xref: /dpdk/drivers/net/ice/ice_dcf_ethdev.c (revision bb85a78d)
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((uint32_t)
70 			      ICE_SUPPORT_CHAIN_NUM * rxq->rx_buf_len,
71 			      dev->data->dev_conf.rxmode.max_rx_pkt_len);
72 
73 	/* Check if the jumbo frame and maximum packet length are set
74 	 * correctly.
75 	 */
76 	if (dev->data->dev_conf.rxmode.offloads & DEV_RX_OFFLOAD_JUMBO_FRAME) {
77 		if (max_pkt_len <= ICE_ETH_MAX_LEN ||
78 		    max_pkt_len > ICE_FRAME_SIZE_MAX) {
79 			PMD_DRV_LOG(ERR, "maximum packet length must be "
80 				    "larger than %u and smaller than %u, "
81 				    "as jumbo frame is enabled",
82 				    (uint32_t)ICE_ETH_MAX_LEN,
83 				    (uint32_t)ICE_FRAME_SIZE_MAX);
84 			return -EINVAL;
85 		}
86 	} else {
87 		if (max_pkt_len < RTE_ETHER_MIN_LEN ||
88 		    max_pkt_len > ICE_ETH_MAX_LEN) {
89 			PMD_DRV_LOG(ERR, "maximum packet length must be "
90 				    "larger than %u and smaller than %u, "
91 				    "as jumbo frame is disabled",
92 				    (uint32_t)RTE_ETHER_MIN_LEN,
93 				    (uint32_t)ICE_ETH_MAX_LEN);
94 			return -EINVAL;
95 		}
96 	}
97 
98 	rxq->max_pkt_len = max_pkt_len;
99 	if ((dev_data->dev_conf.rxmode.offloads & DEV_RX_OFFLOAD_SCATTER) ||
100 	    (rxq->max_pkt_len + 2 * ICE_VLAN_TAG_SIZE) > buf_size) {
101 		dev_data->scattered_rx = 1;
102 	}
103 	rxq->qrx_tail = hw->hw_addr + IAVF_QRX_TAIL1(rxq->queue_id);
104 	IAVF_PCI_REG_WRITE(rxq->qrx_tail, rxq->nb_rx_desc - 1);
105 	IAVF_WRITE_FLUSH(hw);
106 
107 	return 0;
108 }
109 
110 static int
111 ice_dcf_init_rx_queues(struct rte_eth_dev *dev)
112 {
113 	struct ice_rx_queue **rxq =
114 		(struct ice_rx_queue **)dev->data->rx_queues;
115 	int i, ret;
116 
117 	for (i = 0; i < dev->data->nb_rx_queues; i++) {
118 		if (!rxq[i] || !rxq[i]->q_set)
119 			continue;
120 		ret = ice_dcf_init_rxq(dev, rxq[i]);
121 		if (ret)
122 			return ret;
123 	}
124 
125 	ice_set_rx_function(dev);
126 	ice_set_tx_function(dev);
127 
128 	return 0;
129 }
130 
131 #define IAVF_MISC_VEC_ID                RTE_INTR_VEC_ZERO_OFFSET
132 #define IAVF_RX_VEC_START               RTE_INTR_VEC_RXTX_OFFSET
133 
134 #define IAVF_ITR_INDEX_DEFAULT          0
135 #define IAVF_QUEUE_ITR_INTERVAL_DEFAULT 32 /* 32 us */
136 #define IAVF_QUEUE_ITR_INTERVAL_MAX     8160 /* 8160 us */
137 
138 static inline uint16_t
139 iavf_calc_itr_interval(int16_t interval)
140 {
141 	if (interval < 0 || interval > IAVF_QUEUE_ITR_INTERVAL_MAX)
142 		interval = IAVF_QUEUE_ITR_INTERVAL_DEFAULT;
143 
144 	/* Convert to hardware count, as writing each 1 represents 2 us */
145 	return interval / 2;
146 }
147 
148 static int
149 ice_dcf_config_rx_queues_irqs(struct rte_eth_dev *dev,
150 				     struct rte_intr_handle *intr_handle)
151 {
152 	struct ice_dcf_adapter *adapter = dev->data->dev_private;
153 	struct ice_dcf_hw *hw = &adapter->real_hw;
154 	uint16_t interval, i;
155 	int vec;
156 
157 	if (rte_intr_cap_multiple(intr_handle) &&
158 	    dev->data->dev_conf.intr_conf.rxq) {
159 		if (rte_intr_efd_enable(intr_handle, dev->data->nb_rx_queues))
160 			return -1;
161 	}
162 
163 	if (rte_intr_dp_is_en(intr_handle) && !intr_handle->intr_vec) {
164 		intr_handle->intr_vec =
165 			rte_zmalloc("intr_vec",
166 				    dev->data->nb_rx_queues * sizeof(int), 0);
167 		if (!intr_handle->intr_vec) {
168 			PMD_DRV_LOG(ERR, "Failed to allocate %d rx intr_vec",
169 				    dev->data->nb_rx_queues);
170 			return -1;
171 		}
172 	}
173 
174 	if (!dev->data->dev_conf.intr_conf.rxq ||
175 	    !rte_intr_dp_is_en(intr_handle)) {
176 		/* Rx interrupt disabled, Map interrupt only for writeback */
177 		hw->nb_msix = 1;
178 		if (hw->vf_res->vf_cap_flags &
179 		    VIRTCHNL_VF_OFFLOAD_WB_ON_ITR) {
180 			/* If WB_ON_ITR supports, enable it */
181 			hw->msix_base = IAVF_RX_VEC_START;
182 			/* Set the ITR for index zero, to 2us to make sure that
183 			 * we leave time for aggregation to occur, but don't
184 			 * increase latency dramatically.
185 			 */
186 			IAVF_WRITE_REG(&hw->avf,
187 				       IAVF_VFINT_DYN_CTLN1(hw->msix_base - 1),
188 				       (0 << IAVF_VFINT_DYN_CTLN1_ITR_INDX_SHIFT) |
189 				       IAVF_VFINT_DYN_CTLN1_WB_ON_ITR_MASK |
190 				       (2UL << IAVF_VFINT_DYN_CTLN1_INTERVAL_SHIFT));
191 		} else {
192 			/* If no WB_ON_ITR offload flags, need to set
193 			 * interrupt for descriptor write back.
