flag);
967                         }
968                 }
969
970                 PMD_RX_LOG(INFO, rxq, "packet_type 0x%x\n", rx_mb->packet_type);
971
972                 if (unlikely(qede_alloc_rx_buffer(rxq) != 0)) {
973                         PMD_RX_LOG(ERR, rxq,
974                                    "New buffer allocation failed,"
975                                    "dropping incoming packet\n");
976                         qede_recycle_rx_bd_ring(rxq, qdev, fp_cqe->bd_num);
977                         rte_eth_devices[rxq->port_id].
978                             data->rx_mbuf_alloc_failed++;
979                         rxq->rx_alloc_errors++;
980                         break;
981                 }
982
983                 qede_rx_bd_ring_consume(rxq);
984
985                 if (fp_cqe->bd_num > 1) {
986                         pkt_len = rte_le_to_cpu_16(fp_cqe->pkt_len);
987                         num_segs = fp_cqe->bd_num - 1;
988
989                         rxq->rx_segs++;
990
991                         pkt_len -= len;
992                         seg1 = rx_mb;
993                         ret = qede_process_sg_pkts(p_rxq, seg1, num_segs,
994                                                    pkt_len);
995                         if (ret != ECORE_SUCCESS) {
996                                 qede_recycle_rx_bd_ring(rxq, qdev,
997                                                         fp_cqe->bd_num);
998                                 goto next_cqe;
999                         }
1000                 }
1001
1002                 /* Prefetch next mbuf while processing current one. */
1003                 preload_idx = rxq->sw_rx_cons & NUM_RX_BDS(rxq);
1004                 rte_prefetch0(rxq->sw_rx_ring[preload_idx].mbuf);
1005
1006                 /* Update rest of the MBUF fields */
1007                 rx_mb->data_off = pad + RTE_PKTMBUF_HEADROOM;
1008                 rx_mb->nb_segs = fp_cqe->bd_num;
1009                 rx_mb->data_len = len;
1010                 rx_mb->pkt_len = fp_cqe->pkt_len;
1011                 rx_mb->port = rxq->port_id;
1012
1013                 htype = (uint8_t)GET_FIELD(fp_cqe->bitfields,
1014                                 ETH_FAST_PATH_RX_REG_CQE_RSS_HASH_TYPE);
1015                 if (qdev->rss_enable && htype) {
1016                         rx_mb->ol_flags |= PKT_RX_RSS_HASH;
1017                         rx_mb->hash.rss = rte_le_to_cpu_32(fp_cqe->rss_hash);
1018                         PMD_RX_LOG(DEBUG, rxq, "Hash result 0x%x\n",
1019                                    rx_mb->hash.rss);
1020                 }
1021
1022                 rte_prefetch1(rte_pktmbuf_mtod(rx_mb, void *));
1023
1024                 if (CQE_HAS_VLAN(parse_flag)) {
1025                         rx_mb->vlan_tci = rte_le_to_cpu_16(fp_cqe->vlan_tag);
1026                         rx_mb->ol_flags |= PKT_RX_VLAN_PKT;
1027                 }
1028
1029                 if (CQE_HAS_OUTER_VLAN(parse_flag)) {
1030                         /* FW does not provide indication of Outer VLAN tag,
1031                          * which is always stripped, so vlan_tci_outer is set
1032                          * to 0. Here vlan_tag represents inner VLAN tag.
