xref: /dpdk/drivers/net/ice/ice_dcf_ethdev.c (revision e4869c0b)
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 	ad->pf.adapter_stopped = 0;
532 
533 	hw->num_queue_pairs = RTE_MAX(dev->data->nb_rx_queues,
534 				      dev->data->nb_tx_queues);
535 
536 	ret = ice_dcf_init_rx_queues(dev);
537 	if (ret) {
538 		PMD_DRV_LOG(ERR, "Fail to init queues");
539 		return ret;
540 	}
541 
542 	if (hw->vf_res->vf_cap_flags & VIRTCHNL_VF_OFFLOAD_RSS_PF) {
543 		ret = ice_dcf_init_rss(hw);
544 		if (ret) {
545 			PMD_DRV_LOG(ERR, "Failed to configure RSS");
546 			return ret;
547 		}
548 	}
549 
550 	ret = ice_dcf_configure_queues(hw);
551 	if (ret) {
552 		PMD_DRV_LOG(ERR, "Fail to config queues");
553 		return ret;
554 	}
555 
556 	ret = ice_dcf_config_rx_queues_irqs(dev, intr_handle);
557 	if (ret) {
558 		PMD_DRV_LOG(ERR, "Fail to config rx queues' irqs");
559 		return ret;
560 	}
561 
562 	if (dev->data->dev_conf.intr_conf.rxq != 0) {
563 		rte_intr_disable(intr_handle);
564 		rte_intr_enable(intr_handle);
565 	}
566 
567 	ret = ice_dcf_start_queues(dev);
568 	if (ret) {
569 		PMD_DRV_LOG(ERR, "Failed to enable queues");
570 		return ret;
571 	}
572 
573 	ret = ice_dcf_add_del_all_mac_addr(hw, true);
574 	if (ret) {
575 		PMD_DRV_LOG(ERR, "Failed to add mac addr");
576 		return ret;
577 	}
578 
579 	dev->data->dev_link.link_status = ETH_LINK_UP;
580 
581 	return 0;
582 }
583 
584 static void
585 ice_dcf_stop_queues(struct rte_eth_dev *dev)
586 {
587 	struct ice_dcf_adapter *ad = dev->data->dev_private;
588 	struct ice_dcf_hw *hw = &ad->real_hw;
589 	struct ice_rx_queue *rxq;
590 	struct ice_tx_queue *txq;
591 	int ret, i;
592 
593 	/* Stop All queues */
594 	ret = ice_dcf_disable_queues(hw);
595 	if (ret)
596 		PMD_DRV_LOG(WARNING, "Fail to stop queues");
597 
598 	for (i = 0; i < dev->data->nb_tx_queues; i++) {
599 		txq = dev->data->tx_queues[i];
600 		if (!txq)
601 			continue;
602 		txq->tx_rel_mbufs(txq);
603 		reset_tx_queue(txq);
604 		dev->data->tx_queue_state[i] = RTE_ETH_QUEUE_STATE_STOPPED;
605 	}
606 	for (i = 0; i < dev->data->nb_rx_queues; i++) {
607 		rxq = dev->data->rx_queues[i];
608 		if (!rxq)
609 			continue;
610 		rxq->rx_rel_mbufs(rxq);
611 		reset_rx_queue(rxq);
612 		dev->data->rx_queue_state[i] = RTE_ETH_QUEUE_STATE_STOPPED;
613 	}
614 }
615 
616 static int
617 ice_dcf_dev_stop(struct rte_eth_dev *dev)
618 {
619 	struct ice_dcf_adapter *dcf_ad = dev->data->dev_private;
620 	struct rte_intr_handle *intr_handle = dev->intr_handle;
621 	struct ice_adapter *ad = &dcf_ad->parent;
622 
623 	if (ad->pf.adapter_stopped == 1) {
624 		PMD_DRV_LOG(DEBUG, "Port is already stopped");
625 		return 0;
626 	}
627 
628 	/* Stop the VF representors for this device */
629 	ice_dcf_vf_repr_stop_all(dcf_ad);
630 
631 	ice_dcf_stop_queues(dev);
632 
633 	rte_intr_efd_disable(intr_handle);
634 	if (intr_handle->intr_vec) {