194 			 */
195 			hw->msix_base = IAVF_MISC_VEC_ID;
196 
197 			/* set ITR to max */
198 			interval =
199 			iavf_calc_itr_interval(IAVF_QUEUE_ITR_INTERVAL_MAX);
200 			IAVF_WRITE_REG(&hw->avf, IAVF_VFINT_DYN_CTL01,
201 				       IAVF_VFINT_DYN_CTL01_INTENA_MASK |
202 				       (IAVF_ITR_INDEX_DEFAULT <<
203 					IAVF_VFINT_DYN_CTL01_ITR_INDX_SHIFT) |
204 				       (interval <<
205 					IAVF_VFINT_DYN_CTL01_INTERVAL_SHIFT));
206 		}
207 		IAVF_WRITE_FLUSH(&hw->avf);
208 		/* map all queues to the same interrupt */
209 		for (i = 0; i < dev->data->nb_rx_queues; i++)
210 			hw->rxq_map[hw->msix_base] |= 1 << i;
211 	} else {
212 		if (!rte_intr_allow_others(intr_handle)) {
213 			hw->nb_msix = 1;
214 			hw->msix_base = IAVF_MISC_VEC_ID;
215 			for (i = 0; i < dev->data->nb_rx_queues; i++) {
216 				hw->rxq_map[hw->msix_base] |= 1 << i;
217 				intr_handle->intr_vec[i] = IAVF_MISC_VEC_ID;
218 			}
219 			PMD_DRV_LOG(DEBUG,
220 				    "vector %u are mapping to all Rx queues",
221 				    hw->msix_base);
222 		} else {
223 			/* If Rx interrupt is reuquired, and we can use
224 			 * multi interrupts, then the vec is from 1
225 			 */
226 			hw->nb_msix = RTE_MIN(hw->vf_res->max_vectors,
227 					      intr_handle->nb_efd);
228 			hw->msix_base = IAVF_MISC_VEC_ID;
229 			vec = IAVF_MISC_VEC_ID;
230 			for (i = 0; i < dev->data->nb_rx_queues; i++) {
231 				hw->rxq_map[vec] |= 1 << i;
232 				intr_handle->intr_vec[i] = vec++;
233 				if (vec >= hw->nb_msix)
234 					vec = IAVF_RX_VEC_START;
235 			}
236 			PMD_DRV_LOG(DEBUG,
237 				    "%u vectors are mapping to %u Rx queues",
238 				    hw->nb_msix, dev->data->nb_rx_queues);
239 		}
240 	}
241 
242 	if (ice_dcf_config_irq_map(hw)) {
243 		PMD_DRV_LOG(ERR, "config interrupt mapping failed");
244 		return -1;
245 	}
246 	return 0;
247 }
248 
249 static int
250 alloc_rxq_mbufs(struct ice_rx_queue *rxq)
251 {
252 	volatile union ice_rx_flex_desc *rxd;
253 	struct rte_mbuf *mbuf = NULL;
254 	uint64_t dma_addr;
255 	uint16_t i;
256 
257 	for (i = 0; i < rxq->nb_rx_desc; i++) {
258 		mbuf = rte_mbuf_raw_alloc(rxq->mp);
259 		if (unlikely(!mbuf)) {
260 			PMD_DRV_LOG(ERR, "Failed to allocate mbuf for RX");
261 			return -ENOMEM;
262 		}
263 
264 		rte_mbuf_refcnt_set(mbuf, 1);
265 		mbuf->next = NULL;
266 		mbuf->data_off = RTE_PKTMBUF_HEADROOM;
267 		mbuf->nb_segs = 1;
268 		mbuf->port = rxq->port_id;
269 
270 		dma_addr =
271 			rte_cpu_to_le_64(rte_mbuf_data_iova_default(mbuf));
272 
273 		rxd = &rxq->rx_ring[i];
274 		rxd->read.pkt_addr = dma_addr;
275 		rxd->read.hdr_addr = 0;
276 #ifndef RTE_LIBRTE_ICE_16BYTE_RX_DESC
277 		rxd->read.rsvd1 = 0;
278 		rxd->read.rsvd2 = 0;
279 #endif
280 
281 		rxq->sw_ring[i].mbuf = (void *)mbuf;
282 	}
283 
284 	return 0;
285 }
286 
287 static int
288 ice_dcf_rx_queue_start(struct rte_eth_dev *dev, uint16_t rx_queue_id)
289 {
290 	struct ice_dcf_adapter *ad = dev->data->dev_private;
291 	struct iavf_hw *hw = &ad->real_hw.avf;
292 	struct ice_rx_queue *rxq;
293 	int err = 0;
294 
295 	if (rx_queue_id >= dev->data->nb_rx_queues)
296 		return -EINVAL;
297 
298 	rxq = dev->data->rx_queues[rx_queue_id];
299 
300 	err = alloc_rxq_mbufs(rxq);
301 	if (err) {
302 		PMD_DRV_LOG(ERR, "Failed to allocate RX queue mbuf");
303 		return err;
304 	}
305 
306 	rte_wmb();
307 
308 	/* Init the RX tail register. */
309 	IAVF_PCI_REG_WRITE(rxq->qrx_tail, rxq->nb_rx_desc - 1);
310 	IAVF_WRITE_FLUSH(hw);
311 
312 	/* Ready to switch the queue on */
313 	err = ice_dcf_switch_queue(&ad->real_hw, rx_queue_id, true, true);
314 	if (err) {
315 		PMD_DRV_LOG(ERR, "Failed to switch RX queue %u on",
316 			    rx_queue_id);
317 		return err;
318 	}
319 
320 	dev->data->rx_queue_state[rx_queue_id] = RTE_ETH_QUEUE_STATE_STARTED;
321 
322 	return 0;
323 }
324 
325 static inline void
326 reset_rx_queue(struct ice_rx_queue *rxq)
327 {
328 	uint16_t len;
329 	uint32_t i;
330 
331 	if (!rxq)
332 		return;
333 
334 	len = rxq->nb_rx_desc + ICE_RX_MAX_BURST;
335 
336 	for (i = 0; i < len * sizeof(union ice_rx_flex_desc); i++)
337 		((volatile char *)rxq->rx_ring)[i] = 0;
338 
339 	memset(&rxq->fake_mbuf, 0x0, sizeof(rxq->fake_mbuf));
340 
341 	for (i = 0; i < ICE_RX_MAX_BURST; i++)
342 		rxq->sw_ring[rxq->nb_rx_desc + i].mbuf = &rxq->fake_mbuf;
343 
344 	/* for rx bulk */
345 	rxq->rx_nb_avail = 0;
346 	rxq->rx_next_avail = 0;
347 	rxq->rx_free_trigger = (uint16_t)(rxq->rx_free_thresh - 1);
348 
349 	rxq->rx_tail = 0;
350 	rxq->nb_rx_hold = 0;
351 	rxq->pkt_first_seg = NULL;
352 	rxq->pkt_last_seg = NULL;
353 }
354 