1033                          */
1034                         rx_mb->vlan_tci = rte_le_to_cpu_16(fp_cqe->vlan_tag);
1035                         rx_mb->ol_flags |= PKT_RX_QINQ_PKT;
1036                         rx_mb->vlan_tci_outer = 0;
1037                 }
1038
1039                 rx_pkts[rx_pkt] = rx_mb;
1040                 rx_pkt++;
1041 next_cqe:
1042                 ecore_chain_recycle_consumed(&rxq->rx_comp_ring);
1043                 sw_comp_cons = ecore_chain_get_cons_idx(&rxq->rx_comp_ring);
1044                 if (rx_pkt == nb_pkts) {
1045                         PMD_RX_LOG(DEBUG, rxq,
1046                                    "Budget reached nb_pkts=%u received=%u\n",
1047                                    rx_pkt, nb_pkts);
1048                         break;
1049                 }
1050         }
1051
1052         qede_update_rx_prod(qdev, rxq);
1053
1054         rxq->rcv_pkts += rx_pkt;
1055
1056         PMD_RX_LOG(DEBUG, rxq, "rx_pkts=%u core=%d\n", rx_pkt, rte_lcore_id());
1057
1058         return rx_pkt;
1059 }
1060
1061 static inline int
1062 qede_free_tx_pkt(struct ecore_dev *edev, struct qede_tx_queue *txq)
1063 {
1064         uint16_t nb_segs, idx = TX_CONS(txq);
1065         struct eth_tx_bd *tx_data_bd;
1066         struct rte_mbuf *mbuf = txq->sw_tx_ring[idx].mbuf;
1067
1068         if (unlikely(!mbuf)) {
1069                 PMD_TX_LOG(ERR, txq, "null mbuf\n");
1070                 PMD_TX_LOG(ERR, txq,
1071                            "tx_desc %u tx_avail %u tx_cons %u tx_prod %u\n",
1072                            txq->nb_tx_desc, txq->nb_tx_avail, idx,
1073                            TX_PROD(txq));
1074                 return -1;
1075         }
1076
1077         nb_segs = mbuf->nb_segs;
1078         while (nb_segs) {
1079                 /* It's like consuming rxbuf in recv() */
1080                 ecore_chain_consume(&txq->tx_pbl);
1081                 txq->nb_tx_avail++;
1082                 nb_segs--;
1083         }
1084         rte_pktmbuf_free(mbuf);
1085         txq->sw_tx_ring[idx].mbuf = NULL;
1086
1087         return 0;
1088 }
1089
1090 static inline uint16_t
1091 qede_process_tx_compl(struct ecore_dev *edev, struct qede_tx_queue *txq)
1092 {
1093         uint16_t tx_compl = 0;
1094         uint16_t hw_bd_cons;
1095
1096         hw_bd_cons = rte_le_to_cpu_16(*txq->hw_cons_ptr);
1097         rte_compiler_barrier();
1098
1099         while (hw_bd_cons != ecore_chain_get_cons_idx(&txq->tx_pbl)) {
1100                 if (qede_free_tx_pkt(edev, txq)) {
1101                         PMD_TX_LOG(ERR, txq,
1102                                    "hw_bd_cons = %u, chain_cons = %u\n",
1103                                    hw_bd_cons,
1104                                    ecore_chain_get_cons_idx(&txq->tx_pbl));
1105                         break;
1106                 }
1107                 txq->sw_tx_cons++;      /* Making TXD available */
1108                 tx_compl++;
1109         }
1110
1111         PMD_TX_LOG(DEBUG, txq, "Tx compl %u sw_tx_cons %u avail %u\n",
1112                    tx_compl, txq->sw_tx_cons, txq->nb_tx_avail);
1113         return tx_compl;
1114 }
1115
1116 /* Populate scatter gather buffer descriptor fields */
1117 static inline uint16_t qede_encode_sg_bd(struct qede_tx_queue *p_txq,
1118                                          struct rte_mbuf *m_seg,
1119                                          uint16_t count,
1120                                          struct eth_tx_1st_bd *bd1)
1121 {
1122         struct qede_tx_queue *txq = p_txq;
1123         struct eth_tx_2nd_bd *bd2 = NULL;
1124         struct eth_tx_3rd_bd *bd3 = NULL;
1125         struct eth_tx_bd *tx_bd = NULL;
1126         uint16_t nb_segs = count;
1127         dma_addr_t mapping;
1128
1129         /* Check for scattered buffers */
1130         while (m_seg) {
1131                 if (nb_segs == 1) {
1132                         bd2 = (struct eth_tx_2nd_bd *)
1133                                 ecore_chain_produce(&txq->tx_pbl);
1134                         memset(bd2, 0, sizeof(*bd2));
1135                         mapping = rte_mbuf_data_dma_addr(m_seg);
1136                         bd2->addr.hi = rte_cpu_to_le_32(U64_HI(mapping));
1137                         bd2->addr.lo = rte_cpu_to_le_32(U64_LO(mapping));
