xref: /linux/drivers/net/virtio_net.c (revision 7a9b709e7cc5ce1ffb84ce07bf6d157e1de758df)
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 <linux/dim.h>
23 #include <net/route.h>
24 #include <net/xdp.h>
25 #include <net/net_failover.h>
26 #include <net/netdev_rx_queue.h>
27 #include <net/netdev_queues.h>
28 #include <net/xdp_sock_drv.h>
29 
30 static int napi_weight = NAPI_POLL_WEIGHT;
31 module_param(napi_weight, int, 0444);
32 
33 static bool csum = true, gso = true, napi_tx = true;
34 module_param(csum, bool, 0444);
35 module_param(gso, bool, 0444);
36 module_param(napi_tx, bool, 0644);
37 
38 /* FIXME: MTU in config. */
39 #define GOOD_PACKET_LEN (ETH_HLEN + VLAN_HLEN + ETH_DATA_LEN)
40 #define GOOD_COPY_LEN	128
41 
42 #define VIRTNET_RX_PAD (NET_IP_ALIGN + NET_SKB_PAD)
43 
44 /* Separating two types of XDP xmit */
45 #define VIRTIO_XDP_TX		BIT(0)
46 #define VIRTIO_XDP_REDIR	BIT(1)
47 
48 /* RX packet size EWMA. The average packet size is used to determine the packet
49  * buffer size when refilling RX rings. As the entire RX ring may be refilled
50  * at once, the weight is chosen so that the EWMA will be insensitive to short-
51  * term, transient changes in packet size.
52  */
53 DECLARE_EWMA(pkt_len, 0, 64)
54 
55 #define VIRTNET_DRIVER_VERSION "1.0.0"
56 
57 static const unsigned long guest_offloads[] = {
58 	VIRTIO_NET_F_GUEST_TSO4,
59 	VIRTIO_NET_F_GUEST_TSO6,
60 	VIRTIO_NET_F_GUEST_ECN,
61 	VIRTIO_NET_F_GUEST_UFO,
62 	VIRTIO_NET_F_GUEST_CSUM,
63 	VIRTIO_NET_F_GUEST_USO4,
64 	VIRTIO_NET_F_GUEST_USO6,
65 	VIRTIO_NET_F_GUEST_HDRLEN
66 };
67 
68 #define GUEST_OFFLOAD_GRO_HW_MASK ((1ULL << VIRTIO_NET_F_GUEST_TSO4) | \
69 				(1ULL << VIRTIO_NET_F_GUEST_TSO6) | \
70 				(1ULL << VIRTIO_NET_F_GUEST_ECN)  | \
71 				(1ULL << VIRTIO_NET_F_GUEST_UFO)  | \
72 				(1ULL << VIRTIO_NET_F_GUEST_USO4) | \
73 				(1ULL << VIRTIO_NET_F_GUEST_USO6))
74 
75 struct virtnet_stat_desc {
76 	char desc[ETH_GSTRING_LEN];
77 	size_t offset;
78 	size_t qstat_offset;
79 };
80 
81 struct virtnet_sq_free_stats {
82 	u64 packets;
83 	u64 bytes;
84 	u64 napi_packets;
85 	u64 napi_bytes;
86 	u64 xsk;
87 };
88 
89 struct virtnet_sq_stats {
90 	struct u64_stats_sync syncp;
91 	u64_stats_t packets;
92 	u64_stats_t bytes;
93 	u64_stats_t xdp_tx;
94 	u64_stats_t xdp_tx_drops;
95 	u64_stats_t kicks;
96 	u64_stats_t tx_timeouts;
97 	u64_stats_t stop;
98 	u64_stats_t wake;
99 };
100 
101 struct virtnet_rq_stats {
102 	struct u64_stats_sync syncp;
103 	u64_stats_t packets;
104 	u64_stats_t bytes;
105 	u64_stats_t drops;
106 	u64_stats_t xdp_packets;
107 	u64_stats_t xdp_tx;
108 	u64_stats_t xdp_redirects;
109 	u64_stats_t xdp_drops;
110 	u64_stats_t kicks;
111 };
112 
113 #define VIRTNET_SQ_STAT(name, m) {name, offsetof(struct virtnet_sq_stats, m), -1}
114 #define VIRTNET_RQ_STAT(name, m) {name, offsetof(struct virtnet_rq_stats, m), -1}
115 
116 #define VIRTNET_SQ_STAT_QSTAT(name, m)				\
117 	{							\
118 		name,						\
119 		offsetof(struct virtnet_sq_stats, m),		\
120 		offsetof(struct netdev_queue_stats_tx, m),	\
121 	}
122 
123 #define VIRTNET_RQ_STAT_QSTAT(name, m)				\
124 	{							\
125 		name,						\
126 		offsetof(struct virtnet_rq_stats, m),		\
127 		offsetof(struct netdev_queue_stats_rx, m),	\
128 	}
129 
130 static const struct virtnet_stat_desc virtnet_sq_stats_desc[] = {
131 	VIRTNET_SQ_STAT("xdp_tx",       xdp_tx),
132 	VIRTNET_SQ_STAT("xdp_tx_drops", xdp_tx_drops),
133 	VIRTNET_SQ_STAT("kicks",        kicks),
134 	VIRTNET_SQ_STAT("tx_timeouts",  tx_timeouts),
135 };
136 
137 static const struct virtnet_stat_desc virtnet_rq_stats_desc[] = {
138 	VIRTNET_RQ_STAT("drops",         drops),
139 	VIRTNET_RQ_STAT("xdp_packets",   xdp_packets),
140 	VIRTNET_RQ_STAT("xdp_tx",        xdp_tx),
141 	VIRTNET_RQ_STAT("xdp_redirects", xdp_redirects),
142 	VIRTNET_RQ_STAT("xdp_drops",     xdp_drops),
143 	VIRTNET_RQ_STAT("kicks",         kicks),
144 };
145 
146 static const struct virtnet_stat_desc virtnet_sq_stats_desc_qstat[] = {
147 	VIRTNET_SQ_STAT_QSTAT("packets", packets),
148 	VIRTNET_SQ_STAT_QSTAT("bytes",   bytes),
149 	VIRTNET_SQ_STAT_QSTAT("stop",	 stop),
150 	VIRTNET_SQ_STAT_QSTAT("wake",	 wake),
151 };
152 
153 static const struct virtnet_stat_desc virtnet_rq_stats_desc_qstat[] = {
154 	VIRTNET_RQ_STAT_QSTAT("packets", packets),
155 	VIRTNET_RQ_STAT_QSTAT("bytes",   bytes),
156 };
157 
158 #define VIRTNET_STATS_DESC_CQ(name) \
159 	{#name, offsetof(struct virtio_net_stats_cvq, name), -1}
160 
161 #define VIRTNET_STATS_DESC_RX(class, name) \
162 	{#name, offsetof(struct virtio_net_stats_rx_ ## class, rx_ ## name), -1}
163 
164 #define VIRTNET_STATS_DESC_TX(class, name) \
165 	{#name, offsetof(struct virtio_net_stats_tx_ ## class, tx_ ## name), -1}
166 
167 
168 static const struct virtnet_stat_desc virtnet_stats_cvq_desc[] = {
169 	VIRTNET_STATS_DESC_CQ(command_num),
170 	VIRTNET_STATS_DESC_CQ(ok_num),
171 };
172 
173 static const struct virtnet_stat_desc virtnet_stats_rx_basic_desc[] = {
174 	VIRTNET_STATS_DESC_RX(basic, packets),
175 	VIRTNET_STATS_DESC_RX(basic, bytes),
176 
177 	VIRTNET_STATS_DESC_RX(basic, notifications),
178 	VIRTNET_STATS_DESC_RX(basic, interrupts),
179 };
180 
181 static const struct virtnet_stat_desc virtnet_stats_tx_basic_desc[] = {
182 	VIRTNET_STATS_DESC_TX(basic, packets),
183 	VIRTNET_STATS_DESC_TX(basic, bytes),
184 
185 	VIRTNET_STATS_DESC_TX(basic, notifications),
186 	VIRTNET_STATS_DESC_TX(basic, interrupts),
187 };
188 
189 static const struct virtnet_stat_desc virtnet_stats_rx_csum_desc[] = {
190 	VIRTNET_STATS_DESC_RX(csum, needs_csum),
191 };
192 
193 static const struct virtnet_stat_desc virtnet_stats_tx_gso_desc[] = {
194 	VIRTNET_STATS_DESC_TX(gso, gso_packets_noseg),
195 	VIRTNET_STATS_DESC_TX(gso, gso_bytes_noseg),
196 };
197 
198 static const struct virtnet_stat_desc virtnet_stats_rx_speed_desc[] = {
199 	VIRTNET_STATS_DESC_RX(speed, ratelimit_bytes),
200 };
201 
202 static const struct virtnet_stat_desc virtnet_stats_tx_speed_desc[] = {
203 	VIRTNET_STATS_DESC_TX(speed, ratelimit_bytes),
204 };
205 
206 #define VIRTNET_STATS_DESC_RX_QSTAT(class, name, qstat_field)			\
207 	{									\
208 		#name,								\
209 		offsetof(struct virtio_net_stats_rx_ ## class, rx_ ## name),	\
210 		offsetof(struct netdev_queue_stats_rx, qstat_field),		\
211 	}
212 
213 #define VIRTNET_STATS_DESC_TX_QSTAT(class, name, qstat_field)			\
214 	{									\
215 		#name,								\
216 		offsetof(struct virtio_net_stats_tx_ ## class, tx_ ## name),	\
217 		offsetof(struct netdev_queue_stats_tx, qstat_field),		\
218 	}
219 
220 static const struct virtnet_stat_desc virtnet_stats_rx_basic_desc_qstat[] = {
221 	VIRTNET_STATS_DESC_RX_QSTAT(basic, drops,         hw_drops),
222 	VIRTNET_STATS_DESC_RX_QSTAT(basic, drop_overruns, hw_drop_overruns),
223 };
224 
225 static const struct virtnet_stat_desc virtnet_stats_tx_basic_desc_qstat[] = {
226 	VIRTNET_STATS_DESC_TX_QSTAT(basic, drops,          hw_drops),
227 	VIRTNET_STATS_DESC_TX_QSTAT(basic, drop_malformed, hw_drop_errors),
228 };
229 
230 static const struct virtnet_stat_desc virtnet_stats_rx_csum_desc_qstat[] = {
231 	VIRTNET_STATS_DESC_RX_QSTAT(csum, csum_valid, csum_unnecessary),
232 	VIRTNET_STATS_DESC_RX_QSTAT(csum, csum_none,  csum_none),
233 	VIRTNET_STATS_DESC_RX_QSTAT(csum, csum_bad,   csum_bad),
234 };
235 
236 static const struct virtnet_stat_desc virtnet_stats_tx_csum_desc_qstat[] = {
237 	VIRTNET_STATS_DESC_TX_QSTAT(csum, csum_none,  csum_none),
238 	VIRTNET_STATS_DESC_TX_QSTAT(csum, needs_csum, needs_csum),
239 };
240 
241 static const struct virtnet_stat_desc virtnet_stats_rx_gso_desc_qstat[] = {
242 	VIRTNET_STATS_DESC_RX_QSTAT(gso, gso_packets,           hw_gro_packets),
243 	VIRTNET_STATS_DESC_RX_QSTAT(gso, gso_bytes,             hw_gro_bytes),
244 	VIRTNET_STATS_DESC_RX_QSTAT(gso, gso_packets_coalesced, hw_gro_wire_packets),
245 	VIRTNET_STATS_DESC_RX_QSTAT(gso, gso_bytes_coalesced,   hw_gro_wire_bytes),
246 };
247 
248 static const struct virtnet_stat_desc virtnet_stats_tx_gso_desc_qstat[] = {
249 	VIRTNET_STATS_DESC_TX_QSTAT(gso, gso_packets,        hw_gso_packets),
250 	VIRTNET_STATS_DESC_TX_QSTAT(gso, gso_bytes,          hw_gso_bytes),
251 	VIRTNET_STATS_DESC_TX_QSTAT(gso, gso_segments,       hw_gso_wire_packets),
252 	VIRTNET_STATS_DESC_TX_QSTAT(gso, gso_segments_bytes, hw_gso_wire_bytes),
253 };
254 
255 static const struct virtnet_stat_desc virtnet_stats_rx_speed_desc_qstat[] = {
256 	VIRTNET_STATS_DESC_RX_QSTAT(speed, ratelimit_packets, hw_drop_ratelimits),
257 };
258 
259 static const struct virtnet_stat_desc virtnet_stats_tx_speed_desc_qstat[] = {
260 	VIRTNET_STATS_DESC_TX_QSTAT(speed, ratelimit_packets, hw_drop_ratelimits),
261 };
262 
263 #define VIRTNET_Q_TYPE_RX 0
264 #define VIRTNET_Q_TYPE_TX 1
265 #define VIRTNET_Q_TYPE_CQ 2
266 
267 struct virtnet_interrupt_coalesce {
268 	u32 max_packets;
269 	u32 max_usecs;
270 };
271 
272 /* The dma information of pages allocated at a time. */
273 struct virtnet_rq_dma {
274 	dma_addr_t addr;
275 	u32 ref;
276 	u16 len;
277 	u16 need_sync;
278 };
279 
280 /* Internal representation of a send virtqueue */
281 struct send_queue {
282 	/* Virtqueue associated with this send _queue */
283 	struct virtqueue *vq;
284 
285 	/* TX: fragments + linear part + virtio header */
286 	struct scatterlist sg[MAX_SKB_FRAGS + 2];
287 
288 	/* Name of the send queue: output.$index */
289 	char name[16];
290 
291 	struct virtnet_sq_stats stats;
292 
293 	struct virtnet_interrupt_coalesce intr_coal;
294 
295 	struct napi_struct napi;
296 
297 	/* Record whether sq is in reset state. */
298 	bool reset;
299 
300 	struct xsk_buff_pool *xsk_pool;
301 
302 	dma_addr_t xsk_hdr_dma_addr;
303 };
304 
305 /* Internal representation of a receive virtqueue */
306 struct receive_queue {
307 	/* Virtqueue associated with this receive_queue */
308 	struct virtqueue *vq;
309 
310 	struct napi_struct napi;
311 
312 	struct bpf_prog __rcu *xdp_prog;
313 
314 	struct virtnet_rq_stats stats;
315 
316 	/* The number of rx notifications */
317 	u16 calls;
318 
319 	/* Is dynamic interrupt moderation enabled? */
320 	bool dim_enabled;
321 
322 	/* Used to protect dim_enabled and inter_coal */
323 	struct mutex dim_lock;
324 
325 	/* Dynamic Interrupt Moderation */
326 	struct dim dim;
327 
328 	u32 packets_in_napi;
329 
330 	struct virtnet_interrupt_coalesce intr_coal;
331 
332 	/* Chain pages by the private ptr. */
333 	struct page *pages;
334 
335 	/* Average packet length for mergeable receive buffers. */
336 	struct ewma_pkt_len mrg_avg_pkt_len;
337 
338 	/* Page frag for packet buffer allocation. */
339 	struct page_frag alloc_frag;
340 
341 	/* RX: fragments + linear part + virtio header */
342 	struct scatterlist sg[MAX_SKB_FRAGS + 2];
343 
344 	/* Min single buffer size for mergeable buffers case. */
345 	unsigned int min_buf_len;
346 
347 	/* Name of this receive queue: input.$index */
348 	char name[16];
349 
350 	struct xdp_rxq_info xdp_rxq;
351 
352 	/* Record the last dma info to free after new pages is allocated. */
353 	struct virtnet_rq_dma *last_dma;
354 
355 	struct xsk_buff_pool *xsk_pool;
356 
357 	/* xdp rxq used by xsk */
358 	struct xdp_rxq_info xsk_rxq_info;
359 
360 	struct xdp_buff **xsk_buffs;
361 };
362 
363 #define VIRTIO_NET_RSS_MAX_KEY_SIZE     40
364 
365 /* Control VQ buffers: protected by the rtnl lock */
366 struct control_buf {
367 	struct virtio_net_ctrl_hdr hdr;
368 	virtio_net_ctrl_ack status;
369 };
370 
371 struct virtnet_info {
372 	struct virtio_device *vdev;
373 	struct virtqueue *cvq;
374 	struct net_device *dev;
375 	struct send_queue *sq;
376 	struct receive_queue *rq;
377 	unsigned int status;
378 
379 	/* Max # of queue pairs supported by the device */
380 	u16 max_queue_pairs;
381 
382 	/* # of queue pairs currently used by the driver */
383 	u16 curr_queue_pairs;
384 
385 	/* # of XDP queue pairs currently used by the driver */
386 	u16 xdp_queue_pairs;
387 
388 	/* xdp_queue_pairs may be 0, when xdp is already loaded. So add this. */
389 	bool xdp_enabled;
390 
391 	/* I like... big packets and I cannot lie! */
392 	bool big_packets;
393 
394 	/* number of sg entries allocated for big packets */
395 	unsigned int big_packets_num_skbfrags;
396 
397 	/* Host will merge rx buffers for big packets (shake it! shake it!) */
398 	bool mergeable_rx_bufs;
399 
400 	/* Host supports rss and/or hash report */
401 	bool has_rss;
402 	bool has_rss_hash_report;
403 	u8 rss_key_size;
404 	u16 rss_indir_table_size;
405 	u32 rss_hash_types_supported;
406 	u32 rss_hash_types_saved;
407 	struct virtio_net_rss_config_hdr *rss_hdr;
408 	struct virtio_net_rss_config_trailer rss_trailer;
409 	u8 rss_hash_key_data[VIRTIO_NET_RSS_MAX_KEY_SIZE];
410 
411 	/* Has control virtqueue */
412 	bool has_cvq;
413 
414 	/* Lock to protect the control VQ */
415 	struct mutex cvq_lock;
416 
417 	/* Host can handle any s/g split between our header and packet data */
418 	bool any_header_sg;
419 
420 	/* Packet virtio header size */
421 	u8 hdr_len;
422 
423 	/* Work struct for delayed refilling if we run low on memory. */
424 	struct delayed_work refill;
425 
426 	/* Is delayed refill enabled? */
427 	bool refill_enabled;
428 
429 	/* The lock to synchronize the access to refill_enabled */
430 	spinlock_t refill_lock;
431 
432 	/* Work struct for config space updates */
433 	struct work_struct config_work;
434 
435 	/* Work struct for setting rx mode */
436 	struct work_struct rx_mode_work;
437 
438 	/* OK to queue work setting RX mode? */
439 	bool rx_mode_work_enabled;
440 
441 	/* Does the affinity hint is set for virtqueues? */
442 	bool affinity_hint_set;
443 
444 	/* CPU hotplug instances for online & dead */
445 	struct hlist_node node;
446 	struct hlist_node node_dead;
447 
448 	struct control_buf *ctrl;
449 
450 	/* Ethtool settings */
451 	u8 duplex;
452 	u32 speed;
453 
454 	/* Is rx dynamic interrupt moderation enabled? */
455 	bool rx_dim_enabled;
456 
457 	/* Interrupt coalescing settings */
458 	struct virtnet_interrupt_coalesce intr_coal_tx;
459 	struct virtnet_interrupt_coalesce intr_coal_rx;
460 
461 	unsigned long guest_offloads;
462 	unsigned long guest_offloads_capable;
463 
464 	/* failover when STANDBY feature enabled */
465 	struct failover *failover;
466 
467 	u64 device_stats_cap;
468 };
469 
470 struct padded_vnet_hdr {
471 	struct virtio_net_hdr_v1_hash hdr;
472 	/*
473 	 * hdr is in a separate sg buffer, and data sg buffer shares same page
474 	 * with this header sg. This padding makes next sg 16 byte aligned
475 	 * after the header.
476 	 */
477 	char padding[12];
478 };
479 
480 struct virtio_net_common_hdr {
481 	union {
482 		struct virtio_net_hdr hdr;
483 		struct virtio_net_hdr_mrg_rxbuf	mrg_hdr;
484 		struct virtio_net_hdr_v1_hash hash_v1_hdr;
485 	};
486 };
487 
488 static struct virtio_net_common_hdr xsk_hdr;
489 
490 static void virtnet_sq_free_unused_buf(struct virtqueue *vq, void *buf);
491 static void virtnet_sq_free_unused_buf_done(struct virtqueue *vq);
492 static int virtnet_xdp_handler(struct bpf_prog *xdp_prog, struct xdp_buff *xdp,
493 			       struct net_device *dev,
494 			       unsigned int *xdp_xmit,
495 			       struct virtnet_rq_stats *stats);
496 static void virtnet_receive_done(struct virtnet_info *vi, struct receive_queue *rq,
497 				 struct sk_buff *skb, u8 flags);
498 static struct sk_buff *virtnet_skb_append_frag(struct sk_buff *head_skb,
499 					       struct sk_buff *curr_skb,
500 					       struct page *page, void *buf,
501 					       int len, int truesize);
502 static void virtnet_xsk_completed(struct send_queue *sq, int num);
503 
504 enum virtnet_xmit_type {
505 	VIRTNET_XMIT_TYPE_SKB,
506 	VIRTNET_XMIT_TYPE_SKB_ORPHAN,
507 	VIRTNET_XMIT_TYPE_XDP,
508 	VIRTNET_XMIT_TYPE_XSK,
509 };
510 
511 static size_t virtnet_rss_hdr_size(const struct virtnet_info *vi)
512 {
513 	u16 indir_table_size = vi->has_rss ? vi->rss_indir_table_size : 1;
514 
515 	return struct_size(vi->rss_hdr, indirection_table, indir_table_size);
516 }
517 
518 static size_t virtnet_rss_trailer_size(const struct virtnet_info *vi)
519 {
520 	return struct_size(&vi->rss_trailer, hash_key_data, vi->rss_key_size);
521 }
522 
523 /* We use the last two bits of the pointer to distinguish the xmit type. */
524 #define VIRTNET_XMIT_TYPE_MASK (BIT(0) | BIT(1))
525 
526 #define VIRTIO_XSK_FLAG_OFFSET 2
527 
528 static enum virtnet_xmit_type virtnet_xmit_ptr_unpack(void **ptr)
529 {
530 	unsigned long p = (unsigned long)*ptr;
531 
532 	*ptr = (void *)(p & ~VIRTNET_XMIT_TYPE_MASK);
533 
534 	return p & VIRTNET_XMIT_TYPE_MASK;
535 }
536 
537 static void *virtnet_xmit_ptr_pack(void *ptr, enum virtnet_xmit_type type)
538 {
539 	return (void *)((unsigned long)ptr | type);
540 }
541 
542 static int virtnet_add_outbuf(struct send_queue *sq, int num, void *data,
543 			      enum virtnet_xmit_type type)
544 {
545 	return virtqueue_add_outbuf(sq->vq, sq->sg, num,
546 				    virtnet_xmit_ptr_pack(data, type),
547 				    GFP_ATOMIC);
548 }
549 
550 static u32 virtnet_ptr_to_xsk_buff_len(void *ptr)
551 {
552 	return ((unsigned long)ptr) >> VIRTIO_XSK_FLAG_OFFSET;
553 }
554 
555 static void sg_fill_dma(struct scatterlist *sg, dma_addr_t addr, u32 len)
556 {
557 	sg_dma_address(sg) = addr;
558 	sg_dma_len(sg) = len;
559 }
560 
561 static void __free_old_xmit(struct send_queue *sq, struct netdev_queue *txq,
562 			    bool in_napi, struct virtnet_sq_free_stats *stats)
563 {
564 	struct xdp_frame *frame;
565 	struct sk_buff *skb;
566 	unsigned int len;
567 	void *ptr;
568 
569 	while ((ptr = virtqueue_get_buf(sq->vq, &len)) != NULL) {
570 		switch (virtnet_xmit_ptr_unpack(&ptr)) {
571 		case VIRTNET_XMIT_TYPE_SKB:
572 			skb = ptr;
573 
574 			pr_debug("Sent skb %p\n", skb);
575 			stats->napi_packets++;
576 			stats->napi_bytes += skb->len;
577 			napi_consume_skb(skb, in_napi);
578 			break;
579 
580 		case VIRTNET_XMIT_TYPE_SKB_ORPHAN:
581 			skb = ptr;
582 
583 			stats->packets++;
584 			stats->bytes += skb->len;
585 			napi_consume_skb(skb, in_napi);
586 			break;
587 
588 		case VIRTNET_XMIT_TYPE_XDP:
589 			frame = ptr;
590 
591 			stats->packets++;
592 			stats->bytes += xdp_get_frame_len(frame);
593 			xdp_return_frame(frame);
594 			break;
595 
596 		case VIRTNET_XMIT_TYPE_XSK:
597 			stats->bytes += virtnet_ptr_to_xsk_buff_len(ptr);
598 			stats->xsk++;
599 			break;
600 		}
601 	}
602 	netdev_tx_completed_queue(txq, stats->napi_packets, stats->napi_bytes);
603 }
604 
605 static void virtnet_free_old_xmit(struct send_queue *sq,
606 				  struct netdev_queue *txq,
607 				  bool in_napi,
608 				  struct virtnet_sq_free_stats *stats)
609 {
610 	__free_old_xmit(sq, txq, in_napi, stats);
611 
612 	if (stats->xsk)
613 		virtnet_xsk_completed(sq, stats->xsk);
614 }
615 
616 /* Converting between virtqueue no. and kernel tx/rx queue no.
617  * 0:rx0 1:tx0 2:rx1 3:tx1 ... 2N:rxN 2N+1:txN 2N+2:cvq
618  */
619 static int vq2txq(struct virtqueue *vq)
620 {
621 	return (vq->index - 1) / 2;
622 }
623 
624 static int txq2vq(int txq)
625 {
626 	return txq * 2 + 1;
627 }
628 
629 static int vq2rxq(struct virtqueue *vq)
630 {
631 	return vq->index / 2;
632 }
633 
634 static int rxq2vq(int rxq)
635 {
636 	return rxq * 2;
637 }
638 
639 static int vq_type(struct virtnet_info *vi, int qid)
640 {
641 	if (qid == vi->max_queue_pairs * 2)
642 		return VIRTNET_Q_TYPE_CQ;
643 
644 	if (qid % 2)
645 		return VIRTNET_Q_TYPE_TX;
646 
647 	return VIRTNET_Q_TYPE_RX;
648 }
649 
650 static inline struct virtio_net_common_hdr *
651 skb_vnet_common_hdr(struct sk_buff *skb)
652 {
653 	return (struct virtio_net_common_hdr *)skb->cb;
654 }
655 
656 /*
657  * private is used to chain pages for big packets, put the whole
658  * most recent used list in the beginning for reuse
659  */
660 static void give_pages(struct receive_queue *rq, struct page *page)
661 {
662 	struct page *end;
663 
664 	/* Find end of list, sew whole thing into vi->rq.pages. */
665 	for (end = page; end->private; end = (struct page *)end->private);
666 	end->private = (unsigned long)rq->pages;
667 	rq->pages = page;
668 }
669 
670 static struct page *get_a_page(struct receive_queue *rq, gfp_t gfp_mask)
671 {
672 	struct page *p = rq->pages;
673 
674 	if (p) {
675 		rq->pages = (struct page *)p->private;
676 		/* clear private here, it is used to chain pages */
677 		p->private = 0;
678 	} else
679 		p = alloc_page(gfp_mask);
680 	return p;
681 }
682 
683 static void virtnet_rq_free_buf(struct virtnet_info *vi,
684 				struct receive_queue *rq, void *buf)
685 {
686 	if (vi->mergeable_rx_bufs)
687 		put_page(virt_to_head_page(buf));
688 	else if (vi->big_packets)
689 		give_pages(rq, buf);
690 	else
691 		put_page(virt_to_head_page(buf));
692 }
693 
694 static void enable_delayed_refill(struct virtnet_info *vi)
695 {
696 	spin_lock_bh(&vi->refill_lock);
697 	vi->refill_enabled = true;
698 	spin_unlock_bh(&vi->refill_lock);
699 }
700 
701 static void disable_delayed_refill(struct virtnet_info *vi)
702 {
703 	spin_lock_bh(&vi->refill_lock);
704 	vi->refill_enabled = false;
705 	spin_unlock_bh(&vi->refill_lock);
706 }
707 
708 static void enable_rx_mode_work(struct virtnet_info *vi)
709 {
710 	rtnl_lock();
711 	vi->rx_mode_work_enabled = true;
712 	rtnl_unlock();
713 }
714 
715 static void disable_rx_mode_work(struct virtnet_info *vi)
716 {
717 	rtnl_lock();
718 	vi->rx_mode_work_enabled = false;
719 	rtnl_unlock();
720 }
721 
722 static void virtqueue_napi_schedule(struct napi_struct *napi,
723 				    struct virtqueue *vq)
724 {
725 	if (napi_schedule_prep(napi)) {
726 		virtqueue_disable_cb(vq);
727 		__napi_schedule(napi);
728 	}
729 }
730 
731 static bool virtqueue_napi_complete(struct napi_struct *napi,
732 				    struct virtqueue *vq, int processed)
733 {
734 	int opaque;
735 
736 	opaque = virtqueue_enable_cb_prepare(vq);
737 	if (napi_complete_done(napi, processed)) {
738 		if (unlikely(virtqueue_poll(vq, opaque)))
739 			virtqueue_napi_schedule(napi, vq);
740 		else
741 			return true;
742 	} else {
743 		virtqueue_disable_cb(vq);
744 	}
745 
746 	return false;
747 }
748 
749 static void skb_xmit_done(struct virtqueue *vq)
750 {
751 	struct virtnet_info *vi = vq->vdev->priv;
752 	struct napi_struct *napi = &vi->sq[vq2txq(vq)].napi;
753 
754 	/* Suppress further interrupts. */
755 	virtqueue_disable_cb(vq);
756 
757 	if (napi->weight)
758 		virtqueue_napi_schedule(napi, vq);
759 	else
760 		/* We were probably waiting for more output buffers. */
761 		netif_wake_subqueue(vi->dev, vq2txq(vq));
762 }
763 
764 #define MRG_CTX_HEADER_SHIFT 22
765 static void *mergeable_len_to_ctx(unsigned int truesize,
766 				  unsigned int headroom)
767 {
768 	return (void *)(unsigned long)((headroom << MRG_CTX_HEADER_SHIFT) | truesize);
769 }
770 
771 static unsigned int mergeable_ctx_to_headroom(void *mrg_ctx)
772 {
773 	return (unsigned long)mrg_ctx >> MRG_CTX_HEADER_SHIFT;
774 }
775 
776 static unsigned int mergeable_ctx_to_truesize(void *mrg_ctx)
777 {
778 	return (unsigned long)mrg_ctx & ((1 << MRG_CTX_HEADER_SHIFT) - 1);
779 }
780 
781 static struct sk_buff *virtnet_build_skb(void *buf, unsigned int buflen,
782 					 unsigned int headroom,
783 					 unsigned int len)
784 {
785 	struct sk_buff *skb;
786 
787 	skb = build_skb(buf, buflen);
788 	if (unlikely(!skb))
789 		return NULL;
790 
791 	skb_reserve(skb, headroom);
792 	skb_put(skb, len);
793 
794 	return skb;
795 }
796 
797 /* Called from bottom half context */
798 static struct sk_buff *page_to_skb(struct virtnet_info *vi,
799 				   struct receive_queue *rq,
800 				   struct page *page, unsigned int offset,
801 				   unsigned int len, unsigned int truesize,
802 				   unsigned int headroom)
803 {
804 	struct sk_buff *skb;
805 	struct virtio_net_common_hdr *hdr;
806 	unsigned int copy, hdr_len, hdr_padded_len;
807 	struct page *page_to_free = NULL;
808 	int tailroom, shinfo_size;
809 	char *p, *hdr_p, *buf;
810 
811 	p = page_address(page) + offset;
812 	hdr_p = p;
813 
814 	hdr_len = vi->hdr_len;
815 	if (vi->mergeable_rx_bufs)
816 		hdr_padded_len = hdr_len;
817 	else
818 		hdr_padded_len = sizeof(struct padded_vnet_hdr);
819 
820 	buf = p - headroom;
821 	len -= hdr_len;
822 	offset += hdr_padded_len;
823 	p += hdr_padded_len;
824 	tailroom = truesize - headroom  - hdr_padded_len - len;
825 
826 	shinfo_size = SKB_DATA_ALIGN(sizeof(struct skb_shared_info));
827 
828 	if (!NET_IP_ALIGN && len > GOOD_COPY_LEN && tailroom >= shinfo_size) {
829 		skb = virtnet_build_skb(buf, truesize, p - buf, len);
830 		if (unlikely(!skb))
831 			return NULL;
832 
833 		page = (struct page *)page->private;
834 		if (page)
835 			give_pages(rq, page);
836 		goto ok;
837 	}
838 
839 	/* copy small packet so we can reuse these pages for small data */
840 	skb = napi_alloc_skb(&rq->napi, GOOD_COPY_LEN);
841 	if (unlikely(!skb))
842 		return NULL;
843 
844 	/* Copy all frame if it fits skb->head, otherwise
845 	 * we let virtio_net_hdr_to_skb() and GRO pull headers as needed.
846 	 */
847 	if (len <= skb_tailroom(skb))
848 		copy = len;
849 	else
850 		copy = ETH_HLEN;
851 	skb_put_data(skb, p, copy);
852 
853 	len -= copy;
854 	offset += copy;
855 
856 	if (vi->mergeable_rx_bufs) {
857 		if (len)
858 			skb_add_rx_frag(skb, 0, page, offset, len, truesize);
859 		else
860 			page_to_free = page;
861 		goto ok;
862 	}
863 
864 	/*
865 	 * Verify that we can indeed put this data into a skb.
866 	 * This is here to handle cases when the device erroneously
867 	 * tries to receive more than is possible. This is usually
868 	 * the case of a broken device.
869 	 */
870 	if (unlikely(len > MAX_SKB_FRAGS * PAGE_SIZE)) {
871 		net_dbg_ratelimited("%s: too much data\n", skb->dev->name);
872 		dev_kfree_skb(skb);
873 		return NULL;
874 	}
875 	BUG_ON(offset >= PAGE_SIZE);
876 	while (len) {
877 		unsigned int frag_size = min((unsigned)PAGE_SIZE - offset, len);
878 		skb_add_rx_frag(skb, skb_shinfo(skb)->nr_frags, page, offset,
879 				frag_size, truesize);
880 		len -= frag_size;
881 		page = (struct page *)page->private;
882 		offset = 0;
883 	}
884 
885 	if (page)
886 		give_pages(rq, page);
887 
888 ok:
889 	hdr = skb_vnet_common_hdr(skb);
890 	memcpy(hdr, hdr_p, hdr_len);
891 	if (page_to_free)
892 		put_page(page_to_free);
893 
894 	return skb;
895 }
896 
897 static void virtnet_rq_unmap(struct receive_queue *rq, void *buf, u32 len)
898 {
899 	struct virtnet_info *vi = rq->vq->vdev->priv;
900 	struct page *page = virt_to_head_page(buf);
901 	struct virtnet_rq_dma *dma;
902 	void *head;
903 	int offset;
904 
905 	BUG_ON(vi->big_packets && !vi->mergeable_rx_bufs);
906 
907 	head = page_address(page);
908 
909 	dma = head;
910 
911 	--dma->ref;
912 
913 	if (dma->need_sync && len) {
914 		offset = buf - (head + sizeof(*dma));
915 
916 		virtqueue_dma_sync_single_range_for_cpu(rq->vq, dma->addr,
917 							offset, len,
918 							DMA_FROM_DEVICE);
919 	}
920 
921 	if (dma->ref)
922 		return;
923 
924 	virtqueue_dma_unmap_single_attrs(rq->vq, dma->addr, dma->len,
925 					 DMA_FROM_DEVICE, DMA_ATTR_SKIP_CPU_SYNC);
926 	put_page(page);
927 }
928 
929 static void *virtnet_rq_get_buf(struct receive_queue *rq, u32 *len, void **ctx)
930 {
931 	struct virtnet_info *vi = rq->vq->vdev->priv;
932 	void *buf;
933 
934 	BUG_ON(vi->big_packets && !vi->mergeable_rx_bufs);
935 
936 	buf = virtqueue_get_buf_ctx(rq->vq, len, ctx);
937 	if (buf)
938 		virtnet_rq_unmap(rq, buf, *len);
939 
940 	return buf;
941 }
942 
943 static void virtnet_rq_init_one_sg(struct receive_queue *rq, void *buf, u32 len)
944 {
945 	struct virtnet_info *vi = rq->vq->vdev->priv;
946 	struct virtnet_rq_dma *dma;
947 	dma_addr_t addr;
948 	u32 offset;
949 	void *head;
950 
951 	BUG_ON(vi->big_packets && !vi->mergeable_rx_bufs);
952 
953 	head = page_address(rq->alloc_frag.page);
954 
955 	offset = buf - head;
956 
957 	dma = head;
958 
959 	addr = dma->addr - sizeof(*dma) + offset;
960 
961 	sg_init_table(rq->sg, 1);
962 	sg_fill_dma(rq->sg, addr, len);
963 }
964 
965 static void *virtnet_rq_alloc(struct receive_queue *rq, u32 size, gfp_t gfp)
966 {
967 	struct page_frag *alloc_frag = &rq->alloc_frag;
968 	struct virtnet_info *vi = rq->vq->vdev->priv;
969 	struct virtnet_rq_dma *dma;
970 	void *buf, *head;
971 	dma_addr_t addr;
972 
973 	BUG_ON(vi->big_packets && !vi->mergeable_rx_bufs);
974 
975 	head = page_address(alloc_frag->page);
976 
977 	dma = head;
978 
979 	/* new pages */
980 	if (!alloc_frag->offset) {
981 		if (rq->last_dma) {
982 			/* Now, the new page is allocated, the last dma
983 			 * will not be used. So the dma can be unmapped
984 			 * if the ref is 0.
985 			 */
986 			virtnet_rq_unmap(rq, rq->last_dma, 0);
987 			rq->last_dma = NULL;
988 		}
989 
990 		dma->len = alloc_frag->size - sizeof(*dma);
991 
992 		addr = virtqueue_dma_map_single_attrs(rq->vq, dma + 1,
993 						      dma->len, DMA_FROM_DEVICE, 0);
994 		if (virtqueue_dma_mapping_error(rq->vq, addr))
995 			return NULL;
996 
997 		dma->addr = addr;
998 		dma->need_sync = virtqueue_dma_need_sync(rq->vq, addr);
999 
1000 		/* Add a reference to dma to prevent the entire dma from
1001 		 * being released during error handling. This reference
1002 		 * will be freed after the pages are no longer used.