355 static inline void
356 reset_tx_queue(struct ice_tx_queue *txq)
357 {
358 	struct ice_tx_entry *txe;
359 	uint32_t i, size;
360 	uint16_t prev;
361 
362 	if (!txq) {
363 		PMD_DRV_LOG(DEBUG, "Pointer to txq is NULL");
364 		return;
365 	}
366 
367 	txe = txq->sw_ring;
368 	size = sizeof(struct ice_tx_desc) * txq->nb_tx_desc;
369 	for (i = 0; i < size; i++)
370 		((volatile char *)txq->tx_ring)[i] = 0;
371 
372 	prev = (uint16_t)(txq->nb_tx_desc - 1);
373 	for (i = 0; i < txq->nb_tx_desc; i++) {
374 		txq->tx_ring[i].cmd_type_offset_bsz =
375 			rte_cpu_to_le_64(IAVF_TX_DESC_DTYPE_DESC_DONE);
376 		txe[i].mbuf =  NULL;
377 		txe[i].last_id = i;
378 		txe[prev].next_id = i;
379 		prev = i;
380 	}
381 
382 	txq->tx_tail = 0;
383 	txq->nb_tx_used = 0;
384 
385 	txq->last_desc_cleaned = txq->nb_tx_desc - 1;
386 	txq->nb_tx_free = txq->nb_tx_desc - 1;
387 
388 	txq->tx_next_dd = txq->tx_rs_thresh - 1;
389 	txq->tx_next_rs = txq->tx_rs_thresh - 1;
390 }
391 
392 static int
393 ice_dcf_rx_queue_stop(struct rte_eth_dev *dev, uint16_t rx_queue_id)
394 {
395 	struct ice_dcf_adapter *ad = dev->data->dev_private;
396 	struct ice_dcf_hw *hw = &ad->real_hw;
397 	struct ice_rx_queue *rxq;
398 	int err;
399 
400 	if (rx_queue_id >= dev->data->nb_rx_queues)
401 		return -EINVAL;
402 
403 	err = ice_dcf_switch_queue(hw, rx_queue_id, true, false);
404 	if (err) {
405 		PMD_DRV_LOG(ERR, "Failed to switch RX queue %u off",
406 			    rx_queue_id);
407 		return err;
408 	}
409 
410 	rxq = dev->data->rx_queues[rx_queue_id];
411 	rxq->rx_rel_mbufs(rxq);
412 	reset_rx_queue(rxq);
413 	dev->data->rx_queue_state[rx_queue_id] = RTE_ETH_QUEUE_STATE_STOPPED;
414 
415 	return 0;
416 }
417 
418 static int
419 ice_dcf_tx_queue_start(struct rte_eth_dev *dev, uint16_t tx_queue_id)
420 {
421 	struct ice_dcf_adapter *ad = dev->data->dev_private;
422 	struct iavf_hw *hw = &ad->real_hw.avf;
423 	struct ice_tx_queue *txq;
424 	int err = 0;
425 
426 	if (tx_queue_id >= dev->data->nb_tx_queues)
427 		return -EINVAL;
428 
429 	txq = dev->data->tx_queues[tx_queue_id];
430 
431 	/* Init the RX tail register. */
432 	txq->qtx_tail = hw->hw_addr + IAVF_QTX_TAIL1(tx_queue_id);
433 	IAVF_PCI_REG_WRITE(txq->qtx_tail, 0);
434 	IAVF_WRITE_FLUSH(hw);
435 
436 	/* Ready to switch the queue on */
437 	err = ice_dcf_switch_queue(&ad->real_hw, tx_queue_id, false, true);
438 
439 	if (err) {
440 		PMD_DRV_LOG(ERR, "Failed to switch TX queue %u on",
441 			    tx_queue_id);
442 		return err;
443 	}
444 
445 	dev->data->tx_queue_state[tx_queue_id] = RTE_ETH_QUEUE_STATE_STARTED;
446 
447 	return 0;
448 }
449 
450 static int
451 ice_dcf_tx_queue_stop(struct rte_eth_dev *dev, uint16_t tx_queue_id)
452 {
453 	struct ice_dcf_adapter *ad = dev->data->dev_private;
454 	struct ice_dcf_hw *hw = &ad->real_hw;
455 	struct ice_tx_queue *txq;
456 	int err;
457 
458 	if (tx_queue_id >= dev->data->nb_tx_queues)
459 		return -EINVAL;
460 
461 	err = ice_dcf_switch_queue(hw, tx_queue_id, false, false);
462 	if (err) {
463 		PMD_DRV_LOG(ERR, "Failed to switch TX queue %u off",
464 			    tx_queue_id);
465 		return err;
466 	}
467 
468 	txq = dev->data->tx_queues[tx_queue_id];
469 	txq->tx_rel_mbufs(txq);
470 	reset_tx_queue(txq);
471 	dev->data->tx_queue_state[tx_queue_id] = RTE_ETH_QUEUE_STATE_STOPPED;
472 
473 	return 0;
474 }
475 
476 static int
477 ice_dcf_start_queues(struct rte_eth_dev *dev)
478 {
479 	struct ice_rx_queue *rxq;
480 	struct ice_tx_queue *txq;
481 	int nb_rxq = 0;
482 	int nb_txq, i;
483 
484 	for (nb_txq = 0; nb_txq < dev->data->nb_tx_queues; nb_txq++) {
485 		txq = dev->data->tx_queues[nb_txq];
486 		if (txq->tx_deferred_start)
487 			continue;
488 		if (ice_dcf_tx_queue_start(dev, nb_txq) != 0) {
489 			PMD_DRV_LOG(ERR, "Fail to start queue %u", nb_txq);
490 			goto tx_err;
491 		}
492 	}
493 
494 	for (nb_rxq = 0; nb_rxq < dev->data->nb_rx_queues; nb_rxq++) {
495 		rxq = dev->data->rx_queues[nb_rxq];
496 		if (rxq->rx_deferred_start)
497 			continue;
498 		if (ice_dcf_rx_queue_start(dev, nb_rxq) != 0) {
499 			PMD_DRV_LOG(ERR, "Fail to start queue %u", nb_rxq);
500 			goto rx_err;
501 		}
502 	}
503 
504 	return 0;
505 
506 	/* stop the started queues if failed to start all queues */
507 rx_err:
508 	for (i = 0; i < nb_rxq; i++)
509 		ice_dcf_rx_queue_stop(dev, i);
510 tx_err:
511 	for (i = 0; i < nb_txq; i++)
512 		ice_dcf_tx_queue_stop(dev, i);
513 
514 	return -1;
515 }
516 
517 static int
518 ice_dcf_dev_start(struct rte_eth_dev *dev)
519 {
520 	struct ice_dcf_adapter *dcf_ad = dev->data->dev_private;