1138                         bd2->nbytes = rte_cpu_to_le_16(m_seg->data_len);
1139                 } else if (nb_segs == 2) {
1140                         bd3 = (struct eth_tx_3rd_bd *)
1141                                 ecore_chain_produce(&txq->tx_pbl);
1142                         memset(bd3, 0, sizeof(*bd3));
1143                         mapping = rte_mbuf_data_dma_addr(m_seg);
1144                         bd3->addr.hi = rte_cpu_to_le_32(U64_HI(mapping));
1145                         bd3->addr.lo = rte_cpu_to_le_32(U64_LO(mapping));
1146                         bd3->nbytes = rte_cpu_to_le_16(m_seg->data_len);
1147                 } else {
1148                         tx_bd = (struct eth_tx_bd *)
1149                                 ecore_chain_produce(&txq->tx_pbl);
1150                         memset(tx_bd, 0, sizeof(*tx_bd));
1151                         mapping = rte_mbuf_data_dma_addr(m_seg);
1152                         tx_bd->addr.hi = rte_cpu_to_le_32(U64_HI(mapping));
1153                         tx_bd->addr.lo = rte_cpu_to_le_32(U64_LO(mapping));
1154                         tx_bd->nbytes = rte_cpu_to_le_16(m_seg->data_len);
1155                 }
1156                 nb_segs++;
1157                 bd1->data.nbds = nb_segs;
1158                 m_seg = m_seg->next;
1159         }
1160
1161         /* Return total scattered buffers */
1162         return nb_segs;
1163 }
1164
1165 uint16_t
1166 qede_xmit_pkts(void *p_txq, struct rte_mbuf **tx_pkts, uint16_t nb_pkts)
1167 {
1168         struct qede_tx_queue *txq = p_txq;
1169         struct qede_dev *qdev = txq->qdev;
1170         struct ecore_dev *edev = &qdev->edev;
1171         struct qede_fastpath *fp;
1172         struct eth_tx_1st_bd *bd1;
1173         struct rte_mbuf *m_seg = NULL;
1174         uint16_t nb_tx_pkts;
1175         uint16_t nb_pkt_sent = 0;
1176         uint16_t bd_prod;
1177         uint16_t idx;
1178         uint16_t tx_count;
1179         uint16_t nb_segs = 0;
1180
1181         fp = &qdev->fp_array[QEDE_RSS_COUNT(qdev) + txq->queue_id];
1182
1183         if (unlikely(txq->nb_tx_avail < txq->tx_free_thresh)) {
1184                 PMD_TX_LOG(DEBUG, txq, "send=%u avail=%u free_thresh=%u\n",
1185                            nb_pkts, txq->nb_tx_avail, txq->tx_free_thresh);
1186                 (void)qede_process_tx_compl(edev, txq);
1187         }
1188
1189         nb_tx_pkts = RTE_MIN(nb_pkts, (txq->nb_tx_avail /
1190                         ETH_TX_MAX_BDS_PER_NON_LSO_PACKET));
1191         if (unlikely(nb_tx_pkts == 0)) {
1192                 PMD_TX_LOG(DEBUG, txq, "Out of BDs nb_pkts=%u avail=%u\n",
1193                            nb_pkts, txq->nb_tx_avail);
1194                 return 0;
1195         }
1196
1197         tx_count = nb_tx_pkts;
1198         while (nb_tx_pkts--) {
1199                 /* Fill the entry in the SW ring and the BDs in the FW ring */
1200                 idx = TX_PROD(txq);
1201                 struct rte_mbuf *mbuf = *tx_pkts++;
1202
1203                 txq->sw_tx_ring[idx].mbuf = mbuf;
1204                 bd1 = (struct eth_tx_1st_bd *)ecore_chain_produce(&txq->tx_pbl);
1205                 /* Zero init struct fields */
1206                 bd1->data.bd_flags.bitfields = 0;
1207                 bd1->data.bitfields = 0;
1208
1209                 bd1->data.bd_flags.bitfields =
1210                         1 << ETH_TX_1ST_BD_FLAGS_START_BD_SHIFT;
1211                 /* Map MBUF linear data for DMA and set in the first BD */
1212                 QEDE_BD_SET_ADDR_LEN(bd1, rte_mbuf_data_dma_addr(mbuf),
1213                                      mbuf->pkt_len);
1214
1215                 if (RTE_ETH_IS_TUNNEL_PKT(mbuf->packet_type)) {
1216                         PMD_TX_LOG(INFO, txq, "Tx tunnel packet\n");
1217                         /* First indicate its a tunnel pkt */
1218                         bd1->data.bd_flags.bitfields |=
1219                                 ETH_TX_DATA_1ST_BD_TUNN_FLAG_MASK <<
1220                                 ETH_TX_DATA_1ST_BD_TUNN_FLAG_SHIFT;
1221
1222                         /* Legacy FW had flipped behavior in regard to this bit
1223                          * i.e. it needed to set to prevent FW from touching
1224                          * encapsulated packets when it didn't need to.
