?  search:
summary | log | graphiclog1 | graphiclog2 | commit | commitdiff | tree | refs | edit | fork
blame | history | raw | HEAD
mmc: core: Reset HPI enabled state during re-init and in case of errors
[linux/fpc-iii.git] / drivers / net / virtio_net.c
blobea672145f6a66b97ec3572f213939a64323fcdf3
1 /* A network driver using virtio.
2 *
3 * Copyright 2007 Rusty Russell <rusty@rustcorp.com.au> IBM Corporation
4 *
5 * This program is free software; you can redistribute it and/or modify
6 * it under the terms of the GNU General Public License as published by
7 * the Free Software Foundation; either version 2 of the License, or
8 * (at your option) any later version.
9 *
10 * This program is distributed in the hope that it will be useful,
11 * but WITHOUT ANY WARRANTY; without even the implied warranty of
12 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
13 * GNU General Public License for more details.
14 *
15 * You should have received a copy of the GNU General Public License
16 * along with this program; if not, see <http://www.gnu.org/licenses/>.
17 */
18 //#define DEBUG
19 #include <linux/netdevice.h>
20 #include <linux/etherdevice.h>
21 #include <linux/ethtool.h>
22 #include <linux/module.h>
23 #include <linux/virtio.h>
24 #include <linux/virtio_net.h>
25 #include <linux/bpf.h>
26 #include <linux/bpf_trace.h>
27 #include <linux/scatterlist.h>
28 #include <linux/if_vlan.h>
29 #include <linux/slab.h>
30 #include <linux/cpu.h>
31 #include <linux/average.h>
32 #include <linux/filter.h>
33 #include <linux/kernel.h>
34 #include <linux/pci.h>
35 #include <net/route.h>
36 #include <net/xdp.h>
37 #include <net/net_failover.h>
38
39 static int napi_weight = NAPI_POLL_WEIGHT;
40 module_param(napi_weight, int, 0444);
41
42 static bool csum = true, gso = true, napi_tx;
43 module_param(csum, bool, 0444);
44 module_param(gso, bool, 0444);
45 module_param(napi_tx, bool, 0644);
46
47 /* FIXME: MTU in config. */
48 #define GOOD_PACKET_LEN (ETH_HLEN + VLAN_HLEN + ETH_DATA_LEN)
49 #define GOOD_COPY_LEN 128
50
51 #define VIRTNET_RX_PAD (NET_IP_ALIGN + NET_SKB_PAD)
52
53 /* Amount of XDP headroom to prepend to packets for use by xdp_adjust_head */
54 #define VIRTIO_XDP_HEADROOM 256
55
56 /* Separating two types of XDP xmit */
57 #define VIRTIO_XDP_TX BIT(0)
58 #define VIRTIO_XDP_REDIR BIT(1)
59
60 /* RX packet size EWMA. The average packet size is used to determine the packet
61 * buffer size when refilling RX rings. As the entire RX ring may be refilled
62 * at once, the weight is chosen so that the EWMA will be insensitive to short-
63 * term, transient changes in packet size.
