xref: /linux/drivers/net/virtio_net.c (revision 4c630f307455c06f99bdeca7f7a1ab5318604fe0)
1 // SPDX-License-Identifier: GPL-2.0-or-later
2 /* A network driver using virtio.
3  *
4  * Copyright 2007 Rusty Russell <rusty@rustcorp.com.au> IBM Corporation
5  */
6 //#define DEBUG
7 #include <linux/netdevice.h>
8 #include <linux/etherdevice.h>
9 #include <linux/ethtool.h>
10 #include <linux/module.h>
11 #include <linux/virtio.h>
12 #include <linux/virtio_net.h>
13 #include <linux/bpf.h>
14 #include <linux/bpf_trace.h>
15 #include <linux/scatterlist.h>
16 #include <linux/if_vlan.h>
17 #include <linux/slab.h>
18 #include <linux/cpu.h>
19 #include <linux/average.h>
20 #include <linux/filter.h>
21 #include <linux/kernel.h>
22 #include <net/route.h>
23 #include <net/xdp.h>
24 #include <net/net_failover.h>
25 
26 static int napi_weight = NAPI_POLL_WEIGHT;
27 module_param(napi_weight, int, 0444);
28 
29 static bool csum = true, gso = true, napi_tx = true;
30 module_param(csum, bool, 0444);
31 module_param(gso, bool, 0444);
32 module_param(napi_tx, bool, 0644);
33 
34 /* FIXME: MTU in config. */
35 #define GOOD_PACKET_LEN (ETH_HLEN + VLAN_HLEN + ETH_DATA_LEN)
36 #define GOOD_COPY_LEN	128
37 
38 #define VIRTNET_RX_PAD (NET_IP_ALIGN + NET_SKB_PAD)
39 
40 /* Amount of XDP headroom to prepend to packets for use by xdp_adjust_head */
41 #define VIRTIO_XDP_HEADROOM 256
42 
43 /* Separating two types of XDP xmit */
44 #define VIRTIO_XDP_TX		BIT(0)
45 #define VIRTIO_XDP_REDIR	BIT(1)
46 
47 #define VIRTIO_XDP_FLAG	BIT(0)
48 
49 /* RX packet size EWMA. The average packet size is used to determine the packet
50  * buffer size when refilling RX rings. As the entire RX ring may be refilled
51  * at once, the weight is chosen so that the EWMA will be insensitive to short-
52  * term, transient changes in packet size.
53  */
54 DECLARE_EWMA(pkt_len, 0, 64)
55 
56 #define VIRTNET_DRIVER_VERSION "1.0.0"
57 
58 static const unsigned long guest_offloads[] = {
59 	VIRTIO_NET_F_GUEST_TSO4,
60 	VIRTIO_NET_F_GUEST_TSO6,
61 	VIRTIO_NET_F_GUEST_ECN,
62 	VIRTIO_NET_F_GUEST_UFO,
63 	VIRTIO_NET_F_GUEST_CSUM,
64 	VIRTIO_NET_F_GUEST_USO4,
65 	VIRTIO_NET_F_GUEST_USO6,
66 	VIRTIO_NET_F_GUEST_HDRLEN
67 };
68 
69 #define GUEST_OFFLOAD_GRO_HW_MASK ((1ULL << VIRTIO_NET_F_GUEST_TSO4) | \
70 				(1ULL << VIRTIO_NET_F_GUEST_TSO6) | \
71 				(1ULL << VIRTIO_NET_F_GUEST_ECN)  | \
72 				(1ULL << VIRTIO_NET_F_GUEST_UFO)  | \
73 				(1ULL << VIRTIO_NET_F_GUEST_USO4) | \
74 				(1ULL << VIRTIO_NET_F_GUEST_USO6))
75 
76 struct virtnet_stat_desc {
77 	char desc[ETH_GSTRING_LEN];
78 	size_t offset;
79 };
80 
81 struct virtnet_sq_stats {
82 	struct u64_stats_sync syncp;
83 	u64 packets;
84 	u64 bytes;
85 	u64 xdp_tx;
86 	u64 xdp_tx_drops;
87 	u64 kicks;
88 	u64 tx_timeouts;
89 };
90 
91 struct virtnet_rq_stats {
92 	struct u64_stats_sync syncp;
93 	u64 packets;
94 	u64 bytes;
95 	u64 drops;
96 	u64 xdp_packets;
97 	u64 xdp_tx;
98 	u64 xdp_redirects;
99 	u64 xdp_drops;
100 	u64 kicks;
101 };
102 
103 #define VIRTNET_SQ_STAT(m)	offsetof(struct virtnet_sq_stats, m)
104 #define VIRTNET_RQ_STAT(m)	offsetof(struct virtnet_rq_stats, m)
105 
106 static const struct virtnet_stat_desc virtnet_sq_stats_desc[] = {
107 	{ "packets",		VIRTNET_SQ_STAT(packets) },
108 	{ "bytes",		VIRTNET_SQ_STAT(bytes) },
109 	{ "xdp_tx",		VIRTNET_SQ_STAT(xdp_tx) },
110 	{ "xdp_tx_drops",	VIRTNET_SQ_STAT(xdp_tx_drops) },
111 	{ "kicks",		VIRTNET_SQ_STAT(kicks) },
112 	{ "tx_timeouts",	VIRTNET_SQ_STAT(tx_timeouts) },
113 };
114 
115 static const struct virtnet_stat_desc virtnet_rq_stats_desc[] = {
116 	{ "packets",		VIRTNET_RQ_STAT(packets) },
117 	{ "bytes",		VIRTNET_RQ_STAT(bytes) },
118 	{ "drops",		VIRTNET_RQ_STAT(drops) },
119 	{ "xdp_packets",	VIRTNET_RQ_STAT(xdp_packets) },
120 	{ "xdp_tx",		VIRTNET_RQ_STAT(xdp_tx) },
121 	{ "xdp_redirects",	VIRTNET_RQ_STAT(xdp_redirects) },
122 	{ "xdp_drops",		VIRTNET_RQ_STAT(xdp_drops) },
123 	{ "kicks",		VIRTNET_RQ_STAT(kicks) },
124 };
125 
126 #define VIRTNET_SQ_STATS_LEN	ARRAY_SIZE(virtnet_sq_stats_desc)
127 #define VIRTNET_RQ_STATS_LEN	ARRAY_SIZE(virtnet_rq_stats_desc)
128 
129 /* Internal representation of a send virtqueue */
130 struct send_queue {
131 	/* Virtqueue associated with this send _queue */
132 	struct virtqueue *vq;
133 
134 	/* TX: fragments + linear part + virtio header */
135 	struct scatterlist sg[MAX_SKB_FRAGS + 2];
136 
137 	/* Name of the send queue: output.$index */
138 	char name[16];
139 
140 	struct virtnet_sq_stats stats;
141 
142 	struct napi_struct napi;
143 
144 	/* Record whether sq is in reset state. */
145 	bool reset;
146 };
147 
148 /* Internal representation of a receive virtqueue */
149 struct receive_queue {
150 	/* Virtqueue associated with this receive_queue */
151 	struct virtqueue *vq;
152 
153 	struct napi_struct napi;
154 
155 	struct bpf_prog __rcu *xdp_prog;
156 
157 	struct virtnet_rq_stats stats;
158 
159 	/* Chain pages by the private ptr. */
160 	struct page *pages;
161 
162 	/* Average packet length for mergeable receive buffers. */
163 	struct ewma_pkt_len mrg_avg_pkt_len;
164 
165 	/* Page frag for packet buffer allocation. */
166 	struct page_frag alloc_frag;
167 
168 	/* RX: fragments + linear part + virtio header */
169 	struct scatterlist sg[MAX_SKB_FRAGS + 2];
170 
171 	/* Min single buffer size for mergeable buffers case. */
172 	unsigned int min_buf_len;
173 
174 	/* Name of this receive queue: input.$index */
175 	char name[16];
176 
177 	struct xdp_rxq_info xdp_rxq;
178 };
179 
180 /* This structure can contain rss message with maximum settings for indirection table and keysize
181  * Note, that default structure that describes RSS configuration virtio_net_rss_config
182  * contains same info but can't handle table values.
183  * In any case, structure would be passed to virtio hw through sg_buf split by parts
184  * because table sizes may be differ according to the device configuration.
185  */
186 #define VIRTIO_NET_RSS_MAX_KEY_SIZE     40
187 #define VIRTIO_NET_RSS_MAX_TABLE_LEN    128
188 struct virtio_net_ctrl_rss {
189 	u32 hash_types;
190 	u16 indirection_table_mask;
191 	u16 unclassified_queue;
192 	u16 indirection_table[VIRTIO_NET_RSS_MAX_TABLE_LEN];
193 	u16 max_tx_vq;
194 	u8 hash_key_length;
195 	u8 key[VIRTIO_NET_RSS_MAX_KEY_SIZE];
196 };
197 
198 /* Control VQ buffers: protected by the rtnl lock */
199 struct control_buf {
200 	struct virtio_net_ctrl_hdr hdr;
201 	virtio_net_ctrl_ack status;
202 	struct virtio_net_ctrl_mq mq;
203 	u8 promisc;
204 	u8 allmulti;
205 	__virtio16 vid;
206 	__virtio64 offloads;
207 	struct virtio_net_ctrl_rss rss;
208 };
209 
210 struct virtnet_info {
211 	struct virtio_device *vdev;
212 	struct virtqueue *cvq;
213 	struct net_device *dev;
214 	struct send_queue *sq;
215 	struct receive_queue *rq;
216 	unsigned int status;
217 
218 	/* Max # of queue pairs supported by the device */
219 	u16 max_queue_pairs;
220 
221 	/* # of queue pairs currently used by the driver */
222 	u16 curr_queue_pairs;
223 
224 	/* # of XDP queue pairs currently used by the driver */
225 	u16 xdp_queue_pairs;
226 
227 	/* xdp_queue_pairs may be 0, when xdp is already loaded. So add this. */
228 	bool xdp_enabled;
229 
230 	/* I like... big packets and I cannot lie! */
231 	bool big_packets;
232 
233 	/* number of sg entries allocated for big packets */
234 	unsigned int big_packets_num_skbfrags;
235 
236 	/* Host will merge rx buffers for big packets (shake it! shake it!) */
237 	bool mergeable_rx_bufs;
238 
239 	/* Host supports rss and/or hash report */
240 	bool has_rss;
241 	bool has_rss_hash_report;
242 	u8 rss_key_size;
243 	u16 rss_indir_table_size;
244 	u32 rss_hash_types_supported;
245 	u32 rss_hash_types_saved;
246 
247 	/* Has control virtqueue */
248 	bool has_cvq;
249 
250 	/* Host can handle any s/g split between our header and packet data */
251 	bool any_header_sg;
252 
253 	/* Packet virtio header size */
254 	u8 hdr_len;
255 
256 	/* Work struct for delayed refilling if we run low on memory. */
257 	struct delayed_work refill;
258 
259 	/* Is delayed refill enabled? */
260 	bool refill_enabled;
261 
262 	/* The lock to synchronize the access to refill_enabled */
263 	spinlock_t refill_lock;
264 
265 	/* Work struct for config space updates */
266 	struct work_struct config_work;
267 
268 	/* Does the affinity hint is set for virtqueues? */
269 	bool affinity_hint_set;
270 
271 	/* CPU hotplug instances for online & dead */
272 	struct hlist_node node;
273 	struct hlist_node node_dead;
274 
275 	struct control_buf *ctrl;
276 
277 	/* Ethtool settings */
278 	u8 duplex;
279 	u32 speed;
280 
281 	/* Interrupt coalescing settings */
282 	u32 tx_usecs;
283 	u32 rx_usecs;
284 	u32 tx_max_packets;
285 	u32 rx_max_packets;
286 
287 	unsigned long guest_offloads;
288 	unsigned long guest_offloads_capable;
289 
290 	/* failover when STANDBY feature enabled */
291 	struct failover *failover;
292 };
293 
294 struct padded_vnet_hdr {
295 	struct virtio_net_hdr_v1_hash hdr;
296 	/*
297 	 * hdr is in a separate sg buffer, and data sg buffer shares same page
298 	 * with this header sg. This padding makes next sg 16 byte aligned
299 	 * after the header.
300 	 */
301 	char padding[12];
302 };
303 
304 static void virtnet_rq_free_unused_buf(struct virtqueue *vq, void *buf);
305 static void virtnet_sq_free_unused_buf(struct virtqueue *vq, void *buf);
306 
307 static bool is_xdp_frame(void *ptr)
308 {
309 	return (unsigned long)ptr & VIRTIO_XDP_FLAG;
310 }
311 
312 static void *xdp_to_ptr(struct xdp_frame *ptr)
313 {
314 	return (void *)((unsigned long)ptr | VIRTIO_XDP_FLAG);
315 }
316 
317 static struct xdp_frame *ptr_to_xdp(void *ptr)
318 {
319 	return (struct xdp_frame *)((unsigned long)ptr & ~VIRTIO_XDP_FLAG);
320 }
321 
322 /* Converting between virtqueue no. and kernel tx/rx queue no.
323  * 0:rx0 1:tx0 2:rx1 3:tx1 ... 2N:rxN 2N+1:txN 2N+2:cvq
324  */
325 static int vq2txq(struct virtqueue *vq)
326 {
327 	return (vq->index - 1) / 2;
328 }
329 
330 static int txq2vq(int txq)
331 {
332 	return txq * 2 + 1;
333 }
334 
335 static int vq2rxq(struct virtqueue *vq)
336 {
337 	return vq->index / 2;
338 }
339 
340 static int rxq2vq(int rxq)
341 {
342 	return rxq * 2;
343 }
344 
345 static inline struct virtio_net_hdr_mrg_rxbuf *skb_vnet_hdr(struct sk_buff *skb)
346 {
347 	return (struct virtio_net_hdr_mrg_rxbuf *)skb->cb;
348 }
349 
350 /*
351  * private is used to chain pages for big packets, put the whole
352  * most recent used list in the beginning for reuse
353  */
354 static void give_pages(struct receive_queue *rq, struct page *page)
355 {
356 	struct page *end;
357 
358 	/* Find end of list, sew whole thing into vi->rq.pages. */
359 	for (end = page; end->private; end = (struct page *)end->private);
360 	end->private = (unsigned long)rq->pages;
361 	rq->pages = page;
362 }
363 
364 static struct page *get_a_page(struct receive_queue *rq, gfp_t gfp_mask)
365 {
366 	struct page *p = rq->pages;
367 
368 	if (p) {
369 		rq->pages = (struct page *)p->private;
370 		/* clear private here, it is used to chain pages */
371 		p->private = 0;
372 	} else
373 		p = alloc_page(gfp_mask);
374 	return p;
375 }
376 
377 static void enable_delayed_refill(struct virtnet_info *vi)
378 {
379 	spin_lock_bh(&vi->refill_lock);
380 	vi->refill_enabled = true;
381 	spin_unlock_bh(&vi->refill_lock);
382 }
383 
384 static void disable_delayed_refill(struct virtnet_info *vi)
385 {
386 	spin_lock_bh(&vi->refill_lock);
387 	vi->refill_enabled = false;
388 	spin_unlock_bh(&vi->refill_lock);
389 }
390 
391 static void virtqueue_napi_schedule(struct napi_struct *napi,
392 				    struct virtqueue *vq)
393 {
394 	if (napi_schedule_prep(napi)) {
395 		virtqueue_disable_cb(vq);
396 		__napi_schedule(napi);
397 	}
398 }
399 
400 static void virtqueue_napi_complete(struct napi_struct *napi,
401 				    struct virtqueue *vq, int processed)
402 {
403 	int opaque;
404 
405 	opaque = virtqueue_enable_cb_prepare(vq);
406 	if (napi_complete_done(napi, processed)) {
407 		if (unlikely(virtqueue_poll(vq, opaque)))
408 			virtqueue_napi_schedule(napi, vq);
409 	} else {
410 		virtqueue_disable_cb(vq);
411 	}
412 }
413 
414 static void skb_xmit_done(struct virtqueue *vq)
415 {
416 	struct virtnet_info *vi = vq->vdev->priv;
417 	struct napi_struct *napi = &vi->sq[vq2txq(vq)].napi;
418 
419 	/* Suppress further interrupts. */
420 	virtqueue_disable_cb(vq);
421 
422 	if (napi->weight)
423 		virtqueue_napi_schedule(napi, vq);
424 	else
425 		/* We were probably waiting for more output buffers. */
426 		netif_wake_subqueue(vi->dev, vq2txq(vq));
427 }
428 
429 #define MRG_CTX_HEADER_SHIFT 22
430 static void *mergeable_len_to_ctx(unsigned int truesize,
431 				  unsigned int headroom)
432 {
433 	return (void *)(unsigned long)((headroom << MRG_CTX_HEADER_SHIFT) | truesize);
434 }
435 
436 static unsigned int mergeable_ctx_to_headroom(void *mrg_ctx)
437 {
438 	return (unsigned long)mrg_ctx >> MRG_CTX_HEADER_SHIFT;
439 }
440 
441 static unsigned int mergeable_ctx_to_truesize(void *mrg_ctx)
442 {
443 	return (unsigned long)mrg_ctx & ((1 << MRG_CTX_HEADER_SHIFT) - 1);
444 }
445 
446 /* Called from bottom half context */
447 static struct sk_buff *page_to_skb(struct virtnet_info *vi,
448 				   struct receive_queue *rq,
449 				   struct page *page, unsigned int offset,
450 				   unsigned int len, unsigned int truesize,
451 				   unsigned int headroom)
452 {
453 	struct sk_buff *skb;
454 	struct virtio_net_hdr_mrg_rxbuf *hdr;
455 	unsigned int copy, hdr_len, hdr_padded_len;
456 	struct page *page_to_free = NULL;
457 	int tailroom, shinfo_size;
458 	char *p, *hdr_p, *buf;
459 
460 	p = page_address(page) + offset;
461 	hdr_p = p;
462 
463 	hdr_len = vi->hdr_len;
464 	if (vi->mergeable_rx_bufs)
465 		hdr_padded_len = hdr_len;
466 	else
467 		hdr_padded_len = sizeof(struct padded_vnet_hdr);
468 
469 	buf = p - headroom;
470 	len -= hdr_len;
471 	offset += hdr_padded_len;
472 	p += hdr_padded_len;
473 	tailroom = truesize - headroom  - hdr_padded_len - len;
474 
475 	shinfo_size = SKB_DATA_ALIGN(sizeof(struct skb_shared_info));
476 
477 	/* copy small packet so we can reuse these pages */
478 	if (!NET_IP_ALIGN && len > GOOD_COPY_LEN && tailroom >= shinfo_size) {
479 		skb = build_skb(buf, truesize);
480 		if (unlikely(!skb))
481 			return NULL;
482 
483 		skb_reserve(skb, p - buf);
484 		skb_put(skb, len);
485 
486 		page = (struct page *)page->private;
487 		if (page)
488 			give_pages(rq, page);
489 		goto ok;
490 	}
491 
492 	/* copy small packet so we can reuse these pages for small data */
493 	skb = napi_alloc_skb(&rq->napi, GOOD_COPY_LEN);
494 	if (unlikely(!skb))
495 		return NULL;
496 
497 	/* Copy all frame if it fits skb->head, otherwise
498 	 * we let virtio_net_hdr_to_skb() and GRO pull headers as needed.
499 	 */
500 	if (len <= skb_tailroom(skb))
501 		copy = len;
502 	else
503 		copy = ETH_HLEN;
504 	skb_put_data(skb, p, copy);
505 
506 	len -= copy;
507 	offset += copy;
508 
509 	if (vi->mergeable_rx_bufs) {
510 		if (len)
511 			skb_add_rx_frag(skb, 0, page, offset, len, truesize);
512 		else
513 			page_to_free = page;
514 		goto ok;
515 	}
516 
517 	/*
518 	 * Verify that we can indeed put this data into a skb.
519 	 * This is here to handle cases when the device erroneously
520 	 * tries to receive more than is possible. This is usually
521 	 * the case of a broken device.
522 	 */
523 	if (unlikely(len > MAX_SKB_FRAGS * PAGE_SIZE)) {
524 		net_dbg_ratelimited("%s: too much data\n", skb->dev->name);
525 		dev_kfree_skb(skb);
526 		return NULL;
527 	}
528 	BUG_ON(offset >= PAGE_SIZE);
529 	while (len) {
530 		unsigned int frag_size = min((unsigned)PAGE_SIZE - offset, len);
531 		skb_add_rx_frag(skb, skb_shinfo(skb)->nr_frags, page, offset,
532 				frag_size, truesize);
533 		len -= frag_size;
534 		page = (struct page *)page->private;
535 		offset = 0;
536 	}
537 
538 	if (page)
539 		give_pages(rq, page);
540 
541 ok:
542 	hdr = skb_vnet_hdr(skb);
543 	memcpy(hdr, hdr_p, hdr_len);
544 	if (page_to_free)
545 		put_page(page_to_free);
546 
547 	return skb;
548 }
549 
550 static void free_old_xmit_skbs(struct send_queue *sq, bool in_napi)
551 {
552 	unsigned int len;
553 	unsigned int packets = 0;
554 	unsigned int bytes = 0;
555 	void *ptr;
556 
557 	while ((ptr = virtqueue_get_buf(sq->vq, &len)) != NULL) {
558 		if (likely(!is_xdp_frame(ptr))) {
559 			struct sk_buff *skb = ptr;
560 
561 			pr_debug("Sent skb %p\n", skb);
562 
563 			bytes += skb->len;
564 			napi_consume_skb(skb, in_napi);
565 		} else {
566 			struct xdp_frame *frame = ptr_to_xdp(ptr);
567 
568 			bytes += xdp_get_frame_len(frame);
569 			xdp_return_frame(frame);
570 		}
571 		packets++;
572 	}
573 
574 	/* Avoid overhead when no packets have been processed
575 	 * happens when called speculatively from start_xmit.
576 	 */
577 	if (!packets)
578 		return;
579 
580 	u64_stats_update_begin(&sq->stats.syncp);
581 	sq->stats.bytes += bytes;
582 	sq->stats.packets += packets;
583 	u64_stats_update_end(&sq->stats.syncp);
584 }
585 
586 static bool is_xdp_raw_buffer_queue(struct virtnet_info *vi, int q)
587 {
588 	if (q < (vi->curr_queue_pairs - vi->xdp_queue_pairs))
589 		return false;
590 	else if (q < vi->curr_queue_pairs)
591 		return true;
592 	else
593 		return false;
594 }
595 
596 static void check_sq_full_and_disable(struct virtnet_info *vi,
597 				      struct net_device *dev,
598 				      struct send_queue *sq)
599 {
600 	bool use_napi = sq->napi.weight;
601 	int qnum;
602 
603 	qnum = sq - vi->sq;
604 
605 	/* If running out of space, stop queue to avoid getting packets that we
606 	 * are then unable to transmit.
607 	 * An alternative would be to force queuing layer to requeue the skb by
608 	 * returning NETDEV_TX_BUSY. However, NETDEV_TX_BUSY should not be
609 	 * returned in a normal path of operation: it means that driver is not
610 	 * maintaining the TX queue stop/start state properly, and causes
611 	 * the stack to do a non-trivial amount of useless work.
612 	 * Since most packets only take 1 or 2 ring slots, stopping the queue
613 	 * early means 16 slots are typically wasted.
