xref: /dpdk/kernel/linux/kni/kni_net.c (revision 29fd052d)
1 // SPDX-License-Identifier: GPL-2.0
2 /*
3  * Copyright(c) 2010-2014 Intel Corporation.
4  */
5 
6 /*
7  * This code is inspired from the book "Linux Device Drivers" by
8  * Alessandro Rubini and Jonathan Corbet, published by O'Reilly & Associates
9  */
10 
11 #include <linux/device.h>
12 #include <linux/module.h>
13 #include <linux/version.h>
14 #include <linux/netdevice.h>
15 #include <linux/etherdevice.h> /* eth_type_trans */
16 #include <linux/ethtool.h>
17 #include <linux/skbuff.h>
18 #include <linux/kthread.h>
19 #include <linux/delay.h>
20 #include <linux/rtnetlink.h>
21 
22 #include <rte_kni_common.h>
23 #include <kni_fifo.h>
24 
25 #include "compat.h"
26 #include "kni_dev.h"
27 
28 #define WD_TIMEOUT 5 /*jiffies */
29 
30 #define KNI_WAIT_RESPONSE_TIMEOUT 300 /* 3 seconds */
31 
32 /* typedef for rx function */
33 typedef void (*kni_net_rx_t)(struct kni_dev *kni);
34 
35 static void kni_net_rx_normal(struct kni_dev *kni);
36 
37 /* kni rx function pointer, with default to normal rx */
38 static kni_net_rx_t kni_net_rx_func = kni_net_rx_normal;
39 
40 #ifdef HAVE_IOVA_TO_KVA_MAPPING_SUPPORT
41 /* iova to kernel virtual address */
42 static inline void *
43 iova2kva(struct kni_dev *kni, void *iova)
44 {
45 	return phys_to_virt(iova_to_phys(kni->usr_tsk, (unsigned long)iova));
46 }
47 
48 static inline void *
49 iova2data_kva(struct kni_dev *kni, struct rte_kni_mbuf *m)
50 {
51 	return phys_to_virt(iova_to_phys(kni->usr_tsk, m->buf_iova) +
52 			    m->data_off);
53 }
54 #endif
55 
56 /* physical address to kernel virtual address */
57 static void *
58 pa2kva(void *pa)
59 {
60 	return phys_to_virt((unsigned long)pa);
61 }
62 
63 /* physical address to virtual address */
64 static void *
65 pa2va(void *pa, struct rte_kni_mbuf *m)
66 {
67 	void *va;
68 
69 	va = (void *)((unsigned long)pa +
70 			(unsigned long)m->buf_addr -
71 			(unsigned long)m->buf_iova);
72 	return va;
73 }
74 
75 /* mbuf data kernel virtual address from mbuf kernel virtual address */
76 static void *
77 kva2data_kva(struct rte_kni_mbuf *m)
78 {
79 	return phys_to_virt(m->buf_iova + m->data_off);
80 }
81 
82 static inline void *
83 get_kva(struct kni_dev *kni, void *pa)
84 {
85 #ifdef HAVE_IOVA_TO_KVA_MAPPING_SUPPORT
86 	if (kni->iova_mode == 1)
87 		return iova2kva(kni, pa);
88 #endif
89 	return pa2kva(pa);
90 }
91 
92 static inline void *
93 get_data_kva(struct kni_dev *kni, void *pkt_kva)
94 {
95 #ifdef HAVE_IOVA_TO_KVA_MAPPING_SUPPORT
96 	if (kni->iova_mode == 1)
97 		return iova2data_kva(kni, pkt_kva);
98 #endif
99 	return kva2data_kva(pkt_kva);
100 }
101 
102 /*
103  * It can be called to process the request.
