ethdev: fix max Rx packet length
[dpdk.git] / drivers / net / ice / ice_dcf_ethdev.c
1 /* SPDX-License-Identifier: BSD-3-Clause
2  * Copyright(c) 2020 Intel Corporation
3  */
4
5 #include <errno.h>
6 #include <stdbool.h>
7 #include <sys/queue.h>
8 #include <sys/types.h>
9 #include <unistd.h>
10
11 #include <rte_interrupts.h>
12 #include <rte_debug.h>
13 #include <rte_pci.h>
14 #include <rte_atomic.h>
15 #include <rte_eal.h>
16 #include <rte_ether.h>
17 #include <ethdev_pci.h>
18 #include <rte_kvargs.h>
19 #include <rte_malloc.h>
20 #include <rte_memzone.h>
21 #include <rte_dev.h>
22
23 #include <iavf_devids.h>
24
25 #include "ice_generic_flow.h"
26 #include "ice_dcf_ethdev.h"
27 #include "ice_rxtx.h"
28
29 static int
30 ice_dcf_dev_udp_tunnel_port_add(struct rte_eth_dev *dev,
31                                 struct rte_eth_udp_tunnel *udp_tunnel);
32 static int
33 ice_dcf_dev_udp_tunnel_port_del(struct rte_eth_dev *dev,
34                                 struct rte_eth_udp_tunnel *udp_tunnel);
35
36 static int
37 ice_dcf_dev_init(struct rte_eth_dev *eth_dev);
38
39 static int
40 ice_dcf_dev_uninit(struct rte_eth_dev *eth_dev);
41
42 static uint16_t
43 ice_dcf_recv_pkts(__rte_unused void *rx_queue,
44                   __rte_unused struct rte_mbuf **bufs,
45                   __rte_unused uint16_t nb_pkts)
46 {
47         return 0;
48 }
49
50 static uint16_t
51 ice_dcf_xmit_pkts(__rte_unused void *tx_queue,
52                   __rte_unused struct rte_mbuf **bufs,
53                   __rte_unused uint16_t nb_pkts)
54 {
55         return 0;
56 }
57
58 static int
59 ice_dcf_init_rxq(struct rte_eth_dev *dev, struct ice_rx_queue *rxq)
60 {
61         struct ice_dcf_adapter *dcf_ad = dev->data->dev_private;
62         struct rte_eth_dev_data *dev_data = dev->data;
63         struct iavf_hw *hw = &dcf_ad->real_hw.avf;
64         uint16_t buf_size, max_pkt_len;
65
66         buf_size = rte_pktmbuf_data_room_size(rxq->mp) - RTE_PKTMBUF_HEADROOM;
67         rxq->rx_hdr_len = 0;
68         rxq->rx_buf_len = RTE_ALIGN(buf_size, (1 << ICE_RLAN_CTX_DBUF_S));
69         max_pkt_len = RTE_MIN(ICE_SUPPORT_CHAIN_NUM * rxq->rx_buf_len,
70                               dev->data->mtu + ICE_ETH_OVERHEAD);
71
72         /* Check if the jumbo frame and maximum packet length are set
73          * correctly.
74          */
75         if (dev->data->dev_conf.rxmode.offloads & DEV_RX_OFFLOAD_JUMBO_FRAME) {
76                 if (max_pkt_len <= ICE_ETH_MAX_LEN ||
77                     max_pkt_len > ICE_FRAME_SIZE_MAX) {
78                         PMD_DRV_LOG(ERR, "maximum packet length must be "
79                                     "larger than %u and smaller than %u, "
80                                     "as jumbo frame is enabled",
81                                     (uint32_t)ICE_ETH_MAX_LEN,
82                                     (uint32_t)ICE_FRAME_SIZE_MAX);
83                         return -EINVAL;
84                 }
85         } else {
86                 if (max_pkt_len < RTE_ETHER_MIN_LEN ||
87                     max_pkt_len > ICE_ETH_MAX_LEN) {
88                         PMD_DRV_LOG(ERR, "maximum packet length must be "
89                                     "larger than %u and smaller than %u, "
90                                     "as jumbo frame is disabled",
91                                     (uint32_t)RTE_ETHER_MIN_LEN,
92                                     (uint32_t)ICE_ETH_MAX_LEN);
93                         return -EINVAL;
94                 }
95         }
96
97         rxq->max_pkt_len = max_pkt_len;
98         if ((dev_data->dev_conf.rxmode.offloads & DEV_RX_OFFLOAD_SCATTER) ||
99             (rxq->max_pkt_len + 2 * ICE_VLAN_TAG_SIZE) > buf_size) {
100                 dev_data->scattered_rx = 1;
101         }
102         rxq->qrx_tail = hw->hw_addr + IAVF_QRX_TAIL1(rxq->queue_id);
103         IAVF_PCI_REG_WRITE(rxq->qrx_tail, rxq->nb_rx_desc - 1);
104         IAVF_WRITE_FLUSH(hw);
105
106         return 0;
107 }
108
109 static int
110 ice_dcf_init_rx_queues(struct rte_eth_dev *dev)
111 {
112         struct ice_rx_queue **rxq =
113                 (struct ice_rx_queue **)dev->data->rx_queues;
114         int i, ret;
115
116         for (i = 0; i < dev->data->nb_rx_queues; i++) {
117                 if (!rxq[i] || !rxq[i]->q_set)
118                         continue;
119                 ret = ice_dcf_init_rxq(dev, rxq[i]);
120                 if (ret)
121                         return ret;
122         }
123
124         ice_set_rx_function(dev);
125         ice_set_tx_function(dev);
126
127         return 0;
128 }
129
130 #define IAVF_MISC_VEC_ID                RTE_INTR_VEC_ZERO_OFFSET
131 #define IAVF_RX_VEC_START               RTE_INTR_VEC_RXTX_OFFSET
132
133 #define IAVF_ITR_INDEX_DEFAULT          0
134 #define IAVF_QUEUE_ITR_INTERVAL_DEFAULT 32 /* 32 us */
135 #define IAVF_QUEUE_ITR_INTERVAL_MAX     8160 /* 8160 us */
136
137 static inline uint16_t
138 iavf_calc_itr_interval(int16_t interval)
139 {
140         if (interval < 0 || interval > IAVF_QUEUE_ITR_INTERVAL_MAX)
141                 interval = IAVF_QUEUE_ITR_INTERVAL_DEFAULT;
142
143         /* Convert to hardware count, as writing each 1 represents 2 us */
144         return interval / 2;
145 }
146
147 static int
148 ice_dcf_config_rx_queues_irqs(struct rte_eth_dev *dev,
149                                      struct rte_intr_handle *intr_handle)
150 {
151         struct ice_dcf_adapter *adapter = dev->data->dev_private;
152         struct ice_dcf_hw *hw = &adapter->real_hw;
153         uint16_t interval, i;
154         int vec;
155
156         if (rte_intr_cap_multiple(intr_handle) &&
157             dev->data->dev_conf.intr_conf.rxq) {
158                 if (rte_intr_efd_enable(intr_handle, dev->data->nb_rx_queues))
159                         return -1;
160         }
161
162         if (rte_intr_dp_is_en(intr_handle) && !intr_handle->intr_vec) {
163                 intr_handle->intr_vec =
164                         rte_zmalloc("intr_vec",
165                                     dev->data->nb_rx_queues * sizeof(int), 0);
166                 if (!intr_handle->intr_vec) {
167                         PMD_DRV_LOG(ERR, "Failed to allocate %d rx intr_vec",
168                                     dev->data->nb_rx_queues);
169                         return -1;
170                 }
171         }
172
173         if (!dev->data->dev_conf.intr_conf.rxq ||
174             !rte_intr_dp_is_en(intr_handle)) {
175                 /* Rx interrupt disabled, Map interrupt only for writeback */
176                 hw->nb_msix = 1;
177                 if (hw->vf_res->vf_cap_flags &
178                     VIRTCHNL_VF_OFFLOAD_WB_ON_ITR) {
179                         /* If WB_ON_ITR supports, enable it */
180                         hw->msix_base = IAVF_RX_VEC_START;
181                         /* Set the ITR for index zero, to 2us to make sure that
182                          * we leave time for aggregation to occur, but don't
183                          * increase latency dramatically.