1003 		 */
1004 		get_page(alloc_frag->page);
1005 		dma->ref = 1;
1006 		alloc_frag->offset = sizeof(*dma);
1007 
1008 		rq->last_dma = dma;
1009 	}
1010 
1011 	++dma->ref;
1012 
1013 	buf = head + alloc_frag->offset;
1014 
1015 	get_page(alloc_frag->page);
1016 	alloc_frag->offset += size;
1017 
1018 	return buf;
1019 }
1020 
1021 static void virtnet_rq_unmap_free_buf(struct virtqueue *vq, void *buf)
1022 {
1023 	struct virtnet_info *vi = vq->vdev->priv;
1024 	struct receive_queue *rq;
1025 	int i = vq2rxq(vq);
1026 
1027 	rq = &vi->rq[i];
1028 
1029 	if (rq->xsk_pool) {
1030 		xsk_buff_free((struct xdp_buff *)buf);
1031 		return;
1032 	}
1033 
1034 	if (!vi->big_packets || vi->mergeable_rx_bufs)
1035 		virtnet_rq_unmap(rq, buf, 0);
1036 
1037 	virtnet_rq_free_buf(vi, rq, buf);
1038 }
1039 
1040 static void free_old_xmit(struct send_queue *sq, struct netdev_queue *txq,
1041 			  bool in_napi)
1042 {
1043 	struct virtnet_sq_free_stats stats = {0};
1044 
1045 	virtnet_free_old_xmit(sq, txq, in_napi, &stats);
1046 
1047 	/* Avoid overhead when no packets have been processed
1048 	 * happens when called speculatively from start_xmit.
1049 	 */
1050 	if (!stats.packets && !stats.napi_packets)
1051 		return;
1052 
1053 	u64_stats_update_begin(&sq->stats.syncp);
1054 	u64_stats_add(&sq->stats.bytes, stats.bytes + stats.napi_bytes);
1055 	u64_stats_add(&sq->stats.packets, stats.packets + stats.napi_packets);
1056 	u64_stats_update_end(&sq->stats.syncp);
1057 }
1058 
1059 static bool is_xdp_raw_buffer_queue(struct virtnet_info *vi, int q)
1060 {
1061 	if (q < (vi->curr_queue_pairs - vi->xdp_queue_pairs))
1062 		return false;
1063 	else if (q < vi->curr_queue_pairs)
1064 		return true;
1065 	else
1066 		return false;
1067 }
1068 
1069 static bool tx_may_stop(struct virtnet_info *vi,
1070 			struct net_device *dev,
1071 			struct send_queue *sq)
1072 {
1073 	int qnum;
1074 
1075 	qnum = sq - vi->sq;
1076 
1077 	/* If running out of space, stop queue to avoid getting packets that we
1078 	 * are then unable to transmit.
1079 	 * An alternative would be to force queuing layer to requeue the skb by
1080 	 * returning NETDEV_TX_BUSY. However, NETDEV_TX_BUSY should not be
1081 	 * returned in a normal path of operation: it means that driver is not
1082 	 * maintaining the TX queue stop/start state properly, and causes
1083 	 * the stack to do a non-trivial amount of useless work.
1084 	 * Since most packets only take 1 or 2 ring slots, stopping the queue
1085 	 * early means 16 slots are typically wasted.
1086 	 */
1087 	if (sq->vq->num_free < 2+MAX_SKB_FRAGS) {
1088 		struct netdev_queue *txq = netdev_get_tx_queue(dev, qnum);
1089 
1090 		netif_tx_stop_queue(txq);
1091 		u64_stats_update_begin(&sq->stats.syncp);
1092 		u64_stats_inc(&sq->stats.stop);
1093 		u64_stats_update_end(&sq->stats.syncp);
1094 
1095 		return true;
1096 	}
1097 
1098 	return false;
1099 }
1100 
1101 static void check_sq_full_and_disable(struct virtnet_info *vi,
1102 				      struct net_device *dev,
1103 				      struct send_queue *sq)
1104 {
1105 	bool use_napi = sq->napi.weight;
1106 	int qnum;
1107 
1108 	qnum = sq - vi->sq;
1109 
1110 	if (tx_may_stop(vi, dev, sq)) {
1111 		struct netdev_queue *txq = netdev_get_tx_queue(dev, qnum);
1112 
1113 		if (use_napi) {
1114 			if (unlikely(!virtqueue_enable_cb_delayed(sq->vq)))
1115 				virtqueue_napi_schedule(&sq->napi, sq->vq);
1116 		} else if (unlikely(!virtqueue_enable_cb_delayed(sq->vq))) {
1117 			/* More just got used, free them then recheck. */
1118 			free_old_xmit(sq, txq, false);
1119 			if (sq->vq->num_free >= 2+MAX_SKB_FRAGS) {
1120 				netif_start_subqueue(dev, qnum);
1121 				u64_stats_update_begin(&sq->stats.syncp);
1122 				u64_stats_inc(&sq->stats.wake);
1123 				u64_stats_update_end(&sq->stats.syncp);
1124 				virtqueue_disable_cb(sq->vq);
1125 			}
1126 		}
1127 	}
1128 }
1129 
1130 static struct xdp_buff *buf_to_xdp(struct virtnet_info *vi,
1131 				   struct receive_queue *rq, void *buf, u32 len)
1132 {
1133 	struct xdp_buff *xdp;
1134 	u32 bufsize;
1135 
1136 	xdp = (struct xdp_buff *)buf;
1137 
1138 	bufsize = xsk_pool_get_rx_frame_size(rq->xsk_pool) + vi->hdr_len;
1139 
1140 	if (unlikely(len > bufsize)) {
1141 		pr_debug("%s: rx error: len %u exceeds truesize %u\n",
1142 			 vi->dev->name, len, bufsize);
1143 		DEV_STATS_INC(vi->dev, rx_length_errors);
1144 		xsk_buff_free(xdp);
1145 		return NULL;
1146 	}
1147 
1148 	xsk_buff_set_size(xdp, len);
1149 	xsk_buff_dma_sync_for_cpu(xdp);
1150 
1151 	return xdp;
1152 }
1153 
1154 static struct sk_buff *xsk_construct_skb(struct receive_queue *rq,
1155 					 struct xdp_buff *xdp)
1156 {
1157 	unsigned int metasize = xdp->data - xdp->data_meta;
1158 	struct sk_buff *skb;
1159 	unsigned int size;
1160 
1161 	size = xdp->data_end - xdp->data_hard_start;
1162 	skb = napi_alloc_skb(&rq->napi, size);
1163 	if (unlikely(!skb)) {
1164 		xsk_buff_free(xdp);
1165 		return NULL;
1166 	}
1167 
1168 	skb_reserve(skb, xdp->data_meta - xdp->data_hard_start);
1169 
1170 	size = xdp->data_end - xdp->data_meta;
1171 	memcpy(__skb_put(skb, size), xdp->data_meta, size);
1172 
1173 	if (metasize) {
1174 		__skb_pull(skb, metasize);
1175 		skb_metadata_set(skb, metasize);
1176 	}
1177 
1178 	xsk_buff_free(xdp);
1179 
1180 	return skb;
1181 }
1182 
1183 static struct sk_buff *virtnet_receive_xsk_small(struct net_device *dev, struct virtnet_info *vi,
1184 						 struct receive_queue *rq, struct xdp_buff *xdp,
1185 						 unsigned int *xdp_xmit,
1186 						 struct virtnet_rq_stats *stats)
1187 {
1188 	struct bpf_prog *prog;
1189 	u32 ret;
1190 
1191 	ret = XDP_PASS;
1192 	rcu_read_lock();
1193 	prog = rcu_dereference(rq->xdp_prog);
1194 	if (prog)
1195 		ret = virtnet_xdp_handler(prog, xdp, dev, xdp_xmit, stats);
1196 	rcu_read_unlock();
1197 
1198 	switch (ret) {
1199 	case XDP_PASS:
1200 		return xsk_construct_skb(rq, xdp);
1201 
1202 	case XDP_TX:
1203 	case XDP_REDIRECT:
1204 		return NULL;
1205 
1206 	default:
1207 		/* drop packet */
1208 		xsk_buff_free(xdp);
1209 		u64_stats_inc(&stats->drops);
1210 		return NULL;
1211 	}
1212 }
1213 
1214 static void xsk_drop_follow_bufs(struct net_device *dev,
1215 				 struct receive_queue *rq,
1216 				 u32 num_buf,
1217 				 struct virtnet_rq_stats *stats)
1218 {
1219 	struct xdp_buff *xdp;
1220 	u32 len;
1221 
1222 	while (num_buf-- > 1) {
1223 		xdp = virtqueue_get_buf(rq->vq, &len);
1224 		if (unlikely(!xdp)) {
1225 			pr_debug("%s: rx error: %d buffers missing\n",
1226 				 dev->name, num_buf);
1227 			DEV_STATS_INC(dev, rx_length_errors);
1228 			break;
1229 		}
1230 		u64_stats_add(&stats->bytes, len);
1231 		xsk_buff_free(xdp);
1232 	}
1233 }
1234 
1235 static int xsk_append_merge_buffer(struct virtnet_info *vi,
1236 				   struct receive_queue *rq,
1237 				   struct sk_buff *head_skb,
1238 				   u32 num_buf,
1239 				   struct virtio_net_hdr_mrg_rxbuf *hdr,
1240 				   struct virtnet_rq_stats *stats)
1241 {
1242 	struct sk_buff *curr_skb;
1243 	struct xdp_buff *xdp;
1244 	u32 len, truesize;
1245 	struct page *page;
1246 	void *buf;
1247 
1248 	curr_skb = head_skb;
1249 
1250 	while (--num_buf) {
1251 		buf = virtqueue_get_buf(rq->vq, &len);
1252 		if (unlikely(!buf)) {
1253 			pr_debug("%s: rx error: %d buffers out of %d missing\n",
1254 				 vi->dev->name, num_buf,
1255 				 virtio16_to_cpu(vi->vdev,
1256 						 hdr->num_buffers));
1257 			DEV_STATS_INC(vi->dev, rx_length_errors);
1258 			return -EINVAL;
1259 		}
1260 
1261 		u64_stats_add(&stats->bytes, len);
1262 
1263 		xdp = buf_to_xdp(vi, rq, buf, len);
1264 		if (!xdp)
1265 			goto err;
1266 
1267 		buf = napi_alloc_frag(len);
1268 		if (!buf) {
1269 			xsk_buff_free(xdp);
1270 			goto err;
1271 		}
1272 
1273 		memcpy(buf, xdp->data - vi->hdr_len, len);
1274 
1275 		xsk_buff_free(xdp);
1276 
1277 		page = virt_to_page(buf);
1278 
1279 		truesize = len;
1280 
1281 		curr_skb  = virtnet_skb_append_frag(head_skb, curr_skb, page,
1282 						    buf, len, truesize);
1283 		if (!curr_skb) {
1284 			put_page(page);
1285 			goto err;
1286 		}
1287 	}
1288 
1289 	return 0;
1290 
1291 err:
1292 	xsk_drop_follow_bufs(vi->dev, rq, num_buf, stats);
1293 	return -EINVAL;
1294 }
1295 
1296 static struct sk_buff *virtnet_receive_xsk_merge(struct net_device *dev, struct virtnet_info *vi,
1297 						 struct receive_queue *rq, struct xdp_buff *xdp,
1298 						 unsigned int *xdp_xmit,
1299 						 struct virtnet_rq_stats *stats)
1300 {
1301 	struct virtio_net_hdr_mrg_rxbuf *hdr;
1302 	struct bpf_prog *prog;
1303 	struct sk_buff *skb;
1304 	u32 ret, num_buf;
1305 
1306 	hdr = xdp->data - vi->hdr_len;
1307 	num_buf = virtio16_to_cpu(vi->vdev, hdr->num_buffers);
1308 
1309 	ret = XDP_PASS;
1310 	rcu_read_lock();
1311 	prog = rcu_dereference(rq->xdp_prog);
1312 	/* TODO: support multi buffer. */
1313 	if (prog && num_buf == 1)
1314 		ret = virtnet_xdp_handler(prog, xdp, dev, xdp_xmit, stats);
1315 	rcu_read_unlock();
1316 
1317 	switch (ret) {
1318 	case XDP_PASS:
1319 		skb = xsk_construct_skb(rq, xdp);
1320 		if (!skb)
1321 			goto drop_bufs;
1322 
1323 		if (xsk_append_merge_buffer(vi, rq, skb, num_buf, hdr, stats)) {
1324 			dev_kfree_skb(skb);
1325 			goto drop;
1326 		}
1327 
1328 		return skb;
1329 
1330 	case XDP_TX:
1331 	case XDP_REDIRECT:
1332 		return NULL;
1333 
1334 	default:
1335 		/* drop packet */
1336 		xsk_buff_free(xdp);
1337 	}
1338 
1339 drop_bufs:
1340 	xsk_drop_follow_bufs(dev, rq, num_buf, stats);
1341 
1342 drop:
1343 	u64_stats_inc(&stats->drops);
1344 	return NULL;
1345 }
1346 
1347 static void virtnet_receive_xsk_buf(struct virtnet_info *vi, struct receive_queue *rq,
1348 				    void *buf, u32 len,
1349 				    unsigned int *xdp_xmit,
1350 				    struct virtnet_rq_stats *stats)
1351 {
1352 	struct net_device *dev = vi->dev;
1353 	struct sk_buff *skb = NULL;
1354 	struct xdp_buff *xdp;
1355 	u8 flags;
1356 
1357 	len -= vi->hdr_len;
1358 
1359 	u64_stats_add(&stats->bytes, len);
1360 
1361 	xdp = buf_to_xdp(vi, rq, buf, len);
1362 	if (!xdp)
1363 		return;
1364 
1365 	if (unlikely(len < ETH_HLEN)) {
1366 		pr_debug("%s: short packet %i\n", dev->name, len);
1367 		DEV_STATS_INC(dev, rx_length_errors);
1368 		xsk_buff_free(xdp);
1369 		return;
1370 	}
1371 
1372 	flags = ((struct virtio_net_common_hdr *)(xdp->data - vi->hdr_len))->hdr.flags;
1373 
1374 	if (!vi->mergeable_rx_bufs)
1375 		skb = virtnet_receive_xsk_small(dev, vi, rq, xdp, xdp_xmit, stats);
1376 	else
1377 		skb = virtnet_receive_xsk_merge(dev, vi, rq, xdp, xdp_xmit, stats);
1378 
1379 	if (skb)
1380 		virtnet_receive_done(vi, rq, skb, flags);
1381 }
1382 
1383 static int virtnet_add_recvbuf_xsk(struct virtnet_info *vi, struct receive_queue *rq,
1384 				   struct xsk_buff_pool *pool, gfp_t gfp)
1385 {
1386 	struct xdp_buff **xsk_buffs;
1387 	dma_addr_t addr;
1388 	int err = 0;
1389 	u32 len, i;
1390 	int num;
1391 
1392 	xsk_buffs = rq->xsk_buffs;
1393 
1394 	num = xsk_buff_alloc_batch(pool, xsk_buffs, rq->vq->num_free);
1395 	if (!num)
1396 		return -ENOMEM;
1397 
1398 	len = xsk_pool_get_rx_frame_size(pool) + vi->hdr_len;
1399 
1400 	for (i = 0; i < num; ++i) {
1401 		/* Use the part of XDP_PACKET_HEADROOM as the virtnet hdr space.
1402 		 * We assume XDP_PACKET_HEADROOM is larger than hdr->len.
1403 		 * (see function virtnet_xsk_pool_enable)
1404 		 */
1405 		addr = xsk_buff_xdp_get_dma(xsk_buffs[i]) - vi->hdr_len;
1406 
1407 		sg_init_table(rq->sg, 1);
1408 		sg_fill_dma(rq->sg, addr, len);
1409 
1410 		err = virtqueue_add_inbuf_premapped(rq->vq, rq->sg, 1,
1411 						    xsk_buffs[i], NULL, gfp);
1412 		if (err)
1413 			goto err;
1414 	}
1415 
1416 	return num;
1417 
1418 err:
1419 	for (; i < num; ++i)
1420 		xsk_buff_free(xsk_buffs[i]);
1421 
1422 	return err;
1423 }
1424 
1425 static void *virtnet_xsk_to_ptr(u32 len)
1426 {
1427 	unsigned long p;
1428 
1429 	p = len << VIRTIO_XSK_FLAG_OFFSET;
1430 
1431 	return virtnet_xmit_ptr_pack((void *)p, VIRTNET_XMIT_TYPE_XSK);
1432 }
1433 
1434 static int virtnet_xsk_xmit_one(struct send_queue *sq,
1435 				struct xsk_buff_pool *pool,
1436 				struct xdp_desc *desc)
1437 {
1438 	struct virtnet_info *vi;
1439 	dma_addr_t addr;
1440 
1441 	vi = sq->vq->vdev->priv;
1442 
1443 	addr = xsk_buff_raw_get_dma(pool, desc->addr);
1444 	xsk_buff_raw_dma_sync_for_device(pool, addr, desc->len);
1445 
1446 	sg_init_table(sq->sg, 2);
1447 	sg_fill_dma(sq->sg, sq->xsk_hdr_dma_addr, vi->hdr_len);
1448 	sg_fill_dma(sq->sg + 1, addr, desc->len);
1449 
1450 	return virtqueue_add_outbuf_premapped(sq->vq, sq->sg, 2,
1451 					      virtnet_xsk_to_ptr(desc->len),
1452 					      GFP_ATOMIC);
1453 }
1454 
1455 static int virtnet_xsk_xmit_batch(struct send_queue *sq,
1456 				  struct xsk_buff_pool *pool,
1457 				  unsigned int budget,
1458 				  u64 *kicks)
1459 {
1460 	struct xdp_desc *descs = pool->tx_descs;
1461 	bool kick = false;
1462 	u32 nb_pkts, i;
1463 	int err;
1464 
1465 	budget = min_t(u32, budget, sq->vq->num_free);
1466 
1467 	nb_pkts = xsk_tx_peek_release_desc_batch(pool, budget);
1468 	if (!nb_pkts)
1469 		return 0;
1470 
1471 	for (i = 0; i < nb_pkts; i++) {
1472 		err = virtnet_xsk_xmit_one(sq, pool, &descs[i]);
1473 		if (unlikely(err)) {
1474 			xsk_tx_completed(sq->xsk_pool, nb_pkts - i);
1475 			break;
1476 		}
1477 
1478 		kick = true;
1479 	}
1480 
1481 	if (kick && virtqueue_kick_prepare(sq->vq) && virtqueue_notify(sq->vq))
1482 		(*kicks)++;
1483 
1484 	return i;
1485 }
1486 
1487 static bool virtnet_xsk_xmit(struct send_queue *sq, struct xsk_buff_pool *pool,
1488 			     int budget)
1489 {
1490 	struct virtnet_info *vi = sq->vq->vdev->priv;
1491 	struct virtnet_sq_free_stats stats = {};
1492 	struct net_device *dev = vi->dev;
1493 	u64 kicks = 0;
1494 	int sent;
1495 
1496 	/* Avoid to wakeup napi meanless, so call __free_old_xmit instead of
1497 	 * free_old_xmit().
1498 	 */
1499 	__free_old_xmit(sq, netdev_get_tx_queue(dev, sq - vi->sq), true, &stats);
1500 
1501 	if (stats.xsk)
1502 		xsk_tx_completed(sq->xsk_pool, stats.xsk);
1503 
1504 	sent = virtnet_xsk_xmit_batch(sq, pool, budget, &kicks);
1505 
1506 	if (!is_xdp_raw_buffer_queue(vi, sq - vi->sq))
1507 		check_sq_full_and_disable(vi, vi->dev, sq);
1508 
1509 	if (sent) {
1510 		struct netdev_queue *txq;
1511 
1512 		txq = netdev_get_tx_queue(vi->dev, sq - vi->sq);
1513 		txq_trans_cond_update(txq);
1514 	}
1515 
1516 	u64_stats_update_begin(&sq->stats.syncp);
1517 	u64_stats_add(&sq->stats.packets, stats.packets);
1518 	u64_stats_add(&sq->stats.bytes,   stats.bytes);
1519 	u64_stats_add(&sq->stats.kicks,   kicks);
1520 	u64_stats_add(&sq->stats.xdp_tx,  sent);
1521 	u64_stats_update_end(&sq->stats.syncp);
1522 
1523 	if (xsk_uses_need_wakeup(pool))
1524 		xsk_set_tx_need_wakeup(pool);
1525 
1526 	return sent;
1527 }
1528 
1529 static void xsk_wakeup(struct send_queue *sq)
1530 {
1531 	if (napi_if_scheduled_mark_missed(&sq->napi))
1532 		return;
1533 
1534 	local_bh_disable();
1535 	virtqueue_napi_schedule(&sq->napi, sq->vq);
1536 	local_bh_enable();
1537 }
1538 
1539 static int virtnet_xsk_wakeup(struct net_device *dev, u32 qid, u32 flag)
1540 {
1541 	struct virtnet_info *vi = netdev_priv(dev);
1542 	struct send_queue *sq;
1543 
1544 	if (!netif_running(dev))
1545 		return -ENETDOWN;
1546 
1547 	if (qid >= vi->curr_queue_pairs)
1548 		return -EINVAL;
1549 
1550 	sq = &vi->sq[qid];
1551 
1552 	xsk_wakeup(sq);
1553 	return 0;
1554 }
1555 
1556 static void virtnet_xsk_completed(struct send_queue *sq, int num)
1557 {
1558 	xsk_tx_completed(sq->xsk_pool, num);
1559 
1560 	/* If this is called by rx poll, start_xmit and xdp xmit we should
1561 	 * wakeup the tx napi to consume the xsk tx queue, because the tx
1562 	 * interrupt may not be triggered.
1563 	 */
1564 	xsk_wakeup(sq);
1565 }
1566 
1567 static int __virtnet_xdp_xmit_one(struct virtnet_info *vi,
1568 				   struct send_queue *sq,
1569 				   struct xdp_frame *xdpf)
1570 {
1571 	struct virtio_net_hdr_mrg_rxbuf *hdr;
1572 	struct skb_shared_info *shinfo;
1573 	u8 nr_frags = 0;
1574 	int err, i;
1575 
1576 	if (unlikely(xdpf->headroom < vi->hdr_len))
1577 		return -EOVERFLOW;
1578 
1579 	if (unlikely(xdp_frame_has_frags(xdpf))) {
1580 		shinfo = xdp_get_shared_info_from_frame(xdpf);
1581 		nr_frags = shinfo->nr_frags;
1582 	}
1583 
1584 	/* In wrapping function virtnet_xdp_xmit(), we need to free
1585 	 * up the pending old buffers, where we need to calculate the
1586 	 * position of skb_shared_info in xdp_get_frame_len() and
1587 	 * xdp_return_frame(), which will involve to xdpf->data and
1588 	 * xdpf->headroom. Therefore, we need to update the value of
1589 	 * headroom synchronously here.
1590 	 */
1591 	xdpf->headroom -= vi->hdr_len;
1592 	xdpf->data -= vi->hdr_len;
1593 	/* Zero header and leave csum up to XDP layers */
1594 	hdr = xdpf->data;
1595 	memset(hdr, 0, vi->hdr_len);
1596 	xdpf->len   += vi->hdr_len;
1597 
1598 	sg_init_table(sq->sg, nr_frags + 1);
1599 	sg_set_buf(sq->sg, xdpf->data, xdpf->len);
1600 	for (i = 0; i < nr_frags; i++) {
1601 		skb_frag_t *frag = &shinfo->frags[i];
1602 
1603 		sg_set_page(&sq->sg[i + 1], skb_frag_page(frag),
1604 			    skb_frag_size(frag), skb_frag_off(frag));
1605 	}
1606 
1607 	err = virtnet_add_outbuf(sq, nr_frags + 1, xdpf, VIRTNET_XMIT_TYPE_XDP);
1608 	if (unlikely(err))
1609 		return -ENOSPC; /* Caller handle free/refcnt */
1610 
1611 	return 0;
1612 }
1613 
1614 /* when vi->curr_queue_pairs > nr_cpu_ids, the txq/sq is only used for xdp tx on
1615  * the current cpu, so it does not need to be locked.
1616  *
1617  * Here we use marco instead of inline functions because we have to deal with
1618  * three issues at the same time: 1. the choice of sq. 2. judge and execute the
1619  * lock/unlock of txq 3. make sparse happy. It is difficult for two inline
1620  * functions to perfectly solve these three problems at the same time.
1621  */
1622 #define virtnet_xdp_get_sq(vi) ({                                       \
1623 	int cpu = smp_processor_id();                                   \
1624 	struct netdev_queue *txq;                                       \
1625 	typeof(vi) v = (vi);                                            \
1626 	unsigned int qp;                                                \
1627 									\
1628 	if (v->curr_queue_pairs > nr_cpu_ids) {                         \
1629 		qp = v->curr_queue_pairs - v->xdp_queue_pairs;          \
1630 		qp += cpu;                                              \
1631 		txq = netdev_get_tx_queue(v->dev, qp);                  \
1632 		__netif_tx_acquire(txq);                                \
1633 	} else {                                                        \
1634 		qp = cpu % v->curr_queue_pairs;                         \
1635 		txq = netdev_get_tx_queue(v->dev, qp);                  \
1636 		__netif_tx_lock(txq, cpu);                              \
1637 	}                                                               \
1638 	v->sq + qp;                                                     \
1639 })
1640 
1641 #define virtnet_xdp_put_sq(vi, q) {                                     \
1642 	struct netdev_queue *txq;                                       \
1643 	typeof(vi) v = (vi);                                            \
1644 									\
1645 	txq = netdev_get_tx_queue(v->dev, (q) - v->sq);                 \
1646 	if (v->curr_queue_pairs > nr_cpu_ids)                           \
1647 		__netif_tx_release(txq);                                \
1648 	else                                                            \
1649 		__netif_tx_unlock(txq);                                 \
1650 }
1651 
1652 static int virtnet_xdp_xmit(struct net_device *dev,
1653 			    int n, struct xdp_frame **frames, u32 flags)
1654 {
1655 	struct virtnet_info *vi = netdev_priv(dev);
1656 	struct virtnet_sq_free_stats stats = {0};
1657 	struct receive_queue *rq = vi->rq;
1658 	struct bpf_prog *xdp_prog;
1659 	struct send_queue *sq;
1660 	int nxmit = 0;
1661 	int kicks = 0;
1662 	int ret;
1663 	int i;
1664 
1665 	/* Only allow ndo_xdp_xmit if XDP is loaded on dev, as this
1666 	 * indicate XDP resources have been successfully allocated.
1667 	 */
1668 	xdp_prog = rcu_access_pointer(rq->xdp_prog);
1669 	if (!xdp_prog)
1670 		return -ENXIO;
1671 
1672 	sq = virtnet_xdp_get_sq(vi);
1673 
1674 	if (unlikely(flags & ~XDP_XMIT_FLAGS_MASK)) {
1675 		ret = -EINVAL;
1676 		goto out;
1677 	}
1678 
1679 	/* Free up any pending old buffers before queueing new ones. */
1680 	virtnet_free_old_xmit(sq, netdev_get_tx_queue(dev, sq - vi->sq),
1681 			      false, &stats);
1682 
1683 	for (i = 0; i < n; i++) {
1684 		struct xdp_frame *xdpf = frames[i];
1685 
1686 		if (__virtnet_xdp_xmit_one(vi, sq, xdpf))
1687 			break;
1688 		nxmit++;
1689 	}
1690 	ret = nxmit;
1691 
1692 	if (!is_xdp_raw_buffer_queue(vi, sq - vi->sq))
1693 		check_sq_full_and_disable(vi, dev, sq);
1694 
1695 	if (flags & XDP_XMIT_FLUSH) {
1696 		if (virtqueue_kick_prepare(sq->vq) && virtqueue_notify(sq->vq))
1697 			kicks = 1;
1698 	}
1699 out:
1700 	u64_stats_update_begin(&sq->stats.syncp);
1701 	u64_stats_add(&sq->stats.bytes, stats.bytes);
1702 	u64_stats_add(&sq->stats.packets, stats.packets);
1703 	u64_stats_add(&sq->stats.xdp_tx, n);
1704 	u64_stats_add(&sq->stats.xdp_tx_drops, n - nxmit);
1705 	u64_stats_add(&sq->stats.kicks, kicks);
1706 	u64_stats_update_end(&sq->stats.syncp);
1707 
1708 	virtnet_xdp_put_sq(vi, sq);
1709 	return ret;
1710 }
1711 
1712 static void put_xdp_frags(struct xdp_buff *xdp)
1713 {
1714 	struct skb_shared_info *shinfo;
1715 	struct page *xdp_page;
1716 	int i;
1717 
1718 	if (xdp_buff_has_frags(xdp)) {
1719 		shinfo = xdp_get_shared_info_from_buff(xdp);
1720 		for (i = 0; i < shinfo->nr_frags; i++) {
1721 			xdp_page = skb_frag_page(&shinfo->frags[i]);
1722 			put_page(xdp_page);
1723 		}
1724 	}
1725 }
1726 
1727 static int virtnet_xdp_handler(struct bpf_prog *xdp_prog, struct xdp_buff *xdp,
1728 			       struct net_device *dev,
1729 			       unsigned int *xdp_xmit,
1730 			       struct virtnet_rq_stats *stats)
1731 {
1732 	struct xdp_frame *xdpf;
1733 	int err;
1734 	u32 act;
1735 
1736 	act = bpf_prog_run_xdp(xdp_prog, xdp);
1737 	u64_stats_inc(&stats->xdp_packets);
1738 
1739 	switch (act) {
1740 	case XDP_PASS:
1741 		return act;
1742 
1743 	case XDP_TX:
1744 		u64_stats_inc(&stats->xdp_tx);
1745 		xdpf = xdp_convert_buff_to_frame(xdp);
1746 		if (unlikely(!xdpf)) {
1747 			netdev_dbg(dev, "convert buff to frame failed for xdp\n");
1748 			return XDP_DROP;
1749 		}
1750 
1751 		err = virtnet_xdp_xmit(dev, 1, &xdpf, 0);
1752 		if (unlikely(!err)) {
1753 			xdp_return_frame_rx_napi(xdpf);
1754 		} else if (unlikely(err < 0)) {
1755 			trace_xdp_exception(dev, xdp_prog, act);
1756 			return XDP_DROP;
1757 		}
1758 		*xdp_xmit |= VIRTIO_XDP_TX;
1759 		return act;
1760 
1761 	case XDP_REDIRECT:
1762 		u64_stats_inc(&stats->xdp_redirects);
1763 		err = xdp_do_redirect(dev, xdp, xdp_prog);
1764 		if (err)
1765 			return XDP_DROP;
1766 
1767 		*xdp_xmit |= VIRTIO_XDP_REDIR;
1768 		return act;
1769 
1770 	default:
1771 		bpf_warn_invalid_xdp_action(dev, xdp_prog, act);
1772 		fallthrough;
1773 	case XDP_ABORTED:
1774 		trace_xdp_exception(dev, xdp_prog, act);
1775 		fallthrough;
1776 	case XDP_DROP:
1777 		return XDP_DROP;
1778 	}
1779 }
1780 
1781 static unsigned int virtnet_get_headroom(struct virtnet_info *vi)
1782 {
1783 	return vi->xdp_enabled ? XDP_PACKET_HEADROOM : 0;
1784 }
1785 
1786 /* We copy the packet for XDP in the following cases:
1787  *
1788  * 1) Packet is scattered across multiple rx buffers.
1789  * 2) Headroom space is insufficient.
1790  *
1791  * This is inefficient but it's a temporary condition that
1792  * we hit right after XDP is enabled and until queue is refilled
1793  * with large buffers with sufficient headroom - so it should affect
1794  * at most queue size packets.
1795  * Afterwards, the conditions to enable
1796  * XDP should preclude the underlying device from sending packets
1797  * across multiple buffers (num_buf > 1), and we make sure buffers
1798  * have enough headroom.
1799  */
1800 static struct page *xdp_linearize_page(struct receive_queue *rq,
1801 				       int *num_buf,
1802 				       struct page *p,
1803 				       int offset,
1804 				       int page_off,
1805 				       unsigned int *len)
1806 {
1807 	int tailroom = SKB_DATA_ALIGN(sizeof(struct skb_shared_info));
1808 	struct page *page;
1809 
1810 	if (page_off + *len + tailroom > PAGE_SIZE)
1811 		return NULL;
1812 
1813 	page = alloc_page(GFP_ATOMIC);
1814 	if (!page)
1815 		return NULL;
1816 
1817 	memcpy(page_address(page) + page_off, page_address(p) + offset, *len);
1818 	page_off += *len;
1819 
1820 	while (--*num_buf) {
1821 		unsigned int buflen;
1822 		void *buf;
1823 		int off;
1824 
1825 		buf = virtnet_rq_get_buf(rq, &buflen, NULL);
1826 		if (unlikely(!buf))
1827 			goto err_buf;
1828 
1829 		p = virt_to_head_page(buf);
1830 		off = buf - page_address(p);
1831 
1832 		/* guard against a misconfigured or uncooperative backend that
1833 		 * is sending packet larger than the MTU.
1834 		 */
1835 		if ((page_off + buflen + tailroom) > PAGE_SIZE) {
1836 			put_page(p);
1837 			goto err_buf;
1838 		}
1839 
1840 		memcpy(page_address(page) + page_off,
1841 		       page_address(p) + off, buflen);
1842 		page_off += buflen;
1843 		put_page(p);
1844 	}
1845 
1846 	/* Headroom does not contribute to packet length */
1847 	*len = page_off - XDP_PACKET_HEADROOM;
1848 	return page;
1849 err_buf:
1850 	__free_pages(page, 0);
1851 	return NULL;
1852 }
1853 
1854 static struct sk_buff *receive_small_build_skb(struct virtnet_info *vi,
1855 					       unsigned int xdp_headroom,
1856 					       void *buf,
1857 					       unsigned int len)
1858 {
1859 	unsigned int header_offset;
1860 	unsigned int headroom;
1861 	unsigned int buflen;
1862 	struct sk_buff *skb;
1863 
1864 	header_offset = VIRTNET_RX_PAD + xdp_headroom;
1865 	headroom = vi->hdr_len + header_offset;
1866 	buflen = SKB_DATA_ALIGN(GOOD_PACKET_LEN + headroom) +
1867 		SKB_DATA_ALIGN(sizeof(struct skb_shared_info));
1868 
1869 	skb = virtnet_build_skb(buf, buflen, headroom, len);
1870 	if (unlikely(!skb))
1871 		return NULL;
1872 
1873 	buf += header_offset;
1874 	memcpy(skb_vnet_common_hdr(skb), buf, vi->hdr_len);
1875 
1876 	return skb;
1877 }
1878 
1879 static struct sk_buff *receive_small_xdp(struct net_device *dev,
1880 					 struct virtnet_info *vi,
1881 					 struct receive_queue *rq,
1882 					 struct bpf_prog *xdp_prog,
1883 					 void *buf,
1884 					 unsigned int xdp_headroom,
1885 					 unsigned int len,
1886 					 unsigned int *xdp_xmit,
1887 					 struct virtnet_rq_stats *stats)
1888 {
1889 	unsigned int header_offset = VIRTNET_RX_PAD + xdp_headroom;
1890 	unsigned int headroom = vi->hdr_len + header_offset;
1891 	struct virtio_net_hdr_mrg_rxbuf *hdr = buf + header_offset;
1892 	struct page *page = virt_to_head_page(buf);
1893 	struct page *xdp_page;
1894 	unsigned int buflen;
1895 	struct xdp_buff xdp;
1896 	struct sk_buff *skb;
1897 	unsigned int metasize = 0;
1898 	u32 act;
1899 
1900 	if (unlikely(hdr->hdr.gso_type))
1901 		goto err_xdp;
1902 
1903 	/* Partially checksummed packets must be dropped. */
1904 	if (unlikely(hdr->hdr.flags & VIRTIO_NET_HDR_F_NEEDS_CSUM))
1905 		goto err_xdp;
1906 
1907 	buflen = SKB_DATA_ALIGN(GOOD_PACKET_LEN + headroom) +
1908 		SKB_DATA_ALIGN(sizeof(struct skb_shared_info));
1909 
1910 	if (unlikely(xdp_headroom < virtnet_get_headroom(vi))) {
1911 		int offset = buf - page_address(page) + header_offset;
1912 		unsigned int tlen = len + vi->hdr_len;
1913 		int num_buf = 1;
1914 
1915 		xdp_headroom = virtnet_get_headroom(vi);
1916 		header_offset = VIRTNET_RX_PAD + xdp_headroom;
1917 		headroom = vi->hdr_len + header_offset;
1918 		buflen = SKB_DATA_ALIGN(GOOD_PACKET_LEN + headroom) +
1919 			SKB_DATA_ALIGN(sizeof(struct skb_shared_info));
1920 		xdp_page = xdp_linearize_page(rq, &num_buf, page,
1921 					      offset, header_offset,
1922 					      &tlen);
1923 		if (!xdp_page)
1924 			goto err_xdp;
1925 
1926 		buf = page_address(xdp_page);
1927 		put_page(page);
1928 		page = xdp_page;
1929 	}
1930 
1931 	xdp_init_buff(&xdp, buflen, &rq->xdp_rxq);
1932 	xdp_prepare_buff(&xdp, buf + VIRTNET_RX_PAD + vi->hdr_len,
1933 			 xdp_headroom, len, true);
1934 
1935 	act = virtnet_xdp_handler(xdp_prog, &xdp, dev, xdp_xmit, stats);
1936 
1937 	switch (act) {
1938 	case XDP_PASS:
1939 		/* Recalculate length in case bpf program changed it */
1940 		len = xdp.data_end - xdp.data;
1941 		metasize = xdp.data - xdp.data_meta;
1942 		break;
1943 
1944 	case XDP_TX:
1945 	case XDP_REDIRECT:
1946 		goto xdp_xmit;
1947 
1948 	default:
1949 		goto err_xdp;
1950 	}
1951 
1952 	skb = virtnet_build_skb(buf, buflen, xdp.data - buf, len);
1953 	if (unlikely(!skb))
1954 		goto err;
1955 
1956 	if (metasize)
1957 		skb_metadata_set(skb, metasize);
1958 
1959 	return skb;
1960 
1961 err_xdp:
1962 	u64_stats_inc(&stats->xdp_drops);
1963 err:
1964 	u64_stats_inc(&stats->drops);
1965 	put_page(page);
1966 xdp_xmit:
1967 	return NULL;
1968 }
1969 
1970 static struct sk_buff *receive_small(struct net_device *dev,
1971 				     struct virtnet_info *vi,
1972 				     struct receive_queue *rq,
1973 				     void *buf, void *ctx,
1974 				     unsigned int len,
1975 				     unsigned int *xdp_xmit,
1976 				     struct virtnet_rq_stats *stats)
1977 {
1978 	unsigned int xdp_headroom = (unsigned long)ctx;
1979 	struct page *page = virt_to_head_page(buf);
1980 	struct sk_buff *skb;
1981 
1982 	/* We passed the address of virtnet header to virtio-core,
1983 	 * so truncate the padding.