104  */
105 static int
106 kni_net_process_request(struct net_device *dev, struct rte_kni_request *req)
107 {
108 	struct kni_dev *kni = netdev_priv(dev);
109 	int ret = -1;
110 	void *resp_va;
111 	uint32_t num;
112 	int ret_val;
113 
114 	ASSERT_RTNL();
115 
116 	if (bifurcated_support) {
117 		/* If we need to wait and RTNL mutex is held
118 		 * drop the mutex and hold reference to keep device
119 		 */
120 		if (req->async == 0) {
121 			dev_hold(dev);
122 			rtnl_unlock();
123 		}
124 	}
125 
126 	mutex_lock(&kni->sync_lock);
127 
128 	/* Construct data */
129 	memcpy(kni->sync_kva, req, sizeof(struct rte_kni_request));
130 	num = kni_fifo_put(kni->req_q, &kni->sync_va, 1);
131 	if (num < 1) {
132 		pr_err("Cannot send to req_q\n");
133 		ret = -EBUSY;
134 		goto fail;
135 	}
136 
137 	if (bifurcated_support) {
138 		/* No result available since request is handled
139 		 * asynchronously. set response to success.
140 		 */
141 		if (req->async != 0) {
142 			req->result = 0;
143 			goto async;
144 		}
145 	}
146 
147 	ret_val = wait_event_interruptible_timeout(kni->wq,
148 			kni_fifo_count(kni->resp_q), 3 * HZ);
149 	if (signal_pending(current) || ret_val <= 0) {
150 		ret = -ETIME;
151 		goto fail;
152 	}
153 	num = kni_fifo_get(kni->resp_q, (void **)&resp_va, 1);
154 	if (num != 1 || resp_va != kni->sync_va) {
155 		/* This should never happen */
156 		pr_err("No data in resp_q\n");
157 		ret = -ENODATA;
158 		goto fail;
159 	}
160 
161 	memcpy(req, kni->sync_kva, sizeof(struct rte_kni_request));
162 async:
163 	ret = 0;
164 
165 fail:
166 	mutex_unlock(&kni->sync_lock);
167 	if (bifurcated_support) {
168 		if (req->async == 0) {
169 			rtnl_lock();
170 			dev_put(dev);
171 		}
172 	}
173 	return ret;
174 }
175 
176 /*
177  * Open and close
178  */
179 static int
180 kni_net_open(struct net_device *dev)
181 {
182 	int ret;
183 	struct rte_kni_request req;
184 
185 	netif_start_queue(dev);
186 	if (kni_dflt_carrier == 1)
187 		netif_carrier_on(dev);
188 	else
189 		netif_carrier_off(dev);
190 
191 	memset(&req, 0, sizeof(req));
192 	req.req_id = RTE_KNI_REQ_CFG_NETWORK_IF;
193 
194 	/* Setting if_up to non-zero means up */
195 	req.if_up = 1;
196 	ret = kni_net_process_request(dev, &req);
197 
198 	return (ret == 0) ? req.result : ret;
199 }
200 
201 static int
202 kni_net_release(struct net_device *dev)
203 {
204 	int ret;
205 	struct rte_kni_request req;
206 
207 	netif_stop_queue(dev); /* can't transmit any more */
208 	netif_carrier_off(dev);
209 
210 	memset(&req, 0, sizeof(req));
211 	req.req_id = RTE_KNI_REQ_CFG_NETWORK_IF;
212 
213 	/* Setting if_up to 0 means down */
214 	req.if_up = 0;
215 
216 	if (bifurcated_support) {
217 		/* request async because of the deadlock problem */
218 		req.async = 1;
219 	}
220 
221 	ret = kni_net_process_request(dev, &req);
222 
