xref: /dpdk/drivers/net/ice/ice_dcf_ethdev.c (revision fc0ec740)
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 maximum packet length is set correctly.  */
73 	if (max_pkt_len <= RTE_ETHER_MIN_LEN ||
74 	    max_pkt_len > ICE_FRAME_SIZE_MAX) {
75 		PMD_DRV_LOG(ERR, "maximum packet length must be "
76 			    "larger than %u and smaller than %u",
77 			    (uint32_t)RTE_ETHER_MIN_LEN,
78 			    (uint32_t)ICE_FRAME_SIZE_MAX);
79 		return -EINVAL;
80 	}
81 
82 	rxq->max_pkt_len = max_pkt_len;
83 	if ((dev_data->dev_conf.rxmode.offloads & RTE_ETH_RX_OFFLOAD_SCATTER) ||
84 	    (rxq->max_pkt_len + 2 * ICE_VLAN_TAG_SIZE) > buf_size) {
85 		dev_data->scattered_rx = 1;
86 	}
87 	rxq->qrx_tail = hw->hw_addr + IAVF_QRX_TAIL1(rxq->queue_id);
88 	IAVF_PCI_REG_WRITE(rxq->qrx_tail, rxq->nb_rx_desc - 1);
89 	IAVF_WRITE_FLUSH(hw);
90 
91 	return 0;
92 }
93 
94 static int
95 ice_dcf_init_rx_queues(struct rte_eth_dev *dev)
96 {
97 	struct ice_rx_queue **rxq =
98 		(struct ice_rx_queue **)dev->data->rx_queues;
99 	int i, ret;
100 
101 	for (i = 0; i < dev->data->nb_rx_queues; i++) {
102 		if (!rxq[i] || !rxq[i]->q_set)
103 			continue;
104 		ret = ice_dcf_init_rxq(dev, rxq[i]);
105 		if (ret)
106 			return ret;
107 	}
108 
109 	ice_set_rx_function(dev);
110 	ice_set_tx_function(dev);
111 
112 	return 0;
113 }
114 
115 #define IAVF_MISC_VEC_ID                RTE_INTR_VEC_ZERO_OFFSET
116 #define IAVF_RX_VEC_START               RTE_INTR_VEC_RXTX_OFFSET
117 
118 #define IAVF_ITR_INDEX_DEFAULT          0
119 #define IAVF_QUEUE_ITR_INTERVAL_DEFAULT 32 /* 32 us */
120 #define IAVF_QUEUE_ITR_INTERVAL_MAX     8160 /* 8160 us */
121 
122 static inline uint16_t
123 iavf_calc_itr_interval(int16_t interval)
124 {
125 	if (interval < 0 || interval > IAVF_QUEUE_ITR_INTERVAL_MAX)
126 		interval = IAVF_QUEUE_ITR_INTERVAL_DEFAULT;
127 
128 	/* Convert to hardware count, as writing each 1 represents 2 us */
129 	return interval / 2;
130 }
131 
132 static int
133 ice_dcf_config_rx_queues_irqs(struct rte_eth_dev *dev,
134 				     struct rte_intr_handle *intr_handle)
135 {
136 	struct ice_dcf_adapter *adapter = dev->data->dev_private;
137 	struct ice_dcf_hw *hw = &adapter->real_hw;
138 	uint16_t interval, i;
139 	int vec;
140 
141 	if (rte_intr_cap_multiple(intr_handle) &&
142 	    dev->data->dev_conf.intr_conf.rxq) {
143 		if (rte_intr_efd_enable(intr_handle, dev->data->nb_rx_queues))
144 			return -1;
145 	}
146 
147 	if (rte_intr_dp_is_en(intr_handle)) {
148 		if (rte_intr_vec_list_alloc(intr_handle, "intr_vec",
149 						   dev->data->nb_rx_queues)) {
150 			PMD_DRV_LOG(ERR, "Failed to allocate %d rx intr_vec",
151 				    dev->data->nb_rx_queues);
152 			return -1;
153 		}
154 	}
155 
156 	if (!dev->data->dev_conf.intr_conf.rxq ||
157 	    !rte_intr_dp_is_en(intr_handle)) {
158 		/* Rx interrupt disabled, Map interrupt only for writeback */
159 		hw->nb_msix = 1;
160 		if (hw->vf_res->vf_cap_flags &
161 		    VIRTCHNL_VF_OFFLOAD_WB_ON_ITR) {
162 			/* If WB_ON_ITR supports, enable it */
163 			hw->msix_base = IAVF_RX_VEC_START;
164 			/* Set the ITR for index zero, to 2us to make sure that
165 			 * we leave time for aggregation to occur, but don't
166 			 * increase latency dramatically.
167 			 */
168 			IAVF_WRITE_REG(&hw->avf,
169 				       IAVF_VFINT_DYN_CTLN1(hw->msix_base - 1),
170 				       (0 << IAVF_VFINT_DYN_CTLN1_ITR_INDX_SHIFT) |
171 				       IAVF_VFINT_DYN_CTLN1_WB_ON_ITR_MASK |
172 				       (2UL << IAVF_VFINT_DYN_CTLN1_INTERVAL_SHIFT));
173 		} else {
174 			/* If no WB_ON_ITR offload flags, need to set
175 			 * interrupt for descriptor write back.