635 		rte_free(intr_handle->intr_vec);
636 		intr_handle->intr_vec = NULL;
637 	}
638 
639 	ice_dcf_add_del_all_mac_addr(&dcf_ad->real_hw, false);
640 	dev->data->dev_link.link_status = ETH_LINK_DOWN;
641 	ad->pf.adapter_stopped = 1;
642 
643 	return 0;
644 }
645 
646 static int
647 ice_dcf_dev_configure(struct rte_eth_dev *dev)
648 {
649 	struct ice_dcf_adapter *dcf_ad = dev->data->dev_private;
650 	struct ice_adapter *ad = &dcf_ad->parent;
651 
652 	ad->rx_bulk_alloc_allowed = true;
653 	ad->tx_simple_allowed = true;
654 
655 	if (dev->data->dev_conf.rxmode.mq_mode & ETH_MQ_RX_RSS_FLAG)
656 		dev->data->dev_conf.rxmode.offloads |= DEV_RX_OFFLOAD_RSS_HASH;
657 
658 	return 0;
659 }
660 
661 static int
662 ice_dcf_dev_info_get(struct rte_eth_dev *dev,
663 		     struct rte_eth_dev_info *dev_info)
664 {
665 	struct ice_dcf_adapter *adapter = dev->data->dev_private;
666 	struct ice_dcf_hw *hw = &adapter->real_hw;
667 
668 	dev_info->max_mac_addrs = 1;
669 	dev_info->max_rx_queues = hw->vsi_res->num_queue_pairs;
670 	dev_info->max_tx_queues = hw->vsi_res->num_queue_pairs;
671 	dev_info->min_rx_bufsize = ICE_BUF_SIZE_MIN;
672 	dev_info->max_rx_pktlen = ICE_FRAME_SIZE_MAX;
673 	dev_info->hash_key_size = hw->vf_res->rss_key_size;
674 	dev_info->reta_size = hw->vf_res->rss_lut_size;
675 	dev_info->flow_type_rss_offloads = ICE_RSS_OFFLOAD_ALL;
676 
677 	dev_info->rx_offload_capa =
678 		DEV_RX_OFFLOAD_VLAN_STRIP |
679 		DEV_RX_OFFLOAD_IPV4_CKSUM |
680 		DEV_RX_OFFLOAD_UDP_CKSUM |
681 		DEV_RX_OFFLOAD_TCP_CKSUM |
682 		DEV_RX_OFFLOAD_OUTER_IPV4_CKSUM |
683 		DEV_RX_OFFLOAD_SCATTER |
684 		DEV_RX_OFFLOAD_VLAN_FILTER |
685 		DEV_RX_OFFLOAD_RSS_HASH;
686 	dev_info->tx_offload_capa =
687 		DEV_TX_OFFLOAD_VLAN_INSERT |
688 		DEV_TX_OFFLOAD_IPV4_CKSUM |
689 		DEV_TX_OFFLOAD_UDP_CKSUM |
690 		DEV_TX_OFFLOAD_TCP_CKSUM |
691 		DEV_TX_OFFLOAD_SCTP_CKSUM |
692 		DEV_TX_OFFLOAD_OUTER_IPV4_CKSUM |
693 		DEV_TX_OFFLOAD_TCP_TSO |
694 		DEV_TX_OFFLOAD_VXLAN_TNL_TSO |
695 		DEV_TX_OFFLOAD_GRE_TNL_TSO |
696 		DEV_TX_OFFLOAD_IPIP_TNL_TSO |
697 		DEV_TX_OFFLOAD_GENEVE_TNL_TSO |
698 		DEV_TX_OFFLOAD_MULTI_SEGS;
699 
700 	dev_info->default_rxconf = (struct rte_eth_rxconf) {
701 		.rx_thresh = {
702 			.pthresh = ICE_DEFAULT_RX_PTHRESH,
703 			.hthresh = ICE_DEFAULT_RX_HTHRESH,
704 			.wthresh = ICE_DEFAULT_RX_WTHRESH,
705 		},
706 		.rx_free_thresh = ICE_DEFAULT_RX_FREE_THRESH,
707 		.rx_drop_en = 0,
708 		.offloads = 0,
709 	};
710 
711 	dev_info->default_txconf = (struct rte_eth_txconf) {
712 		.tx_thresh = {
713 			.pthresh = ICE_DEFAULT_TX_PTHRESH,
714 			.hthresh = ICE_DEFAULT_TX_HTHRESH,
715 			.wthresh = ICE_DEFAULT_TX_WTHRESH,
716 		},
717 		.tx_free_thresh = ICE_DEFAULT_TX_FREE_THRESH,
718 		.tx_rs_thresh = ICE_DEFAULT_TX_RSBIT_THRESH,
719 		.offloads = 0,
720 	};
721 
722 	dev_info->rx_desc_lim = (struct rte_eth_desc_lim) {