223 	return (ret == 0) ? req.result : ret;
224 }
225 
226 static void
227 kni_fifo_trans_pa2va(struct kni_dev *kni,
228 	struct rte_kni_fifo *src_pa, struct rte_kni_fifo *dst_va)
229 {
230 	uint32_t ret, i, num_dst, num_rx;
231 	struct rte_kni_mbuf *kva, *prev_kva;
232 	int nb_segs;
233 	int kva_nb_segs;
234 
235 	do {
236 		num_dst = kni_fifo_free_count(dst_va);
237 		if (num_dst == 0)
238 			return;
239 
240 		num_rx = min_t(uint32_t, num_dst, MBUF_BURST_SZ);
241 
242 		num_rx = kni_fifo_get(src_pa, kni->pa, num_rx);
243 		if (num_rx == 0)
244 			return;
245 
246 		for (i = 0; i < num_rx; i++) {
247 			kva = get_kva(kni, kni->pa[i]);
248 			kni->va[i] = pa2va(kni->pa[i], kva);
249 
250 			kva_nb_segs = kva->nb_segs;
251 			for (nb_segs = 0; nb_segs < kva_nb_segs; nb_segs++) {
252 				if (!kva->next)
253 					break;
254 
255 				prev_kva = kva;
256 				kva = get_kva(kni, kva->next);
257 				/* Convert physical address to virtual address */
258 				prev_kva->next = pa2va(prev_kva->next, kva);
259 			}
260 		}
261 
262 		ret = kni_fifo_put(dst_va, kni->va, num_rx);
263 		if (ret != num_rx) {
264 			/* Failing should not happen */
265 			pr_err("Fail to enqueue entries into dst_va\n");
266 			return;
267 		}
268 	} while (1);
269 }
270 
271 /* Try to release mbufs when kni release */
272 void kni_net_release_fifo_phy(struct kni_dev *kni)
273 {
274 	/* release rx_q first, because it can't release in userspace */
275 	kni_fifo_trans_pa2va(kni, kni->rx_q, kni->free_q);
276 	/* release alloc_q for speeding up kni release in userspace */
277 	kni_fifo_trans_pa2va(kni, kni->alloc_q, kni->free_q);
278 }
279 
280 /*
281  * Configuration changes (passed on by ifconfig)
282  */
283 static int
284 kni_net_config(struct net_device *dev, struct ifmap *map)
285 {
286 	if (dev->flags & IFF_UP) /* can't act on a running interface */
287 		return -EBUSY;
288 
289 	/* ignore other fields */
290 	return 0;
291 }
292 
293 /*
294  * Transmit a packet (called by the kernel)
295  */
296 static int
297 kni_net_tx(struct sk_buff *skb, struct net_device *dev)
298 {
299 	int len = 0;
300 	uint32_t ret;
301 	struct kni_dev *kni = netdev_priv(dev);
302 	struct rte_kni_mbuf *pkt_kva = NULL;
303 	void *pkt_pa = NULL;
304 	void *pkt_va = NULL;
305 
306 	/* save the timestamp */
307 #ifdef HAVE_TRANS_START_HELPER
308 	netif_trans_update(dev);
309 #else
310 	dev->trans_start = jiffies;
311 #endif
312 
313 	/* Check if the length of skb is less than mbuf size */
314 	if (skb->len > kni->mbuf_size)
315 		goto drop;
316 
317 	/**
318 	 * Check if it has at least one free entry in tx_q and
319 	 * one entry in alloc_q.
320 	 */
321 	if (kni_fifo_free_count(kni->tx_q) == 0 ||
322 			kni_fifo_count(kni->alloc_q) == 0) {
323 		/**
324 		 * If no free entry in tx_q or no entry in alloc_q,
325 		 * drops skb and goes out.