176 			 */
177 			hw->msix_base = IAVF_MISC_VEC_ID;
178 
179 			/* set ITR to max */
180 			interval =
181 			iavf_calc_itr_interval(IAVF_QUEUE_ITR_INTERVAL_MAX);
182 			IAVF_WRITE_REG(&hw->avf, IAVF_VFINT_DYN_CTL01,
183 				       IAVF_VFINT_DYN_CTL01_INTENA_MASK |
184 				       (IAVF_ITR_INDEX_DEFAULT <<
185 					IAVF_VFINT_DYN_CTL01_ITR_INDX_SHIFT) |
186 				       (interval <<
187 					IAVF_VFINT_DYN_CTL01_INTERVAL_SHIFT));
188 		}
189 		IAVF_WRITE_FLUSH(&hw->avf);
190 		/* map all queues to the same interrupt */
191 		for (i = 0; i < dev->data->nb_rx_queues; i++)
192 			hw->rxq_map[hw->msix_base] |= 1 << i;
193 	} else {
194 		if (!rte_intr_allow_others(intr_handle)) {
195 			hw->nb_msix = 1;
196 			hw->msix_base = IAVF_MISC_VEC_ID;
197 			for (i = 0; i < dev->data->nb_rx_queues; i++) {
198 				hw->rxq_map[hw->msix_base] |= 1 << i;
199 				rte_intr_vec_list_index_set(intr_handle,
200 							i, IAVF_MISC_VEC_ID);
201 			}
202 			PMD_DRV_LOG(DEBUG,
203 				    "vector %u are mapping to all Rx queues",
204 				    hw->msix_base);
205 		} else {
206 			/* If Rx interrupt is reuquired, and we can use
207 			 * multi interrupts, then the vec is from 1
208 			 */
209 			hw->nb_msix = RTE_MIN(hw->vf_res->max_vectors,
210 				      rte_intr_nb_efd_get(intr_handle));
211 			hw->msix_base = IAVF_MISC_VEC_ID;
212 			vec = IAVF_MISC_VEC_ID;
213 			for (i = 0; i < dev->data->nb_rx_queues; i++) {
214 				hw->rxq_map[vec] |= 1 << i;
215 				rte_intr_vec_list_index_set(intr_handle,
216 								   i, vec++);
217 				if (vec >= hw->nb_msix)
218 					vec = IAVF_RX_VEC_START;
219 			}
220 			PMD_DRV_LOG(DEBUG,
221 				    "%u vectors are mapping to %u Rx queues",
222 				    hw->nb_msix, dev->data->nb_rx_queues);
223 		}
224 	}
225 
226 	if (ice_dcf_config_irq_map(hw)) {
227 		PMD_DRV_LOG(ERR, "config interrupt mapping failed");
228 		return -1;
229 	}
230 	return 0;
231 }
232 
233 static int
234 alloc_rxq_mbufs(struct ice_rx_queue *rxq)
235 {
236 	volatile union ice_rx_flex_desc *rxd;
237 	struct rte_mbuf *mbuf = NULL;
238 	uint64_t dma_addr;
239 	uint16_t i;
240 
241 	for (i = 0; i < rxq->nb_rx_desc; i++) {
242 		mbuf = rte_mbuf_raw_alloc(rxq->mp);
243 		if (unlikely(!mbuf)) {
244 			PMD_DRV_LOG(ERR, "Failed to allocate mbuf for RX");
245 			return -ENOMEM;
246 		}
247 
248 		rte_mbuf_refcnt_set(mbuf, 1);
249 		mbuf->next = NULL;
250 		mbuf->data_off = RTE_PKTMBUF_HEADROOM;
251 		mbuf->nb_segs = 1;
252 		mbuf->port = rxq->port_id;
253 
254 		dma_addr =
255 			rte_cpu_to_le_64(rte_mbuf_data_iova_default(mbuf));
256 
257 		rxd = &rxq->rx_ring[i];
258 		rxd->read.pkt_addr = dma_addr;
259 		rxd->read.hdr_addr = 0;
260 #ifndef RTE_LIBRTE_ICE_16BYTE_RX_DESC
261 		rxd->read.rsvd1 = 0;
262 		rxd->read.rsvd2 = 0;
263 #endif
264 
265 		rxq->sw_ring[i].mbuf = (void *)mbuf;
266 	}
267 
268 	return 0;
269 }
270 
271 static int
272 ice_dcf_rx_queue_start(struct rte_eth_dev *dev, uint16_t rx_queue_id)
273 {
274 	struct ice_dcf_adapter *ad = dev->data->dev_private;
275 	struct iavf_hw *hw = &ad->real_hw.avf;
276 	struct ice_rx_queue *rxq;
277 	int err = 0;
278 
279 	if (rx_queue_id >= dev->data->nb_rx_queues)
280 		return -EINVAL;
281 
282 	rxq = dev->data->rx_queues[rx_queue_id];
283 
284 	err = alloc_rxq_mbufs(rxq);
285 	if (err) {
286 		PMD_DRV_LOG(ERR, "Failed to allocate RX queue mbuf");
287 		return err;
288 	}
289 
290 	rte_wmb();
291 
292 	/* Init the RX tail register. */
293 	IAVF_PCI_REG_WRITE(rxq->qrx_tail, rxq->nb_rx_desc - 1);
294 	IAVF_WRITE_FLUSH(hw);
295 
296 	/* Ready to switch the queue on */
297 	err = ice_dcf_switch_queue(&ad->real_hw, rx_queue_id, true, true);
298 	if (err) {
299 		PMD_DRV_LOG(ERR, "Failed to switch RX queue %u on",
300 			    rx_queue_id);
301 		return err;
302 	}
303 
304 	dev->data->rx_queue_state[rx_queue_id] = RTE_ETH_QUEUE_STATE_STARTED;
305 
306 	return 0;
307 }
308 
309 static inline void
310 reset_rx_queue(struct ice_rx_queue *rxq)
311 {
312 	uint16_t len;
313 	uint32_t i;
314 
315 	if (!rxq)
316 		return;
317 
318 	len = rxq->nb_rx_desc + ICE_RX_MAX_BURST;
319 
320 	for (i = 0; i < len * sizeof(union ice_rx_flex_desc); i++)
321 		((volatile char *)rxq->rx_ring)[i] = 0;
322 
323 	memset(&rxq->fake_mbuf, 0x0, sizeof(rxq->fake_mbuf));
324 
325 	for (i = 0; i < ICE_RX_MAX_BURST; i++)
326 		rxq->sw_ring[rxq->nb_rx_desc + i].mbuf = &rxq->fake_mbuf;
327 
328 	/* for rx bulk */
329 	rxq->rx_nb_avail = 0;
330 	rxq->rx_next_avail = 0;
331 	rxq->rx_free_trigger = (uint16_t)(rxq->rx_free_thresh - 1);
332 
333 	rxq->rx_tail = 0;
334 	rxq->nb_rx_hold = 0;
335 	rxq->pkt_first_seg = NULL;
336 	rxq->pkt_last_seg = NULL;
337 }
338 
339 static inline void
340 reset_tx_queue(struct ice_tx_queue *txq)
341 {
342 	struct ice_tx_entry *txe;
343 	uint32_t i, size;