723 		.nb_max = ICE_MAX_RING_DESC,
724 		.nb_min = ICE_MIN_RING_DESC,
725 		.nb_align = ICE_ALIGN_RING_DESC,
726 	};
727 
728 	dev_info->tx_desc_lim = (struct rte_eth_desc_lim) {
729 		.nb_max = ICE_MAX_RING_DESC,
730 		.nb_min = ICE_MIN_RING_DESC,
731 		.nb_align = ICE_ALIGN_RING_DESC,
732 	};
733 
734 	return 0;
735 }
736 
737 static int
738 ice_dcf_dev_promiscuous_enable(__rte_unused struct rte_eth_dev *dev)
739 {
740 	return 0;
741 }
742 
743 static int
744 ice_dcf_dev_promiscuous_disable(__rte_unused struct rte_eth_dev *dev)
745 {
746 	return 0;
747 }
748 
749 static int
750 ice_dcf_dev_allmulticast_enable(__rte_unused struct rte_eth_dev *dev)
751 {
752 	return 0;
753 }
754 
755 static int
756 ice_dcf_dev_allmulticast_disable(__rte_unused struct rte_eth_dev *dev)
757 {
758 	return 0;
759 }
760 
761 static int
762 ice_dcf_dev_flow_ops_get(struct rte_eth_dev *dev,
763 			 const struct rte_flow_ops **ops)
764 {
765 	if (!dev)
766 		return -EINVAL;
767 
768 	*ops = &ice_flow_ops;
769 	return 0;
770 }
771 
772 #define ICE_DCF_32_BIT_WIDTH (CHAR_BIT * 4)
773 #define ICE_DCF_48_BIT_WIDTH (CHAR_BIT * 6)
774 #define ICE_DCF_48_BIT_MASK  RTE_LEN2MASK(ICE_DCF_48_BIT_WIDTH, uint64_t)
775 
776 static void
777 ice_dcf_stat_update_48(uint64_t *offset, uint64_t *stat)
778 {
779 	if (*stat >= *offset)
780 		*stat = *stat - *offset;
781 	else
782 		*stat = (uint64_t)((*stat +
783 			((uint64_t)1 << ICE_DCF_48_BIT_WIDTH)) - *offset);
784 
785 	*stat &= ICE_DCF_48_BIT_MASK;
786 }
787 
788 static void
789 ice_dcf_stat_update_32(uint64_t *offset, uint64_t *stat)
790 {
791 	if (*stat >= *offset)
792 		*stat = (uint64_t)(*stat - *offset);
793 	else
794 		*stat = (uint64_t)((*stat +
795 			((uint64_t)1 << ICE_DCF_32_BIT_WIDTH)) - *offset);
796 }
797 
798 static void
799 ice_dcf_update_stats(struct virtchnl_eth_stats *oes,
800 		     struct virtchnl_eth_stats *nes)
801 {
802 	ice_dcf_stat_update_48(&oes->rx_bytes, &nes->rx_bytes);
803 	ice_dcf_stat_update_48(&oes->rx_unicast, &nes->rx_unicast);
804 	ice_dcf_stat_update_48(&oes->rx_multicast, &nes->rx_multicast);
805 	ice_dcf_stat_update_48(&oes->rx_broadcast, &nes->rx_broadcast);
806 	ice_dcf_stat_update_32(&oes->rx_discards, &nes->rx_discards);
807 	ice_dcf_stat_update_48(&oes->tx_bytes, &nes->tx_bytes);
808 	ice_dcf_stat_update_48(&oes->tx_unicast, &nes->tx_unicast);
809 	ice_dcf_stat_update_48(&oes->tx_multicast, &nes->tx_multicast);
810 	ice_dcf_stat_update_48(&oes->tx_broadcast, &nes->tx_broadcast);
811 	ice_dcf_stat_update_32(&oes->tx_errors, &nes->tx_errors);
812 	ice_dcf_stat_update_32(&oes->tx_discards, &nes->tx_discards);
813 }
814 
815 
816 static int
817 ice_dcf_stats_get(struct rte_eth_dev *dev, struct rte_eth_stats *stats)
818 {
819 	struct ice_dcf_adapter *ad = dev->data->dev_private;
820 	struct ice_dcf_hw *hw = &ad->real_hw;
821 	struct virtchnl_eth_stats pstats;
822 	int ret;