326 		 */
327 		goto drop;
328 	}
329 
330 	/* dequeue a mbuf from alloc_q */
331 	ret = kni_fifo_get(kni->alloc_q, &pkt_pa, 1);
332 	if (likely(ret == 1)) {
333 		void *data_kva;
334 
335 		pkt_kva = get_kva(kni, pkt_pa);
336 		data_kva = get_data_kva(kni, pkt_kva);
337 		pkt_va = pa2va(pkt_pa, pkt_kva);
338 
339 		len = skb->len;
340 		memcpy(data_kva, skb->data, len);
341 		if (unlikely(len < ETH_ZLEN)) {
342 			memset(data_kva + len, 0, ETH_ZLEN - len);
343 			len = ETH_ZLEN;
344 		}
345 		pkt_kva->pkt_len = len;
346 		pkt_kva->data_len = len;
347 
348 		/* enqueue mbuf into tx_q */
349 		ret = kni_fifo_put(kni->tx_q, &pkt_va, 1);
350 		if (unlikely(ret != 1)) {
351 			/* Failing should not happen */
352 			pr_err("Fail to enqueue mbuf into tx_q\n");
353 			goto drop;
354 		}
355 	} else {
356 		/* Failing should not happen */
357 		pr_err("Fail to dequeue mbuf from alloc_q\n");
358 		goto drop;
359 	}
360 
361 	/* Free skb and update statistics */
362 	dev_kfree_skb(skb);
363 	dev->stats.tx_bytes += len;
364 	dev->stats.tx_packets++;
365 
366 	return NETDEV_TX_OK;
367 
368 drop:
369 	/* Free skb and update statistics */
370 	dev_kfree_skb(skb);
371 	dev->stats.tx_dropped++;
372 
373 	return NETDEV_TX_OK;
374 }
375 
376 /*
377  * RX: normal working mode
378  */
379 static void
380 kni_net_rx_normal(struct kni_dev *kni)
381 {
382 	uint32_t ret;
383 	uint32_t len;
384 	uint32_t i, num_rx, num_fq;
385 	struct rte_kni_mbuf *kva, *prev_kva;
386 	void *data_kva;
387 	struct sk_buff *skb;
388 	struct net_device *dev = kni->net_dev;
389 
390 	/* Get the number of free entries in free_q */
391 	num_fq = kni_fifo_free_count(kni->free_q);
392 	if (num_fq == 0) {
393 		/* No room on the free_q, bail out */
394 		return;
395 	}
396 
397 	/* Calculate the number of entries to dequeue from rx_q */
398 	num_rx = min_t(uint32_t, num_fq, MBUF_BURST_SZ);
399 
400 	/* Burst dequeue from rx_q */
401 	num_rx = kni_fifo_get(kni->rx_q, kni->pa, num_rx);
402 	if (num_rx == 0)
403 		return;
404 
405 	/* Transfer received packets to netif */
406 	for (i = 0; i < num_rx; i++) {
407 		kva = get_kva(kni, kni->pa[i]);
408 		len = kva->pkt_len;
409 		data_kva = get_data_kva(kni, kva);
410 		kni->va[i] = pa2va(kni->pa[i], kva);
411 
412 		skb = netdev_alloc_skb(dev, len);
413 		if (!skb) {
414 			/* Update statistics */
415 			dev->stats.rx_dropped++;
416 			continue;
417 		}
418 
419 		if (kva->nb_segs == 1) {
420 			memcpy(skb_put(skb, len), data_kva, len);
421 		} else {
422 			int nb_segs;
423 			int kva_nb_segs = kva->nb_segs;
424 
425 			for (nb_segs = 0; nb_segs < kva_nb_segs; nb_segs++) {
426 				memcpy(skb_put(skb, kva->data_len),
427 					data_kva, kva->data_len);
428 
429 				if (!kva->next)
430 					break;
431 
432 				prev_kva = kva;
433 				kva = get_kva(kni, kva->next);
434 				data_kva = kva2data_kva(kva);
435 				/* Convert physical address to virtual address */
436 				prev_kva->next = pa2va(prev_kva->next, kva);
437 			}
438 		}
439 
440 		skb->protocol = eth_type_trans(skb, dev);
441 		skb->ip_summed = CHECKSUM_UNNECESSARY;
442 
443 		/* Call netif interface */
444 		netif_rx_ni(skb);
445 
446 		/* Update statistics */
447 		dev->stats.rx_bytes += len;
448 		dev->stats.rx_packets++;
449 	}
450 
451 	/* Burst enqueue mbufs into free_q */
452 	ret = kni_fifo_put(kni->free_q, kni->va, num_rx);
453 	if (ret != num_rx)
454 		/* Failing should not happen */
455 		pr_err("Fail to enqueue entries into free_q\n");
456 }
457 
458 /*
459  * RX: loopback with enqueue/dequeue fifos.