344 	uint16_t prev;
345 
346 	if (!txq) {
347 		PMD_DRV_LOG(DEBUG, "Pointer to txq is NULL");
348 		return;
349 	}
350 
351 	txe = txq->sw_ring;
352 	size = sizeof(struct ice_tx_desc) * txq->nb_tx_desc;
353 	for (i = 0; i < size; i++)
354 		((volatile char *)txq->tx_ring)[i] = 0;
355 
356 	prev = (uint16_t)(txq->nb_tx_desc - 1);
357 	for (i = 0; i < txq->nb_tx_desc; i++) {
358 		txq->tx_ring[i].cmd_type_offset_bsz =
359 			rte_cpu_to_le_64(IAVF_TX_DESC_DTYPE_DESC_DONE);
360 		txe[i].mbuf =  NULL;
361 		txe[i].last_id = i;
362 		txe[prev].next_id = i;
363 		prev = i;
364 	}
365 
366 	txq->tx_tail = 0;
367 	txq->nb_tx_used = 0;
368 
369 	txq->last_desc_cleaned = txq->nb_tx_desc - 1;
370 	txq->nb_tx_free = txq->nb_tx_desc - 1;
371 
372 	txq->tx_next_dd = txq->tx_rs_thresh - 1;
373 	txq->tx_next_rs = txq->tx_rs_thresh - 1;
374 }
375 
376 static int
377 ice_dcf_rx_queue_stop(struct rte_eth_dev *dev, uint16_t rx_queue_id)
378 {
379 	struct ice_dcf_adapter *ad = dev->data->dev_private;
380 	struct ice_dcf_hw *hw = &ad->real_hw;
381 	struct ice_rx_queue *rxq;
382 	int err;
383 
384 	if (rx_queue_id >= dev->data->nb_rx_queues)
385 		return -EINVAL;
386 
387 	err = ice_dcf_switch_queue(hw, rx_queue_id, true, false);
388 	if (err) {
389 		PMD_DRV_LOG(ERR, "Failed to switch RX queue %u off",
390 			    rx_queue_id);
391 		return err;
392 	}
393 
394 	rxq = dev->data->rx_queues[rx_queue_id];
395 	rxq->rx_rel_mbufs(rxq);
396 	reset_rx_queue(rxq);
397 	dev->data->rx_queue_state[rx_queue_id] = RTE_ETH_QUEUE_STATE_STOPPED;
398 
399 	return 0;
400 }
401 
402 static int
403 ice_dcf_tx_queue_start(struct rte_eth_dev *dev, uint16_t tx_queue_id)
404 {
405 	struct ice_dcf_adapter *ad = dev->data->dev_private;
406 	struct iavf_hw *hw = &ad->real_hw.avf;
407 	struct ice_tx_queue *txq;
408 	int err = 0;
409 
410 	if (tx_queue_id >= dev->data->nb_tx_queues)
411 		return -EINVAL;
412 
413 	txq = dev->data->tx_queues[tx_queue_id];
414 
415 	/* Init the RX tail register. */
416 	txq->qtx_tail = hw->hw_addr + IAVF_QTX_TAIL1(tx_queue_id);
417 	IAVF_PCI_REG_WRITE(txq->qtx_tail, 0);
418 	IAVF_WRITE_FLUSH(hw);
419 
420 	/* Ready to switch the queue on */
421 	err = ice_dcf_switch_queue(&ad->real_hw, tx_queue_id, false, true);
422 
423 	if (err) {
424 		PMD_DRV_LOG(ERR, "Failed to switch TX queue %u on",
425 			    tx_queue_id);
426 		return err;
427 	}
428 
429 	dev->data->tx_queue_state[tx_queue_id] = RTE_ETH_QUEUE_STATE_STARTED;
430 
431 	return 0;
432 }
433 
434 static int
435 ice_dcf_tx_queue_stop(struct rte_eth_dev *dev, uint16_t tx_queue_id)
436 {
437 	struct ice_dcf_adapter *ad = dev->data->dev_private;
438 	struct ice_dcf_hw *hw = &ad->real_hw;
439 	struct ice_tx_queue *txq;
440 	int err;
441 
442 	if (tx_queue_id >= dev->data->nb_tx_queues)
443 		return -EINVAL;
444 
445 	err = ice_dcf_switch_queue(hw, tx_queue_id, false, false);
446 	if (err) {
447 		PMD_DRV_LOG(ERR, "Failed to switch TX queue %u off",
448 			    tx_queue_id);
449 		return err;
450 	}
451 
452 	txq = dev->data->tx_queues[tx_queue_id];
453 	txq->tx_rel_mbufs(txq);
454 	reset_tx_queue(txq);
455 	dev->data->tx_queue_state[tx_queue_id] = RTE_ETH_QUEUE_STATE_STOPPED;
456 
457 	return 0;
458 }
459 
460 static int
461 ice_dcf_start_queues(struct rte_eth_dev *dev)
462 {
463 	struct ice_rx_queue *rxq;
464 	struct ice_tx_queue *txq;
465 	int nb_rxq = 0;
466 	int nb_txq, i;
467 
468 	for (nb_txq = 0; nb_txq < dev->data->nb_tx_queues; nb_txq++) {
469 		txq = dev->data->tx_queues[nb_txq];
470 		if (txq->tx_deferred_start)
471 			continue;
472 		if (ice_dcf_tx_queue_start(dev, nb_txq) != 0) {
473 			PMD_DRV_LOG(ERR, "Fail to start queue %u", nb_txq);
474 			goto tx_err;
475 		}
476 	}
477 
478 	for (nb_rxq = 0; nb_rxq < dev->data->nb_rx_queues; nb_rxq++) {
479 		rxq = dev->data->rx_queues[nb_rxq];
480 		if (rxq->rx_deferred_start)
481 			continue;
482 		if (ice_dcf_rx_queue_start(dev, nb_rxq) != 0) {
483 			PMD_DRV_LOG(ERR, "Fail to start queue %u", nb_rxq);
484 			goto rx_err;
485 		}
486 	}
487 
488 	return 0;
489 
490 	/* stop the started queues if failed to start all queues */
491 rx_err:
492 	for (i = 0; i < nb_rxq; i++)
493 		ice_dcf_rx_queue_stop(dev, i);
494 tx_err:
495 	for (i = 0; i < nb_txq; i++)
496 		ice_dcf_tx_queue_stop(dev, i);
497 
498 	return -1;
499 }
500 
501 static int
502 ice_dcf_dev_start(struct rte_eth_dev *dev)
503 {
504 	struct ice_dcf_adapter *dcf_ad = dev->data->dev_private;
505 	struct rte_intr_handle *intr_handle = dev->intr_handle;
506 	struct ice_adapter *ad = &dcf_ad->parent;
507 	struct ice_dcf_hw *hw = &dcf_ad->real_hw;
508 	int ret;
509 
510 	if (hw->resetting) {
511 		PMD_DRV_LOG(ERR,
512 			    "The DCF has been reset by PF, please reinit first");
513 		return -EIO;
514 	}