823 
824 	if (hw->resetting) {
825 		PMD_DRV_LOG(ERR,
826 			    "The DCF has been reset by PF, please reinit first");
827 		return -EIO;
828 	}
829 
830 	ret = ice_dcf_query_stats(hw, &pstats);
831 	if (ret == 0) {
832 		ice_dcf_update_stats(&hw->eth_stats_offset, &pstats);
833 		stats->ipackets = pstats.rx_unicast + pstats.rx_multicast +
834 				pstats.rx_broadcast - pstats.rx_discards;
835 		stats->opackets = pstats.tx_broadcast + pstats.tx_multicast +
836 						pstats.tx_unicast;
837 		stats->imissed = pstats.rx_discards;
838 		stats->oerrors = pstats.tx_errors + pstats.tx_discards;
839 		stats->ibytes = pstats.rx_bytes;
840 		stats->ibytes -= stats->ipackets * RTE_ETHER_CRC_LEN;
841 		stats->obytes = pstats.tx_bytes;
842 	} else {
843 		PMD_DRV_LOG(ERR, "Get statistics failed");
844 	}
845 	return ret;
846 }
847 
848 static int
849 ice_dcf_stats_reset(struct rte_eth_dev *dev)
850 {
851 	struct ice_dcf_adapter *ad = dev->data->dev_private;
852 	struct ice_dcf_hw *hw = &ad->real_hw;
853 	struct virtchnl_eth_stats pstats;
854 	int ret;
855 
856 	if (hw->resetting)
857 		return 0;
858 
859 	/* read stat values to clear hardware registers */
860 	ret = ice_dcf_query_stats(hw, &pstats);
861 	if (ret != 0)
862 		return ret;
863 
864 	/* set stats offset base on current values */
865 	hw->eth_stats_offset = pstats;
866 
867 	return 0;
868 }
869 
870 static void
871 ice_dcf_free_repr_info(struct ice_dcf_adapter *dcf_adapter)
872 {
873 	if (dcf_adapter->repr_infos) {
874 		rte_free(dcf_adapter->repr_infos);
875 		dcf_adapter->repr_infos = NULL;
876 	}
877 }
878 
879 static int
880 ice_dcf_init_repr_info(struct ice_dcf_adapter *dcf_adapter)
881 {
882 	dcf_adapter->repr_infos =
883 			rte_calloc("ice_dcf_rep_info",
884 				   dcf_adapter->real_hw.num_vfs,
885 				   sizeof(dcf_adapter->repr_infos[0]), 0);
886 	if (!dcf_adapter->repr_infos) {
887 		PMD_DRV_LOG(ERR, "Failed to alloc memory for VF representors\n");
888 		return -ENOMEM;
889 	}
890 
891 	return 0;
892 }
893 
894 static int
895 ice_dcf_dev_close(struct rte_eth_dev *dev)
896 {
897 	struct ice_dcf_adapter *adapter = dev->data->dev_private;
898 
899 	if (rte_eal_process_type() != RTE_PROC_PRIMARY)
900 		return 0;
901 
902 	(void)ice_dcf_dev_stop(dev);
903 
904 	ice_free_queues(dev);
905 
906 	ice_dcf_free_repr_info(adapter);
907 	ice_dcf_uninit_parent_adapter(dev);
908 	ice_dcf_uninit_hw(dev, &adapter->real_hw);
909 
910 	return 0;
911 }
912 
913 int
914 ice_dcf_link_update(struct rte_eth_dev *dev,
915 		    __rte_unused int wait_to_complete)
916 {
917 	struct ice_dcf_adapter *ad = dev->data->dev_private;
918 	struct ice_dcf_hw *hw = &ad->real_hw;
919 	struct rte_eth_link new_link;
920 
921 	memset(&new_link, 0, sizeof(new_link));
922 
923 	/* Only read status info stored in VF, and the info is updated
924 	 * when receive LINK_CHANGE event from PF by virtchnl.