184                          */
185                         IAVF_WRITE_REG(&hw->avf,
186                                        IAVF_VFINT_DYN_CTLN1(hw->msix_base - 1),
187                                        (0 << IAVF_VFINT_DYN_CTLN1_ITR_INDX_SHIFT) |
188                                        IAVF_VFINT_DYN_CTLN1_WB_ON_ITR_MASK |
189                                        (2UL << IAVF_VFINT_DYN_CTLN1_INTERVAL_SHIFT));
190                 } else {
191                         /* If no WB_ON_ITR offload flags, need to set
192                          * interrupt for descriptor write back.
193                          */
194                         hw->msix_base = IAVF_MISC_VEC_ID;
195
196                         /* set ITR to max */
197                         interval =
198                         iavf_calc_itr_interval(IAVF_QUEUE_ITR_INTERVAL_MAX);
199                         IAVF_WRITE_REG(&hw->avf, IAVF_VFINT_DYN_CTL01,
200                                        IAVF_VFINT_DYN_CTL01_INTENA_MASK |
201                                        (IAVF_ITR_INDEX_DEFAULT <<
202                                         IAVF_VFINT_DYN_CTL01_ITR_INDX_SHIFT) |
203                                        (interval <<
204                                         IAVF_VFINT_DYN_CTL01_INTERVAL_SHIFT));
205                 }
206                 IAVF_WRITE_FLUSH(&hw->avf);
207                 /* map all queues to the same interrupt */
208                 for (i = 0; i < dev->data->nb_rx_queues; i++)
209                         hw->rxq_map[hw->msix_base] |= 1 << i;
210         } else {
211                 if (!rte_intr_allow_others(intr_handle)) {
212                         hw->nb_msix = 1;
213                         hw->msix_base = IAVF_MISC_VEC_ID;
214                         for (i = 0; i < dev->data->nb_rx_queues; i++) {
215                                 hw->rxq_map[hw->msix_base] |= 1 << i;
216                                 intr_handle->intr_vec[i] = IAVF_MISC_VEC_ID;
217                         }
218                         PMD_DRV_LOG(DEBUG,
219                                     "vector %u are mapping to all Rx queues",
220                                     hw->msix_base);
221                 } else {
222                         /* If Rx interrupt is reuquired, and we can use
223                          * multi interrupts, then the vec is from 1
224                          */
225                         hw->nb_msix = RTE_MIN(hw->vf_res->max_vectors,
226                                               intr_handle->nb_efd);
227                         hw->msix_base = IAVF_MISC_VEC_ID;
228                         vec = IAVF_MISC_VEC_ID;
229                         for (i = 0; i < dev->data->nb_rx_queues; i++) {
230                                 hw->rxq_map[vec] |= 1 << i;
231                                 intr_handle->intr_vec[i] = vec++;
232                                 if (vec >= hw->nb_msix)
233                                         vec = IAVF_RX_VEC_START;
234                         }
235                         PMD_DRV_LOG(DEBUG,
236                                     "%u vectors are mapping to %u Rx queues",
237                                     hw->nb_msix, dev->data->nb_rx_queues);
238                 }
239         }
240
241         if (ice_dcf_config_irq_map(hw)) {
242                 PMD_DRV_LOG(ERR, "config interrupt mapping failed");
243                 return -1;
244         }
245         return 0;
246 }
247
248 static int
249 alloc_rxq_mbufs(struct ice_rx_queue *rxq)
250 {
251         volatile union ice_rx_flex_desc *rxd;
252         struct rte_mbuf *mbuf = NULL;
253         uint64_t dma_addr;
254         uint16_t i;
255
256         for (i = 0; i < rxq->nb_rx_desc; i++) {
257                 mbuf = rte_mbuf_raw_alloc(rxq->mp);
258                 if (unlikely(!mbuf)) {
259                         PMD_DRV_LOG(ERR, "Failed to allocate mbuf for RX");
260                         return -ENOMEM;
261                 }
262
263                 rte_mbuf_refcnt_set(mbuf, 1);
264                 mbuf->next = NULL;
265                 mbuf->data_off = RTE_PKTMBUF_HEADROOM;
266                 mbuf->nb_segs = 1;
267                 mbuf->port = rxq->port_id;
268
269                 dma_addr =
270                         rte_cpu_to_le_64(rte_mbuf_data_iova_default(mbuf));
271
272                 rxd = &rxq->rx_ring[i];
273                 rxd->read.pkt_addr = dma_addr;
274                 rxd->read.hdr_addr = 0;
275 #ifndef RTE_LIBRTE_ICE_16BYTE_RX_DESC
276                 rxd->read.rsvd1 = 0;
277                 rxd->read.rsvd2 = 0;
278 #endif
279
280                 rxq->sw_ring[i].mbuf = (void *)mbuf;
281         }
282
283         return 0;
284 }
285
286 static int
287 ice_dcf_rx_queue_start(struct rte_eth_dev *dev, uint16_t rx_queue_id)
288 {
289         struct ice_dcf_adapter *ad = dev->data->dev_private;
290         struct iavf_hw *hw = &ad->real_hw.avf;
291         struct ice_rx_queue *rxq;
292         int err = 0;
293
294         if (rx_queue_id >= dev->data->nb_rx_queues)
295                 return -EINVAL;
296
297         rxq = dev->data->rx_queues[rx_queue_id];
298
299         err = alloc_rxq_mbufs(rxq);
300         if (err) {
301                 PMD_DRV_LOG(ERR, "Failed to allocate RX queue mbuf");
302                 return err;
303         }
304
305         rte_wmb();
306