515 
516 	if (hw->tm_conf.root && !hw->tm_conf.committed) {
517 		PMD_DRV_LOG(ERR,
518 			"please call hierarchy_commit() before starting the port");
519 		return -EIO;
520 	}
521 
522 	ad->pf.adapter_stopped = 0;
523 
524 	hw->num_queue_pairs = RTE_MAX(dev->data->nb_rx_queues,
525 				      dev->data->nb_tx_queues);
526 
527 	ret = ice_dcf_init_rx_queues(dev);
528 	if (ret) {
529 		PMD_DRV_LOG(ERR, "Fail to init queues");
530 		return ret;
531 	}
532 
533 	if (hw->vf_res->vf_cap_flags & VIRTCHNL_VF_OFFLOAD_RSS_PF) {
534 		ret = ice_dcf_init_rss(hw);
535 		if (ret) {
536 			PMD_DRV_LOG(ERR, "Failed to configure RSS");
537 			return ret;
538 		}
539 	}
540 
541 	ret = ice_dcf_configure_queues(hw);
542 	if (ret) {
543 		PMD_DRV_LOG(ERR, "Fail to config queues");
544 		return ret;
545 	}
546 
547 	ret = ice_dcf_config_rx_queues_irqs(dev, intr_handle);
548 	if (ret) {
549 		PMD_DRV_LOG(ERR, "Fail to config rx queues' irqs");
550 		return ret;
551 	}
552 
553 	if (dev->data->dev_conf.intr_conf.rxq != 0) {
554 		rte_intr_disable(intr_handle);
555 		rte_intr_enable(intr_handle);
556 	}
557 
558 	ret = ice_dcf_start_queues(dev);
559 	if (ret) {
560 		PMD_DRV_LOG(ERR, "Failed to enable queues");
561 		return ret;
562 	}
563 
564 	ret = ice_dcf_add_del_all_mac_addr(hw, true);
565 	if (ret) {
566 		PMD_DRV_LOG(ERR, "Failed to add mac addr");
567 		return ret;
568 	}
569 
570 	dev->data->dev_link.link_status = RTE_ETH_LINK_UP;
571 
572 	return 0;
573 }
574 
575 static void
576 ice_dcf_stop_queues(struct rte_eth_dev *dev)
577 {
578 	struct ice_dcf_adapter *ad = dev->data->dev_private;
579 	struct ice_dcf_hw *hw = &ad->real_hw;
580 	struct ice_rx_queue *rxq;
581 	struct ice_tx_queue *txq;
582 	int ret, i;
583 
584 	/* Stop All queues */
585 	ret = ice_dcf_disable_queues(hw);
586 	if (ret)
587 		PMD_DRV_LOG(WARNING, "Fail to stop queues");
588 
589 	for (i = 0; i < dev->data->nb_tx_queues; i++) {
590 		txq = dev->data->tx_queues[i];
591 		if (!txq)
592 			continue;
593 		txq->tx_rel_mbufs(txq);
594 		reset_tx_queue(txq);
595 		dev->data->tx_queue_state[i] = RTE_ETH_QUEUE_STATE_STOPPED;
596 	}
597 	for (i = 0; i < dev->data->nb_rx_queues; i++) {
598 		rxq = dev->data->rx_queues[i];
599 		if (!rxq)
600 			continue;
601 		rxq->rx_rel_mbufs(rxq);
602 		reset_rx_queue(rxq);
603 		dev->data->rx_queue_state[i] = RTE_ETH_QUEUE_STATE_STOPPED;
604 	}
605 }
606 
607 static int
608 ice_dcf_dev_stop(struct rte_eth_dev *dev)
609 {
610 	struct ice_dcf_adapter *dcf_ad = dev->data->dev_private;
611 	struct rte_intr_handle *intr_handle = dev->intr_handle;
612 	struct ice_adapter *ad = &dcf_ad->parent;
613 	struct ice_dcf_hw *hw = &dcf_ad->real_hw;
614 
615 	if (ad->pf.adapter_stopped == 1) {
616 		PMD_DRV_LOG(DEBUG, "Port is already stopped");
617 		return 0;
618 	}
619 
620 	/* Stop the VF representors for this device */
621 	ice_dcf_vf_repr_stop_all(dcf_ad);
622 
623 	ice_dcf_stop_queues(dev);
624 
625 	rte_intr_efd_disable(intr_handle);
626 	rte_intr_vec_list_free(intr_handle);
627 
628 	ice_dcf_add_del_all_mac_addr(&dcf_ad->real_hw, false);
629 	dev->data->dev_link.link_status = RTE_ETH_LINK_DOWN;
630 	ad->pf.adapter_stopped = 1;
631 	hw->tm_conf.committed = false;
632 
633 	return 0;
634 }
635 
636 static int
637 ice_dcf_dev_configure(struct rte_eth_dev *dev)
638 {
639 	struct ice_dcf_adapter *dcf_ad = dev->data->dev_private;
640 	struct ice_adapter *ad = &dcf_ad->parent;
641 
642 	ad->rx_bulk_alloc_allowed = true;
643 	ad->tx_simple_allowed = true;
644 
645 	if (dev->data->dev_conf.rxmode.mq_mode & RTE_ETH_MQ_RX_RSS_FLAG)
646 		dev->data->dev_conf.rxmode.offloads |= RTE_ETH_RX_OFFLOAD_RSS_HASH;
647 
648 	return 0;
649 }
650 
651 static int
652 ice_dcf_dev_info_get(struct rte_eth_dev *dev,
653 		     struct rte_eth_dev_info *dev_info)
654 {
655 	struct ice_dcf_adapter *adapter = dev->data->dev_private;
656 	struct ice_dcf_hw *hw = &adapter->real_hw;
657 
658 	dev_info->max_mac_addrs = 1;
659 	dev_info->max_rx_queues = hw->vsi_res->num_queue_pairs;
660 	dev_info->max_tx_queues = hw->vsi_res->num_queue_pairs;
661 	dev_info->min_rx_bufsize = ICE_BUF_SIZE_MIN;
662 	dev_info->max_rx_pktlen = ICE_FRAME_SIZE_MAX;
663 	dev_info->hash_key_size = hw->vf_res->rss_key_size;
664 	dev_info->reta_size = hw->vf_res->rss_lut_size;
665 	dev_info->flow_type_rss_offloads = ICE_RSS_OFFLOAD_ALL;
666 	dev_info->dev_capa &= ~RTE_ETH_DEV_CAPA_FLOW_RULE_KEEP;
667 
668 	dev_info->rx_offload_capa =
669 		RTE_ETH_RX_OFFLOAD_VLAN_STRIP |
670 		RTE_ETH_RX_OFFLOAD_IPV4_CKSUM |
671 		RTE_ETH_RX_OFFLOAD_UDP_CKSUM |
672 		RTE_ETH_RX_OFFLOAD_TCP_CKSUM |
673 		RTE_ETH_RX_OFFLOAD_OUTER_IPV4_CKSUM |
674 		RTE_ETH_RX_OFFLOAD_SCATTER |
675 		RTE_ETH_RX_OFFLOAD_VLAN_FILTER |
676 		RTE_ETH_RX_OFFLOAD_RSS_HASH;