1984 	 */
1985 	buf -= VIRTNET_RX_PAD + xdp_headroom;
1986 
1987 	len -= vi->hdr_len;
1988 	u64_stats_add(&stats->bytes, len);
1989 
1990 	if (unlikely(len > GOOD_PACKET_LEN)) {
1991 		pr_debug("%s: rx error: len %u exceeds max size %d\n",
1992 			 dev->name, len, GOOD_PACKET_LEN);
1993 		DEV_STATS_INC(dev, rx_length_errors);
1994 		goto err;
1995 	}
1996 
1997 	if (unlikely(vi->xdp_enabled)) {
1998 		struct bpf_prog *xdp_prog;
1999 
2000 		rcu_read_lock();
2001 		xdp_prog = rcu_dereference(rq->xdp_prog);
2002 		if (xdp_prog) {
2003 			skb = receive_small_xdp(dev, vi, rq, xdp_prog, buf,
2004 						xdp_headroom, len, xdp_xmit,
2005 						stats);
2006 			rcu_read_unlock();
2007 			return skb;
2008 		}
2009 		rcu_read_unlock();
2010 	}
2011 
2012 	skb = receive_small_build_skb(vi, xdp_headroom, buf, len);
2013 	if (likely(skb))
2014 		return skb;
2015 
2016 err:
2017 	u64_stats_inc(&stats->drops);
2018 	put_page(page);
2019 	return NULL;
2020 }
2021 
2022 static struct sk_buff *receive_big(struct net_device *dev,
2023 				   struct virtnet_info *vi,
2024 				   struct receive_queue *rq,
2025 				   void *buf,
2026 				   unsigned int len,
2027 				   struct virtnet_rq_stats *stats)
2028 {
2029 	struct page *page = buf;
2030 	struct sk_buff *skb =
2031 		page_to_skb(vi, rq, page, 0, len, PAGE_SIZE, 0);
2032 
2033 	u64_stats_add(&stats->bytes, len - vi->hdr_len);
2034 	if (unlikely(!skb))
2035 		goto err;
2036 
2037 	return skb;
2038 
2039 err:
2040 	u64_stats_inc(&stats->drops);
2041 	give_pages(rq, page);
2042 	return NULL;
2043 }
2044 
2045 static void mergeable_buf_free(struct receive_queue *rq, int num_buf,
2046 			       struct net_device *dev,
2047 			       struct virtnet_rq_stats *stats)
2048 {
2049 	struct page *page;
2050 	void *buf;
2051 	int len;
2052 
2053 	while (num_buf-- > 1) {
2054 		buf = virtnet_rq_get_buf(rq, &len, NULL);
2055 		if (unlikely(!buf)) {
2056 			pr_debug("%s: rx error: %d buffers missing\n",
2057 				 dev->name, num_buf);
2058 			DEV_STATS_INC(dev, rx_length_errors);
2059 			break;
2060 		}
2061 		u64_stats_add(&stats->bytes, len);
2062 		page = virt_to_head_page(buf);
2063 		put_page(page);
2064 	}
2065 }
2066 
2067 /* Why not use xdp_build_skb_from_frame() ?
2068  * XDP core assumes that xdp frags are PAGE_SIZE in length, while in
2069  * virtio-net there are 2 points that do not match its requirements:
2070  *  1. The size of the prefilled buffer is not fixed before xdp is set.
2071  *  2. xdp_build_skb_from_frame() does more checks that we don't need,
2072  *     like eth_type_trans() (which virtio-net does in receive_buf()).
2073  */
2074 static struct sk_buff *build_skb_from_xdp_buff(struct net_device *dev,
2075 					       struct virtnet_info *vi,
2076 					       struct xdp_buff *xdp,
2077 					       unsigned int xdp_frags_truesz)
2078 {
2079 	struct skb_shared_info *sinfo = xdp_get_shared_info_from_buff(xdp);
2080 	unsigned int headroom, data_len;
2081 	struct sk_buff *skb;
2082 	int metasize;
2083 	u8 nr_frags;
2084 
2085 	if (unlikely(xdp->data_end > xdp_data_hard_end(xdp))) {
2086 		pr_debug("Error building skb as missing reserved tailroom for xdp");
2087 		return NULL;
2088 	}
2089 
2090 	if (unlikely(xdp_buff_has_frags(xdp)))
2091 		nr_frags = sinfo->nr_frags;
2092 
2093 	skb = build_skb(xdp->data_hard_start, xdp->frame_sz);
2094 	if (unlikely(!skb))
2095 		return NULL;
2096 
2097 	headroom = xdp->data - xdp->data_hard_start;
2098 	data_len = xdp->data_end - xdp->data;
2099 	skb_reserve(skb, headroom);
2100 	__skb_put(skb, data_len);
2101 
2102 	metasize = xdp->data - xdp->data_meta;
2103 	metasize = metasize > 0 ? metasize : 0;
2104 	if (metasize)
2105 		skb_metadata_set(skb, metasize);
2106 
2107 	if (unlikely(xdp_buff_has_frags(xdp)))
2108 		xdp_update_skb_shared_info(skb, nr_frags,
2109 					   sinfo->xdp_frags_size,
2110 					   xdp_frags_truesz,
2111 					   xdp_buff_is_frag_pfmemalloc(xdp));
2112 
2113 	return skb;
2114 }
2115 
2116 /* TODO: build xdp in big mode */
2117 static int virtnet_build_xdp_buff_mrg(struct net_device *dev,
2118 				      struct virtnet_info *vi,
2119 				      struct receive_queue *rq,
2120 				      struct xdp_buff *xdp,
2121 				      void *buf,
2122 				      unsigned int len,
2123 				      unsigned int frame_sz,
2124 				      int *num_buf,
2125 				      unsigned int *xdp_frags_truesize,
2126 				      struct virtnet_rq_stats *stats)
2127 {
2128 	struct virtio_net_hdr_mrg_rxbuf *hdr = buf;
2129 	unsigned int headroom, tailroom, room;
2130 	unsigned int truesize, cur_frag_size;
2131 	struct skb_shared_info *shinfo;
2132 	unsigned int xdp_frags_truesz = 0;
2133 	struct page *page;
2134 	skb_frag_t *frag;
2135 	int offset;
2136 	void *ctx;
2137 
2138 	xdp_init_buff(xdp, frame_sz, &rq->xdp_rxq);
2139 	xdp_prepare_buff(xdp, buf - XDP_PACKET_HEADROOM,
2140 			 XDP_PACKET_HEADROOM + vi->hdr_len, len - vi->hdr_len, true);
2141 
2142 	if (!*num_buf)
2143 		return 0;
2144 
2145 	if (*num_buf > 1) {
2146 		/* If we want to build multi-buffer xdp, we need
2147 		 * to specify that the flags of xdp_buff have the
2148 		 * XDP_FLAGS_HAS_FRAG bit.
2149 		 */
2150 		if (!xdp_buff_has_frags(xdp))
2151 			xdp_buff_set_frags_flag(xdp);
2152 
2153 		shinfo = xdp_get_shared_info_from_buff(xdp);
2154 		shinfo->nr_frags = 0;
2155 		shinfo->xdp_frags_size = 0;
2156 	}
2157 
2158 	if (*num_buf > MAX_SKB_FRAGS + 1)
2159 		return -EINVAL;
2160 
2161 	while (--*num_buf > 0) {
2162 		buf = virtnet_rq_get_buf(rq, &len, &ctx);
2163 		if (unlikely(!buf)) {
2164 			pr_debug("%s: rx error: %d buffers out of %d missing\n",
2165 				 dev->name, *num_buf,
2166 				 virtio16_to_cpu(vi->vdev, hdr->num_buffers));
2167 			DEV_STATS_INC(dev, rx_length_errors);
2168 			goto err;
2169 		}
2170 
2171 		u64_stats_add(&stats->bytes, len);
2172 		page = virt_to_head_page(buf);
2173 		offset = buf - page_address(page);
2174 
2175 		truesize = mergeable_ctx_to_truesize(ctx);
2176 		headroom = mergeable_ctx_to_headroom(ctx);
2177 		tailroom = headroom ? sizeof(struct skb_shared_info) : 0;
2178 		room = SKB_DATA_ALIGN(headroom + tailroom);
2179 
2180 		cur_frag_size = truesize;
2181 		xdp_frags_truesz += cur_frag_size;
2182 		if (unlikely(len > truesize - room || cur_frag_size > PAGE_SIZE)) {
2183 			put_page(page);
2184 			pr_debug("%s: rx error: len %u exceeds truesize %lu\n",
2185 				 dev->name, len, (unsigned long)(truesize - room));
2186 			DEV_STATS_INC(dev, rx_length_errors);
2187 			goto err;
2188 		}
2189 
2190 		frag = &shinfo->frags[shinfo->nr_frags++];
2191 		skb_frag_fill_page_desc(frag, page, offset, len);
2192 		if (page_is_pfmemalloc(page))
2193 			xdp_buff_set_frag_pfmemalloc(xdp);
2194 
2195 		shinfo->xdp_frags_size += len;
2196 	}
2197 
2198 	*xdp_frags_truesize = xdp_frags_truesz;
2199 	return 0;
2200 
2201 err:
2202 	put_xdp_frags(xdp);
2203 	return -EINVAL;
2204 }
2205 
2206 static void *mergeable_xdp_get_buf(struct virtnet_info *vi,
2207 				   struct receive_queue *rq,
2208 				   struct bpf_prog *xdp_prog,
2209 				   void *ctx,
2210 				   unsigned int *frame_sz,
2211 				   int *num_buf,
2212 				   struct page **page,
2213 				   int offset,
2214 				   unsigned int *len,
2215 				   struct virtio_net_hdr_mrg_rxbuf *hdr)
2216 {
2217 	unsigned int truesize = mergeable_ctx_to_truesize(ctx);
2218 	unsigned int headroom = mergeable_ctx_to_headroom(ctx);
2219 	struct page *xdp_page;
2220 	unsigned int xdp_room;
2221 
2222 	/* Transient failure which in theory could occur if
2223 	 * in-flight packets from before XDP was enabled reach
2224 	 * the receive path after XDP is loaded.
2225 	 */
2226 	if (unlikely(hdr->hdr.gso_type))
2227 		return NULL;
2228 
2229 	/* Partially checksummed packets must be dropped. */
2230 	if (unlikely(hdr->hdr.flags & VIRTIO_NET_HDR_F_NEEDS_CSUM))
2231 		return NULL;
2232 
2233 	/* Now XDP core assumes frag size is PAGE_SIZE, but buffers
2234 	 * with headroom may add hole in truesize, which
2235 	 * make their length exceed PAGE_SIZE. So we disabled the
2236 	 * hole mechanism for xdp. See add_recvbuf_mergeable().
2237 	 */
2238 	*frame_sz = truesize;
2239 
2240 	if (likely(headroom >= virtnet_get_headroom(vi) &&
2241 		   (*num_buf == 1 || xdp_prog->aux->xdp_has_frags))) {
2242 		return page_address(*page) + offset;
2243 	}
2244 
2245 	/* This happens when headroom is not enough because
2246 	 * of the buffer was prefilled before XDP is set.
2247 	 * This should only happen for the first several packets.
2248 	 * In fact, vq reset can be used here to help us clean up
2249 	 * the prefilled buffers, but many existing devices do not
2250 	 * support it, and we don't want to bother users who are
2251 	 * using xdp normally.
2252 	 */
2253 	if (!xdp_prog->aux->xdp_has_frags) {
2254 		/* linearize data for XDP */
2255 		xdp_page = xdp_linearize_page(rq, num_buf,
2256 					      *page, offset,
2257 					      XDP_PACKET_HEADROOM,
2258 					      len);
2259 		if (!xdp_page)
2260 			return NULL;
2261 	} else {
2262 		xdp_room = SKB_DATA_ALIGN(XDP_PACKET_HEADROOM +
2263 					  sizeof(struct skb_shared_info));
2264 		if (*len + xdp_room > PAGE_SIZE)
2265 			return NULL;
2266 
2267 		xdp_page = alloc_page(GFP_ATOMIC);
2268 		if (!xdp_page)
2269 			return NULL;
2270 
2271 		memcpy(page_address(xdp_page) + XDP_PACKET_HEADROOM,
2272 		       page_address(*page) + offset, *len);
2273 	}
2274 
2275 	*frame_sz = PAGE_SIZE;
2276 
2277 	put_page(*page);
2278 
2279 	*page = xdp_page;
2280 
2281 	return page_address(*page) + XDP_PACKET_HEADROOM;
2282 }
2283 
2284 static struct sk_buff *receive_mergeable_xdp(struct net_device *dev,
2285 					     struct virtnet_info *vi,
2286 					     struct receive_queue *rq,
2287 					     struct bpf_prog *xdp_prog,
2288 					     void *buf,
2289 					     void *ctx,
2290 					     unsigned int len,
2291 					     unsigned int *xdp_xmit,
2292 					     struct virtnet_rq_stats *stats)
2293 {
2294 	struct virtio_net_hdr_mrg_rxbuf *hdr = buf;
2295 	int num_buf = virtio16_to_cpu(vi->vdev, hdr->num_buffers);
2296 	struct page *page = virt_to_head_page(buf);
2297 	int offset = buf - page_address(page);
2298 	unsigned int xdp_frags_truesz = 0;
2299 	struct sk_buff *head_skb;
2300 	unsigned int frame_sz;
2301 	struct xdp_buff xdp;
2302 	void *data;
2303 	u32 act;
2304 	int err;
2305 
2306 	data = mergeable_xdp_get_buf(vi, rq, xdp_prog, ctx, &frame_sz, &num_buf, &page,
2307 				     offset, &len, hdr);
2308 	if (unlikely(!data))
2309 		goto err_xdp;
2310 
2311 	err = virtnet_build_xdp_buff_mrg(dev, vi, rq, &xdp, data, len, frame_sz,
2312 					 &num_buf, &xdp_frags_truesz, stats);
2313 	if (unlikely(err))
2314 		goto err_xdp;
2315 
2316 	act = virtnet_xdp_handler(xdp_prog, &xdp, dev, xdp_xmit, stats);
2317 
2318 	switch (act) {
2319 	case XDP_PASS:
2320 		head_skb = build_skb_from_xdp_buff(dev, vi, &xdp, xdp_frags_truesz);
2321 		if (unlikely(!head_skb))
2322 			break;
2323 		return head_skb;
2324 
2325 	case XDP_TX:
2326 	case XDP_REDIRECT:
2327 		return NULL;
2328 
2329 	default:
2330 		break;
2331 	}
2332 
2333 	put_xdp_frags(&xdp);
2334 
2335 err_xdp:
2336 	put_page(page);
2337 	mergeable_buf_free(rq, num_buf, dev, stats);
2338 
2339 	u64_stats_inc(&stats->xdp_drops);
2340 	u64_stats_inc(&stats->drops);
2341 	return NULL;
2342 }
2343 
2344 static struct sk_buff *virtnet_skb_append_frag(struct sk_buff *head_skb,
2345 					       struct sk_buff *curr_skb,
2346 					       struct page *page, void *buf,
2347 					       int len, int truesize)
2348 {
2349 	int num_skb_frags;
2350 	int offset;
2351 
2352 	num_skb_frags = skb_shinfo(curr_skb)->nr_frags;
2353 	if (unlikely(num_skb_frags == MAX_SKB_FRAGS)) {
2354 		struct sk_buff *nskb = alloc_skb(0, GFP_ATOMIC);
2355 
2356 		if (unlikely(!nskb))
2357 			return NULL;
2358 
2359 		if (curr_skb == head_skb)
2360 			skb_shinfo(curr_skb)->frag_list = nskb;
2361 		else
2362 			curr_skb->next = nskb;
2363 		curr_skb = nskb;
2364 		head_skb->truesize += nskb->truesize;
2365 		num_skb_frags = 0;
2366 	}
2367 
2368 	if (curr_skb != head_skb) {
2369 		head_skb->data_len += len;
2370 		head_skb->len += len;
2371 		head_skb->truesize += truesize;
2372 	}
2373 
2374 	offset = buf - page_address(page);
2375 	if (skb_can_coalesce(curr_skb, num_skb_frags, page, offset)) {
2376 		put_page(page);
2377 		skb_coalesce_rx_frag(curr_skb, num_skb_frags - 1,
2378 				     len, truesize);
2379 	} else {
2380 		skb_add_rx_frag(curr_skb, num_skb_frags, page,
2381 				offset, len, truesize);
2382 	}
2383 
2384 	return curr_skb;
2385 }
2386 
2387 static struct sk_buff *receive_mergeable(struct net_device *dev,
2388 					 struct virtnet_info *vi,
2389 					 struct receive_queue *rq,
2390 					 void *buf,
2391 					 void *ctx,
2392 					 unsigned int len,
2393 					 unsigned int *xdp_xmit,
2394 					 struct virtnet_rq_stats *stats)
2395 {
2396 	struct virtio_net_hdr_mrg_rxbuf *hdr = buf;
2397 	int num_buf = virtio16_to_cpu(vi->vdev, hdr->num_buffers);
2398 	struct page *page = virt_to_head_page(buf);
2399 	int offset = buf - page_address(page);
2400 	struct sk_buff *head_skb, *curr_skb;
2401 	unsigned int truesize = mergeable_ctx_to_truesize(ctx);
2402 	unsigned int headroom = mergeable_ctx_to_headroom(ctx);
2403 	unsigned int tailroom = headroom ? sizeof(struct skb_shared_info) : 0;
2404 	unsigned int room = SKB_DATA_ALIGN(headroom + tailroom);
2405 
2406 	head_skb = NULL;
2407 	u64_stats_add(&stats->bytes, len - vi->hdr_len);
2408 
2409 	if (unlikely(len > truesize - room)) {
2410 		pr_debug("%s: rx error: len %u exceeds truesize %lu\n",
2411 			 dev->name, len, (unsigned long)(truesize - room));
2412 		DEV_STATS_INC(dev, rx_length_errors);
2413 		goto err_skb;
2414 	}
2415 
2416 	if (unlikely(vi->xdp_enabled)) {
2417 		struct bpf_prog *xdp_prog;
2418 
2419 		rcu_read_lock();
2420 		xdp_prog = rcu_dereference(rq->xdp_prog);
2421 		if (xdp_prog) {
2422 			head_skb = receive_mergeable_xdp(dev, vi, rq, xdp_prog, buf, ctx,
2423 							 len, xdp_xmit, stats);
2424 			rcu_read_unlock();
2425 			return head_skb;
2426 		}
2427 		rcu_read_unlock();
2428 	}
2429 
2430 	head_skb = page_to_skb(vi, rq, page, offset, len, truesize, headroom);
2431 	curr_skb = head_skb;
2432 
2433 	if (unlikely(!curr_skb))
2434 		goto err_skb;
2435 	while (--num_buf) {
2436 		buf = virtnet_rq_get_buf(rq, &len, &ctx);
2437 		if (unlikely(!buf)) {
2438 			pr_debug("%s: rx error: %d buffers out of %d missing\n",
2439 				 dev->name, num_buf,
2440 				 virtio16_to_cpu(vi->vdev,
2441 						 hdr->num_buffers));
2442 			DEV_STATS_INC(dev, rx_length_errors);
2443 			goto err_buf;
2444 		}
2445 
2446 		u64_stats_add(&stats->bytes, len);
2447 		page = virt_to_head_page(buf);
2448 
2449 		truesize = mergeable_ctx_to_truesize(ctx);
2450 		headroom = mergeable_ctx_to_headroom(ctx);
2451 		tailroom = headroom ? sizeof(struct skb_shared_info) : 0;
2452 		room = SKB_DATA_ALIGN(headroom + tailroom);
2453 		if (unlikely(len > truesize - room)) {
2454 			pr_debug("%s: rx error: len %u exceeds truesize %lu\n",
2455 				 dev->name, len, (unsigned long)(truesize - room));
2456 			DEV_STATS_INC(dev, rx_length_errors);
2457 			goto err_skb;
2458 		}
2459 
2460 		curr_skb  = virtnet_skb_append_frag(head_skb, curr_skb, page,
2461 						    buf, len, truesize);
2462 		if (!curr_skb)
2463 			goto err_skb;
2464 	}
2465 
2466 	ewma_pkt_len_add(&rq->mrg_avg_pkt_len, head_skb->len);
2467 	return head_skb;
2468 
2469 err_skb:
2470 	put_page(page);
2471 	mergeable_buf_free(rq, num_buf, dev, stats);
2472 
2473 err_buf:
2474 	u64_stats_inc(&stats->drops);
2475 	dev_kfree_skb(head_skb);
2476 	return NULL;
2477 }
2478 
2479 static void virtio_skb_set_hash(const struct virtio_net_hdr_v1_hash *hdr_hash,
2480 				struct sk_buff *skb)
2481 {
2482 	enum pkt_hash_types rss_hash_type;
2483 
2484 	if (!hdr_hash || !skb)
2485 		return;
2486 
2487 	switch (__le16_to_cpu(hdr_hash->hash_report)) {
2488 	case VIRTIO_NET_HASH_REPORT_TCPv4:
2489 	case VIRTIO_NET_HASH_REPORT_UDPv4:
2490 	case VIRTIO_NET_HASH_REPORT_TCPv6:
2491 	case VIRTIO_NET_HASH_REPORT_UDPv6:
2492 	case VIRTIO_NET_HASH_REPORT_TCPv6_EX:
2493 	case VIRTIO_NET_HASH_REPORT_UDPv6_EX:
2494 		rss_hash_type = PKT_HASH_TYPE_L4;
2495 		break;
2496 	case VIRTIO_NET_HASH_REPORT_IPv4:
2497 	case VIRTIO_NET_HASH_REPORT_IPv6:
2498 	case VIRTIO_NET_HASH_REPORT_IPv6_EX:
2499 		rss_hash_type = PKT_HASH_TYPE_L3;
2500 		break;
2501 	case VIRTIO_NET_HASH_REPORT_NONE:
2502 	default:
2503 		rss_hash_type = PKT_HASH_TYPE_NONE;
2504 	}
2505 	skb_set_hash(skb, __le32_to_cpu(hdr_hash->hash_value), rss_hash_type);
2506 }
2507 
2508 static void virtnet_receive_done(struct virtnet_info *vi, struct receive_queue *rq,
2509 				 struct sk_buff *skb, u8 flags)
2510 {
2511 	struct virtio_net_common_hdr *hdr;
2512 	struct net_device *dev = vi->dev;
2513 
2514 	hdr = skb_vnet_common_hdr(skb);
2515 	if (dev->features & NETIF_F_RXHASH && vi->has_rss_hash_report)
2516 		virtio_skb_set_hash(&hdr->hash_v1_hdr, skb);
2517 
2518 	if (flags & VIRTIO_NET_HDR_F_DATA_VALID)
2519 		skb->ip_summed = CHECKSUM_UNNECESSARY;
2520 
2521 	if (virtio_net_hdr_to_skb(skb, &hdr->hdr,
2522 				  virtio_is_little_endian(vi->vdev))) {
2523 		net_warn_ratelimited("%s: bad gso: type: %u, size: %u\n",
2524 				     dev->name, hdr->hdr.gso_type,
2525 				     hdr->hdr.gso_size);
2526 		goto frame_err;
2527 	}
2528 
2529 	skb_record_rx_queue(skb, vq2rxq(rq->vq));
2530 	skb->protocol = eth_type_trans(skb, dev);
2531 	pr_debug("Receiving skb proto 0x%04x len %i type %i\n",
2532 		 ntohs(skb->protocol), skb->len, skb->pkt_type);
2533 
2534 	napi_gro_receive(&rq->napi, skb);
2535 	return;
2536 
2537 frame_err:
2538 	DEV_STATS_INC(dev, rx_frame_errors);
2539 	dev_kfree_skb(skb);
2540 }
2541 
2542 static void receive_buf(struct virtnet_info *vi, struct receive_queue *rq,
2543 			void *buf, unsigned int len, void **ctx,
2544 			unsigned int *xdp_xmit,
2545 			struct virtnet_rq_stats *stats)
2546 {
2547 	struct net_device *dev = vi->dev;
2548 	struct sk_buff *skb;
2549 	u8 flags;
2550 
2551 	if (unlikely(len < vi->hdr_len + ETH_HLEN)) {
2552 		pr_debug("%s: short packet %i\n", dev->name, len);
2553 		DEV_STATS_INC(dev, rx_length_errors);
2554 		virtnet_rq_free_buf(vi, rq, buf);
2555 		return;
2556 	}
2557 
2558 	/* 1. Save the flags early, as the XDP program might overwrite them.
2559 	 * These flags ensure packets marked as VIRTIO_NET_HDR_F_DATA_VALID
2560 	 * stay valid after XDP processing.
2561 	 * 2. XDP doesn't work with partially checksummed packets (refer to
2562 	 * virtnet_xdp_set()), so packets marked as
2563 	 * VIRTIO_NET_HDR_F_NEEDS_CSUM get dropped during XDP processing.
2564 	 */
2565 	flags = ((struct virtio_net_common_hdr *)buf)->hdr.flags;
2566 
2567 	if (vi->mergeable_rx_bufs)
2568 		skb = receive_mergeable(dev, vi, rq, buf, ctx, len, xdp_xmit,
2569 					stats);
2570 	else if (vi->big_packets)
2571 		skb = receive_big(dev, vi, rq, buf, len, stats);
2572 	else
2573 		skb = receive_small(dev, vi, rq, buf, ctx, len, xdp_xmit, stats);
2574 
2575 	if (unlikely(!skb))
2576 		return;
2577 
2578 	virtnet_receive_done(vi, rq, skb, flags);
2579 }
2580 
2581 /* Unlike mergeable buffers, all buffers are allocated to the
2582  * same size, except for the headroom. For this reason we do
2583  * not need to use  mergeable_len_to_ctx here - it is enough
2584  * to store the headroom as the context ignoring the truesize.
2585  */
2586 static int add_recvbuf_small(struct virtnet_info *vi, struct receive_queue *rq,
2587 			     gfp_t gfp)
2588 {
2589 	char *buf;
2590 	unsigned int xdp_headroom = virtnet_get_headroom(vi);
2591 	void *ctx = (void *)(unsigned long)xdp_headroom;
2592 	int len = vi->hdr_len + VIRTNET_RX_PAD + GOOD_PACKET_LEN + xdp_headroom;
2593 	int err;
2594 
2595 	len = SKB_DATA_ALIGN(len) +
2596 	      SKB_DATA_ALIGN(sizeof(struct skb_shared_info));
2597 
2598 	if (unlikely(!skb_page_frag_refill(len, &rq->alloc_frag, gfp)))
2599 		return -ENOMEM;
2600 
2601 	buf = virtnet_rq_alloc(rq, len, gfp);
2602 	if (unlikely(!buf))
2603 		return -ENOMEM;
2604 
2605 	buf += VIRTNET_RX_PAD + xdp_headroom;
2606 
2607 	virtnet_rq_init_one_sg(rq, buf, vi->hdr_len + GOOD_PACKET_LEN);
2608 
2609 	err = virtqueue_add_inbuf_premapped(rq->vq, rq->sg, 1, buf, ctx, gfp);
2610 	if (err < 0) {
2611 		virtnet_rq_unmap(rq, buf, 0);
2612 		put_page(virt_to_head_page(buf));
2613 	}
2614 
2615 	return err;
2616 }
2617 
2618 static int add_recvbuf_big(struct virtnet_info *vi, struct receive_queue *rq,
2619 			   gfp_t gfp)
2620 {
2621 	struct page *first, *list = NULL;
2622 	char *p;
2623 	int i, err, offset;
2624 
2625 	sg_init_table(rq->sg, vi->big_packets_num_skbfrags + 2);
2626 
2627 	/* page in rq->sg[vi->big_packets_num_skbfrags + 1] is list tail */
2628 	for (i = vi->big_packets_num_skbfrags + 1; i > 1; --i) {
2629 		first = get_a_page(rq, gfp);
2630 		if (!first) {
2631 			if (list)
2632 				give_pages(rq, list);
2633 			return -ENOMEM;
2634 		}
2635 		sg_set_buf(&rq->sg[i], page_address(first), PAGE_SIZE);
2636 
2637 		/* chain new page in list head to match sg */
2638 		first->private = (unsigned long)list;
2639 		list = first;
2640 	}
2641 
2642 	first = get_a_page(rq, gfp);
2643 	if (!first) {
2644 		give_pages(rq, list);
2645 		return -ENOMEM;
2646 	}
2647 	p = page_address(first);
2648 
2649 	/* rq->sg[0], rq->sg[1] share the same page */
2650 	/* a separated rq->sg[0] for header - required in case !any_header_sg */
2651 	sg_set_buf(&rq->sg[0], p, vi->hdr_len);
2652 
2653 	/* rq->sg[1] for data packet, from offset */
2654 	offset = sizeof(struct padded_vnet_hdr);
2655 	sg_set_buf(&rq->sg[1], p + offset, PAGE_SIZE - offset);
2656 
2657 	/* chain first in list head */
2658 	first->private = (unsigned long)list;
2659 	err = virtqueue_add_inbuf(rq->vq, rq->sg, vi->big_packets_num_skbfrags + 2,
2660 				  first, gfp);
2661 	if (err < 0)
2662 		give_pages(rq, first);
2663 
2664 	return err;
2665 }
2666 
2667 static unsigned int get_mergeable_buf_len(struct receive_queue *rq,
2668 					  struct ewma_pkt_len *avg_pkt_len,
2669 					  unsigned int room)
2670 {
2671 	struct virtnet_info *vi = rq->vq->vdev->priv;
2672 	const size_t hdr_len = vi->hdr_len;
2673 	unsigned int len;
2674 
2675 	if (room)
2676 		return PAGE_SIZE - room;
2677 
2678 	len = hdr_len +	clamp_t(unsigned int, ewma_pkt_len_read(avg_pkt_len),
2679 				rq->min_buf_len, PAGE_SIZE - hdr_len);
2680 
2681 	return ALIGN(len, L1_CACHE_BYTES);
2682 }
2683 
2684 static int add_recvbuf_mergeable(struct virtnet_info *vi,
2685 				 struct receive_queue *rq, gfp_t gfp)
2686 {
2687 	struct page_frag *alloc_frag = &rq->alloc_frag;
2688 	unsigned int headroom = virtnet_get_headroom(vi);
2689 	unsigned int tailroom = headroom ? sizeof(struct skb_shared_info) : 0;
2690 	unsigned int room = SKB_DATA_ALIGN(headroom + tailroom);
2691 	unsigned int len, hole;
2692 	void *ctx;
2693 	char *buf;
2694 	int err;
2695 
2696 	/* Extra tailroom is needed to satisfy XDP's assumption. This
2697 	 * means rx frags coalescing won't work, but consider we've
2698 	 * disabled GSO for XDP, it won't be a big issue.
2699 	 */
2700 	len = get_mergeable_buf_len(rq, &rq->mrg_avg_pkt_len, room);
2701 
2702 	if (unlikely(!skb_page_frag_refill(len + room, alloc_frag, gfp)))
2703 		return -ENOMEM;
2704 
2705 	if (!alloc_frag->offset && len + room + sizeof(struct virtnet_rq_dma) > alloc_frag->size)
2706 		len -= sizeof(struct virtnet_rq_dma);
2707 
2708 	buf = virtnet_rq_alloc(rq, len + room, gfp);
2709 	if (unlikely(!buf))
2710 		return -ENOMEM;
2711 
2712 	buf += headroom; /* advance address leaving hole at front of pkt */
2713 	hole = alloc_frag->size - alloc_frag->offset;
2714 	if (hole < len + room) {
2715 		/* To avoid internal fragmentation, if there is very likely not
2716 		 * enough space for another buffer, add the remaining space to
2717 		 * the current buffer.
2718 		 * XDP core assumes that frame_size of xdp_buff and the length
2719 		 * of the frag are PAGE_SIZE, so we disable the hole mechanism.
2720 		 */
2721 		if (!headroom)
2722 			len += hole;
2723 		alloc_frag->offset += hole;
2724 	}
2725 
2726 	virtnet_rq_init_one_sg(rq, buf, len);
2727 
2728 	ctx = mergeable_len_to_ctx(len + room, headroom);
2729 	err = virtqueue_add_inbuf_premapped(rq->vq, rq->sg, 1, buf, ctx, gfp);
2730 	if (err < 0) {
2731 		virtnet_rq_unmap(rq, buf, 0);
2732 		put_page(virt_to_head_page(buf));
2733 	}
2734 
2735 	return err;
2736 }
2737 
2738 /*
2739  * Returns false if we couldn't fill entirely (OOM).
2740  *
2741  * Normally run in the receive path, but can also be run from ndo_open
2742  * before we're receiving packets, or from refill_work which is
2743  * careful to disable receiving (using napi_disable).
2744  */
2745 static bool try_fill_recv(struct virtnet_info *vi, struct receive_queue *rq,
2746 			  gfp_t gfp)
2747 {
2748 	int err;
2749 
2750 	if (rq->xsk_pool) {
2751 		err = virtnet_add_recvbuf_xsk(vi, rq, rq->xsk_pool, gfp);
2752 		goto kick;
2753 	}
2754 
2755 	do {
2756 		if (vi->mergeable_rx_bufs)
2757 			err = add_recvbuf_mergeable(vi, rq, gfp);
2758 		else if (vi->big_packets)
2759 			err = add_recvbuf_big(vi, rq, gfp);
2760 		else
2761 			err = add_recvbuf_small(vi, rq, gfp);
2762 
2763 		if (err)
2764 			break;
2765 	} while (rq->vq->num_free);
2766 
2767 kick:
2768 	if (virtqueue_kick_prepare(rq->vq) && virtqueue_notify(rq->vq)) {
2769 		unsigned long flags;
2770 
2771 		flags = u64_stats_update_begin_irqsave(&rq->stats.syncp);
2772 		u64_stats_inc(&rq->stats.kicks);
2773 		u64_stats_update_end_irqrestore(&rq->stats.syncp, flags);
2774 	}
2775 
2776 	return err != -ENOMEM;
2777 }
2778 
2779 static void skb_recv_done(struct virtqueue *rvq)
2780 {
2781 	struct virtnet_info *vi = rvq->vdev->priv;
2782 	struct receive_queue *rq = &vi->rq[vq2rxq(rvq)];
2783 
2784 	rq->calls++;
2785 	virtqueue_napi_schedule(&rq->napi, rvq);
2786 }
2787 
2788 static void virtnet_napi_do_enable(struct virtqueue *vq,
2789 				   struct napi_struct *napi)
2790 {
2791 	napi_enable(napi);
2792 
2793 	/* If all buffers were filled by other side before we napi_enabled, we
2794 	 * won't get another interrupt, so process any outstanding packets now.
2795 	 * Call local_bh_enable after to trigger softIRQ processing.
2796 	 */
2797 	local_bh_disable();
2798 	virtqueue_napi_schedule(napi, vq);
2799 	local_bh_enable();
2800 }
2801 
2802 static void virtnet_napi_enable(struct receive_queue *rq)
2803 {
2804 	struct virtnet_info *vi = rq->vq->vdev->priv;
2805 	int qidx = vq2rxq(rq->vq);
2806 
2807 	virtnet_napi_do_enable(rq->vq, &rq->napi);
2808 	netif_queue_set_napi(vi->dev, qidx, NETDEV_QUEUE_TYPE_RX, &rq->napi);
2809 }
2810 
2811 static void virtnet_napi_tx_enable(struct send_queue *sq)
2812 {
2813 	struct virtnet_info *vi = sq->vq->vdev->priv;
2814 	struct napi_struct *napi = &sq->napi;
2815 	int qidx = vq2txq(sq->vq);
2816 
2817 	if (!napi->weight)
2818 		return;
2819 
2820 	/* Tx napi touches cachelines on the cpu handling tx interrupts. Only
2821 	 * enable the feature if this is likely affine with the transmit path.
2822 	 */
2823 	if (!vi->affinity_hint_set) {
2824 		napi->weight = 0;
2825 		return;
2826 	}
2827 
2828 	virtnet_napi_do_enable(sq->vq, napi);
2829 	netif_queue_set_napi(vi->dev, qidx, NETDEV_QUEUE_TYPE_TX, napi);
2830 }
2831 
2832 static void virtnet_napi_tx_disable(struct send_queue *sq)
2833 {
2834 	struct virtnet_info *vi = sq->vq->vdev->priv;
2835 	struct napi_struct *napi = &sq->napi;
2836 	int qidx = vq2txq(sq->vq);
2837 
2838 	if (napi->weight) {
2839 		netif_queue_set_napi(vi->dev, qidx, NETDEV_QUEUE_TYPE_TX, NULL);
2840 		napi_disable(napi);
2841 	}
2842 }
2843 
2844 static void virtnet_napi_disable(struct receive_queue *rq)
2845 {
2846 	struct virtnet_info *vi = rq->vq->vdev->priv;
2847 	struct napi_struct *napi = &rq->napi;
2848 	int qidx = vq2rxq(rq->vq);
2849 
2850 	netif_queue_set_napi(vi->dev, qidx, NETDEV_QUEUE_TYPE_RX, NULL);
2851 	napi_disable(napi);
2852 }
2853 
2854 static void refill_work(struct work_struct *work)
2855 {
2856 	struct virtnet_info *vi =
2857 		container_of(work, struct virtnet_info, refill.work);
2858 	bool still_empty;
2859 	int i;
2860 
2861 	for (i = 0; i < vi->curr_queue_pairs; i++) {
2862 		struct receive_queue *rq = &vi->rq[i];
2863 
2864 		/*
2865 		 * When queue API support is added in the future and the call
2866 		 * below becomes napi_disable_locked, this driver will need to
2867 		 * be refactored.
2868 		 *
2869 		 * One possible solution would be to:
2870 		 *   - cancel refill_work with cancel_delayed_work (note:
2871 		 *     non-sync)
2872 		 *   - cancel refill_work with cancel_delayed_work_sync in
2873 		 *     virtnet_remove after the netdev is unregistered
2874 		 *   - wrap all of the work in a lock (perhaps the netdev
2875 		 *     instance lock)
2876 		 *   - check netif_running() and return early to avoid a race
2877 		 */
2878 		napi_disable(&rq->napi);
2879 		still_empty = !try_fill_recv(vi, rq, GFP_KERNEL);
2880 		virtnet_napi_do_enable(rq->vq, &rq->napi);
2881 
2882 		/* In theory, this can happen: if we don't get any buffers in
2883 		 * we will *never* try to fill again.
