xref: /linux/drivers/net/virtio_net.c (revision 31bedc1fb1d93250ae1900ee92ccd56689956d22)
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 #define VIRTIO_XDP_FLAG		BIT(0)
49 #define VIRTIO_ORPHAN_FLAG	BIT(1)
50 
51 /* RX packet size EWMA. The average packet size is used to determine the packet
52  * buffer size when refilling RX rings. As the entire RX ring may be refilled
53  * at once, the weight is chosen so that the EWMA will be insensitive to short-
54  * term, transient changes in packet size.
55  */
56 DECLARE_EWMA(pkt_len, 0, 64)
57 
58 #define VIRTNET_DRIVER_VERSION "1.0.0"
59 
60 static const unsigned long guest_offloads[] = {
61 	VIRTIO_NET_F_GUEST_TSO4,
62 	VIRTIO_NET_F_GUEST_TSO6,
63 	VIRTIO_NET_F_GUEST_ECN,
64 	VIRTIO_NET_F_GUEST_UFO,
65 	VIRTIO_NET_F_GUEST_CSUM,
66 	VIRTIO_NET_F_GUEST_USO4,
67 	VIRTIO_NET_F_GUEST_USO6,
68 	VIRTIO_NET_F_GUEST_HDRLEN
69 };
70 
71 #define GUEST_OFFLOAD_GRO_HW_MASK ((1ULL << VIRTIO_NET_F_GUEST_TSO4) | \
72 				(1ULL << VIRTIO_NET_F_GUEST_TSO6) | \
73 				(1ULL << VIRTIO_NET_F_GUEST_ECN)  | \
74 				(1ULL << VIRTIO_NET_F_GUEST_UFO)  | \
75 				(1ULL << VIRTIO_NET_F_GUEST_USO4) | \
76 				(1ULL << VIRTIO_NET_F_GUEST_USO6))
77 
78 struct virtnet_stat_desc {
79 	char desc[ETH_GSTRING_LEN];
80 	size_t offset;
81 	size_t qstat_offset;
82 };
83 
84 struct virtnet_sq_free_stats {
85 	u64 packets;
86 	u64 bytes;
87 	u64 napi_packets;
88 	u64 napi_bytes;
89 };
90 
91 struct virtnet_sq_stats {
92 	struct u64_stats_sync syncp;
93 	u64_stats_t packets;
94 	u64_stats_t bytes;
95 	u64_stats_t xdp_tx;
96 	u64_stats_t xdp_tx_drops;
97 	u64_stats_t kicks;
98 	u64_stats_t tx_timeouts;
99 	u64_stats_t stop;
100 	u64_stats_t wake;
101 };
102 
103 struct virtnet_rq_stats {
104 	struct u64_stats_sync syncp;
105 	u64_stats_t packets;
106 	u64_stats_t bytes;
107 	u64_stats_t drops;
108 	u64_stats_t xdp_packets;
109 	u64_stats_t xdp_tx;
110 	u64_stats_t xdp_redirects;
111 	u64_stats_t xdp_drops;
112 	u64_stats_t kicks;
113 };
114 
115 #define VIRTNET_SQ_STAT(name, m) {name, offsetof(struct virtnet_sq_stats, m), -1}
116 #define VIRTNET_RQ_STAT(name, m) {name, offsetof(struct virtnet_rq_stats, m), -1}
117 
118 #define VIRTNET_SQ_STAT_QSTAT(name, m)				\
119 	{							\
120 		name,						\
121 		offsetof(struct virtnet_sq_stats, m),		\
122 		offsetof(struct netdev_queue_stats_tx, m),	\
123 	}
124 
125 #define VIRTNET_RQ_STAT_QSTAT(name, m)				\
126 	{							\
127 		name,						\
128 		offsetof(struct virtnet_rq_stats, m),		\
129 		offsetof(struct netdev_queue_stats_rx, m),	\
130 	}
131 
132 static const struct virtnet_stat_desc virtnet_sq_stats_desc[] = {
133 	VIRTNET_SQ_STAT("xdp_tx",       xdp_tx),
134 	VIRTNET_SQ_STAT("xdp_tx_drops", xdp_tx_drops),
135 	VIRTNET_SQ_STAT("kicks",        kicks),
136 	VIRTNET_SQ_STAT("tx_timeouts",  tx_timeouts),
137 };
138 
139 static const struct virtnet_stat_desc virtnet_rq_stats_desc[] = {
140 	VIRTNET_RQ_STAT("drops",         drops),
141 	VIRTNET_RQ_STAT("xdp_packets",   xdp_packets),
142 	VIRTNET_RQ_STAT("xdp_tx",        xdp_tx),
143 	VIRTNET_RQ_STAT("xdp_redirects", xdp_redirects),
144 	VIRTNET_RQ_STAT("xdp_drops",     xdp_drops),
145 	VIRTNET_RQ_STAT("kicks",         kicks),
146 };
147 
148 static const struct virtnet_stat_desc virtnet_sq_stats_desc_qstat[] = {
149 	VIRTNET_SQ_STAT_QSTAT("packets", packets),
150 	VIRTNET_SQ_STAT_QSTAT("bytes",   bytes),
151 	VIRTNET_SQ_STAT_QSTAT("stop",	 stop),
152 	VIRTNET_SQ_STAT_QSTAT("wake",	 wake),
153 };
154 
155 static const struct virtnet_stat_desc virtnet_rq_stats_desc_qstat[] = {
156 	VIRTNET_RQ_STAT_QSTAT("packets", packets),
157 	VIRTNET_RQ_STAT_QSTAT("bytes",   bytes),
158 };
159 
160 #define VIRTNET_STATS_DESC_CQ(name) \
161 	{#name, offsetof(struct virtio_net_stats_cvq, name), -1}
162 
163 #define VIRTNET_STATS_DESC_RX(class, name) \
164 	{#name, offsetof(struct virtio_net_stats_rx_ ## class, rx_ ## name), -1}
165 
166 #define VIRTNET_STATS_DESC_TX(class, name) \
167 	{#name, offsetof(struct virtio_net_stats_tx_ ## class, tx_ ## name), -1}
168 
169 
170 static const struct virtnet_stat_desc virtnet_stats_cvq_desc[] = {
171 	VIRTNET_STATS_DESC_CQ(command_num),
172 	VIRTNET_STATS_DESC_CQ(ok_num),
173 };
174 
175 static const struct virtnet_stat_desc virtnet_stats_rx_basic_desc[] = {
176 	VIRTNET_STATS_DESC_RX(basic, packets),
177 	VIRTNET_STATS_DESC_RX(basic, bytes),
178 
179 	VIRTNET_STATS_DESC_RX(basic, notifications),
180 	VIRTNET_STATS_DESC_RX(basic, interrupts),
181 };
182 
183 static const struct virtnet_stat_desc virtnet_stats_tx_basic_desc[] = {
184 	VIRTNET_STATS_DESC_TX(basic, packets),
185 	VIRTNET_STATS_DESC_TX(basic, bytes),
186 
187 	VIRTNET_STATS_DESC_TX(basic, notifications),
188 	VIRTNET_STATS_DESC_TX(basic, interrupts),
189 };
190 
191 static const struct virtnet_stat_desc virtnet_stats_rx_csum_desc[] = {
192 	VIRTNET_STATS_DESC_RX(csum, needs_csum),
193 };
194 
195 static const struct virtnet_stat_desc virtnet_stats_tx_gso_desc[] = {
196 	VIRTNET_STATS_DESC_TX(gso, gso_packets_noseg),
197 	VIRTNET_STATS_DESC_TX(gso, gso_bytes_noseg),
198 };
199 
200 static const struct virtnet_stat_desc virtnet_stats_rx_speed_desc[] = {
201 	VIRTNET_STATS_DESC_RX(speed, ratelimit_bytes),
202 };
203 
204 static const struct virtnet_stat_desc virtnet_stats_tx_speed_desc[] = {
205 	VIRTNET_STATS_DESC_TX(speed, ratelimit_bytes),
206 };
207 
208 #define VIRTNET_STATS_DESC_RX_QSTAT(class, name, qstat_field)			\
209 	{									\
210 		#name,								\
211 		offsetof(struct virtio_net_stats_rx_ ## class, rx_ ## name),	\
212 		offsetof(struct netdev_queue_stats_rx, qstat_field),		\
213 	}
214 
215 #define VIRTNET_STATS_DESC_TX_QSTAT(class, name, qstat_field)			\
216 	{									\
217 		#name,								\
218 		offsetof(struct virtio_net_stats_tx_ ## class, tx_ ## name),	\
219 		offsetof(struct netdev_queue_stats_tx, qstat_field),		\
220 	}
221 
222 static const struct virtnet_stat_desc virtnet_stats_rx_basic_desc_qstat[] = {
223 	VIRTNET_STATS_DESC_RX_QSTAT(basic, drops,         hw_drops),
224 	VIRTNET_STATS_DESC_RX_QSTAT(basic, drop_overruns, hw_drop_overruns),
225 };
226 
227 static const struct virtnet_stat_desc virtnet_stats_tx_basic_desc_qstat[] = {
228 	VIRTNET_STATS_DESC_TX_QSTAT(basic, drops,          hw_drops),
229 	VIRTNET_STATS_DESC_TX_QSTAT(basic, drop_malformed, hw_drop_errors),
230 };
231 
232 static const struct virtnet_stat_desc virtnet_stats_rx_csum_desc_qstat[] = {
233 	VIRTNET_STATS_DESC_RX_QSTAT(csum, csum_valid, csum_unnecessary),
234 	VIRTNET_STATS_DESC_RX_QSTAT(csum, csum_none,  csum_none),
235 	VIRTNET_STATS_DESC_RX_QSTAT(csum, csum_bad,   csum_bad),
236 };
237 
238 static const struct virtnet_stat_desc virtnet_stats_tx_csum_desc_qstat[] = {
239 	VIRTNET_STATS_DESC_TX_QSTAT(csum, csum_none,  csum_none),
240 	VIRTNET_STATS_DESC_TX_QSTAT(csum, needs_csum, needs_csum),
241 };
242 
243 static const struct virtnet_stat_desc virtnet_stats_rx_gso_desc_qstat[] = {
244 	VIRTNET_STATS_DESC_RX_QSTAT(gso, gso_packets,           hw_gro_packets),
245 	VIRTNET_STATS_DESC_RX_QSTAT(gso, gso_bytes,             hw_gro_bytes),
246 	VIRTNET_STATS_DESC_RX_QSTAT(gso, gso_packets_coalesced, hw_gro_wire_packets),
247 	VIRTNET_STATS_DESC_RX_QSTAT(gso, gso_bytes_coalesced,   hw_gro_wire_bytes),
248 };
249 
250 static const struct virtnet_stat_desc virtnet_stats_tx_gso_desc_qstat[] = {
251 	VIRTNET_STATS_DESC_TX_QSTAT(gso, gso_packets,        hw_gso_packets),
252 	VIRTNET_STATS_DESC_TX_QSTAT(gso, gso_bytes,          hw_gso_bytes),
253 	VIRTNET_STATS_DESC_TX_QSTAT(gso, gso_segments,       hw_gso_wire_packets),
254 	VIRTNET_STATS_DESC_TX_QSTAT(gso, gso_segments_bytes, hw_gso_wire_bytes),
255 };
256 
257 static const struct virtnet_stat_desc virtnet_stats_rx_speed_desc_qstat[] = {
258 	VIRTNET_STATS_DESC_RX_QSTAT(speed, ratelimit_packets, hw_drop_ratelimits),
259 };
260 
261 static const struct virtnet_stat_desc virtnet_stats_tx_speed_desc_qstat[] = {
262 	VIRTNET_STATS_DESC_TX_QSTAT(speed, ratelimit_packets, hw_drop_ratelimits),
263 };
264 
265 #define VIRTNET_Q_TYPE_RX 0
266 #define VIRTNET_Q_TYPE_TX 1
267 #define VIRTNET_Q_TYPE_CQ 2
268 
269 struct virtnet_interrupt_coalesce {
270 	u32 max_packets;
271 	u32 max_usecs;
272 };
273 
274 /* The dma information of pages allocated at a time. */
275 struct virtnet_rq_dma {
276 	dma_addr_t addr;
277 	u32 ref;
278 	u16 len;
279 	u16 need_sync;
280 };
281 
282 /* Internal representation of a send virtqueue */
283 struct send_queue {
284 	/* Virtqueue associated with this send _queue */
285 	struct virtqueue *vq;
286 
287 	/* TX: fragments + linear part + virtio header */
288 	struct scatterlist sg[MAX_SKB_FRAGS + 2];
289 
290 	/* Name of the send queue: output.$index */
291 	char name[16];
292 
293 	struct virtnet_sq_stats stats;
294 
295 	struct virtnet_interrupt_coalesce intr_coal;
296 
297 	struct napi_struct napi;
298 
299 	/* Record whether sq is in reset state. */
300 	bool reset;
301 };
302 
303 /* Internal representation of a receive virtqueue */
304 struct receive_queue {
305 	/* Virtqueue associated with this receive_queue */
306 	struct virtqueue *vq;
307 
308 	struct napi_struct napi;
309 
310 	struct bpf_prog __rcu *xdp_prog;
311 
312 	struct virtnet_rq_stats stats;
313 
314 	/* The number of rx notifications */
315 	u16 calls;
316 
317 	/* Is dynamic interrupt moderation enabled? */
318 	bool dim_enabled;
319 
320 	/* Used to protect dim_enabled and inter_coal */
321 	struct mutex dim_lock;
322 
323 	/* Dynamic Interrupt Moderation */
324 	struct dim dim;
325 
326 	u32 packets_in_napi;
327 
328 	struct virtnet_interrupt_coalesce intr_coal;
329 
330 	/* Chain pages by the private ptr. */
331 	struct page *pages;
332 
333 	/* Average packet length for mergeable receive buffers. */
334 	struct ewma_pkt_len mrg_avg_pkt_len;
335 
336 	/* Page frag for packet buffer allocation. */
337 	struct page_frag alloc_frag;
338 
339 	/* RX: fragments + linear part + virtio header */
340 	struct scatterlist sg[MAX_SKB_FRAGS + 2];
341 
342 	/* Min single buffer size for mergeable buffers case. */
343 	unsigned int min_buf_len;
344 
345 	/* Name of this receive queue: input.$index */
346 	char name[16];
347 
348 	struct xdp_rxq_info xdp_rxq;
349 
350 	/* Record the last dma info to free after new pages is allocated. */
351 	struct virtnet_rq_dma *last_dma;
352 
353 	struct xsk_buff_pool *xsk_pool;
354 
355 	/* xdp rxq used by xsk */
356 	struct xdp_rxq_info xsk_rxq_info;
357 
358 	struct xdp_buff **xsk_buffs;
359 
360 	/* Do dma by self */
361 	bool do_dma;
362 };
363 
364 /* This structure can contain rss message with maximum settings for indirection table and keysize
365  * Note, that default structure that describes RSS configuration virtio_net_rss_config
366  * contains same info but can't handle table values.
367  * In any case, structure would be passed to virtio hw through sg_buf split by parts
368  * because table sizes may be differ according to the device configuration.
369  */
370 #define VIRTIO_NET_RSS_MAX_KEY_SIZE     40
371 #define VIRTIO_NET_RSS_MAX_TABLE_LEN    128
372 struct virtio_net_ctrl_rss {
373 	u32 hash_types;
374 	u16 indirection_table_mask;
375 	u16 unclassified_queue;
376 	u16 indirection_table[VIRTIO_NET_RSS_MAX_TABLE_LEN];
377 	u16 max_tx_vq;
378 	u8 hash_key_length;
379 	u8 key[VIRTIO_NET_RSS_MAX_KEY_SIZE];
380 };
381 
382 /* Control VQ buffers: protected by the rtnl lock */
383 struct control_buf {
384 	struct virtio_net_ctrl_hdr hdr;
385 	virtio_net_ctrl_ack status;
386 };
387 
388 struct virtnet_info {
389 	struct virtio_device *vdev;
390 	struct virtqueue *cvq;
391 	struct net_device *dev;
392 	struct send_queue *sq;
393 	struct receive_queue *rq;
394 	unsigned int status;
395 
396 	/* Max # of queue pairs supported by the device */
397 	u16 max_queue_pairs;
398 
399 	/* # of queue pairs currently used by the driver */
400 	u16 curr_queue_pairs;
401 
402 	/* # of XDP queue pairs currently used by the driver */
403 	u16 xdp_queue_pairs;
404 
405 	/* xdp_queue_pairs may be 0, when xdp is already loaded. So add this. */
406 	bool xdp_enabled;
407 
408 	/* I like... big packets and I cannot lie! */
409 	bool big_packets;
410 
411 	/* number of sg entries allocated for big packets */
412 	unsigned int big_packets_num_skbfrags;
413 
414 	/* Host will merge rx buffers for big packets (shake it! shake it!) */
415 	bool mergeable_rx_bufs;
416 
417 	/* Host supports rss and/or hash report */
418 	bool has_rss;
419 	bool has_rss_hash_report;
420 	u8 rss_key_size;
421 	u16 rss_indir_table_size;
422 	u32 rss_hash_types_supported;
423 	u32 rss_hash_types_saved;
424 	struct virtio_net_ctrl_rss rss;
425 
426 	/* Has control virtqueue */
427 	bool has_cvq;
428 
429 	/* Lock to protect the control VQ */
430 	struct mutex cvq_lock;
431 
432 	/* Host can handle any s/g split between our header and packet data */
433 	bool any_header_sg;
434 
435 	/* Packet virtio header size */
436 	u8 hdr_len;
437 
438 	/* Work struct for delayed refilling if we run low on memory. */
439 	struct delayed_work refill;
440 
441 	/* Is delayed refill enabled? */
442 	bool refill_enabled;
443 
444 	/* The lock to synchronize the access to refill_enabled */
445 	spinlock_t refill_lock;
446 
447 	/* Work struct for config space updates */
448 	struct work_struct config_work;
449 
450 	/* Work struct for setting rx mode */
451 	struct work_struct rx_mode_work;
452 
453 	/* OK to queue work setting RX mode? */
454 	bool rx_mode_work_enabled;
455 
456 	/* Does the affinity hint is set for virtqueues? */
457 	bool affinity_hint_set;
458 
459 	/* CPU hotplug instances for online & dead */
460 	struct hlist_node node;
461 	struct hlist_node node_dead;
462 
463 	struct control_buf *ctrl;
464 
465 	/* Ethtool settings */
466 	u8 duplex;
467 	u32 speed;
468 
469 	/* Is rx dynamic interrupt moderation enabled? */
470 	bool rx_dim_enabled;
471 
472 	/* Interrupt coalescing settings */
473 	struct virtnet_interrupt_coalesce intr_coal_tx;
474 	struct virtnet_interrupt_coalesce intr_coal_rx;
475 
476 	unsigned long guest_offloads;
477 	unsigned long guest_offloads_capable;
478 
479 	/* failover when STANDBY feature enabled */
480 	struct failover *failover;
481 
482 	u64 device_stats_cap;
483 };
484 
485 struct padded_vnet_hdr {
486 	struct virtio_net_hdr_v1_hash hdr;
487 	/*
488 	 * hdr is in a separate sg buffer, and data sg buffer shares same page
489 	 * with this header sg. This padding makes next sg 16 byte aligned
490 	 * after the header.
491 	 */
492 	char padding[12];
493 };
494 
495 struct virtio_net_common_hdr {
496 	union {
497 		struct virtio_net_hdr hdr;
498 		struct virtio_net_hdr_mrg_rxbuf	mrg_hdr;
499 		struct virtio_net_hdr_v1_hash hash_v1_hdr;
500 	};
501 };
502 
503 static void virtnet_sq_free_unused_buf(struct virtqueue *vq, void *buf);
504 static int virtnet_xdp_handler(struct bpf_prog *xdp_prog, struct xdp_buff *xdp,
505 			       struct net_device *dev,
506 			       unsigned int *xdp_xmit,
507 			       struct virtnet_rq_stats *stats);
508 static void virtnet_receive_done(struct virtnet_info *vi, struct receive_queue *rq,
509 				 struct sk_buff *skb, u8 flags);
510 static struct sk_buff *virtnet_skb_append_frag(struct sk_buff *head_skb,
511 					       struct sk_buff *curr_skb,
512 					       struct page *page, void *buf,
513 					       int len, int truesize);
514 
515 static bool is_xdp_frame(void *ptr)
516 {
517 	return (unsigned long)ptr & VIRTIO_XDP_FLAG;
518 }
519 
520 static void *xdp_to_ptr(struct xdp_frame *ptr)
521 {
522 	return (void *)((unsigned long)ptr | VIRTIO_XDP_FLAG);
523 }
524 
525 static struct xdp_frame *ptr_to_xdp(void *ptr)
526 {
527 	return (struct xdp_frame *)((unsigned long)ptr & ~VIRTIO_XDP_FLAG);
528 }
529 
530 static bool is_orphan_skb(void *ptr)
531 {
532 	return (unsigned long)ptr & VIRTIO_ORPHAN_FLAG;
533 }
534 
535 static void *skb_to_ptr(struct sk_buff *skb, bool orphan)
536 {
537 	return (void *)((unsigned long)skb | (orphan ? VIRTIO_ORPHAN_FLAG : 0));
538 }
539 
540 static struct sk_buff *ptr_to_skb(void *ptr)
541 {
542 	return (struct sk_buff *)((unsigned long)ptr & ~VIRTIO_ORPHAN_FLAG);
543 }
544 
545 static void __free_old_xmit(struct send_queue *sq, struct netdev_queue *txq,
546 			    bool in_napi, struct virtnet_sq_free_stats *stats)
547 {
548 	unsigned int len;
549 	void *ptr;
550 
551 	while ((ptr = virtqueue_get_buf(sq->vq, &len)) != NULL) {
552 		if (!is_xdp_frame(ptr)) {
553 			struct sk_buff *skb = ptr_to_skb(ptr);
554 
555 			pr_debug("Sent skb %p\n", skb);
556 
557 			if (is_orphan_skb(ptr)) {
558 				stats->packets++;
559 				stats->bytes += skb->len;
560 			} else {
561 				stats->napi_packets++;
562 				stats->napi_bytes += skb->len;
563 			}
564 			napi_consume_skb(skb, in_napi);
565 		} else {
566 			struct xdp_frame *frame = ptr_to_xdp(ptr);
567 
568 			stats->packets++;
569 			stats->bytes += xdp_get_frame_len(frame);
570 			xdp_return_frame(frame);
571 		}
572 	}
573 	netdev_tx_completed_queue(txq, stats->napi_packets, stats->napi_bytes);
574 }
575 
576 /* Converting between virtqueue no. and kernel tx/rx queue no.
577  * 0:rx0 1:tx0 2:rx1 3:tx1 ... 2N:rxN 2N+1:txN 2N+2:cvq
578  */
579 static int vq2txq(struct virtqueue *vq)
580 {
581 	return (vq->index - 1) / 2;
582 }
583 
584 static int txq2vq(int txq)
585 {
586 	return txq * 2 + 1;
587 }
588 
589 static int vq2rxq(struct virtqueue *vq)
590 {
591 	return vq->index / 2;
592 }
593 
594 static int rxq2vq(int rxq)
595 {
596 	return rxq * 2;
597 }
598 
599 static int vq_type(struct virtnet_info *vi, int qid)
600 {
601 	if (qid == vi->max_queue_pairs * 2)
602 		return VIRTNET_Q_TYPE_CQ;
603 
604 	if (qid % 2)
605 		return VIRTNET_Q_TYPE_TX;
606 
607 	return VIRTNET_Q_TYPE_RX;
608 }
609 
610 static inline struct virtio_net_common_hdr *
611 skb_vnet_common_hdr(struct sk_buff *skb)
612 {
613 	return (struct virtio_net_common_hdr *)skb->cb;
614 }
615 
616 /*
617  * private is used to chain pages for big packets, put the whole
618  * most recent used list in the beginning for reuse
619  */
620 static void give_pages(struct receive_queue *rq, struct page *page)
621 {
622 	struct page *end;
623 
624 	/* Find end of list, sew whole thing into vi->rq.pages. */
625 	for (end = page; end->private; end = (struct page *)end->private);
626 	end->private = (unsigned long)rq->pages;
627 	rq->pages = page;
628 }
629 
630 static struct page *get_a_page(struct receive_queue *rq, gfp_t gfp_mask)
631 {
632 	struct page *p = rq->pages;
633 
634 	if (p) {
635 		rq->pages = (struct page *)p->private;
636 		/* clear private here, it is used to chain pages */
637 		p->private = 0;
638 	} else
639 		p = alloc_page(gfp_mask);
640 	return p;
641 }
642 
643 static void virtnet_rq_free_buf(struct virtnet_info *vi,
644 				struct receive_queue *rq, void *buf)
645 {
646 	if (vi->mergeable_rx_bufs)
647 		put_page(virt_to_head_page(buf));
648 	else if (vi->big_packets)
649 		give_pages(rq, buf);
650 	else
651 		put_page(virt_to_head_page(buf));
652 }
653 
654 static void enable_delayed_refill(struct virtnet_info *vi)
655 {
656 	spin_lock_bh(&vi->refill_lock);
657 	vi->refill_enabled = true;
658 	spin_unlock_bh(&vi->refill_lock);
659 }
660 
661 static void disable_delayed_refill(struct virtnet_info *vi)
662 {
663 	spin_lock_bh(&vi->refill_lock);
664 	vi->refill_enabled = false;
665 	spin_unlock_bh(&vi->refill_lock);
666 }
667 
668 static void enable_rx_mode_work(struct virtnet_info *vi)
669 {
670 	rtnl_lock();
671 	vi->rx_mode_work_enabled = true;
672 	rtnl_unlock();
673 }
674 
675 static void disable_rx_mode_work(struct virtnet_info *vi)
676 {
677 	rtnl_lock();
678 	vi->rx_mode_work_enabled = false;
679 	rtnl_unlock();
680 }
681 
682 static void virtqueue_napi_schedule(struct napi_struct *napi,
683 				    struct virtqueue *vq)
684 {
685 	if (napi_schedule_prep(napi)) {
686 		virtqueue_disable_cb(vq);
687 		__napi_schedule(napi);
688 	}
689 }
690 
691 static bool virtqueue_napi_complete(struct napi_struct *napi,
692 				    struct virtqueue *vq, int processed)
693 {
694 	int opaque;
695 
696 	opaque = virtqueue_enable_cb_prepare(vq);
697 	if (napi_complete_done(napi, processed)) {
698 		if (unlikely(virtqueue_poll(vq, opaque)))
699 			virtqueue_napi_schedule(napi, vq);
700 		else
701 			return true;
702 	} else {
703 		virtqueue_disable_cb(vq);
704 	}
705 
706 	return false;
707 }
708 
709 static void skb_xmit_done(struct virtqueue *vq)
710 {
711 	struct virtnet_info *vi = vq->vdev->priv;
712 	struct napi_struct *napi = &vi->sq[vq2txq(vq)].napi;
713 
714 	/* Suppress further interrupts. */
715 	virtqueue_disable_cb(vq);
716 
717 	if (napi->weight)
718 		virtqueue_napi_schedule(napi, vq);
719 	else
720 		/* We were probably waiting for more output buffers. */
721 		netif_wake_subqueue(vi->dev, vq2txq(vq));
722 }
723 
724 #define MRG_CTX_HEADER_SHIFT 22
725 static void *mergeable_len_to_ctx(unsigned int truesize,
726 				  unsigned int headroom)
727 {
728 	return (void *)(unsigned long)((headroom << MRG_CTX_HEADER_SHIFT) | truesize);
729 }
730 
731 static unsigned int mergeable_ctx_to_headroom(void *mrg_ctx)
732 {
733 	return (unsigned long)mrg_ctx >> MRG_CTX_HEADER_SHIFT;
734 }
735 
736 static unsigned int mergeable_ctx_to_truesize(void *mrg_ctx)
737 {
738 	return (unsigned long)mrg_ctx & ((1 << MRG_CTX_HEADER_SHIFT) - 1);
739 }
740 
741 static struct sk_buff *virtnet_build_skb(void *buf, unsigned int buflen,
742 					 unsigned int headroom,
743 					 unsigned int len)
744 {
745 	struct sk_buff *skb;
746 
747 	skb = build_skb(buf, buflen);
748 	if (unlikely(!skb))
749 		return NULL;
750 
751 	skb_reserve(skb, headroom);
752 	skb_put(skb, len);
753 
754 	return skb;
755 }
756 
757 /* Called from bottom half context */
758 static struct sk_buff *page_to_skb(struct virtnet_info *vi,
759 				   struct receive_queue *rq,
760 				   struct page *page, unsigned int offset,
761 				   unsigned int len, unsigned int truesize,
762 				   unsigned int headroom)
763 {
764 	struct sk_buff *skb;
765 	struct virtio_net_common_hdr *hdr;
766 	unsigned int copy, hdr_len, hdr_padded_len;
767 	struct page *page_to_free = NULL;
768 	int tailroom, shinfo_size;
769 	char *p, *hdr_p, *buf;
770 
771 	p = page_address(page) + offset;
772 	hdr_p = p;
773 
774 	hdr_len = vi->hdr_len;
775 	if (vi->mergeable_rx_bufs)
776 		hdr_padded_len = hdr_len;
777 	else
778 		hdr_padded_len = sizeof(struct padded_vnet_hdr);
779 
780 	buf = p - headroom;
781 	len -= hdr_len;
782 	offset += hdr_padded_len;
783 	p += hdr_padded_len;
784 	tailroom = truesize - headroom  - hdr_padded_len - len;
785 
786 	shinfo_size = SKB_DATA_ALIGN(sizeof(struct skb_shared_info));
787 
788 	if (!NET_IP_ALIGN && len > GOOD_COPY_LEN && tailroom >= shinfo_size) {
789 		skb = virtnet_build_skb(buf, truesize, p - buf, len);
790 		if (unlikely(!skb))
791 			return NULL;
792 
793 		page = (struct page *)page->private;
794 		if (page)
795 			give_pages(rq, page);
796 		goto ok;
797 	}
798 
799 	/* copy small packet so we can reuse these pages for small data */
800 	skb = napi_alloc_skb(&rq->napi, GOOD_COPY_LEN);
801 	if (unlikely(!skb))
802 		return NULL;
803 
804 	/* Copy all frame if it fits skb->head, otherwise
805 	 * we let virtio_net_hdr_to_skb() and GRO pull headers as needed.
806 	 */
807 	if (len <= skb_tailroom(skb))
808 		copy = len;
809 	else
810 		copy = ETH_HLEN;
811 	skb_put_data(skb, p, copy);
812 
813 	len -= copy;
814 	offset += copy;
815 
816 	if (vi->mergeable_rx_bufs) {
817 		if (len)
818 			skb_add_rx_frag(skb, 0, page, offset, len, truesize);
819 		else
820 			page_to_free = page;
821 		goto ok;
822 	}
823 
824 	/*
825 	 * Verify that we can indeed put this data into a skb.
826 	 * This is here to handle cases when the device erroneously
827 	 * tries to receive more than is possible. This is usually
828 	 * the case of a broken device.
829 	 */
830 	if (unlikely(len > MAX_SKB_FRAGS * PAGE_SIZE)) {
831 		net_dbg_ratelimited("%s: too much data\n", skb->dev->name);
832 		dev_kfree_skb(skb);
833 		return NULL;
834 	}
835 	BUG_ON(offset >= PAGE_SIZE);
836 	while (len) {
837 		unsigned int frag_size = min((unsigned)PAGE_SIZE - offset, len);
838 		skb_add_rx_frag(skb, skb_shinfo(skb)->nr_frags, page, offset,
839 				frag_size, truesize);
840 		len -= frag_size;
841 		page = (struct page *)page->private;
842 		offset = 0;
843 	}
844 
845 	if (page)
846 		give_pages(rq, page);
847 
848 ok:
849 	hdr = skb_vnet_common_hdr(skb);
850 	memcpy(hdr, hdr_p, hdr_len);
851 	if (page_to_free)
852 		put_page(page_to_free);
853 
854 	return skb;
855 }
856 
857 static void virtnet_rq_unmap(struct receive_queue *rq, void *buf, u32 len)
858 {
859 	struct page *page = virt_to_head_page(buf);
860 	struct virtnet_rq_dma *dma;
861 	void *head;
862 	int offset;
863 
864 	head = page_address(page);
865 
866 	dma = head;
867 
868 	--dma->ref;
869 
870 	if (dma->need_sync && len) {
871 		offset = buf - (head + sizeof(*dma));
872 
873 		virtqueue_dma_sync_single_range_for_cpu(rq->vq, dma->addr,
874 							offset, len,
875 							DMA_FROM_DEVICE);
876 	}
877 
878 	if (dma->ref)
879 		return;
880 
881 	virtqueue_dma_unmap_single_attrs(rq->vq, dma->addr, dma->len,
882 					 DMA_FROM_DEVICE, DMA_ATTR_SKIP_CPU_SYNC);
883 	put_page(page);
884 }
885 
886 static void *virtnet_rq_get_buf(struct receive_queue *rq, u32 *len, void **ctx)
887 {
888 	void *buf;
889 
890 	buf = virtqueue_get_buf_ctx(rq->vq, len, ctx);
891 	if (buf && rq->do_dma)
892 		virtnet_rq_unmap(rq, buf, *len);
893 
894 	return buf;
895 }
896 
897 static void virtnet_rq_init_one_sg(struct receive_queue *rq, void *buf, u32 len)
898 {
899 	struct virtnet_rq_dma *dma;
900 	dma_addr_t addr;
901 	u32 offset;
902 	void *head;
903 
904 	if (!rq->do_dma) {
905 		sg_init_one(rq->sg, buf, len);
906 		return;
907 	}
908 
909 	head = page_address(rq->alloc_frag.page);
910 
911 	offset = buf - head;
912 
913 	dma = head;
914 
915 	addr = dma->addr - sizeof(*dma) + offset;
916 
917 	sg_init_table(rq->sg, 1);
918 	rq->sg[0].dma_address = addr;
919 	rq->sg[0].length = len;
920 }
921 
922 static void *virtnet_rq_alloc(struct receive_queue *rq, u32 size, gfp_t gfp)
923 {
924 	struct page_frag *alloc_frag = &rq->alloc_frag;
925 	struct virtnet_rq_dma *dma;
926 	void *buf, *head;
927 	dma_addr_t addr;
928 
929 	if (unlikely(!skb_page_frag_refill(size, alloc_frag, gfp)))
930 		return NULL;
931 
932 	head = page_address(alloc_frag->page);
933 
934 	if (rq->do_dma) {
935 		dma = head;
936 
937 		/* new pages */
938 		if (!alloc_frag->offset) {
939 			if (rq->last_dma) {
940 				/* Now, the new page is allocated, the last dma
941 				 * will not be used. So the dma can be unmapped
942 				 * if the ref is 0.
943 				 */
944 				virtnet_rq_unmap(rq, rq->last_dma, 0);
945 				rq->last_dma = NULL;
946 			}
947 
948 			dma->len = alloc_frag->size - sizeof(*dma);
949 
950 			addr = virtqueue_dma_map_single_attrs(rq->vq, dma + 1,
951 							      dma->len, DMA_FROM_DEVICE, 0);
952 			if (virtqueue_dma_mapping_error(rq->vq, addr))
953 				return NULL;
954 
955 			dma->addr = addr;
956 			dma->need_sync = virtqueue_dma_need_sync(rq->vq, addr);
957 
958 			/* Add a reference to dma to prevent the entire dma from
959 			 * being released during error handling. This reference
960 			 * will be freed after the pages are no longer used.
961 			 */
962 			get_page(alloc_frag->page);
963 			dma->ref = 1;
964 			alloc_frag->offset = sizeof(*dma);
965 
966 			rq->last_dma = dma;
967 		}
968 
969 		++dma->ref;
970 	}
971 
972 	buf = head + alloc_frag->offset;
973 
974 	get_page(alloc_frag->page);
975 	alloc_frag->offset += size;
976 
977 	return buf;
978 }
979 
980 static void virtnet_rq_unmap_free_buf(struct virtqueue *vq, void *buf)
981 {
982 	struct virtnet_info *vi = vq->vdev->priv;
983 	struct receive_queue *rq;
984 	int i = vq2rxq(vq);
985 
986 	rq = &vi->rq[i];
987 
988 	if (rq->xsk_pool) {
989 		xsk_buff_free((struct xdp_buff *)buf);
990 		return;
991 	}
992 
993 	if (rq->do_dma)
994 		virtnet_rq_unmap(rq, buf, 0);
995 
996 	virtnet_rq_free_buf(vi, rq, buf);
997 }
998 
999 static void free_old_xmit(struct send_queue *sq, struct netdev_queue *txq,
1000 			  bool in_napi)
1001 {
1002 	struct virtnet_sq_free_stats stats = {0};
1003 
1004 	__free_old_xmit(sq, txq, in_napi, &stats);
1005 
1006 	/* Avoid overhead when no packets have been processed
1007 	 * happens when called speculatively from start_xmit.