521 	struct rte_intr_handle *intr_handle = dev->intr_handle;
522 	struct ice_adapter *ad = &dcf_ad->parent;
523 	struct ice_dcf_hw *hw = &dcf_ad->real_hw;
524 	int ret;
525 
526 	if (hw->resetting) {
527 		PMD_DRV_LOG(ERR,
528 			    "The DCF has been reset by PF, please reinit first");
529 		return -EIO;
530 	}
531 
532 	ad->pf.adapter_stopped = 0;
533 
534 	hw->num_queue_pairs = RTE_MAX(dev->data->nb_rx_queues,
535 				      dev->data->nb_tx_queues);
536 
537 	ret = ice_dcf_init_rx_queues(dev);
538 	if (ret) {
539 		PMD_DRV_LOG(ERR, "Fail to init queues");
540 		return ret;
541 	}
542 
543 	if (hw->vf_res->vf_cap_flags & VIRTCHNL_VF_OFFLOAD_RSS_PF) {
544 		ret = ice_dcf_init_rss(hw);
545 		if (ret) {
546 			PMD_DRV_LOG(ERR, "Failed to configure RSS");
547 			return ret;
548 		}
549 	}
550 
551 	ret = ice_dcf_configure_queues(hw);
552 	if (ret) {
553 		PMD_DRV_LOG(ERR, "Fail to config queues");
554 		return ret;
555 	}
556 
557 	ret = ice_dcf_config_rx_queues_irqs(dev, intr_handle);
558 	if (ret) {
559 		PMD_DRV_LOG(ERR, "Fail to config rx queues' irqs");
560 		return ret;
561 	}
562 
563 	if (dev->data->dev_conf.intr_conf.rxq != 0) {
564 		rte_intr_disable(intr_handle);
565 		rte_intr_enable(intr_handle);
566 	}
567 
568 	ret = ice_dcf_start_queues(dev);
569 	if (ret) {
570 		PMD_DRV_LOG(ERR, "Failed to enable queues");
571 		return ret;
572 	}
573 
574 	ret = ice_dcf_add_del_all_mac_addr(hw, true);
575 	if (ret) {
576 		PMD_DRV_LOG(ERR, "Failed to add mac addr");
577 		return ret;
578 	}
579 
580 	dev->data->dev_link.link_status = ETH_LINK_UP;
581 
582 	return 0;
583 }
584 
585 static void
586 ice_dcf_stop_queues(struct rte_eth_dev *dev)
587 {
588 	struct ice_dcf_adapter *ad = dev->data->dev_private;
589 	struct ice_dcf_hw *hw = &ad->real_hw;
590 	struct ice_rx_queue *rxq;
591 	struct ice_tx_queue *txq;
592 	int ret, i;
593 
594 	/* Stop All queues */
595 	ret = ice_dcf_disable_queues(hw);
596 	if (ret)
597 		PMD_DRV_LOG(WARNING, "Fail to stop queues");
598 
599 	for (i = 0; i < dev->data->nb_tx_queues; i++) {
600 		txq = dev->data->tx_queues[i];
601 		if (!txq)
602 			continue;
603 		txq->tx_rel_mbufs(txq);
604 		reset_tx_queue(txq);
605 		dev->data->tx_queue_state[i] = RTE_ETH_QUEUE_STATE_STOPPED;
606 		dev->data->tx_queues[i] = NULL;
607 	}
608 	for (i = 0; i < dev->data->nb_rx_queues; i++) {
609 		rxq = dev->data->rx_queues[i];
610 		if (!rxq)
611 			continue;
612 		rxq->rx_rel_mbufs(rxq);
613 		reset_rx_queue(rxq);
614 		dev->data->rx_queue_state[i] = RTE_ETH_QUEUE_STATE_STOPPED;
615 		dev->data->rx_queues[i] = NULL;
616 	}
617 }
618 
619 static int
620 ice_dcf_dev_stop(struct rte_eth_dev *dev)
621 {
622 	struct ice_dcf_adapter *dcf_ad = dev->data->dev_private;
623 	struct rte_intr_handle *intr_handle = dev->intr_handle;
624 	struct ice_adapter *ad = &dcf_ad->parent;
625 
626 	if (ad->pf.adapter_stopped == 1) {
627 		PMD_DRV_LOG(DEBUG, "Port is already stopped");
628 		return 0;
629 	}
630 
631 	/* Stop the VF representors for this device */
632 	ice_dcf_vf_repr_stop_all(dcf_ad);
633 
634 	ice_dcf_stop_queues(dev);
635 
636 	rte_intr_efd_disable(intr_handle);
637 	if (intr_handle->intr_vec) {
638 		rte_free(intr_handle->intr_vec);
639 		intr_handle->intr_vec = NULL;
640 	}
641 
642 	ice_dcf_add_del_all_mac_addr(&dcf_ad->real_hw, false);
643 	dev->data->dev_link.link_status = ETH_LINK_DOWN;
644 	ad->pf.adapter_stopped = 1;
645 
646 	return 0;
647 }
648 
649 static int
650 ice_dcf_dev_configure(struct rte_eth_dev *dev)
651 {
652 	struct ice_dcf_adapter *dcf_ad = dev->data->dev_private;
653 	struct ice_adapter *ad = &dcf_ad->parent;
654 
655 	ad->rx_bulk_alloc_allowed = true;
656 	ad->tx_simple_allowed = true;
657 
658 	if (dev->data->dev_conf.rxmode.mq_mode & ETH_MQ_RX_RSS_FLAG)
659 		dev->data->dev_conf.rxmode.offloads |= DEV_RX_OFFLOAD_RSS_HASH;
660 
661 	return 0;
662 }
663 
664 static int
665 ice_dcf_dev_info_get(struct rte_eth_dev *dev,
666 		     struct rte_eth_dev_info *dev_info)
667 {
668 	struct ice_dcf_adapter *adapter = dev->data->dev_private;
669 	struct ice_dcf_hw *hw = &adapter->real_hw;
670 
671 	dev_info->max_mac_addrs = 1;
672 	dev_info->max_rx_queues = hw->vsi_res->num_queue_pairs;
673 	dev_info->max_tx_queues = hw->vsi_res->num_queue_pairs;
674 	dev_info->min_rx_bufsize = ICE_BUF_SIZE_MIN;
675 	dev_info->max_rx_pktlen = ICE_FRAME_SIZE_MAX;
676 	dev_info->hash_key_size = hw->vf_res->rss_key_size;
677 	dev_info->reta_size = hw->vf_res->rss_lut_size;
678 	dev_info->flow_type_rss_offloads = ICE_RSS_OFFLOAD_ALL;
679 
680 	dev_info->rx_offload_capa =