307         /* Init the RX tail register. */
308         IAVF_PCI_REG_WRITE(rxq->qrx_tail, rxq->nb_rx_desc - 1);
309         IAVF_WRITE_FLUSH(hw);
310
311         /* Ready to switch the queue on */
312         err = ice_dcf_switch_queue(&ad->real_hw, rx_queue_id, true, true);
313         if (err) {
314                 PMD_DRV_LOG(ERR, "Failed to switch RX queue %u on",
315                             rx_queue_id);
316                 return err;
317         }
318
319         dev->data->rx_queue_state[rx_queue_id] = RTE_ETH_QUEUE_STATE_STARTED;
320
321         return 0;
322 }
323
324 static inline void
325 reset_rx_queue(struct ice_rx_queue *rxq)
326 {
327         uint16_t len;
328         uint32_t i;
329
330         if (!rxq)
331                 return;
332
333         len = rxq->nb_rx_desc + ICE_RX_MAX_BURST;
334
335         for (i = 0; i < len * sizeof(union ice_rx_flex_desc); i++)
336                 ((volatile char *)rxq->rx_ring)[i] = 0;
337
338         memset(&rxq->fake_mbuf, 0x0, sizeof(rxq->fake_mbuf));
339
340         for (i = 0; i < ICE_RX_MAX_BURST; i++)
341                 rxq->sw_ring[rxq->nb_rx_desc + i].mbuf = &rxq->fake_mbuf;
342
343         /* for rx bulk */
344         rxq->rx_nb_avail = 0;
345         rxq->rx_next_avail = 0;
346         rxq->rx_free_trigger = (uint16_t)(rxq->rx_free_thresh - 1);
347
348         rxq->rx_tail = 0;
349         rxq->nb_rx_hold = 0;
350         rxq->pkt_first_seg = NULL;
351         rxq->pkt_last_seg = NULL;
352 }
353
354 static inline void
355 reset_tx_queue(struct ice_tx_queue *txq)
356 {
357         struct ice_tx_entry *txe;
358         uint32_t i, size;
359         uint16_t prev;
360
361         if (!txq) {
362                 PMD_DRV_LOG(DEBUG, "Pointer to txq is NULL");
363                 return;
364         }
365
366         txe = txq->sw_ring;
367         size = sizeof(struct ice_tx_desc) * txq->nb_tx_desc;
368         for (i = 0; i < size; i++)
369                 ((volatile char *)txq->tx_ring)[i] = 0;
370
371         prev = (uint16_t)(txq->nb_tx_desc - 1);
372         for (i = 0; i < txq->nb_tx_desc; i++) {
373                 txq->tx_ring[i].cmd_type_offset_bsz =
374                         rte_cpu_to_le_64(IAVF_TX_DESC_DTYPE_DESC_DONE);
375                 txe[i].mbuf =  NULL;
376                 txe[i].last_id = i;
377                 txe[prev].next_id = i;
378                 prev = i;
379         }
380
381         txq->tx_tail = 0;
382         txq->nb_tx_used = 0;
383
384         txq->last_desc_cleaned = txq->nb_tx_desc - 1;
385         txq->nb_tx_free = txq->nb_tx_desc - 1;
386
387         txq->tx_next_dd = txq->tx_rs_thresh - 1;
388         txq->tx_next_rs = txq->tx_rs_thresh - 1;
389 }
390
391 static int
392 ice_dcf_rx_queue_stop(struct rte_eth_dev *dev, uint16_t rx_queue_id)
393 {
394         struct ice_dcf_adapter *ad = dev->data->dev_private;
395         struct ice_dcf_hw *hw = &ad->real_hw;
396         struct ice_rx_queue *rxq;
397         int err;
398
399         if (rx_queue_id >= dev->data->nb_rx_queues)
400                 return -EINVAL;
401
402         err = ice_dcf_switch_queue(hw, rx_queue_id, true, false);
403         if (err) {
404                 PMD_DRV_LOG(ERR, "Failed to switch RX queue %u off",
405                             rx_queue_id);
406                 return err;
407         }
408
409         rxq = dev->data->rx_queues[rx_queue_id];
410         rxq->rx_rel_mbufs(rxq);
411         reset_rx_queue(rxq);
412         dev->data->rx_queue_state[rx_queue_id] = RTE_ETH_QUEUE_STATE_STOPPED;
413
414         return 0;
415 }
416
417 static int
418 ice_dcf_tx_queue_start(struct rte_eth_dev *dev, uint16_t tx_queue_id)
419 {
420         struct ice_dcf_adapter *ad = dev->data->dev_private;
421         struct iavf_hw *hw = &ad->real_hw.avf;
422         struct ice_tx_queue *txq;
423         int err = 0;
424
425         if (tx_queue_id >= dev->data->nb_tx_queues)
426                 return -EINVAL;
427
428         txq = dev->data->tx_queues[tx_queue_id];
429
430         /* Init the RX tail register. */
431         txq->qtx_tail = hw->hw_addr + IAVF_QTX_TAIL1(tx_queue_id);
432         IAVF_PCI_REG_WRITE(txq->qtx_tail, 0);
433         IAVF_WRITE_FLUSH(hw);
434
435         /* Ready to switch the queue on */
436         err = ice_dcf_switch_queue(&ad->real_hw, tx_queue_id, false, true);
437
438         if (err) {
439                 PMD_DRV_LOG(ERR, "Failed to switch TX queue %u on",
440                             tx_queue_id);
441                 return err;
442         }
443
444         dev->data->tx_queue_state[tx_queue_id] = RTE_ETH_QUEUE_STATE_STARTED;
445
446         return 0;
447 }
448
449 static int
450 ice_dcf_tx_queue_stop(struct rte_eth_dev *dev, uint16_t tx_queue_id)
451 {
452         struct ice_dcf_adapter *ad = dev->data->dev_private;
453         struct ice_dcf_hw *hw = &ad->real_hw;
454         struct ice_tx_queue *txq;
455         int err;
456
457         if (tx_queue_id >= dev->data->nb_tx_queues)
458                 return -EINVAL;
459
460         err = ice_dcf_switch_queue(hw, tx_queue_id, false, false);
461         if (err) {
462                 PMD_DRV_LOG(ERR, "Failed to switch TX queue %u off",
463                             tx_queue_id);
464                 return err;
465         }
466
467         txq = dev->data->tx_queues[tx_queue_id];
468         txq->tx_rel_mbufs(txq);
469         reset_tx_queue(txq);