614 	 */
615 	if (sq->vq->num_free < 2+MAX_SKB_FRAGS) {
616 		netif_stop_subqueue(dev, qnum);
617 		if (use_napi) {
618 			if (unlikely(!virtqueue_enable_cb_delayed(sq->vq)))
619 				virtqueue_napi_schedule(&sq->napi, sq->vq);
620 		} else if (unlikely(!virtqueue_enable_cb_delayed(sq->vq))) {
621 			/* More just got used, free them then recheck. */
622 			free_old_xmit_skbs(sq, false);
623 			if (sq->vq->num_free >= 2+MAX_SKB_FRAGS) {
624 				netif_start_subqueue(dev, qnum);
625 				virtqueue_disable_cb(sq->vq);
626 			}
627 		}
628 	}
629 }
630 
631 static int __virtnet_xdp_xmit_one(struct virtnet_info *vi,
632 				   struct send_queue *sq,
633 				   struct xdp_frame *xdpf)
634 {
635 	struct virtio_net_hdr_mrg_rxbuf *hdr;
636 	struct skb_shared_info *shinfo;
637 	u8 nr_frags = 0;
638 	int err, i;
639 
640 	if (unlikely(xdpf->headroom < vi->hdr_len))
641 		return -EOVERFLOW;
642 
643 	if (unlikely(xdp_frame_has_frags(xdpf))) {
644 		shinfo = xdp_get_shared_info_from_frame(xdpf);
645 		nr_frags = shinfo->nr_frags;
646 	}
647 
648 	/* In wrapping function virtnet_xdp_xmit(), we need to free
649 	 * up the pending old buffers, where we need to calculate the
650 	 * position of skb_shared_info in xdp_get_frame_len() and
651 	 * xdp_return_frame(), which will involve to xdpf->data and
652 	 * xdpf->headroom. Therefore, we need to update the value of
653 	 * headroom synchronously here.
654 	 */
655 	xdpf->headroom -= vi->hdr_len;
656 	xdpf->data -= vi->hdr_len;
657 	/* Zero header and leave csum up to XDP layers */
658 	hdr = xdpf->data;
659 	memset(hdr, 0, vi->hdr_len);
660 	xdpf->len   += vi->hdr_len;
661 
662 	sg_init_table(sq->sg, nr_frags + 1);
663 	sg_set_buf(sq->sg, xdpf->data, xdpf->len);
664 	for (i = 0; i < nr_frags; i++) {
665 		skb_frag_t *frag = &shinfo->frags[i];
666 
667 		sg_set_page(&sq->sg[i + 1], skb_frag_page(frag),
668 			    skb_frag_size(frag), skb_frag_off(frag));
669 	}
670 
671 	err = virtqueue_add_outbuf(sq->vq, sq->sg, nr_frags + 1,
672 				   xdp_to_ptr(xdpf), GFP_ATOMIC);
673 	if (unlikely(err))
674 		return -ENOSPC; /* Caller handle free/refcnt */
675 
676 	return 0;
677 }
678 
679 /* when vi->curr_queue_pairs > nr_cpu_ids, the txq/sq is only used for xdp tx on
680  * the current cpu, so it does not need to be locked.
681  *
682  * Here we use marco instead of inline functions because we have to deal with
683  * three issues at the same time: 1. the choice of sq. 2. judge and execute the
684  * lock/unlock of txq 3. make sparse happy. It is difficult for two inline
685  * functions to perfectly solve these three problems at the same time.
686  */
687 #define virtnet_xdp_get_sq(vi) ({                                       \
688 	int cpu = smp_processor_id();                                   \
689 	struct netdev_queue *txq;                                       \
690 	typeof(vi) v = (vi);                                            \
691 	unsigned int qp;                                                \
692 									\
693 	if (v->curr_queue_pairs > nr_cpu_ids) {                         \
694 		qp = v->curr_queue_pairs - v->xdp_queue_pairs;          \
695 		qp += cpu;                                              \
696 		txq = netdev_get_tx_queue(v->dev, qp);                  \
697 		__netif_tx_acquire(txq);                                \
698 	} else {                                                        \
699 		qp = cpu % v->curr_queue_pairs;                         \
700 		txq = netdev_get_tx_queue(v->dev, qp);                  \
701 		__netif_tx_lock(txq, cpu);                              \
702 	}                                                               \
703 	v->sq + qp;                                                     \
704 })
705 
706 #define virtnet_xdp_put_sq(vi, q) {                                     \
707 	struct netdev_queue *txq;                                       \
708 	typeof(vi) v = (vi);                                            \
709 									\
710 	txq = netdev_get_tx_queue(v->dev, (q) - v->sq);                 \
711 	if (v->curr_queue_pairs > nr_cpu_ids)                           \
712 		__netif_tx_release(txq);                                \
713 	else                                                            \
714 		__netif_tx_unlock(txq);                                 \
715 }
716 
717 static int virtnet_xdp_xmit(struct net_device *dev,
718 			    int n, struct xdp_frame **frames, u32 flags)
719 {
720 	struct virtnet_info *vi = netdev_priv(dev);
721 	struct receive_queue *rq = vi->rq;
722 	struct bpf_prog *xdp_prog;
723 	struct send_queue *sq;
724 	unsigned int len;
725 	int packets = 0;
726 	int bytes = 0;
727 	int nxmit = 0;
728 	int kicks = 0;
729 	void *ptr;
730 	int ret;
731 	int i;
732 
733 	/* Only allow ndo_xdp_xmit if XDP is loaded on dev, as this
734 	 * indicate XDP resources have been successfully allocated.
735 	 */
736 	xdp_prog = rcu_access_pointer(rq->xdp_prog);
737 	if (!xdp_prog)
738 		return -ENXIO;
739 
740 	sq = virtnet_xdp_get_sq(vi);
741 
742 	if (unlikely(flags & ~XDP_XMIT_FLAGS_MASK)) {
743 		ret = -EINVAL;
744 		goto out;
745 	}
746 
747 	/* Free up any pending old buffers before queueing new ones. */
748 	while ((ptr = virtqueue_get_buf(sq->vq, &len)) != NULL) {
749 		if (likely(is_xdp_frame(ptr))) {
750 			struct xdp_frame *frame = ptr_to_xdp(ptr);
751 
752 			bytes += xdp_get_frame_len(frame);
753 			xdp_return_frame(frame);
754 		} else {
755 			struct sk_buff *skb = ptr;
756 
757 			bytes += skb->len;
758 			napi_consume_skb(skb, false);
759 		}
760 		packets++;
761 	}
762 
763 	for (i = 0; i < n; i++) {
764 		struct xdp_frame *xdpf = frames[i];
765 
766 		if (__virtnet_xdp_xmit_one(vi, sq, xdpf))
767 			break;
768 		nxmit++;
769 	}
770 	ret = nxmit;
771 
772 	if (!is_xdp_raw_buffer_queue(vi, sq - vi->sq))
773 		check_sq_full_and_disable(vi, dev, sq);
774 
775 	if (flags & XDP_XMIT_FLUSH) {
776 		if (virtqueue_kick_prepare(sq->vq) && virtqueue_notify(sq->vq))
777 			kicks = 1;
778 	}
779 out:
780 	u64_stats_update_begin(&sq->stats.syncp);
781 	sq->stats.bytes += bytes;
782 	sq->stats.packets += packets;
783 	sq->stats.xdp_tx += n;
784 	sq->stats.xdp_tx_drops += n - nxmit;
785 	sq->stats.kicks += kicks;
786 	u64_stats_update_end(&sq->stats.syncp);
787 
788 	virtnet_xdp_put_sq(vi, sq);
789 	return ret;
790 }
791 
792 static unsigned int virtnet_get_headroom(struct virtnet_info *vi)
793 {
794 	return vi->xdp_enabled ? VIRTIO_XDP_HEADROOM : 0;
795 }
796 
797 /* We copy the packet for XDP in the following cases:
798  *
799  * 1) Packet is scattered across multiple rx buffers.
800  * 2) Headroom space is insufficient.
801  *
802  * This is inefficient but it's a temporary condition that
803  * we hit right after XDP is enabled and until queue is refilled
804  * with large buffers with sufficient headroom - so it should affect
805  * at most queue size packets.
806  * Afterwards, the conditions to enable
807  * XDP should preclude the underlying device from sending packets
808  * across multiple buffers (num_buf > 1), and we make sure buffers
809  * have enough headroom.
810  */
811 static struct page *xdp_linearize_page(struct receive_queue *rq,
812 				       int *num_buf,
813 				       struct page *p,
814 				       int offset,
815 				       int page_off,
816 				       unsigned int *len)
817 {
818 	int tailroom = SKB_DATA_ALIGN(sizeof(struct skb_shared_info));
819 	struct page *page;
820 
821 	if (page_off + *len + tailroom > PAGE_SIZE)
822 		return NULL;
823 
824 	page = alloc_page(GFP_ATOMIC);
825 	if (!page)
826 		return NULL;
827 
828 	memcpy(page_address(page) + page_off, page_address(p) + offset, *len);
829 	page_off += *len;
830 
831 	while (--*num_buf) {
832 		unsigned int buflen;
833 		void *buf;
834 		int off;
835 
836 		buf = virtqueue_get_buf(rq->vq, &buflen);
837 		if (unlikely(!buf))
838 			goto err_buf;
839 
840 		p = virt_to_head_page(buf);
841 		off = buf - page_address(p);
842 
843 		/* guard against a misconfigured or uncooperative backend that
844 		 * is sending packet larger than the MTU.
845 		 */
846 		if ((page_off + buflen + tailroom) > PAGE_SIZE) {
847 			put_page(p);
848 			goto err_buf;
849 		}
850 
851 		memcpy(page_address(page) + page_off,
852 		       page_address(p) + off, buflen);
853 		page_off += buflen;
854 		put_page(p);
855 	}
856 
857 	/* Headroom does not contribute to packet length */
858 	*len = page_off - VIRTIO_XDP_HEADROOM;
859 	return page;
860 err_buf:
861 	__free_pages(page, 0);
862 	return NULL;
863 }
864 
865 static struct sk_buff *receive_small(struct net_device *dev,
866 				     struct virtnet_info *vi,
867 				     struct receive_queue *rq,
868 				     void *buf, void *ctx,
869 				     unsigned int len,
870 				     unsigned int *xdp_xmit,
871 				     struct virtnet_rq_stats *stats)
872 {
873 	struct sk_buff *skb;
874 	struct bpf_prog *xdp_prog;
875 	unsigned int xdp_headroom = (unsigned long)ctx;
876 	unsigned int header_offset = VIRTNET_RX_PAD + xdp_headroom;
877 	unsigned int headroom = vi->hdr_len + header_offset;
878 	unsigned int buflen = SKB_DATA_ALIGN(GOOD_PACKET_LEN + headroom) +
879 			      SKB_DATA_ALIGN(sizeof(struct skb_shared_info));
880 	struct page *page = virt_to_head_page(buf);
881 	unsigned int delta = 0;
882 	struct page *xdp_page;
883 	int err;
884 	unsigned int metasize = 0;
885 
886 	len -= vi->hdr_len;
887 	stats->bytes += len;
888 
889 	if (unlikely(len > GOOD_PACKET_LEN)) {
890 		pr_debug("%s: rx error: len %u exceeds max size %d\n",
891 			 dev->name, len, GOOD_PACKET_LEN);
892 		dev->stats.rx_length_errors++;
893 		goto err;
894 	}
895 
896 	if (likely(!vi->xdp_enabled)) {
897 		xdp_prog = NULL;
898 		goto skip_xdp;
899 	}
900 
901 	rcu_read_lock();
902 	xdp_prog = rcu_dereference(rq->xdp_prog);
903 	if (xdp_prog) {
904 		struct virtio_net_hdr_mrg_rxbuf *hdr = buf + header_offset;
905 		struct xdp_frame *xdpf;
906 		struct xdp_buff xdp;
907 		void *orig_data;
908 		u32 act;
909 
910 		if (unlikely(hdr->hdr.gso_type))
911 			goto err_xdp;
912 
913 		if (unlikely(xdp_headroom < virtnet_get_headroom(vi))) {
914 			int offset = buf - page_address(page) + header_offset;
915 			unsigned int tlen = len + vi->hdr_len;
916 			int num_buf = 1;
917 
918 			xdp_headroom = virtnet_get_headroom(vi);
919 			header_offset = VIRTNET_RX_PAD + xdp_headroom;
920 			headroom = vi->hdr_len + header_offset;
921 			buflen = SKB_DATA_ALIGN(GOOD_PACKET_LEN + headroom) +
922 				 SKB_DATA_ALIGN(sizeof(struct skb_shared_info));
923 			xdp_page = xdp_linearize_page(rq, &num_buf, page,
924 						      offset, header_offset,
925 						      &tlen);
926 			if (!xdp_page)
927 				goto err_xdp;
928 
929 			buf = page_address(xdp_page);
930 			put_page(page);
931 			page = xdp_page;
932 		}
933 
934 		xdp_init_buff(&xdp, buflen, &rq->xdp_rxq);
935 		xdp_prepare_buff(&xdp, buf + VIRTNET_RX_PAD + vi->hdr_len,
936 				 xdp_headroom, len, true);
937 		orig_data = xdp.data;
938 		act = bpf_prog_run_xdp(xdp_prog, &xdp);
939 		stats->xdp_packets++;
940 
941 		switch (act) {
942 		case XDP_PASS:
943 			/* Recalculate length in case bpf program changed it */
944 			delta = orig_data - xdp.data;
945 			len = xdp.data_end - xdp.data;
946 			metasize = xdp.data - xdp.data_meta;
947 			break;
948 		case XDP_TX:
949 			stats->xdp_tx++;
950 			xdpf = xdp_convert_buff_to_frame(&xdp);
951 			if (unlikely(!xdpf))
952 				goto err_xdp;
953 			err = virtnet_xdp_xmit(dev, 1, &xdpf, 0);
954 			if (unlikely(!err)) {
955 				xdp_return_frame_rx_napi(xdpf);
956 			} else if (unlikely(err < 0)) {
957 				trace_xdp_exception(vi->dev, xdp_prog, act);
958 				goto err_xdp;
959 			}
960 			*xdp_xmit |= VIRTIO_XDP_TX;
961 			rcu_read_unlock();
962 			goto xdp_xmit;
963 		case XDP_REDIRECT:
964 			stats->xdp_redirects++;
965 			err = xdp_do_redirect(dev, &xdp, xdp_prog);
966 			if (err)
967 				goto err_xdp;
968 			*xdp_xmit |= VIRTIO_XDP_REDIR;
969 			rcu_read_unlock();
970 			goto xdp_xmit;
971 		default:
972 			bpf_warn_invalid_xdp_action(vi->dev, xdp_prog, act);
973 			fallthrough;
974 		case XDP_ABORTED:
975 			trace_xdp_exception(vi->dev, xdp_prog, act);
976 			goto err_xdp;
977 		case XDP_DROP:
978 			goto err_xdp;
979 		}
980 	}
981 	rcu_read_unlock();
982 
983 skip_xdp:
984 	skb = build_skb(buf, buflen);
985 	if (!skb)
986 		goto err;
987 	skb_reserve(skb, headroom - delta);
988 	skb_put(skb, len);
989 	if (!xdp_prog) {
990 		buf += header_offset;
991 		memcpy(skb_vnet_hdr(skb), buf, vi->hdr_len);
992 	} /* keep zeroed vnet hdr since XDP is loaded */
993 
994 	if (metasize)
995 		skb_metadata_set(skb, metasize);
996 
997 	return skb;
998 
999 err_xdp:
1000 	rcu_read_unlock();
1001 	stats->xdp_drops++;
1002 err:
1003 	stats->drops++;
1004 	put_page(page);
1005 xdp_xmit:
1006 	return NULL;
1007 }
1008 
1009 static struct sk_buff *receive_big(struct net_device *dev,
1010 				   struct virtnet_info *vi,
1011 				   struct receive_queue *rq,
1012 				   void *buf,
1013 				   unsigned int len,
1014 				   struct virtnet_rq_stats *stats)
1015 {
1016 	struct page *page = buf;
1017 	struct sk_buff *skb =
1018 		page_to_skb(vi, rq, page, 0, len, PAGE_SIZE, 0);
1019 
1020 	stats->bytes += len - vi->hdr_len;
1021 	if (unlikely(!skb))
1022 		goto err;
1023 
1024 	return skb;
1025 
1026 err:
1027 	stats->drops++;
1028 	give_pages(rq, page);
1029 	return NULL;
1030 }
1031 
1032 /* Why not use xdp_build_skb_from_frame() ?
1033  * XDP core assumes that xdp frags are PAGE_SIZE in length, while in
1034  * virtio-net there are 2 points that do not match its requirements:
1035  *  1. The size of the prefilled buffer is not fixed before xdp is set.
1036  *  2. xdp_build_skb_from_frame() does more checks that we don't need,
1037  *     like eth_type_trans() (which virtio-net does in receive_buf()).
1038  */
1039 static struct sk_buff *build_skb_from_xdp_buff(struct net_device *dev,
1040 					       struct virtnet_info *vi,
1041 					       struct xdp_buff *xdp,
1042 					       unsigned int xdp_frags_truesz)
1043 {
1044 	struct skb_shared_info *sinfo = xdp_get_shared_info_from_buff(xdp);
1045 	unsigned int headroom, data_len;
1046 	struct sk_buff *skb;
1047 	int metasize;
1048 	u8 nr_frags;
1049 
1050 	if (unlikely(xdp->data_end > xdp_data_hard_end(xdp))) {
1051 		pr_debug("Error building skb as missing reserved tailroom for xdp");
1052 		return NULL;
1053 	}
1054 
1055 	if (unlikely(xdp_buff_has_frags(xdp)))
1056 		nr_frags = sinfo->nr_frags;
1057 
1058 	skb = build_skb(xdp->data_hard_start, xdp->frame_sz);
1059 	if (unlikely(!skb))
1060 		return NULL;
1061 
1062 	headroom = xdp->data - xdp->data_hard_start;
1063 	data_len = xdp->data_end - xdp->data;
1064 	skb_reserve(skb, headroom);
1065 	__skb_put(skb, data_len);
1066 
1067 	metasize = xdp->data - xdp->data_meta;
1068 	metasize = metasize > 0 ? metasize : 0;
1069 	if (metasize)
1070 		skb_metadata_set(skb, metasize);
1071 
1072 	if (unlikely(xdp_buff_has_frags(xdp)))
1073 		xdp_update_skb_shared_info(skb, nr_frags,
1074 					   sinfo->xdp_frags_size,
1075 					   xdp_frags_truesz,
1076 					   xdp_buff_is_frag_pfmemalloc(xdp));
1077 
1078 	return skb;
1079 }
1080 
1081 /* TODO: build xdp in big mode */
1082 static int virtnet_build_xdp_buff_mrg(struct net_device *dev,
1083 				      struct virtnet_info *vi,
1084 				      struct receive_queue *rq,
1085 				      struct xdp_buff *xdp,
1086 				      void *buf,
1087 				      unsigned int len,
1088 				      unsigned int frame_sz,
1089 				      int *num_buf,
1090 				      unsigned int *xdp_frags_truesize,
1091 				      struct virtnet_rq_stats *stats)
1092 {
1093 	struct virtio_net_hdr_mrg_rxbuf *hdr = buf;
1094 	unsigned int headroom, tailroom, room;
1095 	unsigned int truesize, cur_frag_size;
1096 	struct skb_shared_info *shinfo;
1097 	unsigned int xdp_frags_truesz = 0;
1098 	struct page *page;
1099 	skb_frag_t *frag;
1100 	int offset;
1101 	void *ctx;
1102 
1103 	xdp_init_buff(xdp, frame_sz, &rq->xdp_rxq);
1104 	xdp_prepare_buff(xdp, buf - VIRTIO_XDP_HEADROOM,
1105 			 VIRTIO_XDP_HEADROOM + vi->hdr_len, len - vi->hdr_len, true);
1106 
1107 	if (!*num_buf)
1108 		return 0;
1109 
1110 	if (*num_buf > 1) {
1111 		/* If we want to build multi-buffer xdp, we need
1112 		 * to specify that the flags of xdp_buff have the
1113 		 * XDP_FLAGS_HAS_FRAG bit.
1114 		 */
1115 		if (!xdp_buff_has_frags(xdp))
1116 			xdp_buff_set_frags_flag(xdp);
1117 
1118 		shinfo = xdp_get_shared_info_from_buff(xdp);
1119 		shinfo->nr_frags = 0;
1120 		shinfo->xdp_frags_size = 0;
1121 	}
1122 
1123 	if (*num_buf > MAX_SKB_FRAGS + 1)
1124 		return -EINVAL;
1125 
1126 	while (--*num_buf > 0) {
1127 		buf = virtqueue_get_buf_ctx(rq->vq, &len, &ctx);
1128 		if (unlikely(!buf)) {
1129 			pr_debug("%s: rx error: %d buffers out of %d missing\n",
1130 				 dev->name, *num_buf,
1131 				 virtio16_to_cpu(vi->vdev, hdr->num_buffers));
1132 			dev->stats.rx_length_errors++;
1133 			return -EINVAL;
1134 		}
1135 
1136 		stats->bytes += len;
1137 		page = virt_to_head_page(buf);
1138 		offset = buf - page_address(page);
1139 
1140 		truesize = mergeable_ctx_to_truesize(ctx);
1141 		headroom = mergeable_ctx_to_headroom(ctx);
1142 		tailroom = headroom ? sizeof(struct skb_shared_info) : 0;
1143 		room = SKB_DATA_ALIGN(headroom + tailroom);
1144 
1145 		cur_frag_size = truesize;
1146 		xdp_frags_truesz += cur_frag_size;
1147 		if (unlikely(len > truesize - room || cur_frag_size > PAGE_SIZE)) {
1148 			put_page(page);
1149 			pr_debug("%s: rx error: len %u exceeds truesize %lu\n",
1150 				 dev->name, len, (unsigned long)(truesize - room));
1151 			dev->stats.rx_length_errors++;
1152 			return -EINVAL;
1153 		}
1154 
1155 		frag = &shinfo->frags[shinfo->nr_frags++];
1156 		__skb_frag_set_page(frag, page);
1157 		skb_frag_off_set(frag, offset);
1158 		skb_frag_size_set(frag, len);
1159 		if (page_is_pfmemalloc(page))
1160 			xdp_buff_set_frag_pfmemalloc(xdp);
1161 
1162 		shinfo->xdp_frags_size += len;
1163 	}
1164 
1165 	*xdp_frags_truesize = xdp_frags_truesz;
1166 	return 0;
1167 }
1168 
1169 static struct sk_buff *receive_mergeable(struct net_device *dev,
1170 					 struct virtnet_info *vi,
1171 					 struct receive_queue *rq,
1172 					 void *buf,
1173 					 void *ctx,
1174 					 unsigned int len,
1175 					 unsigned int *xdp_xmit,
1176 					 struct virtnet_rq_stats *stats)
1177 {
1178 	struct virtio_net_hdr_mrg_rxbuf *hdr = buf;
1179 	int num_buf = virtio16_to_cpu(vi->vdev, hdr->num_buffers);
1180 	struct page *page = virt_to_head_page(buf);
1181 	int offset = buf - page_address(page);
1182 	struct sk_buff *head_skb, *curr_skb;
1183 	struct bpf_prog *xdp_prog;
1184 	unsigned int truesize = mergeable_ctx_to_truesize(ctx);
1185 	unsigned int headroom = mergeable_ctx_to_headroom(ctx);
1186 	unsigned int tailroom = headroom ? sizeof(struct skb_shared_info) : 0;
1187 	unsigned int room = SKB_DATA_ALIGN(headroom + tailroom);
1188 	unsigned int frame_sz, xdp_room;
1189 	int err;
1190 
1191 	head_skb = NULL;
1192 	stats->bytes += len - vi->hdr_len;
1193 
1194 	if (unlikely(len > truesize - room)) {
1195 		pr_debug("%s: rx error: len %u exceeds truesize %lu\n",
1196 			 dev->name, len, (unsigned long)(truesize - room));
1197 		dev->stats.rx_length_errors++;
1198 		goto err_skb;
1199 	}
1200 
1201 	if (likely(!vi->xdp_enabled)) {
1202 		xdp_prog = NULL;
1203 		goto skip_xdp;
1204 	}
1205 
1206 	rcu_read_lock();
1207 	xdp_prog = rcu_dereference(rq->xdp_prog);
1208 	if (xdp_prog) {
1209 		unsigned int xdp_frags_truesz = 0;
1210 		struct skb_shared_info *shinfo;
1211 		struct xdp_frame *xdpf;
1212 		struct page *xdp_page;
1213 		struct xdp_buff xdp;
1214 		void *data;
1215 		u32 act;
1216 		int i;
1217 
1218 		/* Transient failure which in theory could occur if
1219 		 * in-flight packets from before XDP was enabled reach
1220 		 * the receive path after XDP is loaded.