460  */
461 static void
462 kni_net_rx_lo_fifo(struct kni_dev *kni)
463 {
464 	uint32_t ret;
465 	uint32_t len;
466 	uint32_t i, num, num_rq, num_tq, num_aq, num_fq;
467 	struct rte_kni_mbuf *kva, *next_kva;
468 	void *data_kva;
469 	struct rte_kni_mbuf *alloc_kva;
470 	void *alloc_data_kva;
471 	struct net_device *dev = kni->net_dev;
472 
473 	/* Get the number of entries in rx_q */
474 	num_rq = kni_fifo_count(kni->rx_q);
475 
476 	/* Get the number of free entries in tx_q */
477 	num_tq = kni_fifo_free_count(kni->tx_q);
478 
479 	/* Get the number of entries in alloc_q */
480 	num_aq = kni_fifo_count(kni->alloc_q);
481 
482 	/* Get the number of free entries in free_q */
483 	num_fq = kni_fifo_free_count(kni->free_q);
484 
485 	/* Calculate the number of entries to be dequeued from rx_q */
486 	num = min(num_rq, num_tq);
487 	num = min(num, num_aq);
488 	num = min(num, num_fq);
489 	num = min_t(uint32_t, num, MBUF_BURST_SZ);
490 
491 	/* Return if no entry to dequeue from rx_q */
492 	if (num == 0)
493 		return;
494 
495 	/* Burst dequeue from rx_q */
496 	ret = kni_fifo_get(kni->rx_q, kni->pa, num);
497 	if (ret == 0)
498 		return; /* Failing should not happen */
499 
500 	/* Dequeue entries from alloc_q */
501 	ret = kni_fifo_get(kni->alloc_q, kni->alloc_pa, num);
502 	if (ret) {
503 		num = ret;
504 		/* Copy mbufs */
505 		for (i = 0; i < num; i++) {
506 			kva = get_kva(kni, kni->pa[i]);
507 			len = kva->data_len;
508 			data_kva = get_data_kva(kni, kva);
509 			kni->va[i] = pa2va(kni->pa[i], kva);
510 
511 			while (kva->next) {
512 				next_kva = get_kva(kni, kva->next);
513 				/* Convert physical address to virtual address */
514 				kva->next = pa2va(kva->next, next_kva);
515 				kva = next_kva;
516 			}
517 
518 			alloc_kva = get_kva(kni, kni->alloc_pa[i]);
519 			alloc_data_kva = get_data_kva(kni, alloc_kva);
520 			kni->alloc_va[i] = pa2va(kni->alloc_pa[i], alloc_kva);
521 
522 			memcpy(alloc_data_kva, data_kva, len);
523 			alloc_kva->pkt_len = len;
524 			alloc_kva->data_len = len;
525 
526 			dev->stats.tx_bytes += len;
527 			dev->stats.rx_bytes += len;
528 		}
529 
530 		/* Burst enqueue mbufs into tx_q */
531 		ret = kni_fifo_put(kni->tx_q, kni->alloc_va, num);
532 		if (ret != num)
533 			/* Failing should not happen */
534 			pr_err("Fail to enqueue mbufs into tx_q\n");
535 	}
536 
537 	/* Burst enqueue mbufs into free_q */
538 	ret = kni_fifo_put(kni->free_q, kni->va, num);
539 	if (ret != num)
540 		/* Failing should not happen */
541 		pr_err("Fail to enqueue mbufs into free_q\n");
542 
543 	/**
544 	 * Update statistic, and enqueue/dequeue failure is impossible,
545 	 * as all queues are checked at first.
546 	 */
547 	dev->stats.tx_packets += num;
548 	dev->stats.rx_packets += num;
549 }
550 
551 /*
552  * RX: loopback with enqueue/dequeue fifos and sk buffer copies.