1225                          */
1226                         if (unlikely(txq->is_legacy))
1227                                 bd1->data.bitfields ^=
1228                                         1 << ETH_TX_DATA_1ST_BD_TUNN_FLAG_SHIFT;
1229
1230                         /* Outer IP checksum offload */
1231                         if (mbuf->ol_flags & PKT_TX_OUTER_IP_CKSUM) {
1232                                 PMD_TX_LOG(INFO, txq, "OuterIP csum offload\n");
1233                                 bd1->data.bd_flags.bitfields |=
1234                                         ETH_TX_1ST_BD_FLAGS_TUNN_IP_CSUM_MASK <<
1235                                         ETH_TX_1ST_BD_FLAGS_TUNN_IP_CSUM_SHIFT;
1236                         }
1237
1238                         /* Outer UDP checksum offload */
1239                         bd1->data.bd_flags.bitfields |=
1240                                 ETH_TX_1ST_BD_FLAGS_TUNN_L4_CSUM_MASK <<
1241                                 ETH_TX_1ST_BD_FLAGS_TUNN_L4_CSUM_SHIFT;
1242                 }
1243
1244                 /* Descriptor based VLAN insertion */
1245                 if (mbuf->ol_flags & (PKT_TX_VLAN_PKT | PKT_TX_QINQ_PKT)) {
1246                         PMD_TX_LOG(INFO, txq, "Insert VLAN 0x%x\n",
1247                                    mbuf->vlan_tci);
1248                         bd1->data.vlan = rte_cpu_to_le_16(mbuf->vlan_tci);
1249                         bd1->data.bd_flags.bitfields |=
1250                             1 << ETH_TX_1ST_BD_FLAGS_VLAN_INSERTION_SHIFT;
1251                 }
1252
1253                 /* Offload the IP checksum in the hardware */
1254                 if (mbuf->ol_flags & PKT_TX_IP_CKSUM) {
1255                         PMD_TX_LOG(INFO, txq, "IP csum offload\n");
1256                         bd1->data.bd_flags.bitfields |=
1257                             1 << ETH_TX_1ST_BD_FLAGS_IP_CSUM_SHIFT;
1258                 }
1259
1260                 /* L4 checksum offload (tcp or udp) */
1261                 if (mbuf->ol_flags & (PKT_TX_TCP_CKSUM | PKT_TX_UDP_CKSUM)) {
1262                         PMD_TX_LOG(INFO, txq, "L4 csum offload\n");
1263                         bd1->data.bd_flags.bitfields |=
1264                             1 << ETH_TX_1ST_BD_FLAGS_L4_CSUM_SHIFT;
1265                         /* IPv6 + extn. -> later */
1266                 }
1267
1268                 /* Handle fragmented MBUF */
1269                 m_seg = mbuf->next;
1270                 nb_segs++;
1271                 bd1->data.nbds = nb_segs;
1272                 /* Encode scatter gather buffer descriptors if required */
1273                 nb_segs = qede_encode_sg_bd(txq, m_seg, nb_segs, bd1);
1274                 txq->nb_tx_avail = txq->nb_tx_avail - nb_segs;
1275                 nb_segs = 0;
1276                 txq->sw_tx_prod++;
1277                 rte_prefetch0(txq->sw_tx_ring[TX_PROD(txq)].mbuf);
1278                 bd_prod =
1279                     rte_cpu_to_le_16(ecore_chain_get_prod_idx(&txq->tx_pbl));
1280                 nb_pkt_sent++;
1281                 txq->xmit_pkts++;
1282         }
1283
1284         /* Write value of prod idx into bd_prod */
1285         txq->tx_db.data.bd_prod = bd_prod;
1286         rte_wmb();
1287         rte_compiler_barrier();
1288         DIRECT_REG_WR(edev, txq->doorbell_addr, txq->tx_db.raw);
1289         rte_wmb();
1290
1291         /* Check again for Tx completions */
1292         (void)qede_process_tx_compl(edev, txq);
1293
1294         PMD_TX_LOG(DEBUG, txq, "to_send=%u can_send=%u sent=%u core=%d\n",
1295                    nb_pkts, tx_count, nb_pkt_sent, rte_lcore_id());
1296
1297         return nb_pkt_sent;
1298 }
1299
1300 static void qede_init_fp_queue(struct rte_eth_dev *eth_dev)
1301 {
1302         struct qede_dev *qdev = eth_dev->data->dev_private;
