xref: /dpdk/drivers/net/ice/ice_dcf_ethdev.c (revision 985e7673)
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->mtu > RTE_ETHER_MTU) {
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 	if (hw->tm_conf.root && !hw->tm_conf.committed) {
532 		PMD_DRV_LOG(ERR,
533 			"please call hierarchy_commit() before starting the port");
534 		return -EIO;
535 	}
536 
537 	ad->pf.adapter_stopped = 0;
538 
539 	hw->num_queue_pairs = RTE_MAX(dev->data->nb_rx_queues,
540 				      dev->data->nb_tx_queues);
541 
542 	ret = ice_dcf_init_rx_queues(dev);
543 	if (ret) {
544 		PMD_DRV_LOG(ERR, "Fail to init queues");
545 		return ret;
546 	}
547 
548 	if (hw->vf_res->vf_cap_flags & VIRTCHNL_VF_OFFLOAD_RSS_PF) {
549 		ret = ice_dcf_init_rss(hw);
550 		if (ret) {
551 			PMD_DRV_LOG(ERR, "Failed to configure RSS");
552 			return ret;
553 		}
554 	}
555 
556 	ret = ice_dcf_configure_queues(hw);
557 	if (ret) {
558 		PMD_DRV_LOG(ERR, "Fail to config queues");
559 		return ret;
560 	}
561 
562 	ret = ice_dcf_config_rx_queues_irqs(dev, intr_handle);
563 	if (ret) {
564 		PMD_DRV_LOG(ERR, "Fail to config rx queues' irqs");
565 		return ret;
566 	}
567 
568 	if (dev->data->dev_conf.intr_conf.rxq != 0) {
569 		rte_intr_disable(intr_handle);
570 		rte_intr_enable(intr_handle);
571 	}
572 
573 	ret = ice_dcf_start_queues(dev);
574 	if (ret) {
575 		PMD_DRV_LOG(ERR, "Failed to enable queues");
576 		return ret;
577 	}
578 
579 	ret = ice_dcf_add_del_all_mac_addr(hw, true);
580 	if (ret) {
581 		PMD_DRV_LOG(ERR, "Failed to add mac addr");
582 		return ret;
583 	}
584 
585 	dev->data->dev_link.link_status = ETH_LINK_UP;
586 
587 	return 0;
588 }
589 
590 static void
591 ice_dcf_stop_queues(struct rte_eth_dev *dev)
592 {
593 	struct ice_dcf_adapter *ad = dev->data->dev_private;
594 	struct ice_dcf_hw *hw = &ad->real_hw;
595 	struct ice_rx_queue *rxq;
596 	struct ice_tx_queue *txq;
597 	int ret, i;
598 
599 	/* Stop All queues */
600 	ret = ice_dcf_disable_queues(hw);
601 	if (ret)
602 		PMD_DRV_LOG(WARNING, "Fail to stop queues");
603 
604 	for (i = 0; i < dev->data->nb_tx_queues; i++) {
605 		txq = dev->data->tx_queues[i];
606 		if (!txq)
607 			continue;
608 		txq->tx_rel_mbufs(txq);
609 		reset_tx_queue(txq);
610 		dev->data->tx_queue_state[i] = RTE_ETH_QUEUE_STATE_STOPPED;
611 	}
612 	for (i = 0; i < dev->data->nb_rx_queues; i++) {
613 		rxq = dev->data->rx_queues[i];
614 		if (!rxq)
615 			continue;
616 		rxq->rx_rel_mbufs(rxq);
617 		reset_rx_queue(rxq);
618 		dev->data->rx_queue_state[i] = RTE_ETH_QUEUE_STATE_STOPPED;
619 	}
620 }
621 
622 static int
623 ice_dcf_dev_stop(struct rte_eth_dev *dev)
624 {
625 	struct ice_dcf_adapter *dcf_ad = dev->data->dev_private;
626 	struct rte_intr_handle *intr_handle = dev->intr_handle;
627 	struct ice_adapter *ad = &dcf_ad->parent;
628 	struct ice_dcf_hw *hw = &dcf_ad->real_hw;
629 
630 	if (ad->pf.adapter_stopped == 1) {
631 		PMD_DRV_LOG(DEBUG, "Port is already stopped");