1221 		 */
1222 		if (unlikely(hdr->hdr.gso_type))
1223 			goto err_xdp;
1224 
1225 		/* Now XDP core assumes frag size is PAGE_SIZE, but buffers
1226 		 * with headroom may add hole in truesize, which
1227 		 * make their length exceed PAGE_SIZE. So we disabled the
1228 		 * hole mechanism for xdp. See add_recvbuf_mergeable().
1229 		 */
1230 		frame_sz = truesize;
1231 
1232 		/* This happens when headroom is not enough because
1233 		 * of the buffer was prefilled before XDP is set.
1234 		 * This should only happen for the first several packets.
1235 		 * In fact, vq reset can be used here to help us clean up
1236 		 * the prefilled buffers, but many existing devices do not
1237 		 * support it, and we don't want to bother users who are
1238 		 * using xdp normally.
1239 		 */
1240 		if (!xdp_prog->aux->xdp_has_frags &&
1241 		    (num_buf > 1 || headroom < virtnet_get_headroom(vi))) {
1242 			/* linearize data for XDP */
1243 			xdp_page = xdp_linearize_page(rq, &num_buf,
1244 						      page, offset,
1245 						      VIRTIO_XDP_HEADROOM,
1246 						      &len);
1247 			frame_sz = PAGE_SIZE;
1248 
1249 			if (!xdp_page)
1250 				goto err_xdp;
1251 			offset = VIRTIO_XDP_HEADROOM;
1252 		} else if (unlikely(headroom < virtnet_get_headroom(vi))) {
1253 			xdp_room = SKB_DATA_ALIGN(VIRTIO_XDP_HEADROOM +
1254 						  sizeof(struct skb_shared_info));
1255 			if (len + xdp_room > PAGE_SIZE)
1256 				goto err_xdp;
1257 
1258 			xdp_page = alloc_page(GFP_ATOMIC);
1259 			if (!xdp_page)
1260 				goto err_xdp;
1261 
1262 			memcpy(page_address(xdp_page) + VIRTIO_XDP_HEADROOM,
1263 			       page_address(page) + offset, len);
1264 			frame_sz = PAGE_SIZE;
1265 			offset = VIRTIO_XDP_HEADROOM;
1266 		} else {
1267 			xdp_page = page;
1268 		}
1269 
1270 		data = page_address(xdp_page) + offset;
1271 		err = virtnet_build_xdp_buff_mrg(dev, vi, rq, &xdp, data, len, frame_sz,
1272 						 &num_buf, &xdp_frags_truesz, stats);
1273 		if (unlikely(err))
1274 			goto err_xdp_frags;
1275 
1276 		act = bpf_prog_run_xdp(xdp_prog, &xdp);
1277 		stats->xdp_packets++;
1278 
1279 		switch (act) {
1280 		case XDP_PASS:
1281 			head_skb = build_skb_from_xdp_buff(dev, vi, &xdp, xdp_frags_truesz);
1282 			if (unlikely(!head_skb))
1283 				goto err_xdp_frags;
1284 
1285 			if (unlikely(xdp_page != page))
1286 				put_page(page);
1287 			rcu_read_unlock();
1288 			return head_skb;
1289 		case XDP_TX:
1290 			stats->xdp_tx++;
1291 			xdpf = xdp_convert_buff_to_frame(&xdp);
1292 			if (unlikely(!xdpf)) {
1293 				netdev_dbg(dev, "convert buff to frame failed for xdp\n");
1294 				goto err_xdp_frags;
1295 			}
1296 			err = virtnet_xdp_xmit(dev, 1, &xdpf, 0);
1297 			if (unlikely(!err)) {
1298 				xdp_return_frame_rx_napi(xdpf);
1299 			} else if (unlikely(err < 0)) {
1300 				trace_xdp_exception(vi->dev, xdp_prog, act);
1301 				goto err_xdp_frags;
1302 			}
1303 			*xdp_xmit |= VIRTIO_XDP_TX;
1304 			if (unlikely(xdp_page != page))
1305 				put_page(page);
1306 			rcu_read_unlock();
1307 			goto xdp_xmit;
1308 		case XDP_REDIRECT:
1309 			stats->xdp_redirects++;
1310 			err = xdp_do_redirect(dev, &xdp, xdp_prog);
1311 			if (err)
1312 				goto err_xdp_frags;
1313 			*xdp_xmit |= VIRTIO_XDP_REDIR;
1314 			if (unlikely(xdp_page != page))
1315 				put_page(page);
1316 			rcu_read_unlock();
1317 			goto xdp_xmit;
1318 		default:
1319 			bpf_warn_invalid_xdp_action(vi->dev, xdp_prog, act);
1320 			fallthrough;
1321 		case XDP_ABORTED:
1322 			trace_xdp_exception(vi->dev, xdp_prog, act);
1323 			fallthrough;
1324 		case XDP_DROP:
1325 			goto err_xdp_frags;
1326 		}
1327 err_xdp_frags:
1328 		if (unlikely(xdp_page != page))
1329 			__free_pages(xdp_page, 0);
1330 
1331 		if (xdp_buff_has_frags(&xdp)) {
1332 			shinfo = xdp_get_shared_info_from_buff(&xdp);
1333 			for (i = 0; i < shinfo->nr_frags; i++) {
1334 				xdp_page = skb_frag_page(&shinfo->frags[i]);
1335 				put_page(xdp_page);
1336 			}
1337 		}
1338 
1339 		goto err_xdp;
1340 	}
1341 	rcu_read_unlock();
1342 
1343 skip_xdp:
1344 	head_skb = page_to_skb(vi, rq, page, offset, len, truesize, headroom);
1345 	curr_skb = head_skb;
1346 
1347 	if (unlikely(!curr_skb))
1348 		goto err_skb;
1349 	while (--num_buf) {
1350 		int num_skb_frags;
1351 
1352 		buf = virtqueue_get_buf_ctx(rq->vq, &len, &ctx);
1353 		if (unlikely(!buf)) {
1354 			pr_debug("%s: rx error: %d buffers out of %d missing\n",
1355 				 dev->name, num_buf,
1356 				 virtio16_to_cpu(vi->vdev,
1357 						 hdr->num_buffers));
1358 			dev->stats.rx_length_errors++;
1359 			goto err_buf;
1360 		}
1361 
1362 		stats->bytes += len;
1363 		page = virt_to_head_page(buf);
1364 
1365 		truesize = mergeable_ctx_to_truesize(ctx);
1366 		headroom = mergeable_ctx_to_headroom(ctx);
1367 		tailroom = headroom ? sizeof(struct skb_shared_info) : 0;
1368 		room = SKB_DATA_ALIGN(headroom + tailroom);
1369 		if (unlikely(len > truesize - room)) {
1370 			pr_debug("%s: rx error: len %u exceeds truesize %lu\n",
1371 				 dev->name, len, (unsigned long)(truesize - room));
1372 			dev->stats.rx_length_errors++;
1373 			goto err_skb;
1374 		}
1375 
1376 		num_skb_frags = skb_shinfo(curr_skb)->nr_frags;
1377 		if (unlikely(num_skb_frags == MAX_SKB_FRAGS)) {
1378 			struct sk_buff *nskb = alloc_skb(0, GFP_ATOMIC);
1379 
1380 			if (unlikely(!nskb))
1381 				goto err_skb;
1382 			if (curr_skb == head_skb)
1383 				skb_shinfo(curr_skb)->frag_list = nskb;
1384 			else
1385 				curr_skb->next = nskb;
1386 			curr_skb = nskb;
1387 			head_skb->truesize += nskb->truesize;
1388 			num_skb_frags = 0;
1389 		}
1390 		if (curr_skb != head_skb) {
1391 			head_skb->data_len += len;
1392 			head_skb->len += len;
1393 			head_skb->truesize += truesize;
1394 		}
1395 		offset = buf - page_address(page);
1396 		if (skb_can_coalesce(curr_skb, num_skb_frags, page, offset)) {
1397 			put_page(page);
1398 			skb_coalesce_rx_frag(curr_skb, num_skb_frags - 1,
1399 					     len, truesize);
1400 		} else {
1401 			skb_add_rx_frag(curr_skb, num_skb_frags, page,
1402 					offset, len, truesize);
1403 		}
1404 	}
1405 
1406 	ewma_pkt_len_add(&rq->mrg_avg_pkt_len, head_skb->len);
1407 	return head_skb;
1408 
1409 err_xdp:
1410 	rcu_read_unlock();
1411 	stats->xdp_drops++;
1412 err_skb:
1413 	put_page(page);
1414 	while (num_buf-- > 1) {
1415 		buf = virtqueue_get_buf(rq->vq, &len);
1416 		if (unlikely(!buf)) {
1417 			pr_debug("%s: rx error: %d buffers missing\n",
1418 				 dev->name, num_buf);
1419 			dev->stats.rx_length_errors++;
1420 			break;
1421 		}
1422 		stats->bytes += len;
1423 		page = virt_to_head_page(buf);
1424 		put_page(page);
1425 	}
1426 err_buf:
1427 	stats->drops++;
1428 	dev_kfree_skb(head_skb);
1429 xdp_xmit:
1430 	return NULL;
1431 }
1432 
1433 static void virtio_skb_set_hash(const struct virtio_net_hdr_v1_hash *hdr_hash,
1434 				struct sk_buff *skb)
1435 {
1436 	enum pkt_hash_types rss_hash_type;
1437 
1438 	if (!hdr_hash || !skb)
1439 		return;
1440 
1441 	switch (__le16_to_cpu(hdr_hash->hash_report)) {
1442 	case VIRTIO_NET_HASH_REPORT_TCPv4:
1443 	case VIRTIO_NET_HASH_REPORT_UDPv4:
1444 	case VIRTIO_NET_HASH_REPORT_TCPv6:
1445 	case VIRTIO_NET_HASH_REPORT_UDPv6:
1446 	case VIRTIO_NET_HASH_REPORT_TCPv6_EX:
1447 	case VIRTIO_NET_HASH_REPORT_UDPv6_EX:
1448 		rss_hash_type = PKT_HASH_TYPE_L4;
1449 		break;
1450 	case VIRTIO_NET_HASH_REPORT_IPv4:
1451 	case VIRTIO_NET_HASH_REPORT_IPv6:
1452 	case VIRTIO_NET_HASH_REPORT_IPv6_EX:
1453 		rss_hash_type = PKT_HASH_TYPE_L3;
1454 		break;
1455 	case VIRTIO_NET_HASH_REPORT_NONE:
1456 	default:
1457 		rss_hash_type = PKT_HASH_TYPE_NONE;
1458 	}
1459 	skb_set_hash(skb, __le32_to_cpu(hdr_hash->hash_value), rss_hash_type);
1460 }
1461 
1462 static void receive_buf(struct virtnet_info *vi, struct receive_queue *rq,
1463 			void *buf, unsigned int len, void **ctx,
1464 			unsigned int *xdp_xmit,
1465 			struct virtnet_rq_stats *stats)
1466 {
1467 	struct net_device *dev = vi->dev;
1468 	struct sk_buff *skb;
1469 	struct virtio_net_hdr_mrg_rxbuf *hdr;
1470 
1471 	if (unlikely(len < vi->hdr_len + ETH_HLEN)) {
1472 		pr_debug("%s: short packet %i\n", dev->name, len);
1473 		dev->stats.rx_length_errors++;
1474 		virtnet_rq_free_unused_buf(rq->vq, buf);
1475 		return;
1476 	}
1477 
1478 	if (vi->mergeable_rx_bufs)
1479 		skb = receive_mergeable(dev, vi, rq, buf, ctx, len, xdp_xmit,
1480 					stats);
1481 	else if (vi->big_packets)
1482 		skb = receive_big(dev, vi, rq, buf, len, stats);
1483 	else
1484 		skb = receive_small(dev, vi, rq, buf, ctx, len, xdp_xmit, stats);
1485 
1486 	if (unlikely(!skb))
1487 		return;
1488 
1489 	hdr = skb_vnet_hdr(skb);
1490 	if (dev->features & NETIF_F_RXHASH && vi->has_rss_hash_report)
1491 		virtio_skb_set_hash((const struct virtio_net_hdr_v1_hash *)hdr, skb);
1492 
1493 	if (hdr->hdr.flags & VIRTIO_NET_HDR_F_DATA_VALID)
1494 		skb->ip_summed = CHECKSUM_UNNECESSARY;
1495 
1496 	if (virtio_net_hdr_to_skb(skb, &hdr->hdr,
1497 				  virtio_is_little_endian(vi->vdev))) {
1498 		net_warn_ratelimited("%s: bad gso: type: %u, size: %u\n",
1499 				     dev->name, hdr->hdr.gso_type,
1500 				     hdr->hdr.gso_size);
1501 		goto frame_err;
1502 	}
1503 
1504 	skb_record_rx_queue(skb, vq2rxq(rq->vq));
1505 	skb->protocol = eth_type_trans(skb, dev);
1506 	pr_debug("Receiving skb proto 0x%04x len %i type %i\n",
1507 		 ntohs(skb->protocol), skb->len, skb->pkt_type);
1508 
1509 	napi_gro_receive(&rq->napi, skb);
1510 	return;
1511 
1512 frame_err:
1513 	dev->stats.rx_frame_errors++;
1514 	dev_kfree_skb(skb);
1515 }
1516 
1517 /* Unlike mergeable buffers, all buffers are allocated to the
1518  * same size, except for the headroom. For this reason we do
1519  * not need to use  mergeable_len_to_ctx here - it is enough
1520  * to store the headroom as the context ignoring the truesize.
1521  */
1522 static int add_recvbuf_small(struct virtnet_info *vi, struct receive_queue *rq,
1523 			     gfp_t gfp)
1524 {
1525 	struct page_frag *alloc_frag = &rq->alloc_frag;
1526 	char *buf;
1527 	unsigned int xdp_headroom = virtnet_get_headroom(vi);
1528 	void *ctx = (void *)(unsigned long)xdp_headroom;
1529 	int len = vi->hdr_len + VIRTNET_RX_PAD + GOOD_PACKET_LEN + xdp_headroom;
1530 	int err;
1531 
1532 	len = SKB_DATA_ALIGN(len) +
1533 	      SKB_DATA_ALIGN(sizeof(struct skb_shared_info));
1534 	if (unlikely(!skb_page_frag_refill(len, alloc_frag, gfp)))
1535 		return -ENOMEM;
1536 
1537 	buf = (char *)page_address(alloc_frag->page) + alloc_frag->offset;
1538 	get_page(alloc_frag->page);
1539 	alloc_frag->offset += len;
1540 	sg_init_one(rq->sg, buf + VIRTNET_RX_PAD + xdp_headroom,
1541 		    vi->hdr_len + GOOD_PACKET_LEN);
1542 	err = virtqueue_add_inbuf_ctx(rq->vq, rq->sg, 1, buf, ctx, gfp);
1543 	if (err < 0)
1544 		put_page(virt_to_head_page(buf));
1545 	return err;
1546 }
1547 
1548 static int add_recvbuf_big(struct virtnet_info *vi, struct receive_queue *rq,
1549 			   gfp_t gfp)
1550 {
1551 	struct page *first, *list = NULL;
1552 	char *p;
1553 	int i, err, offset;
1554 
1555 	sg_init_table(rq->sg, vi->big_packets_num_skbfrags + 2);
1556 
1557 	/* page in rq->sg[vi->big_packets_num_skbfrags + 1] is list tail */
1558 	for (i = vi->big_packets_num_skbfrags + 1; i > 1; --i) {
1559 		first = get_a_page(rq, gfp);
1560 		if (!first) {
1561 			if (list)
1562 				give_pages(rq, list);
1563 			return -ENOMEM;
1564 		}
1565 		sg_set_buf(&rq->sg[i], page_address(first), PAGE_SIZE);
1566 
1567 		/* chain new page in list head to match sg */
1568 		first->private = (unsigned long)list;
1569 		list = first;
1570 	}
1571 
1572 	first = get_a_page(rq, gfp);
1573 	if (!first) {
1574 		give_pages(rq, list);
1575 		return -ENOMEM;
1576 	}
1577 	p = page_address(first);
1578 
1579 	/* rq->sg[0], rq->sg[1] share the same page */
1580 	/* a separated rq->sg[0] for header - required in case !any_header_sg */
1581 	sg_set_buf(&rq->sg[0], p, vi->hdr_len);
1582 
1583 	/* rq->sg[1] for data packet, from offset */
1584 	offset = sizeof(struct padded_vnet_hdr);
1585 	sg_set_buf(&rq->sg[1], p + offset, PAGE_SIZE - offset);
1586 
1587 	/* chain first in list head */
1588 	first->private = (unsigned long)list;
1589 	err = virtqueue_add_inbuf(rq->vq, rq->sg, vi->big_packets_num_skbfrags + 2,
1590 				  first, gfp);
1591 	if (err < 0)
1592 		give_pages(rq, first);
1593 
1594 	return err;
1595 }
1596 
1597 static unsigned int get_mergeable_buf_len(struct receive_queue *rq,
1598 					  struct ewma_pkt_len *avg_pkt_len,
1599 					  unsigned int room)
1600 {
1601 	struct virtnet_info *vi = rq->vq->vdev->priv;
1602 	const size_t hdr_len = vi->hdr_len;
1603 	unsigned int len;
1604 
1605 	if (room)
1606 		return PAGE_SIZE - room;
1607 
1608 	len = hdr_len +	clamp_t(unsigned int, ewma_pkt_len_read(avg_pkt_len),
1609 				rq->min_buf_len, PAGE_SIZE - hdr_len);
1610 
1611 	return ALIGN(len, L1_CACHE_BYTES);
1612 }
1613 
1614 static int add_recvbuf_mergeable(struct virtnet_info *vi,
1615 				 struct receive_queue *rq, gfp_t gfp)
1616 {
1617 	struct page_frag *alloc_frag = &rq->alloc_frag;
1618 	unsigned int headroom = virtnet_get_headroom(vi);
1619 	unsigned int tailroom = headroom ? sizeof(struct skb_shared_info) : 0;
1620 	unsigned int room = SKB_DATA_ALIGN(headroom + tailroom);
1621 	char *buf;
1622 	void *ctx;
1623 	int err;
1624 	unsigned int len, hole;
1625 
1626 	/* Extra tailroom is needed to satisfy XDP's assumption. This
1627 	 * means rx frags coalescing won't work, but consider we've
1628 	 * disabled GSO for XDP, it won't be a big issue.
1629 	 */
1630 	len = get_mergeable_buf_len(rq, &rq->mrg_avg_pkt_len, room);
1631 	if (unlikely(!skb_page_frag_refill(len + room, alloc_frag, gfp)))
1632 		return -ENOMEM;
1633 
1634 	buf = (char *)page_address(alloc_frag->page) + alloc_frag->offset;
1635 	buf += headroom; /* advance address leaving hole at front of pkt */
1636 	get_page(alloc_frag->page);
1637 	alloc_frag->offset += len + room;
1638 	hole = alloc_frag->size - alloc_frag->offset;
1639 	if (hole < len + room) {
1640 		/* To avoid internal fragmentation, if there is very likely not
1641 		 * enough space for another buffer, add the remaining space to
1642 		 * the current buffer.
1643 		 * XDP core assumes that frame_size of xdp_buff and the length
1644 		 * of the frag are PAGE_SIZE, so we disable the hole mechanism.
1645 		 */
1646 		if (!headroom)
1647 			len += hole;
1648 		alloc_frag->offset += hole;
1649 	}
1650 
1651 	sg_init_one(rq->sg, buf, len);
1652 	ctx = mergeable_len_to_ctx(len + room, headroom);
1653 	err = virtqueue_add_inbuf_ctx(rq->vq, rq->sg, 1, buf, ctx, gfp);
1654 	if (err < 0)
1655 		put_page(virt_to_head_page(buf));
1656 
1657 	return err;
1658 }
1659 
1660 /*
1661  * Returns false if we couldn't fill entirely (OOM).
1662  *
1663  * Normally run in the receive path, but can also be run from ndo_open
1664  * before we're receiving packets, or from refill_work which is
1665  * careful to disable receiving (using napi_disable).
1666  */
1667 static bool try_fill_recv(struct virtnet_info *vi, struct receive_queue *rq,
1668 			  gfp_t gfp)
1669 {
1670 	int err;
1671 	bool oom;
1672 
1673 	do {
1674 		if (vi->mergeable_rx_bufs)
1675 			err = add_recvbuf_mergeable(vi, rq, gfp);
1676 		else if (vi->big_packets)
1677 			err = add_recvbuf_big(vi, rq, gfp);
1678 		else
1679 			err = add_recvbuf_small(vi, rq, gfp);
1680 
1681 		oom = err == -ENOMEM;
1682 		if (err)
1683 			break;
1684 	} while (rq->vq->num_free);
1685 	if (virtqueue_kick_prepare(rq->vq) && virtqueue_notify(rq->vq)) {
1686 		unsigned long flags;
1687 
1688 		flags = u64_stats_update_begin_irqsave(&rq->stats.syncp);
1689 		rq->stats.kicks++;
1690 		u64_stats_update_end_irqrestore(&rq->stats.syncp, flags);
1691 	}
1692 
1693 	return !oom;
1694 }
1695 
1696 static void skb_recv_done(struct virtqueue *rvq)
1697 {
1698 	struct virtnet_info *vi = rvq->vdev->priv;
1699 	struct receive_queue *rq = &vi->rq[vq2rxq(rvq)];
1700 
1701 	virtqueue_napi_schedule(&rq->napi, rvq);
1702 }
1703 
1704 static void virtnet_napi_enable(struct virtqueue *vq, struct napi_struct *napi)
1705 {
1706 	napi_enable(napi);
1707 
1708 	/* If all buffers were filled by other side before we napi_enabled, we
1709 	 * won't get another interrupt, so process any outstanding packets now.
1710 	 * Call local_bh_enable after to trigger softIRQ processing.
1711 	 */
1712 	local_bh_disable();
1713 	virtqueue_napi_schedule(napi, vq);
1714 	local_bh_enable();
1715 }
1716 
1717 static void virtnet_napi_tx_enable(struct virtnet_info *vi,
1718 				   struct virtqueue *vq,
1719 				   struct napi_struct *napi)
1720 {
1721 	if (!napi->weight)
1722 		return;
1723 
1724 	/* Tx napi touches cachelines on the cpu handling tx interrupts. Only
1725 	 * enable the feature if this is likely affine with the transmit path.