1008 	 */
1009 	if (!stats.packets && !stats.napi_packets)
1010 		return;
1011 
1012 	u64_stats_update_begin(&sq->stats.syncp);
1013 	u64_stats_add(&sq->stats.bytes, stats.bytes + stats.napi_bytes);
1014 	u64_stats_add(&sq->stats.packets, stats.packets + stats.napi_packets);
1015 	u64_stats_update_end(&sq->stats.syncp);
1016 }
1017 
1018 static bool is_xdp_raw_buffer_queue(struct virtnet_info *vi, int q)
1019 {
1020 	if (q < (vi->curr_queue_pairs - vi->xdp_queue_pairs))
1021 		return false;
1022 	else if (q < vi->curr_queue_pairs)
1023 		return true;
1024 	else
1025 		return false;
1026 }
1027 
1028 static void check_sq_full_and_disable(struct virtnet_info *vi,
1029 				      struct net_device *dev,
1030 				      struct send_queue *sq)
1031 {
1032 	bool use_napi = sq->napi.weight;
1033 	int qnum;
1034 
1035 	qnum = sq - vi->sq;
1036 
1037 	/* If running out of space, stop queue to avoid getting packets that we
1038 	 * are then unable to transmit.
1039 	 * An alternative would be to force queuing layer to requeue the skb by
1040 	 * returning NETDEV_TX_BUSY. However, NETDEV_TX_BUSY should not be
1041 	 * returned in a normal path of operation: it means that driver is not
1042 	 * maintaining the TX queue stop/start state properly, and causes
1043 	 * the stack to do a non-trivial amount of useless work.
1044 	 * Since most packets only take 1 or 2 ring slots, stopping the queue
1045 	 * early means 16 slots are typically wasted.
1046 	 */
1047 	if (sq->vq->num_free < 2+MAX_SKB_FRAGS) {
1048 		struct netdev_queue *txq = netdev_get_tx_queue(dev, qnum);
1049 
1050 		netif_tx_stop_queue(txq);
1051 		u64_stats_update_begin(&sq->stats.syncp);
1052 		u64_stats_inc(&sq->stats.stop);
1053 		u64_stats_update_end(&sq->stats.syncp);
1054 		if (use_napi) {
1055 			if (unlikely(!virtqueue_enable_cb_delayed(sq->vq)))
1056 				virtqueue_napi_schedule(&sq->napi, sq->vq);
1057 		} else if (unlikely(!virtqueue_enable_cb_delayed(sq->vq))) {
1058 			/* More just got used, free them then recheck. */
1059 			free_old_xmit(sq, txq, false);
1060 			if (sq->vq->num_free >= 2+MAX_SKB_FRAGS) {
1061 				netif_start_subqueue(dev, qnum);
1062 				u64_stats_update_begin(&sq->stats.syncp);
1063 				u64_stats_inc(&sq->stats.wake);
1064 				u64_stats_update_end(&sq->stats.syncp);
1065 				virtqueue_disable_cb(sq->vq);
1066 			}
1067 		}
1068 	}
1069 }
1070 
1071 static void sg_fill_dma(struct scatterlist *sg, dma_addr_t addr, u32 len)
1072 {
1073 	sg->dma_address = addr;
1074 	sg->length = len;
1075 }
1076 
1077 static struct xdp_buff *buf_to_xdp(struct virtnet_info *vi,
1078 				   struct receive_queue *rq, void *buf, u32 len)
1079 {
1080 	struct xdp_buff *xdp;
1081 	u32 bufsize;
1082 
1083 	xdp = (struct xdp_buff *)buf;
1084 
1085 	bufsize = xsk_pool_get_rx_frame_size(rq->xsk_pool) + vi->hdr_len;
1086 
1087 	if (unlikely(len > bufsize)) {
1088 		pr_debug("%s: rx error: len %u exceeds truesize %u\n",
1089 			 vi->dev->name, len, bufsize);
1090 		DEV_STATS_INC(vi->dev, rx_length_errors);
1091 		xsk_buff_free(xdp);
1092 		return NULL;
1093 	}
1094 
1095 	xsk_buff_set_size(xdp, len);
1096 	xsk_buff_dma_sync_for_cpu(xdp);
1097 
1098 	return xdp;
1099 }
1100 
1101 static struct sk_buff *xsk_construct_skb(struct receive_queue *rq,
1102 					 struct xdp_buff *xdp)
1103 {
1104 	unsigned int metasize = xdp->data - xdp->data_meta;
1105 	struct sk_buff *skb;
1106 	unsigned int size;
1107 
1108 	size = xdp->data_end - xdp->data_hard_start;
1109 	skb = napi_alloc_skb(&rq->napi, size);
1110 	if (unlikely(!skb)) {
1111 		xsk_buff_free(xdp);
1112 		return NULL;
1113 	}
1114 
1115 	skb_reserve(skb, xdp->data_meta - xdp->data_hard_start);
1116 
1117 	size = xdp->data_end - xdp->data_meta;
1118 	memcpy(__skb_put(skb, size), xdp->data_meta, size);
1119 
1120 	if (metasize) {
1121 		__skb_pull(skb, metasize);
1122 		skb_metadata_set(skb, metasize);
1123 	}
1124 
1125 	xsk_buff_free(xdp);
1126 
1127 	return skb;
1128 }
1129 
1130 static struct sk_buff *virtnet_receive_xsk_small(struct net_device *dev, struct virtnet_info *vi,
1131 						 struct receive_queue *rq, struct xdp_buff *xdp,
1132 						 unsigned int *xdp_xmit,
1133 						 struct virtnet_rq_stats *stats)
1134 {
1135 	struct bpf_prog *prog;
1136 	u32 ret;
1137 
1138 	ret = XDP_PASS;
1139 	rcu_read_lock();
1140 	prog = rcu_dereference(rq->xdp_prog);
1141 	if (prog)
1142 		ret = virtnet_xdp_handler(prog, xdp, dev, xdp_xmit, stats);
1143 	rcu_read_unlock();
1144 
1145 	switch (ret) {
1146 	case XDP_PASS:
1147 		return xsk_construct_skb(rq, xdp);
1148 
1149 	case XDP_TX:
1150 	case XDP_REDIRECT:
1151 		return NULL;
1152 
1153 	default:
1154 		/* drop packet */
1155 		xsk_buff_free(xdp);
1156 		u64_stats_inc(&stats->drops);
1157 		return NULL;
1158 	}
1159 }
1160 
1161 static void xsk_drop_follow_bufs(struct net_device *dev,
1162 				 struct receive_queue *rq,
1163 				 u32 num_buf,
1164 				 struct virtnet_rq_stats *stats)
1165 {
1166 	struct xdp_buff *xdp;
1167 	u32 len;
1168 
1169 	while (num_buf-- > 1) {
1170 		xdp = virtqueue_get_buf(rq->vq, &len);
1171 		if (unlikely(!xdp)) {
1172 			pr_debug("%s: rx error: %d buffers missing\n",
1173 				 dev->name, num_buf);
1174 			DEV_STATS_INC(dev, rx_length_errors);
1175 			break;
1176 		}
1177 		u64_stats_add(&stats->bytes, len);
1178 		xsk_buff_free(xdp);
1179 	}
1180 }
1181 
1182 static int xsk_append_merge_buffer(struct virtnet_info *vi,
1183 				   struct receive_queue *rq,
1184 				   struct sk_buff *head_skb,
1185 				   u32 num_buf,
1186 				   struct virtio_net_hdr_mrg_rxbuf *hdr,
1187 				   struct virtnet_rq_stats *stats)
1188 {
1189 	struct sk_buff *curr_skb;
1190 	struct xdp_buff *xdp;
1191 	u32 len, truesize;
1192 	struct page *page;
1193 	void *buf;
1194 
1195 	curr_skb = head_skb;
1196 
1197 	while (--num_buf) {
1198 		buf = virtqueue_get_buf(rq->vq, &len);
1199 		if (unlikely(!buf)) {
1200 			pr_debug("%s: rx error: %d buffers out of %d missing\n",
1201 				 vi->dev->name, num_buf,
1202 				 virtio16_to_cpu(vi->vdev,
1203 						 hdr->num_buffers));
1204 			DEV_STATS_INC(vi->dev, rx_length_errors);
1205 			return -EINVAL;
1206 		}
1207 
1208 		u64_stats_add(&stats->bytes, len);
1209 
1210 		xdp = buf_to_xdp(vi, rq, buf, len);
1211 		if (!xdp)
1212 			goto err;
1213 
1214 		buf = napi_alloc_frag(len);
1215 		if (!buf) {
1216 			xsk_buff_free(xdp);
1217 			goto err;
1218 		}
1219 
1220 		memcpy(buf, xdp->data - vi->hdr_len, len);
1221 
1222 		xsk_buff_free(xdp);
1223 
1224 		page = virt_to_page(buf);
1225 
1226 		truesize = len;
1227 
1228 		curr_skb  = virtnet_skb_append_frag(head_skb, curr_skb, page,
1229 						    buf, len, truesize);
1230 		if (!curr_skb) {
1231 			put_page(page);
1232 			goto err;
1233 		}
1234 	}
1235 
1236 	return 0;
1237 
1238 err:
1239 	xsk_drop_follow_bufs(vi->dev, rq, num_buf, stats);
1240 	return -EINVAL;
1241 }
1242 
1243 static struct sk_buff *virtnet_receive_xsk_merge(struct net_device *dev, struct virtnet_info *vi,
1244 						 struct receive_queue *rq, struct xdp_buff *xdp,
1245 						 unsigned int *xdp_xmit,
1246 						 struct virtnet_rq_stats *stats)
1247 {
1248 	struct virtio_net_hdr_mrg_rxbuf *hdr;
1249 	struct bpf_prog *prog;
1250 	struct sk_buff *skb;
1251 	u32 ret, num_buf;
1252 
1253 	hdr = xdp->data - vi->hdr_len;
1254 	num_buf = virtio16_to_cpu(vi->vdev, hdr->num_buffers);
1255 
1256 	ret = XDP_PASS;
1257 	rcu_read_lock();
1258 	prog = rcu_dereference(rq->xdp_prog);
1259 	/* TODO: support multi buffer. */
1260 	if (prog && num_buf == 1)
1261 		ret = virtnet_xdp_handler(prog, xdp, dev, xdp_xmit, stats);
1262 	rcu_read_unlock();
1263 
1264 	switch (ret) {
1265 	case XDP_PASS:
1266 		skb = xsk_construct_skb(rq, xdp);
1267 		if (!skb)
1268 			goto drop_bufs;
1269 
1270 		if (xsk_append_merge_buffer(vi, rq, skb, num_buf, hdr, stats)) {
1271 			dev_kfree_skb(skb);
1272 			goto drop;
1273 		}
1274 
1275 		return skb;
1276 
1277 	case XDP_TX:
1278 	case XDP_REDIRECT:
1279 		return NULL;
1280 
1281 	default:
1282 		/* drop packet */
1283 		xsk_buff_free(xdp);
1284 	}
1285 
1286 drop_bufs:
1287 	xsk_drop_follow_bufs(dev, rq, num_buf, stats);
1288 
1289 drop:
1290 	u64_stats_inc(&stats->drops);
1291 	return NULL;
1292 }
1293 
1294 static void virtnet_receive_xsk_buf(struct virtnet_info *vi, struct receive_queue *rq,
1295 				    void *buf, u32 len,
1296 				    unsigned int *xdp_xmit,
1297 				    struct virtnet_rq_stats *stats)
1298 {
1299 	struct net_device *dev = vi->dev;
1300 	struct sk_buff *skb = NULL;
1301 	struct xdp_buff *xdp;
1302 	u8 flags;
1303 
1304 	len -= vi->hdr_len;
1305 
1306 	u64_stats_add(&stats->bytes, len);
1307 
1308 	xdp = buf_to_xdp(vi, rq, buf, len);
1309 	if (!xdp)
1310 		return;
1311 
1312 	if (unlikely(len < ETH_HLEN)) {
1313 		pr_debug("%s: short packet %i\n", dev->name, len);
1314 		DEV_STATS_INC(dev, rx_length_errors);
1315 		xsk_buff_free(xdp);
1316 		return;
1317 	}
1318 
1319 	flags = ((struct virtio_net_common_hdr *)(xdp->data - vi->hdr_len))->hdr.flags;
1320 
1321 	if (!vi->mergeable_rx_bufs)
1322 		skb = virtnet_receive_xsk_small(dev, vi, rq, xdp, xdp_xmit, stats);
1323 	else
1324 		skb = virtnet_receive_xsk_merge(dev, vi, rq, xdp, xdp_xmit, stats);
1325 
1326 	if (skb)
1327 		virtnet_receive_done(vi, rq, skb, flags);
1328 }
1329 
1330 static int virtnet_add_recvbuf_xsk(struct virtnet_info *vi, struct receive_queue *rq,
1331 				   struct xsk_buff_pool *pool, gfp_t gfp)
1332 {
1333 	struct xdp_buff **xsk_buffs;
1334 	dma_addr_t addr;
1335 	int err = 0;
1336 	u32 len, i;
1337 	int num;
1338 
1339 	xsk_buffs = rq->xsk_buffs;
1340 
1341 	num = xsk_buff_alloc_batch(pool, xsk_buffs, rq->vq->num_free);
1342 	if (!num)
1343 		return -ENOMEM;
1344 
1345 	len = xsk_pool_get_rx_frame_size(pool) + vi->hdr_len;
1346 
1347 	for (i = 0; i < num; ++i) {
1348 		/* Use the part of XDP_PACKET_HEADROOM as the virtnet hdr space.
1349 		 * We assume XDP_PACKET_HEADROOM is larger than hdr->len.
1350 		 * (see function virtnet_xsk_pool_enable)
1351 		 */
1352 		addr = xsk_buff_xdp_get_dma(xsk_buffs[i]) - vi->hdr_len;
1353 
1354 		sg_init_table(rq->sg, 1);
1355 		sg_fill_dma(rq->sg, addr, len);
1356 
1357 		err = virtqueue_add_inbuf(rq->vq, rq->sg, 1, xsk_buffs[i], gfp);
1358 		if (err)
1359 			goto err;
1360 	}
1361 
1362 	return num;
1363 
1364 err:
1365 	for (; i < num; ++i)
1366 		xsk_buff_free(xsk_buffs[i]);
1367 
1368 	return err;
1369 }
1370 
1371 static int virtnet_xsk_wakeup(struct net_device *dev, u32 qid, u32 flag)
1372 {
1373 	struct virtnet_info *vi = netdev_priv(dev);
1374 	struct send_queue *sq;
1375 
1376 	if (!netif_running(dev))
1377 		return -ENETDOWN;
1378 
1379 	if (qid >= vi->curr_queue_pairs)
1380 		return -EINVAL;
1381 
1382 	sq = &vi->sq[qid];
1383 
1384 	if (napi_if_scheduled_mark_missed(&sq->napi))
1385 		return 0;
1386 
1387 	local_bh_disable();
1388 	virtqueue_napi_schedule(&sq->napi, sq->vq);
1389 	local_bh_enable();
1390 
1391 	return 0;
1392 }
1393 
1394 static int __virtnet_xdp_xmit_one(struct virtnet_info *vi,
1395 				   struct send_queue *sq,
1396 				   struct xdp_frame *xdpf)
1397 {
1398 	struct virtio_net_hdr_mrg_rxbuf *hdr;
1399 	struct skb_shared_info *shinfo;
1400 	u8 nr_frags = 0;
1401 	int err, i;
1402 
1403 	if (unlikely(xdpf->headroom < vi->hdr_len))
1404 		return -EOVERFLOW;
1405 
1406 	if (unlikely(xdp_frame_has_frags(xdpf))) {
1407 		shinfo = xdp_get_shared_info_from_frame(xdpf);
1408 		nr_frags = shinfo->nr_frags;
1409 	}
1410 
1411 	/* In wrapping function virtnet_xdp_xmit(), we need to free
1412 	 * up the pending old buffers, where we need to calculate the
1413 	 * position of skb_shared_info in xdp_get_frame_len() and
1414 	 * xdp_return_frame(), which will involve to xdpf->data and
1415 	 * xdpf->headroom. Therefore, we need to update the value of
1416 	 * headroom synchronously here.
1417 	 */
1418 	xdpf->headroom -= vi->hdr_len;
1419 	xdpf->data -= vi->hdr_len;
1420 	/* Zero header and leave csum up to XDP layers */
1421 	hdr = xdpf->data;
1422 	memset(hdr, 0, vi->hdr_len);
1423 	xdpf->len   += vi->hdr_len;
1424 
1425 	sg_init_table(sq->sg, nr_frags + 1);
1426 	sg_set_buf(sq->sg, xdpf->data, xdpf->len);
1427 	for (i = 0; i < nr_frags; i++) {
1428 		skb_frag_t *frag = &shinfo->frags[i];
1429 
1430 		sg_set_page(&sq->sg[i + 1], skb_frag_page(frag),
1431 			    skb_frag_size(frag), skb_frag_off(frag));
1432 	}
1433 
1434 	err = virtqueue_add_outbuf(sq->vq, sq->sg, nr_frags + 1,
1435 				   xdp_to_ptr(xdpf), GFP_ATOMIC);
1436 	if (unlikely(err))
1437 		return -ENOSPC; /* Caller handle free/refcnt */
1438 
1439 	return 0;
1440 }
1441 
1442 /* when vi->curr_queue_pairs > nr_cpu_ids, the txq/sq is only used for xdp tx on
1443  * the current cpu, so it does not need to be locked.
1444  *
1445  * Here we use marco instead of inline functions because we have to deal with
1446  * three issues at the same time: 1. the choice of sq. 2. judge and execute the
1447  * lock/unlock of txq 3. make sparse happy. It is difficult for two inline
1448  * functions to perfectly solve these three problems at the same time.
1449  */
1450 #define virtnet_xdp_get_sq(vi) ({                                       \
1451 	int cpu = smp_processor_id();                                   \
1452 	struct netdev_queue *txq;                                       \
1453 	typeof(vi) v = (vi);                                            \
1454 	unsigned int qp;                                                \
1455 									\
1456 	if (v->curr_queue_pairs > nr_cpu_ids) {                         \
1457 		qp = v->curr_queue_pairs - v->xdp_queue_pairs;          \
1458 		qp += cpu;                                              \
1459 		txq = netdev_get_tx_queue(v->dev, qp);                  \
1460 		__netif_tx_acquire(txq);                                \
1461 	} else {                                                        \
1462 		qp = cpu % v->curr_queue_pairs;                         \
1463 		txq = netdev_get_tx_queue(v->dev, qp);                  \
1464 		__netif_tx_lock(txq, cpu);                              \
1465 	}                                                               \
1466 	v->sq + qp;                                                     \
1467 })
1468 
1469 #define virtnet_xdp_put_sq(vi, q) {                                     \
1470 	struct netdev_queue *txq;                                       \
1471 	typeof(vi) v = (vi);                                            \
1472 									\
1473 	txq = netdev_get_tx_queue(v->dev, (q) - v->sq);                 \
1474 	if (v->curr_queue_pairs > nr_cpu_ids)                           \
1475 		__netif_tx_release(txq);                                \
1476 	else                                                            \
1477 		__netif_tx_unlock(txq);                                 \
1478 }
1479 
1480 static int virtnet_xdp_xmit(struct net_device *dev,
1481 			    int n, struct xdp_frame **frames, u32 flags)
1482 {
1483 	struct virtnet_info *vi = netdev_priv(dev);
1484 	struct virtnet_sq_free_stats stats = {0};
1485 	struct receive_queue *rq = vi->rq;
1486 	struct bpf_prog *xdp_prog;
1487 	struct send_queue *sq;
1488 	int nxmit = 0;
1489 	int kicks = 0;
1490 	int ret;
1491 	int i;
1492 
1493 	/* Only allow ndo_xdp_xmit if XDP is loaded on dev, as this
1494 	 * indicate XDP resources have been successfully allocated.
1495 	 */
1496 	xdp_prog = rcu_access_pointer(rq->xdp_prog);
1497 	if (!xdp_prog)
1498 		return -ENXIO;
1499 
1500 	sq = virtnet_xdp_get_sq(vi);
1501 
1502 	if (unlikely(flags & ~XDP_XMIT_FLAGS_MASK)) {
1503 		ret = -EINVAL;
1504 		goto out;
1505 	}
1506 
1507 	/* Free up any pending old buffers before queueing new ones. */
1508 	__free_old_xmit(sq, netdev_get_tx_queue(dev, sq - vi->sq),
1509 			false, &stats);
1510 
1511 	for (i = 0; i < n; i++) {
1512 		struct xdp_frame *xdpf = frames[i];
1513 
1514 		if (__virtnet_xdp_xmit_one(vi, sq, xdpf))
1515 			break;
1516 		nxmit++;
1517 	}
1518 	ret = nxmit;
1519 
1520 	if (!is_xdp_raw_buffer_queue(vi, sq - vi->sq))
1521 		check_sq_full_and_disable(vi, dev, sq);
1522 
1523 	if (flags & XDP_XMIT_FLUSH) {
1524 		if (virtqueue_kick_prepare(sq->vq) && virtqueue_notify(sq->vq))
1525 			kicks = 1;
1526 	}
1527 out:
1528 	u64_stats_update_begin(&sq->stats.syncp);
1529 	u64_stats_add(&sq->stats.bytes, stats.bytes);
1530 	u64_stats_add(&sq->stats.packets, stats.packets);
1531 	u64_stats_add(&sq->stats.xdp_tx, n);
1532 	u64_stats_add(&sq->stats.xdp_tx_drops, n - nxmit);
1533 	u64_stats_add(&sq->stats.kicks, kicks);
1534 	u64_stats_update_end(&sq->stats.syncp);
1535 
1536 	virtnet_xdp_put_sq(vi, sq);
1537 	return ret;
1538 }
1539 
1540 static void put_xdp_frags(struct xdp_buff *xdp)
1541 {
1542 	struct skb_shared_info *shinfo;
1543 	struct page *xdp_page;
1544 	int i;
1545 
1546 	if (xdp_buff_has_frags(xdp)) {
1547 		shinfo = xdp_get_shared_info_from_buff(xdp);
1548 		for (i = 0; i < shinfo->nr_frags; i++) {
1549 			xdp_page = skb_frag_page(&shinfo->frags[i]);
1550 			put_page(xdp_page);
1551 		}
1552 	}
1553 }
1554 
1555 static int virtnet_xdp_handler(struct bpf_prog *xdp_prog, struct xdp_buff *xdp,
1556 			       struct net_device *dev,
1557 			       unsigned int *xdp_xmit,
1558 			       struct virtnet_rq_stats *stats)
1559 {
1560 	struct xdp_frame *xdpf;
1561 	int err;
1562 	u32 act;
1563 
1564 	act = bpf_prog_run_xdp(xdp_prog, xdp);
1565 	u64_stats_inc(&stats->xdp_packets);
1566 
1567 	switch (act) {
1568 	case XDP_PASS:
1569 		return act;
1570 
1571 	case XDP_TX:
1572 		u64_stats_inc(&stats->xdp_tx);
1573 		xdpf = xdp_convert_buff_to_frame(xdp);
1574 		if (unlikely(!xdpf)) {
1575 			netdev_dbg(dev, "convert buff to frame failed for xdp\n");
1576 			return XDP_DROP;
1577 		}
1578 
1579 		err = virtnet_xdp_xmit(dev, 1, &xdpf, 0);
1580 		if (unlikely(!err)) {
1581 			xdp_return_frame_rx_napi(xdpf);
1582 		} else if (unlikely(err < 0)) {
1583 			trace_xdp_exception(dev, xdp_prog, act);
1584 			return XDP_DROP;
1585 		}
1586 		*xdp_xmit |= VIRTIO_XDP_TX;
1587 		return act;
1588 
1589 	case XDP_REDIRECT:
1590 		u64_stats_inc(&stats->xdp_redirects);
1591 		err = xdp_do_redirect(dev, xdp, xdp_prog);
1592 		if (err)
1593 			return XDP_DROP;
1594 
1595 		*xdp_xmit |= VIRTIO_XDP_REDIR;
1596 		return act;
1597 
1598 	default:
1599 		bpf_warn_invalid_xdp_action(dev, xdp_prog, act);
1600 		fallthrough;
1601 	case XDP_ABORTED:
1602 		trace_xdp_exception(dev, xdp_prog, act);
1603 		fallthrough;
1604 	case XDP_DROP:
1605 		return XDP_DROP;
1606 	}
1607 }
1608 
1609 static unsigned int virtnet_get_headroom(struct virtnet_info *vi)
1610 {
1611 	return vi->xdp_enabled ? XDP_PACKET_HEADROOM : 0;
1612 }
1613 
1614 /* We copy the packet for XDP in the following cases:
1615  *
1616  * 1) Packet is scattered across multiple rx buffers.
1617  * 2) Headroom space is insufficient.
1618  *
1619  * This is inefficient but it's a temporary condition that
1620  * we hit right after XDP is enabled and until queue is refilled
1621  * with large buffers with sufficient headroom - so it should affect
1622  * at most queue size packets.
1623  * Afterwards, the conditions to enable
1624  * XDP should preclude the underlying device from sending packets
1625  * across multiple buffers (num_buf > 1), and we make sure buffers
1626  * have enough headroom.
1627  */
1628 static struct page *xdp_linearize_page(struct receive_queue *rq,
1629 				       int *num_buf,
1630 				       struct page *p,
1631 				       int offset,
1632 				       int page_off,
1633 				       unsigned int *len)
1634 {
1635 	int tailroom = SKB_DATA_ALIGN(sizeof(struct skb_shared_info));
1636 	struct page *page;
1637 
1638 	if (page_off + *len + tailroom > PAGE_SIZE)
1639 		return NULL;
1640 
1641 	page = alloc_page(GFP_ATOMIC);
1642 	if (!page)
1643 		return NULL;
1644 
1645 	memcpy(page_address(page) + page_off, page_address(p) + offset, *len);
1646 	page_off += *len;
1647 
1648 	while (--*num_buf) {
1649 		unsigned int buflen;
1650 		void *buf;
1651 		int off;
1652 
1653 		buf = virtnet_rq_get_buf(rq, &buflen, NULL);
1654 		if (unlikely(!buf))
1655 			goto err_buf;
1656 
1657 		p = virt_to_head_page(buf);
1658 		off = buf - page_address(p);
1659 
1660 		/* guard against a misconfigured or uncooperative backend that
1661 		 * is sending packet larger than the MTU.
1662 		 */
1663 		if ((page_off + buflen + tailroom) > PAGE_SIZE) {
1664 			put_page(p);
1665 			goto err_buf;
1666 		}
1667 
1668 		memcpy(page_address(page) + page_off,
1669 		       page_address(p) + off, buflen);
1670 		page_off += buflen;
1671 		put_page(p);
1672 	}
1673 
1674 	/* Headroom does not contribute to packet length */
1675 	*len = page_off - XDP_PACKET_HEADROOM;
1676 	return page;
1677 err_buf:
1678 	__free_pages(page, 0);
1679 	return NULL;
1680 }
1681 
1682 static struct sk_buff *receive_small_build_skb(struct virtnet_info *vi,
1683 					       unsigned int xdp_headroom,
1684 					       void *buf,
1685 					       unsigned int len)
1686 {
1687 	unsigned int header_offset;
1688 	unsigned int headroom;
1689 	unsigned int buflen;
1690 	struct sk_buff *skb;
1691 
1692 	header_offset = VIRTNET_RX_PAD + xdp_headroom;
1693 	headroom = vi->hdr_len + header_offset;
1694 	buflen = SKB_DATA_ALIGN(GOOD_PACKET_LEN + headroom) +
1695 		SKB_DATA_ALIGN(sizeof(struct skb_shared_info));
1696 
1697 	skb = virtnet_build_skb(buf, buflen, headroom, len);
1698 	if (unlikely(!skb))
1699 		return NULL;
1700 
1701 	buf += header_offset;
1702 	memcpy(skb_vnet_common_hdr(skb), buf, vi->hdr_len);
1703 
1704 	return skb;
1705 }
1706 
1707 static struct sk_buff *receive_small_xdp(struct net_device *dev,
1708 					 struct virtnet_info *vi,
1709 					 struct receive_queue *rq,
1710 					 struct bpf_prog *xdp_prog,
1711 					 void *buf,
1712 					 unsigned int xdp_headroom,
1713 					 unsigned int len,
1714 					 unsigned int *xdp_xmit,
1715 					 struct virtnet_rq_stats *stats)
1716 {
1717 	unsigned int header_offset = VIRTNET_RX_PAD + xdp_headroom;
1718 	unsigned int headroom = vi->hdr_len + header_offset;
1719 	struct virtio_net_hdr_mrg_rxbuf *hdr = buf + header_offset;
1720 	struct page *page = virt_to_head_page(buf);
1721 	struct page *xdp_page;
1722 	unsigned int buflen;
1723 	struct xdp_buff xdp;
1724 	struct sk_buff *skb;
1725 	unsigned int metasize = 0;
1726 	u32 act;
1727 
1728 	if (unlikely(hdr->hdr.gso_type))
1729 		goto err_xdp;
1730 
1731 	/* Partially checksummed packets must be dropped. */
1732 	if (unlikely(hdr->hdr.flags & VIRTIO_NET_HDR_F_NEEDS_CSUM))
1733 		goto err_xdp;
1734 
1735 	buflen = SKB_DATA_ALIGN(GOOD_PACKET_LEN + headroom) +
1736 		SKB_DATA_ALIGN(sizeof(struct skb_shared_info));
1737 
1738 	if (unlikely(xdp_headroom < virtnet_get_headroom(vi))) {
1739 		int offset = buf - page_address(page) + header_offset;
1740 		unsigned int tlen = len + vi->hdr_len;
1741 		int num_buf = 1;
1742 
1743 		xdp_headroom = virtnet_get_headroom(vi);
1744 		header_offset = VIRTNET_RX_PAD + xdp_headroom;
1745 		headroom = vi->hdr_len + header_offset;
1746 		buflen = SKB_DATA_ALIGN(GOOD_PACKET_LEN + headroom) +
1747 			SKB_DATA_ALIGN(sizeof(struct skb_shared_info));
1748 		xdp_page = xdp_linearize_page(rq, &num_buf, page,
1749 					      offset, header_offset,
1750 					      &tlen);
1751 		if (!xdp_page)
1752 			goto err_xdp;
1753 
1754 		buf = page_address(xdp_page);
1755 		put_page(page);
1756 		page = xdp_page;
1757 	}
1758 
1759 	xdp_init_buff(&xdp, buflen, &rq->xdp_rxq);
1760 	xdp_prepare_buff(&xdp, buf + VIRTNET_RX_PAD + vi->hdr_len,
1761 			 xdp_headroom, len, true);
1762 
1763 	act = virtnet_xdp_handler(xdp_prog, &xdp, dev, xdp_xmit, stats);
1764 
1765 	switch (act) {
1766 	case XDP_PASS:
1767 		/* Recalculate length in case bpf program changed it */
1768 		len = xdp.data_end - xdp.data;
1769 		metasize = xdp.data - xdp.data_meta;
1770 		break;
1771 
1772 	case XDP_TX:
1773 	case XDP_REDIRECT:
1774 		goto xdp_xmit;
1775 
1776 	default:
1777 		goto err_xdp;
1778 	}
1779 
1780 	skb = virtnet_build_skb(buf, buflen, xdp.data - buf, len);
1781 	if (unlikely(!skb))
1782 		goto err;
1783 
1784 	if (metasize)
1785 		skb_metadata_set(skb, metasize);
1786 
1787 	return skb;
1788 
1789 err_xdp:
1790 	u64_stats_inc(&stats->xdp_drops);
1791 err:
1792 	u64_stats_inc(&stats->drops);
1793 	put_page(page);
1794 xdp_xmit:
1795 	return NULL;
1796 }
1797 
1798 static struct sk_buff *receive_small(struct net_device *dev,
1799 				     struct virtnet_info *vi,
1800 				     struct receive_queue *rq,
1801 				     void *buf, void *ctx,
1802 				     unsigned int len,
1803 				     unsigned int *xdp_xmit,
1804 				     struct virtnet_rq_stats *stats)
1805 {
1806 	unsigned int xdp_headroom = (unsigned long)ctx;
1807 	struct page *page = virt_to_head_page(buf);
1808 	struct sk_buff *skb;
1809 
1810 	/* We passed the address of virtnet header to virtio-core,
1811 	 * so truncate the padding.
1812 	 */
1813 	buf -= VIRTNET_RX_PAD + xdp_headroom;
1814 
1815 	len -= vi->hdr_len;
1816 	u64_stats_add(&stats->bytes, len);
1817 
1818 	if (unlikely(len > GOOD_PACKET_LEN)) {
1819 		pr_debug("%s: rx error: len %u exceeds max size %d\n",
1820 			 dev->name, len, GOOD_PACKET_LEN);
1821 		DEV_STATS_INC(dev, rx_length_errors);
1822 		goto err;
1823 	}
1824 
1825 	if (unlikely(vi->xdp_enabled)) {
1826 		struct bpf_prog *xdp_prog;
1827 
1828 		rcu_read_lock();
1829 		xdp_prog = rcu_dereference(rq->xdp_prog);
1830 		if (xdp_prog) {
1831 			skb = receive_small_xdp(dev, vi, rq, xdp_prog, buf,
1832 						xdp_headroom, len, xdp_xmit,
1833 						stats);
1834 			rcu_read_unlock();
1835 			return skb;
1836 		}
1837 		rcu_read_unlock();
1838 	}
1839 
1840 	skb = receive_small_build_skb(vi, xdp_headroom, buf, len);
1841 	if (likely(skb))
1842 		return skb;
1843 
1844 err:
1845 	u64_stats_inc(&stats->drops);
1846 	put_page(page);
1847 	return NULL;
1848 }
1849 
1850 static struct sk_buff *receive_big(struct net_device *dev,
1851 				   struct virtnet_info *vi,
1852 				   struct receive_queue *rq,
1853 				   void *buf,
1854 				   unsigned int len,
1855 				   struct virtnet_rq_stats *stats)
1856 {
1857 	struct page *page = buf;
1858 	struct sk_buff *skb =
1859 		page_to_skb(vi, rq, page, 0, len, PAGE_SIZE, 0);
1860 
1861 	u64_stats_add(&stats->bytes, len - vi->hdr_len);
1862 	if (unlikely(!skb))
1863 		goto err;
1864 
1865 	return skb;
1866 
1867 err:
1868 	u64_stats_inc(&stats->drops);
1869 	give_pages(rq, page);
1870 	return NULL;
1871 }
1872 
1873 static void mergeable_buf_free(struct receive_queue *rq, int num_buf,
1874 			       struct net_device *dev,
1875 			       struct virtnet_rq_stats *stats)
1876 {
1877 	struct page *page;
1878 	void *buf;
1879 	int len;
1880 
1881 	while (num_buf-- > 1) {
1882 		buf = virtnet_rq_get_buf(rq, &len, NULL);
1883 		if (unlikely(!buf)) {
1884 			pr_debug("%s: rx error: %d buffers missing\n",
1885 				 dev->name, num_buf);
1886 			DEV_STATS_INC(dev, rx_length_errors);
1887 			break;
1888 		}
1889 		u64_stats_add(&stats->bytes, len);
1890 		page = virt_to_head_page(buf);
1891 		put_page(page);
1892 	}
1893 }
1894 
1895 /* Why not use xdp_build_skb_from_frame() ?
1896  * XDP core assumes that xdp frags are PAGE_SIZE in length, while in
1897  * virtio-net there are 2 points that do not match its requirements:
1898  *  1. The size of the prefilled buffer is not fixed before xdp is set.
1899  *  2. xdp_build_skb_from_frame() does more checks that we don't need,
1900  *     like eth_type_trans() (which virtio-net does in receive_buf()).