681 		DEV_RX_OFFLOAD_VLAN_STRIP |
682 		DEV_RX_OFFLOAD_IPV4_CKSUM |
683 		DEV_RX_OFFLOAD_UDP_CKSUM |
684 		DEV_RX_OFFLOAD_TCP_CKSUM |
685 		DEV_RX_OFFLOAD_OUTER_IPV4_CKSUM |
686 		DEV_RX_OFFLOAD_SCATTER |
687 		DEV_RX_OFFLOAD_JUMBO_FRAME |
688 		DEV_RX_OFFLOAD_VLAN_FILTER |
689 		DEV_RX_OFFLOAD_RSS_HASH;
690 	dev_info->tx_offload_capa =
691 		DEV_TX_OFFLOAD_VLAN_INSERT |
692 		DEV_TX_OFFLOAD_IPV4_CKSUM |
693 		DEV_TX_OFFLOAD_UDP_CKSUM |
694 		DEV_TX_OFFLOAD_TCP_CKSUM |
695 		DEV_TX_OFFLOAD_SCTP_CKSUM |
696 		DEV_TX_OFFLOAD_OUTER_IPV4_CKSUM |
697 		DEV_TX_OFFLOAD_TCP_TSO |
698 		DEV_TX_OFFLOAD_VXLAN_TNL_TSO |
699 		DEV_TX_OFFLOAD_GRE_TNL_TSO |
700 		DEV_TX_OFFLOAD_IPIP_TNL_TSO |
701 		DEV_TX_OFFLOAD_GENEVE_TNL_TSO |
702 		DEV_TX_OFFLOAD_MULTI_SEGS;
703 
704 	dev_info->default_rxconf = (struct rte_eth_rxconf) {
705 		.rx_thresh = {
706 			.pthresh = ICE_DEFAULT_RX_PTHRESH,
707 			.hthresh = ICE_DEFAULT_RX_HTHRESH,
708 			.wthresh = ICE_DEFAULT_RX_WTHRESH,
709 		},
710 		.rx_free_thresh = ICE_DEFAULT_RX_FREE_THRESH,
711 		.rx_drop_en = 0,
712 		.offloads = 0,
713 	};
714 
715 	dev_info->default_txconf = (struct rte_eth_txconf) {
716 		.tx_thresh = {
717 			.pthresh = ICE_DEFAULT_TX_PTHRESH,
718 			.hthresh = ICE_DEFAULT_TX_HTHRESH,
719 			.wthresh = ICE_DEFAULT_TX_WTHRESH,
720 		},
721 		.tx_free_thresh = ICE_DEFAULT_TX_FREE_THRESH,
722 		.tx_rs_thresh = ICE_DEFAULT_TX_RSBIT_THRESH,
723 		.offloads = 0,
724 	};
725 
726 	dev_info->rx_desc_lim = (struct rte_eth_desc_lim) {
727 		.nb_max = ICE_MAX_RING_DESC,
728 		.nb_min = ICE_MIN_RING_DESC,
729 		.nb_align = ICE_ALIGN_RING_DESC,
730 	};
731 
732 	dev_info->tx_desc_lim = (struct rte_eth_desc_lim) {
733 		.nb_max = ICE_MAX_RING_DESC,
734 		.nb_min = ICE_MIN_RING_DESC,
735 		.nb_align = ICE_ALIGN_RING_DESC,
736 	};
737 
738 	return 0;
739 }
740 
741 static int
742 ice_dcf_dev_promiscuous_enable(__rte_unused struct rte_eth_dev *dev)
743 {
744 	return 0;
745 }
746 
747 static int
748 ice_dcf_dev_promiscuous_disable(__rte_unused struct rte_eth_dev *dev)
749 {
750 	return 0;
751 }
752 
753 static int
754 ice_dcf_dev_allmulticast_enable(__rte_unused struct rte_eth_dev *dev)
755 {
756 	return 0;
757 }
758 
759 static int
760 ice_dcf_dev_allmulticast_disable(__rte_unused struct rte_eth_dev *dev)
761 {
762 	return 0;
763 }
764 
765 static int
766 ice_dcf_dev_flow_ops_get(struct rte_eth_dev *dev,
767 			 const struct rte_flow_ops **ops)
768 {
769 	if (!dev)
770 		return -EINVAL;
771 
772 	*ops = &ice_flow_ops;
773 	return 0;
774 }
775 
776 #define ICE_DCF_32_BIT_WIDTH (CHAR_BIT * 4)
777 #define ICE_DCF_48_BIT_WIDTH (CHAR_BIT * 6)
778 #define ICE_DCF_48_BIT_MASK  RTE_LEN2MASK(ICE_DCF_48_BIT_WIDTH, uint64_t)
779 
780 static void
781 ice_dcf_stat_update_48(uint64_t *offset, uint64_t *stat)
782 {
783 	if (*stat >= *offset)
784 		*stat = *stat - *offset;
785 	else
786 		*stat = (uint64_t)((*stat +
787 			((uint64_t)1 << ICE_DCF_48_BIT_WIDTH)) - *offset);
788 
789 	*stat &= ICE_DCF_48_BIT_MASK;
790 }
791 
792 static void
793 ice_dcf_stat_update_32(uint64_t *offset, uint64_t *stat)
794 {
795 	if (*stat >= *offset)
796 		*stat = (uint64_t)(*stat - *offset);
797 	else
798 		*stat = (uint64_t)((*stat +
799 			((uint64_t)1 << ICE_DCF_32_BIT_WIDTH)) - *offset);
800 }
801 
802 static void
803 ice_dcf_update_stats(struct virtchnl_eth_stats *oes,
804 		     struct virtchnl_eth_stats *nes)
805 {
806 	ice_dcf_stat_update_48(&oes->rx_bytes, &nes->rx_bytes);
807 	ice_dcf_stat_update_48(&oes->rx_unicast, &nes->rx_unicast);
808 	ice_dcf_stat_update_48(&oes->rx_multicast, &nes->rx_multicast);
809 	ice_dcf_stat_update_48(&oes->rx_broadcast, &nes->rx_broadcast);
810 	ice_dcf_stat_update_32(&oes->rx_discards, &nes->rx_discards);
811 	ice_dcf_stat_update_48(&oes->tx_bytes, &nes->tx_bytes);
812 	ice_dcf_stat_update_48(&oes->tx_unicast, &nes->tx_unicast);
813 	ice_dcf_stat_update_48(&oes->tx_multicast, &nes->tx_multicast);
814 	ice_dcf_stat_update_48(&oes->tx_broadcast, &nes->tx_broadcast);
815 	ice_dcf_stat_update_32(&oes->tx_errors, &nes->tx_errors);
816 	ice_dcf_stat_update_32(&oes->tx_discards, &nes->tx_discards);
817 }
818 
819 
820 static int
821 ice_dcf_stats_get(struct rte_eth_dev *dev, struct rte_eth_stats *stats)
822 {
823 	struct ice_dcf_adapter *ad = dev->data->dev_private;
824 	struct ice_dcf_hw *hw = &ad->real_hw;
825 	struct virtchnl_eth_stats pstats;
826 	int ret;
827 
828 	if (hw->resetting) {
829 		PMD_DRV_LOG(ERR,
830 			    "The DCF has been reset by PF, please reinit first");
831 		return -EIO;
832 	}
833 