2884 		 */
2885 		if (still_empty)
2886 			schedule_delayed_work(&vi->refill, HZ/2);
2887 	}
2888 }
2889 
2890 static int virtnet_receive_xsk_bufs(struct virtnet_info *vi,
2891 				    struct receive_queue *rq,
2892 				    int budget,
2893 				    unsigned int *xdp_xmit,
2894 				    struct virtnet_rq_stats *stats)
2895 {
2896 	unsigned int len;
2897 	int packets = 0;
2898 	void *buf;
2899 
2900 	while (packets < budget) {
2901 		buf = virtqueue_get_buf(rq->vq, &len);
2902 		if (!buf)
2903 			break;
2904 
2905 		virtnet_receive_xsk_buf(vi, rq, buf, len, xdp_xmit, stats);
2906 		packets++;
2907 	}
2908 
2909 	return packets;
2910 }
2911 
2912 static int virtnet_receive_packets(struct virtnet_info *vi,
2913 				   struct receive_queue *rq,
2914 				   int budget,
2915 				   unsigned int *xdp_xmit,
2916 				   struct virtnet_rq_stats *stats)
2917 {
2918 	unsigned int len;
2919 	int packets = 0;
2920 	void *buf;
2921 
2922 	if (!vi->big_packets || vi->mergeable_rx_bufs) {
2923 		void *ctx;
2924 		while (packets < budget &&
2925 		       (buf = virtnet_rq_get_buf(rq, &len, &ctx))) {
2926 			receive_buf(vi, rq, buf, len, ctx, xdp_xmit, stats);
2927 			packets++;
2928 		}
2929 	} else {
2930 		while (packets < budget &&
2931 		       (buf = virtqueue_get_buf(rq->vq, &len)) != NULL) {
2932 			receive_buf(vi, rq, buf, len, NULL, xdp_xmit, stats);
2933 			packets++;
2934 		}
2935 	}
2936 
2937 	return packets;
2938 }
2939 
2940 static int virtnet_receive(struct receive_queue *rq, int budget,
2941 			   unsigned int *xdp_xmit)
2942 {
2943 	struct virtnet_info *vi = rq->vq->vdev->priv;
2944 	struct virtnet_rq_stats stats = {};
2945 	int i, packets;
2946 
2947 	if (rq->xsk_pool)
2948 		packets = virtnet_receive_xsk_bufs(vi, rq, budget, xdp_xmit, &stats);
2949 	else
2950 		packets = virtnet_receive_packets(vi, rq, budget, xdp_xmit, &stats);
2951 
2952 	if (rq->vq->num_free > min((unsigned int)budget, virtqueue_get_vring_size(rq->vq)) / 2) {
2953 		if (!try_fill_recv(vi, rq, GFP_ATOMIC)) {
2954 			spin_lock(&vi->refill_lock);
2955 			if (vi->refill_enabled)
2956 				schedule_delayed_work(&vi->refill, 0);
2957 			spin_unlock(&vi->refill_lock);
2958 		}
2959 	}
2960 
2961 	u64_stats_set(&stats.packets, packets);
2962 	u64_stats_update_begin(&rq->stats.syncp);
2963 	for (i = 0; i < ARRAY_SIZE(virtnet_rq_stats_desc); i++) {
2964 		size_t offset = virtnet_rq_stats_desc[i].offset;
2965 		u64_stats_t *item, *src;
2966 
2967 		item = (u64_stats_t *)((u8 *)&rq->stats + offset);
2968 		src = (u64_stats_t *)((u8 *)&stats + offset);
2969 		u64_stats_add(item, u64_stats_read(src));
2970 	}
2971 
2972 	u64_stats_add(&rq->stats.packets, u64_stats_read(&stats.packets));
2973 	u64_stats_add(&rq->stats.bytes, u64_stats_read(&stats.bytes));
2974 
2975 	u64_stats_update_end(&rq->stats.syncp);
2976 
2977 	return packets;
2978 }
2979 
2980 static void virtnet_poll_cleantx(struct receive_queue *rq, int budget)
2981 {
2982 	struct virtnet_info *vi = rq->vq->vdev->priv;
2983 	unsigned int index = vq2rxq(rq->vq);
2984 	struct send_queue *sq = &vi->sq[index];
2985 	struct netdev_queue *txq = netdev_get_tx_queue(vi->dev, index);
2986 
2987 	if (!sq->napi.weight || is_xdp_raw_buffer_queue(vi, index))
2988 		return;
2989 
2990 	if (__netif_tx_trylock(txq)) {
2991 		if (sq->reset) {
2992 			__netif_tx_unlock(txq);
2993 			return;
2994 		}
2995 
2996 		do {
2997 			virtqueue_disable_cb(sq->vq);
2998 			free_old_xmit(sq, txq, !!budget);
2999 		} while (unlikely(!virtqueue_enable_cb_delayed(sq->vq)));
3000 
3001 		if (sq->vq->num_free >= 2 + MAX_SKB_FRAGS) {
3002 			if (netif_tx_queue_stopped(txq)) {
3003 				u64_stats_update_begin(&sq->stats.syncp);
3004 				u64_stats_inc(&sq->stats.wake);
3005 				u64_stats_update_end(&sq->stats.syncp);
3006 			}
3007 			netif_tx_wake_queue(txq);
3008 		}
3009 
3010 		__netif_tx_unlock(txq);
3011 	}
3012 }
3013 
3014 static void virtnet_rx_dim_update(struct virtnet_info *vi, struct receive_queue *rq)
3015 {
3016 	struct dim_sample cur_sample = {};
3017 
3018 	if (!rq->packets_in_napi)
3019 		return;
3020 
3021 	/* Don't need protection when fetching stats, since fetcher and
3022 	 * updater of the stats are in same context
3023 	 */
3024 	dim_update_sample(rq->calls,
3025 			  u64_stats_read(&rq->stats.packets),
3026 			  u64_stats_read(&rq->stats.bytes),
3027 			  &cur_sample);
3028 
3029 	net_dim(&rq->dim, &cur_sample);
3030 	rq->packets_in_napi = 0;
3031 }
3032 
3033 static int virtnet_poll(struct napi_struct *napi, int budget)
3034 {
3035 	struct receive_queue *rq =
3036 		container_of(napi, struct receive_queue, napi);
3037 	struct virtnet_info *vi = rq->vq->vdev->priv;
3038 	struct send_queue *sq;
3039 	unsigned int received;
3040 	unsigned int xdp_xmit = 0;
3041 	bool napi_complete;
3042 
3043 	virtnet_poll_cleantx(rq, budget);
3044 
3045 	received = virtnet_receive(rq, budget, &xdp_xmit);
3046 	rq->packets_in_napi += received;
3047 
3048 	if (xdp_xmit & VIRTIO_XDP_REDIR)
3049 		xdp_do_flush();
3050 
3051 	/* Out of packets? */
3052 	if (received < budget) {
3053 		napi_complete = virtqueue_napi_complete(napi, rq->vq, received);
3054 		/* Intentionally not taking dim_lock here. This may result in a
3055 		 * spurious net_dim call. But if that happens virtnet_rx_dim_work
3056 		 * will not act on the scheduled work.
3057 		 */
3058 		if (napi_complete && rq->dim_enabled)
3059 			virtnet_rx_dim_update(vi, rq);
3060 	}
3061 
3062 	if (xdp_xmit & VIRTIO_XDP_TX) {
3063 		sq = virtnet_xdp_get_sq(vi);
3064 		if (virtqueue_kick_prepare(sq->vq) && virtqueue_notify(sq->vq)) {
3065 			u64_stats_update_begin(&sq->stats.syncp);
3066 			u64_stats_inc(&sq->stats.kicks);
3067 			u64_stats_update_end(&sq->stats.syncp);
3068 		}
3069 		virtnet_xdp_put_sq(vi, sq);
3070 	}
3071 
3072 	return received;
3073 }
3074 
3075 static void virtnet_disable_queue_pair(struct virtnet_info *vi, int qp_index)
3076 {
3077 	virtnet_napi_tx_disable(&vi->sq[qp_index]);
3078 	virtnet_napi_disable(&vi->rq[qp_index]);
3079 	xdp_rxq_info_unreg(&vi->rq[qp_index].xdp_rxq);
3080 }
3081 
3082 static int virtnet_enable_queue_pair(struct virtnet_info *vi, int qp_index)
3083 {
3084 	struct net_device *dev = vi->dev;
3085 	int err;
3086 
3087 	err = xdp_rxq_info_reg(&vi->rq[qp_index].xdp_rxq, dev, qp_index,
3088 			       vi->rq[qp_index].napi.napi_id);
3089 	if (err < 0)
3090 		return err;
3091 
3092 	err = xdp_rxq_info_reg_mem_model(&vi->rq[qp_index].xdp_rxq,
3093 					 MEM_TYPE_PAGE_SHARED, NULL);
3094 	if (err < 0)
3095 		goto err_xdp_reg_mem_model;
3096 
3097 	virtnet_napi_enable(&vi->rq[qp_index]);
3098 	virtnet_napi_tx_enable(&vi->sq[qp_index]);
3099 
3100 	return 0;
3101 
3102 err_xdp_reg_mem_model:
3103 	xdp_rxq_info_unreg(&vi->rq[qp_index].xdp_rxq);
3104 	return err;
3105 }
3106 
3107 static void virtnet_cancel_dim(struct virtnet_info *vi, struct dim *dim)
3108 {
3109 	if (!virtio_has_feature(vi->vdev, VIRTIO_NET_F_VQ_NOTF_COAL))
3110 		return;
3111 	net_dim_work_cancel(dim);
3112 }
3113 
3114 static void virtnet_update_settings(struct virtnet_info *vi)
3115 {
3116 	u32 speed;
3117 	u8 duplex;
3118 
3119 	if (!virtio_has_feature(vi->vdev, VIRTIO_NET_F_SPEED_DUPLEX))
3120 		return;
3121 
3122 	virtio_cread_le(vi->vdev, struct virtio_net_config, speed, &speed);
3123 
3124 	if (ethtool_validate_speed(speed))
3125 		vi->speed = speed;
3126 
3127 	virtio_cread_le(vi->vdev, struct virtio_net_config, duplex, &duplex);
3128 
3129 	if (ethtool_validate_duplex(duplex))
3130 		vi->duplex = duplex;
3131 }
3132 
3133 static int virtnet_open(struct net_device *dev)
3134 {
3135 	struct virtnet_info *vi = netdev_priv(dev);
3136 	int i, err;
3137 
3138 	enable_delayed_refill(vi);
3139 
3140 	for (i = 0; i < vi->max_queue_pairs; i++) {
3141 		if (i < vi->curr_queue_pairs)
3142 			/* Make sure we have some buffers: if oom use wq. */
3143 			if (!try_fill_recv(vi, &vi->rq[i], GFP_KERNEL))
3144 				schedule_delayed_work(&vi->refill, 0);
3145 
3146 		err = virtnet_enable_queue_pair(vi, i);
3147 		if (err < 0)
3148 			goto err_enable_qp;
3149 	}
3150 
3151 	if (virtio_has_feature(vi->vdev, VIRTIO_NET_F_STATUS)) {
3152 		if (vi->status & VIRTIO_NET_S_LINK_UP)
3153 			netif_carrier_on(vi->dev);
3154 		virtio_config_driver_enable(vi->vdev);
3155 	} else {
3156 		vi->status = VIRTIO_NET_S_LINK_UP;
3157 		netif_carrier_on(dev);
3158 	}
3159 
3160 	return 0;
3161 
3162 err_enable_qp:
3163 	disable_delayed_refill(vi);
3164 	cancel_delayed_work_sync(&vi->refill);
3165 
3166 	for (i--; i >= 0; i--) {
3167 		virtnet_disable_queue_pair(vi, i);
3168 		virtnet_cancel_dim(vi, &vi->rq[i].dim);
3169 	}
3170 
3171 	return err;
3172 }
3173 
3174 static int virtnet_poll_tx(struct napi_struct *napi, int budget)
3175 {
3176 	struct send_queue *sq = container_of(napi, struct send_queue, napi);
3177 	struct virtnet_info *vi = sq->vq->vdev->priv;
3178 	unsigned int index = vq2txq(sq->vq);
3179 	struct netdev_queue *txq;
3180 	int opaque, xsk_done = 0;
3181 	bool done;
3182 
3183 	if (unlikely(is_xdp_raw_buffer_queue(vi, index))) {
3184 		/* We don't need to enable cb for XDP */
3185 		napi_complete_done(napi, 0);
3186 		return 0;
3187 	}
3188 
3189 	txq = netdev_get_tx_queue(vi->dev, index);
3190 	__netif_tx_lock(txq, raw_smp_processor_id());
3191 	virtqueue_disable_cb(sq->vq);
3192 
3193 	if (sq->xsk_pool)
3194 		xsk_done = virtnet_xsk_xmit(sq, sq->xsk_pool, budget);
3195 	else
3196 		free_old_xmit(sq, txq, !!budget);
3197 
3198 	if (sq->vq->num_free >= 2 + MAX_SKB_FRAGS) {
3199 		if (netif_tx_queue_stopped(txq)) {
3200 			u64_stats_update_begin(&sq->stats.syncp);
3201 			u64_stats_inc(&sq->stats.wake);
3202 			u64_stats_update_end(&sq->stats.syncp);
3203 		}
3204 		netif_tx_wake_queue(txq);
3205 	}
3206 
3207 	if (xsk_done >= budget) {
3208 		__netif_tx_unlock(txq);
3209 		return budget;
3210 	}
3211 
3212 	opaque = virtqueue_enable_cb_prepare(sq->vq);
3213 
3214 	done = napi_complete_done(napi, 0);
3215 
3216 	if (!done)
3217 		virtqueue_disable_cb(sq->vq);
3218 
3219 	__netif_tx_unlock(txq);
3220 
3221 	if (done) {
3222 		if (unlikely(virtqueue_poll(sq->vq, opaque))) {
3223 			if (napi_schedule_prep(napi)) {
3224 				__netif_tx_lock(txq, raw_smp_processor_id());
3225 				virtqueue_disable_cb(sq->vq);
3226 				__netif_tx_unlock(txq);
3227 				__napi_schedule(napi);
3228 			}
3229 		}
3230 	}
3231 
3232 	return 0;
3233 }
3234 
3235 static int xmit_skb(struct send_queue *sq, struct sk_buff *skb, bool orphan)
3236 {
3237 	struct virtio_net_hdr_mrg_rxbuf *hdr;
3238 	const unsigned char *dest = ((struct ethhdr *)skb->data)->h_dest;
3239 	struct virtnet_info *vi = sq->vq->vdev->priv;
3240 	int num_sg;
3241 	unsigned hdr_len = vi->hdr_len;
3242 	bool can_push;
3243 
3244 	pr_debug("%s: xmit %p %pM\n", vi->dev->name, skb, dest);
3245 
3246 	can_push = vi->any_header_sg &&
3247 		!((unsigned long)skb->data & (__alignof__(*hdr) - 1)) &&
3248 		!skb_header_cloned(skb) && skb_headroom(skb) >= hdr_len;
3249 	/* Even if we can, don't push here yet as this would skew
3250 	 * csum_start offset below. */
3251 	if (can_push)
3252 		hdr = (struct virtio_net_hdr_mrg_rxbuf *)(skb->data - hdr_len);
3253 	else
3254 		hdr = &skb_vnet_common_hdr(skb)->mrg_hdr;
3255 
3256 	if (virtio_net_hdr_from_skb(skb, &hdr->hdr,
3257 				    virtio_is_little_endian(vi->vdev), false,
3258 				    0))
3259 		return -EPROTO;
3260 
3261 	if (vi->mergeable_rx_bufs)
3262 		hdr->num_buffers = 0;
3263 
3264 	sg_init_table(sq->sg, skb_shinfo(skb)->nr_frags + (can_push ? 1 : 2));
3265 	if (can_push) {
3266 		__skb_push(skb, hdr_len);
3267 		num_sg = skb_to_sgvec(skb, sq->sg, 0, skb->len);
3268 		if (unlikely(num_sg < 0))
3269 			return num_sg;
3270 		/* Pull header back to avoid skew in tx bytes calculations. */
3271 		__skb_pull(skb, hdr_len);
3272 	} else {
3273 		sg_set_buf(sq->sg, hdr, hdr_len);
3274 		num_sg = skb_to_sgvec(skb, sq->sg + 1, 0, skb->len);
3275 		if (unlikely(num_sg < 0))
3276 			return num_sg;
3277 		num_sg++;
3278 	}
3279 
3280 	return virtnet_add_outbuf(sq, num_sg, skb,
3281 				  orphan ? VIRTNET_XMIT_TYPE_SKB_ORPHAN : VIRTNET_XMIT_TYPE_SKB);
3282 }
3283 
3284 static netdev_tx_t start_xmit(struct sk_buff *skb, struct net_device *dev)
3285 {
3286 	struct virtnet_info *vi = netdev_priv(dev);
3287 	int qnum = skb_get_queue_mapping(skb);
3288 	struct send_queue *sq = &vi->sq[qnum];
3289 	int err;
3290 	struct netdev_queue *txq = netdev_get_tx_queue(dev, qnum);
3291 	bool xmit_more = netdev_xmit_more();
3292 	bool use_napi = sq->napi.weight;
3293 	bool kick;
3294 
3295 	if (!use_napi)
3296 		free_old_xmit(sq, txq, false);
3297 	else
3298 		virtqueue_disable_cb(sq->vq);
3299 
3300 	/* timestamp packet in software */
3301 	skb_tx_timestamp(skb);
3302 
3303 	/* Try to transmit */
3304 	err = xmit_skb(sq, skb, !use_napi);
3305 
3306 	/* This should not happen! */
3307 	if (unlikely(err)) {
3308 		DEV_STATS_INC(dev, tx_fifo_errors);
3309 		if (net_ratelimit())
3310 			dev_warn(&dev->dev,
3311 				 "Unexpected TXQ (%d) queue failure: %d\n",
3312 				 qnum, err);
3313 		DEV_STATS_INC(dev, tx_dropped);
3314 		dev_kfree_skb_any(skb);
3315 		return NETDEV_TX_OK;
3316 	}
3317 
3318 	/* Don't wait up for transmitted skbs to be freed. */
3319 	if (!use_napi) {
3320 		skb_orphan(skb);
3321 		nf_reset_ct(skb);
3322 	}
3323 
3324 	if (use_napi)
3325 		tx_may_stop(vi, dev, sq);
3326 	else
3327 		check_sq_full_and_disable(vi, dev,sq);
3328 
3329 	kick = use_napi ? __netdev_tx_sent_queue(txq, skb->len, xmit_more) :
3330 			  !xmit_more || netif_xmit_stopped(txq);
3331 	if (kick) {
3332 		if (virtqueue_kick_prepare(sq->vq) && virtqueue_notify(sq->vq)) {
3333 			u64_stats_update_begin(&sq->stats.syncp);
3334 			u64_stats_inc(&sq->stats.kicks);
3335 			u64_stats_update_end(&sq->stats.syncp);
3336 		}
3337 	}
3338 
3339 	if (use_napi && kick && unlikely(!virtqueue_enable_cb_delayed(sq->vq)))
3340 		virtqueue_napi_schedule(&sq->napi, sq->vq);
3341 
3342 	return NETDEV_TX_OK;
3343 }
3344 
3345 static void __virtnet_rx_pause(struct virtnet_info *vi,
3346 			       struct receive_queue *rq)
3347 {
3348 	bool running = netif_running(vi->dev);
3349 
3350 	if (running) {
3351 		virtnet_napi_disable(rq);
3352 		virtnet_cancel_dim(vi, &rq->dim);
3353 	}
3354 }
3355 
3356 static void virtnet_rx_pause_all(struct virtnet_info *vi)
3357 {
3358 	int i;
3359 
3360 	/*
3361 	 * Make sure refill_work does not run concurrently to
3362 	 * avoid napi_disable race which leads to deadlock.
3363 	 */
3364 	disable_delayed_refill(vi);
3365 	cancel_delayed_work_sync(&vi->refill);
3366 	for (i = 0; i < vi->max_queue_pairs; i++)
3367 		__virtnet_rx_pause(vi, &vi->rq[i]);
3368 }
3369 
3370 static void virtnet_rx_pause(struct virtnet_info *vi, struct receive_queue *rq)
3371 {
3372 	/*
3373 	 * Make sure refill_work does not run concurrently to
3374 	 * avoid napi_disable race which leads to deadlock.
3375 	 */
3376 	disable_delayed_refill(vi);
3377 	cancel_delayed_work_sync(&vi->refill);
3378 	__virtnet_rx_pause(vi, rq);
3379 }
3380 
3381 static void __virtnet_rx_resume(struct virtnet_info *vi,
3382 				struct receive_queue *rq,
3383 				bool refill)
3384 {
3385 	bool running = netif_running(vi->dev);
3386 
3387 	if (refill && !try_fill_recv(vi, rq, GFP_KERNEL))
3388 		schedule_delayed_work(&vi->refill, 0);
3389 
3390 	if (running)
3391 		virtnet_napi_enable(rq);
3392 }
3393 
3394 static void virtnet_rx_resume_all(struct virtnet_info *vi)
3395 {
3396 	int i;
3397 
3398 	enable_delayed_refill(vi);
3399 	for (i = 0; i < vi->max_queue_pairs; i++) {
3400 		if (i < vi->curr_queue_pairs)
3401 			__virtnet_rx_resume(vi, &vi->rq[i], true);
3402 		else
3403 			__virtnet_rx_resume(vi, &vi->rq[i], false);
3404 	}
3405 }
3406 
3407 static void virtnet_rx_resume(struct virtnet_info *vi, struct receive_queue *rq)
3408 {
3409 	enable_delayed_refill(vi);
3410 	__virtnet_rx_resume(vi, rq, true);
3411 }
3412 
3413 static int virtnet_rx_resize(struct virtnet_info *vi,
3414 			     struct receive_queue *rq, u32 ring_num)
3415 {
3416 	int err, qindex;
3417 
3418 	qindex = rq - vi->rq;
3419 
3420 	virtnet_rx_pause(vi, rq);
3421 
3422 	err = virtqueue_resize(rq->vq, ring_num, virtnet_rq_unmap_free_buf, NULL);
3423 	if (err)
3424 		netdev_err(vi->dev, "resize rx fail: rx queue index: %d err: %d\n", qindex, err);
3425 
3426 	virtnet_rx_resume(vi, rq);
3427 	return err;
3428 }
3429 
3430 static void virtnet_tx_pause(struct virtnet_info *vi, struct send_queue *sq)
3431 {
3432 	bool running = netif_running(vi->dev);
3433 	struct netdev_queue *txq;
3434 	int qindex;
3435 
3436 	qindex = sq - vi->sq;
3437 
3438 	if (running)
3439 		virtnet_napi_tx_disable(sq);
3440 
3441 	txq = netdev_get_tx_queue(vi->dev, qindex);
3442 
3443 	/* 1. wait all ximt complete
3444 	 * 2. fix the race of netif_stop_subqueue() vs netif_start_subqueue()
3445 	 */
3446 	__netif_tx_lock_bh(txq);
3447 
3448 	/* Prevent rx poll from accessing sq. */
3449 	sq->reset = true;
3450 
3451 	/* Prevent the upper layer from trying to send packets. */
3452 	netif_stop_subqueue(vi->dev, qindex);
3453 
3454 	__netif_tx_unlock_bh(txq);
3455 }
3456 
3457 static void virtnet_tx_resume(struct virtnet_info *vi, struct send_queue *sq)
3458 {
3459 	bool running = netif_running(vi->dev);
3460 	struct netdev_queue *txq;
3461 	int qindex;
3462 
3463 	qindex = sq - vi->sq;
3464 
3465 	txq = netdev_get_tx_queue(vi->dev, qindex);
3466 
3467 	__netif_tx_lock_bh(txq);
3468 	sq->reset = false;
3469 	netif_tx_wake_queue(txq);
3470 	__netif_tx_unlock_bh(txq);
3471 
3472 	if (running)
3473 		virtnet_napi_tx_enable(sq);
3474 }
3475 
3476 static int virtnet_tx_resize(struct virtnet_info *vi, struct send_queue *sq,
3477 			     u32 ring_num)
3478 {
3479 	int qindex, err;
3480 
3481 	qindex = sq - vi->sq;
3482 
3483 	virtnet_tx_pause(vi, sq);
3484 
3485 	err = virtqueue_resize(sq->vq, ring_num, virtnet_sq_free_unused_buf,
3486 			       virtnet_sq_free_unused_buf_done);
3487 	if (err)
3488 		netdev_err(vi->dev, "resize tx fail: tx queue index: %d err: %d\n", qindex, err);
3489 
3490 	virtnet_tx_resume(vi, sq);
3491 
3492 	return err;
3493 }
3494 
3495 /*
3496  * Send command via the control virtqueue and check status.  Commands
3497  * supported by the hypervisor, as indicated by feature bits, should
3498  * never fail unless improperly formatted.
3499  */
3500 static bool virtnet_send_command_reply(struct virtnet_info *vi, u8 class, u8 cmd,
3501 				       struct scatterlist *out,
3502 				       struct scatterlist *in)
3503 {
3504 	struct scatterlist *sgs[5], hdr, stat;
3505 	u32 out_num = 0, tmp, in_num = 0;
3506 	bool ok;
3507 	int ret;
3508 
3509 	/* Caller should know better */
3510 	BUG_ON(!virtio_has_feature(vi->vdev, VIRTIO_NET_F_CTRL_VQ));
3511 
3512 	mutex_lock(&vi->cvq_lock);
3513 	vi->ctrl->status = ~0;
3514 	vi->ctrl->hdr.class = class;
3515 	vi->ctrl->hdr.cmd = cmd;
3516 	/* Add header */
3517 	sg_init_one(&hdr, &vi->ctrl->hdr, sizeof(vi->ctrl->hdr));
3518 	sgs[out_num++] = &hdr;
3519 
3520 	if (out)
3521 		sgs[out_num++] = out;
3522 
3523 	/* Add return status. */
3524 	sg_init_one(&stat, &vi->ctrl->status, sizeof(vi->ctrl->status));
3525 	sgs[out_num + in_num++] = &stat;
3526 
3527 	if (in)
3528 		sgs[out_num + in_num++] = in;
3529 
3530 	BUG_ON(out_num + in_num > ARRAY_SIZE(sgs));
3531 	ret = virtqueue_add_sgs(vi->cvq, sgs, out_num, in_num, vi, GFP_ATOMIC);
3532 	if (ret < 0) {
3533 		dev_warn(&vi->vdev->dev,
3534 			 "Failed to add sgs for command vq: %d\n.", ret);
3535 		mutex_unlock(&vi->cvq_lock);
3536 		return false;
3537 	}
3538 
3539 	if (unlikely(!virtqueue_kick(vi->cvq)))
3540 		goto unlock;
3541 
3542 	/* Spin for a response, the kick causes an ioport write, trapping
3543 	 * into the hypervisor, so the request should be handled immediately.
3544 	 */
3545 	while (!virtqueue_get_buf(vi->cvq, &tmp) &&
3546 	       !virtqueue_is_broken(vi->cvq)) {
3547 		cond_resched();
3548 		cpu_relax();
3549 	}
3550 
3551 unlock:
3552 	ok = vi->ctrl->status == VIRTIO_NET_OK;
3553 	mutex_unlock(&vi->cvq_lock);
3554 	return ok;
3555 }
3556 
3557 static bool virtnet_send_command(struct virtnet_info *vi, u8 class, u8 cmd,
3558 				 struct scatterlist *out)
3559 {
3560 	return virtnet_send_command_reply(vi, class, cmd, out, NULL);
3561 }
3562 
3563 static int virtnet_set_mac_address(struct net_device *dev, void *p)
3564 {
3565 	struct virtnet_info *vi = netdev_priv(dev);
3566 	struct virtio_device *vdev = vi->vdev;
3567 	int ret;
3568 	struct sockaddr *addr;
3569 	struct scatterlist sg;
3570 
3571 	if (virtio_has_feature(vi->vdev, VIRTIO_NET_F_STANDBY))
3572 		return -EOPNOTSUPP;
3573 
3574 	addr = kmemdup(p, sizeof(*addr), GFP_KERNEL);
3575 	if (!addr)
3576 		return -ENOMEM;
3577 
3578 	ret = eth_prepare_mac_addr_change(dev, addr);
3579 	if (ret)
3580 		goto out;
3581 
3582 	if (virtio_has_feature(vdev, VIRTIO_NET_F_CTRL_MAC_ADDR)) {
3583 		sg_init_one(&sg, addr->sa_data, dev->addr_len);
3584 		if (!virtnet_send_command(vi, VIRTIO_NET_CTRL_MAC,
3585 					  VIRTIO_NET_CTRL_MAC_ADDR_SET, &sg)) {
3586 			dev_warn(&vdev->dev,
3587 				 "Failed to set mac address by vq command.\n");
3588 			ret = -EINVAL;
3589 			goto out;
3590 		}
3591 	} else if (virtio_has_feature(vdev, VIRTIO_NET_F_MAC) &&
3592 		   !virtio_has_feature(vdev, VIRTIO_F_VERSION_1)) {
3593 		unsigned int i;
3594 
3595 		/* Naturally, this has an atomicity problem. */
3596 		for (i = 0; i < dev->addr_len; i++)
3597 			virtio_cwrite8(vdev,
3598 				       offsetof(struct virtio_net_config, mac) +
3599 				       i, addr->sa_data[i]);
3600 	}
3601 
3602 	eth_commit_mac_addr_change(dev, p);
3603 	ret = 0;
3604 
3605 out:
3606 	kfree(addr);
3607 	return ret;
3608 }
3609 
3610 static void virtnet_stats(struct net_device *dev,
3611 			  struct rtnl_link_stats64 *tot)
3612 {
3613 	struct virtnet_info *vi = netdev_priv(dev);
3614 	unsigned int start;
3615 	int i;
3616 
3617 	for (i = 0; i < vi->max_queue_pairs; i++) {
3618 		u64 tpackets, tbytes, terrors, rpackets, rbytes, rdrops;
3619 		struct receive_queue *rq = &vi->rq[i];
3620 		struct send_queue *sq = &vi->sq[i];
3621 
3622 		do {
3623 			start = u64_stats_fetch_begin(&sq->stats.syncp);
3624 			tpackets = u64_stats_read(&sq->stats.packets);
3625 			tbytes   = u64_stats_read(&sq->stats.bytes);
3626 			terrors  = u64_stats_read(&sq->stats.tx_timeouts);
3627 		} while (u64_stats_fetch_retry(&sq->stats.syncp, start));
3628 
3629 		do {
3630 			start = u64_stats_fetch_begin(&rq->stats.syncp);
3631 			rpackets = u64_stats_read(&rq->stats.packets);
3632 			rbytes   = u64_stats_read(&rq->stats.bytes);
3633 			rdrops   = u64_stats_read(&rq->stats.drops);
3634 		} while (u64_stats_fetch_retry(&rq->stats.syncp, start));
3635 
3636 		tot->rx_packets += rpackets;
3637 		tot->tx_packets += tpackets;
3638 		tot->rx_bytes   += rbytes;
3639 		tot->tx_bytes   += tbytes;
3640 		tot->rx_dropped += rdrops;
3641 		tot->tx_errors  += terrors;
3642 	}
3643 
3644 	tot->tx_dropped = DEV_STATS_READ(dev, tx_dropped);
3645 	tot->tx_fifo_errors = DEV_STATS_READ(dev, tx_fifo_errors);
3646 	tot->rx_length_errors = DEV_STATS_READ(dev, rx_length_errors);
3647 	tot->rx_frame_errors = DEV_STATS_READ(dev, rx_frame_errors);
3648 }
3649 
3650 static void virtnet_ack_link_announce(struct virtnet_info *vi)
3651 {
3652 	if (!virtnet_send_command(vi, VIRTIO_NET_CTRL_ANNOUNCE,
3653 				  VIRTIO_NET_CTRL_ANNOUNCE_ACK, NULL))
3654 		dev_warn(&vi->dev->dev, "Failed to ack link announce.\n");
3655 }
3656 
3657 static bool virtnet_commit_rss_command(struct virtnet_info *vi);
3658 
3659 static void virtnet_rss_update_by_qpairs(struct virtnet_info *vi, u16 queue_pairs)
3660 {
3661 	u32 indir_val = 0;
3662 	int i = 0;
3663 
3664 	for (; i < vi->rss_indir_table_size; ++i) {
3665 		indir_val = ethtool_rxfh_indir_default(i, queue_pairs);
3666 		vi->rss_hdr->indirection_table[i] = cpu_to_le16(indir_val);
3667 	}
3668 	vi->rss_trailer.max_tx_vq = cpu_to_le16(queue_pairs);
3669 }
3670 
3671 static int virtnet_set_queues(struct virtnet_info *vi, u16 queue_pairs)
3672 {
3673 	struct virtio_net_ctrl_mq *mq __free(kfree) = NULL;
3674 	struct virtio_net_rss_config_hdr *old_rss_hdr;
3675 	struct virtio_net_rss_config_trailer old_rss_trailer;
3676 	struct net_device *dev = vi->dev;
3677 	struct scatterlist sg;
3678 
3679 	if (!vi->has_cvq || !virtio_has_feature(vi->vdev, VIRTIO_NET_F_MQ))
3680 		return 0;
3681 
3682 	/* Firstly check if we need update rss. Do updating if both (1) rss enabled and
3683 	 * (2) no user configuration.
3684 	 *
3685 	 * During rss command processing, device updates queue_pairs using rss.max_tx_vq. That is,
3686 	 * the device updates queue_pairs together with rss, so we can skip the sperate queue_pairs
3687 	 * update (VIRTIO_NET_CTRL_MQ_VQ_PAIRS_SET below) and return directly.
3688 	 */
3689 	if (vi->has_rss && !netif_is_rxfh_configured(dev)) {
3690 		old_rss_hdr = vi->rss_hdr;
3691 		old_rss_trailer = vi->rss_trailer;
3692 		vi->rss_hdr = devm_kzalloc(&dev->dev, virtnet_rss_hdr_size(vi), GFP_KERNEL);
3693 		if (!vi->rss_hdr) {
3694 			vi->rss_hdr = old_rss_hdr;
3695 			return -ENOMEM;
3696 		}
3697 
3698 		*vi->rss_hdr = *old_rss_hdr;
3699 		virtnet_rss_update_by_qpairs(vi, queue_pairs);
3700 
3701 		if (!virtnet_commit_rss_command(vi)) {
3702 			/* restore ctrl_rss if commit_rss_command failed */
3703 			devm_kfree(&dev->dev, vi->rss_hdr);
3704 			vi->rss_hdr = old_rss_hdr;
3705 			vi->rss_trailer = old_rss_trailer;
3706 
3707 			dev_warn(&dev->dev, "Fail to set num of queue pairs to %d, because committing RSS failed\n",
3708 				 queue_pairs);
3709 			return -EINVAL;
3710 		}
3711 		devm_kfree(&dev->dev, old_rss_hdr);
3712 		goto succ;
3713 	}
3714 
3715 	mq = kzalloc(sizeof(*mq), GFP_KERNEL);
3716 	if (!mq)
3717 		return -ENOMEM;
3718 
3719 	mq->virtqueue_pairs = cpu_to_virtio16(vi->vdev, queue_pairs);
3720 	sg_init_one(&sg, mq, sizeof(*mq));
3721 
3722 	if (!virtnet_send_command(vi, VIRTIO_NET_CTRL_MQ,
3723 				  VIRTIO_NET_CTRL_MQ_VQ_PAIRS_SET, &sg)) {
3724 		dev_warn(&dev->dev, "Fail to set num of queue pairs to %d\n",
3725 			 queue_pairs);
3726 		return -EINVAL;
3727 	}
3728 succ:
3729 	vi->curr_queue_pairs = queue_pairs;
3730 	/* virtnet_open() will refill when device is going to up. */
3731 	if (dev->flags & IFF_UP)
3732 		schedule_delayed_work(&vi->refill, 0);
3733 
3734 	return 0;
3735 }
3736 
3737 static int virtnet_close(struct net_device *dev)
3738 {
3739 	struct virtnet_info *vi = netdev_priv(dev);
3740 	int i;
3741 
3742 	/* Make sure NAPI doesn't schedule refill work */
3743 	disable_delayed_refill(vi);
3744 	/* Make sure refill_work doesn't re-enable napi! */
3745 	cancel_delayed_work_sync(&vi->refill);
3746 	/* Prevent the config change callback from changing carrier
3747 	 * after close
3748 	 */
3749 	virtio_config_driver_disable(vi->vdev);
3750 	/* Stop getting status/speed updates: we don't care until next
3751 	 * open
3752 	 */
3753 	cancel_work_sync(&vi->config_work);
3754 
3755 	for (i = 0; i < vi->max_queue_pairs; i++) {
3756 		virtnet_disable_queue_pair(vi, i);
3757 		virtnet_cancel_dim(vi, &vi->rq[i].dim);
3758 	}
3759 
3760 	netif_carrier_off(dev);
3761 
3762 	return 0;
3763 }
3764 
3765 static void virtnet_rx_mode_work(struct work_struct *work)
3766 {
3767 	struct virtnet_info *vi =
3768 		container_of(work, struct virtnet_info, rx_mode_work);
3769 	u8 *promisc_allmulti  __free(kfree) = NULL;
3770 	struct net_device *dev = vi->dev;
3771 	struct scatterlist sg[2];
3772 	struct virtio_net_ctrl_mac *mac_data;
3773 	struct netdev_hw_addr *ha;
3774 	int uc_count;
3775 	int mc_count;
3776 	void *buf;
3777 	int i;
3778 
3779 	/* We can't dynamically set ndo_set_rx_mode, so return gracefully */
3780 	if (!virtio_has_feature(vi->vdev, VIRTIO_NET_F_CTRL_RX))
3781 		return;
3782 
3783 	promisc_allmulti = kzalloc(sizeof(*promisc_allmulti), GFP_KERNEL);
3784 	if (!promisc_allmulti) {
3785 		dev_warn(&dev->dev, "Failed to set RX mode, no memory.\n");
3786 		return;
3787 	}
3788 
3789 	rtnl_lock();
3790 
3791 	*promisc_allmulti = !!(dev->flags & IFF_PROMISC);
3792 	sg_init_one(sg, promisc_allmulti, sizeof(*promisc_allmulti));
3793 
3794 	if (!virtnet_send_command(vi, VIRTIO_NET_CTRL_RX,
3795 				  VIRTIO_NET_CTRL_RX_PROMISC, sg))
3796 		dev_warn(&dev->dev, "Failed to %sable promisc mode.\n",
3797 			 *promisc_allmulti ? "en" : "dis");
3798 
3799 	*promisc_allmulti = !!(dev->flags & IFF_ALLMULTI);
3800 	sg_init_one(sg, promisc_allmulti, sizeof(*promisc_allmulti));
3801 
3802 	if (!virtnet_send_command(vi, VIRTIO_NET_CTRL_RX,
3803 				  VIRTIO_NET_CTRL_RX_ALLMULTI, sg))
3804 		dev_warn(&dev->dev, "Failed to %sable allmulti mode.\n",
3805 			 *promisc_allmulti ? "en" : "dis");
3806 
3807 	netif_addr_lock_bh(dev);
3808 
3809 	uc_count = netdev_uc_count(dev);
3810 	mc_count = netdev_mc_count(dev);
3811 	/* MAC filter - use one buffer for both lists */
3812 	buf = kzalloc(((uc_count + mc_count) * ETH_ALEN) +
3813 		      (2 * sizeof(mac_data->entries)), GFP_ATOMIC);
3814 	mac_data = buf;
3815 	if (!buf) {
3816 		netif_addr_unlock_bh(dev);
3817 		rtnl_unlock();
3818 		return;
3819 	}
3820 
3821 	sg_init_table(sg, 2);
3822 
3823 	/* Store the unicast list and count in the front of the buffer */
3824 	mac_data->entries = cpu_to_virtio32(vi->vdev, uc_count);
3825 	i = 0;
3826 	netdev_for_each_uc_addr(ha, dev)
3827 		memcpy(&mac_data->macs[i++][0], ha->addr, ETH_ALEN);
3828 
3829 	sg_set_buf(&sg[0], mac_data,
3830 		   sizeof(mac_data->entries) + (uc_count * ETH_ALEN));
3831 
3832 	/* multicast list and count fill the end */
3833 	mac_data = (void *)&mac_data->macs[uc_count][0];
3834 
3835 	mac_data->entries = cpu_to_virtio32(vi->vdev, mc_count);
3836 	i = 0;
3837 	netdev_for_each_mc_addr(ha, dev)
3838 		memcpy(&mac_data->macs[i++][0], ha->addr, ETH_ALEN);
3839 
3840 	netif_addr_unlock_bh(dev);
3841 
3842 	sg_set_buf(&sg[1], mac_data,
3843 		   sizeof(mac_data->entries) + (mc_count * ETH_ALEN));
3844 
3845 	if (!virtnet_send_command(vi, VIRTIO_NET_CTRL_MAC,
3846 				  VIRTIO_NET_CTRL_MAC_TABLE_SET, sg))
3847 		dev_warn(&dev->dev, "Failed to set MAC filter table.\n");
3848 
3849 	rtnl_unlock();
3850 
3851 	kfree(buf);
3852 }
3853 
3854 static void virtnet_set_rx_mode(struct net_device *dev)
3855 {
3856 	struct virtnet_info *vi = netdev_priv(dev);
3857 
3858 	if (vi->rx_mode_work_enabled)
3859 		schedule_work(&vi->rx_mode_work);
3860 }
3861 
3862 static int virtnet_vlan_rx_add_vid(struct net_device *dev,
3863 				   __be16 proto, u16 vid)
3864 {
3865 	struct virtnet_info *vi = netdev_priv(dev);
3866 	__virtio16 *_vid __free(kfree) = NULL;
3867 	struct scatterlist sg;
3868 
3869 	_vid = kzalloc(sizeof(*_vid), GFP_KERNEL);
3870 	if (!_vid)
3871 		return -ENOMEM;
3872 
3873 	*_vid = cpu_to_virtio16(vi->vdev, vid);
3874 	sg_init_one(&sg, _vid, sizeof(*_vid));
3875 
3876 	if (!virtnet_send_command(vi, VIRTIO_NET_CTRL_VLAN,
3877 				  VIRTIO_NET_CTRL_VLAN_ADD, &sg))
3878 		dev_warn(&dev->dev, "Failed to add VLAN ID %d.\n", vid);
3879 	return 0;
3880 }
3881 
3882 static int virtnet_vlan_rx_kill_vid(struct net_device *dev,
3883 				    __be16 proto, u16 vid)
3884 {
3885 	struct virtnet_info *vi = netdev_priv(dev);
3886 	__virtio16 *_vid __free(kfree) = NULL;
3887 	struct scatterlist sg;
3888 
3889 	_vid = kzalloc(sizeof(*_vid), GFP_KERNEL);
3890 	if (!_vid)
3891 		return -ENOMEM;
3892 
3893 	*_vid = cpu_to_virtio16(vi->vdev, vid);
3894 	sg_init_one(&sg, _vid, sizeof(*_vid));
3895 
3896 	if (!virtnet_send_command(vi, VIRTIO_NET_CTRL_VLAN,
3897 				  VIRTIO_NET_CTRL_VLAN_DEL, &sg))
3898 		dev_warn(&dev->dev, "Failed to kill VLAN ID %d.\n", vid);
3899 	return 0;
3900 }
3901 
3902 static void virtnet_clean_affinity(struct virtnet_info *vi)
3903 {
3904 	int i;
3905 
3906 	if (vi->affinity_hint_set) {
3907 		for (i = 0; i < vi->max_queue_pairs; i++) {
3908 			virtqueue_set_affinity(vi->rq[i].vq, NULL);
3909 			virtqueue_set_affinity(vi->sq[i].vq, NULL);
3910 		}
3911 
3912 		vi->affinity_hint_set = false;
3913 	}
3914 }
3915 
3916 static void virtnet_set_affinity(struct virtnet_info *vi)
3917 {
3918 	cpumask_var_t mask;
3919 	int stragglers;
3920 	int group_size;
3921 	int i, start = 0, cpu;
3922 	int num_cpu;
3923 	int stride;
3924 
3925 	if (!zalloc_cpumask_var(&mask, GFP_KERNEL)) {
3926 		virtnet_clean_affinity(vi);
3927 		return;
3928 	}
3929 
3930 	num_cpu = num_online_cpus();
3931 	stride = max_t(int, num_cpu / vi->curr_queue_pairs, 1);
3932 	stragglers = num_cpu >= vi->curr_queue_pairs ?