677 	dev_info->tx_offload_capa =
678 		RTE_ETH_TX_OFFLOAD_VLAN_INSERT |
679 		RTE_ETH_TX_OFFLOAD_IPV4_CKSUM |
680 		RTE_ETH_TX_OFFLOAD_UDP_CKSUM |
681 		RTE_ETH_TX_OFFLOAD_TCP_CKSUM |
682 		RTE_ETH_TX_OFFLOAD_SCTP_CKSUM |
683 		RTE_ETH_TX_OFFLOAD_OUTER_IPV4_CKSUM |
684 		RTE_ETH_TX_OFFLOAD_TCP_TSO |
685 		RTE_ETH_TX_OFFLOAD_VXLAN_TNL_TSO |
686 		RTE_ETH_TX_OFFLOAD_GRE_TNL_TSO |
687 		RTE_ETH_TX_OFFLOAD_IPIP_TNL_TSO |
688 		RTE_ETH_TX_OFFLOAD_GENEVE_TNL_TSO |
689 		RTE_ETH_TX_OFFLOAD_MULTI_SEGS;
690 
691 	dev_info->default_rxconf = (struct rte_eth_rxconf) {
692 		.rx_thresh = {
693 			.pthresh = ICE_DEFAULT_RX_PTHRESH,
694 			.hthresh = ICE_DEFAULT_RX_HTHRESH,
695 			.wthresh = ICE_DEFAULT_RX_WTHRESH,
696 		},
697 		.rx_free_thresh = ICE_DEFAULT_RX_FREE_THRESH,
698 		.rx_drop_en = 0,
699 		.offloads = 0,
700 	};
701 
702 	dev_info->default_txconf = (struct rte_eth_txconf) {
703 		.tx_thresh = {
704 			.pthresh = ICE_DEFAULT_TX_PTHRESH,
705 			.hthresh = ICE_DEFAULT_TX_HTHRESH,
706 			.wthresh = ICE_DEFAULT_TX_WTHRESH,
707 		},
708 		.tx_free_thresh = ICE_DEFAULT_TX_FREE_THRESH,
709 		.tx_rs_thresh = ICE_DEFAULT_TX_RSBIT_THRESH,
710 		.offloads = 0,
711 	};
712 
713 	dev_info->rx_desc_lim = (struct rte_eth_desc_lim) {
714 		.nb_max = ICE_MAX_RING_DESC,
715 		.nb_min = ICE_MIN_RING_DESC,
716 		.nb_align = ICE_ALIGN_RING_DESC,
717 	};
718 
719 	dev_info->tx_desc_lim = (struct rte_eth_desc_lim) {
720 		.nb_max = ICE_MAX_RING_DESC,
721 		.nb_min = ICE_MIN_RING_DESC,
722 		.nb_align = ICE_ALIGN_RING_DESC,
723 	};
724 
725 	return 0;
726 }
727 
728 static int
729 ice_dcf_dev_promiscuous_enable(__rte_unused struct rte_eth_dev *dev)
730 {
731 	return 0;
732 }
733 
734 static int
735 ice_dcf_dev_promiscuous_disable(__rte_unused struct rte_eth_dev *dev)
736 {
737 	return 0;
738 }
739 
740 static int
741 ice_dcf_dev_allmulticast_enable(__rte_unused struct rte_eth_dev *dev)
742 {
743 	return 0;
744 }
745 
746 static int
747 ice_dcf_dev_allmulticast_disable(__rte_unused struct rte_eth_dev *dev)
748 {
749 	return 0;
750 }
751 
752 static int
753 ice_dcf_dev_flow_ops_get(struct rte_eth_dev *dev,
754 			 const struct rte_flow_ops **ops)
755 {
756 	if (!dev)
757 		return -EINVAL;
758 
759 	*ops = &ice_flow_ops;
760 	return 0;
761 }
762 
763 #define ICE_DCF_32_BIT_WIDTH (CHAR_BIT * 4)
764 #define ICE_DCF_48_BIT_WIDTH (CHAR_BIT * 6)
765 #define ICE_DCF_48_BIT_MASK  RTE_LEN2MASK(ICE_DCF_48_BIT_WIDTH, uint64_t)
766 
767 static void
768 ice_dcf_stat_update_48(uint64_t *offset, uint64_t *stat)
769 {
770 	if (*stat >= *offset)
771 		*stat = *stat - *offset;
772 	else
773 		*stat = (uint64_t)((*stat +
774 			((uint64_t)1 << ICE_DCF_48_BIT_WIDTH)) - *offset);
775 
776 	*stat &= ICE_DCF_48_BIT_MASK;
777 }
778 
779 static void
780 ice_dcf_stat_update_32(uint64_t *offset, uint64_t *stat)
781 {
782 	if (*stat >= *offset)
783 		*stat = (uint64_t)(*stat - *offset);
784 	else
785 		*stat = (uint64_t)((*stat +
786 			((uint64_t)1 << ICE_DCF_32_BIT_WIDTH)) - *offset);
787 }
788 
789 static void
790 ice_dcf_update_stats(struct virtchnl_eth_stats *oes,
791 		     struct virtchnl_eth_stats *nes)
792 {
793 	ice_dcf_stat_update_48(&oes->rx_bytes, &nes->rx_bytes);
794 	ice_dcf_stat_update_48(&oes->rx_unicast, &nes->rx_unicast);
795 	ice_dcf_stat_update_48(&oes->rx_multicast, &nes->rx_multicast);
796 	ice_dcf_stat_update_48(&oes->rx_broadcast, &nes->rx_broadcast);
797 	ice_dcf_stat_update_32(&oes->rx_discards, &nes->rx_discards);
798 	ice_dcf_stat_update_48(&oes->tx_bytes, &nes->tx_bytes);
799 	ice_dcf_stat_update_48(&oes->tx_unicast, &nes->tx_unicast);
800 	ice_dcf_stat_update_48(&oes->tx_multicast, &nes->tx_multicast);
801 	ice_dcf_stat_update_48(&oes->tx_broadcast, &nes->tx_broadcast);
802 	ice_dcf_stat_update_32(&oes->tx_errors, &nes->tx_errors);
803 	ice_dcf_stat_update_32(&oes->tx_discards, &nes->tx_discards);
804 }
805 
806 
807 static int
808 ice_dcf_stats_get(struct rte_eth_dev *dev, struct rte_eth_stats *stats)
809 {
810 	struct ice_dcf_adapter *ad = dev->data->dev_private;
811 	struct ice_dcf_hw *hw = &ad->real_hw;
812 	struct virtchnl_eth_stats pstats;
813 	int ret;
814 
815 	if (hw->resetting) {
816 		PMD_DRV_LOG(ERR,
817 			    "The DCF has been reset by PF, please reinit first");
818 		return -EIO;
819 	}
820 
821 	ret = ice_dcf_query_stats(hw, &pstats);
822 	if (ret == 0) {
823 		ice_dcf_update_stats(&hw->eth_stats_offset, &pstats);
824 		stats->ipackets = pstats.rx_unicast + pstats.rx_multicast +
825 				pstats.rx_broadcast - pstats.rx_discards;
826 		stats->opackets = pstats.tx_broadcast + pstats.tx_multicast +
827 						pstats.tx_unicast;
828 		stats->imissed = pstats.rx_discards;
829 		stats->oerrors = pstats.tx_errors + pstats.tx_discards;
830 		stats->ibytes = pstats.rx_bytes;