1726 	 */
1727 	if (!vi->affinity_hint_set) {
1728 		napi->weight = 0;
1729 		return;
1730 	}
1731 
1732 	return virtnet_napi_enable(vq, napi);
1733 }
1734 
1735 static void virtnet_napi_tx_disable(struct napi_struct *napi)
1736 {
1737 	if (napi->weight)
1738 		napi_disable(napi);
1739 }
1740 
1741 static void refill_work(struct work_struct *work)
1742 {
1743 	struct virtnet_info *vi =
1744 		container_of(work, struct virtnet_info, refill.work);
1745 	bool still_empty;
1746 	int i;
1747 
1748 	for (i = 0; i < vi->curr_queue_pairs; i++) {
1749 		struct receive_queue *rq = &vi->rq[i];
1750 
1751 		napi_disable(&rq->napi);
1752 		still_empty = !try_fill_recv(vi, rq, GFP_KERNEL);
1753 		virtnet_napi_enable(rq->vq, &rq->napi);
1754 
1755 		/* In theory, this can happen: if we don't get any buffers in
1756 		 * we will *never* try to fill again.
1757 		 */
1758 		if (still_empty)
1759 			schedule_delayed_work(&vi->refill, HZ/2);
1760 	}
1761 }
1762 
1763 static int virtnet_receive(struct receive_queue *rq, int budget,
1764 			   unsigned int *xdp_xmit)
1765 {
1766 	struct virtnet_info *vi = rq->vq->vdev->priv;
1767 	struct virtnet_rq_stats stats = {};
1768 	unsigned int len;
1769 	void *buf;
1770 	int i;
1771 
1772 	if (!vi->big_packets || vi->mergeable_rx_bufs) {
1773 		void *ctx;
1774 
1775 		while (stats.packets < budget &&
1776 		       (buf = virtqueue_get_buf_ctx(rq->vq, &len, &ctx))) {
1777 			receive_buf(vi, rq, buf, len, ctx, xdp_xmit, &stats);
1778 			stats.packets++;
1779 		}
1780 	} else {
1781 		while (stats.packets < budget &&
1782 		       (buf = virtqueue_get_buf(rq->vq, &len)) != NULL) {
1783 			receive_buf(vi, rq, buf, len, NULL, xdp_xmit, &stats);
1784 			stats.packets++;
1785 		}
1786 	}
1787 
1788 	if (rq->vq->num_free > min((unsigned int)budget, virtqueue_get_vring_size(rq->vq)) / 2) {
1789 		if (!try_fill_recv(vi, rq, GFP_ATOMIC)) {
1790 			spin_lock(&vi->refill_lock);
1791 			if (vi->refill_enabled)
1792 				schedule_delayed_work(&vi->refill, 0);
1793 			spin_unlock(&vi->refill_lock);
1794 		}
1795 	}
1796 
1797 	u64_stats_update_begin(&rq->stats.syncp);
1798 	for (i = 0; i < VIRTNET_RQ_STATS_LEN; i++) {
1799 		size_t offset = virtnet_rq_stats_desc[i].offset;
1800 		u64 *item;
1801 
1802 		item = (u64 *)((u8 *)&rq->stats + offset);
1803 		*item += *(u64 *)((u8 *)&stats + offset);
1804 	}
1805 	u64_stats_update_end(&rq->stats.syncp);
1806 
1807 	return stats.packets;
1808 }
1809 
1810 static void virtnet_poll_cleantx(struct receive_queue *rq)
1811 {
1812 	struct virtnet_info *vi = rq->vq->vdev->priv;
1813 	unsigned int index = vq2rxq(rq->vq);
1814 	struct send_queue *sq = &vi->sq[index];
1815 	struct netdev_queue *txq = netdev_get_tx_queue(vi->dev, index);
1816 
1817 	if (!sq->napi.weight || is_xdp_raw_buffer_queue(vi, index))
1818 		return;
1819 
1820 	if (__netif_tx_trylock(txq)) {
1821 		if (sq->reset) {
1822 			__netif_tx_unlock(txq);
1823 			return;
1824 		}
1825 
1826 		do {
1827 			virtqueue_disable_cb(sq->vq);
1828 			free_old_xmit_skbs(sq, true);
1829 		} while (unlikely(!virtqueue_enable_cb_delayed(sq->vq)));
1830 
1831 		if (sq->vq->num_free >= 2 + MAX_SKB_FRAGS)
1832 			netif_tx_wake_queue(txq);
1833 
1834 		__netif_tx_unlock(txq);
1835 	}
1836 }
1837 
1838 static int virtnet_poll(struct napi_struct *napi, int budget)
1839 {
1840 	struct receive_queue *rq =
1841 		container_of(napi, struct receive_queue, napi);
1842 	struct virtnet_info *vi = rq->vq->vdev->priv;
1843 	struct send_queue *sq;
1844 	unsigned int received;
1845 	unsigned int xdp_xmit = 0;
1846 
1847 	virtnet_poll_cleantx(rq);
1848 
1849 	received = virtnet_receive(rq, budget, &xdp_xmit);
1850 
1851 	if (xdp_xmit & VIRTIO_XDP_REDIR)
1852 		xdp_do_flush();
1853 
1854 	/* Out of packets? */
1855 	if (received < budget)
1856 		virtqueue_napi_complete(napi, rq->vq, received);
1857 
1858 	if (xdp_xmit & VIRTIO_XDP_TX) {
1859 		sq = virtnet_xdp_get_sq(vi);
1860 		if (virtqueue_kick_prepare(sq->vq) && virtqueue_notify(sq->vq)) {
1861 			u64_stats_update_begin(&sq->stats.syncp);
1862 			sq->stats.kicks++;
1863 			u64_stats_update_end(&sq->stats.syncp);
1864 		}
1865 		virtnet_xdp_put_sq(vi, sq);
1866 	}
1867 
1868 	return received;
1869 }
1870 
1871 static void virtnet_disable_queue_pair(struct virtnet_info *vi, int qp_index)
1872 {
1873 	virtnet_napi_tx_disable(&vi->sq[qp_index].napi);
1874 	napi_disable(&vi->rq[qp_index].napi);
1875 	xdp_rxq_info_unreg(&vi->rq[qp_index].xdp_rxq);
1876 }
1877 
1878 static int virtnet_enable_queue_pair(struct virtnet_info *vi, int qp_index)
1879 {
1880 	struct net_device *dev = vi->dev;
1881 	int err;
1882 
1883 	err = xdp_rxq_info_reg(&vi->rq[qp_index].xdp_rxq, dev, qp_index,
1884 			       vi->rq[qp_index].napi.napi_id);
1885 	if (err < 0)
1886 		return err;
1887 
1888 	err = xdp_rxq_info_reg_mem_model(&vi->rq[qp_index].xdp_rxq,
1889 					 MEM_TYPE_PAGE_SHARED, NULL);
1890 	if (err < 0)
1891 		goto err_xdp_reg_mem_model;
1892 
1893 	virtnet_napi_enable(vi->rq[qp_index].vq, &vi->rq[qp_index].napi);
1894 	virtnet_napi_tx_enable(vi, vi->sq[qp_index].vq, &vi->sq[qp_index].napi);
1895 
1896 	return 0;
1897 
1898 err_xdp_reg_mem_model:
1899 	xdp_rxq_info_unreg(&vi->rq[qp_index].xdp_rxq);
1900 	return err;
1901 }
1902 
1903 static int virtnet_open(struct net_device *dev)
1904 {
1905 	struct virtnet_info *vi = netdev_priv(dev);
1906 	int i, err;
1907 
1908 	enable_delayed_refill(vi);
1909 
1910 	for (i = 0; i < vi->max_queue_pairs; i++) {
1911 		if (i < vi->curr_queue_pairs)
1912 			/* Make sure we have some buffers: if oom use wq. */
1913 			if (!try_fill_recv(vi, &vi->rq[i], GFP_KERNEL))
1914 				schedule_delayed_work(&vi->refill, 0);
1915 
1916 		err = virtnet_enable_queue_pair(vi, i);
1917 		if (err < 0)
1918 			goto err_enable_qp;
1919 	}
1920 
1921 	return 0;
1922 
1923 err_enable_qp:
1924 	disable_delayed_refill(vi);
1925 	cancel_delayed_work_sync(&vi->refill);
1926 
1927 	for (i--; i >= 0; i--)
1928 		virtnet_disable_queue_pair(vi, i);
1929 	return err;
1930 }
1931 
1932 static int virtnet_poll_tx(struct napi_struct *napi, int budget)
1933 {
1934 	struct send_queue *sq = container_of(napi, struct send_queue, napi);
1935 	struct virtnet_info *vi = sq->vq->vdev->priv;
1936 	unsigned int index = vq2txq(sq->vq);
1937 	struct netdev_queue *txq;
1938 	int opaque;
1939 	bool done;
1940 
1941 	if (unlikely(is_xdp_raw_buffer_queue(vi, index))) {
1942 		/* We don't need to enable cb for XDP */
1943 		napi_complete_done(napi, 0);
1944 		return 0;
1945 	}
1946 
1947 	txq = netdev_get_tx_queue(vi->dev, index);
1948 	__netif_tx_lock(txq, raw_smp_processor_id());
1949 	virtqueue_disable_cb(sq->vq);
1950 	free_old_xmit_skbs(sq, true);
1951 
1952 	if (sq->vq->num_free >= 2 + MAX_SKB_FRAGS)
1953 		netif_tx_wake_queue(txq);
1954 
1955 	opaque = virtqueue_enable_cb_prepare(sq->vq);
1956 
1957 	done = napi_complete_done(napi, 0);
1958 
1959 	if (!done)
1960 		virtqueue_disable_cb(sq->vq);
1961 
1962 	__netif_tx_unlock(txq);
1963 
1964 	if (done) {
1965 		if (unlikely(virtqueue_poll(sq->vq, opaque))) {
1966 			if (napi_schedule_prep(napi)) {
1967 				__netif_tx_lock(txq, raw_smp_processor_id());
1968 				virtqueue_disable_cb(sq->vq);
1969 				__netif_tx_unlock(txq);
1970 				__napi_schedule(napi);
1971 			}
1972 		}
1973 	}
1974 
1975 	return 0;
1976 }
1977 
1978 static int xmit_skb(struct send_queue *sq, struct sk_buff *skb)
1979 {
1980 	struct virtio_net_hdr_mrg_rxbuf *hdr;
1981 	const unsigned char *dest = ((struct ethhdr *)skb->data)->h_dest;
1982 	struct virtnet_info *vi = sq->vq->vdev->priv;
1983 	int num_sg;
1984 	unsigned hdr_len = vi->hdr_len;
1985 	bool can_push;
1986 
1987 	pr_debug("%s: xmit %p %pM\n", vi->dev->name, skb, dest);
1988 
1989 	can_push = vi->any_header_sg &&
1990 		!((unsigned long)skb->data & (__alignof__(*hdr) - 1)) &&
1991 		!skb_header_cloned(skb) && skb_headroom(skb) >= hdr_len;
1992 	/* Even if we can, don't push here yet as this would skew
1993 	 * csum_start offset below. */
1994 	if (can_push)
1995 		hdr = (struct virtio_net_hdr_mrg_rxbuf *)(skb->data - hdr_len);
1996 	else
1997 		hdr = skb_vnet_hdr(skb);
1998 
1999 	if (virtio_net_hdr_from_skb(skb, &hdr->hdr,
2000 				    virtio_is_little_endian(vi->vdev), false,
2001 				    0))
2002 		return -EPROTO;
2003 
2004 	if (vi->mergeable_rx_bufs)
2005 		hdr->num_buffers = 0;
2006 
2007 	sg_init_table(sq->sg, skb_shinfo(skb)->nr_frags + (can_push ? 1 : 2));
2008 	if (can_push) {
2009 		__skb_push(skb, hdr_len);
2010 		num_sg = skb_to_sgvec(skb, sq->sg, 0, skb->len);
2011 		if (unlikely(num_sg < 0))
2012 			return num_sg;
2013 		/* Pull header back to avoid skew in tx bytes calculations. */
2014 		__skb_pull(skb, hdr_len);
2015 	} else {
2016 		sg_set_buf(sq->sg, hdr, hdr_len);
2017 		num_sg = skb_to_sgvec(skb, sq->sg + 1, 0, skb->len);
2018 		if (unlikely(num_sg < 0))
2019 			return num_sg;
2020 		num_sg++;
2021 	}
2022 	return virtqueue_add_outbuf(sq->vq, sq->sg, num_sg, skb, GFP_ATOMIC);
2023 }
2024 
2025 static netdev_tx_t start_xmit(struct sk_buff *skb, struct net_device *dev)
2026 {
2027 	struct virtnet_info *vi = netdev_priv(dev);
2028 	int qnum = skb_get_queue_mapping(skb);
2029 	struct send_queue *sq = &vi->sq[qnum];
2030 	int err;
2031 	struct netdev_queue *txq = netdev_get_tx_queue(dev, qnum);
2032 	bool kick = !netdev_xmit_more();
2033 	bool use_napi = sq->napi.weight;
2034 
2035 	/* Free up any pending old buffers before queueing new ones. */
2036 	do {
2037 		if (use_napi)
2038 			virtqueue_disable_cb(sq->vq);
2039 
2040 		free_old_xmit_skbs(sq, false);
2041 
2042 	} while (use_napi && kick &&
2043 	       unlikely(!virtqueue_enable_cb_delayed(sq->vq)));
2044 
2045 	/* timestamp packet in software */
2046 	skb_tx_timestamp(skb);
2047 
2048 	/* Try to transmit */
2049 	err = xmit_skb(sq, skb);
2050 
2051 	/* This should not happen! */
2052 	if (unlikely(err)) {
2053 		dev->stats.tx_fifo_errors++;
2054 		if (net_ratelimit())
2055 			dev_warn(&dev->dev,
2056 				 "Unexpected TXQ (%d) queue failure: %d\n",
2057 				 qnum, err);
2058 		dev->stats.tx_dropped++;
2059 		dev_kfree_skb_any(skb);
2060 		return NETDEV_TX_OK;
2061 	}
2062 
2063 	/* Don't wait up for transmitted skbs to be freed. */
2064 	if (!use_napi) {
2065 		skb_orphan(skb);
2066 		nf_reset_ct(skb);
2067 	}
2068 
2069 	check_sq_full_and_disable(vi, dev, sq);
2070 
2071 	if (kick || netif_xmit_stopped(txq)) {
2072 		if (virtqueue_kick_prepare(sq->vq) && virtqueue_notify(sq->vq)) {
2073 			u64_stats_update_begin(&sq->stats.syncp);
2074 			sq->stats.kicks++;
2075 			u64_stats_update_end(&sq->stats.syncp);
2076 		}
2077 	}
2078 
2079 	return NETDEV_TX_OK;
2080 }
2081 
2082 static int virtnet_rx_resize(struct virtnet_info *vi,
2083 			     struct receive_queue *rq, u32 ring_num)
2084 {
2085 	bool running = netif_running(vi->dev);
2086 	int err, qindex;
2087 
2088 	qindex = rq - vi->rq;
2089 
2090 	if (running)
2091 		napi_disable(&rq->napi);
2092 
2093 	err = virtqueue_resize(rq->vq, ring_num, virtnet_rq_free_unused_buf);
2094 	if (err)
2095 		netdev_err(vi->dev, "resize rx fail: rx queue index: %d err: %d\n", qindex, err);
2096 
2097 	if (!try_fill_recv(vi, rq, GFP_KERNEL))
2098 		schedule_delayed_work(&vi->refill, 0);
2099 
2100 	if (running)
2101 		virtnet_napi_enable(rq->vq, &rq->napi);
2102 	return err;
2103 }
2104 
2105 static int virtnet_tx_resize(struct virtnet_info *vi,
2106 			     struct send_queue *sq, u32 ring_num)
2107 {
2108 	bool running = netif_running(vi->dev);
2109 	struct netdev_queue *txq;
2110 	int err, qindex;
2111 
2112 	qindex = sq - vi->sq;
2113 
2114 	if (running)
2115 		virtnet_napi_tx_disable(&sq->napi);
2116 
2117 	txq = netdev_get_tx_queue(vi->dev, qindex);
2118 
2119 	/* 1. wait all ximt complete
2120 	 * 2. fix the race of netif_stop_subqueue() vs netif_start_subqueue()
2121 	 */
2122 	__netif_tx_lock_bh(txq);
2123 
2124 	/* Prevent rx poll from accessing sq. */
2125 	sq->reset = true;
2126 
2127 	/* Prevent the upper layer from trying to send packets. */
2128 	netif_stop_subqueue(vi->dev, qindex);
2129 
2130 	__netif_tx_unlock_bh(txq);
2131 
2132 	err = virtqueue_resize(sq->vq, ring_num, virtnet_sq_free_unused_buf);
2133 	if (err)
2134 		netdev_err(vi->dev, "resize tx fail: tx queue index: %d err: %d\n", qindex, err);
2135 
2136 	__netif_tx_lock_bh(txq);
2137 	sq->reset = false;
2138 	netif_tx_wake_queue(txq);
2139 	__netif_tx_unlock_bh(txq);
2140 
2141 	if (running)
2142 		virtnet_napi_tx_enable(vi, sq->vq, &sq->napi);
2143 	return err;
2144 }
2145 
2146 /*
2147  * Send command via the control virtqueue and check status.  Commands
2148  * supported by the hypervisor, as indicated by feature bits, should
2149  * never fail unless improperly formatted.
2150  */
2151 static bool virtnet_send_command(struct virtnet_info *vi, u8 class, u8 cmd,
2152 				 struct scatterlist *out)
2153 {
2154 	struct scatterlist *sgs[4], hdr, stat;
2155 	unsigned out_num = 0, tmp;
2156 	int ret;
2157 
2158 	/* Caller should know better */
2159 	BUG_ON(!virtio_has_feature(vi->vdev, VIRTIO_NET_F_CTRL_VQ));
2160 
2161 	vi->ctrl->status = ~0;
2162 	vi->ctrl->hdr.class = class;
2163 	vi->ctrl->hdr.cmd = cmd;
2164 	/* Add header */
2165 	sg_init_one(&hdr, &vi->ctrl->hdr, sizeof(vi->ctrl->hdr));
2166 	sgs[out_num++] = &hdr;
2167 
2168 	if (out)
2169 		sgs[out_num++] = out;
2170 
2171 	/* Add return status. */
2172 	sg_init_one(&stat, &vi->ctrl->status, sizeof(vi->ctrl->status));
2173 	sgs[out_num] = &stat;
2174 
2175 	BUG_ON(out_num + 1 > ARRAY_SIZE(sgs));
2176 	ret = virtqueue_add_sgs(vi->cvq, sgs, out_num, 1, vi, GFP_ATOMIC);
2177 	if (ret < 0) {
2178 		dev_warn(&vi->vdev->dev,
2179 			 "Failed to add sgs for command vq: %d\n.", ret);
2180 		return false;
2181 	}
2182 
2183 	if (unlikely(!virtqueue_kick(vi->cvq)))
2184 		return vi->ctrl->status == VIRTIO_NET_OK;
2185 
2186 	/* Spin for a response, the kick causes an ioport write, trapping
2187 	 * into the hypervisor, so the request should be handled immediately.