553  */
554 static void
555 kni_net_rx_lo_fifo_skb(struct kni_dev *kni)
556 {
557 	uint32_t ret;
558 	uint32_t len;
559 	uint32_t i, num_rq, num_fq, num;
560 	struct rte_kni_mbuf *kva, *prev_kva;
561 	void *data_kva;
562 	struct sk_buff *skb;
563 	struct net_device *dev = kni->net_dev;
564 
565 	/* Get the number of entries in rx_q */
566 	num_rq = kni_fifo_count(kni->rx_q);
567 
568 	/* Get the number of free entries in free_q */
569 	num_fq = kni_fifo_free_count(kni->free_q);
570 
571 	/* Calculate the number of entries to dequeue from rx_q */
572 	num = min(num_rq, num_fq);
573 	num = min_t(uint32_t, num, MBUF_BURST_SZ);
574 
575 	/* Return if no entry to dequeue from rx_q */
576 	if (num == 0)
577 		return;
578 
579 	/* Burst dequeue mbufs from rx_q */
580 	ret = kni_fifo_get(kni->rx_q, kni->pa, num);
581 	if (ret == 0)
582 		return;
583 
584 	/* Copy mbufs to sk buffer and then call tx interface */
585 	for (i = 0; i < num; i++) {
586 		kva = get_kva(kni, kni->pa[i]);
587 		len = kva->pkt_len;
588 		data_kva = get_data_kva(kni, kva);
589 		kni->va[i] = pa2va(kni->pa[i], kva);
590 
591 		skb = netdev_alloc_skb(dev, len);
592 		if (skb) {
593 			memcpy(skb_put(skb, len), data_kva, len);
594 			skb->ip_summed = CHECKSUM_UNNECESSARY;
595 			dev_kfree_skb(skb);
596 		}
597 
598 		/* Simulate real usage, allocate/copy skb twice */
599 		skb = netdev_alloc_skb(dev, len);
600 		if (skb == NULL) {
601 			dev->stats.rx_dropped++;
602 			continue;
603 		}
604 
605 		if (kva->nb_segs == 1) {
606 			memcpy(skb_put(skb, len), data_kva, len);
607 		} else {
608 			int nb_segs;
609 			int kva_nb_segs = kva->nb_segs;
610 
611 			for (nb_segs = 0; nb_segs < kva_nb_segs; nb_segs++) {
612 				memcpy(skb_put(skb, kva->data_len),
613 					data_kva, kva->data_len);
614 
615 				if (!kva->next)
616 					break;
617 
618 				prev_kva = kva;
619 				kva = get_kva(kni, kva->next);
620 				data_kva = get_data_kva(kni, kva);
621 				/* Convert physical address to virtual address */
622 				prev_kva->next = pa2va(prev_kva->next, kva);
623 			}
624 		}
625 
626 		skb->ip_summed = CHECKSUM_UNNECESSARY;
627 
628 		dev->stats.rx_bytes += len;
629 		dev->stats.rx_packets++;
630 
631 		/* call tx interface */
632 		kni_net_tx(skb, dev);
633 	}
634 
635 	/* enqueue all the mbufs from rx_q into free_q */
636 	ret = kni_fifo_put(kni->free_q, kni->va, num);
637 	if (ret != num)
638 		/* Failing should not happen */
639 		pr_err("Fail to enqueue mbufs into free_q\n");
640 }
641 
642 /* rx interface */
643 void
644 kni_net_rx(struct kni_dev *kni)
645 {
646 	/**
647 	 * It doesn't need to check if it is NULL pointer,
648 	 * as it has a default value
649 	 */
650 	(*kni_net_rx_func)(kni);
651 }
652 
653 /*
654  * Deal with a transmit timeout.