1901  */
1902 static struct sk_buff *build_skb_from_xdp_buff(struct net_device *dev,
1903 					       struct virtnet_info *vi,
1904 					       struct xdp_buff *xdp,
1905 					       unsigned int xdp_frags_truesz)
1906 {
1907 	struct skb_shared_info *sinfo = xdp_get_shared_info_from_buff(xdp);
1908 	unsigned int headroom, data_len;
1909 	struct sk_buff *skb;
1910 	int metasize;
1911 	u8 nr_frags;
1912 
1913 	if (unlikely(xdp->data_end > xdp_data_hard_end(xdp))) {
1914 		pr_debug("Error building skb as missing reserved tailroom for xdp");
1915 		return NULL;
1916 	}
1917 
1918 	if (unlikely(xdp_buff_has_frags(xdp)))
1919 		nr_frags = sinfo->nr_frags;
1920 
1921 	skb = build_skb(xdp->data_hard_start, xdp->frame_sz);
1922 	if (unlikely(!skb))
1923 		return NULL;
1924 
1925 	headroom = xdp->data - xdp->data_hard_start;
1926 	data_len = xdp->data_end - xdp->data;
1927 	skb_reserve(skb, headroom);
1928 	__skb_put(skb, data_len);
1929 
1930 	metasize = xdp->data - xdp->data_meta;
1931 	metasize = metasize > 0 ? metasize : 0;
1932 	if (metasize)
1933 		skb_metadata_set(skb, metasize);
1934 
1935 	if (unlikely(xdp_buff_has_frags(xdp)))
1936 		xdp_update_skb_shared_info(skb, nr_frags,
1937 					   sinfo->xdp_frags_size,
1938 					   xdp_frags_truesz,
1939 					   xdp_buff_is_frag_pfmemalloc(xdp));
1940 
1941 	return skb;
1942 }
1943 
1944 /* TODO: build xdp in big mode */
1945 static int virtnet_build_xdp_buff_mrg(struct net_device *dev,
1946 				      struct virtnet_info *vi,
1947 				      struct receive_queue *rq,
1948 				      struct xdp_buff *xdp,
1949 				      void *buf,
1950 				      unsigned int len,
1951 				      unsigned int frame_sz,
1952 				      int *num_buf,
1953 				      unsigned int *xdp_frags_truesize,
1954 				      struct virtnet_rq_stats *stats)
1955 {
1956 	struct virtio_net_hdr_mrg_rxbuf *hdr = buf;
1957 	unsigned int headroom, tailroom, room;
1958 	unsigned int truesize, cur_frag_size;
1959 	struct skb_shared_info *shinfo;
1960 	unsigned int xdp_frags_truesz = 0;
1961 	struct page *page;
1962 	skb_frag_t *frag;
1963 	int offset;
1964 	void *ctx;
1965 
1966 	xdp_init_buff(xdp, frame_sz, &rq->xdp_rxq);
1967 	xdp_prepare_buff(xdp, buf - XDP_PACKET_HEADROOM,
1968 			 XDP_PACKET_HEADROOM + vi->hdr_len, len - vi->hdr_len, true);
1969 
1970 	if (!*num_buf)
1971 		return 0;
1972 
1973 	if (*num_buf > 1) {
1974 		/* If we want to build multi-buffer xdp, we need
1975 		 * to specify that the flags of xdp_buff have the
1976 		 * XDP_FLAGS_HAS_FRAG bit.
1977 		 */
1978 		if (!xdp_buff_has_frags(xdp))
1979 			xdp_buff_set_frags_flag(xdp);
1980 
1981 		shinfo = xdp_get_shared_info_from_buff(xdp);
1982 		shinfo->nr_frags = 0;
1983 		shinfo->xdp_frags_size = 0;
1984 	}
1985 
1986 	if (*num_buf > MAX_SKB_FRAGS + 1)
1987 		return -EINVAL;
1988 
1989 	while (--*num_buf > 0) {
1990 		buf = virtnet_rq_get_buf(rq, &len, &ctx);
1991 		if (unlikely(!buf)) {
1992 			pr_debug("%s: rx error: %d buffers out of %d missing\n",
1993 				 dev->name, *num_buf,
1994 				 virtio16_to_cpu(vi->vdev, hdr->num_buffers));
1995 			DEV_STATS_INC(dev, rx_length_errors);
1996 			goto err;
1997 		}
1998 
1999 		u64_stats_add(&stats->bytes, len);
2000 		page = virt_to_head_page(buf);
2001 		offset = buf - page_address(page);
2002 
2003 		truesize = mergeable_ctx_to_truesize(ctx);
2004 		headroom = mergeable_ctx_to_headroom(ctx);
2005 		tailroom = headroom ? sizeof(struct skb_shared_info) : 0;
2006 		room = SKB_DATA_ALIGN(headroom + tailroom);
2007 
2008 		cur_frag_size = truesize;
2009 		xdp_frags_truesz += cur_frag_size;
2010 		if (unlikely(len > truesize - room || cur_frag_size > PAGE_SIZE)) {
2011 			put_page(page);
2012 			pr_debug("%s: rx error: len %u exceeds truesize %lu\n",
2013 				 dev->name, len, (unsigned long)(truesize - room));
2014 			DEV_STATS_INC(dev, rx_length_errors);
2015 			goto err;
2016 		}
2017 
2018 		frag = &shinfo->frags[shinfo->nr_frags++];
2019 		skb_frag_fill_page_desc(frag, page, offset, len);
2020 		if (page_is_pfmemalloc(page))
2021 			xdp_buff_set_frag_pfmemalloc(xdp);
2022 
2023 		shinfo->xdp_frags_size += len;
2024 	}
2025 
2026 	*xdp_frags_truesize = xdp_frags_truesz;
2027 	return 0;
2028 
2029 err:
2030 	put_xdp_frags(xdp);
2031 	return -EINVAL;
2032 }
2033 
2034 static void *mergeable_xdp_get_buf(struct virtnet_info *vi,
2035 				   struct receive_queue *rq,
2036 				   struct bpf_prog *xdp_prog,
2037 				   void *ctx,
2038 				   unsigned int *frame_sz,
2039 				   int *num_buf,
2040 				   struct page **page,
2041 				   int offset,
2042 				   unsigned int *len,
2043 				   struct virtio_net_hdr_mrg_rxbuf *hdr)
2044 {
2045 	unsigned int truesize = mergeable_ctx_to_truesize(ctx);
2046 	unsigned int headroom = mergeable_ctx_to_headroom(ctx);
2047 	struct page *xdp_page;
2048 	unsigned int xdp_room;
2049 
2050 	/* Transient failure which in theory could occur if
2051 	 * in-flight packets from before XDP was enabled reach
2052 	 * the receive path after XDP is loaded.
2053 	 */
2054 	if (unlikely(hdr->hdr.gso_type))
2055 		return NULL;
2056 
2057 	/* Partially checksummed packets must be dropped. */
2058 	if (unlikely(hdr->hdr.flags & VIRTIO_NET_HDR_F_NEEDS_CSUM))
2059 		return NULL;
2060 
2061 	/* Now XDP core assumes frag size is PAGE_SIZE, but buffers
2062 	 * with headroom may add hole in truesize, which
2063 	 * make their length exceed PAGE_SIZE. So we disabled the
2064 	 * hole mechanism for xdp. See add_recvbuf_mergeable().
2065 	 */
2066 	*frame_sz = truesize;
2067 
2068 	if (likely(headroom >= virtnet_get_headroom(vi) &&
2069 		   (*num_buf == 1 || xdp_prog->aux->xdp_has_frags))) {
2070 		return page_address(*page) + offset;
2071 	}
2072 
2073 	/* This happens when headroom is not enough because
2074 	 * of the buffer was prefilled before XDP is set.
2075 	 * This should only happen for the first several packets.
2076 	 * In fact, vq reset can be used here to help us clean up
2077 	 * the prefilled buffers, but many existing devices do not
2078 	 * support it, and we don't want to bother users who are
2079 	 * using xdp normally.
2080 	 */
2081 	if (!xdp_prog->aux->xdp_has_frags) {
2082 		/* linearize data for XDP */
2083 		xdp_page = xdp_linearize_page(rq, num_buf,
2084 					      *page, offset,
2085 					      XDP_PACKET_HEADROOM,
2086 					      len);
2087 		if (!xdp_page)
2088 			return NULL;
2089 	} else {
2090 		xdp_room = SKB_DATA_ALIGN(XDP_PACKET_HEADROOM +
2091 					  sizeof(struct skb_shared_info));
2092 		if (*len + xdp_room > PAGE_SIZE)
2093 			return NULL;
2094 
2095 		xdp_page = alloc_page(GFP_ATOMIC);
2096 		if (!xdp_page)
2097 			return NULL;
2098 
2099 		memcpy(page_address(xdp_page) + XDP_PACKET_HEADROOM,
2100 		       page_address(*page) + offset, *len);
2101 	}
2102 
2103 	*frame_sz = PAGE_SIZE;
2104 
2105 	put_page(*page);
2106 
2107 	*page = xdp_page;
2108 
2109 	return page_address(*page) + XDP_PACKET_HEADROOM;
2110 }
2111 
2112 static struct sk_buff *receive_mergeable_xdp(struct net_device *dev,
2113 					     struct virtnet_info *vi,
2114 					     struct receive_queue *rq,
2115 					     struct bpf_prog *xdp_prog,
2116 					     void *buf,
2117 					     void *ctx,
2118 					     unsigned int len,
2119 					     unsigned int *xdp_xmit,
2120 					     struct virtnet_rq_stats *stats)
2121 {
2122 	struct virtio_net_hdr_mrg_rxbuf *hdr = buf;
2123 	int num_buf = virtio16_to_cpu(vi->vdev, hdr->num_buffers);
2124 	struct page *page = virt_to_head_page(buf);
2125 	int offset = buf - page_address(page);
2126 	unsigned int xdp_frags_truesz = 0;
2127 	struct sk_buff *head_skb;
2128 	unsigned int frame_sz;
2129 	struct xdp_buff xdp;
2130 	void *data;
2131 	u32 act;
2132 	int err;
2133 
2134 	data = mergeable_xdp_get_buf(vi, rq, xdp_prog, ctx, &frame_sz, &num_buf, &page,
2135 				     offset, &len, hdr);
2136 	if (unlikely(!data))
2137 		goto err_xdp;
2138 
2139 	err = virtnet_build_xdp_buff_mrg(dev, vi, rq, &xdp, data, len, frame_sz,
2140 					 &num_buf, &xdp_frags_truesz, stats);
2141 	if (unlikely(err))
2142 		goto err_xdp;
2143 
2144 	act = virtnet_xdp_handler(xdp_prog, &xdp, dev, xdp_xmit, stats);
2145 
2146 	switch (act) {
2147 	case XDP_PASS:
2148 		head_skb = build_skb_from_xdp_buff(dev, vi, &xdp, xdp_frags_truesz);
2149 		if (unlikely(!head_skb))
2150 			break;
2151 		return head_skb;
2152 
2153 	case XDP_TX:
2154 	case XDP_REDIRECT:
2155 		return NULL;
2156 
2157 	default:
2158 		break;
2159 	}
2160 
2161 	put_xdp_frags(&xdp);
2162 
2163 err_xdp:
2164 	put_page(page);
2165 	mergeable_buf_free(rq, num_buf, dev, stats);
2166 
2167 	u64_stats_inc(&stats->xdp_drops);
2168 	u64_stats_inc(&stats->drops);
2169 	return NULL;
2170 }
2171 
2172 static struct sk_buff *virtnet_skb_append_frag(struct sk_buff *head_skb,
2173 					       struct sk_buff *curr_skb,
2174 					       struct page *page, void *buf,
2175 					       int len, int truesize)
2176 {
2177 	int num_skb_frags;
2178 	int offset;
2179 
2180 	num_skb_frags = skb_shinfo(curr_skb)->nr_frags;
2181 	if (unlikely(num_skb_frags == MAX_SKB_FRAGS)) {
2182 		struct sk_buff *nskb = alloc_skb(0, GFP_ATOMIC);
2183 
2184 		if (unlikely(!nskb))
2185 			return NULL;
2186 
2187 		if (curr_skb == head_skb)
2188 			skb_shinfo(curr_skb)->frag_list = nskb;
2189 		else
2190 			curr_skb->next = nskb;
2191 		curr_skb = nskb;
2192 		head_skb->truesize += nskb->truesize;
2193 		num_skb_frags = 0;
2194 	}
2195 
2196 	if (curr_skb != head_skb) {
2197 		head_skb->data_len += len;
2198 		head_skb->len += len;
2199 		head_skb->truesize += truesize;
2200 	}
2201 
2202 	offset = buf - page_address(page);
2203 	if (skb_can_coalesce(curr_skb, num_skb_frags, page, offset)) {
2204 		put_page(page);
2205 		skb_coalesce_rx_frag(curr_skb, num_skb_frags - 1,
2206 				     len, truesize);
2207 	} else {
2208 		skb_add_rx_frag(curr_skb, num_skb_frags, page,
2209 				offset, len, truesize);
2210 	}
2211 
2212 	return curr_skb;
2213 }
2214 
2215 static struct sk_buff *receive_mergeable(struct net_device *dev,
2216 					 struct virtnet_info *vi,
2217 					 struct receive_queue *rq,
2218 					 void *buf,
2219 					 void *ctx,
2220 					 unsigned int len,
2221 					 unsigned int *xdp_xmit,
2222 					 struct virtnet_rq_stats *stats)
2223 {
2224 	struct virtio_net_hdr_mrg_rxbuf *hdr = buf;
2225 	int num_buf = virtio16_to_cpu(vi->vdev, hdr->num_buffers);
2226 	struct page *page = virt_to_head_page(buf);
2227 	int offset = buf - page_address(page);
2228 	struct sk_buff *head_skb, *curr_skb;
2229 	unsigned int truesize = mergeable_ctx_to_truesize(ctx);
2230 	unsigned int headroom = mergeable_ctx_to_headroom(ctx);
2231 	unsigned int tailroom = headroom ? sizeof(struct skb_shared_info) : 0;
2232 	unsigned int room = SKB_DATA_ALIGN(headroom + tailroom);
2233 
2234 	head_skb = NULL;
2235 	u64_stats_add(&stats->bytes, len - vi->hdr_len);
2236 
2237 	if (unlikely(len > truesize - room)) {
2238 		pr_debug("%s: rx error: len %u exceeds truesize %lu\n",
2239 			 dev->name, len, (unsigned long)(truesize - room));
2240 		DEV_STATS_INC(dev, rx_length_errors);
2241 		goto err_skb;
2242 	}
2243 
2244 	if (unlikely(vi->xdp_enabled)) {
2245 		struct bpf_prog *xdp_prog;
2246 
2247 		rcu_read_lock();
2248 		xdp_prog = rcu_dereference(rq->xdp_prog);
2249 		if (xdp_prog) {
2250 			head_skb = receive_mergeable_xdp(dev, vi, rq, xdp_prog, buf, ctx,
2251 							 len, xdp_xmit, stats);
2252 			rcu_read_unlock();
2253 			return head_skb;
2254 		}
2255 		rcu_read_unlock();
2256 	}
2257 
2258 	head_skb = page_to_skb(vi, rq, page, offset, len, truesize, headroom);
2259 	curr_skb = head_skb;
2260 
2261 	if (unlikely(!curr_skb))
2262 		goto err_skb;
2263 	while (--num_buf) {
2264 		buf = virtnet_rq_get_buf(rq, &len, &ctx);
2265 		if (unlikely(!buf)) {
2266 			pr_debug("%s: rx error: %d buffers out of %d missing\n",
2267 				 dev->name, num_buf,
2268 				 virtio16_to_cpu(vi->vdev,
2269 						 hdr->num_buffers));
2270 			DEV_STATS_INC(dev, rx_length_errors);
2271 			goto err_buf;
2272 		}
2273 
2274 		u64_stats_add(&stats->bytes, len);
2275 		page = virt_to_head_page(buf);
2276 
2277 		truesize = mergeable_ctx_to_truesize(ctx);
2278 		headroom = mergeable_ctx_to_headroom(ctx);
2279 		tailroom = headroom ? sizeof(struct skb_shared_info) : 0;
2280 		room = SKB_DATA_ALIGN(headroom + tailroom);
2281 		if (unlikely(len > truesize - room)) {
2282 			pr_debug("%s: rx error: len %u exceeds truesize %lu\n",
2283 				 dev->name, len, (unsigned long)(truesize - room));
2284 			DEV_STATS_INC(dev, rx_length_errors);
2285 			goto err_skb;
2286 		}
2287 
2288 		curr_skb  = virtnet_skb_append_frag(head_skb, curr_skb, page,
2289 						    buf, len, truesize);
2290 		if (!curr_skb)
2291 			goto err_skb;
2292 	}
2293 
2294 	ewma_pkt_len_add(&rq->mrg_avg_pkt_len, head_skb->len);
2295 	return head_skb;
2296 
2297 err_skb:
2298 	put_page(page);
2299 	mergeable_buf_free(rq, num_buf, dev, stats);
2300 
2301 err_buf:
2302 	u64_stats_inc(&stats->drops);
2303 	dev_kfree_skb(head_skb);
2304 	return NULL;
2305 }
2306 
2307 static void virtio_skb_set_hash(const struct virtio_net_hdr_v1_hash *hdr_hash,
2308 				struct sk_buff *skb)
2309 {
2310 	enum pkt_hash_types rss_hash_type;
2311 
2312 	if (!hdr_hash || !skb)
2313 		return;
2314 
2315 	switch (__le16_to_cpu(hdr_hash->hash_report)) {
2316 	case VIRTIO_NET_HASH_REPORT_TCPv4:
2317 	case VIRTIO_NET_HASH_REPORT_UDPv4:
2318 	case VIRTIO_NET_HASH_REPORT_TCPv6:
2319 	case VIRTIO_NET_HASH_REPORT_UDPv6:
2320 	case VIRTIO_NET_HASH_REPORT_TCPv6_EX:
2321 	case VIRTIO_NET_HASH_REPORT_UDPv6_EX:
2322 		rss_hash_type = PKT_HASH_TYPE_L4;
2323 		break;
2324 	case VIRTIO_NET_HASH_REPORT_IPv4:
2325 	case VIRTIO_NET_HASH_REPORT_IPv6:
2326 	case VIRTIO_NET_HASH_REPORT_IPv6_EX:
2327 		rss_hash_type = PKT_HASH_TYPE_L3;
2328 		break;
2329 	case VIRTIO_NET_HASH_REPORT_NONE:
2330 	default:
2331 		rss_hash_type = PKT_HASH_TYPE_NONE;
2332 	}
2333 	skb_set_hash(skb, __le32_to_cpu(hdr_hash->hash_value), rss_hash_type);
2334 }
2335 
2336 static void virtnet_receive_done(struct virtnet_info *vi, struct receive_queue *rq,
2337 				 struct sk_buff *skb, u8 flags)
2338 {
2339 	struct virtio_net_common_hdr *hdr;
2340 	struct net_device *dev = vi->dev;
2341 
2342 	hdr = skb_vnet_common_hdr(skb);
2343 	if (dev->features & NETIF_F_RXHASH && vi->has_rss_hash_report)
2344 		virtio_skb_set_hash(&hdr->hash_v1_hdr, skb);
2345 
2346 	if (flags & VIRTIO_NET_HDR_F_DATA_VALID)
2347 		skb->ip_summed = CHECKSUM_UNNECESSARY;
2348 
2349 	if (virtio_net_hdr_to_skb(skb, &hdr->hdr,
2350 				  virtio_is_little_endian(vi->vdev))) {
2351 		net_warn_ratelimited("%s: bad gso: type: %u, size: %u\n",
2352 				     dev->name, hdr->hdr.gso_type,
2353 				     hdr->hdr.gso_size);
2354 		goto frame_err;
2355 	}
2356 
2357 	skb_record_rx_queue(skb, vq2rxq(rq->vq));
2358 	skb->protocol = eth_type_trans(skb, dev);
2359 	pr_debug("Receiving skb proto 0x%04x len %i type %i\n",
2360 		 ntohs(skb->protocol), skb->len, skb->pkt_type);
2361 
2362 	napi_gro_receive(&rq->napi, skb);
2363 	return;
2364 
2365 frame_err:
2366 	DEV_STATS_INC(dev, rx_frame_errors);
2367 	dev_kfree_skb(skb);
2368 }
2369 
2370 static void receive_buf(struct virtnet_info *vi, struct receive_queue *rq,
2371 			void *buf, unsigned int len, void **ctx,
2372 			unsigned int *xdp_xmit,
2373 			struct virtnet_rq_stats *stats)
2374 {
2375 	struct net_device *dev = vi->dev;
2376 	struct sk_buff *skb;
2377 	u8 flags;
2378 
2379 	if (unlikely(len < vi->hdr_len + ETH_HLEN)) {
2380 		pr_debug("%s: short packet %i\n", dev->name, len);
2381 		DEV_STATS_INC(dev, rx_length_errors);
2382 		virtnet_rq_free_buf(vi, rq, buf);
2383 		return;
2384 	}
2385 
2386 	/* 1. Save the flags early, as the XDP program might overwrite them.
2387 	 * These flags ensure packets marked as VIRTIO_NET_HDR_F_DATA_VALID
2388 	 * stay valid after XDP processing.
2389 	 * 2. XDP doesn't work with partially checksummed packets (refer to
2390 	 * virtnet_xdp_set()), so packets marked as
2391 	 * VIRTIO_NET_HDR_F_NEEDS_CSUM get dropped during XDP processing.
2392 	 */
2393 	flags = ((struct virtio_net_common_hdr *)buf)->hdr.flags;
2394 
2395 	if (vi->mergeable_rx_bufs)
2396 		skb = receive_mergeable(dev, vi, rq, buf, ctx, len, xdp_xmit,
2397 					stats);
2398 	else if (vi->big_packets)
2399 		skb = receive_big(dev, vi, rq, buf, len, stats);
2400 	else
2401 		skb = receive_small(dev, vi, rq, buf, ctx, len, xdp_xmit, stats);
2402 
2403 	if (unlikely(!skb))
2404 		return;
2405 
2406 	virtnet_receive_done(vi, rq, skb, flags);
2407 }
2408 
2409 /* Unlike mergeable buffers, all buffers are allocated to the
2410  * same size, except for the headroom. For this reason we do
2411  * not need to use  mergeable_len_to_ctx here - it is enough
2412  * to store the headroom as the context ignoring the truesize.
2413  */
2414 static int add_recvbuf_small(struct virtnet_info *vi, struct receive_queue *rq,
2415 			     gfp_t gfp)
2416 {
2417 	char *buf;
2418 	unsigned int xdp_headroom = virtnet_get_headroom(vi);
2419 	void *ctx = (void *)(unsigned long)xdp_headroom;
2420 	int len = vi->hdr_len + VIRTNET_RX_PAD + GOOD_PACKET_LEN + xdp_headroom;
2421 	int err;
2422 
2423 	len = SKB_DATA_ALIGN(len) +
2424 	      SKB_DATA_ALIGN(sizeof(struct skb_shared_info));
2425 
2426 	buf = virtnet_rq_alloc(rq, len, gfp);
2427 	if (unlikely(!buf))
2428 		return -ENOMEM;
2429 
2430 	buf += VIRTNET_RX_PAD + xdp_headroom;
2431 
2432 	virtnet_rq_init_one_sg(rq, buf, vi->hdr_len + GOOD_PACKET_LEN);
2433 
2434 	err = virtqueue_add_inbuf_ctx(rq->vq, rq->sg, 1, buf, ctx, gfp);
2435 	if (err < 0) {
2436 		if (rq->do_dma)
2437 			virtnet_rq_unmap(rq, buf, 0);
2438 		put_page(virt_to_head_page(buf));
2439 	}
2440 
2441 	return err;
2442 }
2443 
2444 static int add_recvbuf_big(struct virtnet_info *vi, struct receive_queue *rq,
2445 			   gfp_t gfp)
2446 {
2447 	struct page *first, *list = NULL;
2448 	char *p;
2449 	int i, err, offset;
2450 
2451 	sg_init_table(rq->sg, vi->big_packets_num_skbfrags + 2);
2452 
2453 	/* page in rq->sg[vi->big_packets_num_skbfrags + 1] is list tail */
2454 	for (i = vi->big_packets_num_skbfrags + 1; i > 1; --i) {
2455 		first = get_a_page(rq, gfp);
2456 		if (!first) {
2457 			if (list)
2458 				give_pages(rq, list);
2459 			return -ENOMEM;
2460 		}
2461 		sg_set_buf(&rq->sg[i], page_address(first), PAGE_SIZE);
2462 
2463 		/* chain new page in list head to match sg */
2464 		first->private = (unsigned long)list;
2465 		list = first;
2466 	}
2467 
2468 	first = get_a_page(rq, gfp);
2469 	if (!first) {
2470 		give_pages(rq, list);
2471 		return -ENOMEM;
2472 	}
2473 	p = page_address(first);
2474 
2475 	/* rq->sg[0], rq->sg[1] share the same page */
2476 	/* a separated rq->sg[0] for header - required in case !any_header_sg */
2477 	sg_set_buf(&rq->sg[0], p, vi->hdr_len);
2478 
2479 	/* rq->sg[1] for data packet, from offset */
2480 	offset = sizeof(struct padded_vnet_hdr);
2481 	sg_set_buf(&rq->sg[1], p + offset, PAGE_SIZE - offset);
2482 
2483 	/* chain first in list head */
2484 	first->private = (unsigned long)list;
2485 	err = virtqueue_add_inbuf(rq->vq, rq->sg, vi->big_packets_num_skbfrags + 2,
2486 				  first, gfp);
2487 	if (err < 0)
2488 		give_pages(rq, first);
2489 
2490 	return err;
2491 }
2492 
2493 static unsigned int get_mergeable_buf_len(struct receive_queue *rq,
2494 					  struct ewma_pkt_len *avg_pkt_len,
2495 					  unsigned int room)
2496 {
2497 	struct virtnet_info *vi = rq->vq->vdev->priv;
2498 	const size_t hdr_len = vi->hdr_len;
2499 	unsigned int len;
2500 
2501 	if (room)
2502 		return PAGE_SIZE - room;
2503 
2504 	len = hdr_len +	clamp_t(unsigned int, ewma_pkt_len_read(avg_pkt_len),
2505 				rq->min_buf_len, PAGE_SIZE - hdr_len);
2506 
2507 	return ALIGN(len, L1_CACHE_BYTES);
2508 }
2509 
2510 static int add_recvbuf_mergeable(struct virtnet_info *vi,
2511 				 struct receive_queue *rq, gfp_t gfp)
2512 {
2513 	struct page_frag *alloc_frag = &rq->alloc_frag;
2514 	unsigned int headroom = virtnet_get_headroom(vi);
2515 	unsigned int tailroom = headroom ? sizeof(struct skb_shared_info) : 0;
2516 	unsigned int room = SKB_DATA_ALIGN(headroom + tailroom);
2517 	unsigned int len, hole;
2518 	void *ctx;
2519 	char *buf;
2520 	int err;
2521 
2522 	/* Extra tailroom is needed to satisfy XDP's assumption. This
2523 	 * means rx frags coalescing won't work, but consider we've
2524 	 * disabled GSO for XDP, it won't be a big issue.
2525 	 */
2526 	len = get_mergeable_buf_len(rq, &rq->mrg_avg_pkt_len, room);
2527 
2528 	buf = virtnet_rq_alloc(rq, len + room, gfp);
2529 	if (unlikely(!buf))
2530 		return -ENOMEM;
2531 
2532 	buf += headroom; /* advance address leaving hole at front of pkt */
2533 	hole = alloc_frag->size - alloc_frag->offset;
2534 	if (hole < len + room) {
2535 		/* To avoid internal fragmentation, if there is very likely not
2536 		 * enough space for another buffer, add the remaining space to
2537 		 * the current buffer.
2538 		 * XDP core assumes that frame_size of xdp_buff and the length
2539 		 * of the frag are PAGE_SIZE, so we disable the hole mechanism.
2540 		 */
2541 		if (!headroom)
2542 			len += hole;
2543 		alloc_frag->offset += hole;
2544 	}
2545 
2546 	virtnet_rq_init_one_sg(rq, buf, len);
2547 
2548 	ctx = mergeable_len_to_ctx(len + room, headroom);
2549 	err = virtqueue_add_inbuf_ctx(rq->vq, rq->sg, 1, buf, ctx, gfp);
2550 	if (err < 0) {
2551 		if (rq->do_dma)
2552 			virtnet_rq_unmap(rq, buf, 0);
2553 		put_page(virt_to_head_page(buf));
2554 	}
2555 
2556 	return err;
2557 }
2558 
2559 /*
2560  * Returns false if we couldn't fill entirely (OOM).
2561  *
2562  * Normally run in the receive path, but can also be run from ndo_open
2563  * before we're receiving packets, or from refill_work which is
2564  * careful to disable receiving (using napi_disable).
2565  */
2566 static bool try_fill_recv(struct virtnet_info *vi, struct receive_queue *rq,
2567 			  gfp_t gfp)
2568 {
2569 	int err;
2570 
2571 	if (rq->xsk_pool) {
2572 		err = virtnet_add_recvbuf_xsk(vi, rq, rq->xsk_pool, gfp);
2573 		goto kick;
2574 	}
2575 
2576 	do {
2577 		if (vi->mergeable_rx_bufs)
2578 			err = add_recvbuf_mergeable(vi, rq, gfp);
2579 		else if (vi->big_packets)
2580 			err = add_recvbuf_big(vi, rq, gfp);
2581 		else
2582 			err = add_recvbuf_small(vi, rq, gfp);
2583 
2584 		if (err)
2585 			break;
2586 	} while (rq->vq->num_free);
2587 
2588 kick:
2589 	if (virtqueue_kick_prepare(rq->vq) && virtqueue_notify(rq->vq)) {
2590 		unsigned long flags;
2591 
2592 		flags = u64_stats_update_begin_irqsave(&rq->stats.syncp);
2593 		u64_stats_inc(&rq->stats.kicks);
2594 		u64_stats_update_end_irqrestore(&rq->stats.syncp, flags);
2595 	}
2596 
2597 	return err != -ENOMEM;
2598 }
2599 
2600 static void skb_recv_done(struct virtqueue *rvq)
2601 {
2602 	struct virtnet_info *vi = rvq->vdev->priv;
2603 	struct receive_queue *rq = &vi->rq[vq2rxq(rvq)];
2604 
2605 	rq->calls++;
2606 	virtqueue_napi_schedule(&rq->napi, rvq);
2607 }
2608 
2609 static void virtnet_napi_enable(struct virtqueue *vq, struct napi_struct *napi)
2610 {
2611 	napi_enable(napi);
2612 
2613 	/* If all buffers were filled by other side before we napi_enabled, we
2614 	 * won't get another interrupt, so process any outstanding packets now.
2615 	 * Call local_bh_enable after to trigger softIRQ processing.
2616 	 */
2617 	local_bh_disable();
2618 	virtqueue_napi_schedule(napi, vq);
2619 	local_bh_enable();
2620 }
2621 
2622 static void virtnet_napi_tx_enable(struct virtnet_info *vi,
2623 				   struct virtqueue *vq,
2624 				   struct napi_struct *napi)
2625 {
2626 	if (!napi->weight)
2627 		return;
2628 
2629 	/* Tx napi touches cachelines on the cpu handling tx interrupts. Only
2630 	 * enable the feature if this is likely affine with the transmit path.
2631 	 */
2632 	if (!vi->affinity_hint_set) {
2633 		napi->weight = 0;
2634 		return;
2635 	}
2636 
2637 	return virtnet_napi_enable(vq, napi);
2638 }
2639 
2640 static void virtnet_napi_tx_disable(struct napi_struct *napi)
2641 {
2642 	if (napi->weight)
2643 		napi_disable(napi);
2644 }
2645 
2646 static void refill_work(struct work_struct *work)
2647 {
2648 	struct virtnet_info *vi =
2649 		container_of(work, struct virtnet_info, refill.work);
2650 	bool still_empty;
2651 	int i;
2652 
2653 	for (i = 0; i < vi->curr_queue_pairs; i++) {
2654 		struct receive_queue *rq = &vi->rq[i];
2655 
2656 		napi_disable(&rq->napi);
2657 		still_empty = !try_fill_recv(vi, rq, GFP_KERNEL);
2658 		virtnet_napi_enable(rq->vq, &rq->napi);
2659 
2660 		/* In theory, this can happen: if we don't get any buffers in
2661 		 * we will *never* try to fill again.
2662 		 */
2663 		if (still_empty)
2664 			schedule_delayed_work(&vi->refill, HZ/2);
2665 	}
2666 }
2667 
2668 static int virtnet_receive_xsk_bufs(struct virtnet_info *vi,
2669 				    struct receive_queue *rq,
2670 				    int budget,
2671 				    unsigned int *xdp_xmit,
2672 				    struct virtnet_rq_stats *stats)
2673 {
2674 	unsigned int len;
2675 	int packets = 0;
2676 	void *buf;
2677 
2678 	while (packets < budget) {
2679 		buf = virtqueue_get_buf(rq->vq, &len);
2680 		if (!buf)
2681 			break;
2682 
2683 		virtnet_receive_xsk_buf(vi, rq, buf, len, xdp_xmit, stats);
2684 		packets++;
2685 	}
2686 
2687 	return packets;
2688 }
2689 
2690 static int virtnet_receive_packets(struct virtnet_info *vi,
2691 				   struct receive_queue *rq,
2692 				   int budget,
2693 				   unsigned int *xdp_xmit,
2694 				   struct virtnet_rq_stats *stats)
2695 {
2696 	unsigned int len;
2697 	int packets = 0;
2698 	void *buf;
2699 
2700 	if (!vi->big_packets || vi->mergeable_rx_bufs) {
2701 		void *ctx;
2702 		while (packets < budget &&
2703 		       (buf = virtnet_rq_get_buf(rq, &len, &ctx))) {
2704 			receive_buf(vi, rq, buf, len, ctx, xdp_xmit, stats);
2705 			packets++;
2706 		}
2707 	} else {
2708 		while (packets < budget &&
2709 		       (buf = virtnet_rq_get_buf(rq, &len, NULL)) != NULL) {
2710 			receive_buf(vi, rq, buf, len, NULL, xdp_xmit, stats);
2711 			packets++;
2712 		}
2713 	}
2714 
2715 	return packets;
2716 }
2717 
2718 static int virtnet_receive(struct receive_queue *rq, int budget,
2719 			   unsigned int *xdp_xmit)
2720 {
2721 	struct virtnet_info *vi = rq->vq->vdev->priv;
2722 	struct virtnet_rq_stats stats = {};
2723 	int i, packets;
2724 
2725 	if (rq->xsk_pool)
2726 		packets = virtnet_receive_xsk_bufs(vi, rq, budget, xdp_xmit, &stats);
2727 	else
2728 		packets = virtnet_receive_packets(vi, rq, budget, xdp_xmit, &stats);
2729 
2730 	if (rq->vq->num_free > min((unsigned int)budget, virtqueue_get_vring_size(rq->vq)) / 2) {
2731 		if (!try_fill_recv(vi, rq, GFP_ATOMIC)) {
2732 			spin_lock(&vi->refill_lock);
2733 			if (vi->refill_enabled)
2734 				schedule_delayed_work(&vi->refill, 0);
2735 			spin_unlock(&vi->refill_lock);
2736 		}
2737 	}
2738 
2739 	u64_stats_set(&stats.packets, packets);
2740 	u64_stats_update_begin(&rq->stats.syncp);
2741 	for (i = 0; i < ARRAY_SIZE(virtnet_rq_stats_desc); i++) {
2742 		size_t offset = virtnet_rq_stats_desc[i].offset;
2743 		u64_stats_t *item, *src;
2744 
2745 		item = (u64_stats_t *)((u8 *)&rq->stats + offset);
2746 		src = (u64_stats_t *)((u8 *)&stats + offset);
2747 		u64_stats_add(item, u64_stats_read(src));
2748 	}
2749 
2750 	u64_stats_add(&rq->stats.packets, u64_stats_read(&stats.packets));
2751 	u64_stats_add(&rq->stats.bytes, u64_stats_read(&stats.bytes));
2752 
2753 	u64_stats_update_end(&rq->stats.syncp);
2754 
2755 	return packets;
2756 }
2757 
2758 static void virtnet_poll_cleantx(struct receive_queue *rq, int budget)
2759 {
2760 	struct virtnet_info *vi = rq->vq->vdev->priv;
2761 	unsigned int index = vq2rxq(rq->vq);
2762 	struct send_queue *sq = &vi->sq[index];
2763 	struct netdev_queue *txq = netdev_get_tx_queue(vi->dev, index);
2764 
2765 	if (!sq->napi.weight || is_xdp_raw_buffer_queue(vi, index))
2766 		return;
2767 
2768 	if (__netif_tx_trylock(txq)) {
2769 		if (sq->reset) {
2770 			__netif_tx_unlock(txq);
2771 			return;
2772 		}
2773 
2774 		do {
2775 			virtqueue_disable_cb(sq->vq);
2776 			free_old_xmit(sq, txq, !!budget);
2777 		} while (unlikely(!virtqueue_enable_cb_delayed(sq->vq)));
2778 
2779 		if (sq->vq->num_free >= 2 + MAX_SKB_FRAGS) {
2780 			if (netif_tx_queue_stopped(txq)) {
2781 				u64_stats_update_begin(&sq->stats.syncp);
2782 				u64_stats_inc(&sq->stats.wake);
2783 				u64_stats_update_end(&sq->stats.syncp);
2784 			}
2785 			netif_tx_wake_queue(txq);
2786 		}
2787 
2788 		__netif_tx_unlock(txq);
2789 	}
2790 }
2791 
2792 static void virtnet_rx_dim_update(struct virtnet_info *vi, struct receive_queue *rq)
2793 {
2794 	struct dim_sample cur_sample = {};
2795 
2796 	if (!rq->packets_in_napi)
2797 		return;
2798 
2799 	/* Don't need protection when fetching stats, since fetcher and
2800 	 * updater of the stats are in same context
2801 	 */
2802 	dim_update_sample(rq->calls,
2803 			  u64_stats_read(&rq->stats.packets),
2804 			  u64_stats_read(&rq->stats.bytes),
2805 			  &cur_sample);
2806 
2807 	net_dim(&rq->dim, cur_sample);
2808 	rq->packets_in_napi = 0;
2809 }
2810 
2811 static int virtnet_poll(struct napi_struct *napi, int budget)
2812 {
2813 	struct receive_queue *rq =
2814 		container_of(napi, struct receive_queue, napi);
2815 	struct virtnet_info *vi = rq->vq->vdev->priv;
2816 	struct send_queue *sq;
2817 	unsigned int received;
2818 	unsigned int xdp_xmit = 0;
2819 	bool napi_complete;
2820 
2821 	virtnet_poll_cleantx(rq, budget);
2822 
2823 	received = virtnet_receive(rq, budget, &xdp_xmit);
2824 	rq->packets_in_napi += received;
2825 
2826 	if (xdp_xmit & VIRTIO_XDP_REDIR)
2827 		xdp_do_flush();
2828 
2829 	/* Out of packets? */
2830 	if (received < budget) {
2831 		napi_complete = virtqueue_napi_complete(napi, rq->vq, received);
2832 		/* Intentionally not taking dim_lock here. This may result in a
2833 		 * spurious net_dim call. But if that happens virtnet_rx_dim_work
2834 		 * will not act on the scheduled work.