2188 	 */
2189 	while (!virtqueue_get_buf(vi->cvq, &tmp) &&
2190 	       !virtqueue_is_broken(vi->cvq))
2191 		cpu_relax();
2192 
2193 	return vi->ctrl->status == VIRTIO_NET_OK;
2194 }
2195 
2196 static int virtnet_set_mac_address(struct net_device *dev, void *p)
2197 {
2198 	struct virtnet_info *vi = netdev_priv(dev);
2199 	struct virtio_device *vdev = vi->vdev;
2200 	int ret;
2201 	struct sockaddr *addr;
2202 	struct scatterlist sg;
2203 
2204 	if (virtio_has_feature(vi->vdev, VIRTIO_NET_F_STANDBY))
2205 		return -EOPNOTSUPP;
2206 
2207 	addr = kmemdup(p, sizeof(*addr), GFP_KERNEL);
2208 	if (!addr)
2209 		return -ENOMEM;
2210 
2211 	ret = eth_prepare_mac_addr_change(dev, addr);
2212 	if (ret)
2213 		goto out;
2214 
2215 	if (virtio_has_feature(vdev, VIRTIO_NET_F_CTRL_MAC_ADDR)) {
2216 		sg_init_one(&sg, addr->sa_data, dev->addr_len);
2217 		if (!virtnet_send_command(vi, VIRTIO_NET_CTRL_MAC,
2218 					  VIRTIO_NET_CTRL_MAC_ADDR_SET, &sg)) {
2219 			dev_warn(&vdev->dev,
2220 				 "Failed to set mac address by vq command.\n");
2221 			ret = -EINVAL;
2222 			goto out;
2223 		}
2224 	} else if (virtio_has_feature(vdev, VIRTIO_NET_F_MAC) &&
2225 		   !virtio_has_feature(vdev, VIRTIO_F_VERSION_1)) {
2226 		unsigned int i;
2227 
2228 		/* Naturally, this has an atomicity problem. */
2229 		for (i = 0; i < dev->addr_len; i++)
2230 			virtio_cwrite8(vdev,
2231 				       offsetof(struct virtio_net_config, mac) +
2232 				       i, addr->sa_data[i]);
2233 	}
2234 
2235 	eth_commit_mac_addr_change(dev, p);
2236 	ret = 0;
2237 
2238 out:
2239 	kfree(addr);
2240 	return ret;
2241 }
2242 
2243 static void virtnet_stats(struct net_device *dev,
2244 			  struct rtnl_link_stats64 *tot)
2245 {
2246 	struct virtnet_info *vi = netdev_priv(dev);
2247 	unsigned int start;
2248 	int i;
2249 
2250 	for (i = 0; i < vi->max_queue_pairs; i++) {
2251 		u64 tpackets, tbytes, terrors, rpackets, rbytes, rdrops;
2252 		struct receive_queue *rq = &vi->rq[i];
2253 		struct send_queue *sq = &vi->sq[i];
2254 
2255 		do {
2256 			start = u64_stats_fetch_begin(&sq->stats.syncp);
2257 			tpackets = sq->stats.packets;
2258 			tbytes   = sq->stats.bytes;
2259 			terrors  = sq->stats.tx_timeouts;
2260 		} while (u64_stats_fetch_retry(&sq->stats.syncp, start));
2261 
2262 		do {
2263 			start = u64_stats_fetch_begin(&rq->stats.syncp);
2264 			rpackets = rq->stats.packets;
2265 			rbytes   = rq->stats.bytes;
2266 			rdrops   = rq->stats.drops;
2267 		} while (u64_stats_fetch_retry(&rq->stats.syncp, start));
2268 
2269 		tot->rx_packets += rpackets;
2270 		tot->tx_packets += tpackets;
2271 		tot->rx_bytes   += rbytes;
2272 		tot->tx_bytes   += tbytes;
2273 		tot->rx_dropped += rdrops;
2274 		tot->tx_errors  += terrors;
2275 	}
2276 
2277 	tot->tx_dropped = dev->stats.tx_dropped;
2278 	tot->tx_fifo_errors = dev->stats.tx_fifo_errors;
2279 	tot->rx_length_errors = dev->stats.rx_length_errors;
2280 	tot->rx_frame_errors = dev->stats.rx_frame_errors;
2281 }
2282 
2283 static void virtnet_ack_link_announce(struct virtnet_info *vi)
2284 {
2285 	rtnl_lock();
2286 	if (!virtnet_send_command(vi, VIRTIO_NET_CTRL_ANNOUNCE,
2287 				  VIRTIO_NET_CTRL_ANNOUNCE_ACK, NULL))
2288 		dev_warn(&vi->dev->dev, "Failed to ack link announce.\n");
2289 	rtnl_unlock();
2290 }
2291 
2292 static int _virtnet_set_queues(struct virtnet_info *vi, u16 queue_pairs)
2293 {
2294 	struct scatterlist sg;
2295 	struct net_device *dev = vi->dev;
2296 
2297 	if (!vi->has_cvq || !virtio_has_feature(vi->vdev, VIRTIO_NET_F_MQ))
2298 		return 0;
2299 
2300 	vi->ctrl->mq.virtqueue_pairs = cpu_to_virtio16(vi->vdev, queue_pairs);
2301 	sg_init_one(&sg, &vi->ctrl->mq, sizeof(vi->ctrl->mq));
2302 
2303 	if (!virtnet_send_command(vi, VIRTIO_NET_CTRL_MQ,
2304 				  VIRTIO_NET_CTRL_MQ_VQ_PAIRS_SET, &sg)) {
2305 		dev_warn(&dev->dev, "Fail to set num of queue pairs to %d\n",
2306 			 queue_pairs);
2307 		return -EINVAL;
2308 	} else {
2309 		vi->curr_queue_pairs = queue_pairs;
2310 		/* virtnet_open() will refill when device is going to up. */
2311 		if (dev->flags & IFF_UP)
2312 			schedule_delayed_work(&vi->refill, 0);
2313 	}
2314 
2315 	return 0;
2316 }
2317 
2318 static int virtnet_set_queues(struct virtnet_info *vi, u16 queue_pairs)
2319 {
2320 	int err;
2321 
2322 	rtnl_lock();
2323 	err = _virtnet_set_queues(vi, queue_pairs);
2324 	rtnl_unlock();
2325 	return err;
2326 }
2327 
2328 static int virtnet_close(struct net_device *dev)
2329 {
2330 	struct virtnet_info *vi = netdev_priv(dev);
2331 	int i;
2332 
2333 	/* Make sure NAPI doesn't schedule refill work */
2334 	disable_delayed_refill(vi);
2335 	/* Make sure refill_work doesn't re-enable napi! */
2336 	cancel_delayed_work_sync(&vi->refill);
2337 
2338 	for (i = 0; i < vi->max_queue_pairs; i++)
2339 		virtnet_disable_queue_pair(vi, i);
2340 
2341 	return 0;
2342 }
2343 
2344 static void virtnet_set_rx_mode(struct net_device *dev)
2345 {
2346 	struct virtnet_info *vi = netdev_priv(dev);
2347 	struct scatterlist sg[2];
2348 	struct virtio_net_ctrl_mac *mac_data;
2349 	struct netdev_hw_addr *ha;
2350 	int uc_count;
2351 	int mc_count;
2352 	void *buf;
2353 	int i;
2354 
2355 	/* We can't dynamically set ndo_set_rx_mode, so return gracefully */
2356 	if (!virtio_has_feature(vi->vdev, VIRTIO_NET_F_CTRL_RX))
2357 		return;
2358 
2359 	vi->ctrl->promisc = ((dev->flags & IFF_PROMISC) != 0);
2360 	vi->ctrl->allmulti = ((dev->flags & IFF_ALLMULTI) != 0);
2361 
2362 	sg_init_one(sg, &vi->ctrl->promisc, sizeof(vi->ctrl->promisc));
2363 
2364 	if (!virtnet_send_command(vi, VIRTIO_NET_CTRL_RX,
2365 				  VIRTIO_NET_CTRL_RX_PROMISC, sg))
2366 		dev_warn(&dev->dev, "Failed to %sable promisc mode.\n",
2367 			 vi->ctrl->promisc ? "en" : "dis");
2368 
2369 	sg_init_one(sg, &vi->ctrl->allmulti, sizeof(vi->ctrl->allmulti));
2370 
2371 	if (!virtnet_send_command(vi, VIRTIO_NET_CTRL_RX,
2372 				  VIRTIO_NET_CTRL_RX_ALLMULTI, sg))
2373 		dev_warn(&dev->dev, "Failed to %sable allmulti mode.\n",
2374 			 vi->ctrl->allmulti ? "en" : "dis");
2375 
2376 	uc_count = netdev_uc_count(dev);
2377 	mc_count = netdev_mc_count(dev);
2378 	/* MAC filter - use one buffer for both lists */
2379 	buf = kzalloc(((uc_count + mc_count) * ETH_ALEN) +
2380 		      (2 * sizeof(mac_data->entries)), GFP_ATOMIC);
2381 	mac_data = buf;
2382 	if (!buf)
2383 		return;
2384 
2385 	sg_init_table(sg, 2);
2386 
2387 	/* Store the unicast list and count in the front of the buffer */
2388 	mac_data->entries = cpu_to_virtio32(vi->vdev, uc_count);
2389 	i = 0;
2390 	netdev_for_each_uc_addr(ha, dev)
2391 		memcpy(&mac_data->macs[i++][0], ha->addr, ETH_ALEN);
2392 
2393 	sg_set_buf(&sg[0], mac_data,
2394 		   sizeof(mac_data->entries) + (uc_count * ETH_ALEN));
2395 
2396 	/* multicast list and count fill the end */
2397 	mac_data = (void *)&mac_data->macs[uc_count][0];
2398 
2399 	mac_data->entries = cpu_to_virtio32(vi->vdev, mc_count);
2400 	i = 0;
2401 	netdev_for_each_mc_addr(ha, dev)
2402 		memcpy(&mac_data->macs[i++][0], ha->addr, ETH_ALEN);
2403 
2404 	sg_set_buf(&sg[1], mac_data,
2405 		   sizeof(mac_data->entries) + (mc_count * ETH_ALEN));
2406 
2407 	if (!virtnet_send_command(vi, VIRTIO_NET_CTRL_MAC,
2408 				  VIRTIO_NET_CTRL_MAC_TABLE_SET, sg))
2409 		dev_warn(&dev->dev, "Failed to set MAC filter table.\n");
2410 
2411 	kfree(buf);
2412 }
2413 
2414 static int virtnet_vlan_rx_add_vid(struct net_device *dev,
2415 				   __be16 proto, u16 vid)
2416 {
2417 	struct virtnet_info *vi = netdev_priv(dev);
2418 	struct scatterlist sg;
2419 
2420 	vi->ctrl->vid = cpu_to_virtio16(vi->vdev, vid);
2421 	sg_init_one(&sg, &vi->ctrl->vid, sizeof(vi->ctrl->vid));
2422 
2423 	if (!virtnet_send_command(vi, VIRTIO_NET_CTRL_VLAN,
2424 				  VIRTIO_NET_CTRL_VLAN_ADD, &sg))
2425 		dev_warn(&dev->dev, "Failed to add VLAN ID %d.\n", vid);
2426 	return 0;
2427 }
2428 
2429 static int virtnet_vlan_rx_kill_vid(struct net_device *dev,
2430 				    __be16 proto, u16 vid)
2431 {
2432 	struct virtnet_info *vi = netdev_priv(dev);
2433 	struct scatterlist sg;
2434 
2435 	vi->ctrl->vid = cpu_to_virtio16(vi->vdev, vid);
2436 	sg_init_one(&sg, &vi->ctrl->vid, sizeof(vi->ctrl->vid));
2437 
2438 	if (!virtnet_send_command(vi, VIRTIO_NET_CTRL_VLAN,
2439 				  VIRTIO_NET_CTRL_VLAN_DEL, &sg))
2440 		dev_warn(&dev->dev, "Failed to kill VLAN ID %d.\n", vid);
2441 	return 0;
2442 }
2443 
2444 static void virtnet_clean_affinity(struct virtnet_info *vi)
2445 {
2446 	int i;
2447 
2448 	if (vi->affinity_hint_set) {
2449 		for (i = 0; i < vi->max_queue_pairs; i++) {
2450 			virtqueue_set_affinity(vi->rq[i].vq, NULL);
2451 			virtqueue_set_affinity(vi->sq[i].vq, NULL);
2452 		}
2453 
2454 		vi->affinity_hint_set = false;
2455 	}
2456 }
2457 
2458 static void virtnet_set_affinity(struct virtnet_info *vi)
2459 {
2460 	cpumask_var_t mask;
2461 	int stragglers;
2462 	int group_size;
2463 	int i, j, cpu;
2464 	int num_cpu;
2465 	int stride;
2466 
2467 	if (!zalloc_cpumask_var(&mask, GFP_KERNEL)) {
2468 		virtnet_clean_affinity(vi);
2469 		return;
2470 	}
2471 
2472 	num_cpu = num_online_cpus();
2473 	stride = max_t(int, num_cpu / vi->curr_queue_pairs, 1);
2474 	stragglers = num_cpu >= vi->curr_queue_pairs ?
2475 			num_cpu % vi->curr_queue_pairs :
2476 			0;
2477 	cpu = cpumask_first(cpu_online_mask);
2478 
2479 	for (i = 0; i < vi->curr_queue_pairs; i++) {
2480 		group_size = stride + (i < stragglers ? 1 : 0);
2481 
2482 		for (j = 0; j < group_size; j++) {
2483 			cpumask_set_cpu(cpu, mask);
2484 			cpu = cpumask_next_wrap(cpu, cpu_online_mask,
2485 						nr_cpu_ids, false);
2486 		}
2487 		virtqueue_set_affinity(vi->rq[i].vq, mask);
2488 		virtqueue_set_affinity(vi->sq[i].vq, mask);
2489 		__netif_set_xps_queue(vi->dev, cpumask_bits(mask), i, XPS_CPUS);
2490 		cpumask_clear(mask);
2491 	}
2492 
2493 	vi->affinity_hint_set = true;
2494 	free_cpumask_var(mask);
2495 }
2496 
2497 static int virtnet_cpu_online(unsigned int cpu, struct hlist_node *node)
2498 {
2499 	struct virtnet_info *vi = hlist_entry_safe(node, struct virtnet_info,
2500 						   node);
2501 	virtnet_set_affinity(vi);
2502 	return 0;
2503 }
2504 
2505 static int virtnet_cpu_dead(unsigned int cpu, struct hlist_node *node)
2506 {
2507 	struct virtnet_info *vi = hlist_entry_safe(node, struct virtnet_info,
2508 						   node_dead);
2509 	virtnet_set_affinity(vi);
2510 	return 0;
2511 }
2512 
2513 static int virtnet_cpu_down_prep(unsigned int cpu, struct hlist_node *node)
2514 {
2515 	struct virtnet_info *vi = hlist_entry_safe(node, struct virtnet_info,
2516 						   node);
2517 
2518 	virtnet_clean_affinity(vi);
2519 	return 0;
2520 }
2521 
2522 static enum cpuhp_state virtionet_online;
2523 
2524 static int virtnet_cpu_notif_add(struct virtnet_info *vi)
2525 {
2526 	int ret;
2527 
2528 	ret = cpuhp_state_add_instance_nocalls(virtionet_online, &vi->node);
2529 	if (ret)
2530 		return ret;
2531 	ret = cpuhp_state_add_instance_nocalls(CPUHP_VIRT_NET_DEAD,
2532 					       &vi->node_dead);
2533 	if (!ret)
2534 		return ret;
2535 	cpuhp_state_remove_instance_nocalls(virtionet_online, &vi->node);
2536 	return ret;
2537 }
2538 
2539 static void virtnet_cpu_notif_remove(struct virtnet_info *vi)
2540 {
2541 	cpuhp_state_remove_instance_nocalls(virtionet_online, &vi->node);
2542 	cpuhp_state_remove_instance_nocalls(CPUHP_VIRT_NET_DEAD,
2543 					    &vi->node_dead);
2544 }
2545 
2546 static void virtnet_get_ringparam(struct net_device *dev,
2547 				  struct ethtool_ringparam *ring,
2548 				  struct kernel_ethtool_ringparam *kernel_ring,
2549 				  struct netlink_ext_ack *extack)
2550 {
2551 	struct virtnet_info *vi = netdev_priv(dev);
2552 
2553 	ring->rx_max_pending = vi->rq[0].vq->num_max;
2554 	ring->tx_max_pending = vi->sq[0].vq->num_max;
2555 	ring->rx_pending = virtqueue_get_vring_size(vi->rq[0].vq);
2556 	ring->tx_pending = virtqueue_get_vring_size(vi->sq[0].vq);
2557 }
2558 
2559 static int virtnet_set_ringparam(struct net_device *dev,
2560 				 struct ethtool_ringparam *ring,
2561 				 struct kernel_ethtool_ringparam *kernel_ring,
2562 				 struct netlink_ext_ack *extack)
2563 {
2564 	struct virtnet_info *vi = netdev_priv(dev);
2565 	u32 rx_pending, tx_pending;
2566 	struct receive_queue *rq;
2567 	struct send_queue *sq;
2568 	int i, err;
2569 
2570 	if (ring->rx_mini_pending || ring->rx_jumbo_pending)
2571 		return -EINVAL;
2572 
2573 	rx_pending = virtqueue_get_vring_size(vi->rq[0].vq);
2574 	tx_pending = virtqueue_get_vring_size(vi->sq[0].vq);
2575 
2576 	if (ring->rx_pending == rx_pending &&
2577 	    ring->tx_pending == tx_pending)
2578 		return 0;
2579 
2580 	if (ring->rx_pending > vi->rq[0].vq->num_max)
2581 		return -EINVAL;
2582 
2583 	if (ring->tx_pending > vi->sq[0].vq->num_max)
2584 		return -EINVAL;
2585 
2586 	for (i = 0; i < vi->max_queue_pairs; i++) {
2587 		rq = vi->rq + i;
2588 		sq = vi->sq + i;
2589 
2590 		if (ring->tx_pending != tx_pending) {
2591 			err = virtnet_tx_resize(vi, sq, ring->tx_pending);
2592 			if (err)
2593 				return err;
2594 		}
2595 
2596 		if (ring->rx_pending != rx_pending) {
2597 			err = virtnet_rx_resize(vi, rq, ring->rx_pending);
2598 			if (err)
2599 				return err;
2600 		}
2601 	}
2602 
2603 	return 0;
2604 }
2605 
2606 static bool virtnet_commit_rss_command(struct virtnet_info *vi)
2607 {
2608 	struct net_device *dev = vi->dev;
2609 	struct scatterlist sgs[4];
2610 	unsigned int sg_buf_size;
2611 
2612 	/* prepare sgs */
2613 	sg_init_table(sgs, 4);
2614 
2615 	sg_buf_size = offsetof(struct virtio_net_ctrl_rss, indirection_table);
2616 	sg_set_buf(&sgs[0], &vi->ctrl->rss, sg_buf_size);
2617 
2618 	sg_buf_size = sizeof(uint16_t) * (vi->ctrl->rss.indirection_table_mask + 1);
2619 	sg_set_buf(&sgs[1], vi->ctrl->rss.indirection_table, sg_buf_size);
2620 
2621 	sg_buf_size = offsetof(struct virtio_net_ctrl_rss, key)
2622 			- offsetof(struct virtio_net_ctrl_rss, max_tx_vq);
2623 	sg_set_buf(&sgs[2], &vi->ctrl->rss.max_tx_vq, sg_buf_size);
2624 
2625 	sg_buf_size = vi->rss_key_size;
2626 	sg_set_buf(&sgs[3], vi->ctrl->rss.key, sg_buf_size);
2627 
2628 	if (!virtnet_send_command(vi, VIRTIO_NET_CTRL_MQ,
2629 				  vi->has_rss ? VIRTIO_NET_CTRL_MQ_RSS_CONFIG
2630 				  : VIRTIO_NET_CTRL_MQ_HASH_CONFIG, sgs)) {
2631 		dev_warn(&dev->dev, "VIRTIONET issue with committing RSS sgs\n");
2632 		return false;
2633 	}
2634 	return true;
2635 }
2636 
2637 static void virtnet_init_default_rss(struct virtnet_info *vi)
2638 {
2639 	u32 indir_val = 0;
2640 	int i = 0;
2641 
2642 	vi->ctrl->rss.hash_types = vi->rss_hash_types_supported;
2643 	vi->rss_hash_types_saved = vi->rss_hash_types_supported;
2644 	vi->ctrl->rss.indirection_table_mask = vi->rss_indir_table_size
2645 						? vi->rss_indir_table_size - 1 : 0;
2646 	vi->ctrl->rss.unclassified_queue = 0;
2647 
2648 	for (; i < vi->rss_indir_table_size; ++i) {
2649 		indir_val = ethtool_rxfh_indir_default(i, vi->curr_queue_pairs);
2650 		vi->ctrl->rss.indirection_table[i] = indir_val;
2651 	}
2652 
2653 	vi->ctrl->rss.max_tx_vq = vi->curr_queue_pairs;
2654 	vi->ctrl->rss.hash_key_length = vi->rss_key_size;
2655 
2656 	netdev_rss_key_fill(vi->ctrl->rss.key, vi->rss_key_size);
2657 }
2658 
2659 static void virtnet_get_hashflow(const struct virtnet_info *vi, struct ethtool_rxnfc *info)
2660 {
2661 	info->data = 0;
2662 	switch (info->flow_type) {
2663 	case TCP_V4_FLOW:
2664 		if (vi->rss_hash_types_saved & VIRTIO_NET_RSS_HASH_TYPE_TCPv4) {
2665 			info->data = RXH_IP_SRC | RXH_IP_DST |
2666 						 RXH_L4_B_0_1 | RXH_L4_B_2_3;
2667 		} else if (vi->rss_hash_types_saved & VIRTIO_NET_RSS_HASH_TYPE_IPv4) {
2668 			info->data = RXH_IP_SRC | RXH_IP_DST;
2669 		}
2670 		break;
2671 	case TCP_V6_FLOW:
2672 		if (vi->rss_hash_types_saved & VIRTIO_NET_RSS_HASH_TYPE_TCPv6) {
2673 			info->data = RXH_IP_SRC | RXH_IP_DST |
2674 						 RXH_L4_B_0_1 | RXH_L4_B_2_3;
2675 		} else if (vi->rss_hash_types_saved & VIRTIO_NET_RSS_HASH_TYPE_IPv6) {
2676 			info->data = RXH_IP_SRC | RXH_IP_DST;
2677 		}
2678 		break;
2679 	case UDP_V4_FLOW:
2680 		if (vi->rss_hash_types_saved & VIRTIO_NET_RSS_HASH_TYPE_UDPv4) {
2681 			info->data = RXH_IP_SRC | RXH_IP_DST |
2682 						 RXH_L4_B_0_1 | RXH_L4_B_2_3;
2683 		} else if (vi->rss_hash_types_saved & VIRTIO_NET_RSS_HASH_TYPE_IPv4) {
2684 			info->data = RXH_IP_SRC | RXH_IP_DST;
2685 		}
2686 		break;
2687 	case UDP_V6_FLOW:
2688 		if (vi->rss_hash_types_saved & VIRTIO_NET_RSS_HASH_TYPE_UDPv6) {
2689 			info->data = RXH_IP_SRC | RXH_IP_DST |
2690 						 RXH_L4_B_0_1 | RXH_L4_B_2_3;
2691 		} else if (vi->rss_hash_types_saved & VIRTIO_NET_RSS_HASH_TYPE_IPv6) {
2692 			info->data = RXH_IP_SRC | RXH_IP_DST;
2693 		}
2694 		break;
2695 	case IPV4_FLOW:
2696 		if (vi->rss_hash_types_saved & VIRTIO_NET_RSS_HASH_TYPE_IPv4)
2697 			info->data = RXH_IP_SRC | RXH_IP_DST;
2698 
2699 		break;
2700 	case IPV6_FLOW:
2701 		if (vi->rss_hash_types_saved & VIRTIO_NET_RSS_HASH_TYPE_IPv6)
2702 			info->data = RXH_IP_SRC | RXH_IP_DST;
2703 
2704 		break;
2705 	default:
2706 		info->data = 0;
2707 		break;
2708 	}
2709 }
2710 
2711 static bool virtnet_set_hashflow(struct virtnet_info *vi, struct ethtool_rxnfc *info)
2712 {
2713 	u32 new_hashtypes = vi->rss_hash_types_saved;
2714 	bool is_disable = info->data & RXH_DISCARD;
2715 	bool is_l4 = info->data == (RXH_IP_SRC | RXH_IP_DST | RXH_L4_B_0_1 | RXH_L4_B_2_3);
2716 
2717 	/* supports only 'sd', 'sdfn' and 'r' */
2718 	if (!((info->data == (RXH_IP_SRC | RXH_IP_DST)) | is_l4 | is_disable))
2719 		return false;
2720 
2721 	switch (info->flow_type) {
2722 	case TCP_V4_FLOW:
2723 		new_hashtypes &= ~(VIRTIO_NET_RSS_HASH_TYPE_IPv4 | VIRTIO_NET_RSS_HASH_TYPE_TCPv4);
2724 		if (!is_disable)
2725 			new_hashtypes |= VIRTIO_NET_RSS_HASH_TYPE_IPv4
2726 				| (is_l4 ? VIRTIO_NET_RSS_HASH_TYPE_TCPv4 : 0);
2727 		break;
2728 	case UDP_V4_FLOW:
2729 		new_hashtypes &= ~(VIRTIO_NET_RSS_HASH_TYPE_IPv4 | VIRTIO_NET_RSS_HASH_TYPE_UDPv4);
2730 		if (!is_disable)
2731 			new_hashtypes |= VIRTIO_NET_RSS_HASH_TYPE_IPv4
2732 				| (is_l4 ? VIRTIO_NET_RSS_HASH_TYPE_UDPv4 : 0);
2733 		break;
2734 	case IPV4_FLOW:
2735 		new_hashtypes &= ~VIRTIO_NET_RSS_HASH_TYPE_IPv4;
2736 		if (!is_disable)
2737 			new_hashtypes = VIRTIO_NET_RSS_HASH_TYPE_IPv4;
2738 		break;
2739 	case TCP_V6_FLOW:
2740 		new_hashtypes &= ~(VIRTIO_NET_RSS_HASH_TYPE_IPv6 | VIRTIO_NET_RSS_HASH_TYPE_TCPv6);
2741 		if (!is_disable)
2742 			new_hashtypes |= VIRTIO_NET_RSS_HASH_TYPE_IPv6
2743 				| (is_l4 ? VIRTIO_NET_RSS_HASH_TYPE_TCPv6 : 0);
2744 		break;
2745 	case UDP_V6_FLOW:
2746 		new_hashtypes &= ~(VIRTIO_NET_RSS_HASH_TYPE_IPv6 | VIRTIO_NET_RSS_HASH_TYPE_UDPv6);
2747 		if (!is_disable)
2748 			new_hashtypes |= VIRTIO_NET_RSS_HASH_TYPE_IPv6
2749 				| (is_l4 ? VIRTIO_NET_RSS_HASH_TYPE_UDPv6 : 0);
2750 		break;
2751 	case IPV6_FLOW:
2752 		new_hashtypes &= ~VIRTIO_NET_RSS_HASH_TYPE_IPv6;
2753 		if (!is_disable)
2754 			new_hashtypes = VIRTIO_NET_RSS_HASH_TYPE_IPv6;
2755 		break;
2756 	default:
2757 		/* unsupported flow */
2758 		return false;
2759 	}
2760 
2761 	/* if unsupported hashtype was set */
2762 	if (new_hashtypes != (new_hashtypes & vi->rss_hash_types_supported))
2763 		return false;
2764 
2765 	if (new_hashtypes != vi->rss_hash_types_saved) {
2766 		vi->rss_hash_types_saved = new_hashtypes;
2767 		vi->ctrl->rss.hash_types = vi->rss_hash_types_saved;
2768 		if (vi->dev->features & NETIF_F_RXHASH)
2769 			return virtnet_commit_rss_command(vi);
2770 	}
2771 
2772 	return true;
2773 }
2774 
2775 static void virtnet_get_drvinfo(struct net_device *dev,
2776 				struct ethtool_drvinfo *info)
2777 {
2778 	struct virtnet_info *vi = netdev_priv(dev);
2779 	struct virtio_device *vdev = vi->vdev;
2780 
2781 	strscpy(info->driver, KBUILD_MODNAME, sizeof(info->driver));
2782 	strscpy(info->version, VIRTNET_DRIVER_VERSION, sizeof(info->version));
2783 	strscpy(info->bus_info, virtio_bus_name(vdev), sizeof(info->bus_info));
2784 
2785 }
2786 
2787 /* TODO: Eliminate OOO packets during switching */
2788 static int virtnet_set_channels(struct net_device *dev,
2789 				struct ethtool_channels *channels)
2790 {
2791 	struct virtnet_info *vi = netdev_priv(dev);
2792 	u16 queue_pairs = channels->combined_count;
2793 	int err;
2794 
2795 	/* We don't support separate rx/tx channels.