831 		stats->ibytes -= stats->ipackets * RTE_ETHER_CRC_LEN;
832 		stats->obytes = pstats.tx_bytes;
833 	} else {
834 		PMD_DRV_LOG(ERR, "Get statistics failed");
835 	}
836 	return ret;
837 }
838 
839 static int
840 ice_dcf_stats_reset(struct rte_eth_dev *dev)
841 {
842 	struct ice_dcf_adapter *ad = dev->data->dev_private;
843 	struct ice_dcf_hw *hw = &ad->real_hw;
844 	struct virtchnl_eth_stats pstats;
845 	int ret;
846 
847 	if (hw->resetting)
848 		return 0;
849 
850 	/* read stat values to clear hardware registers */
851 	ret = ice_dcf_query_stats(hw, &pstats);
852 	if (ret != 0)
853 		return ret;
854 
855 	/* set stats offset base on current values */
856 	hw->eth_stats_offset = pstats;
857 
858 	return 0;
859 }
860 
861 static void
862 ice_dcf_free_repr_info(struct ice_dcf_adapter *dcf_adapter)
863 {
864 	if (dcf_adapter->repr_infos) {
865 		rte_free(dcf_adapter->repr_infos);
866 		dcf_adapter->repr_infos = NULL;
867 	}
868 }
869 
870 static int
871 ice_dcf_init_repr_info(struct ice_dcf_adapter *dcf_adapter)
872 {
873 	dcf_adapter->repr_infos =
874 			rte_calloc("ice_dcf_rep_info",
875 				   dcf_adapter->real_hw.num_vfs,
876 				   sizeof(dcf_adapter->repr_infos[0]), 0);
877 	if (!dcf_adapter->repr_infos) {
878 		PMD_DRV_LOG(ERR, "Failed to alloc memory for VF representors\n");
879 		return -ENOMEM;
880 	}
881 
882 	return 0;
883 }
884 
885 static int
886 ice_dcf_dev_close(struct rte_eth_dev *dev)
887 {
888 	struct ice_dcf_adapter *adapter = dev->data->dev_private;
889 
890 	if (rte_eal_process_type() != RTE_PROC_PRIMARY)
891 		return 0;
892 
893 	(void)ice_dcf_dev_stop(dev);
894 
895 	ice_free_queues(dev);
896 
897 	ice_dcf_free_repr_info(adapter);
898 	ice_dcf_uninit_parent_adapter(dev);
899 	ice_dcf_uninit_hw(dev, &adapter->real_hw);
900 
901 	return 0;
902 }
903 
904 int
905 ice_dcf_link_update(struct rte_eth_dev *dev,
906 		    __rte_unused int wait_to_complete)
907 {
908 	struct ice_dcf_adapter *ad = dev->data->dev_private;
909 	struct ice_dcf_hw *hw = &ad->real_hw;
910 	struct rte_eth_link new_link;
911 
912 	memset(&new_link, 0, sizeof(new_link));
913 
914 	/* Only read status info stored in VF, and the info is updated
915 	 * when receive LINK_CHANGE event from PF by virtchnl.
916 	 */
917 	switch (hw->link_speed) {
918 	case 10:
919 		new_link.link_speed = RTE_ETH_SPEED_NUM_10M;
920 		break;
921 	case 100:
922 		new_link.link_speed = RTE_ETH_SPEED_NUM_100M;
923 		break;
924 	case 1000:
925 		new_link.link_speed = RTE_ETH_SPEED_NUM_1G;
926 		break;
927 	case 10000:
928 		new_link.link_speed = RTE_ETH_SPEED_NUM_10G;
929 		break;
930 	case 20000:
931 		new_link.link_speed = RTE_ETH_SPEED_NUM_20G;
932 		break;
933 	case 25000:
934 		new_link.link_speed = RTE_ETH_SPEED_NUM_25G;
935 		break;
936 	case 40000:
937 		new_link.link_speed = RTE_ETH_SPEED_NUM_40G;
938 		break;
939 	case 50000:
940 		new_link.link_speed = RTE_ETH_SPEED_NUM_50G;
941 		break;
942 	case 100000:
943 		new_link.link_speed = RTE_ETH_SPEED_NUM_100G;
944 		break;
945 	default:
946 		new_link.link_speed = RTE_ETH_SPEED_NUM_NONE;
947 		break;
948 	}
949 
950 	new_link.link_duplex = RTE_ETH_LINK_FULL_DUPLEX;
951 	new_link.link_status = hw->link_up ? RTE_ETH_LINK_UP :
952 					     RTE_ETH_LINK_DOWN;
953 	new_link.link_autoneg = !(dev->data->dev_conf.link_speeds &
954 				RTE_ETH_LINK_SPEED_FIXED);
955 
956 	return rte_eth_linkstatus_set(dev, &new_link);
957 }
958 
959 /* Add UDP tunneling port */
960 static int
961 ice_dcf_dev_udp_tunnel_port_add(struct rte_eth_dev *dev,
962 				struct rte_eth_udp_tunnel *udp_tunnel)
963 {
964 	struct ice_dcf_adapter *adapter = dev->data->dev_private;
965 	struct ice_adapter *parent_adapter = &adapter->parent;
966 	struct ice_hw *parent_hw = &parent_adapter->hw;
967 	int ret = 0;
968 
969 	if (!udp_tunnel)
970 		return -EINVAL;
971 
972 	switch (udp_tunnel->prot_type) {
973 	case RTE_ETH_TUNNEL_TYPE_VXLAN:
974 		ret = ice_create_tunnel(parent_hw, TNL_VXLAN,
975 					udp_tunnel->udp_port);
976 		break;
977 	case RTE_ETH_TUNNEL_TYPE_ECPRI:
978 		ret = ice_create_tunnel(parent_hw, TNL_ECPRI,
979 					udp_tunnel->udp_port);
980 		break;
981 	default:
982 		PMD_DRV_LOG(ERR, "Invalid tunnel type");
983 		ret = -EINVAL;
984 		break;
985 	}
986 
987 	return ret;
988 }
989 
990 /* Delete UDP tunneling port */
991 static int
992 ice_dcf_dev_udp_tunnel_port_del(struct rte_eth_dev *dev,
993 				struct rte_eth_udp_tunnel *udp_tunnel)
994 {
995 	struct ice_dcf_adapter *adapter = dev->data->dev_private;
996 	struct ice_adapter *parent_adapter = &adapter->parent;
997 	struct ice_hw *parent_hw = &parent_adapter->hw;
998 	int ret = 0;
999 
1000 	if (!udp_tunnel)
1001 		return -EINVAL;
1002 
1003 	switch (udp_tunnel->prot_type) {
1004 	case RTE_ETH_TUNNEL_TYPE_VXLAN:
1005 	case RTE_ETH_TUNNEL_TYPE_ECPRI:
1006 		ret = ice_destroy_tunnel(parent_hw, udp_tunnel->udp_port, 0);
1007 		break;
1008 	default:
1009 		PMD_DRV_LOG(ERR, "Invalid tunnel type");
1010 		ret = -EINVAL;
1011 		break;
1012 	}
1013 
1014 	return ret;
1015 }
1016 
1017 static int
1018 ice_dcf_tm_ops_get(struct rte_eth_dev *dev __rte_unused,
1019 		void *arg)
1020 {
1021 	if (!arg)
1022 		return -EINVAL;
1023 
1024 	*(const void **)arg = &ice_dcf_tm_ops;
1025 
1026 	return 0;
1027 }
1028 
1029 static inline void
1030 ice_dcf_reset_hw(struct rte_eth_dev *eth_dev, struct ice_dcf_hw *hw)
1031 {
1032 	ice_dcf_uninit_hw(eth_dev, hw);
1033 	ice_dcf_init_hw(eth_dev, hw);
1034 }
1035 
1036 /* Check if reset has been triggered by PF */
1037 static inline bool
1038 ice_dcf_is_reset(struct rte_eth_dev *dev)
1039 {
1040 	struct ice_dcf_adapter *ad = dev->data->dev_private;
1041 	struct iavf_hw *hw = &ad->real_hw.avf;
1042 
1043 	return !(IAVF_READ_REG(hw, IAVF_VF_ARQLEN1) &
1044 		 IAVF_VF_ARQLEN1_ARQENABLE_MASK);
1045 }
1046 
1047 static int
1048 ice_dcf_dev_reset(struct rte_eth_dev *dev)
1049 {
1050 	struct ice_dcf_adapter *ad = dev->data->dev_private;
1051 	struct ice_dcf_hw *hw = &ad->real_hw;
1052 	int ret;
1053 
1054 	if (ice_dcf_is_reset(dev)) {
1055 		if (!ad->real_hw.resetting)
1056 			ad->real_hw.resetting = true;
1057 		PMD_DRV_LOG(ERR, "The DCF has been reset by PF");
1058 
1059 		/*
1060 		 * Simply reset hw to trigger an additional DCF enable/disable
1061 		 * cycle which help to workaround the issue that kernel driver
1062 		 * may not clean up resource during previous reset.
1063 		 */
1064 		ice_dcf_reset_hw(dev, hw);
1065 	}
1066 
1067 	ret = ice_dcf_dev_uninit(dev);
1068 	if (ret)
1069 		return ret;
1070 
1071 	ret = ice_dcf_dev_init(dev);
1072 
1073 	return ret;
1074 }
1075 
1076 static const struct eth_dev_ops ice_dcf_eth_dev_ops = {
1077 	.dev_start               = ice_dcf_dev_start,
1078 	.dev_stop                = ice_dcf_dev_stop,
1079 	.dev_close               = ice_dcf_dev_close,
1080 	.dev_reset               = ice_dcf_dev_reset,
1081 	.dev_configure           = ice_dcf_dev_configure,
1082 	.dev_infos_get           = ice_dcf_dev_info_get,
1083 	.rx_queue_setup          = ice_rx_queue_setup,
1084 	.tx_queue_setup          = ice_tx_queue_setup,
1085 	.rx_queue_release        = ice_dev_rx_queue_release,
1086 	.tx_queue_release        = ice_dev_tx_queue_release,
1087 	.rx_queue_start          = ice_dcf_rx_queue_start,
1088 	.tx_queue_start          = ice_dcf_tx_queue_start,
1089 	.rx_queue_stop           = ice_dcf_rx_queue_stop,
1090 	.tx_queue_stop           = ice_dcf_tx_queue_stop,
1091 	.link_update             = ice_dcf_link_update,
1092 	.stats_get               = ice_dcf_stats_get,
1093 	.stats_reset             = ice_dcf_stats_reset,
1094 	.promiscuous_enable      = ice_dcf_dev_promiscuous_enable,
1095 	.promiscuous_disable     = ice_dcf_dev_promiscuous_disable,
1096 	.allmulticast_enable     = ice_dcf_dev_allmulticast_enable,
1097 	.allmulticast_disable    = ice_dcf_dev_allmulticast_disable,
1098 	.flow_ops_get            = ice_dcf_dev_flow_ops_get,
1099 	.udp_tunnel_port_add	 = ice_dcf_dev_udp_tunnel_port_add,
1100 	.udp_tunnel_port_del	 = ice_dcf_dev_udp_tunnel_port_del,
1101 	.tm_ops_get              = ice_dcf_tm_ops_get,
1102 };
1103 
1104 static int
1105 ice_dcf_dev_init(struct rte_eth_dev *eth_dev)
1106 {
1107 	struct ice_dcf_adapter *adapter = eth_dev->data->dev_private;
1108 
1109 	eth_dev->dev_ops = &ice_dcf_eth_dev_ops;
1110 	eth_dev->rx_pkt_burst = ice_dcf_recv_pkts;
1111 	eth_dev->tx_pkt_burst = ice_dcf_xmit_pkts;
1112 
1113 	if (rte_eal_process_type() != RTE_PROC_PRIMARY)
1114 		return 0;
1115 
1116 	adapter->real_hw.vc_event_msg_cb = ice_dcf_handle_pf_event_msg;
1117 	if (ice_dcf_init_hw(eth_dev, &adapter->real_hw) != 0) {
1118 		PMD_INIT_LOG(ERR, "Failed to init DCF hardware");
1119 		return -1;
1120 	}
1121 
1122 	if (ice_dcf_init_parent_adapter(eth_dev) != 0) {
1123 		PMD_INIT_LOG(ERR, "Failed to init DCF parent adapter");
1124 		ice_dcf_uninit_hw(eth_dev, &adapter->real_hw);
1125 		return -1;
1126 	}
1127 
1128 	return 0;
1129 }
1130 
1131 static int
1132 ice_dcf_dev_uninit(struct rte_eth_dev *eth_dev)
1133 {
1134 	ice_dcf_dev_close(eth_dev);
1135 
1136 	return 0;
1137 }
1138 
1139 static int
1140 ice_dcf_cap_check_handler(__rte_unused const char *key,
1141 			  const char *value, __rte_unused void *opaque)
1142 {
1143 	if (strcmp(value, "dcf"))
1144 		return -1;
1145 
1146 	return 0;
1147 }
1148 
1149 static int
1150 ice_dcf_cap_selected(struct rte_devargs *devargs)
1151 {
1152 	struct rte_kvargs *kvlist;
1153 	const char *key = "cap";
1154 	int ret = 0;
1155 
1156 	if (devargs == NULL)
1157 		return 0;