655  */
656 #ifdef HAVE_TX_TIMEOUT_TXQUEUE
657 static void
658 kni_net_tx_timeout(struct net_device *dev, unsigned int txqueue)
659 #else
660 static void
661 kni_net_tx_timeout(struct net_device *dev)
662 #endif
663 {
664 	pr_debug("Transmit timeout at %ld, latency %ld\n", jiffies,
665 			jiffies - dev_trans_start(dev));
666 
667 	dev->stats.tx_errors++;
668 	netif_wake_queue(dev);
669 }
670 
671 static int
672 kni_net_change_mtu(struct net_device *dev, int new_mtu)
673 {
674 	int ret;
675 	struct rte_kni_request req;
676 
677 	pr_debug("kni_net_change_mtu new mtu %d to be set\n", new_mtu);
678 
679 	memset(&req, 0, sizeof(req));
680 	req.req_id = RTE_KNI_REQ_CHANGE_MTU;
681 	req.new_mtu = new_mtu;
682 	ret = kni_net_process_request(dev, &req);
683 	if (ret == 0 && req.result == 0)
684 		dev->mtu = new_mtu;
685 
686 	return (ret == 0) ? req.result : ret;
687 }
688 
689 static void
690 kni_net_change_rx_flags(struct net_device *netdev, int flags)
691 {
692 	struct rte_kni_request req;
693 
694 	memset(&req, 0, sizeof(req));
695 
696 	if (flags & IFF_ALLMULTI) {
697 		req.req_id = RTE_KNI_REQ_CHANGE_ALLMULTI;
698 
699 		if (netdev->flags & IFF_ALLMULTI)
700 			req.allmulti = 1;
701 		else
702 			req.allmulti = 0;
703 	}
704 
705 	if (flags & IFF_PROMISC) {
706 		req.req_id = RTE_KNI_REQ_CHANGE_PROMISC;
707 
708 		if (netdev->flags & IFF_PROMISC)
709 			req.promiscusity = 1;
710 		else
711 			req.promiscusity = 0;
712 	}
713 
714 	kni_net_process_request(netdev, &req);
715 }
716 
717 /*
718  * Checks if the user space application provided the resp message
719  */
720 void
721 kni_net_poll_resp(struct kni_dev *kni)
722 {
723 	if (kni_fifo_count(kni->resp_q))
724 		wake_up_interruptible(&kni->wq);
725 }
726 
727 /*
728  *  Fill the eth header
729  */
730 static int
731 kni_net_header(struct sk_buff *skb, struct net_device *dev,
732 		unsigned short type, const void *daddr,
733 		const void *saddr, uint32_t len)
734 {
735 	struct ethhdr *eth = (struct ethhdr *) skb_push(skb, ETH_HLEN);
736 
737 	memcpy(eth->h_source, saddr ? saddr : dev->dev_addr, dev->addr_len);
738 	memcpy(eth->h_dest,   daddr ? daddr : dev->dev_addr, dev->addr_len);
739 	eth->h_proto = htons(type);
740 
741 	return dev->hard_header_len;
742 }
743 
744 /*
745  * Re-fill the eth header
746  */
747 #ifdef HAVE_REBUILD_HEADER
748 static int
749 kni_net_rebuild_header(struct sk_buff *skb)
750 {
751 	struct net_device *dev = skb->dev;
752 	struct ethhdr *eth = (struct ethhdr *) skb->data;
753 
754 	memcpy(eth->h_source, dev->dev_addr, dev->addr_len);
755 	memcpy(eth->h_dest, dev->dev_addr, dev->addr_len);
756 
757 	return 0;
758 }
759 #endif /* < 4.1.0  */
760 
761 /**
762  * kni_net_set_mac - Change the Ethernet Address of the KNI NIC
763  * @netdev: network interface device structure
764  * @p: pointer to an address structure
765  *
766  * Returns 0 on success, negative on failure
767  **/
768 static int
769 kni_net_set_mac(struct net_device *netdev, void *p)