632 		return 0;
633 	}
634 
635 	/* Stop the VF representors for this device */
636 	ice_dcf_vf_repr_stop_all(dcf_ad);
637 
638 	ice_dcf_stop_queues(dev);
639 
640 	rte_intr_efd_disable(intr_handle);
641 	if (intr_handle->intr_vec) {
642 		rte_free(intr_handle->intr_vec);
643 		intr_handle->intr_vec = NULL;
644 	}
645 
646 	ice_dcf_add_del_all_mac_addr(&dcf_ad->real_hw, false);
647 	dev->data->dev_link.link_status = ETH_LINK_DOWN;
648 	ad->pf.adapter_stopped = 1;
649 	hw->tm_conf.committed = false;
650 
651 	return 0;
652 }
653 
654 static int
655 ice_dcf_dev_configure(struct rte_eth_dev *dev)
656 {
657 	struct ice_dcf_adapter *dcf_ad = dev->data->dev_private;
658 	struct ice_adapter *ad = &dcf_ad->parent;
659 
660 	ad->rx_bulk_alloc_allowed = true;
661 	ad->tx_simple_allowed = true;
662 
663 	if (dev->data->dev_conf.rxmode.mq_mode & ETH_MQ_RX_RSS_FLAG)
664 		dev->data->dev_conf.rxmode.offloads |= DEV_RX_OFFLOAD_RSS_HASH;
665 
666 	return 0;
667 }
668 
669 static int
670 ice_dcf_dev_info_get(struct rte_eth_dev *dev,
671 		     struct rte_eth_dev_info *dev_info)
672 {
673 	struct ice_dcf_adapter *adapter = dev->data->dev_private;
674 	struct ice_dcf_hw *hw = &adapter->real_hw;
675 
676 	dev_info->max_mac_addrs = 1;
677 	dev_info->max_rx_queues = hw->vsi_res->num_queue_pairs;
678 	dev_info->max_tx_queues = hw->vsi_res->num_queue_pairs;
679 	dev_info->min_rx_bufsize = ICE_BUF_SIZE_MIN;
680 	dev_info->max_rx_pktlen = ICE_FRAME_SIZE_MAX;
681 	dev_info->hash_key_size = hw->vf_res->rss_key_size;
682 	dev_info->reta_size = hw->vf_res->rss_lut_size;
683 	dev_info->flow_type_rss_offloads = ICE_RSS_OFFLOAD_ALL;
684 
685 	dev_info->rx_offload_capa =
686 		DEV_RX_OFFLOAD_VLAN_STRIP |
687 		DEV_RX_OFFLOAD_IPV4_CKSUM |
688 		DEV_RX_OFFLOAD_UDP_CKSUM |
689 		DEV_RX_OFFLOAD_TCP_CKSUM |
690 		DEV_RX_OFFLOAD_OUTER_IPV4_CKSUM |
691 		DEV_RX_OFFLOAD_SCATTER |
692 		DEV_RX_OFFLOAD_VLAN_FILTER |
693 		DEV_RX_OFFLOAD_RSS_HASH;
694 	dev_info->tx_offload_capa =
695 		DEV_TX_OFFLOAD_VLAN_INSERT |
696 		DEV_TX_OFFLOAD_IPV4_CKSUM |
697 		DEV_TX_OFFLOAD_UDP_CKSUM |
698 		DEV_TX_OFFLOAD_TCP_CKSUM |
699 		DEV_TX_OFFLOAD_SCTP_CKSUM |
700 		DEV_TX_OFFLOAD_OUTER_IPV4_CKSUM |
701 		DEV_TX_OFFLOAD_TCP_TSO |
702 		DEV_TX_OFFLOAD_VXLAN_TNL_TSO |
703 		DEV_TX_OFFLOAD_GRE_TNL_TSO |
704 		DEV_TX_OFFLOAD_IPIP_TNL_TSO |
705 		DEV_TX_OFFLOAD_GENEVE_TNL_TSO |
706 		DEV_TX_OFFLOAD_MULTI_SEGS;
707 
708 	dev_info->default_rxconf = (struct rte_eth_rxconf) {
709 		.rx_thresh = {
710 			.pthresh = ICE_DEFAULT_RX_PTHRESH,
711 			.hthresh = ICE_DEFAULT_RX_HTHRESH,
712 			.wthresh = ICE_DEFAULT_RX_WTHRESH,
713 		},
714 		.rx_free_thresh = ICE_DEFAULT_RX_FREE_THRESH,
715 		.rx_drop_en = 0,
716 		.offloads = 0,
717 	};
718 
719 	dev_info->default_txconf = (struct rte_eth_txconf) {
720 		.tx_thresh = {
721 			.pthresh = ICE_DEFAULT_TX_PTHRESH,
722 			.hthresh = ICE_DEFAULT_TX_HTHRESH,
723 			.wthresh = ICE_DEFAULT_TX_WTHRESH,
724 		},
725 		.tx_free_thresh = ICE_DEFAULT_TX_FREE_THRESH,