925 	 */
926 	switch (hw->link_speed) {
927 	case 10:
928 		new_link.link_speed = ETH_SPEED_NUM_10M;
929 		break;
930 	case 100:
931 		new_link.link_speed = ETH_SPEED_NUM_100M;
932 		break;
933 	case 1000:
934 		new_link.link_speed = ETH_SPEED_NUM_1G;
935 		break;
936 	case 10000:
937 		new_link.link_speed = ETH_SPEED_NUM_10G;
938 		break;
939 	case 20000:
940 		new_link.link_speed = ETH_SPEED_NUM_20G;
941 		break;
942 	case 25000:
943 		new_link.link_speed = ETH_SPEED_NUM_25G;
944 		break;
945 	case 40000:
946 		new_link.link_speed = ETH_SPEED_NUM_40G;
947 		break;
948 	case 50000:
949 		new_link.link_speed = ETH_SPEED_NUM_50G;
950 		break;
951 	case 100000:
952 		new_link.link_speed = ETH_SPEED_NUM_100G;
953 		break;
954 	default:
955 		new_link.link_speed = ETH_SPEED_NUM_NONE;
956 		break;
957 	}
958 
959 	new_link.link_duplex = ETH_LINK_FULL_DUPLEX;
960 	new_link.link_status = hw->link_up ? ETH_LINK_UP :
961 					     ETH_LINK_DOWN;
962 	new_link.link_autoneg = !(dev->data->dev_conf.link_speeds &
963 				ETH_LINK_SPEED_FIXED);
964 
965 	return rte_eth_linkstatus_set(dev, &new_link);
966 }
967 
968 /* Add UDP tunneling port */
969 static int
970 ice_dcf_dev_udp_tunnel_port_add(struct rte_eth_dev *dev,
971 				struct rte_eth_udp_tunnel *udp_tunnel)
972 {
973 	struct ice_dcf_adapter *adapter = dev->data->dev_private;
974 	struct ice_adapter *parent_adapter = &adapter->parent;
975 	struct ice_hw *parent_hw = &parent_adapter->hw;
976 	int ret = 0;
977 
978 	if (!udp_tunnel)
979 		return -EINVAL;
980 
981 	switch (udp_tunnel->prot_type) {
982 	case RTE_TUNNEL_TYPE_VXLAN:
983 		ret = ice_create_tunnel(parent_hw, TNL_VXLAN,
984 					udp_tunnel->udp_port);
985 		break;
986 	case RTE_TUNNEL_TYPE_ECPRI:
987 		ret = ice_create_tunnel(parent_hw, TNL_ECPRI,
988 					udp_tunnel->udp_port);
989 		break;
990 	default:
991 		PMD_DRV_LOG(ERR, "Invalid tunnel type");
992 		ret = -EINVAL;
993 		break;
994 	}
995 
996 	return ret;
997 }
998 
999 /* Delete UDP tunneling port */
1000 static int
1001 ice_dcf_dev_udp_tunnel_port_del(struct rte_eth_dev *dev,
1002 				struct rte_eth_udp_tunnel *udp_tunnel)
1003 {
1004 	struct ice_dcf_adapter *adapter = dev->data->dev_private;
1005 	struct ice_adapter *parent_adapter = &adapter->parent;
1006 	struct ice_hw *parent_hw = &parent_adapter->hw;
1007 	int ret = 0;
1008 
1009 	if (!udp_tunnel)
1010 		return -EINVAL;
1011 
1012 	switch (udp_tunnel->prot_type) {
1013 	case RTE_TUNNEL_TYPE_VXLAN:
1014 	case RTE_TUNNEL_TYPE_ECPRI:
1015 		ret = ice_destroy_tunnel(parent_hw, udp_tunnel->udp_port, 0);
1016 		break;
1017 	default:
1018 		PMD_DRV_LOG(ERR, "Invalid tunnel type");
1019 		ret = -EINVAL;
1020 		break;
1021 	}
1022 
1023 	return ret;
1024 }
1025 