2835 		 */
2836 		if (napi_complete && rq->dim_enabled)
2837 			virtnet_rx_dim_update(vi, rq);
2838 	}
2839 
2840 	if (xdp_xmit & VIRTIO_XDP_TX) {
2841 		sq = virtnet_xdp_get_sq(vi);
2842 		if (virtqueue_kick_prepare(sq->vq) && virtqueue_notify(sq->vq)) {
2843 			u64_stats_update_begin(&sq->stats.syncp);
2844 			u64_stats_inc(&sq->stats.kicks);
2845 			u64_stats_update_end(&sq->stats.syncp);
2846 		}
2847 		virtnet_xdp_put_sq(vi, sq);
2848 	}
2849 
2850 	return received;
2851 }
2852 
2853 static void virtnet_disable_queue_pair(struct virtnet_info *vi, int qp_index)
2854 {
2855 	virtnet_napi_tx_disable(&vi->sq[qp_index].napi);
2856 	napi_disable(&vi->rq[qp_index].napi);
2857 	xdp_rxq_info_unreg(&vi->rq[qp_index].xdp_rxq);
2858 }
2859 
2860 static int virtnet_enable_queue_pair(struct virtnet_info *vi, int qp_index)
2861 {
2862 	struct net_device *dev = vi->dev;
2863 	int err;
2864 
2865 	err = xdp_rxq_info_reg(&vi->rq[qp_index].xdp_rxq, dev, qp_index,
2866 			       vi->rq[qp_index].napi.napi_id);
2867 	if (err < 0)
2868 		return err;
2869 
2870 	err = xdp_rxq_info_reg_mem_model(&vi->rq[qp_index].xdp_rxq,
2871 					 MEM_TYPE_PAGE_SHARED, NULL);
2872 	if (err < 0)
2873 		goto err_xdp_reg_mem_model;
2874 
2875 	netdev_tx_reset_queue(netdev_get_tx_queue(vi->dev, qp_index));
2876 	virtnet_napi_enable(vi->rq[qp_index].vq, &vi->rq[qp_index].napi);
2877 	virtnet_napi_tx_enable(vi, vi->sq[qp_index].vq, &vi->sq[qp_index].napi);
2878 
2879 	return 0;
2880 
2881 err_xdp_reg_mem_model:
2882 	xdp_rxq_info_unreg(&vi->rq[qp_index].xdp_rxq);
2883 	return err;
2884 }
2885 
2886 static void virtnet_cancel_dim(struct virtnet_info *vi, struct dim *dim)
2887 {
2888 	if (!virtio_has_feature(vi->vdev, VIRTIO_NET_F_VQ_NOTF_COAL))
2889 		return;
2890 	net_dim_work_cancel(dim);
2891 }
2892 
2893 static void virtnet_update_settings(struct virtnet_info *vi)
2894 {
2895 	u32 speed;
2896 	u8 duplex;
2897 
2898 	if (!virtio_has_feature(vi->vdev, VIRTIO_NET_F_SPEED_DUPLEX))
2899 		return;
2900 
2901 	virtio_cread_le(vi->vdev, struct virtio_net_config, speed, &speed);
2902 
2903 	if (ethtool_validate_speed(speed))
2904 		vi->speed = speed;
2905 
2906 	virtio_cread_le(vi->vdev, struct virtio_net_config, duplex, &duplex);
2907 
2908 	if (ethtool_validate_duplex(duplex))
2909 		vi->duplex = duplex;
2910 }
2911 
2912 static int virtnet_open(struct net_device *dev)
2913 {
2914 	struct virtnet_info *vi = netdev_priv(dev);
2915 	int i, err;
2916 
2917 	enable_delayed_refill(vi);
2918 
2919 	for (i = 0; i < vi->max_queue_pairs; i++) {
2920 		if (i < vi->curr_queue_pairs)
2921 			/* Make sure we have some buffers: if oom use wq. */
2922 			if (!try_fill_recv(vi, &vi->rq[i], GFP_KERNEL))
2923 				schedule_delayed_work(&vi->refill, 0);
2924 
2925 		err = virtnet_enable_queue_pair(vi, i);
2926 		if (err < 0)
2927 			goto err_enable_qp;
2928 	}
2929 
2930 	if (virtio_has_feature(vi->vdev, VIRTIO_NET_F_STATUS)) {
2931 		if (vi->status & VIRTIO_NET_S_LINK_UP)
2932 			netif_carrier_on(vi->dev);
2933 		virtio_config_driver_enable(vi->vdev);
2934 	} else {
2935 		vi->status = VIRTIO_NET_S_LINK_UP;
2936 		netif_carrier_on(dev);
2937 	}
2938 
2939 	return 0;
2940 
2941 err_enable_qp:
2942 	disable_delayed_refill(vi);
2943 	cancel_delayed_work_sync(&vi->refill);
2944 
2945 	for (i--; i >= 0; i--) {
2946 		virtnet_disable_queue_pair(vi, i);
2947 		virtnet_cancel_dim(vi, &vi->rq[i].dim);
2948 	}
2949 
2950 	return err;
2951 }
2952 
2953 static int virtnet_poll_tx(struct napi_struct *napi, int budget)
2954 {
2955 	struct send_queue *sq = container_of(napi, struct send_queue, napi);
2956 	struct virtnet_info *vi = sq->vq->vdev->priv;
2957 	unsigned int index = vq2txq(sq->vq);
2958 	struct netdev_queue *txq;
2959 	int opaque;
2960 	bool done;
2961 
2962 	if (unlikely(is_xdp_raw_buffer_queue(vi, index))) {
2963 		/* We don't need to enable cb for XDP */
2964 		napi_complete_done(napi, 0);
2965 		return 0;
2966 	}
2967 
2968 	txq = netdev_get_tx_queue(vi->dev, index);
2969 	__netif_tx_lock(txq, raw_smp_processor_id());
2970 	virtqueue_disable_cb(sq->vq);
2971 	free_old_xmit(sq, txq, !!budget);
2972 
2973 	if (sq->vq->num_free >= 2 + MAX_SKB_FRAGS) {
2974 		if (netif_tx_queue_stopped(txq)) {
2975 			u64_stats_update_begin(&sq->stats.syncp);
2976 			u64_stats_inc(&sq->stats.wake);
2977 			u64_stats_update_end(&sq->stats.syncp);
2978 		}
2979 		netif_tx_wake_queue(txq);
2980 	}
2981 
2982 	opaque = virtqueue_enable_cb_prepare(sq->vq);
2983 
2984 	done = napi_complete_done(napi, 0);
2985 
2986 	if (!done)
2987 		virtqueue_disable_cb(sq->vq);
2988 
2989 	__netif_tx_unlock(txq);
2990 
2991 	if (done) {
2992 		if (unlikely(virtqueue_poll(sq->vq, opaque))) {
2993 			if (napi_schedule_prep(napi)) {
2994 				__netif_tx_lock(txq, raw_smp_processor_id());
2995 				virtqueue_disable_cb(sq->vq);
2996 				__netif_tx_unlock(txq);
2997 				__napi_schedule(napi);
2998 			}
2999 		}
3000 	}
3001 
3002 	return 0;
3003 }
3004 
3005 static int xmit_skb(struct send_queue *sq, struct sk_buff *skb, bool orphan)
3006 {
3007 	struct virtio_net_hdr_mrg_rxbuf *hdr;
3008 	const unsigned char *dest = ((struct ethhdr *)skb->data)->h_dest;
3009 	struct virtnet_info *vi = sq->vq->vdev->priv;
3010 	int num_sg;
3011 	unsigned hdr_len = vi->hdr_len;
3012 	bool can_push;
3013 
3014 	pr_debug("%s: xmit %p %pM\n", vi->dev->name, skb, dest);
3015 
3016 	can_push = vi->any_header_sg &&
3017 		!((unsigned long)skb->data & (__alignof__(*hdr) - 1)) &&
3018 		!skb_header_cloned(skb) && skb_headroom(skb) >= hdr_len;
3019 	/* Even if we can, don't push here yet as this would skew
3020 	 * csum_start offset below. */
3021 	if (can_push)
3022 		hdr = (struct virtio_net_hdr_mrg_rxbuf *)(skb->data - hdr_len);
3023 	else
3024 		hdr = &skb_vnet_common_hdr(skb)->mrg_hdr;
3025 
3026 	if (virtio_net_hdr_from_skb(skb, &hdr->hdr,
3027 				    virtio_is_little_endian(vi->vdev), false,
3028 				    0))
3029 		return -EPROTO;
3030 
3031 	if (vi->mergeable_rx_bufs)
3032 		hdr->num_buffers = 0;
3033 
3034 	sg_init_table(sq->sg, skb_shinfo(skb)->nr_frags + (can_push ? 1 : 2));
3035 	if (can_push) {
3036 		__skb_push(skb, hdr_len);
3037 		num_sg = skb_to_sgvec(skb, sq->sg, 0, skb->len);
3038 		if (unlikely(num_sg < 0))
3039 			return num_sg;
3040 		/* Pull header back to avoid skew in tx bytes calculations. */
3041 		__skb_pull(skb, hdr_len);
3042 	} else {
3043 		sg_set_buf(sq->sg, hdr, hdr_len);
3044 		num_sg = skb_to_sgvec(skb, sq->sg + 1, 0, skb->len);
3045 		if (unlikely(num_sg < 0))
3046 			return num_sg;
3047 		num_sg++;
3048 	}
3049 	return virtqueue_add_outbuf(sq->vq, sq->sg, num_sg,
3050 				    skb_to_ptr(skb, orphan), GFP_ATOMIC);
3051 }
3052 
3053 static netdev_tx_t start_xmit(struct sk_buff *skb, struct net_device *dev)
3054 {
3055 	struct virtnet_info *vi = netdev_priv(dev);
3056 	int qnum = skb_get_queue_mapping(skb);
3057 	struct send_queue *sq = &vi->sq[qnum];
3058 	int err;
3059 	struct netdev_queue *txq = netdev_get_tx_queue(dev, qnum);
3060 	bool xmit_more = netdev_xmit_more();
3061 	bool use_napi = sq->napi.weight;
3062 	bool kick;
3063 
3064 	/* Free up any pending old buffers before queueing new ones. */
3065 	do {
3066 		if (use_napi)
3067 			virtqueue_disable_cb(sq->vq);
3068 
3069 		free_old_xmit(sq, txq, false);
3070 
3071 	} while (use_napi && !xmit_more &&
3072 	       unlikely(!virtqueue_enable_cb_delayed(sq->vq)));
3073 
3074 	/* timestamp packet in software */
3075 	skb_tx_timestamp(skb);
3076 
3077 	/* Try to transmit */
3078 	err = xmit_skb(sq, skb, !use_napi);
3079 
3080 	/* This should not happen! */
3081 	if (unlikely(err)) {
3082 		DEV_STATS_INC(dev, tx_fifo_errors);
3083 		if (net_ratelimit())
3084 			dev_warn(&dev->dev,
3085 				 "Unexpected TXQ (%d) queue failure: %d\n",
3086 				 qnum, err);
3087 		DEV_STATS_INC(dev, tx_dropped);
3088 		dev_kfree_skb_any(skb);
3089 		return NETDEV_TX_OK;
3090 	}
3091 
3092 	/* Don't wait up for transmitted skbs to be freed. */
3093 	if (!use_napi) {
3094 		skb_orphan(skb);
3095 		nf_reset_ct(skb);
3096 	}
3097 
3098 	check_sq_full_and_disable(vi, dev, sq);
3099 
3100 	kick = use_napi ? __netdev_tx_sent_queue(txq, skb->len, xmit_more) :
3101 			  !xmit_more || netif_xmit_stopped(txq);
3102 	if (kick) {
3103 		if (virtqueue_kick_prepare(sq->vq) && virtqueue_notify(sq->vq)) {
3104 			u64_stats_update_begin(&sq->stats.syncp);
3105 			u64_stats_inc(&sq->stats.kicks);
3106 			u64_stats_update_end(&sq->stats.syncp);
3107 		}
3108 	}
3109 
3110 	return NETDEV_TX_OK;
3111 }
3112 
3113 static void virtnet_rx_pause(struct virtnet_info *vi, struct receive_queue *rq)
3114 {
3115 	bool running = netif_running(vi->dev);
3116 
3117 	if (running) {
3118 		napi_disable(&rq->napi);
3119 		virtnet_cancel_dim(vi, &rq->dim);
3120 	}
3121 }
3122 
3123 static void virtnet_rx_resume(struct virtnet_info *vi, struct receive_queue *rq)
3124 {
3125 	bool running = netif_running(vi->dev);
3126 
3127 	if (!try_fill_recv(vi, rq, GFP_KERNEL))
3128 		schedule_delayed_work(&vi->refill, 0);
3129 
3130 	if (running)
3131 		virtnet_napi_enable(rq->vq, &rq->napi);
3132 }
3133 
3134 static int virtnet_rx_resize(struct virtnet_info *vi,
3135 			     struct receive_queue *rq, u32 ring_num)
3136 {
3137 	int err, qindex;
3138 
3139 	qindex = rq - vi->rq;
3140 
3141 	virtnet_rx_pause(vi, rq);
3142 
3143 	err = virtqueue_resize(rq->vq, ring_num, virtnet_rq_unmap_free_buf);
3144 	if (err)
3145 		netdev_err(vi->dev, "resize rx fail: rx queue index: %d err: %d\n", qindex, err);
3146 
3147 	virtnet_rx_resume(vi, rq);
3148 	return err;
3149 }
3150 
3151 static void virtnet_tx_pause(struct virtnet_info *vi, struct send_queue *sq)
3152 {
3153 	bool running = netif_running(vi->dev);
3154 	struct netdev_queue *txq;
3155 	int qindex;
3156 
3157 	qindex = sq - vi->sq;
3158 
3159 	if (running)
3160 		virtnet_napi_tx_disable(&sq->napi);
3161 
3162 	txq = netdev_get_tx_queue(vi->dev, qindex);
3163 
3164 	/* 1. wait all ximt complete
3165 	 * 2. fix the race of netif_stop_subqueue() vs netif_start_subqueue()
3166 	 */
3167 	__netif_tx_lock_bh(txq);
3168 
3169 	/* Prevent rx poll from accessing sq. */
3170 	sq->reset = true;
3171 
3172 	/* Prevent the upper layer from trying to send packets. */
3173 	netif_stop_subqueue(vi->dev, qindex);
3174 
3175 	__netif_tx_unlock_bh(txq);
3176 }
3177 
3178 static void virtnet_tx_resume(struct virtnet_info *vi, struct send_queue *sq)
3179 {
3180 	bool running = netif_running(vi->dev);
3181 	struct netdev_queue *txq;
3182 	int qindex;
3183 
3184 	qindex = sq - vi->sq;
3185 
3186 	txq = netdev_get_tx_queue(vi->dev, qindex);
3187 
3188 	__netif_tx_lock_bh(txq);
3189 	sq->reset = false;
3190 	netif_tx_wake_queue(txq);
3191 	__netif_tx_unlock_bh(txq);
3192 
3193 	if (running)
3194 		virtnet_napi_tx_enable(vi, sq->vq, &sq->napi);
3195 }
3196 
3197 static int virtnet_tx_resize(struct virtnet_info *vi, struct send_queue *sq,
3198 			     u32 ring_num)
3199 {
3200 	int qindex, err;
3201 
3202 	qindex = sq - vi->sq;
3203 
3204 	virtnet_tx_pause(vi, sq);
3205 
3206 	err = virtqueue_resize(sq->vq, ring_num, virtnet_sq_free_unused_buf);
3207 	if (err)
3208 		netdev_err(vi->dev, "resize tx fail: tx queue index: %d err: %d\n", qindex, err);
3209 
3210 	virtnet_tx_resume(vi, sq);
3211 
3212 	return err;
3213 }
3214 
3215 /*
3216  * Send command via the control virtqueue and check status.  Commands
3217  * supported by the hypervisor, as indicated by feature bits, should
3218  * never fail unless improperly formatted.
3219  */
3220 static bool virtnet_send_command_reply(struct virtnet_info *vi, u8 class, u8 cmd,
3221 				       struct scatterlist *out,
3222 				       struct scatterlist *in)
3223 {
3224 	struct scatterlist *sgs[5], hdr, stat;
3225 	u32 out_num = 0, tmp, in_num = 0;
3226 	bool ok;
3227 	int ret;
3228 
3229 	/* Caller should know better */
3230 	BUG_ON(!virtio_has_feature(vi->vdev, VIRTIO_NET_F_CTRL_VQ));
3231 
3232 	mutex_lock(&vi->cvq_lock);
3233 	vi->ctrl->status = ~0;
3234 	vi->ctrl->hdr.class = class;
3235 	vi->ctrl->hdr.cmd = cmd;
3236 	/* Add header */
3237 	sg_init_one(&hdr, &vi->ctrl->hdr, sizeof(vi->ctrl->hdr));
3238 	sgs[out_num++] = &hdr;
3239 
3240 	if (out)
3241 		sgs[out_num++] = out;
3242 
3243 	/* Add return status. */
3244 	sg_init_one(&stat, &vi->ctrl->status, sizeof(vi->ctrl->status));
3245 	sgs[out_num + in_num++] = &stat;
3246 
3247 	if (in)
3248 		sgs[out_num + in_num++] = in;
3249 
3250 	BUG_ON(out_num + in_num > ARRAY_SIZE(sgs));
3251 	ret = virtqueue_add_sgs(vi->cvq, sgs, out_num, in_num, vi, GFP_ATOMIC);
3252 	if (ret < 0) {
3253 		dev_warn(&vi->vdev->dev,
3254 			 "Failed to add sgs for command vq: %d\n.", ret);
3255 		mutex_unlock(&vi->cvq_lock);
3256 		return false;
3257 	}
3258 
3259 	if (unlikely(!virtqueue_kick(vi->cvq)))
3260 		goto unlock;
3261 
3262 	/* Spin for a response, the kick causes an ioport write, trapping
3263 	 * into the hypervisor, so the request should be handled immediately.
3264 	 */
3265 	while (!virtqueue_get_buf(vi->cvq, &tmp) &&
3266 	       !virtqueue_is_broken(vi->cvq)) {
3267 		cond_resched();
3268 		cpu_relax();
3269 	}
3270 
3271 unlock:
3272 	ok = vi->ctrl->status == VIRTIO_NET_OK;
3273 	mutex_unlock(&vi->cvq_lock);
3274 	return ok;
3275 }
3276 
3277 static bool virtnet_send_command(struct virtnet_info *vi, u8 class, u8 cmd,
3278 				 struct scatterlist *out)
3279 {
3280 	return virtnet_send_command_reply(vi, class, cmd, out, NULL);
3281 }
3282 
3283 static int virtnet_set_mac_address(struct net_device *dev, void *p)
3284 {
3285 	struct virtnet_info *vi = netdev_priv(dev);
3286 	struct virtio_device *vdev = vi->vdev;
3287 	int ret;
3288 	struct sockaddr *addr;
3289 	struct scatterlist sg;
3290 
3291 	if (virtio_has_feature(vi->vdev, VIRTIO_NET_F_STANDBY))
3292 		return -EOPNOTSUPP;
3293 
3294 	addr = kmemdup(p, sizeof(*addr), GFP_KERNEL);
3295 	if (!addr)
3296 		return -ENOMEM;
3297 
3298 	ret = eth_prepare_mac_addr_change(dev, addr);
3299 	if (ret)
3300 		goto out;
3301 
3302 	if (virtio_has_feature(vdev, VIRTIO_NET_F_CTRL_MAC_ADDR)) {
3303 		sg_init_one(&sg, addr->sa_data, dev->addr_len);
3304 		if (!virtnet_send_command(vi, VIRTIO_NET_CTRL_MAC,
3305 					  VIRTIO_NET_CTRL_MAC_ADDR_SET, &sg)) {
3306 			dev_warn(&vdev->dev,
3307 				 "Failed to set mac address by vq command.\n");
3308 			ret = -EINVAL;
3309 			goto out;
3310 		}
3311 	} else if (virtio_has_feature(vdev, VIRTIO_NET_F_MAC) &&
3312 		   !virtio_has_feature(vdev, VIRTIO_F_VERSION_1)) {
3313 		unsigned int i;
3314 
3315 		/* Naturally, this has an atomicity problem. */
3316 		for (i = 0; i < dev->addr_len; i++)
3317 			virtio_cwrite8(vdev,
3318 				       offsetof(struct virtio_net_config, mac) +
3319 				       i, addr->sa_data[i]);
3320 	}
3321 
3322 	eth_commit_mac_addr_change(dev, p);
3323 	ret = 0;
3324 
3325 out:
3326 	kfree(addr);
3327 	return ret;
3328 }
3329 
3330 static void virtnet_stats(struct net_device *dev,
3331 			  struct rtnl_link_stats64 *tot)
3332 {
3333 	struct virtnet_info *vi = netdev_priv(dev);
3334 	unsigned int start;
3335 	int i;
3336 
3337 	for (i = 0; i < vi->max_queue_pairs; i++) {
3338 		u64 tpackets, tbytes, terrors, rpackets, rbytes, rdrops;
3339 		struct receive_queue *rq = &vi->rq[i];
3340 		struct send_queue *sq = &vi->sq[i];
3341 
3342 		do {
3343 			start = u64_stats_fetch_begin(&sq->stats.syncp);
3344 			tpackets = u64_stats_read(&sq->stats.packets);
3345 			tbytes   = u64_stats_read(&sq->stats.bytes);
3346 			terrors  = u64_stats_read(&sq->stats.tx_timeouts);
3347 		} while (u64_stats_fetch_retry(&sq->stats.syncp, start));
3348 
3349 		do {
3350 			start = u64_stats_fetch_begin(&rq->stats.syncp);
3351 			rpackets = u64_stats_read(&rq->stats.packets);
3352 			rbytes   = u64_stats_read(&rq->stats.bytes);
3353 			rdrops   = u64_stats_read(&rq->stats.drops);
3354 		} while (u64_stats_fetch_retry(&rq->stats.syncp, start));
3355 
3356 		tot->rx_packets += rpackets;
3357 		tot->tx_packets += tpackets;
3358 		tot->rx_bytes   += rbytes;
3359 		tot->tx_bytes   += tbytes;
3360 		tot->rx_dropped += rdrops;
3361 		tot->tx_errors  += terrors;
3362 	}
3363 
3364 	tot->tx_dropped = DEV_STATS_READ(dev, tx_dropped);
3365 	tot->tx_fifo_errors = DEV_STATS_READ(dev, tx_fifo_errors);
3366 	tot->rx_length_errors = DEV_STATS_READ(dev, rx_length_errors);
3367 	tot->rx_frame_errors = DEV_STATS_READ(dev, rx_frame_errors);
3368 }
3369 
3370 static void virtnet_ack_link_announce(struct virtnet_info *vi)
3371 {
3372 	if (!virtnet_send_command(vi, VIRTIO_NET_CTRL_ANNOUNCE,
3373 				  VIRTIO_NET_CTRL_ANNOUNCE_ACK, NULL))
3374 		dev_warn(&vi->dev->dev, "Failed to ack link announce.\n");
3375 }
3376 
3377 static int virtnet_set_queues(struct virtnet_info *vi, u16 queue_pairs)
3378 {
3379 	struct virtio_net_ctrl_mq *mq __free(kfree) = NULL;
3380 	struct scatterlist sg;
3381 	struct net_device *dev = vi->dev;
3382 
3383 	if (!vi->has_cvq || !virtio_has_feature(vi->vdev, VIRTIO_NET_F_MQ))
3384 		return 0;
3385 
3386 	mq = kzalloc(sizeof(*mq), GFP_KERNEL);
3387 	if (!mq)
3388 		return -ENOMEM;
3389 
3390 	mq->virtqueue_pairs = cpu_to_virtio16(vi->vdev, queue_pairs);
3391 	sg_init_one(&sg, mq, sizeof(*mq));
3392 
3393 	if (!virtnet_send_command(vi, VIRTIO_NET_CTRL_MQ,
3394 				  VIRTIO_NET_CTRL_MQ_VQ_PAIRS_SET, &sg)) {
3395 		dev_warn(&dev->dev, "Fail to set num of queue pairs to %d\n",
3396 			 queue_pairs);
3397 		return -EINVAL;
3398 	} else {
3399 		vi->curr_queue_pairs = queue_pairs;
3400 		/* virtnet_open() will refill when device is going to up. */
3401 		if (dev->flags & IFF_UP)
3402 			schedule_delayed_work(&vi->refill, 0);
3403 	}
3404 
3405 	return 0;
3406 }
3407 
3408 static int virtnet_close(struct net_device *dev)
3409 {
3410 	struct virtnet_info *vi = netdev_priv(dev);
3411 	int i;
3412 
3413 	/* Make sure NAPI doesn't schedule refill work */
3414 	disable_delayed_refill(vi);
3415 	/* Make sure refill_work doesn't re-enable napi! */
3416 	cancel_delayed_work_sync(&vi->refill);
3417 	/* Prevent the config change callback from changing carrier
3418 	 * after close
3419 	 */
3420 	virtio_config_driver_disable(vi->vdev);
3421 	/* Stop getting status/speed updates: we don't care until next
3422 	 * open
3423 	 */
3424 	cancel_work_sync(&vi->config_work);
3425 
3426 	for (i = 0; i < vi->max_queue_pairs; i++) {
3427 		virtnet_disable_queue_pair(vi, i);
3428 		virtnet_cancel_dim(vi, &vi->rq[i].dim);
3429 	}
3430 
3431 	netif_carrier_off(dev);
3432 
3433 	return 0;
3434 }
3435 
3436 static void virtnet_rx_mode_work(struct work_struct *work)
3437 {
3438 	struct virtnet_info *vi =
3439 		container_of(work, struct virtnet_info, rx_mode_work);
3440 	u8 *promisc_allmulti  __free(kfree) = NULL;
3441 	struct net_device *dev = vi->dev;
3442 	struct scatterlist sg[2];
3443 	struct virtio_net_ctrl_mac *mac_data;
3444 	struct netdev_hw_addr *ha;
3445 	int uc_count;
3446 	int mc_count;
3447 	void *buf;
3448 	int i;
3449 
3450 	/* We can't dynamically set ndo_set_rx_mode, so return gracefully */
3451 	if (!virtio_has_feature(vi->vdev, VIRTIO_NET_F_CTRL_RX))
3452 		return;
3453 
3454 	promisc_allmulti = kzalloc(sizeof(*promisc_allmulti), GFP_KERNEL);
3455 	if (!promisc_allmulti) {
3456 		dev_warn(&dev->dev, "Failed to set RX mode, no memory.\n");
3457 		return;
3458 	}
3459 
3460 	rtnl_lock();
3461 
3462 	*promisc_allmulti = !!(dev->flags & IFF_PROMISC);
3463 	sg_init_one(sg, promisc_allmulti, sizeof(*promisc_allmulti));
3464 
3465 	if (!virtnet_send_command(vi, VIRTIO_NET_CTRL_RX,
3466 				  VIRTIO_NET_CTRL_RX_PROMISC, sg))
3467 		dev_warn(&dev->dev, "Failed to %sable promisc mode.\n",
3468 			 *promisc_allmulti ? "en" : "dis");
3469 
3470 	*promisc_allmulti = !!(dev->flags & IFF_ALLMULTI);
3471 	sg_init_one(sg, promisc_allmulti, sizeof(*promisc_allmulti));
3472 
3473 	if (!virtnet_send_command(vi, VIRTIO_NET_CTRL_RX,
3474 				  VIRTIO_NET_CTRL_RX_ALLMULTI, sg))
3475 		dev_warn(&dev->dev, "Failed to %sable allmulti mode.\n",
3476 			 *promisc_allmulti ? "en" : "dis");
3477 
3478 	netif_addr_lock_bh(dev);
3479 
3480 	uc_count = netdev_uc_count(dev);
3481 	mc_count = netdev_mc_count(dev);
3482 	/* MAC filter - use one buffer for both lists */
3483 	buf = kzalloc(((uc_count + mc_count) * ETH_ALEN) +
3484 		      (2 * sizeof(mac_data->entries)), GFP_ATOMIC);
3485 	mac_data = buf;
3486 	if (!buf) {
3487 		netif_addr_unlock_bh(dev);
3488 		rtnl_unlock();
3489 		return;
3490 	}
3491 
3492 	sg_init_table(sg, 2);
3493 
3494 	/* Store the unicast list and count in the front of the buffer */
3495 	mac_data->entries = cpu_to_virtio32(vi->vdev, uc_count);
3496 	i = 0;
3497 	netdev_for_each_uc_addr(ha, dev)
3498 		memcpy(&mac_data->macs[i++][0], ha->addr, ETH_ALEN);
3499 
3500 	sg_set_buf(&sg[0], mac_data,
3501 		   sizeof(mac_data->entries) + (uc_count * ETH_ALEN));
3502 
3503 	/* multicast list and count fill the end */
3504 	mac_data = (void *)&mac_data->macs[uc_count][0];
3505 
3506 	mac_data->entries = cpu_to_virtio32(vi->vdev, mc_count);
3507 	i = 0;
3508 	netdev_for_each_mc_addr(ha, dev)
3509 		memcpy(&mac_data->macs[i++][0], ha->addr, ETH_ALEN);
3510 
3511 	netif_addr_unlock_bh(dev);
3512 
3513 	sg_set_buf(&sg[1], mac_data,
3514 		   sizeof(mac_data->entries) + (mc_count * ETH_ALEN));
3515 
3516 	if (!virtnet_send_command(vi, VIRTIO_NET_CTRL_MAC,
3517 				  VIRTIO_NET_CTRL_MAC_TABLE_SET, sg))
3518 		dev_warn(&dev->dev, "Failed to set MAC filter table.\n");
3519 
3520 	rtnl_unlock();
3521 
3522 	kfree(buf);
3523 }
3524 
3525 static void virtnet_set_rx_mode(struct net_device *dev)
3526 {
3527 	struct virtnet_info *vi = netdev_priv(dev);
3528 
3529 	if (vi->rx_mode_work_enabled)
3530 		schedule_work(&vi->rx_mode_work);
3531 }
3532 
3533 static int virtnet_vlan_rx_add_vid(struct net_device *dev,
3534 				   __be16 proto, u16 vid)
3535 {
3536 	struct virtnet_info *vi = netdev_priv(dev);
3537 	__virtio16 *_vid __free(kfree) = NULL;
3538 	struct scatterlist sg;
3539 
3540 	_vid = kzalloc(sizeof(*_vid), GFP_KERNEL);
3541 	if (!_vid)
3542 		return -ENOMEM;
3543 
3544 	*_vid = cpu_to_virtio16(vi->vdev, vid);
3545 	sg_init_one(&sg, _vid, sizeof(*_vid));
3546 
3547 	if (!virtnet_send_command(vi, VIRTIO_NET_CTRL_VLAN,
3548 				  VIRTIO_NET_CTRL_VLAN_ADD, &sg))
3549 		dev_warn(&dev->dev, "Failed to add VLAN ID %d.\n", vid);
3550 	return 0;
3551 }
3552 
3553 static int virtnet_vlan_rx_kill_vid(struct net_device *dev,
3554 				    __be16 proto, u16 vid)
3555 {
3556 	struct virtnet_info *vi = netdev_priv(dev);
3557 	__virtio16 *_vid __free(kfree) = NULL;
3558 	struct scatterlist sg;
3559 
3560 	_vid = kzalloc(sizeof(*_vid), GFP_KERNEL);
3561 	if (!_vid)
3562 		return -ENOMEM;
3563 
3564 	*_vid = cpu_to_virtio16(vi->vdev, vid);
3565 	sg_init_one(&sg, _vid, sizeof(*_vid));
3566 
3567 	if (!virtnet_send_command(vi, VIRTIO_NET_CTRL_VLAN,
3568 				  VIRTIO_NET_CTRL_VLAN_DEL, &sg))
3569 		dev_warn(&dev->dev, "Failed to kill VLAN ID %d.\n", vid);
3570 	return 0;
3571 }
3572 
3573 static void virtnet_clean_affinity(struct virtnet_info *vi)
3574 {
3575 	int i;
3576 
3577 	if (vi->affinity_hint_set) {
3578 		for (i = 0; i < vi->max_queue_pairs; i++) {
3579 			virtqueue_set_affinity(vi->rq[i].vq, NULL);
3580 			virtqueue_set_affinity(vi->sq[i].vq, NULL);
3581 		}
3582 
3583 		vi->affinity_hint_set = false;
3584 	}
3585 }
3586 
3587 static void virtnet_set_affinity(struct virtnet_info *vi)
3588 {
3589 	cpumask_var_t mask;
3590 	int stragglers;
3591 	int group_size;
3592 	int i, j, cpu;
3593 	int num_cpu;
3594 	int stride;
3595 
3596 	if (!zalloc_cpumask_var(&mask, GFP_KERNEL)) {
3597 		virtnet_clean_affinity(vi);
3598 		return;
3599 	}
3600 
3601 	num_cpu = num_online_cpus();
3602 	stride = max_t(int, num_cpu / vi->curr_queue_pairs, 1);
3603 	stragglers = num_cpu >= vi->curr_queue_pairs ?