2796 	 * We don't allow setting 'other' channels.
2797 	 */
2798 	if (channels->rx_count || channels->tx_count || channels->other_count)
2799 		return -EINVAL;
2800 
2801 	if (queue_pairs > vi->max_queue_pairs || queue_pairs == 0)
2802 		return -EINVAL;
2803 
2804 	/* For now we don't support modifying channels while XDP is loaded
2805 	 * also when XDP is loaded all RX queues have XDP programs so we only
2806 	 * need to check a single RX queue.
2807 	 */
2808 	if (vi->rq[0].xdp_prog)
2809 		return -EINVAL;
2810 
2811 	cpus_read_lock();
2812 	err = _virtnet_set_queues(vi, queue_pairs);
2813 	if (err) {
2814 		cpus_read_unlock();
2815 		goto err;
2816 	}
2817 	virtnet_set_affinity(vi);
2818 	cpus_read_unlock();
2819 
2820 	netif_set_real_num_tx_queues(dev, queue_pairs);
2821 	netif_set_real_num_rx_queues(dev, queue_pairs);
2822  err:
2823 	return err;
2824 }
2825 
2826 static void virtnet_get_strings(struct net_device *dev, u32 stringset, u8 *data)
2827 {
2828 	struct virtnet_info *vi = netdev_priv(dev);
2829 	unsigned int i, j;
2830 	u8 *p = data;
2831 
2832 	switch (stringset) {
2833 	case ETH_SS_STATS:
2834 		for (i = 0; i < vi->curr_queue_pairs; i++) {
2835 			for (j = 0; j < VIRTNET_RQ_STATS_LEN; j++)
2836 				ethtool_sprintf(&p, "rx_queue_%u_%s", i,
2837 						virtnet_rq_stats_desc[j].desc);
2838 		}
2839 
2840 		for (i = 0; i < vi->curr_queue_pairs; i++) {
2841 			for (j = 0; j < VIRTNET_SQ_STATS_LEN; j++)
2842 				ethtool_sprintf(&p, "tx_queue_%u_%s", i,
2843 						virtnet_sq_stats_desc[j].desc);
2844 		}
2845 		break;
2846 	}
2847 }
2848 
2849 static int virtnet_get_sset_count(struct net_device *dev, int sset)
2850 {
2851 	struct virtnet_info *vi = netdev_priv(dev);
2852 
2853 	switch (sset) {
2854 	case ETH_SS_STATS:
2855 		return vi->curr_queue_pairs * (VIRTNET_RQ_STATS_LEN +
2856 					       VIRTNET_SQ_STATS_LEN);
2857 	default:
2858 		return -EOPNOTSUPP;
2859 	}
2860 }
2861 
2862 static void virtnet_get_ethtool_stats(struct net_device *dev,
2863 				      struct ethtool_stats *stats, u64 *data)
2864 {
2865 	struct virtnet_info *vi = netdev_priv(dev);
2866 	unsigned int idx = 0, start, i, j;
2867 	const u8 *stats_base;
2868 	size_t offset;
2869 
2870 	for (i = 0; i < vi->curr_queue_pairs; i++) {
2871 		struct receive_queue *rq = &vi->rq[i];
2872 
2873 		stats_base = (u8 *)&rq->stats;
2874 		do {
2875 			start = u64_stats_fetch_begin(&rq->stats.syncp);
2876 			for (j = 0; j < VIRTNET_RQ_STATS_LEN; j++) {
2877 				offset = virtnet_rq_stats_desc[j].offset;
2878 				data[idx + j] = *(u64 *)(stats_base + offset);
2879 			}
2880 		} while (u64_stats_fetch_retry(&rq->stats.syncp, start));
2881 		idx += VIRTNET_RQ_STATS_LEN;
2882 	}
2883 
2884 	for (i = 0; i < vi->curr_queue_pairs; i++) {
2885 		struct send_queue *sq = &vi->sq[i];
2886 
2887 		stats_base = (u8 *)&sq->stats;
2888 		do {
2889 			start = u64_stats_fetch_begin(&sq->stats.syncp);
2890 			for (j = 0; j < VIRTNET_SQ_STATS_LEN; j++) {
2891 				offset = virtnet_sq_stats_desc[j].offset;
2892 				data[idx + j] = *(u64 *)(stats_base + offset);
2893 			}
2894 		} while (u64_stats_fetch_retry(&sq->stats.syncp, start));
2895 		idx += VIRTNET_SQ_STATS_LEN;
2896 	}
2897 }
2898 
2899 static void virtnet_get_channels(struct net_device *dev,
2900 				 struct ethtool_channels *channels)
2901 {
2902 	struct virtnet_info *vi = netdev_priv(dev);
2903 
2904 	channels->combined_count = vi->curr_queue_pairs;
2905 	channels->max_combined = vi->max_queue_pairs;
2906 	channels->max_other = 0;
2907 	channels->rx_count = 0;
2908 	channels->tx_count = 0;
2909 	channels->other_count = 0;
2910 }
2911 
2912 static int virtnet_set_link_ksettings(struct net_device *dev,
2913 				      const struct ethtool_link_ksettings *cmd)
2914 {
2915 	struct virtnet_info *vi = netdev_priv(dev);
2916 
2917 	return ethtool_virtdev_set_link_ksettings(dev, cmd,
2918 						  &vi->speed, &vi->duplex);
2919 }
2920 
2921 static int virtnet_get_link_ksettings(struct net_device *dev,
2922 				      struct ethtool_link_ksettings *cmd)
2923 {
2924 	struct virtnet_info *vi = netdev_priv(dev);
2925 
2926 	cmd->base.speed = vi->speed;
2927 	cmd->base.duplex = vi->duplex;
2928 	cmd->base.port = PORT_OTHER;
2929 
2930 	return 0;
2931 }
2932 
2933 static int virtnet_send_notf_coal_cmds(struct virtnet_info *vi,
2934 				       struct ethtool_coalesce *ec)
2935 {
2936 	struct scatterlist sgs_tx, sgs_rx;
2937 	struct virtio_net_ctrl_coal_tx coal_tx;
2938 	struct virtio_net_ctrl_coal_rx coal_rx;
2939 
2940 	coal_tx.tx_usecs = cpu_to_le32(ec->tx_coalesce_usecs);
2941 	coal_tx.tx_max_packets = cpu_to_le32(ec->tx_max_coalesced_frames);
2942 	sg_init_one(&sgs_tx, &coal_tx, sizeof(coal_tx));
2943 
2944 	if (!virtnet_send_command(vi, VIRTIO_NET_CTRL_NOTF_COAL,
2945 				  VIRTIO_NET_CTRL_NOTF_COAL_TX_SET,
2946 				  &sgs_tx))
2947 		return -EINVAL;
2948 
2949 	/* Save parameters */
2950 	vi->tx_usecs = ec->tx_coalesce_usecs;
2951 	vi->tx_max_packets = ec->tx_max_coalesced_frames;
2952 
2953 	coal_rx.rx_usecs = cpu_to_le32(ec->rx_coalesce_usecs);
2954 	coal_rx.rx_max_packets = cpu_to_le32(ec->rx_max_coalesced_frames);
2955 	sg_init_one(&sgs_rx, &coal_rx, sizeof(coal_rx));
2956 
2957 	if (!virtnet_send_command(vi, VIRTIO_NET_CTRL_NOTF_COAL,
2958 				  VIRTIO_NET_CTRL_NOTF_COAL_RX_SET,
2959 				  &sgs_rx))
2960 		return -EINVAL;
2961 
2962 	/* Save parameters */
2963 	vi->rx_usecs = ec->rx_coalesce_usecs;
2964 	vi->rx_max_packets = ec->rx_max_coalesced_frames;
2965 
2966 	return 0;
2967 }
2968 
2969 static int virtnet_coal_params_supported(struct ethtool_coalesce *ec)
2970 {
2971 	/* usecs coalescing is supported only if VIRTIO_NET_F_NOTF_COAL
2972 	 * feature is negotiated.
2973 	 */
2974 	if (ec->rx_coalesce_usecs || ec->tx_coalesce_usecs)
2975 		return -EOPNOTSUPP;
2976 
2977 	if (ec->tx_max_coalesced_frames > 1 ||
2978 	    ec->rx_max_coalesced_frames != 1)
2979 		return -EINVAL;
2980 
2981 	return 0;
2982 }
2983 
2984 static int virtnet_set_coalesce(struct net_device *dev,
2985 				struct ethtool_coalesce *ec,
2986 				struct kernel_ethtool_coalesce *kernel_coal,
2987 				struct netlink_ext_ack *extack)
2988 {
2989 	struct virtnet_info *vi = netdev_priv(dev);
2990 	int ret, i, napi_weight;
2991 	bool update_napi = false;
2992 
2993 	/* Can't change NAPI weight if the link is up */
2994 	napi_weight = ec->tx_max_coalesced_frames ? NAPI_POLL_WEIGHT : 0;
2995 	if (napi_weight ^ vi->sq[0].napi.weight) {
2996 		if (dev->flags & IFF_UP)
2997 			return -EBUSY;
2998 		else
2999 			update_napi = true;
3000 	}
3001 
3002 	if (virtio_has_feature(vi->vdev, VIRTIO_NET_F_NOTF_COAL))
3003 		ret = virtnet_send_notf_coal_cmds(vi, ec);
3004 	else
3005 		ret = virtnet_coal_params_supported(ec);
3006 
3007 	if (ret)
3008 		return ret;
3009 
3010 	if (update_napi) {
3011 		for (i = 0; i < vi->max_queue_pairs; i++)
3012 			vi->sq[i].napi.weight = napi_weight;
3013 	}
3014 
3015 	return ret;
3016 }
3017 
3018 static int virtnet_get_coalesce(struct net_device *dev,
3019 				struct ethtool_coalesce *ec,
3020 				struct kernel_ethtool_coalesce *kernel_coal,
3021 				struct netlink_ext_ack *extack)
3022 {
3023 	struct virtnet_info *vi = netdev_priv(dev);
3024 
3025 	if (virtio_has_feature(vi->vdev, VIRTIO_NET_F_NOTF_COAL)) {
3026 		ec->rx_coalesce_usecs = vi->rx_usecs;
3027 		ec->tx_coalesce_usecs = vi->tx_usecs;
3028 		ec->tx_max_coalesced_frames = vi->tx_max_packets;
3029 		ec->rx_max_coalesced_frames = vi->rx_max_packets;
3030 	} else {
3031 		ec->rx_max_coalesced_frames = 1;
3032 
3033 		if (vi->sq[0].napi.weight)
3034 			ec->tx_max_coalesced_frames = 1;
3035 	}
3036 
3037 	return 0;
3038 }
3039 
3040 static void virtnet_init_settings(struct net_device *dev)
3041 {
3042 	struct virtnet_info *vi = netdev_priv(dev);
3043 
3044 	vi->speed = SPEED_UNKNOWN;
3045 	vi->duplex = DUPLEX_UNKNOWN;
3046 }
3047 
3048 static void virtnet_update_settings(struct virtnet_info *vi)
3049 {
3050 	u32 speed;
3051 	u8 duplex;
3052 
3053 	if (!virtio_has_feature(vi->vdev, VIRTIO_NET_F_SPEED_DUPLEX))
3054 		return;
3055 
3056 	virtio_cread_le(vi->vdev, struct virtio_net_config, speed, &speed);
3057 
3058 	if (ethtool_validate_speed(speed))
3059 		vi->speed = speed;
3060 
3061 	virtio_cread_le(vi->vdev, struct virtio_net_config, duplex, &duplex);
3062 
3063 	if (ethtool_validate_duplex(duplex))
3064 		vi->duplex = duplex;
3065 }
3066 
3067 static u32 virtnet_get_rxfh_key_size(struct net_device *dev)
3068 {
3069 	return ((struct virtnet_info *)netdev_priv(dev))->rss_key_size;
3070 }
3071 
3072 static u32 virtnet_get_rxfh_indir_size(struct net_device *dev)
3073 {
3074 	return ((struct virtnet_info *)netdev_priv(dev))->rss_indir_table_size;
3075 }
3076 
3077 static int virtnet_get_rxfh(struct net_device *dev, u32 *indir, u8 *key, u8 *hfunc)
3078 {
3079 	struct virtnet_info *vi = netdev_priv(dev);
3080 	int i;
3081 
3082 	if (indir) {
3083 		for (i = 0; i < vi->rss_indir_table_size; ++i)
3084 			indir[i] = vi->ctrl->rss.indirection_table[i];
3085 	}
3086 
3087 	if (key)
3088 		memcpy(key, vi->ctrl->rss.key, vi->rss_key_size);
3089 
3090 	if (hfunc)
3091 		*hfunc = ETH_RSS_HASH_TOP;
3092 
3093 	return 0;
3094 }
3095 
3096 static int virtnet_set_rxfh(struct net_device *dev, const u32 *indir, const u8 *key, const u8 hfunc)
3097 {
3098 	struct virtnet_info *vi = netdev_priv(dev);
3099 	int i;
3100 
3101 	if (hfunc != ETH_RSS_HASH_NO_CHANGE && hfunc != ETH_RSS_HASH_TOP)
3102 		return -EOPNOTSUPP;
3103 
3104 	if (indir) {
3105 		for (i = 0; i < vi->rss_indir_table_size; ++i)
3106 			vi->ctrl->rss.indirection_table[i] = indir[i];
3107 	}
3108 	if (key)
3109 		memcpy(vi->ctrl->rss.key, key, vi->rss_key_size);
3110 
3111 	virtnet_commit_rss_command(vi);
3112 
3113 	return 0;
3114 }
3115 
3116 static int virtnet_get_rxnfc(struct net_device *dev, struct ethtool_rxnfc *info, u32 *rule_locs)
3117 {
3118 	struct virtnet_info *vi = netdev_priv(dev);
3119 	int rc = 0;
3120 
3121 	switch (info->cmd) {
3122 	case ETHTOOL_GRXRINGS:
3123 		info->data = vi->curr_queue_pairs;
3124 		break;
3125 	case ETHTOOL_GRXFH:
3126 		virtnet_get_hashflow(vi, info);
3127 		break;
3128 	default:
3129 		rc = -EOPNOTSUPP;
3130 	}
3131 
3132 	return rc;
3133 }
3134 
3135 static int virtnet_set_rxnfc(struct net_device *dev, struct ethtool_rxnfc *info)
3136 {
3137 	struct virtnet_info *vi = netdev_priv(dev);
3138 	int rc = 0;
3139 
3140 	switch (info->cmd) {
3141 	case ETHTOOL_SRXFH:
3142 		if (!virtnet_set_hashflow(vi, info))
3143 			rc = -EINVAL;
3144 
3145 		break;
3146 	default:
3147 		rc = -EOPNOTSUPP;
3148 	}
3149 
3150 	return rc;
3151 }
3152 
3153 static const struct ethtool_ops virtnet_ethtool_ops = {
3154 	.supported_coalesce_params = ETHTOOL_COALESCE_MAX_FRAMES |
3155 		ETHTOOL_COALESCE_USECS,
3156 	.get_drvinfo = virtnet_get_drvinfo,
3157 	.get_link = ethtool_op_get_link,
3158 	.get_ringparam = virtnet_get_ringparam,
3159 	.set_ringparam = virtnet_set_ringparam,
3160 	.get_strings = virtnet_get_strings,
3161 	.get_sset_count = virtnet_get_sset_count,
3162 	.get_ethtool_stats = virtnet_get_ethtool_stats,
3163 	.set_channels = virtnet_set_channels,
3164 	.get_channels = virtnet_get_channels,
3165 	.get_ts_info = ethtool_op_get_ts_info,
3166 	.get_link_ksettings = virtnet_get_link_ksettings,
3167 	.set_link_ksettings = virtnet_set_link_ksettings,
3168 	.set_coalesce = virtnet_set_coalesce,
3169 	.get_coalesce = virtnet_get_coalesce,
3170 	.get_rxfh_key_size = virtnet_get_rxfh_key_size,
3171 	.get_rxfh_indir_size = virtnet_get_rxfh_indir_size,
3172 	.get_rxfh = virtnet_get_rxfh,
3173 	.set_rxfh = virtnet_set_rxfh,
3174 	.get_rxnfc = virtnet_get_rxnfc,
3175 	.set_rxnfc = virtnet_set_rxnfc,
3176 };
3177 
3178 static void virtnet_freeze_down(struct virtio_device *vdev)
3179 {
3180 	struct virtnet_info *vi = vdev->priv;
3181 
3182 	/* Make sure no work handler is accessing the device */
3183 	flush_work(&vi->config_work);
3184 
3185 	netif_tx_lock_bh(vi->dev);
3186 	netif_device_detach(vi->dev);
3187 	netif_tx_unlock_bh(vi->dev);
3188 	if (netif_running(vi->dev))
3189 		virtnet_close(vi->dev);
3190 }
3191 
3192 static int init_vqs(struct virtnet_info *vi);
3193 
3194 static int virtnet_restore_up(struct virtio_device *vdev)
3195 {
3196 	struct virtnet_info *vi = vdev->priv;
3197 	int err;
3198 
3199 	err = init_vqs(vi);
3200 	if (err)
3201 		return err;
3202 
3203 	virtio_device_ready(vdev);
3204 
3205 	enable_delayed_refill(vi);
3206 
3207 	if (netif_running(vi->dev)) {
3208 		err = virtnet_open(vi->dev);
3209 		if (err)
3210 			return err;
3211 	}
3212 
3213 	netif_tx_lock_bh(vi->dev);
3214 	netif_device_attach(vi->dev);
3215 	netif_tx_unlock_bh(vi->dev);
3216 	return err;
3217 }
3218 
3219 static int virtnet_set_guest_offloads(struct virtnet_info *vi, u64 offloads)
3220 {
3221 	struct scatterlist sg;
3222 	vi->ctrl->offloads = cpu_to_virtio64(vi->vdev, offloads);
3223 
3224 	sg_init_one(&sg, &vi->ctrl->offloads, sizeof(vi->ctrl->offloads));
3225 
3226 	if (!virtnet_send_command(vi, VIRTIO_NET_CTRL_GUEST_OFFLOADS,
3227 				  VIRTIO_NET_CTRL_GUEST_OFFLOADS_SET, &sg)) {
3228 		dev_warn(&vi->dev->dev, "Fail to set guest offload.\n");
3229 		return -EINVAL;
3230 	}
3231 
3232 	return 0;
3233 }
3234 
3235 static int virtnet_clear_guest_offloads(struct virtnet_info *vi)
3236 {
3237 	u64 offloads = 0;
3238 
3239 	if (!vi->guest_offloads)
3240 		return 0;
3241 
3242 	return virtnet_set_guest_offloads(vi, offloads);
3243 }
3244 
3245 static int virtnet_restore_guest_offloads(struct virtnet_info *vi)
3246 {
3247 	u64 offloads = vi->guest_offloads;
3248 
3249 	if (!vi->guest_offloads)
3250 		return 0;
3251 
3252 	return virtnet_set_guest_offloads(vi, offloads);
3253 }
3254 
3255 static int virtnet_xdp_set(struct net_device *dev, struct bpf_prog *prog,
3256 			   struct netlink_ext_ack *extack)
3257 {
3258 	unsigned int room = SKB_DATA_ALIGN(VIRTIO_XDP_HEADROOM +
3259 					   sizeof(struct skb_shared_info));
3260 	unsigned int max_sz = PAGE_SIZE - room - ETH_HLEN;
3261 	struct virtnet_info *vi = netdev_priv(dev);
3262 	struct bpf_prog *old_prog;
3263 	u16 xdp_qp = 0, curr_qp;
3264 	int i, err;
3265 
3266 	if (!virtio_has_feature(vi->vdev, VIRTIO_NET_F_CTRL_GUEST_OFFLOADS)
3267 	    && (virtio_has_feature(vi->vdev, VIRTIO_NET_F_GUEST_TSO4) ||
3268 	        virtio_has_feature(vi->vdev, VIRTIO_NET_F_GUEST_TSO6) ||
3269 	        virtio_has_feature(vi->vdev, VIRTIO_NET_F_GUEST_ECN) ||
3270 		virtio_has_feature(vi->vdev, VIRTIO_NET_F_GUEST_UFO) ||
3271 		virtio_has_feature(vi->vdev, VIRTIO_NET_F_GUEST_CSUM) ||
3272 		virtio_has_feature(vi->vdev, VIRTIO_NET_F_GUEST_USO4) ||
3273 		virtio_has_feature(vi->vdev, VIRTIO_NET_F_GUEST_USO6))) {
3274 		NL_SET_ERR_MSG_MOD(extack, "Can't set XDP while host is implementing GRO_HW/CSUM, disable GRO_HW/CSUM first");
3275 		return -EOPNOTSUPP;
3276 	}
3277 
3278 	if (vi->mergeable_rx_bufs && !vi->any_header_sg) {
3279 		NL_SET_ERR_MSG_MOD(extack, "XDP expects header/data in single page, any_header_sg required");
3280 		return -EINVAL;
3281 	}
3282 
3283 	if (prog && !prog->aux->xdp_has_frags && dev->mtu > max_sz) {
3284 		NL_SET_ERR_MSG_MOD(extack, "MTU too large to enable XDP without frags");
3285 		netdev_warn(dev, "single-buffer XDP requires MTU less than %u\n", max_sz);
3286 		return -EINVAL;
3287 	}
3288 
3289 	curr_qp = vi->curr_queue_pairs - vi->xdp_queue_pairs;
3290 	if (prog)
3291 		xdp_qp = nr_cpu_ids;
3292 
3293 	/* XDP requires extra queues for XDP_TX */
3294 	if (curr_qp + xdp_qp > vi->max_queue_pairs) {
3295 		netdev_warn_once(dev, "XDP request %i queues but max is %i. XDP_TX and XDP_REDIRECT will operate in a slower locked tx mode.\n",
3296 				 curr_qp + xdp_qp, vi->max_queue_pairs);
3297 		xdp_qp = 0;
3298 	}
3299 
3300 	old_prog = rtnl_dereference(vi->rq[0].xdp_prog);
3301 	if (!prog && !old_prog)
3302 		return 0;
3303 
3304 	if (prog)
3305 		bpf_prog_add(prog, vi->max_queue_pairs - 1);
3306 
3307 	/* Make sure NAPI is not using any XDP TX queues for RX. */
3308 	if (netif_running(dev)) {
3309 		for (i = 0; i < vi->max_queue_pairs; i++) {
3310 			napi_disable(&vi->rq[i].napi);
3311 			virtnet_napi_tx_disable(&vi->sq[i].napi);
3312 		}
3313 	}
3314 
3315 	if (!prog) {
3316 		for (i = 0; i < vi->max_queue_pairs; i++) {
3317 			rcu_assign_pointer(vi->rq[i].xdp_prog, prog);
3318 			if (i == 0)
3319 				virtnet_restore_guest_offloads(vi);
3320 		}
3321 		synchronize_net();
3322 	}
3323 
3324 	err = _virtnet_set_queues(vi, curr_qp + xdp_qp);
3325 	if (err)
3326 		goto err;
3327 	netif_set_real_num_rx_queues(dev, curr_qp + xdp_qp);
3328 	vi->xdp_queue_pairs = xdp_qp;
3329 
3330 	if (prog) {
3331 		vi->xdp_enabled = true;
3332 		for (i = 0; i < vi->max_queue_pairs; i++) {
3333 			rcu_assign_pointer(vi->rq[i].xdp_prog, prog);
3334 			if (i == 0 && !old_prog)
3335 				virtnet_clear_guest_offloads(vi);
3336 		}
3337 		if (!old_prog)
3338 			xdp_features_set_redirect_target(dev, true);
3339 	} else {
3340 		xdp_features_clear_redirect_target(dev);
3341 		vi->xdp_enabled = false;
3342 	}
3343 