1026 static int
1027 ice_dcf_tm_ops_get(struct rte_eth_dev *dev __rte_unused,
1028 		void *arg)
1029 {
1030 	if (!arg)
1031 		return -EINVAL;
1032 
1033 	*(const void **)arg = &ice_dcf_tm_ops;
1034 
1035 	return 0;
1036 }
1037 
1038 static int
1039 ice_dcf_dev_reset(struct rte_eth_dev *dev)
1040 {
1041 	int ret;
1042 
1043 	ret = ice_dcf_dev_uninit(dev);
1044 	if (ret)
1045 		return ret;
1046 
1047 	ret = ice_dcf_dev_init(dev);
1048 
1049 	return ret;
1050 }
1051 
1052 static const struct eth_dev_ops ice_dcf_eth_dev_ops = {
1053 	.dev_start               = ice_dcf_dev_start,
1054 	.dev_stop                = ice_dcf_dev_stop,
1055 	.dev_close               = ice_dcf_dev_close,
1056 	.dev_reset               = ice_dcf_dev_reset,
1057 	.dev_configure           = ice_dcf_dev_configure,
1058 	.dev_infos_get           = ice_dcf_dev_info_get,
1059 	.rx_queue_setup          = ice_rx_queue_setup,
1060 	.tx_queue_setup          = ice_tx_queue_setup,
1061 	.rx_queue_release        = ice_dev_rx_queue_release,
1062 	.tx_queue_release        = ice_dev_tx_queue_release,
1063 	.rx_queue_start          = ice_dcf_rx_queue_start,
1064 	.tx_queue_start          = ice_dcf_tx_queue_start,
1065 	.rx_queue_stop           = ice_dcf_rx_queue_stop,
1066 	.tx_queue_stop           = ice_dcf_tx_queue_stop,
1067 	.link_update             = ice_dcf_link_update,
1068 	.stats_get               = ice_dcf_stats_get,
1069 	.stats_reset             = ice_dcf_stats_reset,
1070 	.promiscuous_enable      = ice_dcf_dev_promiscuous_enable,
1071 	.promiscuous_disable     = ice_dcf_dev_promiscuous_disable,
1072 	.allmulticast_enable     = ice_dcf_dev_allmulticast_enable,
1073 	.allmulticast_disable    = ice_dcf_dev_allmulticast_disable,
1074 	.flow_ops_get            = ice_dcf_dev_flow_ops_get,
1075 	.udp_tunnel_port_add	 = ice_dcf_dev_udp_tunnel_port_add,
1076 	.udp_tunnel_port_del	 = ice_dcf_dev_udp_tunnel_port_del,
1077 	.tm_ops_get              = ice_dcf_tm_ops_get,
1078 };
1079 
1080 static int
1081 ice_dcf_dev_init(struct rte_eth_dev *eth_dev)
1082 {
1083 	struct ice_dcf_adapter *adapter = eth_dev->data->dev_private;
1084 
1085 	adapter->real_hw.resetting = false;
1086 	eth_dev->dev_ops = &ice_dcf_eth_dev_ops;
1087 	eth_dev->rx_pkt_burst = ice_dcf_recv_pkts;
1088 	eth_dev->tx_pkt_burst = ice_dcf_xmit_pkts;
1089 
1090 	if (rte_eal_process_type() != RTE_PROC_PRIMARY)
1091 		return 0;
1092 
1093 	adapter->real_hw.vc_event_msg_cb = ice_dcf_handle_pf_event_msg;
1094 	if (ice_dcf_init_hw(eth_dev, &adapter->real_hw) != 0) {
1095 		PMD_INIT_LOG(ERR, "Failed to init DCF hardware");
1096 		return -1;
1097 	}
1098 
1099 	if (ice_dcf_init_parent_adapter(eth_dev) != 0) {
1100 		PMD_INIT_LOG(ERR, "Failed to init DCF parent adapter");