3933 			num_cpu % vi->curr_queue_pairs :
3934 			0;
3935 
3936 	for (i = 0; i < vi->curr_queue_pairs; i++) {
3937 		group_size = stride + (i < stragglers ? 1 : 0);
3938 
3939 		for_each_online_cpu_wrap(cpu, start) {
3940 			if (!group_size--) {
3941 				start = cpu;
3942 				break;
3943 			}
3944 			cpumask_set_cpu(cpu, mask);
3945 		}
3946 
3947 		virtqueue_set_affinity(vi->rq[i].vq, mask);
3948 		virtqueue_set_affinity(vi->sq[i].vq, mask);
3949 		__netif_set_xps_queue(vi->dev, cpumask_bits(mask), i, XPS_CPUS);
3950 		cpumask_clear(mask);
3951 	}
3952 
3953 	vi->affinity_hint_set = true;
3954 	free_cpumask_var(mask);
3955 }
3956 
3957 static int virtnet_cpu_online(unsigned int cpu, struct hlist_node *node)
3958 {
3959 	struct virtnet_info *vi = hlist_entry_safe(node, struct virtnet_info,
3960 						   node);
3961 	virtnet_set_affinity(vi);
3962 	return 0;
3963 }
3964 
3965 static int virtnet_cpu_dead(unsigned int cpu, struct hlist_node *node)
3966 {
3967 	struct virtnet_info *vi = hlist_entry_safe(node, struct virtnet_info,
3968 						   node_dead);
3969 	virtnet_set_affinity(vi);
3970 	return 0;
3971 }
3972 
3973 static int virtnet_cpu_down_prep(unsigned int cpu, struct hlist_node *node)
3974 {
3975 	struct virtnet_info *vi = hlist_entry_safe(node, struct virtnet_info,
3976 						   node);
3977 
3978 	virtnet_clean_affinity(vi);
3979 	return 0;
3980 }
3981 
3982 static enum cpuhp_state virtionet_online;
3983 
3984 static int virtnet_cpu_notif_add(struct virtnet_info *vi)
3985 {
3986 	int ret;
3987 
3988 	ret = cpuhp_state_add_instance_nocalls(virtionet_online, &vi->node);
3989 	if (ret)
3990 		return ret;
3991 	ret = cpuhp_state_add_instance_nocalls(CPUHP_VIRT_NET_DEAD,
3992 					       &vi->node_dead);
3993 	if (!ret)
3994 		return ret;
3995 	cpuhp_state_remove_instance_nocalls(virtionet_online, &vi->node);
3996 	return ret;
3997 }
3998 
3999 static void virtnet_cpu_notif_remove(struct virtnet_info *vi)
4000 {
4001 	cpuhp_state_remove_instance_nocalls(virtionet_online, &vi->node);
4002 	cpuhp_state_remove_instance_nocalls(CPUHP_VIRT_NET_DEAD,
4003 					    &vi->node_dead);
4004 }
4005 
4006 static int virtnet_send_ctrl_coal_vq_cmd(struct virtnet_info *vi,
4007 					 u16 vqn, u32 max_usecs, u32 max_packets)
4008 {
4009 	struct virtio_net_ctrl_coal_vq *coal_vq __free(kfree) = NULL;
4010 	struct scatterlist sgs;
4011 
4012 	coal_vq = kzalloc(sizeof(*coal_vq), GFP_KERNEL);
4013 	if (!coal_vq)
4014 		return -ENOMEM;
4015 
4016 	coal_vq->vqn = cpu_to_le16(vqn);
4017 	coal_vq->coal.max_usecs = cpu_to_le32(max_usecs);
4018 	coal_vq->coal.max_packets = cpu_to_le32(max_packets);
4019 	sg_init_one(&sgs, coal_vq, sizeof(*coal_vq));
4020 
4021 	if (!virtnet_send_command(vi, VIRTIO_NET_CTRL_NOTF_COAL,
4022 				  VIRTIO_NET_CTRL_NOTF_COAL_VQ_SET,
4023 				  &sgs))
4024 		return -EINVAL;
4025 
4026 	return 0;
4027 }
4028 
4029 static int virtnet_send_rx_ctrl_coal_vq_cmd(struct virtnet_info *vi,
4030 					    u16 queue, u32 max_usecs,
4031 					    u32 max_packets)
4032 {
4033 	int err;
4034 
4035 	if (!virtio_has_feature(vi->vdev, VIRTIO_NET_F_VQ_NOTF_COAL))
4036 		return -EOPNOTSUPP;
4037 
4038 	err = virtnet_send_ctrl_coal_vq_cmd(vi, rxq2vq(queue),
4039 					    max_usecs, max_packets);
4040 	if (err)
4041 		return err;
4042 
4043 	vi->rq[queue].intr_coal.max_usecs = max_usecs;
4044 	vi->rq[queue].intr_coal.max_packets = max_packets;
4045 
4046 	return 0;
4047 }
4048 
4049 static int virtnet_send_tx_ctrl_coal_vq_cmd(struct virtnet_info *vi,
4050 					    u16 queue, u32 max_usecs,
4051 					    u32 max_packets)
4052 {
4053 	int err;
4054 
4055 	if (!virtio_has_feature(vi->vdev, VIRTIO_NET_F_VQ_NOTF_COAL))
4056 		return -EOPNOTSUPP;
4057 
4058 	err = virtnet_send_ctrl_coal_vq_cmd(vi, txq2vq(queue),
4059 					    max_usecs, max_packets);
4060 	if (err)
4061 		return err;
4062 
4063 	vi->sq[queue].intr_coal.max_usecs = max_usecs;
4064 	vi->sq[queue].intr_coal.max_packets = max_packets;
4065 
4066 	return 0;
4067 }
4068 
4069 static void virtnet_get_ringparam(struct net_device *dev,
4070 				  struct ethtool_ringparam *ring,
4071 				  struct kernel_ethtool_ringparam *kernel_ring,
4072 				  struct netlink_ext_ack *extack)
4073 {
4074 	struct virtnet_info *vi = netdev_priv(dev);
4075 
4076 	ring->rx_max_pending = vi->rq[0].vq->num_max;
4077 	ring->tx_max_pending = vi->sq[0].vq->num_max;
4078 	ring->rx_pending = virtqueue_get_vring_size(vi->rq[0].vq);
4079 	ring->tx_pending = virtqueue_get_vring_size(vi->sq[0].vq);
4080 }
4081 
4082 static int virtnet_set_ringparam(struct net_device *dev,
4083 				 struct ethtool_ringparam *ring,
4084 				 struct kernel_ethtool_ringparam *kernel_ring,
4085 				 struct netlink_ext_ack *extack)
4086 {
4087 	struct virtnet_info *vi = netdev_priv(dev);
4088 	u32 rx_pending, tx_pending;
4089 	struct receive_queue *rq;
4090 	struct send_queue *sq;
4091 	int i, err;
4092 
4093 	if (ring->rx_mini_pending || ring->rx_jumbo_pending)
4094 		return -EINVAL;
4095 
4096 	rx_pending = virtqueue_get_vring_size(vi->rq[0].vq);
4097 	tx_pending = virtqueue_get_vring_size(vi->sq[0].vq);
4098 
4099 	if (ring->rx_pending == rx_pending &&
4100 	    ring->tx_pending == tx_pending)
4101 		return 0;
4102 
4103 	if (ring->rx_pending > vi->rq[0].vq->num_max)
4104 		return -EINVAL;
4105 
4106 	if (ring->tx_pending > vi->sq[0].vq->num_max)
4107 		return -EINVAL;
4108 
4109 	for (i = 0; i < vi->max_queue_pairs; i++) {
4110 		rq = vi->rq + i;
4111 		sq = vi->sq + i;
4112 
4113 		if (ring->tx_pending != tx_pending) {
4114 			err = virtnet_tx_resize(vi, sq, ring->tx_pending);
4115 			if (err)
4116 				return err;
4117 
4118 			/* Upon disabling and re-enabling a transmit virtqueue, the device must
4119 			 * set the coalescing parameters of the virtqueue to those configured
4120 			 * through the VIRTIO_NET_CTRL_NOTF_COAL_TX_SET command, or, if the driver
4121 			 * did not set any TX coalescing parameters, to 0.
4122 			 */
4123 			err = virtnet_send_tx_ctrl_coal_vq_cmd(vi, i,
4124 							       vi->intr_coal_tx.max_usecs,
4125 							       vi->intr_coal_tx.max_packets);
4126 
4127 			/* Don't break the tx resize action if the vq coalescing is not
4128 			 * supported. The same is true for rx resize below.
4129 			 */
4130 			if (err && err != -EOPNOTSUPP)
4131 				return err;
4132 		}
4133 
4134 		if (ring->rx_pending != rx_pending) {
4135 			err = virtnet_rx_resize(vi, rq, ring->rx_pending);
4136 			if (err)
4137 				return err;
4138 
4139 			/* The reason is same as the transmit virtqueue reset */
4140 			mutex_lock(&vi->rq[i].dim_lock);
4141 			err = virtnet_send_rx_ctrl_coal_vq_cmd(vi, i,
4142 							       vi->intr_coal_rx.max_usecs,
4143 							       vi->intr_coal_rx.max_packets);
4144 			mutex_unlock(&vi->rq[i].dim_lock);
4145 			if (err && err != -EOPNOTSUPP)
4146 				return err;
4147 		}
4148 	}
4149 
4150 	return 0;
4151 }
4152 
4153 static bool virtnet_commit_rss_command(struct virtnet_info *vi)
4154 {
4155 	struct net_device *dev = vi->dev;
4156 	struct scatterlist sgs[2];
4157 
4158 	/* prepare sgs */
4159 	sg_init_table(sgs, 2);
4160 	sg_set_buf(&sgs[0], vi->rss_hdr, virtnet_rss_hdr_size(vi));
4161 	sg_set_buf(&sgs[1], &vi->rss_trailer, virtnet_rss_trailer_size(vi));
4162 
4163 	if (!virtnet_send_command(vi, VIRTIO_NET_CTRL_MQ,
4164 				  vi->has_rss ? VIRTIO_NET_CTRL_MQ_RSS_CONFIG
4165 				  : VIRTIO_NET_CTRL_MQ_HASH_CONFIG, sgs))
4166 		goto err;
4167 
4168 	return true;
4169 
4170 err:
4171 	dev_warn(&dev->dev, "VIRTIONET issue with committing RSS sgs\n");
4172 	return false;
4173 
4174 }
4175 
4176 static void virtnet_init_default_rss(struct virtnet_info *vi)
4177 {
4178 	vi->rss_hdr->hash_types = cpu_to_le32(vi->rss_hash_types_supported);
4179 	vi->rss_hash_types_saved = vi->rss_hash_types_supported;
4180 	vi->rss_hdr->indirection_table_mask = vi->rss_indir_table_size
4181 						? cpu_to_le16(vi->rss_indir_table_size - 1) : 0;
4182 	vi->rss_hdr->unclassified_queue = 0;
4183 
4184 	virtnet_rss_update_by_qpairs(vi, vi->curr_queue_pairs);
4185 
4186 	vi->rss_trailer.hash_key_length = vi->rss_key_size;
4187 
4188 	netdev_rss_key_fill(vi->rss_hash_key_data, vi->rss_key_size);
4189 }
4190 
4191 static void virtnet_get_hashflow(const struct virtnet_info *vi, struct ethtool_rxnfc *info)
4192 {
4193 	info->data = 0;
4194 	switch (info->flow_type) {
4195 	case TCP_V4_FLOW:
4196 		if (vi->rss_hash_types_saved & VIRTIO_NET_RSS_HASH_TYPE_TCPv4) {
4197 			info->data = RXH_IP_SRC | RXH_IP_DST |
4198 						 RXH_L4_B_0_1 | RXH_L4_B_2_3;
4199 		} else if (vi->rss_hash_types_saved & VIRTIO_NET_RSS_HASH_TYPE_IPv4) {
4200 			info->data = RXH_IP_SRC | RXH_IP_DST;
4201 		}
4202 		break;
4203 	case TCP_V6_FLOW:
4204 		if (vi->rss_hash_types_saved & VIRTIO_NET_RSS_HASH_TYPE_TCPv6) {
4205 			info->data = RXH_IP_SRC | RXH_IP_DST |
4206 						 RXH_L4_B_0_1 | RXH_L4_B_2_3;
4207 		} else if (vi->rss_hash_types_saved & VIRTIO_NET_RSS_HASH_TYPE_IPv6) {
4208 			info->data = RXH_IP_SRC | RXH_IP_DST;
4209 		}
4210 		break;
4211 	case UDP_V4_FLOW:
4212 		if (vi->rss_hash_types_saved & VIRTIO_NET_RSS_HASH_TYPE_UDPv4) {
4213 			info->data = RXH_IP_SRC | RXH_IP_DST |
4214 						 RXH_L4_B_0_1 | RXH_L4_B_2_3;
4215 		} else if (vi->rss_hash_types_saved & VIRTIO_NET_RSS_HASH_TYPE_IPv4) {
4216 			info->data = RXH_IP_SRC | RXH_IP_DST;
4217 		}
4218 		break;
4219 	case UDP_V6_FLOW:
4220 		if (vi->rss_hash_types_saved & VIRTIO_NET_RSS_HASH_TYPE_UDPv6) {
4221 			info->data = RXH_IP_SRC | RXH_IP_DST |
4222 						 RXH_L4_B_0_1 | RXH_L4_B_2_3;
4223 		} else if (vi->rss_hash_types_saved & VIRTIO_NET_RSS_HASH_TYPE_IPv6) {
4224 			info->data = RXH_IP_SRC | RXH_IP_DST;
4225 		}
4226 		break;
4227 	case IPV4_FLOW:
4228 		if (vi->rss_hash_types_saved & VIRTIO_NET_RSS_HASH_TYPE_IPv4)
4229 			info->data = RXH_IP_SRC | RXH_IP_DST;
4230 
4231 		break;
4232 	case IPV6_FLOW:
4233 		if (vi->rss_hash_types_saved & VIRTIO_NET_RSS_HASH_TYPE_IPv6)
4234 			info->data = RXH_IP_SRC | RXH_IP_DST;
4235 
4236 		break;
4237 	default:
4238 		info->data = 0;
4239 		break;
4240 	}
4241 }
4242 
4243 static bool virtnet_set_hashflow(struct virtnet_info *vi, struct ethtool_rxnfc *info)
4244 {
4245 	u32 new_hashtypes = vi->rss_hash_types_saved;
4246 	bool is_disable = info->data & RXH_DISCARD;
4247 	bool is_l4 = info->data == (RXH_IP_SRC | RXH_IP_DST | RXH_L4_B_0_1 | RXH_L4_B_2_3);
4248 
4249 	/* supports only 'sd', 'sdfn' and 'r' */
4250 	if (!((info->data == (RXH_IP_SRC | RXH_IP_DST)) | is_l4 | is_disable))
4251 		return false;
4252 
4253 	switch (info->flow_type) {
4254 	case TCP_V4_FLOW:
4255 		new_hashtypes &= ~(VIRTIO_NET_RSS_HASH_TYPE_IPv4 | VIRTIO_NET_RSS_HASH_TYPE_TCPv4);
4256 		if (!is_disable)
4257 			new_hashtypes |= VIRTIO_NET_RSS_HASH_TYPE_IPv4
4258 				| (is_l4 ? VIRTIO_NET_RSS_HASH_TYPE_TCPv4 : 0);
4259 		break;
4260 	case UDP_V4_FLOW:
4261 		new_hashtypes &= ~(VIRTIO_NET_RSS_HASH_TYPE_IPv4 | VIRTIO_NET_RSS_HASH_TYPE_UDPv4);
4262 		if (!is_disable)
4263 			new_hashtypes |= VIRTIO_NET_RSS_HASH_TYPE_IPv4
4264 				| (is_l4 ? VIRTIO_NET_RSS_HASH_TYPE_UDPv4 : 0);
4265 		break;
4266 	case IPV4_FLOW:
4267 		new_hashtypes &= ~VIRTIO_NET_RSS_HASH_TYPE_IPv4;
4268 		if (!is_disable)
4269 			new_hashtypes = VIRTIO_NET_RSS_HASH_TYPE_IPv4;
4270 		break;
4271 	case TCP_V6_FLOW:
4272 		new_hashtypes &= ~(VIRTIO_NET_RSS_HASH_TYPE_IPv6 | VIRTIO_NET_RSS_HASH_TYPE_TCPv6);
4273 		if (!is_disable)
4274 			new_hashtypes |= VIRTIO_NET_RSS_HASH_TYPE_IPv6
4275 				| (is_l4 ? VIRTIO_NET_RSS_HASH_TYPE_TCPv6 : 0);
4276 		break;
4277 	case UDP_V6_FLOW:
4278 		new_hashtypes &= ~(VIRTIO_NET_RSS_HASH_TYPE_IPv6 | VIRTIO_NET_RSS_HASH_TYPE_UDPv6);
4279 		if (!is_disable)
4280 			new_hashtypes |= VIRTIO_NET_RSS_HASH_TYPE_IPv6
4281 				| (is_l4 ? VIRTIO_NET_RSS_HASH_TYPE_UDPv6 : 0);
4282 		break;
4283 	case IPV6_FLOW:
4284 		new_hashtypes &= ~VIRTIO_NET_RSS_HASH_TYPE_IPv6;
4285 		if (!is_disable)
4286 			new_hashtypes = VIRTIO_NET_RSS_HASH_TYPE_IPv6;
4287 		break;
4288 	default:
4289 		/* unsupported flow */
4290 		return false;
4291 	}
4292 
4293 	/* if unsupported hashtype was set */
4294 	if (new_hashtypes != (new_hashtypes & vi->rss_hash_types_supported))
4295 		return false;
4296 
4297 	if (new_hashtypes != vi->rss_hash_types_saved) {
4298 		vi->rss_hash_types_saved = new_hashtypes;
4299 		vi->rss_hdr->hash_types = cpu_to_le32(vi->rss_hash_types_saved);
4300 		if (vi->dev->features & NETIF_F_RXHASH)
4301 			return virtnet_commit_rss_command(vi);
4302 	}
4303 
4304 	return true;
4305 }
4306 
4307 static void virtnet_get_drvinfo(struct net_device *dev,
4308 				struct ethtool_drvinfo *info)
4309 {
4310 	struct virtnet_info *vi = netdev_priv(dev);
4311 	struct virtio_device *vdev = vi->vdev;
4312 
4313 	strscpy(info->driver, KBUILD_MODNAME, sizeof(info->driver));
4314 	strscpy(info->version, VIRTNET_DRIVER_VERSION, sizeof(info->version));
4315 	strscpy(info->bus_info, virtio_bus_name(vdev), sizeof(info->bus_info));
4316 
4317 }
4318 
4319 /* TODO: Eliminate OOO packets during switching */
4320 static int virtnet_set_channels(struct net_device *dev,
4321 				struct ethtool_channels *channels)
4322 {
4323 	struct virtnet_info *vi = netdev_priv(dev);
4324 	u16 queue_pairs = channels->combined_count;
4325 	int err;
4326 
4327 	/* We don't support separate rx/tx channels.
4328 	 * We don't allow setting 'other' channels.
4329 	 */
4330 	if (channels->rx_count || channels->tx_count || channels->other_count)
4331 		return -EINVAL;
4332 
4333 	if (queue_pairs > vi->max_queue_pairs || queue_pairs == 0)
4334 		return -EINVAL;
4335 
4336 	/* For now we don't support modifying channels while XDP is loaded
4337 	 * also when XDP is loaded all RX queues have XDP programs so we only
4338 	 * need to check a single RX queue.
4339 	 */
4340 	if (vi->rq[0].xdp_prog)
4341 		return -EINVAL;
4342 
4343 	cpus_read_lock();
4344 	err = virtnet_set_queues(vi, queue_pairs);
4345 	if (err) {
4346 		cpus_read_unlock();
4347 		goto err;
4348 	}
4349 	virtnet_set_affinity(vi);
4350 	cpus_read_unlock();
4351 
4352 	netif_set_real_num_tx_queues(dev, queue_pairs);
4353 	netif_set_real_num_rx_queues(dev, queue_pairs);
4354  err:
4355 	return err;
4356 }
4357 
4358 static void virtnet_stats_sprintf(u8 **p, const char *fmt, const char *noq_fmt,
4359 				  int num, int qid, const struct virtnet_stat_desc *desc)
4360 {
4361 	int i;
4362 
4363 	if (qid < 0) {
4364 		for (i = 0; i < num; ++i)
4365 			ethtool_sprintf(p, noq_fmt, desc[i].desc);
4366 	} else {
4367 		for (i = 0; i < num; ++i)
4368 			ethtool_sprintf(p, fmt, qid, desc[i].desc);
4369 	}
4370 }
4371 
4372 /* qid == -1: for rx/tx queue total field */
4373 static void virtnet_get_stats_string(struct virtnet_info *vi, int type, int qid, u8 **data)
4374 {
4375 	const struct virtnet_stat_desc *desc;
4376 	const char *fmt, *noq_fmt;
4377 	u8 *p = *data;
4378 	u32 num;
4379 
4380 	if (type == VIRTNET_Q_TYPE_CQ && qid >= 0) {
4381 		noq_fmt = "cq_hw_%s";
4382 
4383 		if (vi->device_stats_cap & VIRTIO_NET_STATS_TYPE_CVQ) {
4384 			desc = &virtnet_stats_cvq_desc[0];
4385 			num = ARRAY_SIZE(virtnet_stats_cvq_desc);
4386 
4387 			virtnet_stats_sprintf(&p, NULL, noq_fmt, num, -1, desc);
4388 		}
4389 	}
4390 
4391 	if (type == VIRTNET_Q_TYPE_RX) {
4392 		fmt = "rx%u_%s";
4393 		noq_fmt = "rx_%s";
4394 
4395 		desc = &virtnet_rq_stats_desc[0];
4396 		num = ARRAY_SIZE(virtnet_rq_stats_desc);
4397 
4398 		virtnet_stats_sprintf(&p, fmt, noq_fmt, num, qid, desc);
4399 
4400 		fmt = "rx%u_hw_%s";
4401 		noq_fmt = "rx_hw_%s";
4402 
4403 		if (vi->device_stats_cap & VIRTIO_NET_STATS_TYPE_RX_BASIC) {
4404 			desc = &virtnet_stats_rx_basic_desc[0];
4405 			num = ARRAY_SIZE(virtnet_stats_rx_basic_desc);
4406 
4407 			virtnet_stats_sprintf(&p, fmt, noq_fmt, num, qid, desc);
4408 		}
4409 
4410 		if (vi->device_stats_cap & VIRTIO_NET_STATS_TYPE_RX_CSUM) {
4411 			desc = &virtnet_stats_rx_csum_desc[0];
4412 			num = ARRAY_SIZE(virtnet_stats_rx_csum_desc);
4413 
4414 			virtnet_stats_sprintf(&p, fmt, noq_fmt, num, qid, desc);
4415 		}
4416 
4417 		if (vi->device_stats_cap & VIRTIO_NET_STATS_TYPE_RX_SPEED) {
4418 			desc = &virtnet_stats_rx_speed_desc[0];
4419 			num = ARRAY_SIZE(virtnet_stats_rx_speed_desc);
4420 
4421 			virtnet_stats_sprintf(&p, fmt, noq_fmt, num, qid, desc);
4422 		}
4423 	}
4424 
4425 	if (type == VIRTNET_Q_TYPE_TX) {
4426 		fmt = "tx%u_%s";
4427 		noq_fmt = "tx_%s";
4428 
4429 		desc = &virtnet_sq_stats_desc[0];
4430 		num = ARRAY_SIZE(virtnet_sq_stats_desc);
4431 
4432 		virtnet_stats_sprintf(&p, fmt, noq_fmt, num, qid, desc);
4433 
4434 		fmt = "tx%u_hw_%s";
4435 		noq_fmt = "tx_hw_%s";
4436 
4437 		if (vi->device_stats_cap & VIRTIO_NET_STATS_TYPE_TX_BASIC) {
4438 			desc = &virtnet_stats_tx_basic_desc[0];
4439 			num = ARRAY_SIZE(virtnet_stats_tx_basic_desc);
4440 
4441 			virtnet_stats_sprintf(&p, fmt, noq_fmt, num, qid, desc);
4442 		}
4443 
4444 		if (vi->device_stats_cap & VIRTIO_NET_STATS_TYPE_TX_GSO) {
4445 			desc = &virtnet_stats_tx_gso_desc[0];
4446 			num = ARRAY_SIZE(virtnet_stats_tx_gso_desc);
4447 
4448 			virtnet_stats_sprintf(&p, fmt, noq_fmt, num, qid, desc);
4449 		}
4450 
4451 		if (vi->device_stats_cap & VIRTIO_NET_STATS_TYPE_TX_SPEED) {
4452 			desc = &virtnet_stats_tx_speed_desc[0];
4453 			num = ARRAY_SIZE(virtnet_stats_tx_speed_desc);
4454 
4455 			virtnet_stats_sprintf(&p, fmt, noq_fmt, num, qid, desc);
4456 		}
4457 	}
4458 
4459 	*data = p;
4460 }
4461 
4462 struct virtnet_stats_ctx {
4463 	/* The stats are write to qstats or ethtool -S */
4464 	bool to_qstat;
4465 
4466 	/* Used to calculate the offset inside the output buffer. */
4467 	u32 desc_num[3];
4468 
4469 	/* The actual supported stat types. */
4470 	u64 bitmap[3];
4471 
4472 	/* Used to calculate the reply buffer size. */
4473 	u32 size[3];
4474 
4475 	/* Record the output buffer. */
4476 	u64 *data;
4477 };
4478 
4479 static void virtnet_stats_ctx_init(struct virtnet_info *vi,
4480 				   struct virtnet_stats_ctx *ctx,
4481 				   u64 *data, bool to_qstat)
4482 {
4483 	u32 queue_type;
4484 
4485 	ctx->data = data;
4486 	ctx->to_qstat = to_qstat;
4487 
4488 	if (to_qstat) {
4489 		ctx->desc_num[VIRTNET_Q_TYPE_RX] = ARRAY_SIZE(virtnet_rq_stats_desc_qstat);
4490 		ctx->desc_num[VIRTNET_Q_TYPE_TX] = ARRAY_SIZE(virtnet_sq_stats_desc_qstat);
4491 
4492 		queue_type = VIRTNET_Q_TYPE_RX;
4493 
4494 		if (vi->device_stats_cap & VIRTIO_NET_STATS_TYPE_RX_BASIC) {
4495 			ctx->bitmap[queue_type]   |= VIRTIO_NET_STATS_TYPE_RX_BASIC;
4496 			ctx->desc_num[queue_type] += ARRAY_SIZE(virtnet_stats_rx_basic_desc_qstat);
4497 			ctx->size[queue_type]     += sizeof(struct virtio_net_stats_rx_basic);
4498 		}
4499 
4500 		if (vi->device_stats_cap & VIRTIO_NET_STATS_TYPE_RX_CSUM) {
4501 			ctx->bitmap[queue_type]   |= VIRTIO_NET_STATS_TYPE_RX_CSUM;
4502 			ctx->desc_num[queue_type] += ARRAY_SIZE(virtnet_stats_rx_csum_desc_qstat);
4503 			ctx->size[queue_type]     += sizeof(struct virtio_net_stats_rx_csum);
4504 		}
4505 
4506 		if (vi->device_stats_cap & VIRTIO_NET_STATS_TYPE_RX_GSO) {
4507 			ctx->bitmap[queue_type]   |= VIRTIO_NET_STATS_TYPE_RX_GSO;
4508 			ctx->desc_num[queue_type] += ARRAY_SIZE(virtnet_stats_rx_gso_desc_qstat);
4509 			ctx->size[queue_type]     += sizeof(struct virtio_net_stats_rx_gso);
4510 		}
4511 
4512 		if (vi->device_stats_cap & VIRTIO_NET_STATS_TYPE_RX_SPEED) {
4513 			ctx->bitmap[queue_type]   |= VIRTIO_NET_STATS_TYPE_RX_SPEED;
4514 			ctx->desc_num[queue_type] += ARRAY_SIZE(virtnet_stats_rx_speed_desc_qstat);
4515 			ctx->size[queue_type]     += sizeof(struct virtio_net_stats_rx_speed);
4516 		}
4517 
4518 		queue_type = VIRTNET_Q_TYPE_TX;
4519 
4520 		if (vi->device_stats_cap & VIRTIO_NET_STATS_TYPE_TX_BASIC) {
4521 			ctx->bitmap[queue_type]   |= VIRTIO_NET_STATS_TYPE_TX_BASIC;
4522 			ctx->desc_num[queue_type] += ARRAY_SIZE(virtnet_stats_tx_basic_desc_qstat);
4523 			ctx->size[queue_type]     += sizeof(struct virtio_net_stats_tx_basic);
4524 		}
4525 
4526 		if (vi->device_stats_cap & VIRTIO_NET_STATS_TYPE_TX_CSUM) {
4527 			ctx->bitmap[queue_type]   |= VIRTIO_NET_STATS_TYPE_TX_CSUM;
4528 			ctx->desc_num[queue_type] += ARRAY_SIZE(virtnet_stats_tx_csum_desc_qstat);
4529 			ctx->size[queue_type]     += sizeof(struct virtio_net_stats_tx_csum);
4530 		}
4531 
4532 		if (vi->device_stats_cap & VIRTIO_NET_STATS_TYPE_TX_GSO) {
4533 			ctx->bitmap[queue_type]   |= VIRTIO_NET_STATS_TYPE_TX_GSO;
4534 			ctx->desc_num[queue_type] += ARRAY_SIZE(virtnet_stats_tx_gso_desc_qstat);
4535 			ctx->size[queue_type]     += sizeof(struct virtio_net_stats_tx_gso);
4536 		}
4537 
4538 		if (vi->device_stats_cap & VIRTIO_NET_STATS_TYPE_TX_SPEED) {
4539 			ctx->bitmap[queue_type]   |= VIRTIO_NET_STATS_TYPE_TX_SPEED;
4540 			ctx->desc_num[queue_type] += ARRAY_SIZE(virtnet_stats_tx_speed_desc_qstat);
4541 			ctx->size[queue_type]     += sizeof(struct virtio_net_stats_tx_speed);
4542 		}
4543 
4544 		return;
4545 	}
4546 
4547 	ctx->desc_num[VIRTNET_Q_TYPE_RX] = ARRAY_SIZE(virtnet_rq_stats_desc);
4548 	ctx->desc_num[VIRTNET_Q_TYPE_TX] = ARRAY_SIZE(virtnet_sq_stats_desc);
4549 
4550 	if (vi->device_stats_cap & VIRTIO_NET_STATS_TYPE_CVQ) {
4551 		queue_type = VIRTNET_Q_TYPE_CQ;
4552 
4553 		ctx->bitmap[queue_type]   |= VIRTIO_NET_STATS_TYPE_CVQ;
4554 		ctx->desc_num[queue_type] += ARRAY_SIZE(virtnet_stats_cvq_desc);
4555 		ctx->size[queue_type]     += sizeof(struct virtio_net_stats_cvq);
4556 	}
4557 
4558 	queue_type = VIRTNET_Q_TYPE_RX;
4559 
4560 	if (vi->device_stats_cap & VIRTIO_NET_STATS_TYPE_RX_BASIC) {
4561 		ctx->bitmap[queue_type]   |= VIRTIO_NET_STATS_TYPE_RX_BASIC;
4562 		ctx->desc_num[queue_type] += ARRAY_SIZE(virtnet_stats_rx_basic_desc);
4563 		ctx->size[queue_type]     += sizeof(struct virtio_net_stats_rx_basic);
4564 	}
4565 
4566 	if (vi->device_stats_cap & VIRTIO_NET_STATS_TYPE_RX_CSUM) {
4567 		ctx->bitmap[queue_type]   |= VIRTIO_NET_STATS_TYPE_RX_CSUM;
4568 		ctx->desc_num[queue_type] += ARRAY_SIZE(virtnet_stats_rx_csum_desc);
4569 		ctx->size[queue_type]     += sizeof(struct virtio_net_stats_rx_csum);
4570 	}
4571 
4572 	if (vi->device_stats_cap & VIRTIO_NET_STATS_TYPE_RX_SPEED) {
4573 		ctx->bitmap[queue_type]   |= VIRTIO_NET_STATS_TYPE_RX_SPEED;
4574 		ctx->desc_num[queue_type] += ARRAY_SIZE(virtnet_stats_rx_speed_desc);
4575 		ctx->size[queue_type]     += sizeof(struct virtio_net_stats_rx_speed);
4576 	}
4577 
4578 	queue_type = VIRTNET_Q_TYPE_TX;
4579 
4580 	if (vi->device_stats_cap & VIRTIO_NET_STATS_TYPE_TX_BASIC) {
4581 		ctx->bitmap[queue_type]   |= VIRTIO_NET_STATS_TYPE_TX_BASIC;
4582 		ctx->desc_num[queue_type] += ARRAY_SIZE(virtnet_stats_tx_basic_desc);
4583 		ctx->size[queue_type]     += sizeof(struct virtio_net_stats_tx_basic);
4584 	}
4585 
4586 	if (vi->device_stats_cap & VIRTIO_NET_STATS_TYPE_TX_GSO) {
4587 		ctx->bitmap[queue_type]   |= VIRTIO_NET_STATS_TYPE_TX_GSO;
4588 		ctx->desc_num[queue_type] += ARRAY_SIZE(virtnet_stats_tx_gso_desc);
4589 		ctx->size[queue_type]     += sizeof(struct virtio_net_stats_tx_gso);
4590 	}
4591 
4592 	if (vi->device_stats_cap & VIRTIO_NET_STATS_TYPE_TX_SPEED) {
4593 		ctx->bitmap[queue_type]   |= VIRTIO_NET_STATS_TYPE_TX_SPEED;
4594 		ctx->desc_num[queue_type] += ARRAY_SIZE(virtnet_stats_tx_speed_desc);
4595 		ctx->size[queue_type]     += sizeof(struct virtio_net_stats_tx_speed);
4596 	}
4597 }
4598 
4599 /* stats_sum_queue - Calculate the sum of the same fields in sq or rq.