770 {
771 	int ret;
772 	struct rte_kni_request req;
773 	struct sockaddr *addr = p;
774 
775 	memset(&req, 0, sizeof(req));
776 	req.req_id = RTE_KNI_REQ_CHANGE_MAC_ADDR;
777 
778 	if (!is_valid_ether_addr((unsigned char *)(addr->sa_data)))
779 		return -EADDRNOTAVAIL;
780 
781 	memcpy(req.mac_addr, addr->sa_data, netdev->addr_len);
782 	memcpy(netdev->dev_addr, addr->sa_data, netdev->addr_len);
783 
784 	ret = kni_net_process_request(netdev, &req);
785 
786 	return (ret == 0 ? req.result : ret);
787 }
788 
789 #ifdef HAVE_CHANGE_CARRIER_CB
790 static int
791 kni_net_change_carrier(struct net_device *dev, bool new_carrier)
792 {
793 	if (new_carrier)
794 		netif_carrier_on(dev);
795 	else
796 		netif_carrier_off(dev);
797 	return 0;
798 }
799 #endif
800 
801 static const struct header_ops kni_net_header_ops = {
802 	.create  = kni_net_header,
803 	.parse   = eth_header_parse,
804 #ifdef HAVE_REBUILD_HEADER
805 	.rebuild = kni_net_rebuild_header,
806 #endif /* < 4.1.0  */
807 	.cache   = NULL,  /* disable caching */
808 };
809 
810 static const struct net_device_ops kni_net_netdev_ops = {
811 	.ndo_open = kni_net_open,
812 	.ndo_stop = kni_net_release,
813 	.ndo_set_config = kni_net_config,
814 	.ndo_change_rx_flags = kni_net_change_rx_flags,
815 	.ndo_start_xmit = kni_net_tx,
816 	.ndo_change_mtu = kni_net_change_mtu,
817 	.ndo_tx_timeout = kni_net_tx_timeout,
818 	.ndo_set_mac_address = kni_net_set_mac,
819 #ifdef HAVE_CHANGE_CARRIER_CB
820 	.ndo_change_carrier = kni_net_change_carrier,
821 #endif
822 };
823 
824 static void kni_get_drvinfo(struct net_device *dev,
825 			    struct ethtool_drvinfo *info)
826 {
827 	strlcpy(info->version, KNI_VERSION, sizeof(info->version));
828 	strlcpy(info->driver, "kni", sizeof(info->driver));
829 }
830 
831 static const struct ethtool_ops kni_net_ethtool_ops = {
832 	.get_drvinfo	= kni_get_drvinfo,
833 	.get_link	= ethtool_op_get_link,
834 };
835 
836 void
837 kni_net_init(struct net_device *dev)
838 {
839 	struct kni_dev *kni = netdev_priv(dev);
840 
841 	init_waitqueue_head(&kni->wq);
842 	mutex_init(&kni->sync_lock);
843 
844 	ether_setup(dev); /* assign some of the fields */
845 	dev->netdev_ops      = &kni_net_netdev_ops;
846 	dev->header_ops      = &kni_net_header_ops;
847 	dev->ethtool_ops     = &kni_net_ethtool_ops;
848 	dev->watchdog_timeo = WD_TIMEOUT;
849 }
850 
851 void
852 kni_net_config_lo_mode(char *lo_str)
853 {
854 	if (!lo_str) {
855 		pr_debug("loopback disabled");
856 		return;
857 	}
858 
859 	if (!strcmp(lo_str, "lo_mode_none"))
860 		pr_debug("loopback disabled");
861 	else if (!strcmp(lo_str, "lo_mode_fifo")) {
862 		pr_debug("loopback mode=lo_mode_fifo enabled");
863 		kni_net_rx_func = kni_net_rx_lo_fifo;
864 	} else if (!strcmp(lo_str, "lo_mode_fifo_skb")) {
865 		pr_debug("loopback mode=lo_mode_fifo_skb enabled");
866 		kni_net_rx_func = kni_net_rx_lo_fifo_skb;
867 	} else {
868 		pr_debug("Unknown loopback parameter, disabled");
869 	}
870 }
871