470         dev->data->tx_queue_state[tx_queue_id] = RTE_ETH_QUEUE_STATE_STOPPED;
471
472         return 0;
473 }
474
475 static int
476 ice_dcf_start_queues(struct rte_eth_dev *dev)
477 {
478         struct ice_rx_queue *rxq;
479         struct ice_tx_queue *txq;
480         int nb_rxq = 0;
481         int nb_txq, i;
482
483         for (nb_txq = 0; nb_txq < dev->data->nb_tx_queues; nb_txq++) {
484                 txq = dev->data->tx_queues[nb_txq];
485                 if (txq->tx_deferred_start)
486                         continue;
487                 if (ice_dcf_tx_queue_start(dev, nb_txq) != 0) {
488                         PMD_DRV_LOG(ERR, "Fail to start queue %u", nb_txq);
489                         goto tx_err;
490                 }
491         }
492
493         for (nb_rxq = 0; nb_rxq < dev->data->nb_rx_queues; nb_rxq++) {
494                 rxq = dev->data->rx_queues[nb_rxq];
495                 if (rxq->rx_deferred_start)
496                         continue;
497                 if (ice_dcf_rx_queue_start(dev, nb_rxq) != 0) {
498                         PMD_DRV_LOG(ERR, "Fail to start queue %u", nb_rxq);
499                         goto rx_err;
500                 }
501         }
502
503         return 0;
504
505         /* stop the started queues if failed to start all queues */
506 rx_err:
507         for (i = 0; i < nb_rxq; i++)
508                 ice_dcf_rx_queue_stop(dev, i);
509 tx_err:
510         for (i = 0; i < nb_txq; i++)
511                 ice_dcf_tx_queue_stop(dev, i);
512
513         return -1;
514 }
515
516 static int
517 ice_dcf_dev_start(struct rte_eth_dev *dev)
518 {
519         struct ice_dcf_adapter *dcf_ad = dev->data->dev_private;
520         struct rte_intr_handle *intr_handle = dev->intr_handle;
521         struct ice_adapter *ad = &dcf_ad->parent;
522         struct ice_dcf_hw *hw = &dcf_ad->real_hw;
523         int ret;
524
525         if (hw->resetting) {
526                 PMD_DRV_LOG(ERR,
527                             "The DCF has been reset by PF, please reinit first");
528                 return -EIO;
529         }
530
531         ad->pf.adapter_stopped = 0;
532
533         hw->num_queue_pairs = RTE_MAX(dev->data->nb_rx_queues,
534                                       dev->data->nb_tx_queues);
535
536         ret = ice_dcf_init_rx_queues(dev);
537         if (ret) {
538                 PMD_DRV_LOG(ERR, "Fail to init queues");
539                 return ret;
540         }
541
542         if (hw->vf_res->vf_cap_flags & VIRTCHNL_VF_OFFLOAD_RSS_PF) {
543                 ret = ice_dcf_init_rss(hw);
544                 if (ret) {
545                         PMD_DRV_LOG(ERR, "Failed to configure RSS");
546                         return ret;
547                 }
548         }
549
550         ret = ice_dcf_configure_queues(hw);
551         if (ret) {
552                 PMD_DRV_LOG(ERR, "Fail to config queues");
553                 return ret;
554         }
555
556         ret = ice_dcf_config_rx_queues_irqs(dev, intr_handle);
557         if (ret) {
558                 PMD_DRV_LOG(ERR, "Fail to config rx queues' irqs");
559                 return ret;
560         }
561
562         if (dev->data->dev_conf.intr_conf.rxq != 0) {
563                 rte_intr_disable(intr_handle);
564                 rte_intr_enable(intr_handle);
565         }
566
567         ret = ice_dcf_start_queues(dev);
568         if (ret) {
569                 PMD_DRV_LOG(ERR, "Failed to enable queues");
570                 return ret;
571         }
572
573         ret = ice_dcf_add_del_all_mac_addr(hw, true);
574         if (ret) {
575                 PMD_DRV_LOG(ERR, "Failed to add mac addr");
576                 return ret;
577         }
578
579         dev->data->dev_link.link_status = ETH_LINK_UP;
580
581         return 0;
582 }
583
584 static void
585 ice_dcf_stop_queues(struct rte_eth_dev *dev)
586 {
587         struct ice_dcf_adapter *ad = dev->data->dev_private;
588         struct ice_dcf_hw *hw = &ad->real_hw;
589         struct ice_rx_queue *rxq;
590         struct ice_tx_queue *txq;
591         int ret, i;
592
593         /* Stop All queues */
594         ret = ice_dcf_disable_queues(hw);
595         if (ret)
596                 PMD_DRV_LOG(WARNING, "Fail to stop queues");
597
598         for (i = 0; i < dev->data->nb_tx_queues; i++) {
599                 txq = dev->data->tx_queues[i];
600                 if (!txq)
601                         continue;
602                 txq->tx_rel_mbufs(txq);
603                 reset_tx_queue(txq);
604                 dev->data->tx_queue_state[i] = RTE_ETH_QUEUE_STATE_STOPPED;
605         }
606         for (i = 0; i < dev->data->nb_rx_queues; i++) {
607                 rxq = dev->data->rx_queues[i];
608                 if (!rxq)
609                         continue;
610                 rxq->rx_rel_mbufs(rxq);
611                 reset_rx_queue(rxq);
612                 dev->data->rx_queue_state[i] = RTE_ETH_QUEUE_STATE_STOPPED;
613         }
614 }
615
616 static int
617 ice_dcf_dev_stop(struct rte_eth_dev *dev)
618 {
619         struct ice_dcf_adapter *dcf_ad = dev->data->dev_private;
620         struct rte_intr_handle *intr_handle = dev->intr_handle;
621         struct ice_adapter *ad = &dcf_ad->parent;
622
623         if (ad->pf.adapter_stopped == 1) {
624                 PMD_DRV_LOG(DEBUG, "Port is already stopped");
625                 return 0;