1101 		ice_dcf_uninit_hw(eth_dev, &adapter->real_hw);
1102 		return -1;
1103 	}
1104 
1105 	return 0;
1106 }
1107 
1108 static int
1109 ice_dcf_dev_uninit(struct rte_eth_dev *eth_dev)
1110 {
1111 	ice_dcf_dev_close(eth_dev);
1112 
1113 	return 0;
1114 }
1115 
1116 static int
1117 ice_dcf_cap_check_handler(__rte_unused const char *key,
1118 			  const char *value, __rte_unused void *opaque)
1119 {
1120 	if (strcmp(value, "dcf"))
1121 		return -1;
1122 
1123 	return 0;
1124 }
1125 
1126 static int
1127 ice_dcf_cap_selected(struct rte_devargs *devargs)
1128 {
1129 	struct rte_kvargs *kvlist;
1130 	const char *key = "cap";
1131 	int ret = 0;
1132 
1133 	if (devargs == NULL)
1134 		return 0;
1135 
1136 	kvlist = rte_kvargs_parse(devargs->args, NULL);
1137 	if (kvlist == NULL)
1138 		return 0;
1139 
1140 	if (!rte_kvargs_count(kvlist, key))
1141 		goto exit;
1142 
1143 	/* dcf capability selected when there's a key-value pair: cap=dcf */
1144 	if (rte_kvargs_process(kvlist, key,
1145 			       ice_dcf_cap_check_handler, NULL) < 0)
1146 		goto exit;
1147 
1148 	ret = 1;
1149 
1150 exit:
1151 	rte_kvargs_free(kvlist);
1152 	return ret;
1153 }
1154 
1155 static int
1156 eth_ice_dcf_pci_probe(__rte_unused struct rte_pci_driver *pci_drv,
1157 		      struct rte_pci_device *pci_dev)
1158 {
1159 	struct rte_eth_devargs eth_da = { .nb_representor_ports = 0 };
1160 	struct ice_dcf_vf_repr_param repr_param;
1161 	char repr_name[RTE_ETH_NAME_MAX_LEN];
1162 	struct ice_dcf_adapter *dcf_adapter;
1163 	struct rte_eth_dev *dcf_ethdev;
1164 	uint16_t dcf_vsi_id;
1165 	int i, ret;
1166 
1167 	if (!ice_dcf_cap_selected(pci_dev->device.devargs))
1168 		return 1;
1169 
1170 	ret = rte_eth_devargs_parse(pci_dev->device.devargs->args, &eth_da);
1171 	if (ret)
1172 		return ret;
1173 
1174 	ret = rte_eth_dev_pci_generic_probe(pci_dev,
1175 					    sizeof(struct ice_dcf_adapter),
1176 					    ice_dcf_dev_init);
1177 	if (ret || !eth_da.nb_representor_ports)
1178 		return ret;
1179 	if (eth_da.type != RTE_ETH_REPRESENTOR_VF)
1180 		return -ENOTSUP;
1181 
1182 	dcf_ethdev = rte_eth_dev_allocated(pci_dev->device.name);
1183 	if (dcf_ethdev == NULL)
1184 		return -ENODEV;
1185 
1186 	dcf_adapter = dcf_ethdev->data->dev_private;
1187 	ret = ice_dcf_init_repr_info(dcf_adapter);
1188 	if (ret)
1189 		return ret;
1190 
1191 	if (eth_da.nb_representor_ports > dcf_adapter->real_hw.num_vfs ||
1192 	    eth_da.nb_representor_ports >= RTE_MAX_ETHPORTS) {
1193 		PMD_DRV_LOG(ERR, "the number of port representors is too large: %u",
1194 			    eth_da.nb_representor_ports);
1195 		ice_dcf_free_repr_info(dcf_adapter);
1196 		return -EINVAL;
1197 	}
1198 
1199 	dcf_vsi_id = dcf_adapter->real_hw.vsi_id | VIRTCHNL_DCF_VF_VSI_VALID;