64 */
65 DECLARE_EWMA(pkt_len, 0, 64)
66
67 #define VIRTNET_DRIVER_VERSION "1.0.0"
68
69 static const unsigned long guest_offloads[] = {
70 VIRTIO_NET_F_GUEST_TSO4,
71 VIRTIO_NET_F_GUEST_TSO6,
72 VIRTIO_NET_F_GUEST_ECN,
73 VIRTIO_NET_F_GUEST_UFO,
74 VIRTIO_NET_F_GUEST_CSUM
75 };
76
77 struct virtnet_stat_desc {
78 char desc[ETH_GSTRING_LEN];
79 size_t offset;
80 };
81
82 struct virtnet_sq_stats {
83 struct u64_stats_sync syncp;
84 u64 packets;
85 u64 bytes;
86 u64 xdp_tx;
87 u64 xdp_tx_drops;
88 u64 kicks;
89 };
90
91 struct virtnet_rq_stats {
92 struct u64_stats_sync syncp;
93 u64 packets;
94 u64 bytes;
95 u64 drops;
96 u64 xdp_packets;
97 u64 xdp_tx;
98 u64 xdp_redirects;
99 u64 xdp_drops;
100 u64 kicks;
101 };
102
103 #define VIRTNET_SQ_STAT(m) offsetof(struct virtnet_sq_stats, m)
104 #define VIRTNET_RQ_STAT(m) offsetof(struct virtnet_rq_stats, m)
105
106 static const struct virtnet_stat_desc virtnet_sq_stats_desc[] = {
107 { "packets", VIRTNET_SQ_STAT(packets) },
108 { "bytes", VIRTNET_SQ_STAT(bytes) },
109 { "xdp_tx", VIRTNET_SQ_STAT(xdp_tx) },
110 { "xdp_tx_drops", VIRTNET_SQ_STAT(xdp_tx_drops) },
111 { "kicks", VIRTNET_SQ_STAT(kicks) },
112 };
113
114 static const struct virtnet_stat_desc virtnet_rq_stats_desc[] = {
115 { "packets", VIRTNET_RQ_STAT(packets) },
116 { "bytes", VIRTNET_RQ_STAT(bytes) },
117 { "drops", VIRTNET_RQ_STAT(drops) },
118 { "xdp_packets", VIRTNET_RQ_STAT(xdp_packets) },
119 { "xdp_tx", VIRTNET_RQ_STAT(xdp_tx) },
120 { "xdp_redirects", VIRTNET_RQ_STAT(xdp_redirects) },
121 { "xdp_drops", VIRTNET_RQ_STAT(xdp_drops) },
122 { "kicks", VIRTNET_RQ_STAT(kicks) },
123 };
124
125 #define VIRTNET_SQ_STATS_LEN ARRAY_SIZE(virtnet_sq_stats_desc)
126 #define VIRTNET_RQ_STATS_LEN ARRAY_SIZE(virtnet_rq_stats_desc)
127
128 /* Internal representation of a send virtqueue */
129 struct send_queue {
130 /* Virtqueue associated with this send _queue */
131 struct virtqueue *vq;
132
133 /* TX: fragments + linear part + virtio header */
134 struct scatterlist sg[MAX_SKB_FRAGS + 2];
135
136 /* Name of the send queue: output.$index */
137 char name[40];
138
139 struct virtnet_sq_stats stats;
140
141 struct napi_struct napi;
142 };
143
144 /* Internal representation of a receive virtqueue */
145 struct receive_queue {
146 /* Virtqueue associated with this receive_queue */
147 struct virtqueue *vq;
148
149 struct napi_struct napi;
150
151 struct bpf_prog __rcu *xdp_prog;
152
153 struct virtnet_rq_stats stats;
154
155 /* Chain pages by the private ptr. */
156 struct page *pages;
157
158 /* Average packet length for mergeable receive buffers. */
159 struct ewma_pkt_len mrg_avg_pkt_len;
160
161 /* Page frag for packet buffer allocation. */
162 struct page_frag alloc_frag;
163
164 /* RX: fragments + linear part + virtio header */
165 struct scatterlist sg[MAX_SKB_FRAGS + 2];
166
167 /* Min single buffer size for mergeable buffers case. */
168 unsigned int min_buf_len;
169
170 /* Name of this receive queue: input.$index */
171 char name[40];
172
173 struct xdp_rxq_info xdp_rxq;
174 };
175
176 /* Control VQ buffers: protected by the rtnl lock */
177 struct control_buf {
178 struct virtio_net_ctrl_hdr hdr;
179 virtio_net_ctrl_ack status;
180 struct virtio_net_ctrl_mq mq;
181 u8 promisc;
182 u8 allmulti;
183 __virtio16 vid;
184 __virtio64 offloads;
185 };
186
187 struct virtnet_info {
188 struct virtio_device *vdev;
189 struct virtqueue *cvq;
190 struct net_device *dev;
191 struct send_queue *sq;
192 struct receive_queue *rq;
193 unsigned int status;
194
195 /* Max # of queue pairs supported by the device */
196 u16 max_queue_pairs;
197
198 /* # of queue pairs currently used by the driver */
199 u16 curr_queue_pairs;
200
201 /* # of XDP queue pairs currently used by the driver */
202 u16 xdp_queue_pairs;
203
204 /* I like... big packets and I cannot lie! */
205 bool big_packets;
206
207 /* Host will merge rx buffers for big packets (shake it! shake it!) */
208 bool mergeable_rx_bufs;
209
210 /* Has control virtqueue */
211 bool has_cvq;
212
213 /* Host can handle any s/g split between our header and packet data */
214 bool any_header_sg;
215
216 /* Packet virtio header size */
217 u8 hdr_len;
218
219 /* Work struct for refilling if we run low on memory. */
220 struct delayed_work refill;
221
222 /* Work struct for config space updates */
223 struct work_struct config_work;
224
225 /* Does the affinity hint is set for virtqueues? */
226 bool affinity_hint_set;
227
228 /* CPU hotplug instances for online & dead */
229 struct hlist_node node;
230 struct hlist_node node_dead;
231
232 struct control_buf *ctrl;
233
234 /* Ethtool settings */
235 u8 duplex;
236 u32 speed;
237
238 unsigned long guest_offloads;
239
240 /* failover when STANDBY feature enabled */
241 struct failover *failover;
242 };
243
244 struct padded_vnet_hdr {
245 struct virtio_net_hdr_mrg_rxbuf hdr;
246 /*
247 * hdr is in a separate sg buffer, and data sg buffer shares same page
248 * with this header sg. This padding makes next sg 16 byte aligned
249 * after the header.