626         }
627
628         /* Stop the VF representors for this device */
629         ice_dcf_vf_repr_stop_all(dcf_ad);
630
631         ice_dcf_stop_queues(dev);
632
633         rte_intr_efd_disable(intr_handle);
634         if (intr_handle->intr_vec) {
635                 rte_free(intr_handle->intr_vec);
636                 intr_handle->intr_vec = NULL;
637         }
638
639         ice_dcf_add_del_all_mac_addr(&dcf_ad->real_hw, false);
640         dev->data->dev_link.link_status = ETH_LINK_DOWN;
641         ad->pf.adapter_stopped = 1;
642
643         return 0;
644 }
645
646 static int
647 ice_dcf_dev_configure(struct rte_eth_dev *dev)
648 {
649         struct ice_dcf_adapter *dcf_ad = dev->data->dev_private;
650         struct ice_adapter *ad = &dcf_ad->parent;
651
652         ad->rx_bulk_alloc_allowed = true;
653         ad->tx_simple_allowed = true;
654
655         if (dev->data->dev_conf.rxmode.mq_mode & ETH_MQ_RX_RSS_FLAG)
656                 dev->data->dev_conf.rxmode.offloads |= DEV_RX_OFFLOAD_RSS_HASH;
657
658         return 0;
659 }
660
661 static int
662 ice_dcf_dev_info_get(struct rte_eth_dev *dev,
663                      struct rte_eth_dev_info *dev_info)
664 {
665         struct ice_dcf_adapter *adapter = dev->data->dev_private;
666         struct ice_dcf_hw *hw = &adapter->real_hw;
667
668         dev_info->max_mac_addrs = 1;
669         dev_info->max_rx_queues = hw->vsi_res->num_queue_pairs;
670         dev_info->max_tx_queues = hw->vsi_res->num_queue_pairs;
671         dev_info->min_rx_bufsize = ICE_BUF_SIZE_MIN;
672         dev_info->max_rx_pktlen = ICE_FRAME_SIZE_MAX;
673         dev_info->hash_key_size = hw->vf_res->rss_key_size;
674         dev_info->reta_size = hw->vf_res->rss_lut_size;
675         dev_info->flow_type_rss_offloads = ICE_RSS_OFFLOAD_ALL;
676
677         dev_info->rx_offload_capa =
678                 DEV_RX_OFFLOAD_VLAN_STRIP |
679                 DEV_RX_OFFLOAD_IPV4_CKSUM |
680                 DEV_RX_OFFLOAD_UDP_CKSUM |
681                 DEV_RX_OFFLOAD_TCP_CKSUM |
682                 DEV_RX_OFFLOAD_OUTER_IPV4_CKSUM |
683                 DEV_RX_OFFLOAD_SCATTER |
684                 DEV_RX_OFFLOAD_JUMBO_FRAME |
685                 DEV_RX_OFFLOAD_VLAN_FILTER |
686                 DEV_RX_OFFLOAD_RSS_HASH;
687         dev_info->tx_offload_capa =
688                 DEV_TX_OFFLOAD_VLAN_INSERT |
689                 DEV_TX_OFFLOAD_IPV4_CKSUM |
690                 DEV_TX_OFFLOAD_UDP_CKSUM |
691                 DEV_TX_OFFLOAD_TCP_CKSUM |
692                 DEV_TX_OFFLOAD_SCTP_CKSUM |
693                 DEV_TX_OFFLOAD_OUTER_IPV4_CKSUM |
694                 DEV_TX_OFFLOAD_TCP_TSO |
695                 DEV_TX_OFFLOAD_VXLAN_TNL_TSO |
696                 DEV_TX_OFFLOAD_GRE_TNL_TSO |
697                 DEV_TX_OFFLOAD_IPIP_TNL_TSO |
698                 DEV_TX_OFFLOAD_GENEVE_TNL_TSO |
699                 DEV_TX_OFFLOAD_MULTI_SEGS;
700
701         dev_info->default_rxconf = (struct rte_eth_rxconf) {
702                 .rx_thresh = {
703                         .pthresh = ICE_DEFAULT_RX_PTHRESH,
704                         .hthresh = ICE_DEFAULT_RX_HTHRESH,
705                         .wthresh = ICE_DEFAULT_RX_WTHRESH,
706                 },
707                 .rx_free_thresh = ICE_DEFAULT_RX_FREE_THRESH,
708                 .rx_drop_en = 0,
709                 .offloads = 0,
710         };
711
712         dev_info->default_txconf = (struct rte_eth_txconf) {
713                 .tx_thresh = {
714                         .pthresh = ICE_DEFAULT_TX_PTHRESH,
715                         .hthresh = ICE_DEFAULT_TX_HTHRESH,
716                         .wthresh = ICE_DEFAULT_TX_WTHRESH,
717                 },
718                 .tx_free_thresh = ICE_DEFAULT_TX_FREE_THRESH,
719                 .tx_rs_thresh = ICE_DEFAULT_TX_RSBIT_THRESH,
720                 .offloads = 0,
721         };
722
723         dev_info->rx_desc_lim = (struct rte_eth_desc_lim) {
724                 .nb_max = ICE_MAX_RING_DESC,
725                 .nb_min = ICE_MIN_RING_DESC,
726                 .nb_align = ICE_ALIGN_RING_DESC,
727         };
728
729         dev_info->tx_desc_lim = (struct rte_eth_desc_lim) {
730                 .nb_max = ICE_MAX_RING_DESC,
731                 .nb_min = ICE_MIN_RING_DESC,
732                 .nb_align = ICE_ALIGN_RING_DESC,
733         };
734
735         return 0;
736 }
737
738 static int
739 ice_dcf_dev_promiscuous_enable(__rte_unused struct rte_eth_dev *dev)
740 {
741         return 0;
742 }
743
744 static int
745 ice_dcf_dev_promiscuous_disable(__rte_unused struct rte_eth_dev *dev)
746 {
747         return 0;
748 }
749
750 static int
751 ice_dcf_dev_allmulticast_enable(__rte_unused struct rte_eth_dev *dev)
752 {
753         return 0;
754 }
755
756 static int
757 ice_dcf_dev_allmulticast_disable(__rte_unused struct rte_eth_dev *dev)
758 {
759         return 0;
760 }
761
762 static int
763 ice_dcf_dev_flow_ops_get(struct rte_eth_dev *dev,
764                          const struct rte_flow_ops **ops)
765 {
766         if (!dev)
767                 return -EINVAL;
768
769         *ops = &ice_flow_ops;
770         return 0;
771 }
772
773 #define ICE_DCF_32_BIT_WIDTH (CHAR_BIT * 4)
774 #define ICE_DCF_48_BIT_WIDTH (CHAR_BIT * 6)
775 #define ICE_DCF_48_BIT_MASK  RTE_LEN2MASK(ICE_DCF_48_BIT_WIDTH, uint64_t)
776
777 static void
778 ice_dcf_stat_update_48(uint64_t *offset, uint64_t *stat)