3604 			num_cpu % vi->curr_queue_pairs :
3605 			0;
3606 	cpu = cpumask_first(cpu_online_mask);
3607 
3608 	for (i = 0; i < vi->curr_queue_pairs; i++) {
3609 		group_size = stride + (i < stragglers ? 1 : 0);
3610 
3611 		for (j = 0; j < group_size; j++) {
3612 			cpumask_set_cpu(cpu, mask);
3613 			cpu = cpumask_next_wrap(cpu, cpu_online_mask,
3614 						nr_cpu_ids, false);
3615 		}
3616 		virtqueue_set_affinity(vi->rq[i].vq, mask);
3617 		virtqueue_set_affinity(vi->sq[i].vq, mask);
3618 		__netif_set_xps_queue(vi->dev, cpumask_bits(mask), i, XPS_CPUS);
3619 		cpumask_clear(mask);
3620 	}
3621 
3622 	vi->affinity_hint_set = true;
3623 	free_cpumask_var(mask);
3624 }
3625 
3626 static int virtnet_cpu_online(unsigned int cpu, struct hlist_node *node)
3627 {
3628 	struct virtnet_info *vi = hlist_entry_safe(node, struct virtnet_info,
3629 						   node);
3630 	virtnet_set_affinity(vi);
3631 	return 0;
3632 }
3633 
3634 static int virtnet_cpu_dead(unsigned int cpu, struct hlist_node *node)
3635 {
3636 	struct virtnet_info *vi = hlist_entry_safe(node, struct virtnet_info,
3637 						   node_dead);
3638 	virtnet_set_affinity(vi);
3639 	return 0;
3640 }
3641 
3642 static int virtnet_cpu_down_prep(unsigned int cpu, struct hlist_node *node)
3643 {
3644 	struct virtnet_info *vi = hlist_entry_safe(node, struct virtnet_info,
3645 						   node);
3646 
3647 	virtnet_clean_affinity(vi);
3648 	return 0;
3649 }
3650 
3651 static enum cpuhp_state virtionet_online;
3652 
3653 static int virtnet_cpu_notif_add(struct virtnet_info *vi)
3654 {
3655 	int ret;
3656 
3657 	ret = cpuhp_state_add_instance_nocalls(virtionet_online, &vi->node);
3658 	if (ret)
3659 		return ret;
3660 	ret = cpuhp_state_add_instance_nocalls(CPUHP_VIRT_NET_DEAD,
3661 					       &vi->node_dead);
3662 	if (!ret)
3663 		return ret;
3664 	cpuhp_state_remove_instance_nocalls(virtionet_online, &vi->node);
3665 	return ret;
3666 }
3667 
3668 static void virtnet_cpu_notif_remove(struct virtnet_info *vi)
3669 {
3670 	cpuhp_state_remove_instance_nocalls(virtionet_online, &vi->node);
3671 	cpuhp_state_remove_instance_nocalls(CPUHP_VIRT_NET_DEAD,
3672 					    &vi->node_dead);
3673 }
3674 
3675 static int virtnet_send_ctrl_coal_vq_cmd(struct virtnet_info *vi,
3676 					 u16 vqn, u32 max_usecs, u32 max_packets)
3677 {
3678 	struct virtio_net_ctrl_coal_vq *coal_vq __free(kfree) = NULL;
3679 	struct scatterlist sgs;
3680 
3681 	coal_vq = kzalloc(sizeof(*coal_vq), GFP_KERNEL);
3682 	if (!coal_vq)
3683 		return -ENOMEM;
3684 
3685 	coal_vq->vqn = cpu_to_le16(vqn);
3686 	coal_vq->coal.max_usecs = cpu_to_le32(max_usecs);
3687 	coal_vq->coal.max_packets = cpu_to_le32(max_packets);
3688 	sg_init_one(&sgs, coal_vq, sizeof(*coal_vq));
3689 
3690 	if (!virtnet_send_command(vi, VIRTIO_NET_CTRL_NOTF_COAL,
3691 				  VIRTIO_NET_CTRL_NOTF_COAL_VQ_SET,
3692 				  &sgs))
3693 		return -EINVAL;
3694 
3695 	return 0;
3696 }
3697 
3698 static int virtnet_send_rx_ctrl_coal_vq_cmd(struct virtnet_info *vi,
3699 					    u16 queue, u32 max_usecs,
3700 					    u32 max_packets)
3701 {
3702 	int err;
3703 
3704 	if (!virtio_has_feature(vi->vdev, VIRTIO_NET_F_VQ_NOTF_COAL))
3705 		return -EOPNOTSUPP;
3706 
3707 	err = virtnet_send_ctrl_coal_vq_cmd(vi, rxq2vq(queue),
3708 					    max_usecs, max_packets);
3709 	if (err)
3710 		return err;
3711 
3712 	vi->rq[queue].intr_coal.max_usecs = max_usecs;
3713 	vi->rq[queue].intr_coal.max_packets = max_packets;
3714 
3715 	return 0;
3716 }
3717 
3718 static int virtnet_send_tx_ctrl_coal_vq_cmd(struct virtnet_info *vi,
3719 					    u16 queue, u32 max_usecs,
3720 					    u32 max_packets)
3721 {
3722 	int err;
3723 
3724 	if (!virtio_has_feature(vi->vdev, VIRTIO_NET_F_VQ_NOTF_COAL))
3725 		return -EOPNOTSUPP;
3726 
3727 	err = virtnet_send_ctrl_coal_vq_cmd(vi, txq2vq(queue),
3728 					    max_usecs, max_packets);
3729 	if (err)
3730 		return err;
3731 
3732 	vi->sq[queue].intr_coal.max_usecs = max_usecs;
3733 	vi->sq[queue].intr_coal.max_packets = max_packets;
3734 
3735 	return 0;
3736 }
3737 
3738 static void virtnet_get_ringparam(struct net_device *dev,
3739 				  struct ethtool_ringparam *ring,
3740 				  struct kernel_ethtool_ringparam *kernel_ring,
3741 				  struct netlink_ext_ack *extack)
3742 {
3743 	struct virtnet_info *vi = netdev_priv(dev);
3744 
3745 	ring->rx_max_pending = vi->rq[0].vq->num_max;
3746 	ring->tx_max_pending = vi->sq[0].vq->num_max;
3747 	ring->rx_pending = virtqueue_get_vring_size(vi->rq[0].vq);
3748 	ring->tx_pending = virtqueue_get_vring_size(vi->sq[0].vq);
3749 }
3750 
3751 static int virtnet_set_ringparam(struct net_device *dev,
3752 				 struct ethtool_ringparam *ring,
3753 				 struct kernel_ethtool_ringparam *kernel_ring,
3754 				 struct netlink_ext_ack *extack)
3755 {
3756 	struct virtnet_info *vi = netdev_priv(dev);
3757 	u32 rx_pending, tx_pending;
3758 	struct receive_queue *rq;
3759 	struct send_queue *sq;
3760 	int i, err;
3761 
3762 	if (ring->rx_mini_pending || ring->rx_jumbo_pending)
3763 		return -EINVAL;
3764 
3765 	rx_pending = virtqueue_get_vring_size(vi->rq[0].vq);
3766 	tx_pending = virtqueue_get_vring_size(vi->sq[0].vq);
3767 
3768 	if (ring->rx_pending == rx_pending &&
3769 	    ring->tx_pending == tx_pending)
3770 		return 0;
3771 
3772 	if (ring->rx_pending > vi->rq[0].vq->num_max)
3773 		return -EINVAL;
3774 
3775 	if (ring->tx_pending > vi->sq[0].vq->num_max)
3776 		return -EINVAL;
3777 
3778 	for (i = 0; i < vi->max_queue_pairs; i++) {
3779 		rq = vi->rq + i;
3780 		sq = vi->sq + i;
3781 
3782 		if (ring->tx_pending != tx_pending) {
3783 			err = virtnet_tx_resize(vi, sq, ring->tx_pending);
3784 			if (err)
3785 				return err;
3786 
3787 			/* Upon disabling and re-enabling a transmit virtqueue, the device must
3788 			 * set the coalescing parameters of the virtqueue to those configured
3789 			 * through the VIRTIO_NET_CTRL_NOTF_COAL_TX_SET command, or, if the driver
3790 			 * did not set any TX coalescing parameters, to 0.
3791 			 */
3792 			err = virtnet_send_tx_ctrl_coal_vq_cmd(vi, i,
3793 							       vi->intr_coal_tx.max_usecs,
3794 							       vi->intr_coal_tx.max_packets);
3795 
3796 			/* Don't break the tx resize action if the vq coalescing is not
3797 			 * supported. The same is true for rx resize below.
3798 			 */
3799 			if (err && err != -EOPNOTSUPP)
3800 				return err;
3801 		}
3802 
3803 		if (ring->rx_pending != rx_pending) {
3804 			err = virtnet_rx_resize(vi, rq, ring->rx_pending);
3805 			if (err)
3806 				return err;
3807 
3808 			/* The reason is same as the transmit virtqueue reset */
3809 			mutex_lock(&vi->rq[i].dim_lock);
3810 			err = virtnet_send_rx_ctrl_coal_vq_cmd(vi, i,
3811 							       vi->intr_coal_rx.max_usecs,
3812 							       vi->intr_coal_rx.max_packets);
3813 			mutex_unlock(&vi->rq[i].dim_lock);
3814 			if (err && err != -EOPNOTSUPP)
3815 				return err;
3816 		}
3817 	}
3818 
3819 	return 0;
3820 }
3821 
3822 static bool virtnet_commit_rss_command(struct virtnet_info *vi)
3823 {
3824 	struct net_device *dev = vi->dev;
3825 	struct scatterlist sgs[4];
3826 	unsigned int sg_buf_size;
3827 
3828 	/* prepare sgs */
3829 	sg_init_table(sgs, 4);
3830 
3831 	sg_buf_size = offsetof(struct virtio_net_ctrl_rss, indirection_table);
3832 	sg_set_buf(&sgs[0], &vi->rss, sg_buf_size);
3833 
3834 	sg_buf_size = sizeof(uint16_t) * (vi->rss.indirection_table_mask + 1);
3835 	sg_set_buf(&sgs[1], vi->rss.indirection_table, sg_buf_size);
3836 
3837 	sg_buf_size = offsetof(struct virtio_net_ctrl_rss, key)
3838 			- offsetof(struct virtio_net_ctrl_rss, max_tx_vq);
3839 	sg_set_buf(&sgs[2], &vi->rss.max_tx_vq, sg_buf_size);
3840 
3841 	sg_buf_size = vi->rss_key_size;
3842 	sg_set_buf(&sgs[3], vi->rss.key, sg_buf_size);
3843 
3844 	if (!virtnet_send_command(vi, VIRTIO_NET_CTRL_MQ,
3845 				  vi->has_rss ? VIRTIO_NET_CTRL_MQ_RSS_CONFIG
3846 				  : VIRTIO_NET_CTRL_MQ_HASH_CONFIG, sgs))
3847 		goto err;
3848 
3849 	return true;
3850 
3851 err:
3852 	dev_warn(&dev->dev, "VIRTIONET issue with committing RSS sgs\n");
3853 	return false;
3854 
3855 }
3856 
3857 static void virtnet_init_default_rss(struct virtnet_info *vi)
3858 {
3859 	u32 indir_val = 0;
3860 	int i = 0;
3861 
3862 	vi->rss.hash_types = vi->rss_hash_types_supported;
3863 	vi->rss_hash_types_saved = vi->rss_hash_types_supported;
3864 	vi->rss.indirection_table_mask = vi->rss_indir_table_size
3865 						? vi->rss_indir_table_size - 1 : 0;
3866 	vi->rss.unclassified_queue = 0;
3867 
3868 	for (; i < vi->rss_indir_table_size; ++i) {
3869 		indir_val = ethtool_rxfh_indir_default(i, vi->curr_queue_pairs);
3870 		vi->rss.indirection_table[i] = indir_val;
3871 	}
3872 
3873 	vi->rss.max_tx_vq = vi->has_rss ? vi->curr_queue_pairs : 0;
3874 	vi->rss.hash_key_length = vi->rss_key_size;
3875 
3876 	netdev_rss_key_fill(vi->rss.key, vi->rss_key_size);
3877 }
3878 
3879 static void virtnet_get_hashflow(const struct virtnet_info *vi, struct ethtool_rxnfc *info)
3880 {
3881 	info->data = 0;
3882 	switch (info->flow_type) {
3883 	case TCP_V4_FLOW:
3884 		if (vi->rss_hash_types_saved & VIRTIO_NET_RSS_HASH_TYPE_TCPv4) {
3885 			info->data = RXH_IP_SRC | RXH_IP_DST |
3886 						 RXH_L4_B_0_1 | RXH_L4_B_2_3;
3887 		} else if (vi->rss_hash_types_saved & VIRTIO_NET_RSS_HASH_TYPE_IPv4) {
3888 			info->data = RXH_IP_SRC | RXH_IP_DST;
3889 		}
3890 		break;
3891 	case TCP_V6_FLOW:
3892 		if (vi->rss_hash_types_saved & VIRTIO_NET_RSS_HASH_TYPE_TCPv6) {
3893 			info->data = RXH_IP_SRC | RXH_IP_DST |
3894 						 RXH_L4_B_0_1 | RXH_L4_B_2_3;
3895 		} else if (vi->rss_hash_types_saved & VIRTIO_NET_RSS_HASH_TYPE_IPv6) {
3896 			info->data = RXH_IP_SRC | RXH_IP_DST;
3897 		}
3898 		break;
3899 	case UDP_V4_FLOW:
3900 		if (vi->rss_hash_types_saved & VIRTIO_NET_RSS_HASH_TYPE_UDPv4) {
3901 			info->data = RXH_IP_SRC | RXH_IP_DST |
3902 						 RXH_L4_B_0_1 | RXH_L4_B_2_3;
3903 		} else if (vi->rss_hash_types_saved & VIRTIO_NET_RSS_HASH_TYPE_IPv4) {
3904 			info->data = RXH_IP_SRC | RXH_IP_DST;
3905 		}
3906 		break;
3907 	case UDP_V6_FLOW:
3908 		if (vi->rss_hash_types_saved & VIRTIO_NET_RSS_HASH_TYPE_UDPv6) {
3909 			info->data = RXH_IP_SRC | RXH_IP_DST |
3910 						 RXH_L4_B_0_1 | RXH_L4_B_2_3;
3911 		} else if (vi->rss_hash_types_saved & VIRTIO_NET_RSS_HASH_TYPE_IPv6) {
3912 			info->data = RXH_IP_SRC | RXH_IP_DST;
3913 		}
3914 		break;
3915 	case IPV4_FLOW:
3916 		if (vi->rss_hash_types_saved & VIRTIO_NET_RSS_HASH_TYPE_IPv4)
3917 			info->data = RXH_IP_SRC | RXH_IP_DST;
3918 
3919 		break;
3920 	case IPV6_FLOW:
3921 		if (vi->rss_hash_types_saved & VIRTIO_NET_RSS_HASH_TYPE_IPv6)
3922 			info->data = RXH_IP_SRC | RXH_IP_DST;
3923 
3924 		break;
3925 	default:
3926 		info->data = 0;
3927 		break;
3928 	}
3929 }
3930 
3931 static bool virtnet_set_hashflow(struct virtnet_info *vi, struct ethtool_rxnfc *info)
3932 {
3933 	u32 new_hashtypes = vi->rss_hash_types_saved;
3934 	bool is_disable = info->data & RXH_DISCARD;
3935 	bool is_l4 = info->data == (RXH_IP_SRC | RXH_IP_DST | RXH_L4_B_0_1 | RXH_L4_B_2_3);
3936 
3937 	/* supports only 'sd', 'sdfn' and 'r' */
3938 	if (!((info->data == (RXH_IP_SRC | RXH_IP_DST)) | is_l4 | is_disable))
3939 		return false;
3940 
3941 	switch (info->flow_type) {
3942 	case TCP_V4_FLOW:
3943 		new_hashtypes &= ~(VIRTIO_NET_RSS_HASH_TYPE_IPv4 | VIRTIO_NET_RSS_HASH_TYPE_TCPv4);
3944 		if (!is_disable)
3945 			new_hashtypes |= VIRTIO_NET_RSS_HASH_TYPE_IPv4
3946 				| (is_l4 ? VIRTIO_NET_RSS_HASH_TYPE_TCPv4 : 0);
3947 		break;
3948 	case UDP_V4_FLOW:
3949 		new_hashtypes &= ~(VIRTIO_NET_RSS_HASH_TYPE_IPv4 | VIRTIO_NET_RSS_HASH_TYPE_UDPv4);
3950 		if (!is_disable)
3951 			new_hashtypes |= VIRTIO_NET_RSS_HASH_TYPE_IPv4
3952 				| (is_l4 ? VIRTIO_NET_RSS_HASH_TYPE_UDPv4 : 0);
3953 		break;
3954 	case IPV4_FLOW:
3955 		new_hashtypes &= ~VIRTIO_NET_RSS_HASH_TYPE_IPv4;
3956 		if (!is_disable)
3957 			new_hashtypes = VIRTIO_NET_RSS_HASH_TYPE_IPv4;
3958 		break;
3959 	case TCP_V6_FLOW:
3960 		new_hashtypes &= ~(VIRTIO_NET_RSS_HASH_TYPE_IPv6 | VIRTIO_NET_RSS_HASH_TYPE_TCPv6);
3961 		if (!is_disable)
3962 			new_hashtypes |= VIRTIO_NET_RSS_HASH_TYPE_IPv6
3963 				| (is_l4 ? VIRTIO_NET_RSS_HASH_TYPE_TCPv6 : 0);
3964 		break;
3965 	case UDP_V6_FLOW:
3966 		new_hashtypes &= ~(VIRTIO_NET_RSS_HASH_TYPE_IPv6 | VIRTIO_NET_RSS_HASH_TYPE_UDPv6);
3967 		if (!is_disable)
3968 			new_hashtypes |= VIRTIO_NET_RSS_HASH_TYPE_IPv6
3969 				| (is_l4 ? VIRTIO_NET_RSS_HASH_TYPE_UDPv6 : 0);
3970 		break;
3971 	case IPV6_FLOW:
3972 		new_hashtypes &= ~VIRTIO_NET_RSS_HASH_TYPE_IPv6;
3973 		if (!is_disable)
3974 			new_hashtypes = VIRTIO_NET_RSS_HASH_TYPE_IPv6;
3975 		break;
3976 	default:
3977 		/* unsupported flow */
3978 		return false;
3979 	}
3980 
3981 	/* if unsupported hashtype was set */
3982 	if (new_hashtypes != (new_hashtypes & vi->rss_hash_types_supported))
3983 		return false;
3984 
3985 	if (new_hashtypes != vi->rss_hash_types_saved) {
3986 		vi->rss_hash_types_saved = new_hashtypes;
3987 		vi->rss.hash_types = vi->rss_hash_types_saved;
3988 		if (vi->dev->features & NETIF_F_RXHASH)
3989 			return virtnet_commit_rss_command(vi);
3990 	}
3991 
3992 	return true;
3993 }
3994 
3995 static void virtnet_get_drvinfo(struct net_device *dev,
3996 				struct ethtool_drvinfo *info)
3997 {
3998 	struct virtnet_info *vi = netdev_priv(dev);
3999 	struct virtio_device *vdev = vi->vdev;
4000 
4001 	strscpy(info->driver, KBUILD_MODNAME, sizeof(info->driver));
4002 	strscpy(info->version, VIRTNET_DRIVER_VERSION, sizeof(info->version));
4003 	strscpy(info->bus_info, virtio_bus_name(vdev), sizeof(info->bus_info));
4004 
4005 }
4006 
4007 /* TODO: Eliminate OOO packets during switching */
4008 static int virtnet_set_channels(struct net_device *dev,
4009 				struct ethtool_channels *channels)
4010 {
4011 	struct virtnet_info *vi = netdev_priv(dev);
4012 	u16 queue_pairs = channels->combined_count;
4013 	int err;
4014 
4015 	/* We don't support separate rx/tx channels.
4016 	 * We don't allow setting 'other' channels.
4017 	 */
4018 	if (channels->rx_count || channels->tx_count || channels->other_count)
4019 		return -EINVAL;
4020 
4021 	if (queue_pairs > vi->max_queue_pairs || queue_pairs == 0)
4022 		return -EINVAL;
4023 
4024 	/* For now we don't support modifying channels while XDP is loaded
4025 	 * also when XDP is loaded all RX queues have XDP programs so we only
4026 	 * need to check a single RX queue.
4027 	 */
4028 	if (vi->rq[0].xdp_prog)
4029 		return -EINVAL;
4030 
4031 	cpus_read_lock();
4032 	err = virtnet_set_queues(vi, queue_pairs);
4033 	if (err) {
4034 		cpus_read_unlock();
4035 		goto err;
4036 	}
4037 	virtnet_set_affinity(vi);
4038 	cpus_read_unlock();
4039 
4040 	netif_set_real_num_tx_queues(dev, queue_pairs);
4041 	netif_set_real_num_rx_queues(dev, queue_pairs);
4042  err:
4043 	return err;
4044 }
4045 
4046 static void virtnet_stats_sprintf(u8 **p, const char *fmt, const char *noq_fmt,
4047 				  int num, int qid, const struct virtnet_stat_desc *desc)
4048 {
4049 	int i;
4050 
4051 	if (qid < 0) {
4052 		for (i = 0; i < num; ++i)
4053 			ethtool_sprintf(p, noq_fmt, desc[i].desc);
4054 	} else {
4055 		for (i = 0; i < num; ++i)
4056 			ethtool_sprintf(p, fmt, qid, desc[i].desc);
4057 	}
4058 }
4059 
4060 /* qid == -1: for rx/tx queue total field */
4061 static void virtnet_get_stats_string(struct virtnet_info *vi, int type, int qid, u8 **data)
4062 {
4063 	const struct virtnet_stat_desc *desc;
4064 	const char *fmt, *noq_fmt;
4065 	u8 *p = *data;
4066 	u32 num;
4067 
4068 	if (type == VIRTNET_Q_TYPE_CQ && qid >= 0) {
4069 		noq_fmt = "cq_hw_%s";
4070 
4071 		if (vi->device_stats_cap & VIRTIO_NET_STATS_TYPE_CVQ) {
4072 			desc = &virtnet_stats_cvq_desc[0];
4073 			num = ARRAY_SIZE(virtnet_stats_cvq_desc);
4074 
4075 			virtnet_stats_sprintf(&p, NULL, noq_fmt, num, -1, desc);
4076 		}
4077 	}
4078 
4079 	if (type == VIRTNET_Q_TYPE_RX) {
4080 		fmt = "rx%u_%s";
4081 		noq_fmt = "rx_%s";
4082 
4083 		desc = &virtnet_rq_stats_desc[0];
4084 		num = ARRAY_SIZE(virtnet_rq_stats_desc);
4085 
4086 		virtnet_stats_sprintf(&p, fmt, noq_fmt, num, qid, desc);
4087 
4088 		fmt = "rx%u_hw_%s";
4089 		noq_fmt = "rx_hw_%s";
4090 
4091 		if (vi->device_stats_cap & VIRTIO_NET_STATS_TYPE_RX_BASIC) {
4092 			desc = &virtnet_stats_rx_basic_desc[0];
4093 			num = ARRAY_SIZE(virtnet_stats_rx_basic_desc);
4094 
4095 			virtnet_stats_sprintf(&p, fmt, noq_fmt, num, qid, desc);
4096 		}
4097 
4098 		if (vi->device_stats_cap & VIRTIO_NET_STATS_TYPE_RX_CSUM) {
4099 			desc = &virtnet_stats_rx_csum_desc[0];
4100 			num = ARRAY_SIZE(virtnet_stats_rx_csum_desc);
4101 
4102 			virtnet_stats_sprintf(&p, fmt, noq_fmt, num, qid, desc);
4103 		}
4104 
4105 		if (vi->device_stats_cap & VIRTIO_NET_STATS_TYPE_RX_SPEED) {
4106 			desc = &virtnet_stats_rx_speed_desc[0];
4107 			num = ARRAY_SIZE(virtnet_stats_rx_speed_desc);
4108 
4109 			virtnet_stats_sprintf(&p, fmt, noq_fmt, num, qid, desc);
4110 		}
4111 	}
4112 
4113 	if (type == VIRTNET_Q_TYPE_TX) {
4114 		fmt = "tx%u_%s";
4115 		noq_fmt = "tx_%s";
4116 
4117 		desc = &virtnet_sq_stats_desc[0];
4118 		num = ARRAY_SIZE(virtnet_sq_stats_desc);
4119 
4120 		virtnet_stats_sprintf(&p, fmt, noq_fmt, num, qid, desc);
4121 
4122 		fmt = "tx%u_hw_%s";
4123 		noq_fmt = "tx_hw_%s";
4124 
4125 		if (vi->device_stats_cap & VIRTIO_NET_STATS_TYPE_TX_BASIC) {
4126 			desc = &virtnet_stats_tx_basic_desc[0];
4127 			num = ARRAY_SIZE(virtnet_stats_tx_basic_desc);
4128 
4129 			virtnet_stats_sprintf(&p, fmt, noq_fmt, num, qid, desc);
4130 		}
4131 
4132 		if (vi->device_stats_cap & VIRTIO_NET_STATS_TYPE_TX_GSO) {
4133 			desc = &virtnet_stats_tx_gso_desc[0];
4134 			num = ARRAY_SIZE(virtnet_stats_tx_gso_desc);
4135 
4136 			virtnet_stats_sprintf(&p, fmt, noq_fmt, num, qid, desc);
4137 		}
4138 
4139 		if (vi->device_stats_cap & VIRTIO_NET_STATS_TYPE_TX_SPEED) {
4140 			desc = &virtnet_stats_tx_speed_desc[0];
4141 			num = ARRAY_SIZE(virtnet_stats_tx_speed_desc);
4142 
4143 			virtnet_stats_sprintf(&p, fmt, noq_fmt, num, qid, desc);
4144 		}
4145 	}
4146 
4147 	*data = p;
4148 }
4149 
4150 struct virtnet_stats_ctx {
4151 	/* The stats are write to qstats or ethtool -S */
4152 	bool to_qstat;
4153 
4154 	/* Used to calculate the offset inside the output buffer. */
4155 	u32 desc_num[3];
4156 
4157 	/* The actual supported stat types. */
4158 	u64 bitmap[3];
4159 
4160 	/* Used to calculate the reply buffer size. */
4161 	u32 size[3];
4162 
4163 	/* Record the output buffer. */
4164 	u64 *data;
4165 };
4166 
4167 static void virtnet_stats_ctx_init(struct virtnet_info *vi,
4168 				   struct virtnet_stats_ctx *ctx,
4169 				   u64 *data, bool to_qstat)
4170 {
4171 	u32 queue_type;
4172 
4173 	ctx->data = data;
4174 	ctx->to_qstat = to_qstat;
4175 
4176 	if (to_qstat) {
4177 		ctx->desc_num[VIRTNET_Q_TYPE_RX] = ARRAY_SIZE(virtnet_rq_stats_desc_qstat);
4178 		ctx->desc_num[VIRTNET_Q_TYPE_TX] = ARRAY_SIZE(virtnet_sq_stats_desc_qstat);
4179 
4180 		queue_type = VIRTNET_Q_TYPE_RX;
4181 
4182 		if (vi->device_stats_cap & VIRTIO_NET_STATS_TYPE_RX_BASIC) {
4183 			ctx->bitmap[queue_type]   |= VIRTIO_NET_STATS_TYPE_RX_BASIC;
4184 			ctx->desc_num[queue_type] += ARRAY_SIZE(virtnet_stats_rx_basic_desc_qstat);
4185 			ctx->size[queue_type]     += sizeof(struct virtio_net_stats_rx_basic);
4186 		}
4187 
4188 		if (vi->device_stats_cap & VIRTIO_NET_STATS_TYPE_RX_CSUM) {
4189 			ctx->bitmap[queue_type]   |= VIRTIO_NET_STATS_TYPE_RX_CSUM;
4190 			ctx->desc_num[queue_type] += ARRAY_SIZE(virtnet_stats_rx_csum_desc_qstat);
4191 			ctx->size[queue_type]     += sizeof(struct virtio_net_stats_rx_csum);
4192 		}
4193 
4194 		if (vi->device_stats_cap & VIRTIO_NET_STATS_TYPE_RX_GSO) {
4195 			ctx->bitmap[queue_type]   |= VIRTIO_NET_STATS_TYPE_RX_GSO;
4196 			ctx->desc_num[queue_type] += ARRAY_SIZE(virtnet_stats_rx_gso_desc_qstat);
4197 			ctx->size[queue_type]     += sizeof(struct virtio_net_stats_rx_gso);
4198 		}
4199 
4200 		if (vi->device_stats_cap & VIRTIO_NET_STATS_TYPE_RX_SPEED) {
4201 			ctx->bitmap[queue_type]   |= VIRTIO_NET_STATS_TYPE_RX_SPEED;
4202 			ctx->desc_num[queue_type] += ARRAY_SIZE(virtnet_stats_rx_speed_desc_qstat);
4203 			ctx->size[queue_type]     += sizeof(struct virtio_net_stats_rx_speed);
4204 		}
4205 
4206 		queue_type = VIRTNET_Q_TYPE_TX;
4207 
4208 		if (vi->device_stats_cap & VIRTIO_NET_STATS_TYPE_TX_BASIC) {
4209 			ctx->bitmap[queue_type]   |= VIRTIO_NET_STATS_TYPE_TX_BASIC;
4210 			ctx->desc_num[queue_type] += ARRAY_SIZE(virtnet_stats_tx_basic_desc_qstat);
4211 			ctx->size[queue_type]     += sizeof(struct virtio_net_stats_tx_basic);
4212 		}
4213 
4214 		if (vi->device_stats_cap & VIRTIO_NET_STATS_TYPE_TX_CSUM) {
4215 			ctx->bitmap[queue_type]   |= VIRTIO_NET_STATS_TYPE_TX_CSUM;
4216 			ctx->desc_num[queue_type] += ARRAY_SIZE(virtnet_stats_tx_csum_desc_qstat);
4217 			ctx->size[queue_type]     += sizeof(struct virtio_net_stats_tx_csum);
4218 		}
4219 
4220 		if (vi->device_stats_cap & VIRTIO_NET_STATS_TYPE_TX_GSO) {
4221 			ctx->bitmap[queue_type]   |= VIRTIO_NET_STATS_TYPE_TX_GSO;
4222 			ctx->desc_num[queue_type] += ARRAY_SIZE(virtnet_stats_tx_gso_desc_qstat);
4223 			ctx->size[queue_type]     += sizeof(struct virtio_net_stats_tx_gso);
4224 		}
4225 
4226 		if (vi->device_stats_cap & VIRTIO_NET_STATS_TYPE_TX_SPEED) {
4227 			ctx->bitmap[queue_type]   |= VIRTIO_NET_STATS_TYPE_TX_SPEED;
4228 			ctx->desc_num[queue_type] += ARRAY_SIZE(virtnet_stats_tx_speed_desc_qstat);
4229 			ctx->size[queue_type]     += sizeof(struct virtio_net_stats_tx_speed);
4230 		}
4231 
4232 		return;
4233 	}
4234 
4235 	ctx->desc_num[VIRTNET_Q_TYPE_RX] = ARRAY_SIZE(virtnet_rq_stats_desc);
4236 	ctx->desc_num[VIRTNET_Q_TYPE_TX] = ARRAY_SIZE(virtnet_sq_stats_desc);
4237 
4238 	if (vi->device_stats_cap & VIRTIO_NET_STATS_TYPE_CVQ) {
4239 		queue_type = VIRTNET_Q_TYPE_CQ;
4240 
4241 		ctx->bitmap[queue_type]   |= VIRTIO_NET_STATS_TYPE_CVQ;
4242 		ctx->desc_num[queue_type] += ARRAY_SIZE(virtnet_stats_cvq_desc);
4243 		ctx->size[queue_type]     += sizeof(struct virtio_net_stats_cvq);
4244 	}
4245 
4246 	queue_type = VIRTNET_Q_TYPE_RX;
4247 
4248 	if (vi->device_stats_cap & VIRTIO_NET_STATS_TYPE_RX_BASIC) {
4249 		ctx->bitmap[queue_type]   |= VIRTIO_NET_STATS_TYPE_RX_BASIC;
4250 		ctx->desc_num[queue_type] += ARRAY_SIZE(virtnet_stats_rx_basic_desc);
4251 		ctx->size[queue_type]     += sizeof(struct virtio_net_stats_rx_basic);
4252 	}
4253 
4254 	if (vi->device_stats_cap & VIRTIO_NET_STATS_TYPE_RX_CSUM) {
4255 		ctx->bitmap[queue_type]   |= VIRTIO_NET_STATS_TYPE_RX_CSUM;
4256 		ctx->desc_num[queue_type] += ARRAY_SIZE(virtnet_stats_rx_csum_desc);
4257 		ctx->size[queue_type]     += sizeof(struct virtio_net_stats_rx_csum);
4258 	}
4259 
4260 	if (vi->device_stats_cap & VIRTIO_NET_STATS_TYPE_RX_SPEED) {
4261 		ctx->bitmap[queue_type]   |= VIRTIO_NET_STATS_TYPE_RX_SPEED;
4262 		ctx->desc_num[queue_type] += ARRAY_SIZE(virtnet_stats_rx_speed_desc);
4263 		ctx->size[queue_type]     += sizeof(struct virtio_net_stats_rx_speed);
4264 	}
4265 
4266 	queue_type = VIRTNET_Q_TYPE_TX;
4267 
4268 	if (vi->device_stats_cap & VIRTIO_NET_STATS_TYPE_TX_BASIC) {
4269 		ctx->bitmap[queue_type]   |= VIRTIO_NET_STATS_TYPE_TX_BASIC;
4270 		ctx->desc_num[queue_type] += ARRAY_SIZE(virtnet_stats_tx_basic_desc);
4271 		ctx->size[queue_type]     += sizeof(struct virtio_net_stats_tx_basic);
4272 	}
4273 
4274 	if (vi->device_stats_cap & VIRTIO_NET_STATS_TYPE_TX_GSO) {
4275 		ctx->bitmap[queue_type]   |= VIRTIO_NET_STATS_TYPE_TX_GSO;
4276 		ctx->desc_num[queue_type] += ARRAY_SIZE(virtnet_stats_tx_gso_desc);
4277 		ctx->size[queue_type]     += sizeof(struct virtio_net_stats_tx_gso);
4278 	}
4279 
4280 	if (vi->device_stats_cap & VIRTIO_NET_STATS_TYPE_TX_SPEED) {
4281 		ctx->bitmap[queue_type]   |= VIRTIO_NET_STATS_TYPE_TX_SPEED;
4282 		ctx->desc_num[queue_type] += ARRAY_SIZE(virtnet_stats_tx_speed_desc);
4283 		ctx->size[queue_type]     += sizeof(struct virtio_net_stats_tx_speed);
4284 	}
4285 }
4286 
4287 /* stats_sum_queue - Calculate the sum of the same fields in sq or rq.