726 		.tx_rs_thresh = ICE_DEFAULT_TX_RSBIT_THRESH,
727 		.offloads = 0,
728 	};
729 
730 	dev_info->rx_desc_lim = (struct rte_eth_desc_lim) {
731 		.nb_max = ICE_MAX_RING_DESC,
732 		.nb_min = ICE_MIN_RING_DESC,
733 		.nb_align = ICE_ALIGN_RING_DESC,
734 	};
735 
736 	dev_info->tx_desc_lim = (struct rte_eth_desc_lim) {
737 		.nb_max = ICE_MAX_RING_DESC,
738 		.nb_min = ICE_MIN_RING_DESC,
739 		.nb_align = ICE_ALIGN_RING_DESC,
740 	};
741 
742 	return 0;
743 }
744 
745 static int
746 ice_dcf_dev_promiscuous_enable(__rte_unused struct rte_eth_dev *dev)
747 {
748 	return 0;
749 }
750 
751 static int
752 ice_dcf_dev_promiscuous_disable(__rte_unused struct rte_eth_dev *dev)
753 {
754 	return 0;
755 }
756 
757 static int
758 ice_dcf_dev_allmulticast_enable(__rte_unused struct rte_eth_dev *dev)
759 {
760 	return 0;
761 }
762 
763 static int
764 ice_dcf_dev_allmulticast_disable(__rte_unused struct rte_eth_dev *dev)
765 {
766 	return 0;
767 }
768 
769 static int
770 ice_dcf_dev_flow_ops_get(struct rte_eth_dev *dev,
771 			 const struct rte_flow_ops **ops)
772 {
773 	if (!dev)
774 		return -EINVAL;
775 
776 	*ops = &ice_flow_ops;
777 	return 0;
778 }
779 
780 #define ICE_DCF_32_BIT_WIDTH (CHAR_BIT * 4)
781 #define ICE_DCF_48_BIT_WIDTH (CHAR_BIT * 6)
782 #define ICE_DCF_48_BIT_MASK  RTE_LEN2MASK(ICE_DCF_48_BIT_WIDTH, uint64_t)
783 
784 static void
785 ice_dcf_stat_update_48(uint64_t *offset, uint64_t *stat)
786 {
787 	if (*stat >= *offset)
788 		*stat = *stat - *offset;
789 	else
790 		*stat = (uint64_t)((*stat +
791 			((uint64_t)1 << ICE_DCF_48_BIT_WIDTH)) - *offset);
792 
793 	*stat &= ICE_DCF_48_BIT_MASK;
794 }
795 
796 static void
797 ice_dcf_stat_update_32(uint64_t *offset, uint64_t *stat)
798 {
799 	if (*stat >= *offset)
800 		*stat = (uint64_t)(*stat - *offset);
801 	else
802 		*stat = (uint64_t)((*stat +
803 			((uint64_t)1 << ICE_DCF_32_BIT_WIDTH)) - *offset);
804 }
805 
806 static void
807 ice_dcf_update_stats(struct virtchnl_eth_stats *oes,
808 		     struct virtchnl_eth_stats *nes)
809 {
810 	ice_dcf_stat_update_48(&oes->rx_bytes, &nes->rx_bytes);
811 	ice_dcf_stat_update_48(&oes->rx_unicast, &nes->rx_unicast);
812 	ice_dcf_stat_update_48(&oes->rx_multicast, &nes->rx_multicast);
813 	ice_dcf_stat_update_48(&oes->rx_broadcast, &nes->rx_broadcast);
814 	ice_dcf_stat_update_32(&oes->rx_discards, &nes->rx_discards);
815 	ice_dcf_stat_update_48(&oes->tx_bytes, &nes->tx_bytes);
816 	ice_dcf_stat_update_48(&oes->tx_unicast, &nes->tx_unicast);
817 	ice_dcf_stat_update_48(&oes->tx_multicast, &nes->tx_multicast);
818 	ice_dcf_stat_update_48(&oes->tx_broadcast, &nes->tx_broadcast);
819 	ice_dcf_stat_update_32(&oes->tx_errors, &nes->tx_errors);
820 	ice_dcf_stat_update_32(&oes->tx_discards, &nes->tx_discards);
821 }
822 
823 
824 static int
825 ice_dcf_stats_get(struct rte_eth_dev *dev, struct rte_eth_stats *stats)
826 {
827 	struct ice_dcf_adapter *ad = dev->data->dev_private;