834 	ret = ice_dcf_query_stats(hw, &pstats);
835 	if (ret == 0) {
836 		ice_dcf_update_stats(&hw->eth_stats_offset, &pstats);
837 		stats->ipackets = pstats.rx_unicast + pstats.rx_multicast +
838 				pstats.rx_broadcast - pstats.rx_discards;
839 		stats->opackets = pstats.tx_broadcast + pstats.tx_multicast +
840 						pstats.tx_unicast;
841 		stats->imissed = pstats.rx_discards;
842 		stats->oerrors = pstats.tx_errors + pstats.tx_discards;
843 		stats->ibytes = pstats.rx_bytes;
844 		stats->ibytes -= stats->ipackets * RTE_ETHER_CRC_LEN;
845 		stats->obytes = pstats.tx_bytes;
846 	} else {
847 		PMD_DRV_LOG(ERR, "Get statistics failed");
848 	}
849 	return ret;
850 }
851 
852 static int
853 ice_dcf_stats_reset(struct rte_eth_dev *dev)
854 {
855 	struct ice_dcf_adapter *ad = dev->data->dev_private;
856 	struct ice_dcf_hw *hw = &ad->real_hw;
857 	struct virtchnl_eth_stats pstats;
858 	int ret;
859 
860 	if (hw->resetting)
861 		return 0;
862 
863 	/* read stat values to clear hardware registers */
864 	ret = ice_dcf_query_stats(hw, &pstats);
865 	if (ret != 0)
866 		return ret;
867 
868 	/* set stats offset base on current values */
869 	hw->eth_stats_offset = pstats;
870 
871 	return 0;
872 }
873 
874 static void
875 ice_dcf_free_repr_info(struct ice_dcf_adapter *dcf_adapter)
876 {
877 	if (dcf_adapter->repr_infos) {
878 		rte_free(dcf_adapter->repr_infos);
879 		dcf_adapter->repr_infos = NULL;
880 	}
881 }
882 
883 static int
884 ice_dcf_init_repr_info(struct ice_dcf_adapter *dcf_adapter)
885 {
886 	dcf_adapter->repr_infos =
887 			rte_calloc("ice_dcf_rep_info",
888 				   dcf_adapter->real_hw.num_vfs,
889 				   sizeof(dcf_adapter->repr_infos[0]), 0);
890 	if (!dcf_adapter->repr_infos) {
891 		PMD_DRV_LOG(ERR, "Failed to alloc memory for VF representors\n");
892 		return -ENOMEM;
893 	}
894 
895 	return 0;
896 }
897 
898 static int
899 ice_dcf_dev_close(struct rte_eth_dev *dev)
900 {
901 	struct ice_dcf_adapter *adapter = dev->data->dev_private;
902 
903 	if (rte_eal_process_type() != RTE_PROC_PRIMARY)
904 		return 0;
905 
906 	(void)ice_dcf_dev_stop(dev);
907 
908 	ice_dcf_free_repr_info(adapter);
909 	ice_dcf_uninit_parent_adapter(dev);
910 	ice_dcf_uninit_hw(dev, &adapter->real_hw);
911 
912 	return 0;
913 }
914 
915 int
916 ice_dcf_link_update(struct rte_eth_dev *dev,
917 		    __rte_unused int wait_to_complete)
918 {
919 	struct ice_dcf_adapter *ad = dev->data->dev_private;
920 	struct ice_dcf_hw *hw = &ad->real_hw;
921 	struct rte_eth_link new_link;
922 
923 	memset(&new_link, 0, sizeof(new_link));
924 
925 	/* Only read status info stored in VF, and the info is updated
926 	 * when receive LINK_CHANGE event from PF by virtchnl.
927 	 */
928 	switch (hw->link_speed) {
929 	case 10:
930 		new_link.link_speed = ETH_SPEED_NUM_10M;
931 		break;
932 	case 100:
933 		new_link.link_speed = ETH_SPEED_NUM_100M;
934 		break;
935 	case 1000:
936 		new_link.link_speed = ETH_SPEED_NUM_1G;
937 		break;
938 	case 10000:
939 		new_link.link_speed = ETH_SPEED_NUM_10G;
940 		break;
941 	case 20000:
942 		new_link.link_speed = ETH_SPEED_NUM_20G;
943 		break;
944 	case 25000:
945 		new_link.link_speed = ETH_SPEED_NUM_25G;
946 		break;
947 	case 40000:
948 		new_link.link_speed = ETH_SPEED_NUM_40G;
949 		break;
950 	case 50000:
951 		new_link.link_speed = ETH_SPEED_NUM_50G;
952 		break;
953 	case 100000:
954 		new_link.link_speed = ETH_SPEED_NUM_100G;
955 		break;
956 	default:
957 		new_link.link_speed = ETH_SPEED_NUM_NONE;
958 		break;
959 	}
960 
961 	new_link.link_duplex = ETH_LINK_FULL_DUPLEX;
962 	new_link.link_status = hw->link_up ? ETH_LINK_UP :
963 					     ETH_LINK_DOWN;
964 	new_link.link_autoneg = !(dev->data->dev_conf.link_speeds &
965 				ETH_LINK_SPEED_FIXED);
966 
967 	return rte_eth_linkstatus_set(dev, &new_link);
968 }
969 
970 /* Add UDP tunneling port */
971 static int
972 ice_dcf_dev_udp_tunnel_port_add(struct rte_eth_dev *dev,
973 				struct rte_eth_udp_tunnel *udp_tunnel)
974 {
975 	struct ice_dcf_adapter *adapter = dev->data->dev_private;
976 	struct ice_adapter *parent_adapter = &adapter->parent;
977 	struct ice_hw *parent_hw = &parent_adapter->hw;
978 	int ret = 0;
979 
980 	if (!udp_tunnel)
981 		return -EINVAL;
982 
983 	switch (udp_tunnel->prot_type) {
984 	case RTE_TUNNEL_TYPE_VXLAN:
985 		ret = ice_create_tunnel(parent_hw, TNL_VXLAN,
986 					udp_tunnel->udp_port);
987 		break;
988 	case RTE_TUNNEL_TYPE_ECPRI:
989 		ret = ice_create_tunnel(parent_hw, TNL_ECPRI,
990 					udp_tunnel->udp_port);
991 		break;
992 	default:
993 		PMD_DRV_LOG(ERR, "Invalid tunnel type");