4288  * @sum: the position to store the sum values
4289  * @num: field num
4290  * @q_value: the first queue fields
4291  * @q_num: number of the queues
4292  */
4293 static void stats_sum_queue(u64 *sum, u32 num, u64 *q_value, u32 q_num)
4294 {
4295 	u32 step = num;
4296 	int i, j;
4297 	u64 *p;
4298 
4299 	for (i = 0; i < num; ++i) {
4300 		p = sum + i;
4301 		*p = 0;
4302 
4303 		for (j = 0; j < q_num; ++j)
4304 			*p += *(q_value + i + j * step);
4305 	}
4306 }
4307 
4308 static void virtnet_fill_total_fields(struct virtnet_info *vi,
4309 				      struct virtnet_stats_ctx *ctx)
4310 {
4311 	u64 *data, *first_rx_q, *first_tx_q;
4312 	u32 num_cq, num_rx, num_tx;
4313 
4314 	num_cq = ctx->desc_num[VIRTNET_Q_TYPE_CQ];
4315 	num_rx = ctx->desc_num[VIRTNET_Q_TYPE_RX];
4316 	num_tx = ctx->desc_num[VIRTNET_Q_TYPE_TX];
4317 
4318 	first_rx_q = ctx->data + num_rx + num_tx + num_cq;
4319 	first_tx_q = first_rx_q + vi->curr_queue_pairs * num_rx;
4320 
4321 	data = ctx->data;
4322 
4323 	stats_sum_queue(data, num_rx, first_rx_q, vi->curr_queue_pairs);
4324 
4325 	data = ctx->data + num_rx;
4326 
4327 	stats_sum_queue(data, num_tx, first_tx_q, vi->curr_queue_pairs);
4328 }
4329 
4330 static void virtnet_fill_stats_qstat(struct virtnet_info *vi, u32 qid,
4331 				     struct virtnet_stats_ctx *ctx,
4332 				     const u8 *base, bool drv_stats, u8 reply_type)
4333 {
4334 	const struct virtnet_stat_desc *desc;
4335 	const u64_stats_t *v_stat;
4336 	u64 offset, bitmap;
4337 	const __le64 *v;
4338 	u32 queue_type;
4339 	int i, num;
4340 
4341 	queue_type = vq_type(vi, qid);
4342 	bitmap = ctx->bitmap[queue_type];
4343 
4344 	if (drv_stats) {
4345 		if (queue_type == VIRTNET_Q_TYPE_RX) {
4346 			desc = &virtnet_rq_stats_desc_qstat[0];
4347 			num = ARRAY_SIZE(virtnet_rq_stats_desc_qstat);
4348 		} else {
4349 			desc = &virtnet_sq_stats_desc_qstat[0];
4350 			num = ARRAY_SIZE(virtnet_sq_stats_desc_qstat);
4351 		}
4352 
4353 		for (i = 0; i < num; ++i) {
4354 			offset = desc[i].qstat_offset / sizeof(*ctx->data);
4355 			v_stat = (const u64_stats_t *)(base + desc[i].offset);
4356 			ctx->data[offset] = u64_stats_read(v_stat);
4357 		}
4358 		return;
4359 	}
4360 
4361 	if (bitmap & VIRTIO_NET_STATS_TYPE_RX_BASIC) {
4362 		desc = &virtnet_stats_rx_basic_desc_qstat[0];
4363 		num = ARRAY_SIZE(virtnet_stats_rx_basic_desc_qstat);
4364 		if (reply_type == VIRTIO_NET_STATS_TYPE_REPLY_RX_BASIC)
4365 			goto found;
4366 	}
4367 
4368 	if (bitmap & VIRTIO_NET_STATS_TYPE_RX_CSUM) {
4369 		desc = &virtnet_stats_rx_csum_desc_qstat[0];
4370 		num = ARRAY_SIZE(virtnet_stats_rx_csum_desc_qstat);
4371 		if (reply_type == VIRTIO_NET_STATS_TYPE_REPLY_RX_CSUM)
4372 			goto found;
4373 	}
4374 
4375 	if (bitmap & VIRTIO_NET_STATS_TYPE_RX_GSO) {
4376 		desc = &virtnet_stats_rx_gso_desc_qstat[0];
4377 		num = ARRAY_SIZE(virtnet_stats_rx_gso_desc_qstat);
4378 		if (reply_type == VIRTIO_NET_STATS_TYPE_REPLY_RX_GSO)
4379 			goto found;
4380 	}
4381 
4382 	if (bitmap & VIRTIO_NET_STATS_TYPE_RX_SPEED) {
4383 		desc = &virtnet_stats_rx_speed_desc_qstat[0];
4384 		num = ARRAY_SIZE(virtnet_stats_rx_speed_desc_qstat);
4385 		if (reply_type == VIRTIO_NET_STATS_TYPE_REPLY_RX_SPEED)
4386 			goto found;
4387 	}
4388 
4389 	if (bitmap & VIRTIO_NET_STATS_TYPE_TX_BASIC) {
4390 		desc = &virtnet_stats_tx_basic_desc_qstat[0];
4391 		num = ARRAY_SIZE(virtnet_stats_tx_basic_desc_qstat);
4392 		if (reply_type == VIRTIO_NET_STATS_TYPE_REPLY_TX_BASIC)
4393 			goto found;
4394 	}
4395 
4396 	if (bitmap & VIRTIO_NET_STATS_TYPE_TX_CSUM) {
4397 		desc = &virtnet_stats_tx_csum_desc_qstat[0];
4398 		num = ARRAY_SIZE(virtnet_stats_tx_csum_desc_qstat);
4399 		if (reply_type == VIRTIO_NET_STATS_TYPE_REPLY_TX_CSUM)
4400 			goto found;
4401 	}
4402 
4403 	if (bitmap & VIRTIO_NET_STATS_TYPE_TX_GSO) {
4404 		desc = &virtnet_stats_tx_gso_desc_qstat[0];
4405 		num = ARRAY_SIZE(virtnet_stats_tx_gso_desc_qstat);
4406 		if (reply_type == VIRTIO_NET_STATS_TYPE_REPLY_TX_GSO)
4407 			goto found;
4408 	}
4409 
4410 	if (bitmap & VIRTIO_NET_STATS_TYPE_TX_SPEED) {
4411 		desc = &virtnet_stats_tx_speed_desc_qstat[0];
4412 		num = ARRAY_SIZE(virtnet_stats_tx_speed_desc_qstat);
4413 		if (reply_type == VIRTIO_NET_STATS_TYPE_REPLY_TX_SPEED)
4414 			goto found;
4415 	}
4416 
4417 	return;
4418 
4419 found:
4420 	for (i = 0; i < num; ++i) {
4421 		offset = desc[i].qstat_offset / sizeof(*ctx->data);
4422 		v = (const __le64 *)(base + desc[i].offset);
4423 		ctx->data[offset] = le64_to_cpu(*v);
4424 	}
4425 }
4426 
4427 /* virtnet_fill_stats - copy the stats to qstats or ethtool -S
4428  * The stats source is the device or the driver.
4429  *
4430  * @vi: virtio net info
4431  * @qid: the vq id
4432  * @ctx: stats ctx (initiated by virtnet_stats_ctx_init())
4433  * @base: pointer to the device reply or the driver stats structure.
4434  * @drv_stats: designate the base type (device reply, driver stats)
4435  * @type: the type of the device reply (if drv_stats is true, this must be zero)
4436  */
4437 static void virtnet_fill_stats(struct virtnet_info *vi, u32 qid,
4438 			       struct virtnet_stats_ctx *ctx,
4439 			       const u8 *base, bool drv_stats, u8 reply_type)
4440 {
4441 	u32 queue_type, num_rx, num_tx, num_cq;
4442 	const struct virtnet_stat_desc *desc;
4443 	const u64_stats_t *v_stat;
4444 	u64 offset, bitmap;
4445 	const __le64 *v;
4446 	int i, num;
4447 
4448 	if (ctx->to_qstat)
4449 		return virtnet_fill_stats_qstat(vi, qid, ctx, base, drv_stats, reply_type);
4450 
4451 	num_cq = ctx->desc_num[VIRTNET_Q_TYPE_CQ];
4452 	num_rx = ctx->desc_num[VIRTNET_Q_TYPE_RX];
4453 	num_tx = ctx->desc_num[VIRTNET_Q_TYPE_TX];
4454 
4455 	queue_type = vq_type(vi, qid);
4456 	bitmap = ctx->bitmap[queue_type];
4457 
4458 	/* skip the total fields of pairs */
4459 	offset = num_rx + num_tx;
4460 
4461 	if (queue_type == VIRTNET_Q_TYPE_TX) {
4462 		offset += num_cq + num_rx * vi->curr_queue_pairs + num_tx * (qid / 2);
4463 
4464 		num = ARRAY_SIZE(virtnet_sq_stats_desc);
4465 		if (drv_stats) {
4466 			desc = &virtnet_sq_stats_desc[0];
4467 			goto drv_stats;
4468 		}
4469 
4470 		offset += num;
4471 
4472 	} else if (queue_type == VIRTNET_Q_TYPE_RX) {
4473 		offset += num_cq + num_rx * (qid / 2);
4474 
4475 		num = ARRAY_SIZE(virtnet_rq_stats_desc);
4476 		if (drv_stats) {
4477 			desc = &virtnet_rq_stats_desc[0];
4478 			goto drv_stats;
4479 		}
4480 
4481 		offset += num;
4482 	}
4483 
4484 	if (bitmap & VIRTIO_NET_STATS_TYPE_CVQ) {
4485 		desc = &virtnet_stats_cvq_desc[0];
4486 		num = ARRAY_SIZE(virtnet_stats_cvq_desc);
4487 		if (reply_type == VIRTIO_NET_STATS_TYPE_REPLY_CVQ)
4488 			goto found;
4489 
4490 		offset += num;
4491 	}
4492 
4493 	if (bitmap & VIRTIO_NET_STATS_TYPE_RX_BASIC) {
4494 		desc = &virtnet_stats_rx_basic_desc[0];
4495 		num = ARRAY_SIZE(virtnet_stats_rx_basic_desc);
4496 		if (reply_type == VIRTIO_NET_STATS_TYPE_REPLY_RX_BASIC)
4497 			goto found;
4498 
4499 		offset += num;
4500 	}
4501 
4502 	if (bitmap & VIRTIO_NET_STATS_TYPE_RX_CSUM) {
4503 		desc = &virtnet_stats_rx_csum_desc[0];
4504 		num = ARRAY_SIZE(virtnet_stats_rx_csum_desc);
4505 		if (reply_type == VIRTIO_NET_STATS_TYPE_REPLY_RX_CSUM)
4506 			goto found;
4507 
4508 		offset += num;
4509 	}
4510 
4511 	if (bitmap & VIRTIO_NET_STATS_TYPE_RX_SPEED) {
4512 		desc = &virtnet_stats_rx_speed_desc[0];
4513 		num = ARRAY_SIZE(virtnet_stats_rx_speed_desc);
4514 		if (reply_type == VIRTIO_NET_STATS_TYPE_REPLY_RX_SPEED)
4515 			goto found;
4516 
4517 		offset += num;
4518 	}
4519 
4520 	if (bitmap & VIRTIO_NET_STATS_TYPE_TX_BASIC) {
4521 		desc = &virtnet_stats_tx_basic_desc[0];
4522 		num = ARRAY_SIZE(virtnet_stats_tx_basic_desc);
4523 		if (reply_type == VIRTIO_NET_STATS_TYPE_REPLY_TX_BASIC)
4524 			goto found;
4525 
4526 		offset += num;
4527 	}
4528 
4529 	if (bitmap & VIRTIO_NET_STATS_TYPE_TX_GSO) {
4530 		desc = &virtnet_stats_tx_gso_desc[0];
4531 		num = ARRAY_SIZE(virtnet_stats_tx_gso_desc);
4532 		if (reply_type == VIRTIO_NET_STATS_TYPE_REPLY_TX_GSO)
4533 			goto found;
4534 
4535 		offset += num;
4536 	}
4537 
4538 	if (bitmap & VIRTIO_NET_STATS_TYPE_TX_SPEED) {
4539 		desc = &virtnet_stats_tx_speed_desc[0];
4540 		num = ARRAY_SIZE(virtnet_stats_tx_speed_desc);
4541 		if (reply_type == VIRTIO_NET_STATS_TYPE_REPLY_TX_SPEED)
4542 			goto found;
4543 
4544 		offset += num;
4545 	}
4546 
4547 	return;
4548 
4549 found:
4550 	for (i = 0; i < num; ++i) {
4551 		v = (const __le64 *)(base + desc[i].offset);
4552 		ctx->data[offset + i] = le64_to_cpu(*v);
4553 	}
4554 
4555 	return;
4556 
4557 drv_stats:
4558 	for (i = 0; i < num; ++i) {
4559 		v_stat = (const u64_stats_t *)(base + desc[i].offset);
4560 		ctx->data[offset + i] = u64_stats_read(v_stat);
4561 	}
4562 }
4563 
4564 static int __virtnet_get_hw_stats(struct virtnet_info *vi,
4565 				  struct virtnet_stats_ctx *ctx,
4566 				  struct virtio_net_ctrl_queue_stats *req,
4567 				  int req_size, void *reply, int res_size)
4568 {
4569 	struct virtio_net_stats_reply_hdr *hdr;
4570 	struct scatterlist sgs_in, sgs_out;
4571 	void *p;
4572 	u32 qid;
4573 	int ok;
4574 
4575 	sg_init_one(&sgs_out, req, req_size);
4576 	sg_init_one(&sgs_in, reply, res_size);
4577 
4578 	ok = virtnet_send_command_reply(vi, VIRTIO_NET_CTRL_STATS,
4579 					VIRTIO_NET_CTRL_STATS_GET,
4580 					&sgs_out, &sgs_in);
4581 
4582 	if (!ok)
4583 		return ok;
4584 
4585 	for (p = reply; p - reply < res_size; p += le16_to_cpu(hdr->size)) {
4586 		hdr = p;
4587 		qid = le16_to_cpu(hdr->vq_index);
4588 		virtnet_fill_stats(vi, qid, ctx, p, false, hdr->type);
4589 	}
4590 
4591 	return 0;
4592 }
4593 
4594 static void virtnet_make_stat_req(struct virtnet_info *vi,
4595 				  struct virtnet_stats_ctx *ctx,
4596 				  struct virtio_net_ctrl_queue_stats *req,
4597 				  int qid, int *idx)
4598 {
4599 	int qtype = vq_type(vi, qid);
4600 	u64 bitmap = ctx->bitmap[qtype];
4601 
4602 	if (!bitmap)
4603 		return;
4604 
4605 	req->stats[*idx].vq_index = cpu_to_le16(qid);
4606 	req->stats[*idx].types_bitmap[0] = cpu_to_le64(bitmap);
4607 	*idx += 1;
4608 }
4609 
4610 /* qid: -1: get stats of all vq.
4611  *     > 0: get the stats for the special vq. This must not be cvq.
4612  */
4613 static int virtnet_get_hw_stats(struct virtnet_info *vi,
4614 				struct virtnet_stats_ctx *ctx, int qid)
4615 {
4616 	int qnum, i, j, res_size, qtype, last_vq, first_vq;
4617 	struct virtio_net_ctrl_queue_stats *req;
4618 	bool enable_cvq;
4619 	void *reply;
4620 	int ok;
4621 
4622 	if (!virtio_has_feature(vi->vdev, VIRTIO_NET_F_DEVICE_STATS))
4623 		return 0;
4624 
4625 	if (qid == -1) {
4626 		last_vq = vi->curr_queue_pairs * 2 - 1;
4627 		first_vq = 0;
4628 		enable_cvq = true;
4629 	} else {
4630 		last_vq = qid;
4631 		first_vq = qid;
4632 		enable_cvq = false;
4633 	}
4634 
4635 	qnum = 0;
4636 	res_size = 0;
4637 	for (i = first_vq; i <= last_vq ; ++i) {
4638 		qtype = vq_type(vi, i);
4639 		if (ctx->bitmap[qtype]) {
4640 			++qnum;
4641 			res_size += ctx->size[qtype];
4642 		}
4643 	}
4644 
4645 	if (enable_cvq && ctx->bitmap[VIRTNET_Q_TYPE_CQ]) {
4646 		res_size += ctx->size[VIRTNET_Q_TYPE_CQ];
4647 		qnum += 1;
4648 	}
4649 
4650 	req = kcalloc(qnum, sizeof(*req), GFP_KERNEL);
4651 	if (!req)
4652 		return -ENOMEM;
4653 
4654 	reply = kmalloc(res_size, GFP_KERNEL);
4655 	if (!reply) {
4656 		kfree(req);
4657 		return -ENOMEM;
4658 	}
4659 
4660 	j = 0;
4661 	for (i = first_vq; i <= last_vq ; ++i)
4662 		virtnet_make_stat_req(vi, ctx, req, i, &j);
4663 
4664 	if (enable_cvq)
4665 		virtnet_make_stat_req(vi, ctx, req, vi->max_queue_pairs * 2, &j);
4666 
4667 	ok = __virtnet_get_hw_stats(vi, ctx, req, sizeof(*req) * j, reply, res_size);
4668 
4669 	kfree(req);
4670 	kfree(reply);
4671 
4672 	return ok;
4673 }
4674 
4675 static void virtnet_get_strings(struct net_device *dev, u32 stringset, u8 *data)
4676 {
4677 	struct virtnet_info *vi = netdev_priv(dev);
4678 	unsigned int i;
4679 	u8 *p = data;
4680 
4681 	switch (stringset) {
4682 	case ETH_SS_STATS:
4683 		/* Generate the total field names. */
4684 		virtnet_get_stats_string(vi, VIRTNET_Q_TYPE_RX, -1, &p);
4685 		virtnet_get_stats_string(vi, VIRTNET_Q_TYPE_TX, -1, &p);
4686 
4687 		virtnet_get_stats_string(vi, VIRTNET_Q_TYPE_CQ, 0, &p);
4688 
4689 		for (i = 0; i < vi->curr_queue_pairs; ++i)
4690 			virtnet_get_stats_string(vi, VIRTNET_Q_TYPE_RX, i, &p);
4691 
4692 		for (i = 0; i < vi->curr_queue_pairs; ++i)
4693 			virtnet_get_stats_string(vi, VIRTNET_Q_TYPE_TX, i, &p);
4694 		break;
4695 	}
4696 }
4697 
4698 static int virtnet_get_sset_count(struct net_device *dev, int sset)
4699 {
4700 	struct virtnet_info *vi = netdev_priv(dev);
4701 	struct virtnet_stats_ctx ctx = {0};
4702 	u32 pair_count;
4703 
4704 	switch (sset) {
4705 	case ETH_SS_STATS:
4706 		virtnet_stats_ctx_init(vi, &ctx, NULL, false);
4707 
4708 		pair_count = ctx.desc_num[VIRTNET_Q_TYPE_RX] + ctx.desc_num[VIRTNET_Q_TYPE_TX];
4709 
4710 		return pair_count + ctx.desc_num[VIRTNET_Q_TYPE_CQ] +
4711 			vi->curr_queue_pairs * pair_count;
4712 	default:
4713 		return -EOPNOTSUPP;
4714 	}
4715 }
4716 
4717 static void virtnet_get_ethtool_stats(struct net_device *dev,
4718 				      struct ethtool_stats *stats, u64 *data)
4719 {
4720 	struct virtnet_info *vi = netdev_priv(dev);
4721 	struct virtnet_stats_ctx ctx = {0};
4722 	unsigned int start, i;
4723 	const u8 *stats_base;
4724 
4725 	virtnet_stats_ctx_init(vi, &ctx, data, false);
4726 	if (virtnet_get_hw_stats(vi, &ctx, -1))
4727 		dev_warn(&vi->dev->dev, "Failed to get hw stats.\n");
4728 
4729 	for (i = 0; i < vi->curr_queue_pairs; i++) {
4730 		struct receive_queue *rq = &vi->rq[i];
4731 		struct send_queue *sq = &vi->sq[i];
4732 
4733 		stats_base = (const u8 *)&rq->stats;
4734 		do {
4735 			start = u64_stats_fetch_begin(&rq->stats.syncp);
4736 			virtnet_fill_stats(vi, i * 2, &ctx, stats_base, true, 0);
4737 		} while (u64_stats_fetch_retry(&rq->stats.syncp, start));
4738 
4739 		stats_base = (const u8 *)&sq->stats;
4740 		do {
4741 			start = u64_stats_fetch_begin(&sq->stats.syncp);
4742 			virtnet_fill_stats(vi, i * 2 + 1, &ctx, stats_base, true, 0);
4743 		} while (u64_stats_fetch_retry(&sq->stats.syncp, start));
4744 	}
4745 
4746 	virtnet_fill_total_fields(vi, &ctx);
4747 }
4748 
4749 static void virtnet_get_channels(struct net_device *dev,
4750 				 struct ethtool_channels *channels)
4751 {
4752 	struct virtnet_info *vi = netdev_priv(dev);
4753 
4754 	channels->combined_count = vi->curr_queue_pairs;
4755 	channels->max_combined = vi->max_queue_pairs;
4756 	channels->max_other = 0;
4757 	channels->rx_count = 0;
4758 	channels->tx_count = 0;
4759 	channels->other_count = 0;
4760 }
4761 
4762 static int virtnet_set_link_ksettings(struct net_device *dev,
4763 				      const struct ethtool_link_ksettings *cmd)
4764 {
4765 	struct virtnet_info *vi = netdev_priv(dev);
4766 
4767 	return ethtool_virtdev_set_link_ksettings(dev, cmd,
4768 						  &vi->speed, &vi->duplex);
4769 }
4770 
4771 static int virtnet_get_link_ksettings(struct net_device *dev,
4772 				      struct ethtool_link_ksettings *cmd)
4773 {
4774 	struct virtnet_info *vi = netdev_priv(dev);
4775 
4776 	cmd->base.speed = vi->speed;
4777 	cmd->base.duplex = vi->duplex;
4778 	cmd->base.port = PORT_OTHER;
4779 
4780 	return 0;
4781 }
4782 
4783 static int virtnet_send_tx_notf_coal_cmds(struct virtnet_info *vi,
4784 					  struct ethtool_coalesce *ec)
4785 {
4786 	struct virtio_net_ctrl_coal_tx *coal_tx __free(kfree) = NULL;
4787 	struct scatterlist sgs_tx;
4788 	int i;
4789 
4790 	coal_tx = kzalloc(sizeof(*coal_tx), GFP_KERNEL);
4791 	if (!coal_tx)
4792 		return -ENOMEM;
4793 
4794 	coal_tx->tx_usecs = cpu_to_le32(ec->tx_coalesce_usecs);
4795 	coal_tx->tx_max_packets = cpu_to_le32(ec->tx_max_coalesced_frames);
4796 	sg_init_one(&sgs_tx, coal_tx, sizeof(*coal_tx));
4797 
4798 	if (!virtnet_send_command(vi, VIRTIO_NET_CTRL_NOTF_COAL,
4799 				  VIRTIO_NET_CTRL_NOTF_COAL_TX_SET,
4800 				  &sgs_tx))
4801 		return -EINVAL;
4802 
4803 	vi->intr_coal_tx.max_usecs = ec->tx_coalesce_usecs;
4804 	vi->intr_coal_tx.max_packets = ec->tx_max_coalesced_frames;
4805 	for (i = 0; i < vi->max_queue_pairs; i++) {
4806 		vi->sq[i].intr_coal.max_usecs = ec->tx_coalesce_usecs;
4807 		vi->sq[i].intr_coal.max_packets = ec->tx_max_coalesced_frames;
4808 	}
4809 
4810 	return 0;
4811 }
4812 
4813 static int virtnet_send_rx_notf_coal_cmds(struct virtnet_info *vi,
4814 					  struct ethtool_coalesce *ec)
4815 {
4816 	struct virtio_net_ctrl_coal_rx *coal_rx __free(kfree) = NULL;
4817 	bool rx_ctrl_dim_on = !!ec->use_adaptive_rx_coalesce;
4818 	struct scatterlist sgs_rx;
4819 	int i;
4820 
4821 	if (rx_ctrl_dim_on && !virtio_has_feature(vi->vdev, VIRTIO_NET_F_VQ_NOTF_COAL))
4822 		return -EOPNOTSUPP;
4823 
4824 	if (rx_ctrl_dim_on && (ec->rx_coalesce_usecs != vi->intr_coal_rx.max_usecs ||
4825 			       ec->rx_max_coalesced_frames != vi->intr_coal_rx.max_packets))
4826 		return -EINVAL;
4827 
4828 	if (rx_ctrl_dim_on && !vi->rx_dim_enabled) {
4829 		vi->rx_dim_enabled = true;
4830 		for (i = 0; i < vi->max_queue_pairs; i++) {
4831 			mutex_lock(&vi->rq[i].dim_lock);
4832 			vi->rq[i].dim_enabled = true;
4833 			mutex_unlock(&vi->rq[i].dim_lock);
4834 		}
4835 		return 0;
4836 	}
4837 
4838 	coal_rx = kzalloc(sizeof(*coal_rx), GFP_KERNEL);
4839 	if (!coal_rx)
4840 		return -ENOMEM;
4841 
4842 	if (!rx_ctrl_dim_on && vi->rx_dim_enabled) {
4843 		vi->rx_dim_enabled = false;
4844 		for (i = 0; i < vi->max_queue_pairs; i++) {
4845 			mutex_lock(&vi->rq[i].dim_lock);
4846 			vi->rq[i].dim_enabled = false;
4847 			mutex_unlock(&vi->rq[i].dim_lock);
4848 		}
4849 	}
4850 
4851 	/* Since the per-queue coalescing params can be set,
4852 	 * we need apply the global new params even if they
4853 	 * are not updated.
4854 	 */
4855 	coal_rx->rx_usecs = cpu_to_le32(ec->rx_coalesce_usecs);
4856 	coal_rx->rx_max_packets = cpu_to_le32(ec->rx_max_coalesced_frames);
4857 	sg_init_one(&sgs_rx, coal_rx, sizeof(*coal_rx));
4858 
4859 	if (!virtnet_send_command(vi, VIRTIO_NET_CTRL_NOTF_COAL,
4860 				  VIRTIO_NET_CTRL_NOTF_COAL_RX_SET,
4861 				  &sgs_rx))
4862 		return -EINVAL;
4863 
4864 	vi->intr_coal_rx.max_usecs = ec->rx_coalesce_usecs;
4865 	vi->intr_coal_rx.max_packets = ec->rx_max_coalesced_frames;
4866 	for (i = 0; i < vi->max_queue_pairs; i++) {
4867 		mutex_lock(&vi->rq[i].dim_lock);
4868 		vi->rq[i].intr_coal.max_usecs = ec->rx_coalesce_usecs;
4869 		vi->rq[i].intr_coal.max_packets = ec->rx_max_coalesced_frames;
4870 		mutex_unlock(&vi->rq[i].dim_lock);
4871 	}
4872 
4873 	return 0;
4874 }
4875 
4876 static int virtnet_send_notf_coal_cmds(struct virtnet_info *vi,
4877 				       struct ethtool_coalesce *ec)
4878 {
4879 	int err;
4880 
4881 	err = virtnet_send_tx_notf_coal_cmds(vi, ec);
4882 	if (err)
4883 		return err;
4884 
4885 	err = virtnet_send_rx_notf_coal_cmds(vi, ec);
4886 	if (err)
4887 		return err;
4888 
4889 	return 0;
4890 }
4891 
4892 static int virtnet_send_rx_notf_coal_vq_cmds(struct virtnet_info *vi,
4893 					     struct ethtool_coalesce *ec,
4894 					     u16 queue)
4895 {
4896 	bool rx_ctrl_dim_on = !!ec->use_adaptive_rx_coalesce;
4897 	u32 max_usecs, max_packets;
4898 	bool cur_rx_dim;
4899 	int err;
4900 
4901 	mutex_lock(&vi->rq[queue].dim_lock);
4902 	cur_rx_dim = vi->rq[queue].dim_enabled;
4903 	max_usecs = vi->rq[queue].intr_coal.max_usecs;
4904 	max_packets = vi->rq[queue].intr_coal.max_packets;
4905 
4906 	if (rx_ctrl_dim_on && (ec->rx_coalesce_usecs != max_usecs ||
4907 			       ec->rx_max_coalesced_frames != max_packets)) {
4908 		mutex_unlock(&vi->rq[queue].dim_lock);
4909 		return -EINVAL;
4910 	}
4911 
4912 	if (rx_ctrl_dim_on && !cur_rx_dim) {
4913 		vi->rq[queue].dim_enabled = true;
4914 		mutex_unlock(&vi->rq[queue].dim_lock);
4915 		return 0;
4916 	}
4917 
4918 	if (!rx_ctrl_dim_on && cur_rx_dim)
4919 		vi->rq[queue].dim_enabled = false;
4920 
4921 	/* If no params are updated, userspace ethtool will
4922 	 * reject the modification.
4923 	 */
4924 	err = virtnet_send_rx_ctrl_coal_vq_cmd(vi, queue,
4925 					       ec->rx_coalesce_usecs,
4926 					       ec->rx_max_coalesced_frames);
4927 	mutex_unlock(&vi->rq[queue].dim_lock);
4928 	return err;
4929 }
4930 
4931 static int virtnet_send_notf_coal_vq_cmds(struct virtnet_info *vi,
4932 					  struct ethtool_coalesce *ec,
4933 					  u16 queue)
4934 {
4935 	int err;
4936 
4937 	err = virtnet_send_rx_notf_coal_vq_cmds(vi, ec, queue);
4938 	if (err)
4939 		return err;
4940 
4941 	err = virtnet_send_tx_ctrl_coal_vq_cmd(vi, queue,
4942 					       ec->tx_coalesce_usecs,
4943 					       ec->tx_max_coalesced_frames);
4944 	if (err)
4945 		return err;
4946 
4947 	return 0;
4948 }
4949 
4950 static void virtnet_rx_dim_work(struct work_struct *work)
4951 {
4952 	struct dim *dim = container_of(work, struct dim, work);
4953 	struct receive_queue *rq = container_of(dim,
4954 			struct receive_queue, dim);
4955 	struct virtnet_info *vi = rq->vq->vdev->priv;
4956 	struct net_device *dev = vi->dev;
4957 	struct dim_cq_moder update_moder;
4958 	int qnum, err;
4959 
4960 	qnum = rq - vi->rq;
4961 
4962 	mutex_lock(&rq->dim_lock);
4963 	if (!rq->dim_enabled)
4964 		goto out;
4965 
4966 	update_moder = net_dim_get_rx_irq_moder(dev, dim);
4967 	if (update_moder.usec != rq->intr_coal.max_usecs ||
4968 	    update_moder.pkts != rq->intr_coal.max_packets) {
4969 		err = virtnet_send_rx_ctrl_coal_vq_cmd(vi, qnum,
4970 						       update_moder.usec,
4971 						       update_moder.pkts);
4972 		if (err)
4973 			pr_debug("%s: Failed to send dim parameters on rxq%d\n",
4974 				 dev->name, qnum);
4975 	}
4976 out:
4977 	dim->state = DIM_START_MEASURE;
4978 	mutex_unlock(&rq->dim_lock);
4979 }
4980 
4981 static int virtnet_coal_params_supported(struct ethtool_coalesce *ec)
4982 {
4983 	/* usecs coalescing is supported only if VIRTIO_NET_F_NOTF_COAL
4984 	 * or VIRTIO_NET_F_VQ_NOTF_COAL feature is negotiated.
4985 	 */
4986 	if (ec->rx_coalesce_usecs || ec->tx_coalesce_usecs)
4987 		return -EOPNOTSUPP;
4988 
4989 	if (ec->tx_max_coalesced_frames > 1 ||
4990 	    ec->rx_max_coalesced_frames != 1)
4991 		return -EINVAL;
4992 
4993 	return 0;
4994 }
4995 
4996 static int virtnet_should_update_vq_weight(int dev_flags, int weight,
4997 					   int vq_weight, bool *should_update)
4998 {
4999 	if (weight ^ vq_weight) {
5000 		if (dev_flags & IFF_UP)
5001 			return -EBUSY;
5002 		*should_update = true;
5003 	}
5004 
5005 	return 0;
5006 }
5007 
5008 static int virtnet_set_coalesce(struct net_device *dev,
5009 				struct ethtool_coalesce *ec,
5010 				struct kernel_ethtool_coalesce *kernel_coal,
5011 				struct netlink_ext_ack *extack)
5012 {
5013 	struct virtnet_info *vi = netdev_priv(dev);
5014 	int ret, queue_number, napi_weight;
5015 	bool update_napi = false;
5016 
5017 	/* Can't change NAPI weight if the link is up */
5018 	napi_weight = ec->tx_max_coalesced_frames ? NAPI_POLL_WEIGHT : 0;
5019 	for (queue_number = 0; queue_number < vi->max_queue_pairs; queue_number++) {
5020 		ret = virtnet_should_update_vq_weight(dev->flags, napi_weight,
5021 						      vi->sq[queue_number].napi.weight,
5022 						      &update_napi);
5023 		if (ret)
5024 			return ret;
5025 
5026 		if (update_napi) {
5027 			/* All queues that belong to [queue_number, vi->max_queue_pairs] will be
5028 			 * updated for the sake of simplicity, which might not be necessary
5029 			 */
5030 			break;
5031 		}
5032 	}
5033 
5034 	if (virtio_has_feature(vi->vdev, VIRTIO_NET_F_NOTF_COAL))
5035 		ret = virtnet_send_notf_coal_cmds(vi, ec);
5036 	else
5037 		ret = virtnet_coal_params_supported(ec);
5038 
5039 	if (ret)
5040 		return ret;
5041 
5042 	if (update_napi) {
5043 		for (; queue_number < vi->max_queue_pairs; queue_number++)
5044 			vi->sq[queue_number].napi.weight = napi_weight;
5045 	}
5046 
5047 	return ret;
5048 }
5049 
5050 static int virtnet_get_coalesce(struct net_device *dev,
5051 				struct ethtool_coalesce *ec,
5052 				struct kernel_ethtool_coalesce *kernel_coal,
5053 				struct netlink_ext_ack *extack)
5054 {
5055 	struct virtnet_info *vi = netdev_priv(dev);
5056 
5057 	if (virtio_has_feature(vi->vdev, VIRTIO_NET_F_NOTF_COAL)) {
5058 		ec->rx_coalesce_usecs = vi->intr_coal_rx.max_usecs;
5059 		ec->tx_coalesce_usecs = vi->intr_coal_tx.max_usecs;
5060 		ec->tx_max_coalesced_frames = vi->intr_coal_tx.max_packets;
5061 		ec->rx_max_coalesced_frames = vi->intr_coal_rx.max_packets;
5062 		ec->use_adaptive_rx_coalesce = vi->rx_dim_enabled;
5063 	} else {
5064 		ec->rx_max_coalesced_frames = 1;
5065 
5066 		if (vi->sq[0].napi.weight)
5067 			ec->tx_max_coalesced_frames = 1;
5068 	}
5069 
5070 	return 0;
5071 }
5072 
5073 static int virtnet_set_per_queue_coalesce(struct net_device *dev,
5074 					  u32 queue,
5075 					  struct ethtool_coalesce *ec)
5076 {
5077 	struct virtnet_info *vi = netdev_priv(dev);
5078 	int ret, napi_weight;
5079 	bool update_napi = false;
5080 
5081 	if (queue >= vi->max_queue_pairs)
5082 		return -EINVAL;
5083 
5084 	/* Can't change NAPI weight if the link is up */
5085 	napi_weight = ec->tx_max_coalesced_frames ? NAPI_POLL_WEIGHT : 0;
5086 	ret = virtnet_should_update_vq_weight(dev->flags, napi_weight,
5087 					      vi->sq[queue].napi.weight,
5088 					      &update_napi);
5089 	if (ret)
5090 		return ret;
5091 
5092 	if (virtio_has_feature(vi->vdev, VIRTIO_NET_F_VQ_NOTF_COAL))
5093 		ret = virtnet_send_notf_coal_vq_cmds(vi, ec, queue);
5094 	else
5095 		ret = virtnet_coal_params_supported(ec);
5096 
5097 	if (ret)
5098 		return ret;
5099 
5100 	if (update_napi)
5101 		vi->sq[queue].napi.weight = napi_weight;
5102 
5103 	return 0;
5104 }
5105 
5106 static int virtnet_get_per_queue_coalesce(struct net_device *dev,
5107 					  u32 queue,
5108 					  struct ethtool_coalesce *ec)
5109 {
5110 	struct virtnet_info *vi = netdev_priv(dev);
5111 
5112 	if (queue >= vi->max_queue_pairs)
5113 		return -EINVAL;
5114 
5115 	if (virtio_has_feature(vi->vdev, VIRTIO_NET_F_VQ_NOTF_COAL)) {
5116 		mutex_lock(&vi->rq[queue].dim_lock);
5117 		ec->rx_coalesce_usecs = vi->rq[queue].intr_coal.max_usecs;
5118 		ec->tx_coalesce_usecs = vi->sq[queue].intr_coal.max_usecs;
5119 		ec->tx_max_coalesced_frames = vi->sq[queue].intr_coal.max_packets;
5120 		ec->rx_max_coalesced_frames = vi->rq[queue].intr_coal.max_packets;
5121 		ec->use_adaptive_rx_coalesce = vi->rq[queue].dim_enabled;
5122 		mutex_unlock(&vi->rq[queue].dim_lock);
5123 	} else {
5124 		ec->rx_max_coalesced_frames = 1;
5125 
5126 		if (vi->sq[queue].napi.weight)
5127 			ec->tx_max_coalesced_frames = 1;
5128 	}
5129 
5130 	return 0;
5131 }
5132 
5133 static void virtnet_init_settings(struct net_device *dev)
5134 {
5135 	struct virtnet_info *vi = netdev_priv(dev);
5136 
5137 	vi->speed = SPEED_UNKNOWN;
5138 	vi->duplex = DUPLEX_UNKNOWN;
5139 }
5140 
5141 static u32 virtnet_get_rxfh_key_size(struct net_device *dev)
5142 {
5143 	return ((struct virtnet_info *)netdev_priv(dev))->rss_key_size;
5144 }
5145 
5146 static u32 virtnet_get_rxfh_indir_size(struct net_device *dev)
5147 {
5148 	return ((struct virtnet_info *)netdev_priv(dev))->rss_indir_table_size;
5149 }
5150 
5151 static int virtnet_get_rxfh(struct net_device *dev,
5152 			    struct ethtool_rxfh_param *rxfh)
5153 {
5154 	struct virtnet_info *vi = netdev_priv(dev);
5155 	int i;
5156 
5157 	if (rxfh->indir) {
5158 		for (i = 0; i < vi->rss_indir_table_size; ++i)
5159 			rxfh->indir[i] = vi->rss.indirection_table[i];
5160 	}
5161 
5162 	if (rxfh->key)
5163 		memcpy(rxfh->key, vi->rss.key, vi->rss_key_size);
5164 
5165 	rxfh->hfunc = ETH_RSS_HASH_TOP;
5166 
5167 	return 0;
5168 }
5169 
5170 static int virtnet_set_rxfh(struct net_device *dev,
5171 			    struct ethtool_rxfh_param *rxfh,
5172 			    struct netlink_ext_ack *extack)
5173 {
5174 	struct virtnet_info *vi = netdev_priv(dev);
5175 	bool update = false;
5176 	int i;
5177 
5178 	if (rxfh->hfunc != ETH_RSS_HASH_NO_CHANGE &&
5179 	    rxfh->hfunc != ETH_RSS_HASH_TOP)
5180 		return -EOPNOTSUPP;
5181 
5182 	if (rxfh->indir) {
5183 		if (!vi->has_rss)
5184 			return -EOPNOTSUPP;
5185 
5186 		for (i = 0; i < vi->rss_indir_table_size; ++i)
5187 			vi->rss.indirection_table[i] = rxfh->indir[i];
5188 		update = true;
5189 	}
5190 
5191 	if (rxfh->key) {
5192 		/* If either _F_HASH_REPORT or _F_RSS are negotiated, the
5193 		 * device provides hash calculation capabilities, that is,
5194 		 * hash_key is configured.