828 	struct ice_dcf_hw *hw = &ad->real_hw;
829 	struct virtchnl_eth_stats pstats;
830 	int ret;
831 
832 	if (hw->resetting) {
833 		PMD_DRV_LOG(ERR,
834 			    "The DCF has been reset by PF, please reinit first");
835 		return -EIO;
836 	}
837 
838 	ret = ice_dcf_query_stats(hw, &pstats);
839 	if (ret == 0) {
840 		ice_dcf_update_stats(&hw->eth_stats_offset, &pstats);
841 		stats->ipackets = pstats.rx_unicast + pstats.rx_multicast +
842 				pstats.rx_broadcast - pstats.rx_discards;
843 		stats->opackets = pstats.tx_broadcast + pstats.tx_multicast +
844 						pstats.tx_unicast;
845 		stats->imissed = pstats.rx_discards;
846 		stats->oerrors = pstats.tx_errors + pstats.tx_discards;
847 		stats->ibytes = pstats.rx_bytes;
848 		stats->ibytes -= stats->ipackets * RTE_ETHER_CRC_LEN;
849 		stats->obytes = pstats.tx_bytes;
850 	} else {
851 		PMD_DRV_LOG(ERR, "Get statistics failed");
852 	}
853 	return ret;
854 }
855 
856 static int
857 ice_dcf_stats_reset(struct rte_eth_dev *dev)
858 {
859 	struct ice_dcf_adapter *ad = dev->data->dev_private;
860 	struct ice_dcf_hw *hw = &ad->real_hw;
861 	struct virtchnl_eth_stats pstats;
862 	int ret;
863 
864 	if (hw->resetting)
865 		return 0;
866 
867 	/* read stat values to clear hardware registers */
868 	ret = ice_dcf_query_stats(hw, &pstats);
869 	if (ret != 0)
870 		return ret;
871 
872 	/* set stats offset base on current values */
873 	hw->eth_stats_offset = pstats;
874 
875 	return 0;
876 }
877 
878 static void
879 ice_dcf_free_repr_info(struct ice_dcf_adapter *dcf_adapter)
880 {
881 	if (dcf_adapter->repr_infos) {
882 		rte_free(dcf_adapter->repr_infos);
883 		dcf_adapter->repr_infos = NULL;
884 	}
885 }
886 
887 static int
888 ice_dcf_init_repr_info(struct ice_dcf_adapter *dcf_adapter)
889 {
890 	dcf_adapter->repr_infos =
891 			rte_calloc("ice_dcf_rep_info",
892 				   dcf_adapter->real_hw.num_vfs,
893 				   sizeof(dcf_adapter->repr_infos[0]), 0);
894 	if (!dcf_adapter->repr_infos) {
895 		PMD_DRV_LOG(ERR, "Failed to alloc memory for VF representors\n");
896 		return -ENOMEM;
897 	}
898 
899 	return 0;
900 }
901 
902 static int
903 ice_dcf_dev_close(struct rte_eth_dev *dev)
904 {
905 	struct ice_dcf_adapter *adapter = dev->data->dev_private;
906 
907 	if (rte_eal_process_type() != RTE_PROC_PRIMARY)
908 		return 0;
909 
910 	(void)ice_dcf_dev_stop(dev);
911 
912 	ice_free_queues(dev);
913 
914 	ice_dcf_free_repr_info(adapter);
915 	ice_dcf_uninit_parent_adapter(dev);
916 	ice_dcf_uninit_hw(dev, &adapter->real_hw);
917 
918 	return 0;
919 }
920 
921 int
922 ice_dcf_link_update(struct rte_eth_dev *dev,
923 		    __rte_unused int wait_to_complete)
924 {
925 	struct ice_dcf_adapter *ad = dev->data->dev_private;
926 	struct ice_dcf_hw *hw = &ad->real_hw;
927 	struct rte_eth_link new_link;
928 
929 	memset(&new_link, 0, sizeof(new_link));
930 
931 	/* Only read status info stored in VF, and the info is updated
932 	 * when receive LINK_CHANGE event from PF by virtchnl.