1158 
1159 	kvlist = rte_kvargs_parse(devargs->args, NULL);
1160 	if (kvlist == NULL)
1161 		return 0;
1162 
1163 	if (!rte_kvargs_count(kvlist, key))
1164 		goto exit;
1165 
1166 	/* dcf capability selected when there's a key-value pair: cap=dcf */
1167 	if (rte_kvargs_process(kvlist, key,
1168 			       ice_dcf_cap_check_handler, NULL) < 0)
1169 		goto exit;
1170 
1171 	ret = 1;
1172 
1173 exit:
1174 	rte_kvargs_free(kvlist);
1175 	return ret;
1176 }
1177 
1178 static int
1179 eth_ice_dcf_pci_probe(__rte_unused struct rte_pci_driver *pci_drv,
1180 		      struct rte_pci_device *pci_dev)
1181 {
1182 	struct rte_eth_devargs eth_da = { .nb_representor_ports = 0 };
1183 	struct ice_dcf_vf_repr_param repr_param;
1184 	char repr_name[RTE_ETH_NAME_MAX_LEN];
1185 	struct ice_dcf_adapter *dcf_adapter;
1186 	struct rte_eth_dev *dcf_ethdev;
1187 	uint16_t dcf_vsi_id;
1188 	int i, ret;
1189 
1190 	if (!ice_dcf_cap_selected(pci_dev->device.devargs))
1191 		return 1;
1192 
1193 	ret = rte_eth_devargs_parse(pci_dev->device.devargs->args, &eth_da);
1194 	if (ret)
1195 		return ret;
1196 
1197 	ret = rte_eth_dev_pci_generic_probe(pci_dev,
1198 					    sizeof(struct ice_dcf_adapter),
1199 					    ice_dcf_dev_init);
1200 	if (ret || !eth_da.nb_representor_ports)
1201 		return ret;
1202 	if (eth_da.type != RTE_ETH_REPRESENTOR_VF)
1203 		return -ENOTSUP;
1204 
1205 	dcf_ethdev = rte_eth_dev_allocated(pci_dev->device.name);
1206 	if (dcf_ethdev == NULL)
1207 		return -ENODEV;
1208 
1209 	dcf_adapter = dcf_ethdev->data->dev_private;
1210 	ret = ice_dcf_init_repr_info(dcf_adapter);
1211 	if (ret)
1212 		return ret;
1213 
1214 	if (eth_da.nb_representor_ports > dcf_adapter->real_hw.num_vfs ||
1215 	    eth_da.nb_representor_ports >= RTE_MAX_ETHPORTS) {
1216 		PMD_DRV_LOG(ERR, "the number of port representors is too large: %u",
1217 			    eth_da.nb_representor_ports);
1218 		ice_dcf_free_repr_info(dcf_adapter);
1219 		return -EINVAL;
1220 	}
1221 
1222 	dcf_vsi_id = dcf_adapter->real_hw.vsi_id | VIRTCHNL_DCF_VF_VSI_VALID;
1223 
1224 	repr_param.dcf_eth_dev = dcf_ethdev;
1225 	repr_param.switch_domain_id = 0;
1226 
1227 	for (i = 0; i < eth_da.nb_representor_ports; i++) {
1228 		uint16_t vf_id = eth_da.representor_ports[i];
1229 		struct rte_eth_dev *vf_rep_eth_dev;
1230 
1231 		if (vf_id >= dcf_adapter->real_hw.num_vfs) {
1232 			PMD_DRV_LOG(ERR, "VF ID %u is out of range (0 ~ %u)",
1233 				    vf_id, dcf_adapter->real_hw.num_vfs - 1);
1234 			ret = -EINVAL;
1235 			break;
1236 		}
1237 
1238 		if (dcf_adapter->real_hw.vf_vsi_map[vf_id] == dcf_vsi_id) {
1239 			PMD_DRV_LOG(ERR, "VF ID %u is DCF's ID.\n", vf_id);
1240 			ret = -EINVAL;
1241 			break;
1242 		}
1243 
1244 		repr_param.vf_id = vf_id;
1245 		snprintf(repr_name, sizeof(repr_name), "net_%s_representor_%u",
1246 			 pci_dev->device.name, vf_id);
1247 		ret = rte_eth_dev_create(&pci_dev->device, repr_name,
1248 					 sizeof(struct ice_dcf_vf_repr),
1249 					 NULL, NULL, ice_dcf_vf_repr_init,
1250 					 &repr_param);
1251 		if (ret) {
1252 			PMD_DRV_LOG(ERR, "failed to create DCF VF representor %s",
1253 				    repr_name);
1254 			break;
1255 		}
1256 
1257 		vf_rep_eth_dev = rte_eth_dev_allocated(repr_name);
1258 		if (!vf_rep_eth_dev) {
1259 			PMD_DRV_LOG(ERR,
1260 				    "Failed to find the ethdev for DCF VF representor: %s",
1261 				    repr_name);
1262 			ret = -ENODEV;
1263 			break;
1264 		}
1265 
1266 		dcf_adapter->repr_infos[vf_id].vf_rep_eth_dev = vf_rep_eth_dev;
1267 		dcf_adapter->num_reprs++;
1268 	}
1269 
1270 	return ret;
1271 }
1272 
1273 static int
1274 eth_ice_dcf_pci_remove(struct rte_pci_device *pci_dev)
1275 {
1276 	struct rte_eth_dev *eth_dev;
1277 
1278 	eth_dev = rte_eth_dev_allocated(pci_dev->device.name);
1279 	if (!eth_dev)
1280 		return 0;
1281 
1282 	if (eth_dev->data->dev_flags & RTE_ETH_DEV_REPRESENTOR)
1283 		return rte_eth_dev_pci_generic_remove(pci_dev,
1284 						      ice_dcf_vf_repr_uninit);
1285 	else
1286 		return rte_eth_dev_pci_generic_remove(pci_dev,
1287 						      ice_dcf_dev_uninit);
1288 }
1289 
1290 static const struct rte_pci_id pci_id_ice_dcf_map[] = {
1291 	{ RTE_PCI_DEVICE(IAVF_INTEL_VENDOR_ID, IAVF_DEV_ID_ADAPTIVE_VF) },
1292 	{ .vendor_id = 0, /* sentinel */ },
1293 };
1294 
1295 static struct rte_pci_driver rte_ice_dcf_pmd = {
1296 	.id_table = pci_id_ice_dcf_map,
1297 	.drv_flags = RTE_PCI_DRV_NEED_MAPPING,
1298 	.probe = eth_ice_dcf_pci_probe,
1299 	.remove = eth_ice_dcf_pci_remove,
1300 };
1301 
1302 RTE_PMD_REGISTER_PCI(net_ice_dcf, rte_ice_dcf_pmd);
1303 RTE_PMD_REGISTER_PCI_TABLE(net_ice_dcf, pci_id_ice_dcf_map);
1304 RTE_PMD_REGISTER_KMOD_DEP(net_ice_dcf, "* igb_uio | vfio-pci");
1305 RTE_PMD_REGISTER_PARAM_STRING(net_ice_dcf, "cap=dcf");
1306