1303         struct qede_fastpath *fp;
1304         uint8_t i, rss_id, txq_index, tc;
1305         int rxq = 0, txq = 0;
1306
1307         for_each_queue(i) {
1308                 fp = &qdev->fp_array[i];
1309                 if (fp->type & QEDE_FASTPATH_RX) {
1310                         fp->rxq = eth_dev->data->rx_queues[i];
1311                         fp->rxq->queue_id = rxq++;
1312                 }
1313
1314                 if (fp->type & QEDE_FASTPATH_TX) {
1315                         for (tc = 0; tc < qdev->num_tc; tc++) {
1316                                 txq_index = tc * QEDE_TSS_COUNT(qdev) + txq;
1317                                 fp->txqs[tc] =
1318                                         eth_dev->data->tx_queues[txq_index];
1319                                 fp->txqs[tc]->queue_id = txq_index;
1320                                 if (qdev->dev_info.is_legacy)
1321                                         fp->txqs[tc]->is_legacy = true;
1322                         }
1323                         txq++;
1324                 }
1325         }
1326 }
1327
1328 int qede_dev_start(struct rte_eth_dev *eth_dev)
1329 {
1330         struct qede_dev *qdev = eth_dev->data->dev_private;
1331         struct ecore_dev *edev = &qdev->edev;
1332         struct qed_link_output link_output;
1333         struct qede_fastpath *fp;
1334         int rc;
1335
1336         DP_INFO(edev, "Device state is %d\n", qdev->state);
1337
1338         if (qdev->state == QEDE_DEV_START) {
1339                 DP_INFO(edev, "Port is already started\n");
1340                 return 0;
1341         }
1342
1343         if (qdev->state == QEDE_DEV_CONFIG)
1344                 qede_init_fp_queue(eth_dev);
1345
1346         rc = qede_start_queues(eth_dev, true);
1347         if (rc) {
1348                 DP_ERR(edev, "Failed to start queues\n");
1349                 /* TBD: free */
1350                 return rc;
1351         }
1352
1353         /* Bring-up the link */
1354         qede_dev_set_link_state(eth_dev, true);
1355
1356         /* Reset ring */
1357         if (qede_reset_fp_rings(qdev))
1358                 return -ENOMEM;
1359
1360         /* Start/resume traffic */
1361         qdev->ops->fastpath_start(edev);
1362
1363         qdev->state = QEDE_DEV_START;
1364
1365         DP_INFO(edev, "dev_state is QEDE_DEV_START\n");
1366
1367         return 0;
1368 }
1369
1370 static int qede_drain_txq(struct qede_dev *qdev,
1371                           struct qede_tx_queue *txq, bool allow_drain)
1372 {
1373         struct ecore_dev *edev = &qdev->edev;
1374         int rc, cnt = 1000;
1375
1376         while (txq->sw_tx_cons != txq->sw_tx_prod) {
1377                 qede_process_tx_compl(edev, txq);
1378                 if (!cnt) {
1379                         if (allow_drain) {
1380                                 DP_NOTICE(edev, false,
1381                                           "Tx queue[%u] is stuck,"
1382                                           "requesting MCP to drain\n",
1383                                           txq->queue_id);
1384                                 rc = qdev->ops->common->drain(edev);
1385                                 if (rc)
1386                                         return rc;
1387                                 return qede_drain_txq(qdev, txq, false);
1388                         }
1389
1390                         DP_NOTICE(edev, false,
1391                                   "Timeout waiting for tx queue[%d]:"
1392                                   "PROD=%d, CONS=%d\n",
1393                                   txq->queue_id, txq->sw_tx_prod,
1394                                   txq->sw_tx_cons);
1395                         return -ENODEV;
1396                 }
1397                 cnt--;
1398                 DELAY(1000);