933 	 */
934 	switch (hw->link_speed) {
935 	case 10:
936 		new_link.link_speed = ETH_SPEED_NUM_10M;
937 		break;
938 	case 100:
939 		new_link.link_speed = ETH_SPEED_NUM_100M;
940 		break;
941 	case 1000:
942 		new_link.link_speed = ETH_SPEED_NUM_1G;
943 		break;
944 	case 10000:
945 		new_link.link_speed = ETH_SPEED_NUM_10G;
946 		break;
947 	case 20000:
948 		new_link.link_speed = ETH_SPEED_NUM_20G;
949 		break;
950 	case 25000:
951 		new_link.link_speed = ETH_SPEED_NUM_25G;
952 		break;
953 	case 40000:
954 		new_link.link_speed = ETH_SPEED_NUM_40G;
955 		break;
956 	case 50000:
957 		new_link.link_speed = ETH_SPEED_NUM_50G;
958 		break;
959 	case 100000:
960 		new_link.link_speed = ETH_SPEED_NUM_100G;
961 		break;
962 	default:
963 		new_link.link_speed = ETH_SPEED_NUM_NONE;
964 		break;
965 	}
966 
967 	new_link.link_duplex = ETH_LINK_FULL_DUPLEX;
968 	new_link.link_status = hw->link_up ? ETH_LINK_UP :
969 					     ETH_LINK_DOWN;
970 	new_link.link_autoneg = !(dev->data->dev_conf.link_speeds &
971 				ETH_LINK_SPEED_FIXED);
972 
973 	return rte_eth_linkstatus_set(dev, &new_link);
974 }
975 
976 /* Add UDP tunneling port */
977 static int
978 ice_dcf_dev_udp_tunnel_port_add(struct rte_eth_dev *dev,
979 				struct rte_eth_udp_tunnel *udp_tunnel)
980 {
981 	struct ice_dcf_adapter *adapter = dev->data->dev_private;
982 	struct ice_adapter *parent_adapter = &adapter->parent;
983 	struct ice_hw *parent_hw = &parent_adapter->hw;
984 	int ret = 0;
985 
986 	if (!udp_tunnel)
987 		return -EINVAL;
988 
989 	switch (udp_tunnel->prot_type) {
990 	case RTE_TUNNEL_TYPE_VXLAN:
991 		ret = ice_create_tunnel(parent_hw, TNL_VXLAN,
992 					udp_tunnel->udp_port);
993 		break;
994 	case RTE_TUNNEL_TYPE_ECPRI:
995 		ret = ice_create_tunnel(parent_hw, TNL_ECPRI,
996 					udp_tunnel->udp_port);
997 		break;
998 	default:
999 		PMD_DRV_LOG(ERR, "Invalid tunnel type");
1000 		ret = -EINVAL;
1001 		break;
1002 	}
1003 
1004 	return ret;
1005 }
1006 
1007 /* Delete UDP tunneling port */
1008 static int
1009 ice_dcf_dev_udp_tunnel_port_del(struct rte_eth_dev *dev,
1010 				struct rte_eth_udp_tunnel *udp_tunnel)
1011 {
1012 	struct ice_dcf_adapter *adapter = dev->data->dev_private;
1013 	struct ice_adapter *parent_adapter = &adapter->parent;
1014 	struct ice_hw *parent_hw = &parent_adapter->hw;
1015 	int ret = 0;
1016 
1017 	if (!udp_tunnel)
1018 		return -EINVAL;
1019 
1020 	switch (udp_tunnel->prot_type) {
1021 	case RTE_TUNNEL_TYPE_VXLAN:
1022 	case RTE_TUNNEL_TYPE_ECPRI:
1023 		ret = ice_destroy_tunnel(parent_hw, udp_tunnel->udp_port, 0);
1024 		break;
1025 	default:
1026 		PMD_DRV_LOG(ERR, "Invalid tunnel type");
1027 		ret = -EINVAL;
1028 		break;
1029 	}
1030 
1031 	return ret;
1032 }