4600  * @sum: the position to store the sum values
4601  * @num: field num
4602  * @q_value: the first queue fields
4603  * @q_num: number of the queues
4604  */
4605 static void stats_sum_queue(u64 *sum, u32 num, u64 *q_value, u32 q_num)
4606 {
4607 	u32 step = num;
4608 	int i, j;
4609 	u64 *p;
4610 
4611 	for (i = 0; i < num; ++i) {
4612 		p = sum + i;
4613 		*p = 0;
4614 
4615 		for (j = 0; j < q_num; ++j)
4616 			*p += *(q_value + i + j * step);
4617 	}
4618 }
4619 
4620 static void virtnet_fill_total_fields(struct virtnet_info *vi,
4621 				      struct virtnet_stats_ctx *ctx)
4622 {
4623 	u64 *data, *first_rx_q, *first_tx_q;
4624 	u32 num_cq, num_rx, num_tx;
4625 
4626 	num_cq = ctx->desc_num[VIRTNET_Q_TYPE_CQ];
4627 	num_rx = ctx->desc_num[VIRTNET_Q_TYPE_RX];
4628 	num_tx = ctx->desc_num[VIRTNET_Q_TYPE_TX];
4629 
4630 	first_rx_q = ctx->data + num_rx + num_tx + num_cq;
4631 	first_tx_q = first_rx_q + vi->curr_queue_pairs * num_rx;
4632 
4633 	data = ctx->data;
4634 
4635 	stats_sum_queue(data, num_rx, first_rx_q, vi->curr_queue_pairs);
4636 
4637 	data = ctx->data + num_rx;
4638 
4639 	stats_sum_queue(data, num_tx, first_tx_q, vi->curr_queue_pairs);
4640 }
4641 
4642 static void virtnet_fill_stats_qstat(struct virtnet_info *vi, u32 qid,
4643 				     struct virtnet_stats_ctx *ctx,
4644 				     const u8 *base, bool drv_stats, u8 reply_type)
4645 {
4646 	const struct virtnet_stat_desc *desc;
4647 	const u64_stats_t *v_stat;
4648 	u64 offset, bitmap;
4649 	const __le64 *v;
4650 	u32 queue_type;
4651 	int i, num;
4652 
4653 	queue_type = vq_type(vi, qid);
4654 	bitmap = ctx->bitmap[queue_type];
4655 
4656 	if (drv_stats) {
4657 		if (queue_type == VIRTNET_Q_TYPE_RX) {
4658 			desc = &virtnet_rq_stats_desc_qstat[0];
4659 			num = ARRAY_SIZE(virtnet_rq_stats_desc_qstat);
4660 		} else {
4661 			desc = &virtnet_sq_stats_desc_qstat[0];
4662 			num = ARRAY_SIZE(virtnet_sq_stats_desc_qstat);
4663 		}
4664 
4665 		for (i = 0; i < num; ++i) {
4666 			offset = desc[i].qstat_offset / sizeof(*ctx->data);
4667 			v_stat = (const u64_stats_t *)(base + desc[i].offset);
4668 			ctx->data[offset] = u64_stats_read(v_stat);
4669 		}
4670 		return;
4671 	}
4672 
4673 	if (bitmap & VIRTIO_NET_STATS_TYPE_RX_BASIC) {
4674 		desc = &virtnet_stats_rx_basic_desc_qstat[0];
4675 		num = ARRAY_SIZE(virtnet_stats_rx_basic_desc_qstat);
4676 		if (reply_type == VIRTIO_NET_STATS_TYPE_REPLY_RX_BASIC)
4677 			goto found;
4678 	}
4679 
4680 	if (bitmap & VIRTIO_NET_STATS_TYPE_RX_CSUM) {
4681 		desc = &virtnet_stats_rx_csum_desc_qstat[0];
4682 		num = ARRAY_SIZE(virtnet_stats_rx_csum_desc_qstat);
4683 		if (reply_type == VIRTIO_NET_STATS_TYPE_REPLY_RX_CSUM)
4684 			goto found;
4685 	}
4686 
4687 	if (bitmap & VIRTIO_NET_STATS_TYPE_RX_GSO) {
4688 		desc = &virtnet_stats_rx_gso_desc_qstat[0];
4689 		num = ARRAY_SIZE(virtnet_stats_rx_gso_desc_qstat);
4690 		if (reply_type == VIRTIO_NET_STATS_TYPE_REPLY_RX_GSO)
4691 			goto found;
4692 	}
4693 
4694 	if (bitmap & VIRTIO_NET_STATS_TYPE_RX_SPEED) {
4695 		desc = &virtnet_stats_rx_speed_desc_qstat[0];
4696 		num = ARRAY_SIZE(virtnet_stats_rx_speed_desc_qstat);
4697 		if (reply_type == VIRTIO_NET_STATS_TYPE_REPLY_RX_SPEED)
4698 			goto found;
4699 	}
4700 
4701 	if (bitmap & VIRTIO_NET_STATS_TYPE_TX_BASIC) {
4702 		desc = &virtnet_stats_tx_basic_desc_qstat[0];
4703 		num = ARRAY_SIZE(virtnet_stats_tx_basic_desc_qstat);
4704 		if (reply_type == VIRTIO_NET_STATS_TYPE_REPLY_TX_BASIC)
4705 			goto found;
4706 	}
4707 
4708 	if (bitmap & VIRTIO_NET_STATS_TYPE_TX_CSUM) {
4709 		desc = &virtnet_stats_tx_csum_desc_qstat[0];
4710 		num = ARRAY_SIZE(virtnet_stats_tx_csum_desc_qstat);
4711 		if (reply_type == VIRTIO_NET_STATS_TYPE_REPLY_TX_CSUM)
4712 			goto found;
4713 	}
4714 
4715 	if (bitmap & VIRTIO_NET_STATS_TYPE_TX_GSO) {
4716 		desc = &virtnet_stats_tx_gso_desc_qstat[0];
4717 		num = ARRAY_SIZE(virtnet_stats_tx_gso_desc_qstat);
4718 		if (reply_type == VIRTIO_NET_STATS_TYPE_REPLY_TX_GSO)
4719 			goto found;
4720 	}
4721 
4722 	if (bitmap & VIRTIO_NET_STATS_TYPE_TX_SPEED) {
4723 		desc = &virtnet_stats_tx_speed_desc_qstat[0];
4724 		num = ARRAY_SIZE(virtnet_stats_tx_speed_desc_qstat);
4725 		if (reply_type == VIRTIO_NET_STATS_TYPE_REPLY_TX_SPEED)
4726 			goto found;
4727 	}
4728 
4729 	return;
4730 
4731 found:
4732 	for (i = 0; i < num; ++i) {
4733 		offset = desc[i].qstat_offset / sizeof(*ctx->data);
4734 		v = (const __le64 *)(base + desc[i].offset);
4735 		ctx->data[offset] = le64_to_cpu(*v);
4736 	}
4737 }
4738 
4739 /* virtnet_fill_stats - copy the stats to qstats or ethtool -S
4740  * The stats source is the device or the driver.
4741  *
4742  * @vi: virtio net info
4743  * @qid: the vq id
4744  * @ctx: stats ctx (initiated by virtnet_stats_ctx_init())
4745  * @base: pointer to the device reply or the driver stats structure.
4746  * @drv_stats: designate the base type (device reply, driver stats)
4747  * @type: the type of the device reply (if drv_stats is true, this must be zero)
4748  */
4749 static void virtnet_fill_stats(struct virtnet_info *vi, u32 qid,
4750 			       struct virtnet_stats_ctx *ctx,
4751 			       const u8 *base, bool drv_stats, u8 reply_type)
4752 {
4753 	u32 queue_type, num_rx, num_tx, num_cq;
4754 	const struct virtnet_stat_desc *desc;
4755 	const u64_stats_t *v_stat;
4756 	u64 offset, bitmap;
4757 	const __le64 *v;
4758 	int i, num;
4759 
4760 	if (ctx->to_qstat)
4761 		return virtnet_fill_stats_qstat(vi, qid, ctx, base, drv_stats, reply_type);
4762 
4763 	num_cq = ctx->desc_num[VIRTNET_Q_TYPE_CQ];
4764 	num_rx = ctx->desc_num[VIRTNET_Q_TYPE_RX];
4765 	num_tx = ctx->desc_num[VIRTNET_Q_TYPE_TX];
4766 
4767 	queue_type = vq_type(vi, qid);
4768 	bitmap = ctx->bitmap[queue_type];
4769 
4770 	/* skip the total fields of pairs */
4771 	offset = num_rx + num_tx;
4772 
4773 	if (queue_type == VIRTNET_Q_TYPE_TX) {
4774 		offset += num_cq + num_rx * vi->curr_queue_pairs + num_tx * (qid / 2);
4775 
4776 		num = ARRAY_SIZE(virtnet_sq_stats_desc);
4777 		if (drv_stats) {
4778 			desc = &virtnet_sq_stats_desc[0];
4779 			goto drv_stats;
4780 		}
4781 
4782 		offset += num;
4783 
4784 	} else if (queue_type == VIRTNET_Q_TYPE_RX) {
4785 		offset += num_cq + num_rx * (qid / 2);
4786 
4787 		num = ARRAY_SIZE(virtnet_rq_stats_desc);
4788 		if (drv_stats) {
4789 			desc = &virtnet_rq_stats_desc[0];
4790 			goto drv_stats;
4791 		}
4792 
4793 		offset += num;
4794 	}
4795 
4796 	if (bitmap & VIRTIO_NET_STATS_TYPE_CVQ) {
4797 		desc = &virtnet_stats_cvq_desc[0];
4798 		num = ARRAY_SIZE(virtnet_stats_cvq_desc);
4799 		if (reply_type == VIRTIO_NET_STATS_TYPE_REPLY_CVQ)
4800 			goto found;
4801 
4802 		offset += num;
4803 	}
4804 
4805 	if (bitmap & VIRTIO_NET_STATS_TYPE_RX_BASIC) {
4806 		desc = &virtnet_stats_rx_basic_desc[0];
4807 		num = ARRAY_SIZE(virtnet_stats_rx_basic_desc);
4808 		if (reply_type == VIRTIO_NET_STATS_TYPE_REPLY_RX_BASIC)
4809 			goto found;
4810 
4811 		offset += num;
4812 	}
4813 
4814 	if (bitmap & VIRTIO_NET_STATS_TYPE_RX_CSUM) {
4815 		desc = &virtnet_stats_rx_csum_desc[0];
4816 		num = ARRAY_SIZE(virtnet_stats_rx_csum_desc);
4817 		if (reply_type == VIRTIO_NET_STATS_TYPE_REPLY_RX_CSUM)
4818 			goto found;
4819 
4820 		offset += num;
4821 	}
4822 
4823 	if (bitmap & VIRTIO_NET_STATS_TYPE_RX_SPEED) {
4824 		desc = &virtnet_stats_rx_speed_desc[0];
4825 		num = ARRAY_SIZE(virtnet_stats_rx_speed_desc);
4826 		if (reply_type == VIRTIO_NET_STATS_TYPE_REPLY_RX_SPEED)
4827 			goto found;
4828 
4829 		offset += num;
4830 	}
4831 
4832 	if (bitmap & VIRTIO_NET_STATS_TYPE_TX_BASIC) {
4833 		desc = &virtnet_stats_tx_basic_desc[0];
4834 		num = ARRAY_SIZE(virtnet_stats_tx_basic_desc);
4835 		if (reply_type == VIRTIO_NET_STATS_TYPE_REPLY_TX_BASIC)
4836 			goto found;
4837 
4838 		offset += num;
4839 	}
4840 
4841 	if (bitmap & VIRTIO_NET_STATS_TYPE_TX_GSO) {
4842 		desc = &virtnet_stats_tx_gso_desc[0];
4843 		num = ARRAY_SIZE(virtnet_stats_tx_gso_desc);
4844 		if (reply_type == VIRTIO_NET_STATS_TYPE_REPLY_TX_GSO)
4845 			goto found;
4846 
4847 		offset += num;
4848 	}
4849 
4850 	if (bitmap & VIRTIO_NET_STATS_TYPE_TX_SPEED) {
4851 		desc = &virtnet_stats_tx_speed_desc[0];
4852 		num = ARRAY_SIZE(virtnet_stats_tx_speed_desc);
4853 		if (reply_type == VIRTIO_NET_STATS_TYPE_REPLY_TX_SPEED)
4854 			goto found;
4855 
4856 		offset += num;
4857 	}
4858 
4859 	return;
4860 
4861 found:
4862 	for (i = 0; i < num; ++i) {
4863 		v = (const __le64 *)(base + desc[i].offset);
4864 		ctx->data[offset + i] = le64_to_cpu(*v);
4865 	}
4866 
4867 	return;
4868 
4869 drv_stats:
4870 	for (i = 0; i < num; ++i) {
4871 		v_stat = (const u64_stats_t *)(base + desc[i].offset);
4872 		ctx->data[offset + i] = u64_stats_read(v_stat);
4873 	}
4874 }
4875 
4876 static int __virtnet_get_hw_stats(struct virtnet_info *vi,
4877 				  struct virtnet_stats_ctx *ctx,
4878 				  struct virtio_net_ctrl_queue_stats *req,
4879 				  int req_size, void *reply, int res_size)
4880 {
4881 	struct virtio_net_stats_reply_hdr *hdr;
4882 	struct scatterlist sgs_in, sgs_out;
4883 	void *p;
4884 	u32 qid;
4885 	int ok;
4886 
4887 	sg_init_one(&sgs_out, req, req_size);
4888 	sg_init_one(&sgs_in, reply, res_size);
4889 
4890 	ok = virtnet_send_command_reply(vi, VIRTIO_NET_CTRL_STATS,
4891 					VIRTIO_NET_CTRL_STATS_GET,
4892 					&sgs_out, &sgs_in);
4893 
4894 	if (!ok)
4895 		return ok;
4896 
4897 	for (p = reply; p - reply < res_size; p += le16_to_cpu(hdr->size)) {
4898 		hdr = p;
4899 		qid = le16_to_cpu(hdr->vq_index);
4900 		virtnet_fill_stats(vi, qid, ctx, p, false, hdr->type);
4901 	}
4902 
4903 	return 0;
4904 }
4905 
4906 static void virtnet_make_stat_req(struct virtnet_info *vi,
4907 				  struct virtnet_stats_ctx *ctx,
4908 				  struct virtio_net_ctrl_queue_stats *req,
4909 				  int qid, int *idx)
4910 {
4911 	int qtype = vq_type(vi, qid);
4912 	u64 bitmap = ctx->bitmap[qtype];
4913 
4914 	if (!bitmap)
4915 		return;
4916 
4917 	req->stats[*idx].vq_index = cpu_to_le16(qid);
4918 	req->stats[*idx].types_bitmap[0] = cpu_to_le64(bitmap);
4919 	*idx += 1;
4920 }
4921 
4922 /* qid: -1: get stats of all vq.
4923  *     > 0: get the stats for the special vq. This must not be cvq.
4924  */
4925 static int virtnet_get_hw_stats(struct virtnet_info *vi,
4926 				struct virtnet_stats_ctx *ctx, int qid)
4927 {
4928 	int qnum, i, j, res_size, qtype, last_vq, first_vq;
4929 	struct virtio_net_ctrl_queue_stats *req;
4930 	bool enable_cvq;
4931 	void *reply;
4932 	int ok;
4933 
4934 	if (!virtio_has_feature(vi->vdev, VIRTIO_NET_F_DEVICE_STATS))
4935 		return 0;
4936 
4937 	if (qid == -1) {
4938 		last_vq = vi->curr_queue_pairs * 2 - 1;
4939 		first_vq = 0;
4940 		enable_cvq = true;
4941 	} else {
4942 		last_vq = qid;
4943 		first_vq = qid;
4944 		enable_cvq = false;
4945 	}
4946 
4947 	qnum = 0;
4948 	res_size = 0;
4949 	for (i = first_vq; i <= last_vq ; ++i) {
4950 		qtype = vq_type(vi, i);
4951 		if (ctx->bitmap[qtype]) {
4952 			++qnum;
4953 			res_size += ctx->size[qtype];
4954 		}
4955 	}
4956 
4957 	if (enable_cvq && ctx->bitmap[VIRTNET_Q_TYPE_CQ]) {
4958 		res_size += ctx->size[VIRTNET_Q_TYPE_CQ];
4959 		qnum += 1;
4960 	}
4961 
4962 	req = kcalloc(qnum, sizeof(*req), GFP_KERNEL);
4963 	if (!req)
4964 		return -ENOMEM;
4965 
4966 	reply = kmalloc(res_size, GFP_KERNEL);
4967 	if (!reply) {
4968 		kfree(req);
4969 		return -ENOMEM;
4970 	}
4971 
4972 	j = 0;
4973 	for (i = first_vq; i <= last_vq ; ++i)
4974 		virtnet_make_stat_req(vi, ctx, req, i, &j);
4975 
4976 	if (enable_cvq)
4977 		virtnet_make_stat_req(vi, ctx, req, vi->max_queue_pairs * 2, &j);
4978 
4979 	ok = __virtnet_get_hw_stats(vi, ctx, req, sizeof(*req) * j, reply, res_size);
4980 
4981 	kfree(req);
4982 	kfree(reply);
4983 
4984 	return ok;
4985 }
4986 
4987 static void virtnet_get_strings(struct net_device *dev, u32 stringset, u8 *data)
4988 {
4989 	struct virtnet_info *vi = netdev_priv(dev);
4990 	unsigned int i;
4991 	u8 *p = data;
4992 
4993 	switch (stringset) {
4994 	case ETH_SS_STATS:
4995 		/* Generate the total field names. */
4996 		virtnet_get_stats_string(vi, VIRTNET_Q_TYPE_RX, -1, &p);
4997 		virtnet_get_stats_string(vi, VIRTNET_Q_TYPE_TX, -1, &p);
4998 
4999 		virtnet_get_stats_string(vi, VIRTNET_Q_TYPE_CQ, 0, &p);
5000 
5001 		for (i = 0; i < vi->curr_queue_pairs; ++i)
5002 			virtnet_get_stats_string(vi, VIRTNET_Q_TYPE_RX, i, &p);
5003 
5004 		for (i = 0; i < vi->curr_queue_pairs; ++i)
5005 			virtnet_get_stats_string(vi, VIRTNET_Q_TYPE_TX, i, &p);
5006 		break;
5007 	}
5008 }
5009 
5010 static int virtnet_get_sset_count(struct net_device *dev, int sset)
5011 {
5012 	struct virtnet_info *vi = netdev_priv(dev);
5013 	struct virtnet_stats_ctx ctx = {0};
5014 	u32 pair_count;
5015 
5016 	switch (sset) {
5017 	case ETH_SS_STATS:
5018 		virtnet_stats_ctx_init(vi, &ctx, NULL, false);
5019 
5020 		pair_count = ctx.desc_num[VIRTNET_Q_TYPE_RX] + ctx.desc_num[VIRTNET_Q_TYPE_TX];
5021 
5022 		return pair_count + ctx.desc_num[VIRTNET_Q_TYPE_CQ] +
5023 			vi->curr_queue_pairs * pair_count;
5024 	default:
5025 		return -EOPNOTSUPP;
5026 	}
5027 }
5028 
5029 static void virtnet_get_ethtool_stats(struct net_device *dev,
5030 				      struct ethtool_stats *stats, u64 *data)
5031 {
5032 	struct virtnet_info *vi = netdev_priv(dev);
5033 	struct virtnet_stats_ctx ctx = {0};
5034 	unsigned int start, i;
5035 	const u8 *stats_base;
5036 
5037 	virtnet_stats_ctx_init(vi, &ctx, data, false);
5038 	if (virtnet_get_hw_stats(vi, &ctx, -1))
5039 		dev_warn(&vi->dev->dev, "Failed to get hw stats.\n");
5040 
5041 	for (i = 0; i < vi->curr_queue_pairs; i++) {
5042 		struct receive_queue *rq = &vi->rq[i];
5043 		struct send_queue *sq = &vi->sq[i];
5044 
5045 		stats_base = (const u8 *)&rq->stats;
5046 		do {
5047 			start = u64_stats_fetch_begin(&rq->stats.syncp);
5048 			virtnet_fill_stats(vi, i * 2, &ctx, stats_base, true, 0);
5049 		} while (u64_stats_fetch_retry(&rq->stats.syncp, start));
5050 
5051 		stats_base = (const u8 *)&sq->stats;
5052 		do {
5053 			start = u64_stats_fetch_begin(&sq->stats.syncp);
5054 			virtnet_fill_stats(vi, i * 2 + 1, &ctx, stats_base, true, 0);
5055 		} while (u64_stats_fetch_retry(&sq->stats.syncp, start));
5056 	}
5057 
5058 	virtnet_fill_total_fields(vi, &ctx);
5059 }
5060 
5061 static void virtnet_get_channels(struct net_device *dev,
5062 				 struct ethtool_channels *channels)
5063 {
5064 	struct virtnet_info *vi = netdev_priv(dev);
5065 
5066 	channels->combined_count = vi->curr_queue_pairs;
5067 	channels->max_combined = vi->max_queue_pairs;
5068 	channels->max_other = 0;
5069 	channels->rx_count = 0;
5070 	channels->tx_count = 0;
5071 	channels->other_count = 0;
5072 }
5073 
5074 static int virtnet_set_link_ksettings(struct net_device *dev,
5075 				      const struct ethtool_link_ksettings *cmd)
5076 {
5077 	struct virtnet_info *vi = netdev_priv(dev);
5078 
5079 	return ethtool_virtdev_set_link_ksettings(dev, cmd,
5080 						  &vi->speed, &vi->duplex);
5081 }
5082 
5083 static int virtnet_get_link_ksettings(struct net_device *dev,
5084 				      struct ethtool_link_ksettings *cmd)
5085 {
5086 	struct virtnet_info *vi = netdev_priv(dev);
5087 
5088 	cmd->base.speed = vi->speed;
5089 	cmd->base.duplex = vi->duplex;
5090 	cmd->base.port = PORT_OTHER;
5091 
5092 	return 0;
5093 }
5094 
5095 static int virtnet_send_tx_notf_coal_cmds(struct virtnet_info *vi,
5096 					  struct ethtool_coalesce *ec)
5097 {
5098 	struct virtio_net_ctrl_coal_tx *coal_tx __free(kfree) = NULL;
5099 	struct scatterlist sgs_tx;
5100 	int i;
5101 
5102 	coal_tx = kzalloc(sizeof(*coal_tx), GFP_KERNEL);
5103 	if (!coal_tx)
5104 		return -ENOMEM;
5105 
5106 	coal_tx->tx_usecs = cpu_to_le32(ec->tx_coalesce_usecs);
5107 	coal_tx->tx_max_packets = cpu_to_le32(ec->tx_max_coalesced_frames);
5108 	sg_init_one(&sgs_tx, coal_tx, sizeof(*coal_tx));
5109 
5110 	if (!virtnet_send_command(vi, VIRTIO_NET_CTRL_NOTF_COAL,
5111 				  VIRTIO_NET_CTRL_NOTF_COAL_TX_SET,
5112 				  &sgs_tx))
5113 		return -EINVAL;
5114 
5115 	vi->intr_coal_tx.max_usecs = ec->tx_coalesce_usecs;
5116 	vi->intr_coal_tx.max_packets = ec->tx_max_coalesced_frames;
5117 	for (i = 0; i < vi->max_queue_pairs; i++) {
5118 		vi->sq[i].intr_coal.max_usecs = ec->tx_coalesce_usecs;
5119 		vi->sq[i].intr_coal.max_packets = ec->tx_max_coalesced_frames;
5120 	}
5121 
5122 	return 0;
5123 }
5124 
5125 static int virtnet_send_rx_notf_coal_cmds(struct virtnet_info *vi,
5126 					  struct ethtool_coalesce *ec)
5127 {
5128 	struct virtio_net_ctrl_coal_rx *coal_rx __free(kfree) = NULL;
5129 	bool rx_ctrl_dim_on = !!ec->use_adaptive_rx_coalesce;
5130 	struct scatterlist sgs_rx;
5131 	int i;
5132 
5133 	if (rx_ctrl_dim_on && !virtio_has_feature(vi->vdev, VIRTIO_NET_F_VQ_NOTF_COAL))
5134 		return -EOPNOTSUPP;
5135 
5136 	if (rx_ctrl_dim_on && (ec->rx_coalesce_usecs != vi->intr_coal_rx.max_usecs ||
5137 			       ec->rx_max_coalesced_frames != vi->intr_coal_rx.max_packets))
5138 		return -EINVAL;
5139 
5140 	if (rx_ctrl_dim_on && !vi->rx_dim_enabled) {
5141 		vi->rx_dim_enabled = true;
5142 		for (i = 0; i < vi->max_queue_pairs; i++) {
5143 			mutex_lock(&vi->rq[i].dim_lock);
5144 			vi->rq[i].dim_enabled = true;
5145 			mutex_unlock(&vi->rq[i].dim_lock);
5146 		}
5147 		return 0;
5148 	}
5149 
5150 	coal_rx = kzalloc(sizeof(*coal_rx), GFP_KERNEL);
5151 	if (!coal_rx)
5152 		return -ENOMEM;
5153 
5154 	if (!rx_ctrl_dim_on && vi->rx_dim_enabled) {
5155 		vi->rx_dim_enabled = false;
5156 		for (i = 0; i < vi->max_queue_pairs; i++) {
5157 			mutex_lock(&vi->rq[i].dim_lock);
5158 			vi->rq[i].dim_enabled = false;
5159 			mutex_unlock(&vi->rq[i].dim_lock);
5160 		}
5161 	}
5162 
5163 	/* Since the per-queue coalescing params can be set,
5164 	 * we need apply the global new params even if they
5165 	 * are not updated.
5166 	 */
5167 	coal_rx->rx_usecs = cpu_to_le32(ec->rx_coalesce_usecs);
5168 	coal_rx->rx_max_packets = cpu_to_le32(ec->rx_max_coalesced_frames);
5169 	sg_init_one(&sgs_rx, coal_rx, sizeof(*coal_rx));
5170 
5171 	if (!virtnet_send_command(vi, VIRTIO_NET_CTRL_NOTF_COAL,
5172 				  VIRTIO_NET_CTRL_NOTF_COAL_RX_SET,
5173 				  &sgs_rx))
5174 		return -EINVAL;
5175 
5176 	vi->intr_coal_rx.max_usecs = ec->rx_coalesce_usecs;
5177 	vi->intr_coal_rx.max_packets = ec->rx_max_coalesced_frames;
5178 	for (i = 0; i < vi->max_queue_pairs; i++) {
5179 		mutex_lock(&vi->rq[i].dim_lock);
5180 		vi->rq[i].intr_coal.max_usecs = ec->rx_coalesce_usecs;
5181 		vi->rq[i].intr_coal.max_packets = ec->rx_max_coalesced_frames;
5182 		mutex_unlock(&vi->rq[i].dim_lock);
5183 	}
5184 
5185 	return 0;
5186 }
5187 
5188 static int virtnet_send_notf_coal_cmds(struct virtnet_info *vi,
5189 				       struct ethtool_coalesce *ec)
5190 {
5191 	int err;
5192 
5193 	err = virtnet_send_tx_notf_coal_cmds(vi, ec);
5194 	if (err)
5195 		return err;
5196 
5197 	err = virtnet_send_rx_notf_coal_cmds(vi, ec);
5198 	if (err)
5199 		return err;
5200 
5201 	return 0;
5202 }
5203 
5204 static int virtnet_send_rx_notf_coal_vq_cmds(struct virtnet_info *vi,
5205 					     struct ethtool_coalesce *ec,
5206 					     u16 queue)
5207 {
5208 	bool rx_ctrl_dim_on = !!ec->use_adaptive_rx_coalesce;
5209 	u32 max_usecs, max_packets;
5210 	bool cur_rx_dim;
5211 	int err;
5212 
5213 	mutex_lock(&vi->rq[queue].dim_lock);
5214 	cur_rx_dim = vi->rq[queue].dim_enabled;
5215 	max_usecs = vi->rq[queue].intr_coal.max_usecs;
5216 	max_packets = vi->rq[queue].intr_coal.max_packets;
5217 
5218 	if (rx_ctrl_dim_on && (ec->rx_coalesce_usecs != max_usecs ||
5219 			       ec->rx_max_coalesced_frames != max_packets)) {
5220 		mutex_unlock(&vi->rq[queue].dim_lock);
5221 		return -EINVAL;
5222 	}
5223 
5224 	if (rx_ctrl_dim_on && !cur_rx_dim) {
5225 		vi->rq[queue].dim_enabled = true;
5226 		mutex_unlock(&vi->rq[queue].dim_lock);
5227 		return 0;
5228 	}
5229 
5230 	if (!rx_ctrl_dim_on && cur_rx_dim)
5231 		vi->rq[queue].dim_enabled = false;
5232 
5233 	/* If no params are updated, userspace ethtool will
5234 	 * reject the modification.
5235 	 */
5236 	err = virtnet_send_rx_ctrl_coal_vq_cmd(vi, queue,
5237 					       ec->rx_coalesce_usecs,
5238 					       ec->rx_max_coalesced_frames);
5239 	mutex_unlock(&vi->rq[queue].dim_lock);
5240 	return err;
5241 }
5242 
5243 static int virtnet_send_notf_coal_vq_cmds(struct virtnet_info *vi,
5244 					  struct ethtool_coalesce *ec,
5245 					  u16 queue)
5246 {
5247 	int err;
5248 
5249 	err = virtnet_send_rx_notf_coal_vq_cmds(vi, ec, queue);
5250 	if (err)
5251 		return err;
5252 
5253 	err = virtnet_send_tx_ctrl_coal_vq_cmd(vi, queue,
5254 					       ec->tx_coalesce_usecs,
5255 					       ec->tx_max_coalesced_frames);
5256 	if (err)
5257 		return err;
5258 
5259 	return 0;
5260 }
5261 
5262 static void virtnet_rx_dim_work(struct work_struct *work)
5263 {
5264 	struct dim *dim = container_of(work, struct dim, work);
5265 	struct receive_queue *rq = container_of(dim,
5266 			struct receive_queue, dim);
5267 	struct virtnet_info *vi = rq->vq->vdev->priv;
5268 	struct net_device *dev = vi->dev;
5269 	struct dim_cq_moder update_moder;
5270 	int qnum, err;
5271 
5272 	qnum = rq - vi->rq;
5273 
5274 	mutex_lock(&rq->dim_lock);
5275 	if (!rq->dim_enabled)
5276 		goto out;
5277 
5278 	update_moder = net_dim_get_rx_irq_moder(dev, dim);
5279 	if (update_moder.usec != rq->intr_coal.max_usecs ||
5280 	    update_moder.pkts != rq->intr_coal.max_packets) {
5281 		err = virtnet_send_rx_ctrl_coal_vq_cmd(vi, qnum,
5282 						       update_moder.usec,
5283 						       update_moder.pkts);
5284 		if (err)
5285 			pr_debug("%s: Failed to send dim parameters on rxq%d\n",
5286 				 dev->name, qnum);
5287 	}
5288 out:
5289 	dim->state = DIM_START_MEASURE;
5290 	mutex_unlock(&rq->dim_lock);
5291 }
5292 
5293 static int virtnet_coal_params_supported(struct ethtool_coalesce *ec)
5294 {
5295 	/* usecs coalescing is supported only if VIRTIO_NET_F_NOTF_COAL
5296 	 * or VIRTIO_NET_F_VQ_NOTF_COAL feature is negotiated.
5297 	 */
5298 	if (ec->rx_coalesce_usecs || ec->tx_coalesce_usecs)
5299 		return -EOPNOTSUPP;
5300 
5301 	if (ec->tx_max_coalesced_frames > 1 ||
5302 	    ec->rx_max_coalesced_frames != 1)
5303 		return -EINVAL;
5304 
5305 	return 0;
5306 }
5307 
5308 static int virtnet_should_update_vq_weight(int dev_flags, int weight,
5309 					   int vq_weight, bool *should_update)
5310 {
5311 	if (weight ^ vq_weight) {
5312 		if (dev_flags & IFF_UP)
5313 			return -EBUSY;
5314 		*should_update = true;
5315 	}
5316 
5317 	return 0;
5318 }
5319 
5320 static int virtnet_set_coalesce(struct net_device *dev,
5321 				struct ethtool_coalesce *ec,
5322 				struct kernel_ethtool_coalesce *kernel_coal,
5323 				struct netlink_ext_ack *extack)
5324 {
5325 	struct virtnet_info *vi = netdev_priv(dev);
5326 	int ret, queue_number, napi_weight, i;
5327 	bool update_napi = false;
5328 
5329 	/* Can't change NAPI weight if the link is up */
5330 	napi_weight = ec->tx_max_coalesced_frames ? NAPI_POLL_WEIGHT : 0;
5331 	for (queue_number = 0; queue_number < vi->max_queue_pairs; queue_number++) {
5332 		ret = virtnet_should_update_vq_weight(dev->flags, napi_weight,
5333 						      vi->sq[queue_number].napi.weight,
5334 						      &update_napi);
5335 		if (ret)
5336 			return ret;
5337 
5338 		if (update_napi) {
5339 			/* All queues that belong to [queue_number, vi->max_queue_pairs] will be
5340 			 * updated for the sake of simplicity, which might not be necessary
5341 			 */
5342 			break;
5343 		}
5344 	}
5345 
5346 	if (virtio_has_feature(vi->vdev, VIRTIO_NET_F_NOTF_COAL))
5347 		ret = virtnet_send_notf_coal_cmds(vi, ec);
5348 	else
5349 		ret = virtnet_coal_params_supported(ec);
5350 
5351 	if (ret)
5352 		return ret;
5353 
5354 	if (update_napi) {
5355 		/* xsk xmit depends on the tx napi. So if xsk is active,
5356 		 * prevent modifications to tx napi.
5357 		 */
5358 		for (i = queue_number; i < vi->max_queue_pairs; i++) {
5359 			if (vi->sq[i].xsk_pool)
5360 				return -EBUSY;
5361 		}
5362 
5363 		for (; queue_number < vi->max_queue_pairs; queue_number++)
5364 			vi->sq[queue_number].napi.weight = napi_weight;
5365 	}
5366 
5367 	return ret;
5368 }
5369 
5370 static int virtnet_get_coalesce(struct net_device *dev,
5371 				struct ethtool_coalesce *ec,
5372 				struct kernel_ethtool_coalesce *kernel_coal,
5373 				struct netlink_ext_ack *extack)
5374 {
5375 	struct virtnet_info *vi = netdev_priv(dev);
5376 
5377 	if (virtio_has_feature(vi->vdev, VIRTIO_NET_F_NOTF_COAL)) {
5378 		ec->rx_coalesce_usecs = vi->intr_coal_rx.max_usecs;
5379 		ec->tx_coalesce_usecs = vi->intr_coal_tx.max_usecs;
5380 		ec->tx_max_coalesced_frames = vi->intr_coal_tx.max_packets;
5381 		ec->rx_max_coalesced_frames = vi->intr_coal_rx.max_packets;
5382 		ec->use_adaptive_rx_coalesce = vi->rx_dim_enabled;
5383 	} else {
5384 		ec->rx_max_coalesced_frames = 1;
5385 
5386 		if (vi->sq[0].napi.weight)
5387 			ec->tx_max_coalesced_frames = 1;
5388 	}
5389 
5390 	return 0;
5391 }
5392 
5393 static int virtnet_set_per_queue_coalesce(struct net_device *dev,
5394 					  u32 queue,
5395 					  struct ethtool_coalesce *ec)
5396 {
5397 	struct virtnet_info *vi = netdev_priv(dev);
5398 	int ret, napi_weight;
5399 	bool update_napi = false;
5400 
5401 	if (queue >= vi->max_queue_pairs)
5402 		return -EINVAL;
5403 
5404 	/* Can't change NAPI weight if the link is up */
5405 	napi_weight = ec->tx_max_coalesced_frames ? NAPI_POLL_WEIGHT : 0;
5406 	ret = virtnet_should_update_vq_weight(dev->flags, napi_weight,
5407 					      vi->sq[queue].napi.weight,
5408 					      &update_napi);
5409 	if (ret)
5410 		return ret;
5411 
5412 	if (virtio_has_feature(vi->vdev, VIRTIO_NET_F_VQ_NOTF_COAL))
5413 		ret = virtnet_send_notf_coal_vq_cmds(vi, ec, queue);
5414 	else
5415 		ret = virtnet_coal_params_supported(ec);
5416 
5417 	if (ret)
5418 		return ret;
5419 
5420 	if (update_napi)
5421 		vi->sq[queue].napi.weight = napi_weight;
5422 
5423 	return 0;
5424 }
5425 
5426 static int virtnet_get_per_queue_coalesce(struct net_device *dev,
5427 					  u32 queue,
5428 					  struct ethtool_coalesce *ec)
5429 {
5430 	struct virtnet_info *vi = netdev_priv(dev);
5431 
5432 	if (queue >= vi->max_queue_pairs)
5433 		return -EINVAL;
5434 
5435 	if (virtio_has_feature(vi->vdev, VIRTIO_NET_F_VQ_NOTF_COAL)) {
5436 		mutex_lock(&vi->rq[queue].dim_lock);
5437 		ec->rx_coalesce_usecs = vi->rq[queue].intr_coal.max_usecs;
5438 		ec->tx_coalesce_usecs = vi->sq[queue].intr_coal.max_usecs;
5439 		ec->tx_max_coalesced_frames = vi->sq[queue].intr_coal.max_packets;
5440 		ec->rx_max_coalesced_frames = vi->rq[queue].intr_coal.max_packets;
5441 		ec->use_adaptive_rx_coalesce = vi->rq[queue].dim_enabled;
5442 		mutex_unlock(&vi->rq[queue].dim_lock);
5443 	} else {
5444 		ec->rx_max_coalesced_frames = 1;
5445 
5446 		if (vi->sq[queue].napi.weight)
5447 			ec->tx_max_coalesced_frames = 1;
5448 	}
5449 
5450 	return 0;
5451 }
5452 
5453 static void virtnet_init_settings(struct net_device *dev)
5454 {
5455 	struct virtnet_info *vi = netdev_priv(dev);
5456 
5457 	vi->speed = SPEED_UNKNOWN;
5458 	vi->duplex = DUPLEX_UNKNOWN;
5459 }
5460 
5461 static u32 virtnet_get_rxfh_key_size(struct net_device *dev)
5462 {
5463 	return ((struct virtnet_info *)netdev_priv(dev))->rss_key_size;
5464 }
5465 
5466 static u32 virtnet_get_rxfh_indir_size(struct net_device *dev)
5467 {
5468 	return ((struct virtnet_info *)netdev_priv(dev))->rss_indir_table_size;
5469 }
5470 
5471 static int virtnet_get_rxfh(struct net_device *dev,
5472 			    struct ethtool_rxfh_param *rxfh)
5473 {
5474 	struct virtnet_info *vi = netdev_priv(dev);
5475 	int i;
5476 
5477 	if (rxfh->indir) {
5478 		for (i = 0; i < vi->rss_indir_table_size; ++i)
5479 			rxfh->indir[i] = le16_to_cpu(vi->rss_hdr->indirection_table[i]);
5480 	}
5481 
5482 	if (rxfh->key)
5483 		memcpy(rxfh->key, vi->rss_hash_key_data, vi->rss_key_size);
5484 
5485 	rxfh->hfunc = ETH_RSS_HASH_TOP;
5486 
5487 	return 0;
5488 }
5489 
5490 static int virtnet_set_rxfh(struct net_device *dev,
5491 			    struct ethtool_rxfh_param *rxfh,
5492 			    struct netlink_ext_ack *extack)
5493 {
5494 	struct virtnet_info *vi = netdev_priv(dev);
5495 	bool update = false;
5496 	int i;
5497 
5498 	if (rxfh->hfunc != ETH_RSS_HASH_NO_CHANGE &&
5499 	    rxfh->hfunc != ETH_RSS_HASH_TOP)
5500 		return -EOPNOTSUPP;
5501 
5502 	if (rxfh->indir) {
5503 		if (!vi->has_rss)
5504 			return -EOPNOTSUPP;
5505 
5506 		for (i = 0; i < vi->rss_indir_table_size; ++i)
5507 			vi->rss_hdr->indirection_table[i] = cpu_to_le16(rxfh->indir[i]);
5508 		update = true;
5509 	}
5510 
5511 	if (rxfh->key) {
5512 		/* If either _F_HASH_REPORT or _F_RSS are negotiated, the
5513 		 * device provides hash calculation capabilities, that is,
5514 		 * hash_key is configured.