250 */
251 char padding[4];
252 };
253
254 /* Converting between virtqueue no. and kernel tx/rx queue no.
255 * 0:rx0 1:tx0 2:rx1 3:tx1 ... 2N:rxN 2N+1:txN 2N+2:cvq
256 */
257 static int vq2txq(struct virtqueue *vq)
258 {
259 return (vq->index - 1) / 2;
260 }
261
262 static int txq2vq(int txq)
263 {
264 return txq * 2 + 1;
265 }
266
267 static int vq2rxq(struct virtqueue *vq)
268 {
269 return vq->index / 2;
270 }
271
272 static int rxq2vq(int rxq)
273 {
274 return rxq * 2;
275 }
276
277 static inline struct virtio_net_hdr_mrg_rxbuf *skb_vnet_hdr(struct sk_buff *skb)
278 {
279 return (struct virtio_net_hdr_mrg_rxbuf *)skb->cb;
280 }
281
282 /*
283 * private is used to chain pages for big packets, put the whole
284 * most recent used list in the beginning for reuse
285 */
286 static void give_pages(struct receive_queue *rq, struct page *page)
287 {
288 struct page *end;
289
290 /* Find end of list, sew whole thing into vi->rq.pages. */
291 for (end = page; end->private; end = (struct page *)end->private);
292 end->private = (unsigned long)rq->pages;
293 rq->pages = page;
294 }
295
296 static struct page *get_a_page(struct receive_queue *rq, gfp_t gfp_mask)
297 {
298 struct page *p = rq->pages;
299
300 if (p) {
301 rq->pages = (struct page *)p->private;
302 /* clear private here, it is used to chain pages */
303 p->private = 0;
304 } else
305 p = alloc_page(gfp_mask);
306 return p;
307 }
308
309 static void virtqueue_napi_schedule(struct napi_struct *napi,
310 struct virtqueue *vq)
311 {
312 if (napi_schedule_prep(napi)) {
313 virtqueue_disable_cb(vq);
314 __napi_schedule(napi);
315 }
316 }
317
318 static void virtqueue_napi_complete(struct napi_struct *napi,
319 struct virtqueue *vq, int processed)
320 {
321 int opaque;
322
323 opaque = virtqueue_enable_cb_prepare(vq);
324 if (napi_complete_done(napi, processed)) {
325 if (unlikely(virtqueue_poll(vq, opaque)))
326 virtqueue_napi_schedule(napi, vq);
327 } else {
328 virtqueue_disable_cb(vq);
329 }
330 }
331
332 static void skb_xmit_done(struct virtqueue *vq)
333 {
334 struct virtnet_info *vi = vq->vdev->priv;
335 struct napi_struct *napi = &vi->sq[vq2txq(vq)].napi;
336
337 /* Suppress further interrupts. */
338 virtqueue_disable_cb(vq);
339
340 if (napi->weight)
341 virtqueue_napi_schedule(napi, vq);
342 else
343 /* We were probably waiting for more output buffers. */
344 netif_wake_subqueue(vi->dev, vq2txq(vq));
345 }
346
347 #define MRG_CTX_HEADER_SHIFT 22
348 static void *mergeable_len_to_ctx(unsigned int truesize,
349 unsigned int headroom)
350 {
351 return (void *)(unsigned long)((headroom << MRG_CTX_HEADER_SHIFT) | truesize);
352 }
353
354 static unsigned int mergeable_ctx_to_headroom(void *mrg_ctx)
355 {
356 return (unsigned long)mrg_ctx >> MRG_CTX_HEADER_SHIFT;
357 }