1033 
1034 static int
1035 ice_dcf_tm_ops_get(struct rte_eth_dev *dev __rte_unused,
1036 		void *arg)
1037 {
1038 	if (!arg)
1039 		return -EINVAL;
1040 
1041 	*(const void **)arg = &ice_dcf_tm_ops;
1042 
1043 	return 0;
1044 }
1045 
1046 static int
1047 ice_dcf_dev_reset(struct rte_eth_dev *dev)
1048 {
1049 	int ret;
1050 
1051 	ret = ice_dcf_dev_uninit(dev);
1052 	if (ret)
1053 		return ret;
1054 
1055 	ret = ice_dcf_dev_init(dev);
1056 
1057 	return ret;
1058 }
1059 
1060 static const struct eth_dev_ops ice_dcf_eth_dev_ops = {
1061 	.dev_start               = ice_dcf_dev_start,
1062 	.dev_stop                = ice_dcf_dev_stop,
1063 	.dev_close               = ice_dcf_dev_close,
1064 	.dev_reset               = ice_dcf_dev_reset,
1065 	.dev_configure           = ice_dcf_dev_configure,
1066 	.dev_infos_get           = ice_dcf_dev_info_get,
1067 	.rx_queue_setup          = ice_rx_queue_setup,
1068 	.tx_queue_setup          = ice_tx_queue_setup,
1069 	.rx_queue_release        = ice_dev_rx_queue_release,
1070 	.tx_queue_release        = ice_dev_tx_queue_release,
1071 	.rx_queue_start          = ice_dcf_rx_queue_start,
1072 	.tx_queue_start          = ice_dcf_tx_queue_start,
1073 	.rx_queue_stop           = ice_dcf_rx_queue_stop,
1074 	.tx_queue_stop           = ice_dcf_tx_queue_stop,
1075 	.link_update             = ice_dcf_link_update,
1076 	.stats_get               = ice_dcf_stats_get,
1077 	.stats_reset             = ice_dcf_stats_reset,
1078 	.promiscuous_enable      = ice_dcf_dev_promiscuous_enable,
1079 	.promiscuous_disable     = ice_dcf_dev_promiscuous_disable,
1080 	.allmulticast_enable     = ice_dcf_dev_allmulticast_enable,
1081 	.allmulticast_disable    = ice_dcf_dev_allmulticast_disable,
1082 	.flow_ops_get            = ice_dcf_dev_flow_ops_get,
1083 	.udp_tunnel_port_add	 = ice_dcf_dev_udp_tunnel_port_add,
1084 	.udp_tunnel_port_del	 = ice_dcf_dev_udp_tunnel_port_del,
1085 	.tm_ops_get              = ice_dcf_tm_ops_get,
1086 };
1087 
1088 static int
1089 ice_dcf_dev_init(struct rte_eth_dev *eth_dev)
1090 {
1091 	struct ice_dcf_adapter *adapter = eth_dev->data->dev_private;
1092 
1093 	adapter->real_hw.resetting = false;
1094 	eth_dev->dev_ops = &ice_dcf_eth_dev_ops;
1095 	eth_dev->rx_pkt_burst = ice_dcf_recv_pkts;
1096 	eth_dev->tx_pkt_burst = ice_dcf_xmit_pkts;
1097 
1098 	if (rte_eal_process_type() != RTE_PROC_PRIMARY)
1099 		return 0;
1100 
1101 	adapter->real_hw.vc_event_msg_cb = ice_dcf_handle_pf_event_msg;
1102 	if (ice_dcf_init_hw(eth_dev, &adapter->real_hw) != 0) {
1103 		PMD_INIT_LOG(ERR, "Failed to init DCF hardware");
1104 		return -1;
1105 	}
1106 
1107 	if (ice_dcf_init_parent_adapter(eth_dev) != 0) {
1108 		PMD_INIT_LOG(ERR, "Failed to init DCF parent adapter");