1200 
1201 	repr_param.dcf_eth_dev = dcf_ethdev;
1202 	repr_param.switch_domain_id = 0;
1203 
1204 	for (i = 0; i < eth_da.nb_representor_ports; i++) {
1205 		uint16_t vf_id = eth_da.representor_ports[i];
1206 		struct rte_eth_dev *vf_rep_eth_dev;
1207 
1208 		if (vf_id >= dcf_adapter->real_hw.num_vfs) {
1209 			PMD_DRV_LOG(ERR, "VF ID %u is out of range (0 ~ %u)",
1210 				    vf_id, dcf_adapter->real_hw.num_vfs - 1);
1211 			ret = -EINVAL;
1212 			break;
1213 		}
1214 
1215 		if (dcf_adapter->real_hw.vf_vsi_map[vf_id] == dcf_vsi_id) {
1216 			PMD_DRV_LOG(ERR, "VF ID %u is DCF's ID.\n", vf_id);
1217 			ret = -EINVAL;
1218 			break;
1219 		}
1220 
1221 		repr_param.vf_id = vf_id;
1222 		snprintf(repr_name, sizeof(repr_name), "net_%s_representor_%u",
1223 			 pci_dev->device.name, vf_id);
1224 		ret = rte_eth_dev_create(&pci_dev->device, repr_name,
1225 					 sizeof(struct ice_dcf_vf_repr),
1226 					 NULL, NULL, ice_dcf_vf_repr_init,
1227 					 &repr_param);
1228 		if (ret) {
1229 			PMD_DRV_LOG(ERR, "failed to create DCF VF representor %s",
1230 				    repr_name);
1231 			break;
1232 		}
1233 
1234 		vf_rep_eth_dev = rte_eth_dev_allocated(repr_name);
1235 		if (!vf_rep_eth_dev) {
1236 			PMD_DRV_LOG(ERR,
1237 				    "Failed to find the ethdev for DCF VF representor: %s",
1238 				    repr_name);
1239 			ret = -ENODEV;
1240 			break;
1241 		}
1242 
1243 		dcf_adapter->repr_infos[vf_id].vf_rep_eth_dev = vf_rep_eth_dev;
1244 		dcf_adapter->num_reprs++;
1245 	}
1246 
1247 	return ret;
1248 }
1249 
1250 static int
1251 eth_ice_dcf_pci_remove(struct rte_pci_device *pci_dev)
1252 {
1253 	struct rte_eth_dev *eth_dev;
1254 
1255 	eth_dev = rte_eth_dev_allocated(pci_dev->device.name);
1256 	if (!eth_dev)
1257 		return 0;
1258 
1259 	if (eth_dev->data->dev_flags & RTE_ETH_DEV_REPRESENTOR)
1260 		return rte_eth_dev_pci_generic_remove(pci_dev,
1261 						      ice_dcf_vf_repr_uninit);
1262 	else
1263 		return rte_eth_dev_pci_generic_remove(pci_dev,
1264 						      ice_dcf_dev_uninit);
1265 }
1266 
1267 static const struct rte_pci_id pci_id_ice_dcf_map[] = {
1268 	{ RTE_PCI_DEVICE(IAVF_INTEL_VENDOR_ID, IAVF_DEV_ID_ADAPTIVE_VF) },
1269 	{ .vendor_id = 0, /* sentinel */ },
1270 };
1271 
1272 static struct rte_pci_driver rte_ice_dcf_pmd = {
1273 	.id_table = pci_id_ice_dcf_map,
1274 	.drv_flags = RTE_PCI_DRV_NEED_MAPPING,
1275 	.probe = eth_ice_dcf_pci_probe,
1276 	.remove = eth_ice_dcf_pci_remove,
1277 };
1278 
1279 RTE_PMD_REGISTER_PCI(net_ice_dcf, rte_ice_dcf_pmd);
1280 RTE_PMD_REGISTER_PCI_TABLE(net_ice_dcf, pci_id_ice_dcf_map);
1281 RTE_PMD_REGISTER_KMOD_DEP(net_ice_dcf, "* igb_uio | vfio-pci");
1282 RTE_PMD_REGISTER_PARAM_STRING(net_ice_dcf, "cap=dcf");
1283