779 {
780         if (*stat >= *offset)
781                 *stat = *stat - *offset;
782         else
783                 *stat = (uint64_t)((*stat +
784                         ((uint64_t)1 << ICE_DCF_48_BIT_WIDTH)) - *offset);
785
786         *stat &= ICE_DCF_48_BIT_MASK;
787 }
788
789 static void
790 ice_dcf_stat_update_32(uint64_t *offset, uint64_t *stat)
791 {
792         if (*stat >= *offset)
793                 *stat = (uint64_t)(*stat - *offset);
794         else
795                 *stat = (uint64_t)((*stat +
796                         ((uint64_t)1 << ICE_DCF_32_BIT_WIDTH)) - *offset);
797 }
798
799 static void
800 ice_dcf_update_stats(struct virtchnl_eth_stats *oes,
801                      struct virtchnl_eth_stats *nes)
802 {
803         ice_dcf_stat_update_48(&oes->rx_bytes, &nes->rx_bytes);
804         ice_dcf_stat_update_48(&oes->rx_unicast, &nes->rx_unicast);
805         ice_dcf_stat_update_48(&oes->rx_multicast, &nes->rx_multicast);
806         ice_dcf_stat_update_48(&oes->rx_broadcast, &nes->rx_broadcast);
807         ice_dcf_stat_update_32(&oes->rx_discards, &nes->rx_discards);
808         ice_dcf_stat_update_48(&oes->tx_bytes, &nes->tx_bytes);
809         ice_dcf_stat_update_48(&oes->tx_unicast, &nes->tx_unicast);
810         ice_dcf_stat_update_48(&oes->tx_multicast, &nes->tx_multicast);
811         ice_dcf_stat_update_48(&oes->tx_broadcast, &nes->tx_broadcast);
812         ice_dcf_stat_update_32(&oes->tx_errors, &nes->tx_errors);
813         ice_dcf_stat_update_32(&oes->tx_discards, &nes->tx_discards);
814 }
815
816
817 static int
818 ice_dcf_stats_get(struct rte_eth_dev *dev, struct rte_eth_stats *stats)
819 {
820         struct ice_dcf_adapter *ad = dev->data->dev_private;
821         struct ice_dcf_hw *hw = &ad->real_hw;
822         struct virtchnl_eth_stats pstats;
823         int ret;
824
825         if (hw->resetting) {
826                 PMD_DRV_LOG(ERR,
827                             "The DCF has been reset by PF, please reinit first");
828                 return -EIO;
829         }
830
831         ret = ice_dcf_query_stats(hw, &pstats);
832         if (ret == 0) {
833                 ice_dcf_update_stats(&hw->eth_stats_offset, &pstats);
834                 stats->ipackets = pstats.rx_unicast + pstats.rx_multicast +
835                                 pstats.rx_broadcast - pstats.rx_discards;
836                 stats->opackets = pstats.tx_broadcast + pstats.tx_multicast +
837                                                 pstats.tx_unicast;
838                 stats->imissed = pstats.rx_discards;
839                 stats->oerrors = pstats.tx_errors + pstats.tx_discards;
840                 stats->ibytes = pstats.rx_bytes;
841                 stats->ibytes -= stats->ipackets * RTE_ETHER_CRC_LEN;
842                 stats->obytes = pstats.tx_bytes;
843         } else {
844                 PMD_DRV_LOG(ERR, "Get statistics failed");
845         }
846         return ret;
847 }
848
849 static int
850 ice_dcf_stats_reset(struct rte_eth_dev *dev)
851 {
852         struct ice_dcf_adapter *ad = dev->data->dev_private;
853         struct ice_dcf_hw *hw = &ad->real_hw;
854         struct virtchnl_eth_stats pstats;
855         int ret;
856
857         if (hw->resetting)
858                 return 0;
859
860         /* read stat values to clear hardware registers */
861         ret = ice_dcf_query_stats(hw, &pstats);
862         if (ret != 0)
863                 return ret;
864
865         /* set stats offset base on current values */
866         hw->eth_stats_offset = pstats;
867
868         return 0;
869 }
870
871 static void
872 ice_dcf_free_repr_info(struct ice_dcf_adapter *dcf_adapter)
873 {
874         if (dcf_adapter->repr_infos) {
875                 rte_free(dcf_adapter->repr_infos);
876                 dcf_adapter->repr_infos = NULL;
877         }
878 }
879
880 static int
881 ice_dcf_init_repr_info(struct ice_dcf_adapter *dcf_adapter)
882 {
883         dcf_adapter->repr_infos =
884                         rte_calloc("ice_dcf_rep_info",
885                                    dcf_adapter->real_hw.num_vfs,
886                                    sizeof(dcf_adapter->repr_infos[0]), 0);
887         if (!dcf_adapter->repr_infos) {
888                 PMD_DRV_LOG(ERR, "Failed to alloc memory for VF representors\n");
889                 return -ENOMEM;
890         }
891
892         return 0;
893 }
894
895 static int
896 ice_dcf_dev_close(struct rte_eth_dev *dev)
897 {
898         struct ice_dcf_adapter *adapter = dev->data->dev_private;
899
900         if (rte_eal_process_type() != RTE_PROC_PRIMARY)
901                 return 0;
902
903         (void)ice_dcf_dev_stop(dev);
904
905         ice_free_queues(dev);
906
907         ice_dcf_free_repr_info(adapter);
908         ice_dcf_uninit_parent_adapter(dev);
909         ice_dcf_uninit_hw(dev, &adapter->real_hw);
910
911         return 0;
912 }
913
914 int
915 ice_dcf_link_update(struct rte_eth_dev *dev,
916                     __rte_unused int wait_to_complete)
917 {
918         struct ice_dcf_adapter *ad = dev->data->dev_private;
919         struct ice_dcf_hw *hw = &ad->real_hw;
920         struct rte_eth_link new_link;
921
922         memset(&new_link, 0, sizeof(new_link));
923
924         /* Only read status info stored in VF, and the info is updated
925          * when receive LINK_CHANGE event from PF by virtchnl.