1399                 rte_compiler_barrier();
1400         }
1401
1402         /* FW finished processing, wait for HW to transmit all tx packets */
1403         DELAY(2000);
1404
1405         return 0;
1406 }
1407
1408 static int qede_stop_queues(struct qede_dev *qdev)
1409 {
1410         struct qed_update_vport_params vport_update_params;
1411         struct ecore_dev *edev = &qdev->edev;
1412         int rc, tc, i;
1413
1414         /* Disable the vport */
1415         memset(&vport_update_params, 0, sizeof(vport_update_params));
1416         vport_update_params.vport_id = 0;
1417         vport_update_params.update_vport_active_flg = 1;
1418         vport_update_params.vport_active_flg = 0;
1419         vport_update_params.update_rss_flg = 0;
1420
1421         DP_INFO(edev, "Deactivate vport\n");
1422
1423         rc = qdev->ops->vport_update(edev, &vport_update_params);
1424         if (rc) {
1425                 DP_ERR(edev, "Failed to update vport\n");
1426                 return rc;
1427         }
1428
1429         DP_INFO(edev, "Flushing tx queues\n");
1430
1431         /* Flush Tx queues. If needed, request drain from MCP */
1432         for_each_queue(i) {
1433                 struct qede_fastpath *fp = &qdev->fp_array[i];
1434
1435                 if (fp->type & QEDE_FASTPATH_TX) {
1436                         for (tc = 0; tc < qdev->num_tc; tc++) {
1437                                 struct qede_tx_queue *txq = fp->txqs[tc];
1438
1439                                 rc = qede_drain_txq(qdev, txq, true);
1440                                 if (rc)
1441                                         return rc;
1442                         }
1443                 }
1444         }
1445
1446         /* Stop all Queues in reverse order */
1447         for (i = QEDE_QUEUE_CNT(qdev) - 1; i >= 0; i--) {
1448                 struct qed_stop_rxq_params rx_params;
1449
1450                 /* Stop the Tx Queue(s) */
1451                 if (qdev->fp_array[i].type & QEDE_FASTPATH_TX) {
1452                         for (tc = 0; tc < qdev->num_tc; tc++) {
1453                                 struct qed_stop_txq_params tx_params;
1454                                 u8 val;
1455
1456                                 tx_params.rss_id = i;
1457                                 val = qdev->fp_array[i].txqs[tc]->queue_id;
1458                                 tx_params.tx_queue_id = val;
1459
1460                                 DP_INFO(edev, "Stopping tx queues\n");
1461                                 rc = qdev->ops->q_tx_stop(edev, &tx_params);
1462                                 if (rc) {
1463                                         DP_ERR(edev, "Failed to stop TXQ #%d\n",
1464                                                tx_params.tx_queue_id);
1465                                         return rc;
1466                                 }
1467                         }
1468                 }
1469
1470                 /* Stop the Rx Queue */
1471                 if (qdev->fp_array[i].type & QEDE_FASTPATH_RX) {
1472                         memset(&rx_params, 0, sizeof(rx_params));
1473                         rx_params.rss_id = i;
1474                         rx_params.rx_queue_id = qdev->fp_array[i].rxq->queue_id;
1475                         rx_params.eq_completion_only = 1;
1476
1477                         DP_INFO(edev, "Stopping rx queues\n");
1478
1479                         rc = qdev->ops->q_rx_stop(edev, &rx_params);
1480                         if (rc) {
1481                                 DP_ERR(edev, "Failed to stop RXQ #%d\n", i);
1482                                 return rc;
1483                         }
1484                 }
1485         }
1486
1487         return 0;
1488 }
1489
1490 int qede_reset_fp_rings(struct qede_dev *qdev)
1491 {
1492         struct qede_fastpath *fp;