5195 		 */
5196 		if (!vi->has_rss && !vi->has_rss_hash_report)
5197 			return -EOPNOTSUPP;
5198 
5199 		memcpy(vi->rss.key, rxfh->key, vi->rss_key_size);
5200 		update = true;
5201 	}
5202 
5203 	if (update)
5204 		virtnet_commit_rss_command(vi);
5205 
5206 	return 0;
5207 }
5208 
5209 static int virtnet_get_rxnfc(struct net_device *dev, struct ethtool_rxnfc *info, u32 *rule_locs)
5210 {
5211 	struct virtnet_info *vi = netdev_priv(dev);
5212 	int rc = 0;
5213 
5214 	switch (info->cmd) {
5215 	case ETHTOOL_GRXRINGS:
5216 		info->data = vi->curr_queue_pairs;
5217 		break;
5218 	case ETHTOOL_GRXFH:
5219 		virtnet_get_hashflow(vi, info);
5220 		break;
5221 	default:
5222 		rc = -EOPNOTSUPP;
5223 	}
5224 
5225 	return rc;
5226 }
5227 
5228 static int virtnet_set_rxnfc(struct net_device *dev, struct ethtool_rxnfc *info)
5229 {
5230 	struct virtnet_info *vi = netdev_priv(dev);
5231 	int rc = 0;
5232 
5233 	switch (info->cmd) {
5234 	case ETHTOOL_SRXFH:
5235 		if (!virtnet_set_hashflow(vi, info))
5236 			rc = -EINVAL;
5237 
5238 		break;
5239 	default:
5240 		rc = -EOPNOTSUPP;
5241 	}
5242 
5243 	return rc;
5244 }
5245 
5246 static const struct ethtool_ops virtnet_ethtool_ops = {
5247 	.supported_coalesce_params = ETHTOOL_COALESCE_MAX_FRAMES |
5248 		ETHTOOL_COALESCE_USECS | ETHTOOL_COALESCE_USE_ADAPTIVE_RX,
5249 	.get_drvinfo = virtnet_get_drvinfo,
5250 	.get_link = ethtool_op_get_link,
5251 	.get_ringparam = virtnet_get_ringparam,
5252 	.set_ringparam = virtnet_set_ringparam,
5253 	.get_strings = virtnet_get_strings,
5254 	.get_sset_count = virtnet_get_sset_count,
5255 	.get_ethtool_stats = virtnet_get_ethtool_stats,
5256 	.set_channels = virtnet_set_channels,
5257 	.get_channels = virtnet_get_channels,
5258 	.get_ts_info = ethtool_op_get_ts_info,
5259 	.get_link_ksettings = virtnet_get_link_ksettings,
5260 	.set_link_ksettings = virtnet_set_link_ksettings,
5261 	.set_coalesce = virtnet_set_coalesce,
5262 	.get_coalesce = virtnet_get_coalesce,
5263 	.set_per_queue_coalesce = virtnet_set_per_queue_coalesce,
5264 	.get_per_queue_coalesce = virtnet_get_per_queue_coalesce,
5265 	.get_rxfh_key_size = virtnet_get_rxfh_key_size,
5266 	.get_rxfh_indir_size = virtnet_get_rxfh_indir_size,
5267 	.get_rxfh = virtnet_get_rxfh,
5268 	.set_rxfh = virtnet_set_rxfh,
5269 	.get_rxnfc = virtnet_get_rxnfc,
5270 	.set_rxnfc = virtnet_set_rxnfc,
5271 };
5272 
5273 static void virtnet_get_queue_stats_rx(struct net_device *dev, int i,
5274 				       struct netdev_queue_stats_rx *stats)
5275 {
5276 	struct virtnet_info *vi = netdev_priv(dev);
5277 	struct receive_queue *rq = &vi->rq[i];
5278 	struct virtnet_stats_ctx ctx = {0};
5279 
5280 	virtnet_stats_ctx_init(vi, &ctx, (void *)stats, true);
5281 
5282 	virtnet_get_hw_stats(vi, &ctx, i * 2);
5283 	virtnet_fill_stats(vi, i * 2, &ctx, (void *)&rq->stats, true, 0);
5284 }
5285 
5286 static void virtnet_get_queue_stats_tx(struct net_device *dev, int i,
5287 				       struct netdev_queue_stats_tx *stats)
5288 {
5289 	struct virtnet_info *vi = netdev_priv(dev);
5290 	struct send_queue *sq = &vi->sq[i];
5291 	struct virtnet_stats_ctx ctx = {0};
5292 
5293 	virtnet_stats_ctx_init(vi, &ctx, (void *)stats, true);
5294 
5295 	virtnet_get_hw_stats(vi, &ctx, i * 2 + 1);
5296 	virtnet_fill_stats(vi, i * 2 + 1, &ctx, (void *)&sq->stats, true, 0);
5297 }
5298 
5299 static void virtnet_get_base_stats(struct net_device *dev,
5300 				   struct netdev_queue_stats_rx *rx,
5301 				   struct netdev_queue_stats_tx *tx)
5302 {
5303 	struct virtnet_info *vi = netdev_priv(dev);
5304 
5305 	/* The queue stats of the virtio-net will not be reset. So here we
5306 	 * return 0.
5307 	 */
5308 	rx->bytes = 0;
5309 	rx->packets = 0;
5310 
5311 	if (vi->device_stats_cap & VIRTIO_NET_STATS_TYPE_RX_BASIC) {
5312 		rx->hw_drops = 0;
5313 		rx->hw_drop_overruns = 0;
5314 	}
5315 
5316 	if (vi->device_stats_cap & VIRTIO_NET_STATS_TYPE_RX_CSUM) {
5317 		rx->csum_unnecessary = 0;
5318 		rx->csum_none = 0;
5319 		rx->csum_bad = 0;
5320 	}
5321 
5322 	if (vi->device_stats_cap & VIRTIO_NET_STATS_TYPE_RX_GSO) {
5323 		rx->hw_gro_packets = 0;
5324 		rx->hw_gro_bytes = 0;
5325 		rx->hw_gro_wire_packets = 0;
5326 		rx->hw_gro_wire_bytes = 0;
5327 	}
5328 
5329 	if (vi->device_stats_cap & VIRTIO_NET_STATS_TYPE_RX_SPEED)
5330 		rx->hw_drop_ratelimits = 0;
5331 
5332 	tx->bytes = 0;
5333 	tx->packets = 0;
5334 	tx->stop = 0;
5335 	tx->wake = 0;
5336 
5337 	if (vi->device_stats_cap & VIRTIO_NET_STATS_TYPE_TX_BASIC) {
5338 		tx->hw_drops = 0;
5339 		tx->hw_drop_errors = 0;
5340 	}
5341 
5342 	if (vi->device_stats_cap & VIRTIO_NET_STATS_TYPE_TX_CSUM) {
5343 		tx->csum_none = 0;
5344 		tx->needs_csum = 0;
5345 	}
5346 
5347 	if (vi->device_stats_cap & VIRTIO_NET_STATS_TYPE_TX_GSO) {
5348 		tx->hw_gso_packets = 0;
5349 		tx->hw_gso_bytes = 0;
5350 		tx->hw_gso_wire_packets = 0;
5351 		tx->hw_gso_wire_bytes = 0;
5352 	}
5353 
5354 	if (vi->device_stats_cap & VIRTIO_NET_STATS_TYPE_TX_SPEED)
5355 		tx->hw_drop_ratelimits = 0;
5356 }
5357 
5358 static const struct netdev_stat_ops virtnet_stat_ops = {
5359 	.get_queue_stats_rx	= virtnet_get_queue_stats_rx,
5360 	.get_queue_stats_tx	= virtnet_get_queue_stats_tx,
5361 	.get_base_stats		= virtnet_get_base_stats,
5362 };
5363 
5364 static void virtnet_freeze_down(struct virtio_device *vdev)
5365 {
5366 	struct virtnet_info *vi = vdev->priv;
5367 
5368 	/* Make sure no work handler is accessing the device */
5369 	flush_work(&vi->config_work);
5370 	disable_rx_mode_work(vi);
5371 	flush_work(&vi->rx_mode_work);
5372 
5373 	netif_tx_lock_bh(vi->dev);
5374 	netif_device_detach(vi->dev);
5375 	netif_tx_unlock_bh(vi->dev);
5376 	if (netif_running(vi->dev))
5377 		virtnet_close(vi->dev);
5378 }
5379 
5380 static int init_vqs(struct virtnet_info *vi);
5381 
5382 static int virtnet_restore_up(struct virtio_device *vdev)
5383 {
5384 	struct virtnet_info *vi = vdev->priv;
5385 	int err;
5386 
5387 	err = init_vqs(vi);
5388 	if (err)
5389 		return err;
5390 
5391 	virtio_device_ready(vdev);
5392 
5393 	enable_delayed_refill(vi);
5394 	enable_rx_mode_work(vi);
5395 
5396 	if (netif_running(vi->dev)) {
5397 		err = virtnet_open(vi->dev);
5398 		if (err)
5399 			return err;
5400 	}
5401 
5402 	netif_tx_lock_bh(vi->dev);
5403 	netif_device_attach(vi->dev);
5404 	netif_tx_unlock_bh(vi->dev);
5405 	return err;
5406 }
5407 
5408 static int virtnet_set_guest_offloads(struct virtnet_info *vi, u64 offloads)
5409 {
5410 	__virtio64 *_offloads __free(kfree) = NULL;
5411 	struct scatterlist sg;
5412 
5413 	_offloads = kzalloc(sizeof(*_offloads), GFP_KERNEL);
5414 	if (!_offloads)
5415 		return -ENOMEM;
5416 
5417 	*_offloads = cpu_to_virtio64(vi->vdev, offloads);
5418 
5419 	sg_init_one(&sg, _offloads, sizeof(*_offloads));
5420 
5421 	if (!virtnet_send_command(vi, VIRTIO_NET_CTRL_GUEST_OFFLOADS,
5422 				  VIRTIO_NET_CTRL_GUEST_OFFLOADS_SET, &sg)) {
5423 		dev_warn(&vi->dev->dev, "Fail to set guest offload.\n");
5424 		return -EINVAL;
5425 	}
5426 
5427 	return 0;
5428 }
5429 
5430 static int virtnet_clear_guest_offloads(struct virtnet_info *vi)
5431 {
5432 	u64 offloads = 0;
5433 
5434 	if (!vi->guest_offloads)
5435 		return 0;
5436 
5437 	return virtnet_set_guest_offloads(vi, offloads);
5438 }
5439 
5440 static int virtnet_restore_guest_offloads(struct virtnet_info *vi)
5441 {
5442 	u64 offloads = vi->guest_offloads;
5443 
5444 	if (!vi->guest_offloads)
5445 		return 0;
5446 
5447 	return virtnet_set_guest_offloads(vi, offloads);
5448 }
5449 
5450 static int virtnet_rq_bind_xsk_pool(struct virtnet_info *vi, struct receive_queue *rq,
5451 				    struct xsk_buff_pool *pool)
5452 {
5453 	int err, qindex;
5454 
5455 	qindex = rq - vi->rq;
5456 
5457 	if (pool) {
5458 		err = xdp_rxq_info_reg(&rq->xsk_rxq_info, vi->dev, qindex, rq->napi.napi_id);
5459 		if (err < 0)
5460 			return err;
5461 
5462 		err = xdp_rxq_info_reg_mem_model(&rq->xsk_rxq_info,
5463 						 MEM_TYPE_XSK_BUFF_POOL, NULL);
5464 		if (err < 0)
5465 			goto unreg;
5466 
5467 		xsk_pool_set_rxq_info(pool, &rq->xsk_rxq_info);
5468 	}
5469 
5470 	virtnet_rx_pause(vi, rq);
5471 
5472 	err = virtqueue_reset(rq->vq, virtnet_rq_unmap_free_buf);
5473 	if (err) {
5474 		netdev_err(vi->dev, "reset rx fail: rx queue index: %d err: %d\n", qindex, err);
5475 
5476 		pool = NULL;
5477 	}
5478 
5479 	rq->xsk_pool = pool;
5480 
5481 	virtnet_rx_resume(vi, rq);
5482 
5483 	if (pool)
5484 		return 0;
5485 
5486 unreg:
5487 	xdp_rxq_info_unreg(&rq->xsk_rxq_info);
5488 	return err;
5489 }
5490 
5491 static int virtnet_xsk_pool_enable(struct net_device *dev,
5492 				   struct xsk_buff_pool *pool,
5493 				   u16 qid)
5494 {
5495 	struct virtnet_info *vi = netdev_priv(dev);
5496 	struct receive_queue *rq;
5497 	struct device *dma_dev;
5498 	struct send_queue *sq;
5499 	int err, size;
5500 
5501 	if (vi->hdr_len > xsk_pool_get_headroom(pool))
5502 		return -EINVAL;
5503 
5504 	/* In big_packets mode, xdp cannot work, so there is no need to
5505 	 * initialize xsk of rq.
5506 	 */
5507 	if (vi->big_packets && !vi->mergeable_rx_bufs)
5508 		return -ENOENT;
5509 
5510 	if (qid >= vi->curr_queue_pairs)
5511 		return -EINVAL;
5512 
5513 	sq = &vi->sq[qid];
5514 	rq = &vi->rq[qid];
5515 
5516 	/* xsk assumes that tx and rx must have the same dma device. The af-xdp
5517 	 * may use one buffer to receive from the rx and reuse this buffer to
5518 	 * send by the tx. So the dma dev of sq and rq must be the same one.
5519 	 *
5520 	 * But vq->dma_dev allows every vq has the respective dma dev. So I
5521 	 * check the dma dev of vq and sq is the same dev.
5522 	 */
5523 	if (virtqueue_dma_dev(rq->vq) != virtqueue_dma_dev(sq->vq))
5524 		return -EINVAL;
5525 
5526 	dma_dev = virtqueue_dma_dev(rq->vq);
5527 	if (!dma_dev)
5528 		return -EINVAL;
5529 
5530 	size = virtqueue_get_vring_size(rq->vq);
5531 
5532 	rq->xsk_buffs = kvcalloc(size, sizeof(*rq->xsk_buffs), GFP_KERNEL);
5533 	if (!rq->xsk_buffs)
5534 		return -ENOMEM;
5535 
5536 	err = xsk_pool_dma_map(pool, dma_dev, 0);
5537 	if (err)
5538 		goto err_xsk_map;
5539 
5540 	err = virtnet_rq_bind_xsk_pool(vi, rq, pool);
5541 	if (err)
5542 		goto err_rq;
5543 
5544 	return 0;
5545 
5546 err_rq:
5547 	xsk_pool_dma_unmap(pool, 0);
5548 err_xsk_map:
5549 	return err;
5550 }
5551 
5552 static int virtnet_xsk_pool_disable(struct net_device *dev, u16 qid)
5553 {
5554 	struct virtnet_info *vi = netdev_priv(dev);
5555 	struct xsk_buff_pool *pool;
5556 	struct receive_queue *rq;
5557 	int err;
5558 
5559 	if (qid >= vi->curr_queue_pairs)
5560 		return -EINVAL;
5561 
5562 	rq = &vi->rq[qid];
5563 
5564 	pool = rq->xsk_pool;
5565 
5566 	err = virtnet_rq_bind_xsk_pool(vi, rq, NULL);
5567 
5568 	xsk_pool_dma_unmap(pool, 0);
5569 
5570 	kvfree(rq->xsk_buffs);
5571 
5572 	return err;
5573 }
5574 
5575 static int virtnet_xsk_pool_setup(struct net_device *dev, struct netdev_bpf *xdp)
5576 {
5577 	if (xdp->xsk.pool)
5578 		return virtnet_xsk_pool_enable(dev, xdp->xsk.pool,
5579 					       xdp->xsk.queue_id);
5580 	else
5581 		return virtnet_xsk_pool_disable(dev, xdp->xsk.queue_id);
5582 }
5583 
5584 static int virtnet_xdp_set(struct net_device *dev, struct bpf_prog *prog,
5585 			   struct netlink_ext_ack *extack)
5586 {
5587 	unsigned int room = SKB_DATA_ALIGN(XDP_PACKET_HEADROOM +
5588 					   sizeof(struct skb_shared_info));
5589 	unsigned int max_sz = PAGE_SIZE - room - ETH_HLEN;
5590 	struct virtnet_info *vi = netdev_priv(dev);
5591 	struct bpf_prog *old_prog;
5592 	u16 xdp_qp = 0, curr_qp;
5593 	int i, err;
5594 
5595 	if (!virtio_has_feature(vi->vdev, VIRTIO_NET_F_CTRL_GUEST_OFFLOADS)
5596 	    && (virtio_has_feature(vi->vdev, VIRTIO_NET_F_GUEST_TSO4) ||
5597 	        virtio_has_feature(vi->vdev, VIRTIO_NET_F_GUEST_TSO6) ||
5598 	        virtio_has_feature(vi->vdev, VIRTIO_NET_F_GUEST_ECN) ||
5599 		virtio_has_feature(vi->vdev, VIRTIO_NET_F_GUEST_UFO) ||
5600 		virtio_has_feature(vi->vdev, VIRTIO_NET_F_GUEST_CSUM) ||
5601 		virtio_has_feature(vi->vdev, VIRTIO_NET_F_GUEST_USO4) ||
5602 		virtio_has_feature(vi->vdev, VIRTIO_NET_F_GUEST_USO6))) {
5603 		NL_SET_ERR_MSG_MOD(extack, "Can't set XDP while host is implementing GRO_HW/CSUM, disable GRO_HW/CSUM first");
5604 		return -EOPNOTSUPP;
5605 	}
5606 
5607 	if (vi->mergeable_rx_bufs && !vi->any_header_sg) {
5608 		NL_SET_ERR_MSG_MOD(extack, "XDP expects header/data in single page, any_header_sg required");
5609 		return -EINVAL;
5610 	}
5611 
5612 	if (prog && !prog->aux->xdp_has_frags && dev->mtu > max_sz) {
5613 		NL_SET_ERR_MSG_MOD(extack, "MTU too large to enable XDP without frags");
5614 		netdev_warn(dev, "single-buffer XDP requires MTU less than %u\n", max_sz);
5615 		return -EINVAL;
5616 	}
5617 
5618 	curr_qp = vi->curr_queue_pairs - vi->xdp_queue_pairs;
5619 	if (prog)
5620 		xdp_qp = nr_cpu_ids;
5621 
5622 	/* XDP requires extra queues for XDP_TX */
5623 	if (curr_qp + xdp_qp > vi->max_queue_pairs) {
5624 		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",
5625 				 curr_qp + xdp_qp, vi->max_queue_pairs);
5626 		xdp_qp = 0;
5627 	}
5628 
5629 	old_prog = rtnl_dereference(vi->rq[0].xdp_prog);
5630 	if (!prog && !old_prog)
5631 		return 0;
5632 
5633 	if (prog)
5634 		bpf_prog_add(prog, vi->max_queue_pairs - 1);
5635 
5636 	/* Make sure NAPI is not using any XDP TX queues for RX. */
5637 	if (netif_running(dev)) {
5638 		for (i = 0; i < vi->max_queue_pairs; i++) {
5639 			napi_disable(&vi->rq[i].napi);
5640 			virtnet_napi_tx_disable(&vi->sq[i].napi);
5641 		}
5642 	}
5643 
5644 	if (!prog) {
5645 		for (i = 0; i < vi->max_queue_pairs; i++) {
5646 			rcu_assign_pointer(vi->rq[i].xdp_prog, prog);
5647 			if (i == 0)
5648 				virtnet_restore_guest_offloads(vi);
5649 		}
5650 		synchronize_net();
5651 	}
5652 
5653 	err = virtnet_set_queues(vi, curr_qp + xdp_qp);
5654 	if (err)
5655 		goto err;
5656 	netif_set_real_num_rx_queues(dev, curr_qp + xdp_qp);
5657 	vi->xdp_queue_pairs = xdp_qp;
5658 
5659 	if (prog) {
5660 		vi->xdp_enabled = true;
5661 		for (i = 0; i < vi->max_queue_pairs; i++) {
5662 			rcu_assign_pointer(vi->rq[i].xdp_prog, prog);
5663 			if (i == 0 && !old_prog)
5664 				virtnet_clear_guest_offloads(vi);
5665 		}
5666 		if (!old_prog)
5667 			xdp_features_set_redirect_target(dev, true);
5668 	} else {
5669 		xdp_features_clear_redirect_target(dev);
5670 		vi->xdp_enabled = false;
5671 	}
5672 
5673 	for (i = 0; i < vi->max_queue_pairs; i++) {
5674 		if (old_prog)
5675 			bpf_prog_put(old_prog);
5676 		if (netif_running(dev)) {
5677 			virtnet_napi_enable(vi->rq[i].vq, &vi->rq[i].napi);
5678 			virtnet_napi_tx_enable(vi, vi->sq[i].vq,
5679 					       &vi->sq[i].napi);
5680 		}
5681 	}
5682 
5683 	return 0;
5684 
5685 err:
5686 	if (!prog) {
5687 		virtnet_clear_guest_offloads(vi);
5688 		for (i = 0; i < vi->max_queue_pairs; i++)
5689 			rcu_assign_pointer(vi->rq[i].xdp_prog, old_prog);
5690 	}
5691 
5692 	if (netif_running(dev)) {
5693 		for (i = 0; i < vi->max_queue_pairs; i++) {
5694 			virtnet_napi_enable(vi->rq[i].vq, &vi->rq[i].napi);
5695 			virtnet_napi_tx_enable(vi, vi->sq[i].vq,
5696 					       &vi->sq[i].napi);
5697 		}
5698 	}
5699 	if (prog)
5700 		bpf_prog_sub(prog, vi->max_queue_pairs - 1);
5701 	return err;
5702 }
5703 
5704 static int virtnet_xdp(struct net_device *dev, struct netdev_bpf *xdp)
5705 {
5706 	switch (xdp->command) {
5707 	case XDP_SETUP_PROG:
5708 		return virtnet_xdp_set(dev, xdp->prog, xdp->extack);
5709 	case XDP_SETUP_XSK_POOL:
5710 		return virtnet_xsk_pool_setup(dev, xdp);
5711 	default:
5712 		return -EINVAL;
5713 	}
5714 }
5715 
5716 static int virtnet_get_phys_port_name(struct net_device *dev, char *buf,
5717 				      size_t len)
5718 {
5719 	struct virtnet_info *vi = netdev_priv(dev);
5720 	int ret;
5721 
5722 	if (!virtio_has_feature(vi->vdev, VIRTIO_NET_F_STANDBY))
5723 		return -EOPNOTSUPP;
5724 
5725 	ret = snprintf(buf, len, "sby");
5726 	if (ret >= len)
5727 		return -EOPNOTSUPP;
5728 
5729 	return 0;
5730 }
5731 
5732 static int virtnet_set_features(struct net_device *dev,
5733 				netdev_features_t features)
5734 {
5735 	struct virtnet_info *vi = netdev_priv(dev);
5736 	u64 offloads;
5737 	int err;
5738 
5739 	if ((dev->features ^ features) & NETIF_F_GRO_HW) {
5740 		if (vi->xdp_enabled)
5741 			return -EBUSY;
5742 
5743 		if (features & NETIF_F_GRO_HW)
5744 			offloads = vi->guest_offloads_capable;
5745 		else
5746 			offloads = vi->guest_offloads_capable &
5747 				   ~GUEST_OFFLOAD_GRO_HW_MASK;
5748 
5749 		err = virtnet_set_guest_offloads(vi, offloads);
5750 		if (err)
5751 			return err;
5752 		vi->guest_offloads = offloads;
5753 	}
5754 
5755 	if ((dev->features ^ features) & NETIF_F_RXHASH) {
5756 		if (features & NETIF_F_RXHASH)
5757 			vi->rss.hash_types = vi->rss_hash_types_saved;
5758 		else
5759 			vi->rss.hash_types = VIRTIO_NET_HASH_REPORT_NONE;
5760 
5761 		if (!virtnet_commit_rss_command(vi))
5762 			return -EINVAL;
5763 	}
5764 
5765 	return 0;
5766 }
5767 
5768 static void virtnet_tx_timeout(struct net_device *dev, unsigned int txqueue)
5769 {
5770 	struct virtnet_info *priv = netdev_priv(dev);
5771 	struct send_queue *sq = &priv->sq[txqueue];
5772 	struct netdev_queue *txq = netdev_get_tx_queue(dev, txqueue);
5773 
5774 	u64_stats_update_begin(&sq->stats.syncp);
5775 	u64_stats_inc(&sq->stats.tx_timeouts);
5776 	u64_stats_update_end(&sq->stats.syncp);
5777 
5778 	netdev_err(dev, "TX timeout on queue: %u, sq: %s, vq: 0x%x, name: %s, %u usecs ago\n",
5779 		   txqueue, sq->name, sq->vq->index, sq->vq->name,
5780 		   jiffies_to_usecs(jiffies - READ_ONCE(txq->trans_start)));
5781 }
5782 
5783 static int virtnet_init_irq_moder(struct virtnet_info *vi)
5784 {
5785 	u8 profile_flags = 0, coal_flags = 0;
5786 	int ret, i;
5787 
5788 	profile_flags |= DIM_PROFILE_RX;
5789 	coal_flags |= DIM_COALESCE_USEC | DIM_COALESCE_PKTS;
5790 	ret = net_dim_init_irq_moder(vi->dev, profile_flags, coal_flags,
5791 				     DIM_CQ_PERIOD_MODE_START_FROM_EQE,
5792 				     0, virtnet_rx_dim_work, NULL);
5793 
5794 	if (ret)
5795 		return ret;
5796 
5797 	for (i = 0; i < vi->max_queue_pairs; i++)
5798 		net_dim_setting(vi->dev, &vi->rq[i].dim, false);
5799 
5800 	return 0;
5801 }
5802 
5803 static void virtnet_free_irq_moder(struct virtnet_info *vi)
5804 {
5805 	if (!virtio_has_feature(vi->vdev, VIRTIO_NET_F_VQ_NOTF_COAL))
5806 		return;
5807 
5808 	rtnl_lock();
5809 	net_dim_free_irq_moder(vi->dev);
5810 	rtnl_unlock();
5811 }
5812 
5813 static const struct net_device_ops virtnet_netdev = {
5814 	.ndo_open            = virtnet_open,
5815 	.ndo_stop   	     = virtnet_close,
5816 	.ndo_start_xmit      = start_xmit,
5817 	.ndo_validate_addr   = eth_validate_addr,
5818 	.ndo_set_mac_address = virtnet_set_mac_address,
5819 	.ndo_set_rx_mode     = virtnet_set_rx_mode,
5820 	.ndo_get_stats64     = virtnet_stats,
5821 	.ndo_vlan_rx_add_vid = virtnet_vlan_rx_add_vid,
5822 	.ndo_vlan_rx_kill_vid = virtnet_vlan_rx_kill_vid,
5823 	.ndo_bpf		= virtnet_xdp,
5824 	.ndo_xdp_xmit		= virtnet_xdp_xmit,
5825 	.ndo_xsk_wakeup         = virtnet_xsk_wakeup,
5826 	.ndo_features_check	= passthru_features_check,
5827 	.ndo_get_phys_port_name	= virtnet_get_phys_port_name,
5828 	.ndo_set_features	= virtnet_set_features,
5829 	.ndo_tx_timeout		= virtnet_tx_timeout,
5830 };
5831 
5832 static void virtnet_config_changed_work(struct work_struct *work)
5833 {
5834 	struct virtnet_info *vi =
5835 		container_of(work, struct virtnet_info, config_work);
5836 	u16 v;
5837 
5838 	if (virtio_cread_feature(vi->vdev, VIRTIO_NET_F_STATUS,
5839 				 struct virtio_net_config, status, &v) < 0)
5840 		return;
5841 
5842 	if (v & VIRTIO_NET_S_ANNOUNCE) {
5843 		netdev_notify_peers(vi->dev);
5844 		virtnet_ack_link_announce(vi);
5845 	}
5846 
5847 	/* Ignore unknown (future) status bits */
5848 	v &= VIRTIO_NET_S_LINK_UP;
5849 
5850 	if (vi->status == v)
5851 		return;
5852 
5853 	vi->status = v;
5854 
5855 	if (vi->status & VIRTIO_NET_S_LINK_UP) {
5856 		virtnet_update_settings(vi);
5857 		netif_carrier_on(vi->dev);
5858 		netif_tx_wake_all_queues(vi->dev);
5859 	} else {
5860 		netif_carrier_off(vi->dev);
5861 		netif_tx_stop_all_queues(vi->dev);
5862 	}
5863 }
5864 
5865 static void virtnet_config_changed(struct virtio_device *vdev)
5866 {
5867 	struct virtnet_info *vi = vdev->priv;
5868 
5869 	schedule_work(&vi->config_work);
5870 }
5871 
5872 static void virtnet_free_queues(struct virtnet_info *vi)
5873 {
5874 	int i;
5875 
5876 	for (i = 0; i < vi->max_queue_pairs; i++) {
5877 		__netif_napi_del(&vi->rq[i].napi);
5878 		__netif_napi_del(&vi->sq[i].napi);
5879 	}
5880 
5881 	/* We called __netif_napi_del(),
5882 	 * we need to respect an RCU grace period before freeing vi->rq
5883 	 */
5884 	synchronize_net();
5885 
5886 	kfree(vi->rq);
5887 	kfree(vi->sq);
5888 	kfree(vi->ctrl);
5889 }
5890 
5891 static void _free_receive_bufs(struct virtnet_info *vi)
5892 {
5893 	struct bpf_prog *old_prog;
5894 	int i;
5895 
5896 	for (i = 0; i < vi->max_queue_pairs; i++) {
5897 		while (vi->rq[i].pages)
5898 			__free_pages(get_a_page(&vi->rq[i], GFP_KERNEL), 0);
5899 
5900 		old_prog = rtnl_dereference(vi->rq[i].xdp_prog);
5901 		RCU_INIT_POINTER(vi->rq[i].xdp_prog, NULL);
5902 		if (old_prog)
5903 			bpf_prog_put(old_prog);
5904 	}
5905 }
5906 
5907 static void free_receive_bufs(struct virtnet_info *vi)
5908 {
5909 	rtnl_lock();
5910 	_free_receive_bufs(vi);
5911 	rtnl_unlock();
5912 }
5913 
5914 static void free_receive_page_frags(struct virtnet_info *vi)
5915 {
5916 	int i;
5917 	for (i = 0; i < vi->max_queue_pairs; i++)
5918 		if (vi->rq[i].alloc_frag.page) {
5919 			if (vi->rq[i].do_dma && vi->rq[i].last_dma)
5920 				virtnet_rq_unmap(&vi->rq[i], vi->rq[i].last_dma, 0);
5921 			put_page(vi->rq[i].alloc_frag.page);
5922 		}
5923 }
5924 
5925 static void virtnet_sq_free_unused_buf(struct virtqueue *vq, void *buf)
5926 {
5927 	if (!is_xdp_frame(buf))
5928 		dev_kfree_skb(buf);
5929 	else
5930 		xdp_return_frame(ptr_to_xdp(buf));
5931 }
5932 
5933 static void free_unused_bufs(struct virtnet_info *vi)
5934 {
5935 	void *buf;
5936 	int i;
5937 
5938 	for (i = 0; i < vi->max_queue_pairs; i++) {
5939 		struct virtqueue *vq = vi->sq[i].vq;
5940 		while ((buf = virtqueue_detach_unused_buf(vq)) != NULL)
5941 			virtnet_sq_free_unused_buf(vq, buf);
5942 		cond_resched();
5943 	}
5944 
5945 	for (i = 0; i < vi->max_queue_pairs; i++) {
5946 		struct virtqueue *vq = vi->rq[i].vq;
5947 
5948 		while ((buf = virtqueue_detach_unused_buf(vq)) != NULL)
5949 			virtnet_rq_unmap_free_buf(vq, buf);
5950 		cond_resched();
5951 	}
5952 }
5953 
5954 static void virtnet_del_vqs(struct virtnet_info *vi)
5955 {
5956 	struct virtio_device *vdev = vi->vdev;
5957 
5958 	virtnet_clean_affinity(vi);
5959 
5960 	vdev->config->del_vqs(vdev);
5961 
5962 	virtnet_free_queues(vi);
5963 }
5964 
5965 /* How large should a single buffer be so a queue full of these can fit at
5966  * least one full packet?
5967  * Logic below assumes the mergeable buffer header is used.
5968  */
5969 static unsigned int mergeable_min_buf_len(struct virtnet_info *vi, struct virtqueue *vq)
5970 {
5971 	const unsigned int hdr_len = vi->hdr_len;
5972 	unsigned int rq_size = virtqueue_get_vring_size(vq);
5973 	unsigned int packet_len = vi->big_packets ? IP_MAX_MTU : vi->dev->max_mtu;
5974 	unsigned int buf_len = hdr_len + ETH_HLEN + VLAN_HLEN + packet_len;
5975 	unsigned int min_buf_len = DIV_ROUND_UP(buf_len, rq_size);
5976 
5977 	return max(max(min_buf_len, hdr_len) - hdr_len,
5978 		   (unsigned int)GOOD_PACKET_LEN);
5979 }
5980 
5981 static int virtnet_find_vqs(struct virtnet_info *vi)
5982 {
5983 	struct virtqueue_info *vqs_info;
5984 	struct virtqueue **vqs;
5985 	int ret = -ENOMEM;
5986 	int total_vqs;
5987 	bool *ctx;
5988 	u16 i;
5989 
5990 	/* We expect 1 RX virtqueue followed by 1 TX virtqueue, followed by
5991 	 * possible N-1 RX/TX queue pairs used in multiqueue mode, followed by
5992 	 * possible control vq.