926          */
927         switch (hw->link_speed) {
928         case 10:
929                 new_link.link_speed = ETH_SPEED_NUM_10M;
930                 break;
931         case 100:
932                 new_link.link_speed = ETH_SPEED_NUM_100M;
933                 break;
934         case 1000:
935                 new_link.link_speed = ETH_SPEED_NUM_1G;
936                 break;
937         case 10000:
938                 new_link.link_speed = ETH_SPEED_NUM_10G;
939                 break;
940         case 20000:
941                 new_link.link_speed = ETH_SPEED_NUM_20G;
942                 break;
943         case 25000:
944                 new_link.link_speed = ETH_SPEED_NUM_25G;
945                 break;
946         case 40000:
947                 new_link.link_speed = ETH_SPEED_NUM_40G;
948                 break;
949         case 50000:
950                 new_link.link_speed = ETH_SPEED_NUM_50G;
951                 break;
952         case 100000:
953                 new_link.link_speed = ETH_SPEED_NUM_100G;
954                 break;
955         default:
956                 new_link.link_speed = ETH_SPEED_NUM_NONE;
957                 break;
958         }
959
960         new_link.link_duplex = ETH_LINK_FULL_DUPLEX;
961         new_link.link_status = hw->link_up ? ETH_LINK_UP :
962                                              ETH_LINK_DOWN;
963         new_link.link_autoneg = !(dev->data->dev_conf.link_speeds &
964                                 ETH_LINK_SPEED_FIXED);
965
966         return rte_eth_linkstatus_set(dev, &new_link);
967 }
968
969 /* Add UDP tunneling port */
970 static int
971 ice_dcf_dev_udp_tunnel_port_add(struct rte_eth_dev *dev,
972                                 struct rte_eth_udp_tunnel *udp_tunnel)
973 {
974         struct ice_dcf_adapter *adapter = dev->data->dev_private;
975         struct ice_adapter *parent_adapter = &adapter->parent;
976         struct ice_hw *parent_hw = &parent_adapter->hw;
977         int ret = 0;
978
979         if (!udp_tunnel)
980                 return -EINVAL;
981
982         switch (udp_tunnel->prot_type) {
983         case RTE_TUNNEL_TYPE_VXLAN:
984                 ret = ice_create_tunnel(parent_hw, TNL_VXLAN,
985                                         udp_tunnel->udp_port);
986                 break;
987         case RTE_TUNNEL_TYPE_ECPRI:
988                 ret = ice_create_tunnel(parent_hw, TNL_ECPRI,
989                                         udp_tunnel->udp_port);
990                 break;
991         default:
992                 PMD_DRV_LOG(ERR, "Invalid tunnel type");
993                 ret = -EINVAL;
994                 break;
995         }
996
997         return ret;
998 }
999
1000 /* Delete UDP tunneling port */
1001 static int
1002 ice_dcf_dev_udp_tunnel_port_del(struct rte_eth_dev *dev,
1003                                 struct rte_eth_udp_tunnel *udp_tunnel)
1004 {
1005         struct ice_dcf_adapter *adapter = dev->data->dev_private;
1006         struct ice_adapter *parent_adapter = &adapter->parent;
1007         struct ice_hw *parent_hw = &parent_adapter->hw;
1008         int ret = 0;
1009
1010         if (!udp_tunnel)
1011                 return -EINVAL;
1012
1013         switch (udp_tunnel->prot_type) {
1014         case RTE_TUNNEL_TYPE_VXLAN:
1015         case RTE_TUNNEL_TYPE_ECPRI:
1016                 ret = ice_destroy_tunnel(parent_hw, udp_tunnel->udp_port, 0);
1017                 break;
1018         default:
1019                 PMD_DRV_LOG(ERR, "Invalid tunnel type");
1020                 ret = -EINVAL;
1021                 break;
1022         }
1023
1024         return ret;
1025 }
1026
1027 static int
1028 ice_dcf_tm_ops_get(struct rte_eth_dev *dev __rte_unused,
1029                 void *arg)
1030 {
1031         if (!arg)
1032                 return -EINVAL;
1033
1034         *(const void **)arg = &ice_dcf_tm_ops;
1035
1036         return 0;
1037 }
1038
1039 static int
1040 ice_dcf_dev_reset(struct rte_eth_dev *dev)
1041 {
1042         int ret;
1043
1044         ret = ice_dcf_dev_uninit(dev);
1045         if (ret)
1046                 return ret;
1047
1048         ret = ice_dcf_dev_init(dev);
1049
1050         return ret;
1051 }
1052
1053 static const struct eth_dev_ops ice_dcf_eth_dev_ops = {
1054         .dev_start               = ice_dcf_dev_start,
1055         .dev_stop                = ice_dcf_dev_stop,
1056         .dev_close               = ice_dcf_dev_close,
1057         .dev_reset               = ice_dcf_dev_reset,
1058         .dev_configure           = ice_dcf_dev_configure,
1059         .dev_infos_get           = ice_dcf_dev_info_get,
1060         .rx_queue_setup          = ice_rx_queue_setup,
1061         .tx_queue_setup          = ice_tx_queue_setup,
1062         .rx_queue_release        = ice_dev_rx_queue_release,
1063         .tx_queue_release        = ice_dev_tx_queue_release,
1064         .rx_queue_start          = ice_dcf_rx_queue_start,
1065         .tx_queue_start          = ice_dcf_tx_queue_start,
1066         .rx_queue_stop           = ice_dcf_rx_queue_stop,
1067         .tx_queue_stop           = ice_dcf_tx_queue_stop,
1068         .link_update             = ice_dcf_link_update,
1069         .stats_get               = ice_dcf_stats_get,
1070         .stats_reset             = ice_dcf_stats_reset,
1071         .promiscuous_enable      = ice_dcf_dev_promiscuous_enable,
1072         .promiscuous_disable     = ice_dcf_dev_promiscuous_disable,
1073         .allmulticast_enable     = ice_dcf_dev_allmulticast_enable,
1074         .allmulticast_disable    = ice_dcf_dev_allmulticast_disable,
1075         .flow_ops_get            = ice_dcf_dev_flow_ops_get,
1076         .udp_tunnel_port_add     = ice_dcf_dev_udp_tunnel_port_add,
1077         .udp_tunnel_port_del     = ice_dcf_dev_udp_tunnel_port_del,
1078         .tm_ops_get              = ice_dcf_tm_ops_get,
1079 };
1080
1081 static int
1082 ice_dcf_dev_init(struct rte_eth_dev *eth_dev)
1083 {
1084         struct ice_dcf_adapter *adapter = eth_dev->data->dev_private;
1085
1086         adapter->real_hw.resetting = false;
1087         eth_dev->dev_ops = &ice_dcf_eth_dev_ops;
1088         eth_dev->rx_pkt_burst = ice_dcf_recv_pkts;
1089         eth_dev->tx_pkt_burst = ice_dcf_xmit_pkts;
1090
1091         if (rte_eal_process_type() != RTE_PROC_PRIMARY)
1092                 return 0;
1093
1094         adapter->real_hw.vc_event_msg_cb = ice_dcf_handle_pf_event_msg;
1095         if (ice_dcf_init_hw(eth_dev, &adapter->real_hw) != 0) {
1096                 PMD_INIT_LOG(ERR, "Failed to init DCF hardware");
1097                 return -1;
1098         }
1099
1100         if (ice_dcf_init_parent_adapter(eth_dev) != 0) {
1101                 PMD_INIT_LOG(ERR, "Failed to init DCF parent adapter");
1102                 ice_dcf_uninit_hw(eth_dev, &adapter->real_hw);
1103                 return -1;