994 		ret = -EINVAL;
995 		break;
996 	}
997 
998 	return ret;
999 }
1000 
1001 /* Delete UDP tunneling port */
1002 static int
1003 ice_dcf_dev_udp_tunnel_port_del(struct rte_eth_dev *dev,
1004 				struct rte_eth_udp_tunnel *udp_tunnel)
1005 {
1006 	struct ice_dcf_adapter *adapter = dev->data->dev_private;
1007 	struct ice_adapter *parent_adapter = &adapter->parent;
1008 	struct ice_hw *parent_hw = &parent_adapter->hw;
1009 	int ret = 0;
1010 
1011 	if (!udp_tunnel)
1012 		return -EINVAL;
1013 
1014 	switch (udp_tunnel->prot_type) {
1015 	case RTE_TUNNEL_TYPE_VXLAN:
1016 	case RTE_TUNNEL_TYPE_ECPRI:
1017 		ret = ice_destroy_tunnel(parent_hw, udp_tunnel->udp_port, 0);
1018 		break;
1019 	default:
1020 		PMD_DRV_LOG(ERR, "Invalid tunnel type");
1021 		ret = -EINVAL;
1022 		break;
1023 	}
1024 
1025 	return ret;
1026 }
1027 
1028 static int
1029 ice_dcf_tm_ops_get(struct rte_eth_dev *dev __rte_unused,
1030 		void *arg)
1031 {
1032 	if (!arg)
1033 		return -EINVAL;
1034 
1035 	*(const void **)arg = &ice_dcf_tm_ops;
1036 
1037 	return 0;
1038 }
1039 
1040 static int
1041 ice_dcf_dev_reset(struct rte_eth_dev *dev)
1042 {
1043 	int ret;
1044 
1045 	ret = ice_dcf_dev_uninit(dev);
1046 	if (ret)
1047 		return ret;
1048 
1049 	ret = ice_dcf_dev_init(dev);
1050 
1051 	return ret;
1052 }
1053 
1054 static const struct eth_dev_ops ice_dcf_eth_dev_ops = {
1055 	.dev_start               = ice_dcf_dev_start,
1056 	.dev_stop                = ice_dcf_dev_stop,
1057 	.dev_close               = ice_dcf_dev_close,
1058 	.dev_reset               = ice_dcf_dev_reset,
1059 	.dev_configure           = ice_dcf_dev_configure,
1060 	.dev_infos_get           = ice_dcf_dev_info_get,
1061 	.rx_queue_setup          = ice_rx_queue_setup,
1062 	.tx_queue_setup          = ice_tx_queue_setup,
1063 	.rx_queue_release        = ice_dev_rx_queue_release,
1064 	.tx_queue_release        = ice_dev_tx_queue_release,
1065 	.rx_queue_start          = ice_dcf_rx_queue_start,
1066 	.tx_queue_start          = ice_dcf_tx_queue_start,
1067 	.rx_queue_stop           = ice_dcf_rx_queue_stop,
1068 	.tx_queue_stop           = ice_dcf_tx_queue_stop,
1069 	.link_update             = ice_dcf_link_update,
1070 	.stats_get               = ice_dcf_stats_get,
1071 	.stats_reset             = ice_dcf_stats_reset,
1072 	.promiscuous_enable      = ice_dcf_dev_promiscuous_enable,
1073 	.promiscuous_disable     = ice_dcf_dev_promiscuous_disable,
1074 	.allmulticast_enable     = ice_dcf_dev_allmulticast_enable,
1075 	.allmulticast_disable    = ice_dcf_dev_allmulticast_disable,
1076 	.flow_ops_get            = ice_dcf_dev_flow_ops_get,
1077 	.udp_tunnel_port_add	 = ice_dcf_dev_udp_tunnel_port_add,
1078 	.udp_tunnel_port_del	 = ice_dcf_dev_udp_tunnel_port_del,
1079 	.tm_ops_get              = ice_dcf_tm_ops_get,
1080 };
1081 
1082 static int
1083 ice_dcf_dev_init(struct rte_eth_dev *eth_dev)
1084 {
1085 	struct ice_dcf_adapter *adapter = eth_dev->data->dev_private;
1086 
1087 	adapter->real_hw.resetting = false;
1088 	eth_dev->dev_ops = &ice_dcf_eth_dev_ops;
1089 	eth_dev->rx_pkt_burst = ice_dcf_recv_pkts;
1090 	eth_dev->tx_pkt_burst = ice_dcf_xmit_pkts;
1091 
1092 	if (rte_eal_process_type() != RTE_PROC_PRIMARY)
1093 		return 0;
1094 
1095 	eth_dev->data->dev_flags |= RTE_ETH_DEV_AUTOFILL_QUEUE_XSTATS;
1096 
1097 	adapter->real_hw.vc_event_msg_cb = ice_dcf_handle_pf_event_msg;
1098 	if (ice_dcf_init_hw(eth_dev, &adapter->real_hw) != 0) {
1099 		PMD_INIT_LOG(ERR, "Failed to init DCF hardware");
1100 		return -1;
1101 	}
1102 
1103 	if (ice_dcf_init_parent_adapter(eth_dev) != 0) {
1104 		PMD_INIT_LOG(ERR, "Failed to init DCF parent adapter");
1105 		ice_dcf_uninit_hw(eth_dev, &adapter->real_hw);
1106 		return -1;
1107 	}
1108 
1109 	return 0;
1110 }
1111 
1112 static int
1113 ice_dcf_dev_uninit(struct rte_eth_dev *eth_dev)
1114 {
1115 	ice_dcf_dev_close(eth_dev);
1116 
1117 	return 0;
1118 }
1119 
1120 static int
1121 ice_dcf_cap_check_handler(__rte_unused const char *key,
1122 			  const char *value, __rte_unused void *opaque)
1123 {
1124 	if (strcmp(value, "dcf"))
1125 		return -1;
1126 
1127 	return 0;
1128 }
1129 
1130 static int
1131 ice_dcf_cap_selected(struct rte_devargs *devargs)
1132 {
1133 	struct rte_kvargs *kvlist;
1134 	const char *key = "cap";
1135 	int ret = 0;
1136 
1137 	if (devargs == NULL)
1138 		return 0;
1139 
1140 	kvlist = rte_kvargs_parse(devargs->args, NULL);
1141 	if (kvlist == NULL)
1142 		return 0;
1143 
1144 	if (!rte_kvargs_count(kvlist, key))
1145 		goto exit;
1146 