1493         struct qede_tx_queue *txq;
1494         uint8_t tc;
1495         uint16_t id, i;
1496
1497         for_each_queue(id) {
1498                 fp = &qdev->fp_array[id];
1499
1500                 if (fp->type & QEDE_FASTPATH_RX) {
1501                         DP_INFO(&qdev->edev,
1502                                 "Reset FP chain for RSS %u\n", id);
1503                         qede_rx_queue_release_mbufs(fp->rxq);
1504                         ecore_chain_reset(&fp->rxq->rx_bd_ring);
1505                         ecore_chain_reset(&fp->rxq->rx_comp_ring);
1506                         fp->rxq->sw_rx_prod = 0;
1507                         fp->rxq->sw_rx_cons = 0;
1508                         *fp->rxq->hw_cons_ptr = 0;
1509                         for (i = 0; i < fp->rxq->nb_rx_desc; i++) {
1510                                 if (qede_alloc_rx_buffer(fp->rxq)) {
1511                                         DP_ERR(&qdev->edev,
1512                                                "RX buffer allocation failed\n");
1513                                         return -ENOMEM;
1514                                 }
1515                         }
1516                 }
1517                 if (fp->type & QEDE_FASTPATH_TX) {
1518                         for (tc = 0; tc < qdev->num_tc; tc++) {
1519                                 txq = fp->txqs[tc];
1520                                 qede_tx_queue_release_mbufs(txq);
1521                                 ecore_chain_reset(&txq->tx_pbl);
1522                                 txq->sw_tx_cons = 0;
1523                                 txq->sw_tx_prod = 0;
1524                                 *txq->hw_cons_ptr = 0;
1525                         }
1526                 }
1527         }
1528
1529         return 0;
1530 }
1531
1532 /* This function frees all memory of a single fp */
1533 void qede_free_mem_load(struct rte_eth_dev *eth_dev)
1534 {
1535         struct qede_dev *qdev = QEDE_INIT_QDEV(eth_dev);
1536         struct qede_fastpath *fp;
1537         uint16_t txq_idx;
1538         uint8_t id;
1539         uint8_t tc;
1540
1541         for_each_queue(id) {
1542                 fp = &qdev->fp_array[id];
1543                 if (fp->type & QEDE_FASTPATH_RX) {
1544                         if (!fp->rxq)
1545                                 continue;
1546                         qede_rx_queue_release(fp->rxq);
1547                         eth_dev->data->rx_queues[id] = NULL;
1548                 } else {
1549                         for (tc = 0; tc < qdev->num_tc; tc++) {
1550                                 if (!fp->txqs[tc])
1551                                         continue;
1552                                 txq_idx = fp->txqs[tc]->queue_id;
1553                                 qede_tx_queue_release(fp->txqs[tc]);
1554                                 eth_dev->data->tx_queues[txq_idx] = NULL;
1555                         }
1556                 }
1557         }
1558 }
1559
1560 void qede_dev_stop(struct rte_eth_dev *eth_dev)
1561 {
1562         struct qede_dev *qdev = eth_dev->data->dev_private;
1563         struct ecore_dev *edev = &qdev->edev;
1564
1565         DP_INFO(edev, "port %u\n", eth_dev->data->port_id);
1566
1567         if (qdev->state != QEDE_DEV_START) {
1568                 DP_INFO(edev, "Device not yet started\n");
1569                 return;
1570         }
1571
1572         if (qede_stop_queues(qdev))
1573                 DP_ERR(edev, "Didn't succeed to close queues\n");
1574
1575         DP_INFO(edev, "Stopped queues\n");
1576
1577         qdev->ops->fastpath_stop(edev);
1578
1579         /* Bring the link down */
1580         qede_dev_set_link_state(eth_dev, false);
1581
1582         qdev->state = QEDE_DEV_STOP;
1583
1584         DP_INFO(edev, "dev_state is QEDE_DEV_STOP\n");
1585 }