1104         }
1105
1106         return 0;
1107 }
1108
1109 static int
1110 ice_dcf_dev_uninit(struct rte_eth_dev *eth_dev)
1111 {
1112         ice_dcf_dev_close(eth_dev);
1113
1114         return 0;
1115 }
1116
1117 static int
1118 ice_dcf_cap_check_handler(__rte_unused const char *key,
1119                           const char *value, __rte_unused void *opaque)
1120 {
1121         if (strcmp(value, "dcf"))
1122                 return -1;
1123
1124         return 0;
1125 }
1126
1127 static int
1128 ice_dcf_cap_selected(struct rte_devargs *devargs)
1129 {
1130         struct rte_kvargs *kvlist;
1131         const char *key = "cap";
1132         int ret = 0;
1133
1134         if (devargs == NULL)
1135                 return 0;
1136
1137         kvlist = rte_kvargs_parse(devargs->args, NULL);
1138         if (kvlist == NULL)
1139                 return 0;
1140
1141         if (!rte_kvargs_count(kvlist, key))
1142                 goto exit;
1143
1144         /* dcf capability selected when there's a key-value pair: cap=dcf */
1145         if (rte_kvargs_process(kvlist, key,
1146                                ice_dcf_cap_check_handler, NULL) < 0)
1147                 goto exit;
1148
1149         ret = 1;
1150
1151 exit:
1152         rte_kvargs_free(kvlist);
1153         return ret;
1154 }
1155
1156 static int
1157 eth_ice_dcf_pci_probe(__rte_unused struct rte_pci_driver *pci_drv,
1158                       struct rte_pci_device *pci_dev)
1159 {
1160         struct rte_eth_devargs eth_da = { .nb_representor_ports = 0 };
1161         struct ice_dcf_vf_repr_param repr_param;
1162         char repr_name[RTE_ETH_NAME_MAX_LEN];
1163         struct ice_dcf_adapter *dcf_adapter;
1164         struct rte_eth_dev *dcf_ethdev;
1165         uint16_t dcf_vsi_id;
1166         int i, ret;
1167
1168         if (!ice_dcf_cap_selected(pci_dev->device.devargs))
1169                 return 1;
1170
1171         ret = rte_eth_devargs_parse(pci_dev->device.devargs->args, &eth_da);
1172         if (ret)
1173                 return ret;
1174
1175         ret = rte_eth_dev_pci_generic_probe(pci_dev,
1176                                             sizeof(struct ice_dcf_adapter),
1177                                             ice_dcf_dev_init);
1178         if (ret || !eth_da.nb_representor_ports)
1179                 return ret;
1180         if (eth_da.type != RTE_ETH_REPRESENTOR_VF)
1181                 return -ENOTSUP;
1182
1183         dcf_ethdev = rte_eth_dev_allocated(pci_dev->device.name);
1184         if (dcf_ethdev == NULL)
1185                 return -ENODEV;
1186
1187         dcf_adapter = dcf_ethdev->data->dev_private;
1188         ret = ice_dcf_init_repr_info(dcf_adapter);
1189         if (ret)
1190                 return ret;
1191
1192         if (eth_da.nb_representor_ports > dcf_adapter->real_hw.num_vfs ||
1193             eth_da.nb_representor_ports >= RTE_MAX_ETHPORTS) {
1194                 PMD_DRV_LOG(ERR, "the number of port representors is too large: %u",
1195                             eth_da.nb_representor_ports);
1196                 ice_dcf_free_repr_info(dcf_adapter);
1197                 return -EINVAL;
1198         }
1199
1200         dcf_vsi_id = dcf_adapter->real_hw.vsi_id | VIRTCHNL_DCF_VF_VSI_VALID;
1201
1202         repr_param.dcf_eth_dev = dcf_ethdev;
1203         repr_param.switch_domain_id = 0;
1204
1205         for (i = 0; i < eth_da.nb_representor_ports; i++) {
1206                 uint16_t vf_id = eth_da.representor_ports[i];
1207                 struct rte_eth_dev *vf_rep_eth_dev;
1208
1209                 if (vf_id >= dcf_adapter->real_hw.num_vfs) {
1210                         PMD_DRV_LOG(ERR, "VF ID %u is out of range (0 ~ %u)",
1211                                     vf_id, dcf_adapter->real_hw.num_vfs - 1);
1212                         ret = -EINVAL;
1213                         break;
1214                 }
1215
1216                 if (dcf_adapter->real_hw.vf_vsi_map[vf_id] == dcf_vsi_id) {
1217                         PMD_DRV_LOG(ERR, "VF ID %u is DCF's ID.\n", vf_id);
1218                         ret = -EINVAL;
1219                         break;
1220                 }
1221
1222                 repr_param.vf_id = vf_id;
1223                 snprintf(repr_name, sizeof(repr_name), "net_%s_representor_%u",
1224                          pci_dev->device.name, vf_id);
1225                 ret = rte_eth_dev_create(&pci_dev->device, repr_name,
1226                                          sizeof(struct ice_dcf_vf_repr),
1227                                          NULL, NULL, ice_dcf_vf_repr_init,
1228                                          &repr_param);
1229                 if (ret) {
1230                         PMD_DRV_LOG(ERR, "failed to create DCF VF representor %s",
1231                                     repr_name);
1232                         break;
1233                 }
1234
1235                 vf_rep_eth_dev = rte_eth_dev_allocated(repr_name);
1236                 if (!vf_rep_eth_dev) {
1237                         PMD_DRV_LOG(ERR,
1238                                     "Failed to find the ethdev for DCF VF representor: %s",
1239                                     repr_name);
1240                         ret = -ENODEV;
1241                         break;
1242                 }
1243
1244                 dcf_adapter->repr_infos[vf_id].vf_rep_eth_dev = vf_rep_eth_dev;
1245                 dcf_adapter->num_reprs++;
1246         }
1247
1248         return ret;
1249 }
1250
1251 static int
1252 eth_ice_dcf_pci_remove(struct rte_pci_device *pci_dev)
1253 {
1254         struct rte_eth_dev *eth_dev;
1255
1256         eth_dev = rte_eth_dev_allocated(pci_dev->device.name);
1257         if (!eth_dev)
1258                 return 0;
1259
1260         if (eth_dev->data->dev_flags & RTE_ETH_DEV_REPRESENTOR)
1261                 return rte_eth_dev_pci_generic_remove(pci_dev,
1262                                                       ice_dcf_vf_repr_uninit);
1263         else
1264                 return rte_eth_dev_pci_generic_remove(pci_dev,
1265                                                       ice_dcf_dev_uninit);
1266 }
1267
1268 static const struct rte_pci_id pci_id_ice_dcf_map[] = {
1269         { RTE_PCI_DEVICE(IAVF_INTEL_VENDOR_ID, IAVF_DEV_ID_ADAPTIVE_VF) },
1270         { .vendor_id = 0, /* sentinel */ },
1271 };
1272
1273 static struct rte_pci_driver rte_ice_dcf_pmd = {
1274         .id_table = pci_id_ice_dcf_map,
1275         .drv_flags = RTE_PCI_DRV_NEED_MAPPING,
1276         .probe = eth_ice_dcf_pci_probe,
1277         .remove = eth_ice_dcf_pci_remove,
1278 };
1279
1280 RTE_PMD_REGISTER_PCI(net_ice_dcf, rte_ice_dcf_pmd);
1281 RTE_PMD_REGISTER_PCI_TABLE(net_ice_dcf, pci_id_ice_dcf_map);
1282 RTE_PMD_REGISTER_KMOD_DEP(net_ice_dcf, "* igb_uio | vfio-pci");
1283 RTE_PMD_REGISTER_PARAM_STRING(net_ice_dcf, "cap=dcf");