3344 	for (i = 0; i < vi->max_queue_pairs; i++) {
3345 		if (old_prog)
3346 			bpf_prog_put(old_prog);
3347 		if (netif_running(dev)) {
3348 			virtnet_napi_enable(vi->rq[i].vq, &vi->rq[i].napi);
3349 			virtnet_napi_tx_enable(vi, vi->sq[i].vq,
3350 					       &vi->sq[i].napi);
3351 		}
3352 	}
3353 
3354 	return 0;
3355 
3356 err:
3357 	if (!prog) {
3358 		virtnet_clear_guest_offloads(vi);
3359 		for (i = 0; i < vi->max_queue_pairs; i++)
3360 			rcu_assign_pointer(vi->rq[i].xdp_prog, old_prog);
3361 	}
3362 
3363 	if (netif_running(dev)) {
3364 		for (i = 0; i < vi->max_queue_pairs; i++) {
3365 			virtnet_napi_enable(vi->rq[i].vq, &vi->rq[i].napi);
3366 			virtnet_napi_tx_enable(vi, vi->sq[i].vq,
3367 					       &vi->sq[i].napi);
3368 		}
3369 	}
3370 	if (prog)
3371 		bpf_prog_sub(prog, vi->max_queue_pairs - 1);
3372 	return err;
3373 }
3374 
3375 static int virtnet_xdp(struct net_device *dev, struct netdev_bpf *xdp)
3376 {
3377 	switch (xdp->command) {
3378 	case XDP_SETUP_PROG:
3379 		return virtnet_xdp_set(dev, xdp->prog, xdp->extack);
3380 	default:
3381 		return -EINVAL;
3382 	}
3383 }
3384 
3385 static int virtnet_get_phys_port_name(struct net_device *dev, char *buf,
3386 				      size_t len)
3387 {
3388 	struct virtnet_info *vi = netdev_priv(dev);
3389 	int ret;
3390 
3391 	if (!virtio_has_feature(vi->vdev, VIRTIO_NET_F_STANDBY))
3392 		return -EOPNOTSUPP;
3393 
3394 	ret = snprintf(buf, len, "sby");
3395 	if (ret >= len)
3396 		return -EOPNOTSUPP;
3397 
3398 	return 0;
3399 }
3400 
3401 static int virtnet_set_features(struct net_device *dev,
3402 				netdev_features_t features)
3403 {
3404 	struct virtnet_info *vi = netdev_priv(dev);
3405 	u64 offloads;
3406 	int err;
3407 
3408 	if ((dev->features ^ features) & NETIF_F_GRO_HW) {
3409 		if (vi->xdp_enabled)
3410 			return -EBUSY;
3411 
3412 		if (features & NETIF_F_GRO_HW)
3413 			offloads = vi->guest_offloads_capable;
3414 		else
3415 			offloads = vi->guest_offloads_capable &
3416 				   ~GUEST_OFFLOAD_GRO_HW_MASK;
3417 
3418 		err = virtnet_set_guest_offloads(vi, offloads);
3419 		if (err)
3420 			return err;
3421 		vi->guest_offloads = offloads;
3422 	}
3423 
3424 	if ((dev->features ^ features) & NETIF_F_RXHASH) {
3425 		if (features & NETIF_F_RXHASH)
3426 			vi->ctrl->rss.hash_types = vi->rss_hash_types_saved;
3427 		else
3428 			vi->ctrl->rss.hash_types = VIRTIO_NET_HASH_REPORT_NONE;
3429 
3430 		if (!virtnet_commit_rss_command(vi))
3431 			return -EINVAL;
3432 	}
3433 
3434 	return 0;
3435 }
3436 
3437 static void virtnet_tx_timeout(struct net_device *dev, unsigned int txqueue)
3438 {
3439 	struct virtnet_info *priv = netdev_priv(dev);
3440 	struct send_queue *sq = &priv->sq[txqueue];
3441 	struct netdev_queue *txq = netdev_get_tx_queue(dev, txqueue);
3442 
3443 	u64_stats_update_begin(&sq->stats.syncp);
3444 	sq->stats.tx_timeouts++;
3445 	u64_stats_update_end(&sq->stats.syncp);
3446 
3447 	netdev_err(dev, "TX timeout on queue: %u, sq: %s, vq: 0x%x, name: %s, %u usecs ago\n",
3448 		   txqueue, sq->name, sq->vq->index, sq->vq->name,
3449 		   jiffies_to_usecs(jiffies - READ_ONCE(txq->trans_start)));
3450 }
3451 
3452 static const struct net_device_ops virtnet_netdev = {
3453 	.ndo_open            = virtnet_open,
3454 	.ndo_stop   	     = virtnet_close,
3455 	.ndo_start_xmit      = start_xmit,
3456 	.ndo_validate_addr   = eth_validate_addr,
3457 	.ndo_set_mac_address = virtnet_set_mac_address,
3458 	.ndo_set_rx_mode     = virtnet_set_rx_mode,
3459 	.ndo_get_stats64     = virtnet_stats,
3460 	.ndo_vlan_rx_add_vid = virtnet_vlan_rx_add_vid,
3461 	.ndo_vlan_rx_kill_vid = virtnet_vlan_rx_kill_vid,
3462 	.ndo_bpf		= virtnet_xdp,
3463 	.ndo_xdp_xmit		= virtnet_xdp_xmit,
3464 	.ndo_features_check	= passthru_features_check,
3465 	.ndo_get_phys_port_name	= virtnet_get_phys_port_name,
3466 	.ndo_set_features	= virtnet_set_features,
3467 	.ndo_tx_timeout		= virtnet_tx_timeout,
3468 };
3469 
3470 static void virtnet_config_changed_work(struct work_struct *work)
3471 {
3472 	struct virtnet_info *vi =
3473 		container_of(work, struct virtnet_info, config_work);
3474 	u16 v;
3475 
3476 	if (virtio_cread_feature(vi->vdev, VIRTIO_NET_F_STATUS,
3477 				 struct virtio_net_config, status, &v) < 0)
3478 		return;
3479 
3480 	if (v & VIRTIO_NET_S_ANNOUNCE) {
3481 		netdev_notify_peers(vi->dev);
3482 		virtnet_ack_link_announce(vi);
3483 	}
3484 
3485 	/* Ignore unknown (future) status bits */
3486 	v &= VIRTIO_NET_S_LINK_UP;
3487 
3488 	if (vi->status == v)
3489 		return;
3490 
3491 	vi->status = v;
3492 
3493 	if (vi->status & VIRTIO_NET_S_LINK_UP) {
3494 		virtnet_update_settings(vi);
3495 		netif_carrier_on(vi->dev);
3496 		netif_tx_wake_all_queues(vi->dev);
3497 	} else {
3498 		netif_carrier_off(vi->dev);
3499 		netif_tx_stop_all_queues(vi->dev);
3500 	}
3501 }
3502 
3503 static void virtnet_config_changed(struct virtio_device *vdev)
3504 {
3505 	struct virtnet_info *vi = vdev->priv;
3506 
3507 	schedule_work(&vi->config_work);
3508 }
3509 
3510 static void virtnet_free_queues(struct virtnet_info *vi)
3511 {
3512 	int i;
3513 
3514 	for (i = 0; i < vi->max_queue_pairs; i++) {
3515 		__netif_napi_del(&vi->rq[i].napi);
3516 		__netif_napi_del(&vi->sq[i].napi);
3517 	}
3518 
3519 	/* We called __netif_napi_del(),
3520 	 * we need to respect an RCU grace period before freeing vi->rq
3521 	 */
3522 	synchronize_net();
3523 
3524 	kfree(vi->rq);
3525 	kfree(vi->sq);
3526 	kfree(vi->ctrl);
3527 }
3528 
3529 static void _free_receive_bufs(struct virtnet_info *vi)
3530 {
3531 	struct bpf_prog *old_prog;
3532 	int i;
3533 
3534 	for (i = 0; i < vi->max_queue_pairs; i++) {
3535 		while (vi->rq[i].pages)
3536 			__free_pages(get_a_page(&vi->rq[i], GFP_KERNEL), 0);
3537 
3538 		old_prog = rtnl_dereference(vi->rq[i].xdp_prog);
3539 		RCU_INIT_POINTER(vi->rq[i].xdp_prog, NULL);
3540 		if (old_prog)
3541 			bpf_prog_put(old_prog);
3542 	}
3543 }
3544 
3545 static void free_receive_bufs(struct virtnet_info *vi)
3546 {
3547 	rtnl_lock();
3548 	_free_receive_bufs(vi);
3549 	rtnl_unlock();
3550 }
3551 
3552 static void free_receive_page_frags(struct virtnet_info *vi)
3553 {
3554 	int i;
3555 	for (i = 0; i < vi->max_queue_pairs; i++)
3556 		if (vi->rq[i].alloc_frag.page)
3557 			put_page(vi->rq[i].alloc_frag.page);
3558 }
3559 
3560 static void virtnet_sq_free_unused_buf(struct virtqueue *vq, void *buf)
3561 {
3562 	if (!is_xdp_frame(buf))
3563 		dev_kfree_skb(buf);
3564 	else
3565 		xdp_return_frame(ptr_to_xdp(buf));
3566 }
3567 
3568 static void virtnet_rq_free_unused_buf(struct virtqueue *vq, void *buf)
3569 {
3570 	struct virtnet_info *vi = vq->vdev->priv;
3571 	int i = vq2rxq(vq);
3572 
3573 	if (vi->mergeable_rx_bufs)
3574 		put_page(virt_to_head_page(buf));
3575 	else if (vi->big_packets)
3576 		give_pages(&vi->rq[i], buf);
3577 	else
3578 		put_page(virt_to_head_page(buf));
3579 }
3580 
3581 static void free_unused_bufs(struct virtnet_info *vi)
3582 {
3583 	void *buf;
3584 	int i;
3585 
3586 	for (i = 0; i < vi->max_queue_pairs; i++) {
3587 		struct virtqueue *vq = vi->sq[i].vq;
3588 		while ((buf = virtqueue_detach_unused_buf(vq)) != NULL)
3589 			virtnet_sq_free_unused_buf(vq, buf);
3590 		cond_resched();
3591 	}
3592 
3593 	for (i = 0; i < vi->max_queue_pairs; i++) {
3594 		struct virtqueue *vq = vi->rq[i].vq;
3595 		while ((buf = virtqueue_detach_unused_buf(vq)) != NULL)
3596 			virtnet_rq_free_unused_buf(vq, buf);
3597 		cond_resched();
3598 	}
3599 }
3600 
3601 static void virtnet_del_vqs(struct virtnet_info *vi)
3602 {
3603 	struct virtio_device *vdev = vi->vdev;
3604 
3605 	virtnet_clean_affinity(vi);
3606 
3607 	vdev->config->del_vqs(vdev);
3608 
3609 	virtnet_free_queues(vi);
3610 }
3611 
3612 /* How large should a single buffer be so a queue full of these can fit at
3613  * least one full packet?
3614  * Logic below assumes the mergeable buffer header is used.
3615  */
3616 static unsigned int mergeable_min_buf_len(struct virtnet_info *vi, struct virtqueue *vq)
3617 {
3618 	const unsigned int hdr_len = vi->hdr_len;
3619 	unsigned int rq_size = virtqueue_get_vring_size(vq);
3620 	unsigned int packet_len = vi->big_packets ? IP_MAX_MTU : vi->dev->max_mtu;
3621 	unsigned int buf_len = hdr_len + ETH_HLEN + VLAN_HLEN + packet_len;
3622 	unsigned int min_buf_len = DIV_ROUND_UP(buf_len, rq_size);
3623 
3624 	return max(max(min_buf_len, hdr_len) - hdr_len,
3625 		   (unsigned int)GOOD_PACKET_LEN);
3626 }
3627 
3628 static int virtnet_find_vqs(struct virtnet_info *vi)
3629 {
3630 	vq_callback_t **callbacks;
3631 	struct virtqueue **vqs;
3632 	int ret = -ENOMEM;
3633 	int i, total_vqs;
3634 	const char **names;
3635 	bool *ctx;
3636 
3637 	/* We expect 1 RX virtqueue followed by 1 TX virtqueue, followed by
3638 	 * possible N-1 RX/TX queue pairs used in multiqueue mode, followed by
3639 	 * possible control vq.
3640 	 */
3641 	total_vqs = vi->max_queue_pairs * 2 +
3642 		    virtio_has_feature(vi->vdev, VIRTIO_NET_F_CTRL_VQ);
3643 
3644 	/* Allocate space for find_vqs parameters */
3645 	vqs = kcalloc(total_vqs, sizeof(*vqs), GFP_KERNEL);
3646 	if (!vqs)
3647 		goto err_vq;
3648 	callbacks = kmalloc_array(total_vqs, sizeof(*callbacks), GFP_KERNEL);
3649 	if (!callbacks)
3650 		goto err_callback;
3651 	names = kmalloc_array(total_vqs, sizeof(*names), GFP_KERNEL);
3652 	if (!names)
3653 		goto err_names;
3654 	if (!vi->big_packets || vi->mergeable_rx_bufs) {
3655 		ctx = kcalloc(total_vqs, sizeof(*ctx), GFP_KERNEL);
3656 		if (!ctx)
3657 			goto err_ctx;
3658 	} else {
3659 		ctx = NULL;
3660 	}
3661 
3662 	/* Parameters for control virtqueue, if any */
3663 	if (vi->has_cvq) {
3664 		callbacks[total_vqs - 1] = NULL;
3665 		names[total_vqs - 1] = "control";
3666 	}
3667 
3668 	/* Allocate/initialize parameters for send/receive virtqueues */
3669 	for (i = 0; i < vi->max_queue_pairs; i++) {
3670 		callbacks[rxq2vq(i)] = skb_recv_done;
3671 		callbacks[txq2vq(i)] = skb_xmit_done;
3672 		sprintf(vi->rq[i].name, "input.%d", i);
3673 		sprintf(vi->sq[i].name, "output.%d", i);
3674 		names[rxq2vq(i)] = vi->rq[i].name;
3675 		names[txq2vq(i)] = vi->sq[i].name;
3676 		if (ctx)
3677 			ctx[rxq2vq(i)] = true;
3678 	}
3679 
3680 	ret = virtio_find_vqs_ctx(vi->vdev, total_vqs, vqs, callbacks,
3681 				  names, ctx, NULL);
3682 	if (ret)
3683 		goto err_find;
3684 
3685 	if (vi->has_cvq) {
3686 		vi->cvq = vqs[total_vqs - 1];
3687 		if (virtio_has_feature(vi->vdev, VIRTIO_NET_F_CTRL_VLAN))
3688 			vi->dev->features |= NETIF_F_HW_VLAN_CTAG_FILTER;
3689 	}
3690 
3691 	for (i = 0; i < vi->max_queue_pairs; i++) {
3692 		vi->rq[i].vq = vqs[rxq2vq(i)];
3693 		vi->rq[i].min_buf_len = mergeable_min_buf_len(vi, vi->rq[i].vq);
3694 		vi->sq[i].vq = vqs[txq2vq(i)];
3695 	}
3696 
3697 	/* run here: ret == 0. */
3698 
3699 
3700 err_find:
3701 	kfree(ctx);
3702 err_ctx:
3703 	kfree(names);
3704 err_names:
3705 	kfree(callbacks);
3706 err_callback:
3707 	kfree(vqs);
3708 err_vq:
3709 	return ret;
3710 }
3711 
3712 static int virtnet_alloc_queues(struct virtnet_info *vi)
3713 {
3714 	int i;
3715 
3716 	if (vi->has_cvq) {
3717 		vi->ctrl = kzalloc(sizeof(*vi->ctrl), GFP_KERNEL);
3718 		if (!vi->ctrl)
3719 			goto err_ctrl;
3720 	} else {
3721 		vi->ctrl = NULL;
3722 	}
3723 	vi->sq = kcalloc(vi->max_queue_pairs, sizeof(*vi->sq), GFP_KERNEL);
3724 	if (!vi->sq)
3725 		goto err_sq;
3726 	vi->rq = kcalloc(vi->max_queue_pairs, sizeof(*vi->rq), GFP_KERNEL);
3727 	if (!vi->rq)
3728 		goto err_rq;
3729 
3730 	INIT_DELAYED_WORK(&vi->refill, refill_work);
3731 	for (i = 0; i < vi->max_queue_pairs; i++) {
3732 		vi->rq[i].pages = NULL;
3733 		netif_napi_add_weight(vi->dev, &vi->rq[i].napi, virtnet_poll,
3734 				      napi_weight);
3735 		netif_napi_add_tx_weight(vi->dev, &vi->sq[i].napi,
3736 					 virtnet_poll_tx,
3737 					 napi_tx ? napi_weight : 0);
3738 
3739 		sg_init_table(vi->rq[i].sg, ARRAY_SIZE(vi->rq[i].sg));
3740 		ewma_pkt_len_init(&vi->rq[i].mrg_avg_pkt_len);
3741 		sg_init_table(vi->sq[i].sg, ARRAY_SIZE(vi->sq[i].sg));
3742 
3743 		u64_stats_init(&vi->rq[i].stats.syncp);
3744 		u64_stats_init(&vi->sq[i].stats.syncp);
3745 	}
3746 
3747 	return 0;
3748 
3749 err_rq:
3750 	kfree(vi->sq);
3751 err_sq:
3752 	kfree(vi->ctrl);
3753 err_ctrl:
3754 	return -ENOMEM;
3755 }
3756 
3757 static int init_vqs(struct virtnet_info *vi)
3758 {
3759 	int ret;
3760 
3761 	/* Allocate send & receive queues */
3762 	ret = virtnet_alloc_queues(vi);
3763 	if (ret)
3764 		goto err;
3765 
3766 	ret = virtnet_find_vqs(vi);
3767 	if (ret)
3768 		goto err_free;
3769 
3770 	cpus_read_lock();
3771 	virtnet_set_affinity(vi);
3772 	cpus_read_unlock();
3773 
3774 	return 0;
3775 
3776 err_free:
3777 	virtnet_free_queues(vi);
3778 err:
3779 	return ret;
3780 }
3781 
3782 #ifdef CONFIG_SYSFS
3783 static ssize_t mergeable_rx_buffer_size_show(struct netdev_rx_queue *queue,
3784 		char *buf)
3785 {
3786 	struct virtnet_info *vi = netdev_priv(queue->dev);
3787 	unsigned int queue_index = get_netdev_rx_queue_index(queue);
3788 	unsigned int headroom = virtnet_get_headroom(vi);
3789 	unsigned int tailroom = headroom ? sizeof(struct skb_shared_info) : 0;
3790 	struct ewma_pkt_len *avg;
3791 
3792 	BUG_ON(queue_index >= vi->max_queue_pairs);
3793 	avg = &vi->rq[queue_index].mrg_avg_pkt_len;
3794 	return sprintf(buf, "%u\n",
3795 		       get_mergeable_buf_len(&vi->rq[queue_index], avg,
3796 				       SKB_DATA_ALIGN(headroom + tailroom)));
3797 }
3798 
3799 static struct rx_queue_attribute mergeable_rx_buffer_size_attribute =
3800 	__ATTR_RO(mergeable_rx_buffer_size);
3801 
3802 static struct attribute *virtio_net_mrg_rx_attrs[] = {
3803 	&mergeable_rx_buffer_size_attribute.attr,
3804 	NULL
3805 };
3806 
3807 static const struct attribute_group virtio_net_mrg_rx_group = {
3808 	.name = "virtio_net",
3809 	.attrs = virtio_net_mrg_rx_attrs
3810 };
3811 #endif
3812 
3813 static bool virtnet_fail_on_feature(struct virtio_device *vdev,
3814 				    unsigned int fbit,
3815 				    const char *fname, const char *dname)
3816 {
3817 	if (!virtio_has_feature(vdev, fbit))
3818 		return false;
3819 
3820 	dev_err(&vdev->dev, "device advertises feature %s but not %s",
3821 		fname, dname);
3822 
3823 	return true;
3824 }
3825 
3826 #define VIRTNET_FAIL_ON(vdev, fbit, dbit)			\
3827 	virtnet_fail_on_feature(vdev, fbit, #fbit, dbit)
3828 
3829 static bool virtnet_validate_features(struct virtio_device *vdev)
3830 {
3831 	if (!virtio_has_feature(vdev, VIRTIO_NET_F_CTRL_VQ) &&
3832 	    (VIRTNET_FAIL_ON(vdev, VIRTIO_NET_F_CTRL_RX,
3833 			     "VIRTIO_NET_F_CTRL_VQ") ||
3834 	     VIRTNET_FAIL_ON(vdev, VIRTIO_NET_F_CTRL_VLAN,
3835 			     "VIRTIO_NET_F_CTRL_VQ") ||
3836 	     VIRTNET_FAIL_ON(vdev, VIRTIO_NET_F_GUEST_ANNOUNCE,
3837 			     "VIRTIO_NET_F_CTRL_VQ") ||
3838 	     VIRTNET_FAIL_ON(vdev, VIRTIO_NET_F_MQ, "VIRTIO_NET_F_CTRL_VQ") ||
3839 	     VIRTNET_FAIL_ON(vdev, VIRTIO_NET_F_CTRL_MAC_ADDR,
3840 			     "VIRTIO_NET_F_CTRL_VQ") ||
3841 	     VIRTNET_FAIL_ON(vdev, VIRTIO_NET_F_RSS,
3842 			     "VIRTIO_NET_F_CTRL_VQ") ||
3843 	     VIRTNET_FAIL_ON(vdev, VIRTIO_NET_F_HASH_REPORT,
3844 			     "VIRTIO_NET_F_CTRL_VQ") ||
3845 	     VIRTNET_FAIL_ON(vdev, VIRTIO_NET_F_NOTF_COAL,
3846 			     "VIRTIO_NET_F_CTRL_VQ"))) {
3847 		return false;
3848 	}
3849 
3850 	return true;
3851 }
3852 
3853 #define MIN_MTU ETH_MIN_MTU
3854 #define MAX_MTU ETH_MAX_MTU
3855 
3856 static int virtnet_validate(struct virtio_device *vdev)
3857 {
3858 	if (!vdev->config->get) {
3859 		dev_err(&vdev->dev, "%s failure: config access disabled\n",
3860 			__func__);
3861 		return -EINVAL;
3862 	}
3863 
3864 	if (!virtnet_validate_features(vdev))
3865 		return -EINVAL;
3866 
3867 	if (virtio_has_feature(vdev, VIRTIO_NET_F_MTU)) {
3868 		int mtu = virtio_cread16(vdev,
3869 					 offsetof(struct virtio_net_config,
3870 						  mtu));
3871 		if (mtu < MIN_MTU)
3872 			__virtio_clear_bit(vdev, VIRTIO_NET_F_MTU);
3873 	}
3874 
3875 	if (virtio_has_feature(vdev, VIRTIO_NET_F_STANDBY) &&
3876 	    !virtio_has_feature(vdev, VIRTIO_NET_F_MAC)) {
3877 		dev_warn(&vdev->dev, "device advertises feature VIRTIO_NET_F_STANDBY but not VIRTIO_NET_F_MAC, disabling standby");
3878 		__virtio_clear_bit(vdev, VIRTIO_NET_F_STANDBY);
3879 	}
3880 
3881 	return 0;
3882 }
3883 
3884 static bool virtnet_check_guest_gso(const struct virtnet_info *vi)
3885 {
3886 	return virtio_has_feature(vi->vdev, VIRTIO_NET_F_GUEST_TSO4) ||
3887 		virtio_has_feature(vi->vdev, VIRTIO_NET_F_GUEST_TSO6) ||
3888 		virtio_has_feature(vi->vdev, VIRTIO_NET_F_GUEST_ECN) ||
3889 		virtio_has_feature(vi->vdev, VIRTIO_NET_F_GUEST_UFO) ||
3890 		(virtio_has_feature(vi->vdev, VIRTIO_NET_F_GUEST_USO4) &&
3891 		virtio_has_feature(vi->vdev, VIRTIO_NET_F_GUEST_USO6));
3892 }
3893 
3894 static void virtnet_set_big_packets(struct virtnet_info *vi, const int mtu)
3895 {
3896 	bool guest_gso = virtnet_check_guest_gso(vi);
3897 
3898 	/* If device can receive ANY guest GSO packets, regardless of mtu,
3899 	 * allocate packets of maximum size, otherwise limit it to only
3900 	 * mtu size worth only.