1109 		ice_dcf_uninit_hw(eth_dev, &adapter->real_hw);
1110 		return -1;
1111 	}
1112 
1113 	return 0;
1114 }
1115 
1116 static int
1117 ice_dcf_dev_uninit(struct rte_eth_dev *eth_dev)
1118 {
1119 	ice_dcf_dev_close(eth_dev);
1120 
1121 	return 0;
1122 }
1123 
1124 static int
1125 ice_dcf_cap_check_handler(__rte_unused const char *key,
1126 			  const char *value, __rte_unused void *opaque)
1127 {
1128 	if (strcmp(value, "dcf"))
1129 		return -1;
1130 
1131 	return 0;
1132 }
1133 
1134 static int
1135 ice_dcf_cap_selected(struct rte_devargs *devargs)
1136 {
1137 	struct rte_kvargs *kvlist;
1138 	const char *key = "cap";
1139 	int ret = 0;
1140 
1141 	if (devargs == NULL)
1142 		return 0;
1143 
1144 	kvlist = rte_kvargs_parse(devargs->args, NULL);
1145 	if (kvlist == NULL)
1146 		return 0;
1147 
1148 	if (!rte_kvargs_count(kvlist, key))
1149 		goto exit;
1150 
1151 	/* dcf capability selected when there's a key-value pair: cap=dcf */
1152 	if (rte_kvargs_process(kvlist, key,
1153 			       ice_dcf_cap_check_handler, NULL) < 0)
1154 		goto exit;
1155 
1156 	ret = 1;
1157 
1158 exit:
1159 	rte_kvargs_free(kvlist);
1160 	return ret;
1161 }
1162 
1163 static int
1164 eth_ice_dcf_pci_probe(__rte_unused struct rte_pci_driver *pci_drv,
1165 		      struct rte_pci_device *pci_dev)
1166 {
1167 	struct rte_eth_devargs eth_da = { .nb_representor_ports = 0 };
1168 	struct ice_dcf_vf_repr_param repr_param;
1169 	char repr_name[RTE_ETH_NAME_MAX_LEN];
1170 	struct ice_dcf_adapter *dcf_adapter;
1171 	struct rte_eth_dev *dcf_ethdev;
1172 	uint16_t dcf_vsi_id;
1173 	int i, ret;
1174 
1175 	if (!ice_dcf_cap_selected(pci_dev->device.devargs))
1176 		return 1;
1177 
1178 	ret = rte_eth_devargs_parse(pci_dev->device.devargs->args, &eth_da);
1179 	if (ret)
1180 		return ret;
1181 
1182 	ret = rte_eth_dev_pci_generic_probe(pci_dev,
1183 					    sizeof(struct ice_dcf_adapter),
1184 					    ice_dcf_dev_init);
1185 	if (ret || !eth_da.nb_representor_ports)
1186 		return ret;
1187 	if (eth_da.type != RTE_ETH_REPRESENTOR_VF)
1188 		return -ENOTSUP;
1189 
1190 	dcf_ethdev = rte_eth_dev_allocated(pci_dev->device.name);
1191 	if (dcf_ethdev == NULL)
1192 		return -ENODEV;
1193 
1194 	dcf_adapter = dcf_ethdev->data->dev_private;
1195 	ret = ice_dcf_init_repr_info(dcf_adapter);
1196 	if (ret)
1197 		return ret;
1198 
1199 	if (eth_da.nb_representor_ports > dcf_adapter->real_hw.num_vfs ||
1200 	    eth_da.nb_representor_ports >= RTE_MAX_ETHPORTS) {
1201 		PMD_DRV_LOG(ERR, "the number of port representors is too large: %u",
1202 			    eth_da.nb_representor_ports);
1203 		ice_dcf_free_repr_info(dcf_adapter);
1204 		return -EINVAL;
1205 	}
1206 
1207 	dcf_vsi_id = dcf_adapter->real_hw.vsi_id | VIRTCHNL_DCF_VF_VSI_VALID;