5515 		 */
5516 		if (!vi->has_rss && !vi->has_rss_hash_report)
5517 			return -EOPNOTSUPP;
5518 
5519 		memcpy(vi->rss_hash_key_data, rxfh->key, vi->rss_key_size);
5520 		update = true;
5521 	}
5522 
5523 	if (update)
5524 		virtnet_commit_rss_command(vi);
5525 
5526 	return 0;
5527 }
5528 
5529 static int virtnet_get_rxnfc(struct net_device *dev, struct ethtool_rxnfc *info, u32 *rule_locs)
5530 {
5531 	struct virtnet_info *vi = netdev_priv(dev);
5532 	int rc = 0;
5533 
5534 	switch (info->cmd) {
5535 	case ETHTOOL_GRXRINGS:
5536 		info->data = vi->curr_queue_pairs;
5537 		break;
5538 	case ETHTOOL_GRXFH:
5539 		virtnet_get_hashflow(vi, info);
5540 		break;
5541 	default:
5542 		rc = -EOPNOTSUPP;
5543 	}
5544 
5545 	return rc;
5546 }
5547 
5548 static int virtnet_set_rxnfc(struct net_device *dev, struct ethtool_rxnfc *info)
5549 {
5550 	struct virtnet_info *vi = netdev_priv(dev);
5551 	int rc = 0;
5552 
5553 	switch (info->cmd) {
5554 	case ETHTOOL_SRXFH:
5555 		if (!virtnet_set_hashflow(vi, info))
5556 			rc = -EINVAL;
5557 
5558 		break;
5559 	default:
5560 		rc = -EOPNOTSUPP;
5561 	}
5562 
5563 	return rc;
5564 }
5565 
5566 static const struct ethtool_ops virtnet_ethtool_ops = {
5567 	.supported_coalesce_params = ETHTOOL_COALESCE_MAX_FRAMES |
5568 		ETHTOOL_COALESCE_USECS | ETHTOOL_COALESCE_USE_ADAPTIVE_RX,
5569 	.get_drvinfo = virtnet_get_drvinfo,
5570 	.get_link = ethtool_op_get_link,
5571 	.get_ringparam = virtnet_get_ringparam,
5572 	.set_ringparam = virtnet_set_ringparam,
5573 	.get_strings = virtnet_get_strings,
5574 	.get_sset_count = virtnet_get_sset_count,
5575 	.get_ethtool_stats = virtnet_get_ethtool_stats,
5576 	.set_channels = virtnet_set_channels,
5577 	.get_channels = virtnet_get_channels,
5578 	.get_ts_info = ethtool_op_get_ts_info,
5579 	.get_link_ksettings = virtnet_get_link_ksettings,
5580 	.set_link_ksettings = virtnet_set_link_ksettings,
5581 	.set_coalesce = virtnet_set_coalesce,
5582 	.get_coalesce = virtnet_get_coalesce,
5583 	.set_per_queue_coalesce = virtnet_set_per_queue_coalesce,
5584 	.get_per_queue_coalesce = virtnet_get_per_queue_coalesce,
5585 	.get_rxfh_key_size = virtnet_get_rxfh_key_size,
5586 	.get_rxfh_indir_size = virtnet_get_rxfh_indir_size,
5587 	.get_rxfh = virtnet_get_rxfh,
5588 	.set_rxfh = virtnet_set_rxfh,
5589 	.get_rxnfc = virtnet_get_rxnfc,
5590 	.set_rxnfc = virtnet_set_rxnfc,
5591 };
5592 
5593 static void virtnet_get_queue_stats_rx(struct net_device *dev, int i,
5594 				       struct netdev_queue_stats_rx *stats)
5595 {
5596 	struct virtnet_info *vi = netdev_priv(dev);
5597 	struct receive_queue *rq = &vi->rq[i];
5598 	struct virtnet_stats_ctx ctx = {0};
5599 
5600 	virtnet_stats_ctx_init(vi, &ctx, (void *)stats, true);
5601 
5602 	virtnet_get_hw_stats(vi, &ctx, i * 2);
5603 	virtnet_fill_stats(vi, i * 2, &ctx, (void *)&rq->stats, true, 0);
5604 }
5605 
5606 static void virtnet_get_queue_stats_tx(struct net_device *dev, int i,
5607 				       struct netdev_queue_stats_tx *stats)
5608 {
5609 	struct virtnet_info *vi = netdev_priv(dev);
5610 	struct send_queue *sq = &vi->sq[i];
5611 	struct virtnet_stats_ctx ctx = {0};
5612 
5613 	virtnet_stats_ctx_init(vi, &ctx, (void *)stats, true);
5614 
5615 	virtnet_get_hw_stats(vi, &ctx, i * 2 + 1);
5616 	virtnet_fill_stats(vi, i * 2 + 1, &ctx, (void *)&sq->stats, true, 0);
5617 }
5618 
5619 static void virtnet_get_base_stats(struct net_device *dev,
5620 				   struct netdev_queue_stats_rx *rx,
5621 				   struct netdev_queue_stats_tx *tx)
5622 {
5623 	struct virtnet_info *vi = netdev_priv(dev);
5624 
5625 	/* The queue stats of the virtio-net will not be reset. So here we
5626 	 * return 0.
5627 	 */
5628 	rx->bytes = 0;
5629 	rx->packets = 0;
5630 
5631 	if (vi->device_stats_cap & VIRTIO_NET_STATS_TYPE_RX_BASIC) {
5632 		rx->hw_drops = 0;
5633 		rx->hw_drop_overruns = 0;
5634 	}
5635 
5636 	if (vi->device_stats_cap & VIRTIO_NET_STATS_TYPE_RX_CSUM) {
5637 		rx->csum_unnecessary = 0;
5638 		rx->csum_none = 0;
5639 		rx->csum_bad = 0;
5640 	}
5641 
5642 	if (vi->device_stats_cap & VIRTIO_NET_STATS_TYPE_RX_GSO) {
5643 		rx->hw_gro_packets = 0;
5644 		rx->hw_gro_bytes = 0;
5645 		rx->hw_gro_wire_packets = 0;
5646 		rx->hw_gro_wire_bytes = 0;
5647 	}
5648 
5649 	if (vi->device_stats_cap & VIRTIO_NET_STATS_TYPE_RX_SPEED)
5650 		rx->hw_drop_ratelimits = 0;
5651 
5652 	tx->bytes = 0;
5653 	tx->packets = 0;
5654 	tx->stop = 0;
5655 	tx->wake = 0;
5656 
5657 	if (vi->device_stats_cap & VIRTIO_NET_STATS_TYPE_TX_BASIC) {
5658 		tx->hw_drops = 0;
5659 		tx->hw_drop_errors = 0;
5660 	}
5661 
5662 	if (vi->device_stats_cap & VIRTIO_NET_STATS_TYPE_TX_CSUM) {
5663 		tx->csum_none = 0;
5664 		tx->needs_csum = 0;
5665 	}
5666 
5667 	if (vi->device_stats_cap & VIRTIO_NET_STATS_TYPE_TX_GSO) {
5668 		tx->hw_gso_packets = 0;
5669 		tx->hw_gso_bytes = 0;
5670 		tx->hw_gso_wire_packets = 0;
5671 		tx->hw_gso_wire_bytes = 0;
5672 	}
5673 
5674 	if (vi->device_stats_cap & VIRTIO_NET_STATS_TYPE_TX_SPEED)
5675 		tx->hw_drop_ratelimits = 0;
5676 }
5677 
5678 static const struct netdev_stat_ops virtnet_stat_ops = {
5679 	.get_queue_stats_rx	= virtnet_get_queue_stats_rx,
5680 	.get_queue_stats_tx	= virtnet_get_queue_stats_tx,
5681 	.get_base_stats		= virtnet_get_base_stats,
5682 };
5683 
5684 static void virtnet_freeze_down(struct virtio_device *vdev)
5685 {
5686 	struct virtnet_info *vi = vdev->priv;
5687 
5688 	/* Make sure no work handler is accessing the device */
5689 	flush_work(&vi->config_work);
5690 	disable_rx_mode_work(vi);
5691 	flush_work(&vi->rx_mode_work);
5692 
5693 	netif_tx_lock_bh(vi->dev);
5694 	netif_device_detach(vi->dev);
5695 	netif_tx_unlock_bh(vi->dev);
5696 	if (netif_running(vi->dev)) {
5697 		rtnl_lock();
5698 		virtnet_close(vi->dev);
5699 		rtnl_unlock();
5700 	}
5701 }
5702 
5703 static int init_vqs(struct virtnet_info *vi);
5704 
5705 static int virtnet_restore_up(struct virtio_device *vdev)
5706 {
5707 	struct virtnet_info *vi = vdev->priv;
5708 	int err;
5709 
5710 	err = init_vqs(vi);
5711 	if (err)
5712 		return err;
5713 
5714 	virtio_device_ready(vdev);
5715 
5716 	enable_delayed_refill(vi);
5717 	enable_rx_mode_work(vi);
5718 
5719 	if (netif_running(vi->dev)) {
5720 		rtnl_lock();
5721 		err = virtnet_open(vi->dev);
5722 		rtnl_unlock();
5723 		if (err)
5724 			return err;
5725 	}
5726 
5727 	netif_tx_lock_bh(vi->dev);
5728 	netif_device_attach(vi->dev);
5729 	netif_tx_unlock_bh(vi->dev);
5730 	return err;
5731 }
5732 
5733 static int virtnet_set_guest_offloads(struct virtnet_info *vi, u64 offloads)
5734 {
5735 	__virtio64 *_offloads __free(kfree) = NULL;
5736 	struct scatterlist sg;
5737 
5738 	_offloads = kzalloc(sizeof(*_offloads), GFP_KERNEL);
5739 	if (!_offloads)
5740 		return -ENOMEM;
5741 
5742 	*_offloads = cpu_to_virtio64(vi->vdev, offloads);
5743 
5744 	sg_init_one(&sg, _offloads, sizeof(*_offloads));
5745 
5746 	if (!virtnet_send_command(vi, VIRTIO_NET_CTRL_GUEST_OFFLOADS,
5747 				  VIRTIO_NET_CTRL_GUEST_OFFLOADS_SET, &sg)) {
5748 		dev_warn(&vi->dev->dev, "Fail to set guest offload.\n");
5749 		return -EINVAL;
5750 	}
5751 
5752 	return 0;
5753 }
5754 
5755 static int virtnet_clear_guest_offloads(struct virtnet_info *vi)
5756 {
5757 	u64 offloads = 0;
5758 
5759 	if (!vi->guest_offloads)
5760 		return 0;
5761 
5762 	return virtnet_set_guest_offloads(vi, offloads);
5763 }
5764 
5765 static int virtnet_restore_guest_offloads(struct virtnet_info *vi)
5766 {
5767 	u64 offloads = vi->guest_offloads;
5768 
5769 	if (!vi->guest_offloads)
5770 		return 0;
5771 
5772 	return virtnet_set_guest_offloads(vi, offloads);
5773 }
5774 
5775 static int virtnet_rq_bind_xsk_pool(struct virtnet_info *vi, struct receive_queue *rq,
5776 				    struct xsk_buff_pool *pool)
5777 {
5778 	int err, qindex;
5779 
5780 	qindex = rq - vi->rq;
5781 
5782 	if (pool) {
5783 		err = xdp_rxq_info_reg(&rq->xsk_rxq_info, vi->dev, qindex, rq->napi.napi_id);
5784 		if (err < 0)
5785 			return err;
5786 
5787 		err = xdp_rxq_info_reg_mem_model(&rq->xsk_rxq_info,
5788 						 MEM_TYPE_XSK_BUFF_POOL, NULL);
5789 		if (err < 0)
5790 			goto unreg;
5791 
5792 		xsk_pool_set_rxq_info(pool, &rq->xsk_rxq_info);
5793 	}
5794 
5795 	virtnet_rx_pause(vi, rq);
5796 
5797 	err = virtqueue_reset(rq->vq, virtnet_rq_unmap_free_buf, NULL);
5798 	if (err) {
5799 		netdev_err(vi->dev, "reset rx fail: rx queue index: %d err: %d\n", qindex, err);
5800 
5801 		pool = NULL;
5802 	}
5803 
5804 	rq->xsk_pool = pool;
5805 
5806 	virtnet_rx_resume(vi, rq);
5807 
5808 	if (pool)
5809 		return 0;
5810 
5811 unreg:
5812 	xdp_rxq_info_unreg(&rq->xsk_rxq_info);
5813 	return err;
5814 }
5815 
5816 static int virtnet_sq_bind_xsk_pool(struct virtnet_info *vi,
5817 				    struct send_queue *sq,
5818 				    struct xsk_buff_pool *pool)
5819 {
5820 	int err, qindex;
5821 
5822 	qindex = sq - vi->sq;
5823 
5824 	virtnet_tx_pause(vi, sq);
5825 
5826 	err = virtqueue_reset(sq->vq, virtnet_sq_free_unused_buf,
5827 			      virtnet_sq_free_unused_buf_done);
5828 	if (err) {
5829 		netdev_err(vi->dev, "reset tx fail: tx queue index: %d err: %d\n", qindex, err);
5830 		pool = NULL;
5831 	}
5832 
5833 	sq->xsk_pool = pool;
5834 
5835 	virtnet_tx_resume(vi, sq);
5836 
5837 	return err;
5838 }
5839 
5840 static int virtnet_xsk_pool_enable(struct net_device *dev,
5841 				   struct xsk_buff_pool *pool,
5842 				   u16 qid)
5843 {
5844 	struct virtnet_info *vi = netdev_priv(dev);
5845 	struct receive_queue *rq;
5846 	struct device *dma_dev;
5847 	struct send_queue *sq;
5848 	dma_addr_t hdr_dma;
5849 	int err, size;
5850 
5851 	if (vi->hdr_len > xsk_pool_get_headroom(pool))
5852 		return -EINVAL;
5853 
5854 	/* In big_packets mode, xdp cannot work, so there is no need to
5855 	 * initialize xsk of rq.
5856 	 */
5857 	if (vi->big_packets && !vi->mergeable_rx_bufs)
5858 		return -ENOENT;
5859 
5860 	if (qid >= vi->curr_queue_pairs)
5861 		return -EINVAL;
5862 
5863 	sq = &vi->sq[qid];
5864 	rq = &vi->rq[qid];
5865 
5866 	/* xsk assumes that tx and rx must have the same dma device. The af-xdp
5867 	 * may use one buffer to receive from the rx and reuse this buffer to
5868 	 * send by the tx. So the dma dev of sq and rq must be the same one.
5869 	 *
5870 	 * But vq->dma_dev allows every vq has the respective dma dev. So I
5871 	 * check the dma dev of vq and sq is the same dev.
5872 	 */
5873 	if (virtqueue_dma_dev(rq->vq) != virtqueue_dma_dev(sq->vq))
5874 		return -EINVAL;
5875 
5876 	dma_dev = virtqueue_dma_dev(rq->vq);
5877 	if (!dma_dev)
5878 		return -EINVAL;
5879 
5880 	size = virtqueue_get_vring_size(rq->vq);
5881 
5882 	rq->xsk_buffs = kvcalloc(size, sizeof(*rq->xsk_buffs), GFP_KERNEL);
5883 	if (!rq->xsk_buffs)
5884 		return -ENOMEM;
5885 
5886 	hdr_dma = virtqueue_dma_map_single_attrs(sq->vq, &xsk_hdr, vi->hdr_len,
5887 						 DMA_TO_DEVICE, 0);
5888 	if (virtqueue_dma_mapping_error(sq->vq, hdr_dma))
5889 		return -ENOMEM;
5890 
5891 	err = xsk_pool_dma_map(pool, dma_dev, 0);
5892 	if (err)
5893 		goto err_xsk_map;
5894 
5895 	err = virtnet_rq_bind_xsk_pool(vi, rq, pool);
5896 	if (err)
5897 		goto err_rq;
5898 
5899 	err = virtnet_sq_bind_xsk_pool(vi, sq, pool);
5900 	if (err)
5901 		goto err_sq;
5902 
5903 	/* Now, we do not support tx offload(such as tx csum), so all the tx
5904 	 * virtnet hdr is zero. So all the tx packets can share a single hdr.
5905 	 */
5906 	sq->xsk_hdr_dma_addr = hdr_dma;
5907 
5908 	return 0;
5909 
5910 err_sq:
5911 	virtnet_rq_bind_xsk_pool(vi, rq, NULL);
5912 err_rq:
5913 	xsk_pool_dma_unmap(pool, 0);
5914 err_xsk_map:
5915 	virtqueue_dma_unmap_single_attrs(rq->vq, hdr_dma, vi->hdr_len,
5916 					 DMA_TO_DEVICE, 0);
5917 	return err;
5918 }
5919 
5920 static int virtnet_xsk_pool_disable(struct net_device *dev, u16 qid)
5921 {
5922 	struct virtnet_info *vi = netdev_priv(dev);
5923 	struct xsk_buff_pool *pool;
5924 	struct receive_queue *rq;
5925 	struct send_queue *sq;
5926 	int err;
5927 
5928 	if (qid >= vi->curr_queue_pairs)
5929 		return -EINVAL;
5930 
5931 	sq = &vi->sq[qid];
5932 	rq = &vi->rq[qid];
5933 
5934 	pool = rq->xsk_pool;
5935 
5936 	err = virtnet_rq_bind_xsk_pool(vi, rq, NULL);
5937 	err |= virtnet_sq_bind_xsk_pool(vi, sq, NULL);
5938 
5939 	xsk_pool_dma_unmap(pool, 0);
5940 
5941 	virtqueue_dma_unmap_single_attrs(sq->vq, sq->xsk_hdr_dma_addr,
5942 					 vi->hdr_len, DMA_TO_DEVICE, 0);
5943 	kvfree(rq->xsk_buffs);
5944 
5945 	return err;
5946 }
5947 
5948 static int virtnet_xsk_pool_setup(struct net_device *dev, struct netdev_bpf *xdp)
5949 {
5950 	if (xdp->xsk.pool)
5951 		return virtnet_xsk_pool_enable(dev, xdp->xsk.pool,
5952 					       xdp->xsk.queue_id);
5953 	else
5954 		return virtnet_xsk_pool_disable(dev, xdp->xsk.queue_id);
5955 }
5956 
5957 static int virtnet_xdp_set(struct net_device *dev, struct bpf_prog *prog,
5958 			   struct netlink_ext_ack *extack)
5959 {
5960 	unsigned int room = SKB_DATA_ALIGN(XDP_PACKET_HEADROOM +
5961 					   sizeof(struct skb_shared_info));
5962 	unsigned int max_sz = PAGE_SIZE - room - ETH_HLEN;
5963 	struct virtnet_info *vi = netdev_priv(dev);
5964 	struct bpf_prog *old_prog;
5965 	u16 xdp_qp = 0, curr_qp;
5966 	int i, err;
5967 
5968 	if (!virtio_has_feature(vi->vdev, VIRTIO_NET_F_CTRL_GUEST_OFFLOADS)
5969 	    && (virtio_has_feature(vi->vdev, VIRTIO_NET_F_GUEST_TSO4) ||
5970 	        virtio_has_feature(vi->vdev, VIRTIO_NET_F_GUEST_TSO6) ||
5971 	        virtio_has_feature(vi->vdev, VIRTIO_NET_F_GUEST_ECN) ||
5972 		virtio_has_feature(vi->vdev, VIRTIO_NET_F_GUEST_UFO) ||
5973 		virtio_has_feature(vi->vdev, VIRTIO_NET_F_GUEST_CSUM) ||
5974 		virtio_has_feature(vi->vdev, VIRTIO_NET_F_GUEST_USO4) ||
5975 		virtio_has_feature(vi->vdev, VIRTIO_NET_F_GUEST_USO6))) {
5976 		NL_SET_ERR_MSG_MOD(extack, "Can't set XDP while host is implementing GRO_HW/CSUM, disable GRO_HW/CSUM first");
5977 		return -EOPNOTSUPP;
5978 	}
5979 
5980 	if (vi->mergeable_rx_bufs && !vi->any_header_sg) {
5981 		NL_SET_ERR_MSG_MOD(extack, "XDP expects header/data in single page, any_header_sg required");
5982 		return -EINVAL;
5983 	}
5984 
5985 	if (prog && !prog->aux->xdp_has_frags && dev->mtu > max_sz) {
5986 		NL_SET_ERR_MSG_MOD(extack, "MTU too large to enable XDP without frags");
5987 		netdev_warn(dev, "single-buffer XDP requires MTU less than %u\n", max_sz);
5988 		return -EINVAL;
5989 	}
5990 
5991 	curr_qp = vi->curr_queue_pairs - vi->xdp_queue_pairs;
5992 	if (prog)
5993 		xdp_qp = nr_cpu_ids;
5994 
5995 	/* XDP requires extra queues for XDP_TX */
5996 	if (curr_qp + xdp_qp > vi->max_queue_pairs) {
5997 		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",
5998 				 curr_qp + xdp_qp, vi->max_queue_pairs);
5999 		xdp_qp = 0;
6000 	}
6001 
6002 	old_prog = rtnl_dereference(vi->rq[0].xdp_prog);
6003 	if (!prog && !old_prog)
6004 		return 0;
6005 
6006 	if (prog)
6007 		bpf_prog_add(prog, vi->max_queue_pairs - 1);
6008 
6009 	virtnet_rx_pause_all(vi);
6010 
6011 	/* Make sure NAPI is not using any XDP TX queues for RX. */
6012 	if (netif_running(dev)) {
6013 		for (i = 0; i < vi->max_queue_pairs; i++)
6014 			virtnet_napi_tx_disable(&vi->sq[i]);
6015 	}
6016 
6017 	if (!prog) {
6018 		for (i = 0; i < vi->max_queue_pairs; i++) {
6019 			rcu_assign_pointer(vi->rq[i].xdp_prog, prog);
6020 			if (i == 0)
6021 				virtnet_restore_guest_offloads(vi);
6022 		}
6023 		synchronize_net();
6024 	}
6025 
6026 	err = virtnet_set_queues(vi, curr_qp + xdp_qp);
6027 	if (err)
6028 		goto err;
6029 	netif_set_real_num_rx_queues(dev, curr_qp + xdp_qp);
6030 	vi->xdp_queue_pairs = xdp_qp;
6031 
6032 	if (prog) {
6033 		vi->xdp_enabled = true;
6034 		for (i = 0; i < vi->max_queue_pairs; i++) {
6035 			rcu_assign_pointer(vi->rq[i].xdp_prog, prog);
6036 			if (i == 0 && !old_prog)
6037 				virtnet_clear_guest_offloads(vi);
6038 		}
6039 		if (!old_prog)
6040 			xdp_features_set_redirect_target(dev, true);
6041 	} else {
6042 		xdp_features_clear_redirect_target(dev);
6043 		vi->xdp_enabled = false;
6044 	}
6045 
6046 	virtnet_rx_resume_all(vi);
6047 	for (i = 0; i < vi->max_queue_pairs; i++) {
6048 		if (old_prog)
6049 			bpf_prog_put(old_prog);
6050 		if (netif_running(dev))
6051 			virtnet_napi_tx_enable(&vi->sq[i]);
6052 	}
6053 
6054 	return 0;
6055 
6056 err:
6057 	if (!prog) {
6058 		virtnet_clear_guest_offloads(vi);
6059 		for (i = 0; i < vi->max_queue_pairs; i++)
6060 			rcu_assign_pointer(vi->rq[i].xdp_prog, old_prog);
6061 	}
6062 
6063 	virtnet_rx_resume_all(vi);
6064 	if (netif_running(dev)) {
6065 		for (i = 0; i < vi->max_queue_pairs; i++)
6066 			virtnet_napi_tx_enable(&vi->sq[i]);
6067 	}
6068 	if (prog)
6069 		bpf_prog_sub(prog, vi->max_queue_pairs - 1);
6070 	return err;
6071 }
6072 
6073 static int virtnet_xdp(struct net_device *dev, struct netdev_bpf *xdp)
6074 {
6075 	switch (xdp->command) {
6076 	case XDP_SETUP_PROG:
6077 		return virtnet_xdp_set(dev, xdp->prog, xdp->extack);
6078 	case XDP_SETUP_XSK_POOL:
6079 		return virtnet_xsk_pool_setup(dev, xdp);
6080 	default:
6081 		return -EINVAL;
6082 	}
6083 }
6084 
6085 static int virtnet_get_phys_port_name(struct net_device *dev, char *buf,
6086 				      size_t len)
6087 {
6088 	struct virtnet_info *vi = netdev_priv(dev);
6089 	int ret;
6090 
6091 	if (!virtio_has_feature(vi->vdev, VIRTIO_NET_F_STANDBY))
6092 		return -EOPNOTSUPP;
6093 
6094 	ret = snprintf(buf, len, "sby");
6095 	if (ret >= len)
6096 		return -EOPNOTSUPP;
6097 
6098 	return 0;
6099 }
6100 
6101 static int virtnet_set_features(struct net_device *dev,
6102 				netdev_features_t features)
6103 {
6104 	struct virtnet_info *vi = netdev_priv(dev);
6105 	u64 offloads;
6106 	int err;
6107 
6108 	if ((dev->features ^ features) & NETIF_F_GRO_HW) {
6109 		if (vi->xdp_enabled)
6110 			return -EBUSY;
6111 
6112 		if (features & NETIF_F_GRO_HW)
6113 			offloads = vi->guest_offloads_capable;
6114 		else
6115 			offloads = vi->guest_offloads_capable &
6116 				   ~GUEST_OFFLOAD_GRO_HW_MASK;
6117 
6118 		err = virtnet_set_guest_offloads(vi, offloads);
6119 		if (err)
6120 			return err;
6121 		vi->guest_offloads = offloads;
6122 	}
6123 
6124 	if ((dev->features ^ features) & NETIF_F_RXHASH) {
6125 		if (features & NETIF_F_RXHASH)
6126 			vi->rss_hdr->hash_types = cpu_to_le32(vi->rss_hash_types_saved);
6127 		else
6128 			vi->rss_hdr->hash_types = cpu_to_le32(VIRTIO_NET_HASH_REPORT_NONE);
6129 
6130 		if (!virtnet_commit_rss_command(vi))
6131 			return -EINVAL;
6132 	}
6133 
6134 	return 0;
6135 }
6136 
6137 static void virtnet_tx_timeout(struct net_device *dev, unsigned int txqueue)
6138 {
6139 	struct virtnet_info *priv = netdev_priv(dev);
6140 	struct send_queue *sq = &priv->sq[txqueue];
6141 	struct netdev_queue *txq = netdev_get_tx_queue(dev, txqueue);
6142 
6143 	u64_stats_update_begin(&sq->stats.syncp);
6144 	u64_stats_inc(&sq->stats.tx_timeouts);
6145 	u64_stats_update_end(&sq->stats.syncp);
6146 
6147 	netdev_err(dev, "TX timeout on queue: %u, sq: %s, vq: 0x%x, name: %s, %u usecs ago\n",
6148 		   txqueue, sq->name, sq->vq->index, sq->vq->name,
6149 		   jiffies_to_usecs(jiffies - READ_ONCE(txq->trans_start)));
6150 }
6151 
6152 static int virtnet_init_irq_moder(struct virtnet_info *vi)
6153 {
6154 	u8 profile_flags = 0, coal_flags = 0;
6155 	int ret, i;
6156 
6157 	profile_flags |= DIM_PROFILE_RX;
6158 	coal_flags |= DIM_COALESCE_USEC | DIM_COALESCE_PKTS;
6159 	ret = net_dim_init_irq_moder(vi->dev, profile_flags, coal_flags,
6160 				     DIM_CQ_PERIOD_MODE_START_FROM_EQE,
6161 				     0, virtnet_rx_dim_work, NULL);
6162 
6163 	if (ret)
6164 		return ret;
6165 
6166 	for (i = 0; i < vi->max_queue_pairs; i++)
6167 		net_dim_setting(vi->dev, &vi->rq[i].dim, false);
6168 
6169 	return 0;
6170 }
6171 
6172 static void virtnet_free_irq_moder(struct virtnet_info *vi)
6173 {
6174 	if (!virtio_has_feature(vi->vdev, VIRTIO_NET_F_VQ_NOTF_COAL))
6175 		return;
6176 
6177 	rtnl_lock();
6178 	net_dim_free_irq_moder(vi->dev);
6179 	rtnl_unlock();
6180 }
6181 
6182 static const struct net_device_ops virtnet_netdev = {
6183 	.ndo_open            = virtnet_open,
6184 	.ndo_stop   	     = virtnet_close,
6185 	.ndo_start_xmit      = start_xmit,
6186 	.ndo_validate_addr   = eth_validate_addr,
6187 	.ndo_set_mac_address = virtnet_set_mac_address,
6188 	.ndo_set_rx_mode     = virtnet_set_rx_mode,
6189 	.ndo_get_stats64     = virtnet_stats,
6190 	.ndo_vlan_rx_add_vid = virtnet_vlan_rx_add_vid,
6191 	.ndo_vlan_rx_kill_vid = virtnet_vlan_rx_kill_vid,
6192 	.ndo_bpf		= virtnet_xdp,
6193 	.ndo_xdp_xmit		= virtnet_xdp_xmit,
6194 	.ndo_xsk_wakeup         = virtnet_xsk_wakeup,
6195 	.ndo_features_check	= passthru_features_check,
6196 	.ndo_get_phys_port_name	= virtnet_get_phys_port_name,
6197 	.ndo_set_features	= virtnet_set_features,
6198 	.ndo_tx_timeout		= virtnet_tx_timeout,
6199 };
6200 
6201 static void virtnet_config_changed_work(struct work_struct *work)
6202 {
6203 	struct virtnet_info *vi =
6204 		container_of(work, struct virtnet_info, config_work);
6205 	u16 v;
6206 
6207 	if (virtio_cread_feature(vi->vdev, VIRTIO_NET_F_STATUS,
6208 				 struct virtio_net_config, status, &v) < 0)
6209 		return;
6210 
6211 	if (v & VIRTIO_NET_S_ANNOUNCE) {
6212 		netdev_notify_peers(vi->dev);
6213 		virtnet_ack_link_announce(vi);
6214 	}
6215 
6216 	/* Ignore unknown (future) status bits */
6217 	v &= VIRTIO_NET_S_LINK_UP;
6218 
6219 	if (vi->status == v)
6220 		return;
6221 
6222 	vi->status = v;
6223 
6224 	if (vi->status & VIRTIO_NET_S_LINK_UP) {
6225 		virtnet_update_settings(vi);
6226 		netif_carrier_on(vi->dev);
6227 		netif_tx_wake_all_queues(vi->dev);
6228 	} else {
6229 		netif_carrier_off(vi->dev);
6230 		netif_tx_stop_all_queues(vi->dev);
6231 	}
6232 }
6233 
6234 static void virtnet_config_changed(struct virtio_device *vdev)
6235 {
6236 	struct virtnet_info *vi = vdev->priv;
6237 
6238 	schedule_work(&vi->config_work);
6239 }
6240 
6241 static void virtnet_free_queues(struct virtnet_info *vi)
6242 {
6243 	int i;
6244 
6245 	for (i = 0; i < vi->max_queue_pairs; i++) {
6246 		__netif_napi_del(&vi->rq[i].napi);
6247 		__netif_napi_del(&vi->sq[i].napi);
6248 	}
6249 
6250 	/* We called __netif_napi_del(),
6251 	 * we need to respect an RCU grace period before freeing vi->rq
6252 	 */
6253 	synchronize_net();
6254 
6255 	kfree(vi->rq);
6256 	kfree(vi->sq);
6257 	kfree(vi->ctrl);
6258 }
6259 
6260 static void _free_receive_bufs(struct virtnet_info *vi)
6261 {
6262 	struct bpf_prog *old_prog;
6263 	int i;
6264 
6265 	for (i = 0; i < vi->max_queue_pairs; i++) {
6266 		while (vi->rq[i].pages)
6267 			__free_pages(get_a_page(&vi->rq[i], GFP_KERNEL), 0);
6268 
6269 		old_prog = rtnl_dereference(vi->rq[i].xdp_prog);
6270 		RCU_INIT_POINTER(vi->rq[i].xdp_prog, NULL);
6271 		if (old_prog)
6272 			bpf_prog_put(old_prog);
6273 	}
6274 }
6275 
6276 static void free_receive_bufs(struct virtnet_info *vi)
6277 {
6278 	rtnl_lock();
6279 	_free_receive_bufs(vi);
6280 	rtnl_unlock();
6281 }
6282 
6283 static void free_receive_page_frags(struct virtnet_info *vi)
6284 {
6285 	int i;
6286 	for (i = 0; i < vi->max_queue_pairs; i++)
6287 		if (vi->rq[i].alloc_frag.page) {
6288 			if (vi->rq[i].last_dma)
6289 				virtnet_rq_unmap(&vi->rq[i], vi->rq[i].last_dma, 0);
6290 			put_page(vi->rq[i].alloc_frag.page);
6291 		}
6292 }
6293 
6294 static void virtnet_sq_free_unused_buf(struct virtqueue *vq, void *buf)
6295 {
6296 	struct virtnet_info *vi = vq->vdev->priv;
6297 	struct send_queue *sq;
6298 	int i = vq2txq(vq);
6299 
6300 	sq = &vi->sq[i];
6301 
6302 	switch (virtnet_xmit_ptr_unpack(&buf)) {
6303 	case VIRTNET_XMIT_TYPE_SKB:
6304 	case VIRTNET_XMIT_TYPE_SKB_ORPHAN:
6305 		dev_kfree_skb(buf);
6306 		break;
6307 
6308 	case VIRTNET_XMIT_TYPE_XDP:
6309 		xdp_return_frame(buf);
6310 		break;
6311 
6312 	case VIRTNET_XMIT_TYPE_XSK:
6313 		xsk_tx_completed(sq->xsk_pool, 1);
6314 		break;
6315 	}
6316 }
6317 
6318 static void virtnet_sq_free_unused_buf_done(struct virtqueue *vq)
6319 {
6320 	struct virtnet_info *vi = vq->vdev->priv;
6321 	int i = vq2txq(vq);
6322 
6323 	netdev_tx_reset_queue(netdev_get_tx_queue(vi->dev, i));
6324 }
6325 
6326 static void free_unused_bufs(struct virtnet_info *vi)
6327 {
6328 	void *buf;
6329 	int i;
6330 
6331 	for (i = 0; i < vi->max_queue_pairs; i++) {
6332 		struct virtqueue *vq = vi->sq[i].vq;
6333 		while ((buf = virtqueue_detach_unused_buf(vq)) != NULL)
6334 			virtnet_sq_free_unused_buf(vq, buf);
6335 		cond_resched();
6336 	}
6337 
6338 	for (i = 0; i < vi->max_queue_pairs; i++) {
6339 		struct virtqueue *vq = vi->rq[i].vq;
6340 
6341 		while ((buf = virtqueue_detach_unused_buf(vq)) != NULL)
6342 			virtnet_rq_unmap_free_buf(vq, buf);
6343 		cond_resched();
6344 	}
6345 }
6346 
6347 static void virtnet_del_vqs(struct virtnet_info *vi)
6348 {
6349 	struct virtio_device *vdev = vi->vdev;
6350 
6351 	virtnet_clean_affinity(vi);
6352 
6353 	vdev->config->del_vqs(vdev);
6354 
6355 	virtnet_free_queues(vi);
6356 }
6357 
6358 /* How large should a single buffer be so a queue full of these can fit at
6359  * least one full packet?
6360  * Logic below assumes the mergeable buffer header is used.
6361  */
6362 static unsigned int mergeable_min_buf_len(struct virtnet_info *vi, struct virtqueue *vq)
6363 {
6364 	const unsigned int hdr_len = vi->hdr_len;
6365 	unsigned int rq_size = virtqueue_get_vring_size(vq);
6366 	unsigned int packet_len = vi->big_packets ? IP_MAX_MTU : vi->dev->max_mtu;
6367 	unsigned int buf_len = hdr_len + ETH_HLEN + VLAN_HLEN + packet_len;
6368 	unsigned int min_buf_len = DIV_ROUND_UP(buf_len, rq_size);
6369 
6370 	return max(max(min_buf_len, hdr_len) - hdr_len,
6371 		   (unsigned int)GOOD_PACKET_LEN);
6372 }
6373 
6374 static int virtnet_find_vqs(struct virtnet_info *vi)
6375 {
6376 	struct virtqueue_info *vqs_info;
6377 	struct virtqueue **vqs;
6378 	int ret = -ENOMEM;
6379 	int total_vqs;
6380 	bool *ctx;
6381 	u16 i;
6382 
6383 	/* We expect 1 RX virtqueue followed by 1 TX virtqueue, followed by
6384 	 * possible N-1 RX/TX queue pairs used in multiqueue mode, followed by
6385 	 * possible control vq.