3901 	 */
3902 	if (mtu > ETH_DATA_LEN || guest_gso) {
3903 		vi->big_packets = true;
3904 		vi->big_packets_num_skbfrags = guest_gso ? MAX_SKB_FRAGS : DIV_ROUND_UP(mtu, PAGE_SIZE);
3905 	}
3906 }
3907 
3908 static int virtnet_probe(struct virtio_device *vdev)
3909 {
3910 	int i, err = -ENOMEM;
3911 	struct net_device *dev;
3912 	struct virtnet_info *vi;
3913 	u16 max_queue_pairs;
3914 	int mtu = 0;
3915 
3916 	/* Find if host supports multiqueue/rss virtio_net device */
3917 	max_queue_pairs = 1;
3918 	if (virtio_has_feature(vdev, VIRTIO_NET_F_MQ) || virtio_has_feature(vdev, VIRTIO_NET_F_RSS))
3919 		max_queue_pairs =
3920 		     virtio_cread16(vdev, offsetof(struct virtio_net_config, max_virtqueue_pairs));
3921 
3922 	/* We need at least 2 queue's */
3923 	if (max_queue_pairs < VIRTIO_NET_CTRL_MQ_VQ_PAIRS_MIN ||
3924 	    max_queue_pairs > VIRTIO_NET_CTRL_MQ_VQ_PAIRS_MAX ||
3925 	    !virtio_has_feature(vdev, VIRTIO_NET_F_CTRL_VQ))
3926 		max_queue_pairs = 1;
3927 
3928 	/* Allocate ourselves a network device with room for our info */
3929 	dev = alloc_etherdev_mq(sizeof(struct virtnet_info), max_queue_pairs);
3930 	if (!dev)
3931 		return -ENOMEM;
3932 
3933 	/* Set up network device as normal. */
3934 	dev->priv_flags |= IFF_UNICAST_FLT | IFF_LIVE_ADDR_CHANGE |
3935 			   IFF_TX_SKB_NO_LINEAR;
3936 	dev->netdev_ops = &virtnet_netdev;
3937 	dev->features = NETIF_F_HIGHDMA;
3938 
3939 	dev->ethtool_ops = &virtnet_ethtool_ops;
3940 	SET_NETDEV_DEV(dev, &vdev->dev);
3941 
3942 	/* Do we support "hardware" checksums? */
3943 	if (virtio_has_feature(vdev, VIRTIO_NET_F_CSUM)) {
3944 		/* This opens up the world of extra features. */
3945 		dev->hw_features |= NETIF_F_HW_CSUM | NETIF_F_SG;
3946 		if (csum)
3947 			dev->features |= NETIF_F_HW_CSUM | NETIF_F_SG;
3948 
3949 		if (virtio_has_feature(vdev, VIRTIO_NET_F_GSO)) {
3950 			dev->hw_features |= NETIF_F_TSO
3951 				| NETIF_F_TSO_ECN | NETIF_F_TSO6;
3952 		}
3953 		/* Individual feature bits: what can host handle? */
3954 		if (virtio_has_feature(vdev, VIRTIO_NET_F_HOST_TSO4))
3955 			dev->hw_features |= NETIF_F_TSO;
3956 		if (virtio_has_feature(vdev, VIRTIO_NET_F_HOST_TSO6))
3957 			dev->hw_features |= NETIF_F_TSO6;
3958 		if (virtio_has_feature(vdev, VIRTIO_NET_F_HOST_ECN))
3959 			dev->hw_features |= NETIF_F_TSO_ECN;
3960 		if (virtio_has_feature(vdev, VIRTIO_NET_F_HOST_USO))
3961 			dev->hw_features |= NETIF_F_GSO_UDP_L4;
3962 
3963 		dev->features |= NETIF_F_GSO_ROBUST;
3964 
3965 		if (gso)
3966 			dev->features |= dev->hw_features & NETIF_F_ALL_TSO;
3967 		/* (!csum && gso) case will be fixed by register_netdev() */
3968 	}
3969 	if (virtio_has_feature(vdev, VIRTIO_NET_F_GUEST_CSUM))
3970 		dev->features |= NETIF_F_RXCSUM;
3971 	if (virtio_has_feature(vdev, VIRTIO_NET_F_GUEST_TSO4) ||
3972 	    virtio_has_feature(vdev, VIRTIO_NET_F_GUEST_TSO6))
3973 		dev->features |= NETIF_F_GRO_HW;
3974 	if (virtio_has_feature(vdev, VIRTIO_NET_F_CTRL_GUEST_OFFLOADS))
3975 		dev->hw_features |= NETIF_F_GRO_HW;
3976 
3977 	dev->vlan_features = dev->features;
3978 	dev->xdp_features = NETDEV_XDP_ACT_BASIC | NETDEV_XDP_ACT_REDIRECT;
3979 
3980 	/* MTU range: 68 - 65535 */
3981 	dev->min_mtu = MIN_MTU;
3982 	dev->max_mtu = MAX_MTU;
3983 
3984 	/* Configuration may specify what MAC to use.  Otherwise random. */
3985 	if (virtio_has_feature(vdev, VIRTIO_NET_F_MAC)) {
3986 		u8 addr[ETH_ALEN];
3987 
3988 		virtio_cread_bytes(vdev,
3989 				   offsetof(struct virtio_net_config, mac),
3990 				   addr, ETH_ALEN);
3991 		eth_hw_addr_set(dev, addr);
3992 	} else {
3993 		eth_hw_addr_random(dev);
3994 		dev_info(&vdev->dev, "Assigned random MAC address %pM\n",
3995 			 dev->dev_addr);
3996 	}
3997 
3998 	/* Set up our device-specific information */
3999 	vi = netdev_priv(dev);
4000 	vi->dev = dev;
4001 	vi->vdev = vdev;
4002 	vdev->priv = vi;
4003 
4004 	INIT_WORK(&vi->config_work, virtnet_config_changed_work);
4005 	spin_lock_init(&vi->refill_lock);
4006 
4007 	if (virtio_has_feature(vdev, VIRTIO_NET_F_MRG_RXBUF)) {
4008 		vi->mergeable_rx_bufs = true;
4009 		dev->xdp_features |= NETDEV_XDP_ACT_RX_SG;
4010 	}
4011 
4012 	if (virtio_has_feature(vi->vdev, VIRTIO_NET_F_NOTF_COAL)) {
4013 		vi->rx_usecs = 0;
4014 		vi->tx_usecs = 0;
4015 		vi->tx_max_packets = 0;
4016 		vi->rx_max_packets = 0;
4017 	}
4018 
4019 	if (virtio_has_feature(vdev, VIRTIO_NET_F_HASH_REPORT))
4020 		vi->has_rss_hash_report = true;
4021 
4022 	if (virtio_has_feature(vdev, VIRTIO_NET_F_RSS))
4023 		vi->has_rss = true;
4024 
4025 	if (vi->has_rss || vi->has_rss_hash_report) {
4026 		vi->rss_indir_table_size =
4027 			virtio_cread16(vdev, offsetof(struct virtio_net_config,
4028 				rss_max_indirection_table_length));
4029 		vi->rss_key_size =
4030 			virtio_cread8(vdev, offsetof(struct virtio_net_config, rss_max_key_size));
4031 
4032 		vi->rss_hash_types_supported =
4033 		    virtio_cread32(vdev, offsetof(struct virtio_net_config, supported_hash_types));
4034 		vi->rss_hash_types_supported &=
4035 				~(VIRTIO_NET_RSS_HASH_TYPE_IP_EX |
4036 				  VIRTIO_NET_RSS_HASH_TYPE_TCP_EX |
4037 				  VIRTIO_NET_RSS_HASH_TYPE_UDP_EX);
4038 
4039 		dev->hw_features |= NETIF_F_RXHASH;
4040 	}
4041 
4042 	if (vi->has_rss_hash_report)
4043 		vi->hdr_len = sizeof(struct virtio_net_hdr_v1_hash);
4044 	else if (virtio_has_feature(vdev, VIRTIO_NET_F_MRG_RXBUF) ||
4045 		 virtio_has_feature(vdev, VIRTIO_F_VERSION_1))
4046 		vi->hdr_len = sizeof(struct virtio_net_hdr_mrg_rxbuf);
4047 	else
4048 		vi->hdr_len = sizeof(struct virtio_net_hdr);
4049 
4050 	if (virtio_has_feature(vdev, VIRTIO_F_ANY_LAYOUT) ||
4051 	    virtio_has_feature(vdev, VIRTIO_F_VERSION_1))
4052 		vi->any_header_sg = true;
4053 
4054 	if (virtio_has_feature(vdev, VIRTIO_NET_F_CTRL_VQ))
4055 		vi->has_cvq = true;
4056 
4057 	if (virtio_has_feature(vdev, VIRTIO_NET_F_MTU)) {
4058 		mtu = virtio_cread16(vdev,
4059 				     offsetof(struct virtio_net_config,
4060 					      mtu));
4061 		if (mtu < dev->min_mtu) {
4062 			/* Should never trigger: MTU was previously validated
4063 			 * in virtnet_validate.
4064 			 */
4065 			dev_err(&vdev->dev,
4066 				"device MTU appears to have changed it is now %d < %d",
4067 				mtu, dev->min_mtu);
4068 			err = -EINVAL;
4069 			goto free;
4070 		}
4071 
4072 		dev->mtu = mtu;
4073 		dev->max_mtu = mtu;
4074 	}
4075 
4076 	virtnet_set_big_packets(vi, mtu);
4077 
4078 	if (vi->any_header_sg)
4079 		dev->needed_headroom = vi->hdr_len;
4080 
4081 	/* Enable multiqueue by default */
4082 	if (num_online_cpus() >= max_queue_pairs)
4083 		vi->curr_queue_pairs = max_queue_pairs;
4084 	else
4085 		vi->curr_queue_pairs = num_online_cpus();
4086 	vi->max_queue_pairs = max_queue_pairs;
4087 
4088 	/* Allocate/initialize the rx/tx queues, and invoke find_vqs */
4089 	err = init_vqs(vi);
4090 	if (err)
4091 		goto free;
4092 
4093 #ifdef CONFIG_SYSFS
4094 	if (vi->mergeable_rx_bufs)
4095 		dev->sysfs_rx_queue_group = &virtio_net_mrg_rx_group;
4096 #endif
4097 	netif_set_real_num_tx_queues(dev, vi->curr_queue_pairs);
4098 	netif_set_real_num_rx_queues(dev, vi->curr_queue_pairs);
4099 
4100 	virtnet_init_settings(dev);
4101 
4102 	if (virtio_has_feature(vdev, VIRTIO_NET_F_STANDBY)) {
4103 		vi->failover = net_failover_create(vi->dev);
4104 		if (IS_ERR(vi->failover)) {
4105 			err = PTR_ERR(vi->failover);
4106 			goto free_vqs;
4107 		}
4108 	}
4109 
4110 	if (vi->has_rss || vi->has_rss_hash_report)
4111 		virtnet_init_default_rss(vi);
4112 
4113 	/* serialize netdev register + virtio_device_ready() with ndo_open() */
4114 	rtnl_lock();
4115 
4116 	err = register_netdevice(dev);
4117 	if (err) {
4118 		pr_debug("virtio_net: registering device failed\n");
4119 		rtnl_unlock();
4120 		goto free_failover;
4121 	}
4122 
4123 	virtio_device_ready(vdev);
4124 
4125 	/* a random MAC address has been assigned, notify the device.
4126 	 * We don't fail probe if VIRTIO_NET_F_CTRL_MAC_ADDR is not there
4127 	 * because many devices work fine without getting MAC explicitly
4128 	 */
4129 	if (!virtio_has_feature(vdev, VIRTIO_NET_F_MAC) &&
4130 	    virtio_has_feature(vi->vdev, VIRTIO_NET_F_CTRL_MAC_ADDR)) {
4131 		struct scatterlist sg;
4132 
4133 		sg_init_one(&sg, dev->dev_addr, dev->addr_len);
4134 		if (!virtnet_send_command(vi, VIRTIO_NET_CTRL_MAC,
4135 					  VIRTIO_NET_CTRL_MAC_ADDR_SET, &sg)) {
4136 			pr_debug("virtio_net: setting MAC address failed\n");
4137 			rtnl_unlock();
4138 			err = -EINVAL;
4139 			goto free_unregister_netdev;
4140 		}
4141 	}
4142 
4143 	rtnl_unlock();
4144 
4145 	err = virtnet_cpu_notif_add(vi);
4146 	if (err) {
4147 		pr_debug("virtio_net: registering cpu notifier failed\n");
4148 		goto free_unregister_netdev;
4149 	}
4150 
4151 	virtnet_set_queues(vi, vi->curr_queue_pairs);
4152 
4153 	/* Assume link up if device can't report link status,
4154 	   otherwise get link status from config. */
4155 	netif_carrier_off(dev);
4156 	if (virtio_has_feature(vi->vdev, VIRTIO_NET_F_STATUS)) {
4157 		schedule_work(&vi->config_work);
4158 	} else {
4159 		vi->status = VIRTIO_NET_S_LINK_UP;
4160 		virtnet_update_settings(vi);
4161 		netif_carrier_on(dev);
4162 	}
4163 
4164 	for (i = 0; i < ARRAY_SIZE(guest_offloads); i++)
4165 		if (virtio_has_feature(vi->vdev, guest_offloads[i]))
4166 			set_bit(guest_offloads[i], &vi->guest_offloads);
4167 	vi->guest_offloads_capable = vi->guest_offloads;
4168 
4169 	pr_debug("virtnet: registered device %s with %d RX and TX vq's\n",
4170 		 dev->name, max_queue_pairs);
4171 
4172 	return 0;
4173 
4174 free_unregister_netdev:
4175 	unregister_netdev(dev);
4176 free_failover:
4177 	net_failover_destroy(vi->failover);
4178 free_vqs:
4179 	virtio_reset_device(vdev);
4180 	cancel_delayed_work_sync(&vi->refill);
4181 	free_receive_page_frags(vi);
4182 	virtnet_del_vqs(vi);
4183 free:
4184 	free_netdev(dev);
4185 	return err;
4186 }
4187 
4188 static void remove_vq_common(struct virtnet_info *vi)
4189 {
4190 	virtio_reset_device(vi->vdev);
4191 
4192 	/* Free unused buffers in both send and recv, if any. */
4193 	free_unused_bufs(vi);
4194 
4195 	free_receive_bufs(vi);
4196 
4197 	free_receive_page_frags(vi);
4198 
4199 	virtnet_del_vqs(vi);
4200 }
4201 
4202 static void virtnet_remove(struct virtio_device *vdev)
4203 {
4204 	struct virtnet_info *vi = vdev->priv;
4205 
4206 	virtnet_cpu_notif_remove(vi);
4207 
4208 	/* Make sure no work handler is accessing the device. */
4209 	flush_work(&vi->config_work);
4210 
4211 	unregister_netdev(vi->dev);
4212 
4213 	net_failover_destroy(vi->failover);
4214 
4215 	remove_vq_common(vi);
4216 
4217 	free_netdev(vi->dev);
4218 }
4219 
4220 static __maybe_unused int virtnet_freeze(struct virtio_device *vdev)
4221 {
4222 	struct virtnet_info *vi = vdev->priv;
4223 
4224 	virtnet_cpu_notif_remove(vi);
4225 	virtnet_freeze_down(vdev);
4226 	remove_vq_common(vi);
4227 
4228 	return 0;
4229 }
4230 
4231 static __maybe_unused int virtnet_restore(struct virtio_device *vdev)
4232 {
4233 	struct virtnet_info *vi = vdev->priv;
4234 	int err;
4235 
4236 	err = virtnet_restore_up(vdev);
4237 	if (err)
4238 		return err;
4239 	virtnet_set_queues(vi, vi->curr_queue_pairs);
4240 
4241 	err = virtnet_cpu_notif_add(vi);
4242 	if (err) {
4243 		virtnet_freeze_down(vdev);
4244 		remove_vq_common(vi);
4245 		return err;
4246 	}
4247 
4248 	return 0;
4249 }
4250 
4251 static struct virtio_device_id id_table[] = {
4252 	{ VIRTIO_ID_NET, VIRTIO_DEV_ANY_ID },
4253 	{ 0 },
4254 };
4255 
4256 #define VIRTNET_FEATURES \
4257 	VIRTIO_NET_F_CSUM, VIRTIO_NET_F_GUEST_CSUM, \
4258 	VIRTIO_NET_F_MAC, \
4259 	VIRTIO_NET_F_HOST_TSO4, VIRTIO_NET_F_HOST_UFO, VIRTIO_NET_F_HOST_TSO6, \
4260 	VIRTIO_NET_F_HOST_ECN, VIRTIO_NET_F_GUEST_TSO4, VIRTIO_NET_F_GUEST_TSO6, \
4261 	VIRTIO_NET_F_GUEST_ECN, VIRTIO_NET_F_GUEST_UFO, \
4262 	VIRTIO_NET_F_HOST_USO, VIRTIO_NET_F_GUEST_USO4, VIRTIO_NET_F_GUEST_USO6, \
4263 	VIRTIO_NET_F_MRG_RXBUF, VIRTIO_NET_F_STATUS, VIRTIO_NET_F_CTRL_VQ, \
4264 	VIRTIO_NET_F_CTRL_RX, VIRTIO_NET_F_CTRL_VLAN, \
4265 	VIRTIO_NET_F_GUEST_ANNOUNCE, VIRTIO_NET_F_MQ, \
4266 	VIRTIO_NET_F_CTRL_MAC_ADDR, \
4267 	VIRTIO_NET_F_MTU, VIRTIO_NET_F_CTRL_GUEST_OFFLOADS, \
4268 	VIRTIO_NET_F_SPEED_DUPLEX, VIRTIO_NET_F_STANDBY, \
4269 	VIRTIO_NET_F_RSS, VIRTIO_NET_F_HASH_REPORT, VIRTIO_NET_F_NOTF_COAL, \
4270 	VIRTIO_NET_F_GUEST_HDRLEN
4271 
4272 static unsigned int features[] = {
4273 	VIRTNET_FEATURES,
4274 };
4275 
4276 static unsigned int features_legacy[] = {
4277 	VIRTNET_FEATURES,
4278 	VIRTIO_NET_F_GSO,
4279 	VIRTIO_F_ANY_LAYOUT,
4280 };
4281 
4282 static struct virtio_driver virtio_net_driver = {
4283 	.feature_table = features,
4284 	.feature_table_size = ARRAY_SIZE(features),
4285 	.feature_table_legacy = features_legacy,
4286 	.feature_table_size_legacy = ARRAY_SIZE(features_legacy),
4287 	.driver.name =	KBUILD_MODNAME,
4288 	.driver.owner =	THIS_MODULE,
4289 	.id_table =	id_table,
4290 	.validate =	virtnet_validate,
4291 	.probe =	virtnet_probe,
4292 	.remove =	virtnet_remove,
4293 	.config_changed = virtnet_config_changed,
4294 #ifdef CONFIG_PM_SLEEP
4295 	.freeze =	virtnet_freeze,
4296 	.restore =	virtnet_restore,
4297 #endif
4298 };
4299 
4300 static __init int virtio_net_driver_init(void)
4301 {
4302 	int ret;
4303 
4304 	ret = cpuhp_setup_state_multi(CPUHP_AP_ONLINE_DYN, "virtio/net:online",
4305 				      virtnet_cpu_online,
4306 				      virtnet_cpu_down_prep);
4307 	if (ret < 0)
4308 		goto out;
4309 	virtionet_online = ret;
4310 	ret = cpuhp_setup_state_multi(CPUHP_VIRT_NET_DEAD, "virtio/net:dead",
4311 				      NULL, virtnet_cpu_dead);
4312 	if (ret)
4313 		goto err_dead;
4314 	ret = register_virtio_driver(&virtio_net_driver);
4315 	if (ret)
4316 		goto err_virtio;
4317 	return 0;
4318 err_virtio:
4319 	cpuhp_remove_multi_state(CPUHP_VIRT_NET_DEAD);
4320 err_dead:
4321 	cpuhp_remove_multi_state(virtionet_online);
4322 out:
4323 	return ret;
4324 }
4325 module_init(virtio_net_driver_init);
4326 
4327 static __exit void virtio_net_driver_exit(void)
4328 {
4329 	unregister_virtio_driver(&virtio_net_driver);
4330 	cpuhp_remove_multi_state(CPUHP_VIRT_NET_DEAD);
4331 	cpuhp_remove_multi_state(virtionet_online);
4332 }
4333 module_exit(virtio_net_driver_exit);
4334 
4335 MODULE_DEVICE_TABLE(virtio, id_table);
4336 MODULE_DESCRIPTION("Virtio network driver");
4337 MODULE_LICENSE("GPL");
4338