1147 	/* dcf capability selected when there's a key-value pair: cap=dcf */
1148 	if (rte_kvargs_process(kvlist, key,
1149 			       ice_dcf_cap_check_handler, NULL) < 0)
1150 		goto exit;
1151 
1152 	ret = 1;
1153 
1154 exit:
1155 	rte_kvargs_free(kvlist);
1156 	return ret;
1157 }
1158 
1159 static int
1160 eth_ice_dcf_pci_probe(__rte_unused struct rte_pci_driver *pci_drv,
1161 		      struct rte_pci_device *pci_dev)
1162 {
1163 	struct rte_eth_devargs eth_da = { .nb_representor_ports = 0 };
1164 	struct ice_dcf_vf_repr_param repr_param;
1165 	char repr_name[RTE_ETH_NAME_MAX_LEN];
1166 	struct ice_dcf_adapter *dcf_adapter;
1167 	struct rte_eth_dev *dcf_ethdev;
1168 	uint16_t dcf_vsi_id;
1169 	int i, ret;
1170 
1171 	if (!ice_dcf_cap_selected(pci_dev->device.devargs))
1172 		return 1;
1173 
1174 	ret = rte_eth_devargs_parse(pci_dev->device.devargs->args, &eth_da);
1175 	if (ret)
1176 		return ret;
1177 
1178 	ret = rte_eth_dev_pci_generic_probe(pci_dev,
1179 					    sizeof(struct ice_dcf_adapter),
1180 					    ice_dcf_dev_init);
1181 	if (ret || !eth_da.nb_representor_ports)
1182 		return ret;
1183 	if (eth_da.type != RTE_ETH_REPRESENTOR_VF)
1184 		return -ENOTSUP;
1185 
1186 	dcf_ethdev = rte_eth_dev_allocated(pci_dev->device.name);
1187 	if (dcf_ethdev == NULL)
1188 		return -ENODEV;
1189 
1190 	dcf_adapter = dcf_ethdev->data->dev_private;
1191 	ret = ice_dcf_init_repr_info(dcf_adapter);
1192 	if (ret)
1193 		return ret;
1194 
1195 	if (eth_da.nb_representor_ports > dcf_adapter->real_hw.num_vfs ||
1196 	    eth_da.nb_representor_ports >= RTE_MAX_ETHPORTS) {
1197 		PMD_DRV_LOG(ERR, "the number of port representors is too large: %u",
1198 			    eth_da.nb_representor_ports);
1199 		ice_dcf_free_repr_info(dcf_adapter);
1200 		return -EINVAL;
1201 	}
1202 
1203 	dcf_vsi_id = dcf_adapter->real_hw.vsi_id | VIRTCHNL_DCF_VF_VSI_VALID;
1204 
1205 	repr_param.dcf_eth_dev = dcf_ethdev;
1206 	repr_param.switch_domain_id = 0;
1207 
1208 	for (i = 0; i < eth_da.nb_representor_ports; i++) {
1209 		uint16_t vf_id = eth_da.representor_ports[i];
1210 		struct rte_eth_dev *vf_rep_eth_dev;
1211 
1212 		if (vf_id >= dcf_adapter->real_hw.num_vfs) {
1213 			PMD_DRV_LOG(ERR, "VF ID %u is out of range (0 ~ %u)",
1214 				    vf_id, dcf_adapter->real_hw.num_vfs - 1);
1215 			ret = -EINVAL;
1216 			break;
1217 		}
1218 
1219 		if (dcf_adapter->real_hw.vf_vsi_map[vf_id] == dcf_vsi_id) {
1220 			PMD_DRV_LOG(ERR, "VF ID %u is DCF's ID.\n", vf_id);
1221 			ret = -EINVAL;
1222 			break;
1223 		}
1224 
1225 		repr_param.vf_id = vf_id;
1226 		snprintf(repr_name, sizeof(repr_name), "net_%s_representor_%u",
1227 			 pci_dev->device.name, vf_id);
1228 		ret = rte_eth_dev_create(&pci_dev->device, repr_name,
1229 					 sizeof(struct ice_dcf_vf_repr),
1230 					 NULL, NULL, ice_dcf_vf_repr_init,
1231 					 &repr_param);
1232 		if (ret) {
1233 			PMD_DRV_LOG(ERR, "failed to create DCF VF representor %s",
1234 				    repr_name);
1235 			break;
1236 		}
1237 
1238 		vf_rep_eth_dev = rte_eth_dev_allocated(repr_name);
1239 		if (!vf_rep_eth_dev) {
1240 			PMD_DRV_LOG(ERR,
1241 				    "Failed to find the ethdev for DCF VF representor: %s",
1242 				    repr_name);
1243 			ret = -ENODEV;
1244 			break;
1245 		}
1246 
1247 		dcf_adapter->repr_infos[vf_id].vf_rep_eth_dev = vf_rep_eth_dev;
1248 		dcf_adapter->num_reprs++;
1249 	}
1250 
1251 	return ret;
1252 }
1253 
1254 static int
1255 eth_ice_dcf_pci_remove(struct rte_pci_device *pci_dev)
1256 {
1257 	struct rte_eth_dev *eth_dev;
1258 
1259 	eth_dev = rte_eth_dev_allocated(pci_dev->device.name);
1260 	if (!eth_dev)
1261 		return 0;
1262 
1263 	if (eth_dev->data->dev_flags & RTE_ETH_DEV_REPRESENTOR)
1264 		return rte_eth_dev_pci_generic_remove(pci_dev,
1265 						      ice_dcf_vf_repr_uninit);
1266 	else
1267 		return rte_eth_dev_pci_generic_remove(pci_dev,
1268 						      ice_dcf_dev_uninit);
1269 }
1270 
1271 static const struct rte_pci_id pci_id_ice_dcf_map[] = {
1272 	{ RTE_PCI_DEVICE(IAVF_INTEL_VENDOR_ID, IAVF_DEV_ID_ADAPTIVE_VF) },
1273 	{ .vendor_id = 0, /* sentinel */ },
1274 };
1275 
1276 static struct rte_pci_driver rte_ice_dcf_pmd = {
1277 	.id_table = pci_id_ice_dcf_map,
1278 	.drv_flags = RTE_PCI_DRV_NEED_MAPPING,
1279 	.probe = eth_ice_dcf_pci_probe,
1280 	.remove = eth_ice_dcf_pci_remove,
1281 };
1282 
1283 RTE_PMD_REGISTER_PCI(net_ice_dcf, rte_ice_dcf_pmd);
1284 RTE_PMD_REGISTER_PCI_TABLE(net_ice_dcf, pci_id_ice_dcf_map);
1285 RTE_PMD_REGISTER_KMOD_DEP(net_ice_dcf, "* igb_uio | vfio-pci");
1286 RTE_PMD_REGISTER_PARAM_STRING(net_ice_dcf, "cap=dcf");
1287