1208 
1209 	repr_param.dcf_eth_dev = dcf_ethdev;
1210 	repr_param.switch_domain_id = 0;
1211 
1212 	for (i = 0; i < eth_da.nb_representor_ports; i++) {
1213 		uint16_t vf_id = eth_da.representor_ports[i];
1214 		struct rte_eth_dev *vf_rep_eth_dev;
1215 
1216 		if (vf_id >= dcf_adapter->real_hw.num_vfs) {
1217 			PMD_DRV_LOG(ERR, "VF ID %u is out of range (0 ~ %u)",
1218 				    vf_id, dcf_adapter->real_hw.num_vfs - 1);
1219 			ret = -EINVAL;
1220 			break;
1221 		}
1222 
1223 		if (dcf_adapter->real_hw.vf_vsi_map[vf_id] == dcf_vsi_id) {
1224 			PMD_DRV_LOG(ERR, "VF ID %u is DCF's ID.\n", vf_id);
1225 			ret = -EINVAL;
1226 			break;
1227 		}
1228 
1229 		repr_param.vf_id = vf_id;
1230 		snprintf(repr_name, sizeof(repr_name), "net_%s_representor_%u",
1231 			 pci_dev->device.name, vf_id);
1232 		ret = rte_eth_dev_create(&pci_dev->device, repr_name,
1233 					 sizeof(struct ice_dcf_vf_repr),
1234 					 NULL, NULL, ice_dcf_vf_repr_init,
1235 					 &repr_param);
1236 		if (ret) {
1237 			PMD_DRV_LOG(ERR, "failed to create DCF VF representor %s",
1238 				    repr_name);
1239 			break;
1240 		}
1241 
1242 		vf_rep_eth_dev = rte_eth_dev_allocated(repr_name);
1243 		if (!vf_rep_eth_dev) {
1244 			PMD_DRV_LOG(ERR,
1245 				    "Failed to find the ethdev for DCF VF representor: %s",
1246 				    repr_name);
1247 			ret = -ENODEV;
1248 			break;
1249 		}
1250 
1251 		dcf_adapter->repr_infos[vf_id].vf_rep_eth_dev = vf_rep_eth_dev;
1252 		dcf_adapter->num_reprs++;
1253 	}
1254 
1255 	return ret;
1256 }
1257 
1258 static int
1259 eth_ice_dcf_pci_remove(struct rte_pci_device *pci_dev)
1260 {
1261 	struct rte_eth_dev *eth_dev;
1262 
1263 	eth_dev = rte_eth_dev_allocated(pci_dev->device.name);
1264 	if (!eth_dev)
1265 		return 0;
1266 
1267 	if (eth_dev->data->dev_flags & RTE_ETH_DEV_REPRESENTOR)
1268 		return rte_eth_dev_pci_generic_remove(pci_dev,
1269 						      ice_dcf_vf_repr_uninit);
1270 	else
1271 		return rte_eth_dev_pci_generic_remove(pci_dev,
1272 						      ice_dcf_dev_uninit);
1273 }
1274 
1275 static const struct rte_pci_id pci_id_ice_dcf_map[] = {
1276 	{ RTE_PCI_DEVICE(IAVF_INTEL_VENDOR_ID, IAVF_DEV_ID_ADAPTIVE_VF) },
1277 	{ .vendor_id = 0, /* sentinel */ },
1278 };
1279 
1280 static struct rte_pci_driver rte_ice_dcf_pmd = {
1281 	.id_table = pci_id_ice_dcf_map,
1282 	.drv_flags = RTE_PCI_DRV_NEED_MAPPING,
1283 	.probe = eth_ice_dcf_pci_probe,
1284 	.remove = eth_ice_dcf_pci_remove,
1285 };
1286 
1287 RTE_PMD_REGISTER_PCI(net_ice_dcf, rte_ice_dcf_pmd);
1288 RTE_PMD_REGISTER_PCI_TABLE(net_ice_dcf, pci_id_ice_dcf_map);
1289 RTE_PMD_REGISTER_KMOD_DEP(net_ice_dcf, "* igb_uio | vfio-pci");
1290 RTE_PMD_REGISTER_PARAM_STRING(net_ice_dcf, "cap=dcf");
1291