6386 	 */
6387 	total_vqs = vi->max_queue_pairs * 2 +
6388 		    virtio_has_feature(vi->vdev, VIRTIO_NET_F_CTRL_VQ);
6389 
6390 	/* Allocate space for find_vqs parameters */
6391 	vqs = kcalloc(total_vqs, sizeof(*vqs), GFP_KERNEL);
6392 	if (!vqs)
6393 		goto err_vq;
6394 	vqs_info = kcalloc(total_vqs, sizeof(*vqs_info), GFP_KERNEL);
6395 	if (!vqs_info)
6396 		goto err_vqs_info;
6397 	if (!vi->big_packets || vi->mergeable_rx_bufs) {
6398 		ctx = kcalloc(total_vqs, sizeof(*ctx), GFP_KERNEL);
6399 		if (!ctx)
6400 			goto err_ctx;
6401 	} else {
6402 		ctx = NULL;
6403 	}
6404 
6405 	/* Parameters for control virtqueue, if any */
6406 	if (vi->has_cvq) {
6407 		vqs_info[total_vqs - 1].name = "control";
6408 	}
6409 
6410 	/* Allocate/initialize parameters for send/receive virtqueues */
6411 	for (i = 0; i < vi->max_queue_pairs; i++) {
6412 		vqs_info[rxq2vq(i)].callback = skb_recv_done;
6413 		vqs_info[txq2vq(i)].callback = skb_xmit_done;
6414 		sprintf(vi->rq[i].name, "input.%u", i);
6415 		sprintf(vi->sq[i].name, "output.%u", i);
6416 		vqs_info[rxq2vq(i)].name = vi->rq[i].name;
6417 		vqs_info[txq2vq(i)].name = vi->sq[i].name;
6418 		if (ctx)
6419 			vqs_info[rxq2vq(i)].ctx = true;
6420 	}
6421 
6422 	ret = virtio_find_vqs(vi->vdev, total_vqs, vqs, vqs_info, NULL);
6423 	if (ret)
6424 		goto err_find;
6425 
6426 	if (vi->has_cvq) {
6427 		vi->cvq = vqs[total_vqs - 1];
6428 		if (virtio_has_feature(vi->vdev, VIRTIO_NET_F_CTRL_VLAN))
6429 			vi->dev->features |= NETIF_F_HW_VLAN_CTAG_FILTER;
6430 	}
6431 
6432 	for (i = 0; i < vi->max_queue_pairs; i++) {
6433 		vi->rq[i].vq = vqs[rxq2vq(i)];
6434 		vi->rq[i].min_buf_len = mergeable_min_buf_len(vi, vi->rq[i].vq);
6435 		vi->sq[i].vq = vqs[txq2vq(i)];
6436 	}
6437 
6438 	/* run here: ret == 0. */
6439 
6440 
6441 err_find:
6442 	kfree(ctx);
6443 err_ctx:
6444 	kfree(vqs_info);
6445 err_vqs_info:
6446 	kfree(vqs);
6447 err_vq:
6448 	return ret;
6449 }
6450 
6451 static int virtnet_alloc_queues(struct virtnet_info *vi)
6452 {
6453 	int i;
6454 
6455 	if (vi->has_cvq) {
6456 		vi->ctrl = kzalloc(sizeof(*vi->ctrl), GFP_KERNEL);
6457 		if (!vi->ctrl)
6458 			goto err_ctrl;
6459 	} else {
6460 		vi->ctrl = NULL;
6461 	}
6462 	vi->sq = kcalloc(vi->max_queue_pairs, sizeof(*vi->sq), GFP_KERNEL);
6463 	if (!vi->sq)
6464 		goto err_sq;
6465 	vi->rq = kcalloc(vi->max_queue_pairs, sizeof(*vi->rq), GFP_KERNEL);
6466 	if (!vi->rq)
6467 		goto err_rq;
6468 
6469 	INIT_DELAYED_WORK(&vi->refill, refill_work);
6470 	for (i = 0; i < vi->max_queue_pairs; i++) {
6471 		vi->rq[i].pages = NULL;
6472 		netif_napi_add_config(vi->dev, &vi->rq[i].napi, virtnet_poll,
6473 				      i);
6474 		vi->rq[i].napi.weight = napi_weight;
6475 		netif_napi_add_tx_weight(vi->dev, &vi->sq[i].napi,
6476 					 virtnet_poll_tx,
6477 					 napi_tx ? napi_weight : 0);
6478 
6479 		sg_init_table(vi->rq[i].sg, ARRAY_SIZE(vi->rq[i].sg));
6480 		ewma_pkt_len_init(&vi->rq[i].mrg_avg_pkt_len);
6481 		sg_init_table(vi->sq[i].sg, ARRAY_SIZE(vi->sq[i].sg));
6482 
6483 		u64_stats_init(&vi->rq[i].stats.syncp);
6484 		u64_stats_init(&vi->sq[i].stats.syncp);
6485 		mutex_init(&vi->rq[i].dim_lock);
6486 	}
6487 
6488 	return 0;
6489 
6490 err_rq:
6491 	kfree(vi->sq);
6492 err_sq:
6493 	kfree(vi->ctrl);
6494 err_ctrl:
6495 	return -ENOMEM;
6496 }
6497 
6498 static int init_vqs(struct virtnet_info *vi)
6499 {
6500 	int ret;
6501 
6502 	/* Allocate send & receive queues */
6503 	ret = virtnet_alloc_queues(vi);
6504 	if (ret)
6505 		goto err;
6506 
6507 	ret = virtnet_find_vqs(vi);
6508 	if (ret)
6509 		goto err_free;
6510 
6511 	cpus_read_lock();
6512 	virtnet_set_affinity(vi);
6513 	cpus_read_unlock();
6514 
6515 	return 0;
6516 
6517 err_free:
6518 	virtnet_free_queues(vi);
6519 err:
6520 	return ret;
6521 }
6522 
6523 #ifdef CONFIG_SYSFS
6524 static ssize_t mergeable_rx_buffer_size_show(struct netdev_rx_queue *queue,
6525 		char *buf)
6526 {
6527 	struct virtnet_info *vi = netdev_priv(queue->dev);
6528 	unsigned int queue_index = get_netdev_rx_queue_index(queue);
6529 	unsigned int headroom = virtnet_get_headroom(vi);
6530 	unsigned int tailroom = headroom ? sizeof(struct skb_shared_info) : 0;
6531 	struct ewma_pkt_len *avg;
6532 
6533 	BUG_ON(queue_index >= vi->max_queue_pairs);
6534 	avg = &vi->rq[queue_index].mrg_avg_pkt_len;
6535 	return sprintf(buf, "%u\n",
6536 		       get_mergeable_buf_len(&vi->rq[queue_index], avg,
6537 				       SKB_DATA_ALIGN(headroom + tailroom)));
6538 }
6539 
6540 static struct rx_queue_attribute mergeable_rx_buffer_size_attribute =
6541 	__ATTR_RO(mergeable_rx_buffer_size);
6542 
6543 static struct attribute *virtio_net_mrg_rx_attrs[] = {
6544 	&mergeable_rx_buffer_size_attribute.attr,
6545 	NULL
6546 };
6547 
6548 static const struct attribute_group virtio_net_mrg_rx_group = {
6549 	.name = "virtio_net",
6550 	.attrs = virtio_net_mrg_rx_attrs
6551 };
6552 #endif
6553 
6554 static bool virtnet_fail_on_feature(struct virtio_device *vdev,
6555 				    unsigned int fbit,
6556 				    const char *fname, const char *dname)
6557 {
6558 	if (!virtio_has_feature(vdev, fbit))
6559 		return false;
6560 
6561 	dev_err(&vdev->dev, "device advertises feature %s but not %s",
6562 		fname, dname);
6563 
6564 	return true;
6565 }
6566 
6567 #define VIRTNET_FAIL_ON(vdev, fbit, dbit)			\
6568 	virtnet_fail_on_feature(vdev, fbit, #fbit, dbit)
6569 
6570 static bool virtnet_validate_features(struct virtio_device *vdev)
6571 {
6572 	if (!virtio_has_feature(vdev, VIRTIO_NET_F_CTRL_VQ) &&
6573 	    (VIRTNET_FAIL_ON(vdev, VIRTIO_NET_F_CTRL_RX,
6574 			     "VIRTIO_NET_F_CTRL_VQ") ||
6575 	     VIRTNET_FAIL_ON(vdev, VIRTIO_NET_F_CTRL_VLAN,
6576 			     "VIRTIO_NET_F_CTRL_VQ") ||
6577 	     VIRTNET_FAIL_ON(vdev, VIRTIO_NET_F_GUEST_ANNOUNCE,
6578 			     "VIRTIO_NET_F_CTRL_VQ") ||
6579 	     VIRTNET_FAIL_ON(vdev, VIRTIO_NET_F_MQ, "VIRTIO_NET_F_CTRL_VQ") ||
6580 	     VIRTNET_FAIL_ON(vdev, VIRTIO_NET_F_CTRL_MAC_ADDR,
6581 			     "VIRTIO_NET_F_CTRL_VQ") ||
6582 	     VIRTNET_FAIL_ON(vdev, VIRTIO_NET_F_RSS,
6583 			     "VIRTIO_NET_F_CTRL_VQ") ||
6584 	     VIRTNET_FAIL_ON(vdev, VIRTIO_NET_F_HASH_REPORT,
6585 			     "VIRTIO_NET_F_CTRL_VQ") ||
6586 	     VIRTNET_FAIL_ON(vdev, VIRTIO_NET_F_NOTF_COAL,
6587 			     "VIRTIO_NET_F_CTRL_VQ") ||
6588 	     VIRTNET_FAIL_ON(vdev, VIRTIO_NET_F_VQ_NOTF_COAL,
6589 			     "VIRTIO_NET_F_CTRL_VQ"))) {
6590 		return false;
6591 	}
6592 
6593 	return true;
6594 }
6595 
6596 #define MIN_MTU ETH_MIN_MTU
6597 #define MAX_MTU ETH_MAX_MTU
6598 
6599 static int virtnet_validate(struct virtio_device *vdev)
6600 {
6601 	if (!vdev->config->get) {
6602 		dev_err(&vdev->dev, "%s failure: config access disabled\n",
6603 			__func__);
6604 		return -EINVAL;
6605 	}
6606 
6607 	if (!virtnet_validate_features(vdev))
6608 		return -EINVAL;
6609 
6610 	if (virtio_has_feature(vdev, VIRTIO_NET_F_MTU)) {
6611 		int mtu = virtio_cread16(vdev,
6612 					 offsetof(struct virtio_net_config,
6613 						  mtu));
6614 		if (mtu < MIN_MTU)
6615 			__virtio_clear_bit(vdev, VIRTIO_NET_F_MTU);
6616 	}
6617 
6618 	if (virtio_has_feature(vdev, VIRTIO_NET_F_STANDBY) &&
6619 	    !virtio_has_feature(vdev, VIRTIO_NET_F_MAC)) {
6620 		dev_warn(&vdev->dev, "device advertises feature VIRTIO_NET_F_STANDBY but not VIRTIO_NET_F_MAC, disabling standby");
6621 		__virtio_clear_bit(vdev, VIRTIO_NET_F_STANDBY);
6622 	}
6623 
6624 	return 0;
6625 }
6626 
6627 static bool virtnet_check_guest_gso(const struct virtnet_info *vi)
6628 {
6629 	return virtio_has_feature(vi->vdev, VIRTIO_NET_F_GUEST_TSO4) ||
6630 		virtio_has_feature(vi->vdev, VIRTIO_NET_F_GUEST_TSO6) ||
6631 		virtio_has_feature(vi->vdev, VIRTIO_NET_F_GUEST_ECN) ||
6632 		virtio_has_feature(vi->vdev, VIRTIO_NET_F_GUEST_UFO) ||
6633 		(virtio_has_feature(vi->vdev, VIRTIO_NET_F_GUEST_USO4) &&
6634 		virtio_has_feature(vi->vdev, VIRTIO_NET_F_GUEST_USO6));
6635 }
6636 
6637 static void virtnet_set_big_packets(struct virtnet_info *vi, const int mtu)
6638 {
6639 	bool guest_gso = virtnet_check_guest_gso(vi);
6640 
6641 	/* If device can receive ANY guest GSO packets, regardless of mtu,
6642 	 * allocate packets of maximum size, otherwise limit it to only
6643 	 * mtu size worth only.
6644 	 */
6645 	if (mtu > ETH_DATA_LEN || guest_gso) {
6646 		vi->big_packets = true;
6647 		vi->big_packets_num_skbfrags = guest_gso ? MAX_SKB_FRAGS : DIV_ROUND_UP(mtu, PAGE_SIZE);
6648 	}
6649 }
6650 
6651 #define VIRTIO_NET_HASH_REPORT_MAX_TABLE      10
6652 static enum xdp_rss_hash_type
6653 virtnet_xdp_rss_type[VIRTIO_NET_HASH_REPORT_MAX_TABLE] = {
6654 	[VIRTIO_NET_HASH_REPORT_NONE] = XDP_RSS_TYPE_NONE,
6655 	[VIRTIO_NET_HASH_REPORT_IPv4] = XDP_RSS_TYPE_L3_IPV4,
6656 	[VIRTIO_NET_HASH_REPORT_TCPv4] = XDP_RSS_TYPE_L4_IPV4_TCP,
6657 	[VIRTIO_NET_HASH_REPORT_UDPv4] = XDP_RSS_TYPE_L4_IPV4_UDP,
6658 	[VIRTIO_NET_HASH_REPORT_IPv6] = XDP_RSS_TYPE_L3_IPV6,
6659 	[VIRTIO_NET_HASH_REPORT_TCPv6] = XDP_RSS_TYPE_L4_IPV6_TCP,
6660 	[VIRTIO_NET_HASH_REPORT_UDPv6] = XDP_RSS_TYPE_L4_IPV6_UDP,
6661 	[VIRTIO_NET_HASH_REPORT_IPv6_EX] = XDP_RSS_TYPE_L3_IPV6_EX,
6662 	[VIRTIO_NET_HASH_REPORT_TCPv6_EX] = XDP_RSS_TYPE_L4_IPV6_TCP_EX,
6663 	[VIRTIO_NET_HASH_REPORT_UDPv6_EX] = XDP_RSS_TYPE_L4_IPV6_UDP_EX
6664 };
6665 
6666 static int virtnet_xdp_rx_hash(const struct xdp_md *_ctx, u32 *hash,
6667 			       enum xdp_rss_hash_type *rss_type)
6668 {
6669 	const struct xdp_buff *xdp = (void *)_ctx;
6670 	struct virtio_net_hdr_v1_hash *hdr_hash;
6671 	struct virtnet_info *vi;
6672 	u16 hash_report;
6673 
6674 	if (!(xdp->rxq->dev->features & NETIF_F_RXHASH))
6675 		return -ENODATA;
6676 
6677 	vi = netdev_priv(xdp->rxq->dev);
6678 	hdr_hash = (struct virtio_net_hdr_v1_hash *)(xdp->data - vi->hdr_len);
6679 	hash_report = __le16_to_cpu(hdr_hash->hash_report);
6680 
6681 	if (hash_report >= VIRTIO_NET_HASH_REPORT_MAX_TABLE)
6682 		hash_report = VIRTIO_NET_HASH_REPORT_NONE;
6683 
6684 	*rss_type = virtnet_xdp_rss_type[hash_report];
6685 	*hash = __le32_to_cpu(hdr_hash->hash_value);
6686 	return 0;
6687 }
6688 
6689 static const struct xdp_metadata_ops virtnet_xdp_metadata_ops = {
6690 	.xmo_rx_hash			= virtnet_xdp_rx_hash,
6691 };
6692 
6693 static int virtnet_probe(struct virtio_device *vdev)
6694 {
6695 	int i, err = -ENOMEM;
6696 	struct net_device *dev;
6697 	struct virtnet_info *vi;
6698 	u16 max_queue_pairs;
6699 	int mtu = 0;
6700 
6701 	/* Find if host supports multiqueue/rss virtio_net device */
6702 	max_queue_pairs = 1;
6703 	if (virtio_has_feature(vdev, VIRTIO_NET_F_MQ) || virtio_has_feature(vdev, VIRTIO_NET_F_RSS))
6704 		max_queue_pairs =
6705 		     virtio_cread16(vdev, offsetof(struct virtio_net_config, max_virtqueue_pairs));
6706 
6707 	/* We need at least 2 queue's */
6708 	if (max_queue_pairs < VIRTIO_NET_CTRL_MQ_VQ_PAIRS_MIN ||
6709 	    max_queue_pairs > VIRTIO_NET_CTRL_MQ_VQ_PAIRS_MAX ||
6710 	    !virtio_has_feature(vdev, VIRTIO_NET_F_CTRL_VQ))
6711 		max_queue_pairs = 1;
6712 
6713 	/* Allocate ourselves a network device with room for our info */
6714 	dev = alloc_etherdev_mq(sizeof(struct virtnet_info), max_queue_pairs);
6715 	if (!dev)
6716 		return -ENOMEM;
6717 
6718 	/* Set up network device as normal. */
6719 	dev->priv_flags |= IFF_UNICAST_FLT | IFF_LIVE_ADDR_CHANGE |
6720 			   IFF_TX_SKB_NO_LINEAR;
6721 	dev->netdev_ops = &virtnet_netdev;
6722 	dev->stat_ops = &virtnet_stat_ops;
6723 	dev->features = NETIF_F_HIGHDMA;
6724 
6725 	dev->ethtool_ops = &virtnet_ethtool_ops;
6726 	SET_NETDEV_DEV(dev, &vdev->dev);
6727 
6728 	/* Do we support "hardware" checksums? */
6729 	if (virtio_has_feature(vdev, VIRTIO_NET_F_CSUM)) {
6730 		/* This opens up the world of extra features. */
6731 		dev->hw_features |= NETIF_F_HW_CSUM | NETIF_F_SG;
6732 		if (csum)
6733 			dev->features |= NETIF_F_HW_CSUM | NETIF_F_SG;
6734 
6735 		if (virtio_has_feature(vdev, VIRTIO_NET_F_GSO)) {
6736 			dev->hw_features |= NETIF_F_TSO
6737 				| NETIF_F_TSO_ECN | NETIF_F_TSO6;
6738 		}
6739 		/* Individual feature bits: what can host handle? */
6740 		if (virtio_has_feature(vdev, VIRTIO_NET_F_HOST_TSO4))
6741 			dev->hw_features |= NETIF_F_TSO;
6742 		if (virtio_has_feature(vdev, VIRTIO_NET_F_HOST_TSO6))
6743 			dev->hw_features |= NETIF_F_TSO6;
6744 		if (virtio_has_feature(vdev, VIRTIO_NET_F_HOST_ECN))
6745 			dev->hw_features |= NETIF_F_TSO_ECN;
6746 		if (virtio_has_feature(vdev, VIRTIO_NET_F_HOST_USO))
6747 			dev->hw_features |= NETIF_F_GSO_UDP_L4;
6748 
6749 		dev->features |= NETIF_F_GSO_ROBUST;
6750 
6751 		if (gso)
6752 			dev->features |= dev->hw_features & NETIF_F_ALL_TSO;
6753 		/* (!csum && gso) case will be fixed by register_netdev() */
6754 	}
6755 
6756 	/* 1. With VIRTIO_NET_F_GUEST_CSUM negotiation, the driver doesn't
6757 	 * need to calculate checksums for partially checksummed packets,
6758 	 * as they're considered valid by the upper layer.
6759 	 * 2. Without VIRTIO_NET_F_GUEST_CSUM negotiation, the driver only
6760 	 * receives fully checksummed packets. The device may assist in
6761 	 * validating these packets' checksums, so the driver won't have to.
6762 	 */
6763 	dev->features |= NETIF_F_RXCSUM;
6764 
6765 	if (virtio_has_feature(vdev, VIRTIO_NET_F_GUEST_TSO4) ||
6766 	    virtio_has_feature(vdev, VIRTIO_NET_F_GUEST_TSO6))
6767 		dev->features |= NETIF_F_GRO_HW;
6768 	if (virtio_has_feature(vdev, VIRTIO_NET_F_CTRL_GUEST_OFFLOADS))
6769 		dev->hw_features |= NETIF_F_GRO_HW;
6770 
6771 	dev->vlan_features = dev->features;
6772 	dev->xdp_features = NETDEV_XDP_ACT_BASIC | NETDEV_XDP_ACT_REDIRECT |
6773 		NETDEV_XDP_ACT_XSK_ZEROCOPY;
6774 
6775 	/* MTU range: 68 - 65535 */
6776 	dev->min_mtu = MIN_MTU;
6777 	dev->max_mtu = MAX_MTU;
6778 
6779 	/* Configuration may specify what MAC to use.  Otherwise random. */
6780 	if (virtio_has_feature(vdev, VIRTIO_NET_F_MAC)) {
6781 		u8 addr[ETH_ALEN];
6782 
6783 		virtio_cread_bytes(vdev,
6784 				   offsetof(struct virtio_net_config, mac),
6785 				   addr, ETH_ALEN);
6786 		eth_hw_addr_set(dev, addr);
6787 	} else {
6788 		eth_hw_addr_random(dev);
6789 		dev_info(&vdev->dev, "Assigned random MAC address %pM\n",
6790 			 dev->dev_addr);
6791 	}
6792 
6793 	/* Set up our device-specific information */
6794 	vi = netdev_priv(dev);
6795 	vi->dev = dev;
6796 	vi->vdev = vdev;
6797 	vdev->priv = vi;
6798 
6799 	INIT_WORK(&vi->config_work, virtnet_config_changed_work);
6800 	INIT_WORK(&vi->rx_mode_work, virtnet_rx_mode_work);
6801 	spin_lock_init(&vi->refill_lock);
6802 
6803 	if (virtio_has_feature(vdev, VIRTIO_NET_F_MRG_RXBUF)) {
6804 		vi->mergeable_rx_bufs = true;
6805 		dev->xdp_features |= NETDEV_XDP_ACT_RX_SG;
6806 	}
6807 
6808 	if (virtio_has_feature(vdev, VIRTIO_NET_F_HASH_REPORT))
6809 		vi->has_rss_hash_report = true;
6810 
6811 	if (virtio_has_feature(vdev, VIRTIO_NET_F_RSS)) {
6812 		vi->has_rss = true;
6813 
6814 		vi->rss_indir_table_size =
6815 			virtio_cread16(vdev, offsetof(struct virtio_net_config,
6816 				rss_max_indirection_table_length));
6817 	}
6818 	vi->rss_hdr = devm_kzalloc(&vdev->dev, virtnet_rss_hdr_size(vi), GFP_KERNEL);
6819 	if (!vi->rss_hdr) {
6820 		err = -ENOMEM;
6821 		goto free;
6822 	}
6823 
6824 	if (vi->has_rss || vi->has_rss_hash_report) {
6825 		vi->rss_key_size =
6826 			virtio_cread8(vdev, offsetof(struct virtio_net_config, rss_max_key_size));
6827 		if (vi->rss_key_size > VIRTIO_NET_RSS_MAX_KEY_SIZE) {
6828 			dev_err(&vdev->dev, "rss_max_key_size=%u exceeds the limit %u.\n",
6829 				vi->rss_key_size, VIRTIO_NET_RSS_MAX_KEY_SIZE);
6830 			err = -EINVAL;
6831 			goto free;
6832 		}
6833 
6834 		vi->rss_hash_types_supported =
6835 		    virtio_cread32(vdev, offsetof(struct virtio_net_config, supported_hash_types));
6836 		vi->rss_hash_types_supported &=
6837 				~(VIRTIO_NET_RSS_HASH_TYPE_IP_EX |
6838 				  VIRTIO_NET_RSS_HASH_TYPE_TCP_EX |
6839 				  VIRTIO_NET_RSS_HASH_TYPE_UDP_EX);
6840 
6841 		dev->hw_features |= NETIF_F_RXHASH;
6842 		dev->xdp_metadata_ops = &virtnet_xdp_metadata_ops;
6843 	}
6844 
6845 	if (vi->has_rss_hash_report)
6846 		vi->hdr_len = sizeof(struct virtio_net_hdr_v1_hash);
6847 	else if (virtio_has_feature(vdev, VIRTIO_NET_F_MRG_RXBUF) ||
6848 		 virtio_has_feature(vdev, VIRTIO_F_VERSION_1))
6849 		vi->hdr_len = sizeof(struct virtio_net_hdr_mrg_rxbuf);
6850 	else
6851 		vi->hdr_len = sizeof(struct virtio_net_hdr);
6852 
6853 	if (virtio_has_feature(vdev, VIRTIO_F_ANY_LAYOUT) ||
6854 	    virtio_has_feature(vdev, VIRTIO_F_VERSION_1))
6855 		vi->any_header_sg = true;
6856 
6857 	if (virtio_has_feature(vdev, VIRTIO_NET_F_CTRL_VQ))
6858 		vi->has_cvq = true;
6859 
6860 	mutex_init(&vi->cvq_lock);
6861 
6862 	if (virtio_has_feature(vdev, VIRTIO_NET_F_MTU)) {
6863 		mtu = virtio_cread16(vdev,
6864 				     offsetof(struct virtio_net_config,
6865 					      mtu));
6866 		if (mtu < dev->min_mtu) {
6867 			/* Should never trigger: MTU was previously validated
6868 			 * in virtnet_validate.
6869 			 */
6870 			dev_err(&vdev->dev,
6871 				"device MTU appears to have changed it is now %d < %d",
6872 				mtu, dev->min_mtu);
6873 			err = -EINVAL;
6874 			goto free;
6875 		}
6876 
6877 		dev->mtu = mtu;
6878 		dev->max_mtu = mtu;
6879 	}
6880 
6881 	virtnet_set_big_packets(vi, mtu);
6882 
6883 	if (vi->any_header_sg)
6884 		dev->needed_headroom = vi->hdr_len;
6885 
6886 	/* Enable multiqueue by default */
6887 	if (num_online_cpus() >= max_queue_pairs)
6888 		vi->curr_queue_pairs = max_queue_pairs;
6889 	else
6890 		vi->curr_queue_pairs = num_online_cpus();
6891 	vi->max_queue_pairs = max_queue_pairs;
6892 
6893 	/* Allocate/initialize the rx/tx queues, and invoke find_vqs */
6894 	err = init_vqs(vi);
6895 	if (err)
6896 		goto free;
6897 
6898 	if (virtio_has_feature(vi->vdev, VIRTIO_NET_F_NOTF_COAL)) {
6899 		vi->intr_coal_rx.max_usecs = 0;
6900 		vi->intr_coal_tx.max_usecs = 0;
6901 		vi->intr_coal_rx.max_packets = 0;
6902 
6903 		/* Keep the default values of the coalescing parameters
6904 		 * aligned with the default napi_tx state.
6905 		 */
6906 		if (vi->sq[0].napi.weight)
6907 			vi->intr_coal_tx.max_packets = 1;
6908 		else
6909 			vi->intr_coal_tx.max_packets = 0;
6910 	}
6911 
6912 	if (virtio_has_feature(vi->vdev, VIRTIO_NET_F_VQ_NOTF_COAL)) {
6913 		/* The reason is the same as VIRTIO_NET_F_NOTF_COAL. */
6914 		for (i = 0; i < vi->max_queue_pairs; i++)
6915 			if (vi->sq[i].napi.weight)
6916 				vi->sq[i].intr_coal.max_packets = 1;
6917 
6918 		err = virtnet_init_irq_moder(vi);
6919 		if (err)
6920 			goto free;
6921 	}
6922 
6923 #ifdef CONFIG_SYSFS
6924 	if (vi->mergeable_rx_bufs)
6925 		dev->sysfs_rx_queue_group = &virtio_net_mrg_rx_group;
6926 #endif
6927 	netif_set_real_num_tx_queues(dev, vi->curr_queue_pairs);
6928 	netif_set_real_num_rx_queues(dev, vi->curr_queue_pairs);
6929 
6930 	virtnet_init_settings(dev);
6931 
6932 	if (virtio_has_feature(vdev, VIRTIO_NET_F_STANDBY)) {
6933 		vi->failover = net_failover_create(vi->dev);
6934 		if (IS_ERR(vi->failover)) {
6935 			err = PTR_ERR(vi->failover);
6936 			goto free_vqs;
6937 		}
6938 	}
6939 
6940 	if (vi->has_rss || vi->has_rss_hash_report)
6941 		virtnet_init_default_rss(vi);
6942 
6943 	enable_rx_mode_work(vi);
6944 
6945 	/* serialize netdev register + virtio_device_ready() with ndo_open() */
6946 	rtnl_lock();
6947 
6948 	err = register_netdevice(dev);
6949 	if (err) {
6950 		pr_debug("virtio_net: registering device failed\n");
6951 		rtnl_unlock();
6952 		goto free_failover;
6953 	}
6954 
6955 	/* Disable config change notification until ndo_open. */
6956 	virtio_config_driver_disable(vi->vdev);
6957 
6958 	virtio_device_ready(vdev);
6959 
6960 	if (vi->has_rss || vi->has_rss_hash_report) {
6961 		if (!virtnet_commit_rss_command(vi)) {
6962 			dev_warn(&vdev->dev, "RSS disabled because committing failed.\n");
6963 			dev->hw_features &= ~NETIF_F_RXHASH;
6964 			vi->has_rss_hash_report = false;
6965 			vi->has_rss = false;
6966 		}
6967 	}
6968 
6969 	virtnet_set_queues(vi, vi->curr_queue_pairs);
6970 
6971 	/* a random MAC address has been assigned, notify the device.
6972 	 * We don't fail probe if VIRTIO_NET_F_CTRL_MAC_ADDR is not there
6973 	 * because many devices work fine without getting MAC explicitly
6974 	 */
6975 	if (!virtio_has_feature(vdev, VIRTIO_NET_F_MAC) &&
6976 	    virtio_has_feature(vi->vdev, VIRTIO_NET_F_CTRL_MAC_ADDR)) {
6977 		struct scatterlist sg;
6978 
6979 		sg_init_one(&sg, dev->dev_addr, dev->addr_len);
6980 		if (!virtnet_send_command(vi, VIRTIO_NET_CTRL_MAC,
6981 					  VIRTIO_NET_CTRL_MAC_ADDR_SET, &sg)) {
6982 			pr_debug("virtio_net: setting MAC address failed\n");
6983 			rtnl_unlock();
6984 			err = -EINVAL;
6985 			goto free_unregister_netdev;
6986 		}
6987 	}
6988 
6989 	if (virtio_has_feature(vi->vdev, VIRTIO_NET_F_DEVICE_STATS)) {
6990 		struct virtio_net_stats_capabilities *stats_cap  __free(kfree) = NULL;
6991 		struct scatterlist sg;
6992 		__le64 v;
6993 
6994 		stats_cap = kzalloc(sizeof(*stats_cap), GFP_KERNEL);
6995 		if (!stats_cap) {
6996 			rtnl_unlock();
6997 			err = -ENOMEM;
6998 			goto free_unregister_netdev;
6999 		}
7000 
7001 		sg_init_one(&sg, stats_cap, sizeof(*stats_cap));
7002 
7003 		if (!virtnet_send_command_reply(vi, VIRTIO_NET_CTRL_STATS,
7004 						VIRTIO_NET_CTRL_STATS_QUERY,
7005 						NULL, &sg)) {
7006 			pr_debug("virtio_net: fail to get stats capability\n");
7007 			rtnl_unlock();
7008 			err = -EINVAL;
7009 			goto free_unregister_netdev;
7010 		}
7011 
7012 		v = stats_cap->supported_stats_types[0];
7013 		vi->device_stats_cap = le64_to_cpu(v);
7014 	}
7015 
7016 	/* Assume link up if device can't report link status,
7017 	   otherwise get link status from config. */
7018 	netif_carrier_off(dev);
7019 	if (virtio_has_feature(vi->vdev, VIRTIO_NET_F_STATUS)) {
7020 		virtnet_config_changed_work(&vi->config_work);
7021 	} else {
7022 		vi->status = VIRTIO_NET_S_LINK_UP;
7023 		virtnet_update_settings(vi);
7024 		netif_carrier_on(dev);
7025 	}
7026 
7027 	for (i = 0; i < ARRAY_SIZE(guest_offloads); i++)
7028 		if (virtio_has_feature(vi->vdev, guest_offloads[i]))
7029 			set_bit(guest_offloads[i], &vi->guest_offloads);
7030 	vi->guest_offloads_capable = vi->guest_offloads;
7031 
7032 	rtnl_unlock();
7033 
7034 	err = virtnet_cpu_notif_add(vi);
7035 	if (err) {
7036 		pr_debug("virtio_net: registering cpu notifier failed\n");
7037 		goto free_unregister_netdev;
7038 	}
7039 
7040 	pr_debug("virtnet: registered device %s with %d RX and TX vq's\n",
7041 		 dev->name, max_queue_pairs);
7042 
7043 	return 0;
7044 
7045 free_unregister_netdev:
7046 	unregister_netdev(dev);
7047 free_failover:
7048 	net_failover_destroy(vi->failover);
7049 free_vqs:
7050 	virtio_reset_device(vdev);
7051 	cancel_delayed_work_sync(&vi->refill);
7052 	free_receive_page_frags(vi);
7053 	virtnet_del_vqs(vi);
7054 free:
7055 	free_netdev(dev);
7056 	return err;
7057 }
7058 
7059 static void remove_vq_common(struct virtnet_info *vi)
7060 {
7061 	int i;
7062 
7063 	virtio_reset_device(vi->vdev);
7064 
7065 	/* Free unused buffers in both send and recv, if any. */
7066 	free_unused_bufs(vi);
7067 
7068 	/*
7069 	 * Rule of thumb is netdev_tx_reset_queue() should follow any
7070 	 * skb freeing not followed by netdev_tx_completed_queue()
7071 	 */
7072 	for (i = 0; i < vi->max_queue_pairs; i++)
7073 		netdev_tx_reset_queue(netdev_get_tx_queue(vi->dev, i));
7074 
7075 	free_receive_bufs(vi);
7076 
7077 	free_receive_page_frags(vi);
7078 
7079 	virtnet_del_vqs(vi);
7080 }
7081 
7082 static void virtnet_remove(struct virtio_device *vdev)
7083 {
7084 	struct virtnet_info *vi = vdev->priv;
7085 
7086 	virtnet_cpu_notif_remove(vi);
7087 
7088 	/* Make sure no work handler is accessing the device. */
7089 	flush_work(&vi->config_work);
7090 	disable_rx_mode_work(vi);
7091 	flush_work(&vi->rx_mode_work);
7092 
7093 	virtnet_free_irq_moder(vi);
7094 
7095 	unregister_netdev(vi->dev);
7096 
7097 	net_failover_destroy(vi->failover);
7098 
7099 	remove_vq_common(vi);
7100 
7101 	free_netdev(vi->dev);
7102 }
7103 
7104 static __maybe_unused int virtnet_freeze(struct virtio_device *vdev)
7105 {
7106 	struct virtnet_info *vi = vdev->priv;
7107 
7108 	virtnet_cpu_notif_remove(vi);
7109 	virtnet_freeze_down(vdev);
7110 	remove_vq_common(vi);
7111 
7112 	return 0;
7113 }
7114 
7115 static __maybe_unused int virtnet_restore(struct virtio_device *vdev)
7116 {
7117 	struct virtnet_info *vi = vdev->priv;
7118 	int err;
7119 
7120 	err = virtnet_restore_up(vdev);
7121 	if (err)
7122 		return err;
7123 	virtnet_set_queues(vi, vi->curr_queue_pairs);
7124 
7125 	err = virtnet_cpu_notif_add(vi);
7126 	if (err) {
7127 		virtnet_freeze_down(vdev);
7128 		remove_vq_common(vi);
7129 		return err;
7130 	}
7131 
7132 	return 0;
7133 }
7134 
7135 static struct virtio_device_id id_table[] = {
7136 	{ VIRTIO_ID_NET, VIRTIO_DEV_ANY_ID },
7137 	{ 0 },
7138 };
7139 
7140 #define VIRTNET_FEATURES \
7141 	VIRTIO_NET_F_CSUM, VIRTIO_NET_F_GUEST_CSUM, \
7142 	VIRTIO_NET_F_MAC, \
7143 	VIRTIO_NET_F_HOST_TSO4, VIRTIO_NET_F_HOST_UFO, VIRTIO_NET_F_HOST_TSO6, \
7144 	VIRTIO_NET_F_HOST_ECN, VIRTIO_NET_F_GUEST_TSO4, VIRTIO_NET_F_GUEST_TSO6, \
7145 	VIRTIO_NET_F_GUEST_ECN, VIRTIO_NET_F_GUEST_UFO, \
7146 	VIRTIO_NET_F_HOST_USO, VIRTIO_NET_F_GUEST_USO4, VIRTIO_NET_F_GUEST_USO6, \
7147 	VIRTIO_NET_F_MRG_RXBUF, VIRTIO_NET_F_STATUS, VIRTIO_NET_F_CTRL_VQ, \
7148 	VIRTIO_NET_F_CTRL_RX, VIRTIO_NET_F_CTRL_VLAN, \
7149 	VIRTIO_NET_F_GUEST_ANNOUNCE, VIRTIO_NET_F_MQ, \
7150 	VIRTIO_NET_F_CTRL_MAC_ADDR, \
7151 	VIRTIO_NET_F_MTU, VIRTIO_NET_F_CTRL_GUEST_OFFLOADS, \
7152 	VIRTIO_NET_F_SPEED_DUPLEX, VIRTIO_NET_F_STANDBY, \
7153 	VIRTIO_NET_F_RSS, VIRTIO_NET_F_HASH_REPORT, VIRTIO_NET_F_NOTF_COAL, \
7154 	VIRTIO_NET_F_VQ_NOTF_COAL, \
7155 	VIRTIO_NET_F_GUEST_HDRLEN, VIRTIO_NET_F_DEVICE_STATS
7156 
7157 static unsigned int features[] = {
7158 	VIRTNET_FEATURES,
7159 };
7160 
7161 static unsigned int features_legacy[] = {
7162 	VIRTNET_FEATURES,
7163 	VIRTIO_NET_F_GSO,
7164 	VIRTIO_F_ANY_LAYOUT,
7165 };
7166 
7167 static struct virtio_driver virtio_net_driver = {
7168 	.feature_table = features,
7169 	.feature_table_size = ARRAY_SIZE(features),
7170 	.feature_table_legacy = features_legacy,
7171 	.feature_table_size_legacy = ARRAY_SIZE(features_legacy),
7172 	.driver.name =	KBUILD_MODNAME,
7173 	.id_table =	id_table,
7174 	.validate =	virtnet_validate,
7175 	.probe =	virtnet_probe,
7176 	.remove =	virtnet_remove,
7177 	.config_changed = virtnet_config_changed,
7178 #ifdef CONFIG_PM_SLEEP
7179 	.freeze =	virtnet_freeze,
7180 	.restore =	virtnet_restore,
7181 #endif
7182 };
7183 
7184 static __init int virtio_net_driver_init(void)
7185 {
7186 	int ret;
7187 
7188 	ret = cpuhp_setup_state_multi(CPUHP_AP_ONLINE_DYN, "virtio/net:online",
7189 				      virtnet_cpu_online,
7190 				      virtnet_cpu_down_prep);
7191 	if (ret < 0)
7192 		goto out;
7193 	virtionet_online = ret;
7194 	ret = cpuhp_setup_state_multi(CPUHP_VIRT_NET_DEAD, "virtio/net:dead",
7195 				      NULL, virtnet_cpu_dead);
7196 	if (ret)
7197 		goto err_dead;
7198 	ret = register_virtio_driver(&virtio_net_driver);
7199 	if (ret)
7200 		goto err_virtio;
7201 	return 0;
7202 err_virtio:
7203 	cpuhp_remove_multi_state(CPUHP_VIRT_NET_DEAD);
7204 err_dead:
7205 	cpuhp_remove_multi_state(virtionet_online);
7206 out:
7207 	return ret;
7208 }
7209 module_init(virtio_net_driver_init);
7210 
7211 static __exit void virtio_net_driver_exit(void)
7212 {
7213 	unregister_virtio_driver(&virtio_net_driver);
7214 	cpuhp_remove_multi_state(CPUHP_VIRT_NET_DEAD);
7215 	cpuhp_remove_multi_state(virtionet_online);
7216 }
7217 module_exit(virtio_net_driver_exit);
7218 
7219 MODULE_DEVICE_TABLE(virtio, id_table);
7220 MODULE_DESCRIPTION("Virtio network driver");
7221 MODULE_LICENSE("GPL");
7222