5993 	 */
5994 	total_vqs = vi->max_queue_pairs * 2 +
5995 		    virtio_has_feature(vi->vdev, VIRTIO_NET_F_CTRL_VQ);
5996 
5997 	/* Allocate space for find_vqs parameters */
5998 	vqs = kcalloc(total_vqs, sizeof(*vqs), GFP_KERNEL);
5999 	if (!vqs)
6000 		goto err_vq;
6001 	vqs_info = kcalloc(total_vqs, sizeof(*vqs_info), GFP_KERNEL);
6002 	if (!vqs_info)
6003 		goto err_vqs_info;
6004 	if (!vi->big_packets || vi->mergeable_rx_bufs) {
6005 		ctx = kcalloc(total_vqs, sizeof(*ctx), GFP_KERNEL);
6006 		if (!ctx)
6007 			goto err_ctx;
6008 	} else {
6009 		ctx = NULL;
6010 	}
6011 
6012 	/* Parameters for control virtqueue, if any */
6013 	if (vi->has_cvq) {
6014 		vqs_info[total_vqs - 1].name = "control";
6015 	}
6016 
6017 	/* Allocate/initialize parameters for send/receive virtqueues */
6018 	for (i = 0; i < vi->max_queue_pairs; i++) {
6019 		vqs_info[rxq2vq(i)].callback = skb_recv_done;
6020 		vqs_info[txq2vq(i)].callback = skb_xmit_done;
6021 		sprintf(vi->rq[i].name, "input.%u", i);
6022 		sprintf(vi->sq[i].name, "output.%u", i);
6023 		vqs_info[rxq2vq(i)].name = vi->rq[i].name;
6024 		vqs_info[txq2vq(i)].name = vi->sq[i].name;
6025 		if (ctx)
6026 			vqs_info[rxq2vq(i)].ctx = true;
6027 	}
6028 
6029 	ret = virtio_find_vqs(vi->vdev, total_vqs, vqs, vqs_info, NULL);
6030 	if (ret)
6031 		goto err_find;
6032 
6033 	if (vi->has_cvq) {
6034 		vi->cvq = vqs[total_vqs - 1];
6035 		if (virtio_has_feature(vi->vdev, VIRTIO_NET_F_CTRL_VLAN))
6036 			vi->dev->features |= NETIF_F_HW_VLAN_CTAG_FILTER;
6037 	}
6038 
6039 	for (i = 0; i < vi->max_queue_pairs; i++) {
6040 		vi->rq[i].vq = vqs[rxq2vq(i)];
6041 		vi->rq[i].min_buf_len = mergeable_min_buf_len(vi, vi->rq[i].vq);
6042 		vi->sq[i].vq = vqs[txq2vq(i)];
6043 	}
6044 
6045 	/* run here: ret == 0. */
6046 
6047 
6048 err_find:
6049 	kfree(ctx);
6050 err_ctx:
6051 	kfree(vqs_info);
6052 err_vqs_info:
6053 	kfree(vqs);
6054 err_vq:
6055 	return ret;
6056 }
6057 
6058 static int virtnet_alloc_queues(struct virtnet_info *vi)
6059 {
6060 	int i;
6061 
6062 	if (vi->has_cvq) {
6063 		vi->ctrl = kzalloc(sizeof(*vi->ctrl), GFP_KERNEL);
6064 		if (!vi->ctrl)
6065 			goto err_ctrl;
6066 	} else {
6067 		vi->ctrl = NULL;
6068 	}
6069 	vi->sq = kcalloc(vi->max_queue_pairs, sizeof(*vi->sq), GFP_KERNEL);
6070 	if (!vi->sq)
6071 		goto err_sq;
6072 	vi->rq = kcalloc(vi->max_queue_pairs, sizeof(*vi->rq), GFP_KERNEL);
6073 	if (!vi->rq)
6074 		goto err_rq;
6075 
6076 	INIT_DELAYED_WORK(&vi->refill, refill_work);
6077 	for (i = 0; i < vi->max_queue_pairs; i++) {
6078 		vi->rq[i].pages = NULL;
6079 		netif_napi_add_weight(vi->dev, &vi->rq[i].napi, virtnet_poll,
6080 				      napi_weight);
6081 		netif_napi_add_tx_weight(vi->dev, &vi->sq[i].napi,
6082 					 virtnet_poll_tx,
6083 					 napi_tx ? napi_weight : 0);
6084 
6085 		sg_init_table(vi->rq[i].sg, ARRAY_SIZE(vi->rq[i].sg));
6086 		ewma_pkt_len_init(&vi->rq[i].mrg_avg_pkt_len);
6087 		sg_init_table(vi->sq[i].sg, ARRAY_SIZE(vi->sq[i].sg));
6088 
6089 		u64_stats_init(&vi->rq[i].stats.syncp);
6090 		u64_stats_init(&vi->sq[i].stats.syncp);
6091 		mutex_init(&vi->rq[i].dim_lock);
6092 	}
6093 
6094 	return 0;
6095 
6096 err_rq:
6097 	kfree(vi->sq);
6098 err_sq:
6099 	kfree(vi->ctrl);
6100 err_ctrl:
6101 	return -ENOMEM;
6102 }
6103 
6104 static int init_vqs(struct virtnet_info *vi)
6105 {
6106 	int ret;
6107 
6108 	/* Allocate send & receive queues */
6109 	ret = virtnet_alloc_queues(vi);
6110 	if (ret)
6111 		goto err;
6112 
6113 	ret = virtnet_find_vqs(vi);
6114 	if (ret)
6115 		goto err_free;
6116 
6117 	cpus_read_lock();
6118 	virtnet_set_affinity(vi);
6119 	cpus_read_unlock();
6120 
6121 	return 0;
6122 
6123 err_free:
6124 	virtnet_free_queues(vi);
6125 err:
6126 	return ret;
6127 }
6128 
6129 #ifdef CONFIG_SYSFS
6130 static ssize_t mergeable_rx_buffer_size_show(struct netdev_rx_queue *queue,
6131 		char *buf)
6132 {
6133 	struct virtnet_info *vi = netdev_priv(queue->dev);
6134 	unsigned int queue_index = get_netdev_rx_queue_index(queue);
6135 	unsigned int headroom = virtnet_get_headroom(vi);
6136 	unsigned int tailroom = headroom ? sizeof(struct skb_shared_info) : 0;
6137 	struct ewma_pkt_len *avg;
6138 
6139 	BUG_ON(queue_index >= vi->max_queue_pairs);
6140 	avg = &vi->rq[queue_index].mrg_avg_pkt_len;
6141 	return sprintf(buf, "%u\n",
6142 		       get_mergeable_buf_len(&vi->rq[queue_index], avg,
6143 				       SKB_DATA_ALIGN(headroom + tailroom)));
6144 }
6145 
6146 static struct rx_queue_attribute mergeable_rx_buffer_size_attribute =
6147 	__ATTR_RO(mergeable_rx_buffer_size);
6148 
6149 static struct attribute *virtio_net_mrg_rx_attrs[] = {
6150 	&mergeable_rx_buffer_size_attribute.attr,
6151 	NULL
6152 };
6153 
6154 static const struct attribute_group virtio_net_mrg_rx_group = {
6155 	.name = "virtio_net",
6156 	.attrs = virtio_net_mrg_rx_attrs
6157 };
6158 #endif
6159 
6160 static bool virtnet_fail_on_feature(struct virtio_device *vdev,
6161 				    unsigned int fbit,
6162 				    const char *fname, const char *dname)
6163 {
6164 	if (!virtio_has_feature(vdev, fbit))
6165 		return false;
6166 
6167 	dev_err(&vdev->dev, "device advertises feature %s but not %s",
6168 		fname, dname);
6169 
6170 	return true;
6171 }
6172 
6173 #define VIRTNET_FAIL_ON(vdev, fbit, dbit)			\
6174 	virtnet_fail_on_feature(vdev, fbit, #fbit, dbit)
6175 
6176 static bool virtnet_validate_features(struct virtio_device *vdev)
6177 {
6178 	if (!virtio_has_feature(vdev, VIRTIO_NET_F_CTRL_VQ) &&
6179 	    (VIRTNET_FAIL_ON(vdev, VIRTIO_NET_F_CTRL_RX,
6180 			     "VIRTIO_NET_F_CTRL_VQ") ||
6181 	     VIRTNET_FAIL_ON(vdev, VIRTIO_NET_F_CTRL_VLAN,
6182 			     "VIRTIO_NET_F_CTRL_VQ") ||
6183 	     VIRTNET_FAIL_ON(vdev, VIRTIO_NET_F_GUEST_ANNOUNCE,
6184 			     "VIRTIO_NET_F_CTRL_VQ") ||
6185 	     VIRTNET_FAIL_ON(vdev, VIRTIO_NET_F_MQ, "VIRTIO_NET_F_CTRL_VQ") ||
6186 	     VIRTNET_FAIL_ON(vdev, VIRTIO_NET_F_CTRL_MAC_ADDR,
6187 			     "VIRTIO_NET_F_CTRL_VQ") ||
6188 	     VIRTNET_FAIL_ON(vdev, VIRTIO_NET_F_RSS,
6189 			     "VIRTIO_NET_F_CTRL_VQ") ||
6190 	     VIRTNET_FAIL_ON(vdev, VIRTIO_NET_F_HASH_REPORT,
6191 			     "VIRTIO_NET_F_CTRL_VQ") ||
6192 	     VIRTNET_FAIL_ON(vdev, VIRTIO_NET_F_NOTF_COAL,
6193 			     "VIRTIO_NET_F_CTRL_VQ") ||
6194 	     VIRTNET_FAIL_ON(vdev, VIRTIO_NET_F_VQ_NOTF_COAL,
6195 			     "VIRTIO_NET_F_CTRL_VQ"))) {
6196 		return false;
6197 	}
6198 
6199 	return true;
6200 }
6201 
6202 #define MIN_MTU ETH_MIN_MTU
6203 #define MAX_MTU ETH_MAX_MTU
6204 
6205 static int virtnet_validate(struct virtio_device *vdev)
6206 {
6207 	if (!vdev->config->get) {
6208 		dev_err(&vdev->dev, "%s failure: config access disabled\n",
6209 			__func__);
6210 		return -EINVAL;
6211 	}
6212 
6213 	if (!virtnet_validate_features(vdev))
6214 		return -EINVAL;
6215 
6216 	if (virtio_has_feature(vdev, VIRTIO_NET_F_MTU)) {
6217 		int mtu = virtio_cread16(vdev,
6218 					 offsetof(struct virtio_net_config,
6219 						  mtu));
6220 		if (mtu < MIN_MTU)
6221 			__virtio_clear_bit(vdev, VIRTIO_NET_F_MTU);
6222 	}
6223 
6224 	if (virtio_has_feature(vdev, VIRTIO_NET_F_STANDBY) &&
6225 	    !virtio_has_feature(vdev, VIRTIO_NET_F_MAC)) {
6226 		dev_warn(&vdev->dev, "device advertises feature VIRTIO_NET_F_STANDBY but not VIRTIO_NET_F_MAC, disabling standby");
6227 		__virtio_clear_bit(vdev, VIRTIO_NET_F_STANDBY);
6228 	}
6229 
6230 	return 0;
6231 }
6232 
6233 static bool virtnet_check_guest_gso(const struct virtnet_info *vi)
6234 {
6235 	return virtio_has_feature(vi->vdev, VIRTIO_NET_F_GUEST_TSO4) ||
6236 		virtio_has_feature(vi->vdev, VIRTIO_NET_F_GUEST_TSO6) ||
6237 		virtio_has_feature(vi->vdev, VIRTIO_NET_F_GUEST_ECN) ||
6238 		virtio_has_feature(vi->vdev, VIRTIO_NET_F_GUEST_UFO) ||
6239 		(virtio_has_feature(vi->vdev, VIRTIO_NET_F_GUEST_USO4) &&
6240 		virtio_has_feature(vi->vdev, VIRTIO_NET_F_GUEST_USO6));
6241 }
6242 
6243 static void virtnet_set_big_packets(struct virtnet_info *vi, const int mtu)
6244 {
6245 	bool guest_gso = virtnet_check_guest_gso(vi);
6246 
6247 	/* If device can receive ANY guest GSO packets, regardless of mtu,
6248 	 * allocate packets of maximum size, otherwise limit it to only
6249 	 * mtu size worth only.
6250 	 */
6251 	if (mtu > ETH_DATA_LEN || guest_gso) {
6252 		vi->big_packets = true;
6253 		vi->big_packets_num_skbfrags = guest_gso ? MAX_SKB_FRAGS : DIV_ROUND_UP(mtu, PAGE_SIZE);
6254 	}
6255 }
6256 
6257 #define VIRTIO_NET_HASH_REPORT_MAX_TABLE      10
6258 static enum xdp_rss_hash_type
6259 virtnet_xdp_rss_type[VIRTIO_NET_HASH_REPORT_MAX_TABLE] = {
6260 	[VIRTIO_NET_HASH_REPORT_NONE] = XDP_RSS_TYPE_NONE,
6261 	[VIRTIO_NET_HASH_REPORT_IPv4] = XDP_RSS_TYPE_L3_IPV4,
6262 	[VIRTIO_NET_HASH_REPORT_TCPv4] = XDP_RSS_TYPE_L4_IPV4_TCP,
6263 	[VIRTIO_NET_HASH_REPORT_UDPv4] = XDP_RSS_TYPE_L4_IPV4_UDP,
6264 	[VIRTIO_NET_HASH_REPORT_IPv6] = XDP_RSS_TYPE_L3_IPV6,
6265 	[VIRTIO_NET_HASH_REPORT_TCPv6] = XDP_RSS_TYPE_L4_IPV6_TCP,
6266 	[VIRTIO_NET_HASH_REPORT_UDPv6] = XDP_RSS_TYPE_L4_IPV6_UDP,
6267 	[VIRTIO_NET_HASH_REPORT_IPv6_EX] = XDP_RSS_TYPE_L3_IPV6_EX,
6268 	[VIRTIO_NET_HASH_REPORT_TCPv6_EX] = XDP_RSS_TYPE_L4_IPV6_TCP_EX,
6269 	[VIRTIO_NET_HASH_REPORT_UDPv6_EX] = XDP_RSS_TYPE_L4_IPV6_UDP_EX
6270 };
6271 
6272 static int virtnet_xdp_rx_hash(const struct xdp_md *_ctx, u32 *hash,
6273 			       enum xdp_rss_hash_type *rss_type)
6274 {
6275 	const struct xdp_buff *xdp = (void *)_ctx;
6276 	struct virtio_net_hdr_v1_hash *hdr_hash;
6277 	struct virtnet_info *vi;
6278 	u16 hash_report;
6279 
6280 	if (!(xdp->rxq->dev->features & NETIF_F_RXHASH))
6281 		return -ENODATA;
6282 
6283 	vi = netdev_priv(xdp->rxq->dev);
6284 	hdr_hash = (struct virtio_net_hdr_v1_hash *)(xdp->data - vi->hdr_len);
6285 	hash_report = __le16_to_cpu(hdr_hash->hash_report);
6286 
6287 	if (hash_report >= VIRTIO_NET_HASH_REPORT_MAX_TABLE)
6288 		hash_report = VIRTIO_NET_HASH_REPORT_NONE;
6289 
6290 	*rss_type = virtnet_xdp_rss_type[hash_report];
6291 	*hash = __le32_to_cpu(hdr_hash->hash_value);
6292 	return 0;
6293 }
6294 
6295 static const struct xdp_metadata_ops virtnet_xdp_metadata_ops = {
6296 	.xmo_rx_hash			= virtnet_xdp_rx_hash,
6297 };
6298 
6299 static int virtnet_probe(struct virtio_device *vdev)
6300 {
6301 	int i, err = -ENOMEM;
6302 	struct net_device *dev;
6303 	struct virtnet_info *vi;
6304 	u16 max_queue_pairs;
6305 	int mtu = 0;
6306 
6307 	/* Find if host supports multiqueue/rss virtio_net device */
6308 	max_queue_pairs = 1;
6309 	if (virtio_has_feature(vdev, VIRTIO_NET_F_MQ) || virtio_has_feature(vdev, VIRTIO_NET_F_RSS))
6310 		max_queue_pairs =
6311 		     virtio_cread16(vdev, offsetof(struct virtio_net_config, max_virtqueue_pairs));
6312 
6313 	/* We need at least 2 queue's */
6314 	if (max_queue_pairs < VIRTIO_NET_CTRL_MQ_VQ_PAIRS_MIN ||
6315 	    max_queue_pairs > VIRTIO_NET_CTRL_MQ_VQ_PAIRS_MAX ||
6316 	    !virtio_has_feature(vdev, VIRTIO_NET_F_CTRL_VQ))
6317 		max_queue_pairs = 1;
6318 
6319 	/* Allocate ourselves a network device with room for our info */
6320 	dev = alloc_etherdev_mq(sizeof(struct virtnet_info), max_queue_pairs);
6321 	if (!dev)
6322 		return -ENOMEM;
6323 
6324 	/* Set up network device as normal. */
6325 	dev->priv_flags |= IFF_UNICAST_FLT | IFF_LIVE_ADDR_CHANGE |
6326 			   IFF_TX_SKB_NO_LINEAR;
6327 	dev->netdev_ops = &virtnet_netdev;
6328 	dev->stat_ops = &virtnet_stat_ops;
6329 	dev->features = NETIF_F_HIGHDMA;
6330 
6331 	dev->ethtool_ops = &virtnet_ethtool_ops;
6332 	SET_NETDEV_DEV(dev, &vdev->dev);
6333 
6334 	/* Do we support "hardware" checksums? */
6335 	if (virtio_has_feature(vdev, VIRTIO_NET_F_CSUM)) {
6336 		/* This opens up the world of extra features. */
6337 		dev->hw_features |= NETIF_F_HW_CSUM | NETIF_F_SG;
6338 		if (csum)
6339 			dev->features |= NETIF_F_HW_CSUM | NETIF_F_SG;
6340 
6341 		if (virtio_has_feature(vdev, VIRTIO_NET_F_GSO)) {
6342 			dev->hw_features |= NETIF_F_TSO
6343 				| NETIF_F_TSO_ECN | NETIF_F_TSO6;
6344 		}
6345 		/* Individual feature bits: what can host handle? */
6346 		if (virtio_has_feature(vdev, VIRTIO_NET_F_HOST_TSO4))
6347 			dev->hw_features |= NETIF_F_TSO;
6348 		if (virtio_has_feature(vdev, VIRTIO_NET_F_HOST_TSO6))
6349 			dev->hw_features |= NETIF_F_TSO6;
6350 		if (virtio_has_feature(vdev, VIRTIO_NET_F_HOST_ECN))
6351 			dev->hw_features |= NETIF_F_TSO_ECN;
6352 		if (virtio_has_feature(vdev, VIRTIO_NET_F_HOST_USO))
6353 			dev->hw_features |= NETIF_F_GSO_UDP_L4;
6354 
6355 		dev->features |= NETIF_F_GSO_ROBUST;
6356 
6357 		if (gso)
6358 			dev->features |= dev->hw_features & NETIF_F_ALL_TSO;
6359 		/* (!csum && gso) case will be fixed by register_netdev() */
6360 	}
6361 
6362 	/* 1. With VIRTIO_NET_F_GUEST_CSUM negotiation, the driver doesn't
6363 	 * need to calculate checksums for partially checksummed packets,
6364 	 * as they're considered valid by the upper layer.
6365 	 * 2. Without VIRTIO_NET_F_GUEST_CSUM negotiation, the driver only
6366 	 * receives fully checksummed packets. The device may assist in
6367 	 * validating these packets' checksums, so the driver won't have to.
6368 	 */
6369 	dev->features |= NETIF_F_RXCSUM;
6370 
6371 	if (virtio_has_feature(vdev, VIRTIO_NET_F_GUEST_TSO4) ||
6372 	    virtio_has_feature(vdev, VIRTIO_NET_F_GUEST_TSO6))
6373 		dev->features |= NETIF_F_GRO_HW;
6374 	if (virtio_has_feature(vdev, VIRTIO_NET_F_CTRL_GUEST_OFFLOADS))
6375 		dev->hw_features |= NETIF_F_GRO_HW;
6376 
6377 	dev->vlan_features = dev->features;
6378 	dev->xdp_features = NETDEV_XDP_ACT_BASIC | NETDEV_XDP_ACT_REDIRECT;
6379 
6380 	/* MTU range: 68 - 65535 */
6381 	dev->min_mtu = MIN_MTU;
6382 	dev->max_mtu = MAX_MTU;
6383 
6384 	/* Configuration may specify what MAC to use.  Otherwise random. */
6385 	if (virtio_has_feature(vdev, VIRTIO_NET_F_MAC)) {
6386 		u8 addr[ETH_ALEN];
6387 
6388 		virtio_cread_bytes(vdev,
6389 				   offsetof(struct virtio_net_config, mac),
6390 				   addr, ETH_ALEN);
6391 		eth_hw_addr_set(dev, addr);
6392 	} else {
6393 		eth_hw_addr_random(dev);
6394 		dev_info(&vdev->dev, "Assigned random MAC address %pM\n",
6395 			 dev->dev_addr);
6396 	}
6397 
6398 	/* Set up our device-specific information */
6399 	vi = netdev_priv(dev);
6400 	vi->dev = dev;
6401 	vi->vdev = vdev;
6402 	vdev->priv = vi;
6403 
6404 	INIT_WORK(&vi->config_work, virtnet_config_changed_work);
6405 	INIT_WORK(&vi->rx_mode_work, virtnet_rx_mode_work);
6406 	spin_lock_init(&vi->refill_lock);
6407 
6408 	if (virtio_has_feature(vdev, VIRTIO_NET_F_MRG_RXBUF)) {
6409 		vi->mergeable_rx_bufs = true;
6410 		dev->xdp_features |= NETDEV_XDP_ACT_RX_SG;
6411 	}
6412 
6413 	if (virtio_has_feature(vdev, VIRTIO_NET_F_HASH_REPORT))
6414 		vi->has_rss_hash_report = true;
6415 
6416 	if (virtio_has_feature(vdev, VIRTIO_NET_F_RSS)) {
6417 		vi->has_rss = true;
6418 
6419 		vi->rss_indir_table_size =
6420 			virtio_cread16(vdev, offsetof(struct virtio_net_config,
6421 				rss_max_indirection_table_length));
6422 	}
6423 
6424 	if (vi->has_rss || vi->has_rss_hash_report) {
6425 		vi->rss_key_size =
6426 			virtio_cread8(vdev, offsetof(struct virtio_net_config, rss_max_key_size));
6427 
6428 		vi->rss_hash_types_supported =
6429 		    virtio_cread32(vdev, offsetof(struct virtio_net_config, supported_hash_types));
6430 		vi->rss_hash_types_supported &=
6431 				~(VIRTIO_NET_RSS_HASH_TYPE_IP_EX |
6432 				  VIRTIO_NET_RSS_HASH_TYPE_TCP_EX |
6433 				  VIRTIO_NET_RSS_HASH_TYPE_UDP_EX);
6434 
6435 		dev->hw_features |= NETIF_F_RXHASH;
6436 		dev->xdp_metadata_ops = &virtnet_xdp_metadata_ops;
6437 	}
6438 
6439 	if (vi->has_rss_hash_report)
6440 		vi->hdr_len = sizeof(struct virtio_net_hdr_v1_hash);
6441 	else if (virtio_has_feature(vdev, VIRTIO_NET_F_MRG_RXBUF) ||
6442 		 virtio_has_feature(vdev, VIRTIO_F_VERSION_1))
6443 		vi->hdr_len = sizeof(struct virtio_net_hdr_mrg_rxbuf);
6444 	else
6445 		vi->hdr_len = sizeof(struct virtio_net_hdr);
6446 
6447 	if (virtio_has_feature(vdev, VIRTIO_F_ANY_LAYOUT) ||
6448 	    virtio_has_feature(vdev, VIRTIO_F_VERSION_1))
6449 		vi->any_header_sg = true;
6450 
6451 	if (virtio_has_feature(vdev, VIRTIO_NET_F_CTRL_VQ))
6452 		vi->has_cvq = true;
6453 
6454 	mutex_init(&vi->cvq_lock);
6455 
6456 	if (virtio_has_feature(vdev, VIRTIO_NET_F_MTU)) {
6457 		mtu = virtio_cread16(vdev,
6458 				     offsetof(struct virtio_net_config,
6459 					      mtu));
6460 		if (mtu < dev->min_mtu) {
6461 			/* Should never trigger: MTU was previously validated
6462 			 * in virtnet_validate.
6463 			 */
6464 			dev_err(&vdev->dev,
6465 				"device MTU appears to have changed it is now %d < %d",
6466 				mtu, dev->min_mtu);
6467 			err = -EINVAL;
6468 			goto free;
6469 		}
6470 
6471 		dev->mtu = mtu;
6472 		dev->max_mtu = mtu;
6473 	}
6474 
6475 	virtnet_set_big_packets(vi, mtu);
6476 
6477 	if (vi->any_header_sg)
6478 		dev->needed_headroom = vi->hdr_len;
6479 
6480 	/* Enable multiqueue by default */
6481 	if (num_online_cpus() >= max_queue_pairs)
6482 		vi->curr_queue_pairs = max_queue_pairs;
6483 	else
6484 		vi->curr_queue_pairs = num_online_cpus();
6485 	vi->max_queue_pairs = max_queue_pairs;
6486 
6487 	/* Allocate/initialize the rx/tx queues, and invoke find_vqs */
6488 	err = init_vqs(vi);
6489 	if (err)
6490 		goto free;
6491 
6492 	if (virtio_has_feature(vi->vdev, VIRTIO_NET_F_NOTF_COAL)) {
6493 		vi->intr_coal_rx.max_usecs = 0;
6494 		vi->intr_coal_tx.max_usecs = 0;
6495 		vi->intr_coal_rx.max_packets = 0;
6496 
6497 		/* Keep the default values of the coalescing parameters
6498 		 * aligned with the default napi_tx state.
6499 		 */
6500 		if (vi->sq[0].napi.weight)
6501 			vi->intr_coal_tx.max_packets = 1;
6502 		else
6503 			vi->intr_coal_tx.max_packets = 0;
6504 	}
6505 
6506 	if (virtio_has_feature(vi->vdev, VIRTIO_NET_F_VQ_NOTF_COAL)) {
6507 		/* The reason is the same as VIRTIO_NET_F_NOTF_COAL. */
6508 		for (i = 0; i < vi->max_queue_pairs; i++)
6509 			if (vi->sq[i].napi.weight)
6510 				vi->sq[i].intr_coal.max_packets = 1;
6511 
6512 		err = virtnet_init_irq_moder(vi);
6513 		if (err)
6514 			goto free;
6515 	}
6516 
6517 #ifdef CONFIG_SYSFS
6518 	if (vi->mergeable_rx_bufs)
6519 		dev->sysfs_rx_queue_group = &virtio_net_mrg_rx_group;
6520 #endif
6521 	netif_set_real_num_tx_queues(dev, vi->curr_queue_pairs);
6522 	netif_set_real_num_rx_queues(dev, vi->curr_queue_pairs);
6523 
6524 	virtnet_init_settings(dev);
6525 
6526 	if (virtio_has_feature(vdev, VIRTIO_NET_F_STANDBY)) {
6527 		vi->failover = net_failover_create(vi->dev);
6528 		if (IS_ERR(vi->failover)) {
6529 			err = PTR_ERR(vi->failover);
6530 			goto free_vqs;
6531 		}
6532 	}
6533 
6534 	if (vi->has_rss || vi->has_rss_hash_report)
6535 		virtnet_init_default_rss(vi);
6536 
6537 	enable_rx_mode_work(vi);
6538 
6539 	/* serialize netdev register + virtio_device_ready() with ndo_open() */
6540 	rtnl_lock();
6541 
6542 	err = register_netdevice(dev);
6543 	if (err) {
6544 		pr_debug("virtio_net: registering device failed\n");
6545 		rtnl_unlock();
6546 		goto free_failover;
6547 	}
6548 
6549 	/* Disable config change notification until ndo_open. */
6550 	virtio_config_driver_disable(vi->vdev);
6551 
6552 	virtio_device_ready(vdev);
6553 
6554 	virtnet_set_queues(vi, vi->curr_queue_pairs);
6555 
6556 	/* a random MAC address has been assigned, notify the device.
6557 	 * We don't fail probe if VIRTIO_NET_F_CTRL_MAC_ADDR is not there
6558 	 * because many devices work fine without getting MAC explicitly
6559 	 */
6560 	if (!virtio_has_feature(vdev, VIRTIO_NET_F_MAC) &&
6561 	    virtio_has_feature(vi->vdev, VIRTIO_NET_F_CTRL_MAC_ADDR)) {
6562 		struct scatterlist sg;
6563 
6564 		sg_init_one(&sg, dev->dev_addr, dev->addr_len);
6565 		if (!virtnet_send_command(vi, VIRTIO_NET_CTRL_MAC,
6566 					  VIRTIO_NET_CTRL_MAC_ADDR_SET, &sg)) {
6567 			pr_debug("virtio_net: setting MAC address failed\n");
6568 			rtnl_unlock();
6569 			err = -EINVAL;
6570 			goto free_unregister_netdev;
6571 		}
6572 	}
6573 
6574 	if (virtio_has_feature(vi->vdev, VIRTIO_NET_F_DEVICE_STATS)) {
6575 		struct virtio_net_stats_capabilities *stats_cap  __free(kfree) = NULL;
6576 		struct scatterlist sg;
6577 		__le64 v;
6578 
6579 		stats_cap = kzalloc(sizeof(*stats_cap), GFP_KERNEL);
6580 		if (!stats_cap) {
6581 			rtnl_unlock();
6582 			err = -ENOMEM;
6583 			goto free_unregister_netdev;
6584 		}
6585 
6586 		sg_init_one(&sg, stats_cap, sizeof(*stats_cap));
6587 
6588 		if (!virtnet_send_command_reply(vi, VIRTIO_NET_CTRL_STATS,
6589 						VIRTIO_NET_CTRL_STATS_QUERY,
6590 						NULL, &sg)) {
6591 			pr_debug("virtio_net: fail to get stats capability\n");
6592 			rtnl_unlock();
6593 			err = -EINVAL;
6594 			goto free_unregister_netdev;
6595 		}
6596 
6597 		v = stats_cap->supported_stats_types[0];
6598 		vi->device_stats_cap = le64_to_cpu(v);
6599 	}
6600 
6601 	/* Assume link up if device can't report link status,
6602 	   otherwise get link status from config. */
6603 	netif_carrier_off(dev);
6604 	if (virtio_has_feature(vi->vdev, VIRTIO_NET_F_STATUS)) {
6605 		virtnet_config_changed_work(&vi->config_work);
6606 	} else {
6607 		vi->status = VIRTIO_NET_S_LINK_UP;
6608 		virtnet_update_settings(vi);
6609 		netif_carrier_on(dev);
6610 	}
6611 
6612 	for (i = 0; i < ARRAY_SIZE(guest_offloads); i++)
6613 		if (virtio_has_feature(vi->vdev, guest_offloads[i]))
6614 			set_bit(guest_offloads[i], &vi->guest_offloads);
6615 	vi->guest_offloads_capable = vi->guest_offloads;
6616 
6617 	rtnl_unlock();
6618 
6619 	err = virtnet_cpu_notif_add(vi);
6620 	if (err) {
6621 		pr_debug("virtio_net: registering cpu notifier failed\n");
6622 		goto free_unregister_netdev;
6623 	}
6624 
6625 	pr_debug("virtnet: registered device %s with %d RX and TX vq's\n",
6626 		 dev->name, max_queue_pairs);
6627 
6628 	return 0;
6629 
6630 free_unregister_netdev:
6631 	unregister_netdev(dev);
6632 free_failover:
6633 	net_failover_destroy(vi->failover);
6634 free_vqs:
6635 	virtio_reset_device(vdev);
6636 	cancel_delayed_work_sync(&vi->refill);
6637 	free_receive_page_frags(vi);
6638 	virtnet_del_vqs(vi);
6639 free:
6640 	free_netdev(dev);
6641 	return err;
6642 }
6643 
6644 static void remove_vq_common(struct virtnet_info *vi)
6645 {
6646 	virtio_reset_device(vi->vdev);
6647 
6648 	/* Free unused buffers in both send and recv, if any. */
6649 	free_unused_bufs(vi);
6650 
6651 	free_receive_bufs(vi);
6652 
6653 	free_receive_page_frags(vi);
6654 
6655 	virtnet_del_vqs(vi);
6656 }
6657 
6658 static void virtnet_remove(struct virtio_device *vdev)
6659 {
6660 	struct virtnet_info *vi = vdev->priv;
6661 
6662 	virtnet_cpu_notif_remove(vi);
6663 
6664 	/* Make sure no work handler is accessing the device. */
6665 	flush_work(&vi->config_work);
6666 	disable_rx_mode_work(vi);
6667 	flush_work(&vi->rx_mode_work);
6668 
6669 	virtnet_free_irq_moder(vi);
6670 
6671 	unregister_netdev(vi->dev);
6672 
6673 	net_failover_destroy(vi->failover);
6674 
6675 	remove_vq_common(vi);
6676 
6677 	free_netdev(vi->dev);
6678 }
6679 
6680 static __maybe_unused int virtnet_freeze(struct virtio_device *vdev)
6681 {
6682 	struct virtnet_info *vi = vdev->priv;
6683 
6684 	virtnet_cpu_notif_remove(vi);
6685 	virtnet_freeze_down(vdev);
6686 	remove_vq_common(vi);
6687 
6688 	return 0;
6689 }
6690 
6691 static __maybe_unused int virtnet_restore(struct virtio_device *vdev)
6692 {
6693 	struct virtnet_info *vi = vdev->priv;
6694 	int err;
6695 
6696 	err = virtnet_restore_up(vdev);
6697 	if (err)
6698 		return err;
6699 	virtnet_set_queues(vi, vi->curr_queue_pairs);
6700 
6701 	err = virtnet_cpu_notif_add(vi);
6702 	if (err) {
6703 		virtnet_freeze_down(vdev);
6704 		remove_vq_common(vi);
6705 		return err;
6706 	}
6707 
6708 	return 0;
6709 }
6710 
6711 static struct virtio_device_id id_table[] = {
6712 	{ VIRTIO_ID_NET, VIRTIO_DEV_ANY_ID },
6713 	{ 0 },
6714 };
6715 
6716 #define VIRTNET_FEATURES \
6717 	VIRTIO_NET_F_CSUM, VIRTIO_NET_F_GUEST_CSUM, \
6718 	VIRTIO_NET_F_MAC, \
6719 	VIRTIO_NET_F_HOST_TSO4, VIRTIO_NET_F_HOST_UFO, VIRTIO_NET_F_HOST_TSO6, \
6720 	VIRTIO_NET_F_HOST_ECN, VIRTIO_NET_F_GUEST_TSO4, VIRTIO_NET_F_GUEST_TSO6, \
6721 	VIRTIO_NET_F_GUEST_ECN, VIRTIO_NET_F_GUEST_UFO, \
6722 	VIRTIO_NET_F_HOST_USO, VIRTIO_NET_F_GUEST_USO4, VIRTIO_NET_F_GUEST_USO6, \
6723 	VIRTIO_NET_F_MRG_RXBUF, VIRTIO_NET_F_STATUS, VIRTIO_NET_F_CTRL_VQ, \
6724 	VIRTIO_NET_F_CTRL_RX, VIRTIO_NET_F_CTRL_VLAN, \
6725 	VIRTIO_NET_F_GUEST_ANNOUNCE, VIRTIO_NET_F_MQ, \
6726 	VIRTIO_NET_F_CTRL_MAC_ADDR, \
6727 	VIRTIO_NET_F_MTU, VIRTIO_NET_F_CTRL_GUEST_OFFLOADS, \
6728 	VIRTIO_NET_F_SPEED_DUPLEX, VIRTIO_NET_F_STANDBY, \
6729 	VIRTIO_NET_F_RSS, VIRTIO_NET_F_HASH_REPORT, VIRTIO_NET_F_NOTF_COAL, \
6730 	VIRTIO_NET_F_VQ_NOTF_COAL, \
6731 	VIRTIO_NET_F_GUEST_HDRLEN, VIRTIO_NET_F_DEVICE_STATS
6732 
6733 static unsigned int features[] = {
6734 	VIRTNET_FEATURES,
6735 };
6736 
6737 static unsigned int features_legacy[] = {
6738 	VIRTNET_FEATURES,
6739 	VIRTIO_NET_F_GSO,
6740 	VIRTIO_F_ANY_LAYOUT,
6741 };
6742 
6743 static struct virtio_driver virtio_net_driver = {
6744 	.feature_table = features,
6745 	.feature_table_size = ARRAY_SIZE(features),
6746 	.feature_table_legacy = features_legacy,
6747 	.feature_table_size_legacy = ARRAY_SIZE(features_legacy),
6748 	.driver.name =	KBUILD_MODNAME,
6749 	.id_table =	id_table,
6750 	.validate =	virtnet_validate,
6751 	.probe =	virtnet_probe,
6752 	.remove =	virtnet_remove,
6753 	.config_changed = virtnet_config_changed,
6754 #ifdef CONFIG_PM_SLEEP
6755 	.freeze =	virtnet_freeze,
6756 	.restore =	virtnet_restore,
6757 #endif
6758 };
6759 
6760 static __init int virtio_net_driver_init(void)
6761 {
6762 	int ret;
6763 
6764 	ret = cpuhp_setup_state_multi(CPUHP_AP_ONLINE_DYN, "virtio/net:online",
6765 				      virtnet_cpu_online,
6766 				      virtnet_cpu_down_prep);
6767 	if (ret < 0)
6768 		goto out;
6769 	virtionet_online = ret;
6770 	ret = cpuhp_setup_state_multi(CPUHP_VIRT_NET_DEAD, "virtio/net:dead",
6771 				      NULL, virtnet_cpu_dead);
6772 	if (ret)
6773 		goto err_dead;
6774 	ret = register_virtio_driver(&virtio_net_driver);
6775 	if (ret)
6776 		goto err_virtio;
6777 	return 0;
6778 err_virtio:
6779 	cpuhp_remove_multi_state(CPUHP_VIRT_NET_DEAD);
6780 err_dead:
6781 	cpuhp_remove_multi_state(virtionet_online);
6782 out:
6783 	return ret;
6784 }
6785 module_init(virtio_net_driver_init);
6786 
6787 static __exit void virtio_net_driver_exit(void)
6788 {
6789 	unregister_virtio_driver(&virtio_net_driver);
6790 	cpuhp_remove_multi_state(CPUHP_VIRT_NET_DEAD);
6791 	cpuhp_remove_multi_state(virtionet_online);
6792 }
6793 module_exit(virtio_net_driver_exit);
6794 
6795 MODULE_DEVICE_TABLE(virtio, id_table);
6796 MODULE_DESCRIPTION("Virtio network driver");
6797 MODULE_LICENSE("GPL");
6798