358
359 static unsigned int mergeable_ctx_to_truesize(void *mrg_ctx)
360 {
361 return (unsigned long)mrg_ctx & ((1 << MRG_CTX_HEADER_SHIFT) - 1);
362 }
363
364 /* Called from bottom half context */
365 static struct sk_buff *page_to_skb(struct virtnet_info *vi,
366 struct receive_queue *rq,
367 struct page *page, unsigned int offset,
368 unsigned int len, unsigned int truesize,
369 bool hdr_valid)
370 {
371 struct sk_buff *skb;
372 struct virtio_net_hdr_mrg_rxbuf *hdr;
373 unsigned int copy, hdr_len, hdr_padded_len;
374 char *p;
375
376 p = page_address(page) + offset;
377
378 /* copy small packet so we can reuse these pages for small data */
379 skb = napi_alloc_skb(&rq->napi, GOOD_COPY_LEN);
380 if (unlikely(!skb))
381 return NULL;
382
383 hdr = skb_vnet_hdr(skb);
384
385 hdr_len = vi->hdr_len;
386 if (vi->mergeable_rx_bufs)
387 hdr_padded_len = sizeof(*hdr);
388 else
389 hdr_padded_len = sizeof(struct padded_vnet_hdr);
390
391 if (hdr_valid)
392 memcpy(hdr, p, hdr_len);
393
394 len -= hdr_len;
395 offset += hdr_padded_len;
396 p += hdr_padded_len;
397
398 copy = len;
399 if (copy > skb_tailroom(skb))
400 copy = skb_tailroom(skb);
401 skb_put_data(skb, p, copy);
402
403 len -= copy;
404 offset += copy;
405
406 if (vi->mergeable_rx_bufs) {
407 if (len)
408 skb_add_rx_frag(skb, 0, page, offset, len, truesize);
409 else
410 put_page(page);
411 return skb;
412 }
413
414 /*
415 * Verify that we can indeed put this data into a skb.
416 * This is here to handle cases when the device erroneously
417 * tries to receive more than is possible. This is usually
418 * the case of a broken device.
419 */
420 if (unlikely(len > MAX_SKB_FRAGS * PAGE_SIZE)) {
421 net_dbg_ratelimited("%s: too much data\n", skb->dev->name);
422 dev_kfree_skb(skb);
423 return NULL;
424 }
425 BUG_ON(offset >= PAGE_SIZE);
426 while (len) {
427 unsigned int frag_size = min((unsigned)PAGE_SIZE - offset, len);
428 skb_add_rx_frag(skb, skb_shinfo(skb)->nr_frags, page, offset,
429 frag_size, truesize);
430 len -= frag_size;
431 page = (struct page *)page->private;
432 offset = 0;
433 }
434
435 if (page)
436 give_pages(rq, page);
437
438 return skb;
439 }
440
441 static int __virtnet_xdp_xmit_one(struct virtnet_info *vi,
442 struct send_queue *sq,
443 struct xdp_frame *xdpf)
444 {
445 struct virtio_net_hdr_mrg_rxbuf *hdr;
446 int err;
447
448 /* virtqueue want to use data area in-front of packet */
449 if (unlikely(xdpf->metasize > 0))
450 return -EOPNOTSUPP;
451
452 if (unlikely(xdpf->headroom < vi->hdr_len))
453 return -EOVERFLOW;
454
455 /* Make room for virtqueue hdr (also change xdpf->headroom?) */
456 xdpf->data -= vi->hdr_len;
457 /* Zero header and leave csum up to XDP layers */
458 hdr = xdpf->data;
459 memset(hdr, 0, vi->hdr_len);
460 xdpf->len += vi->hdr_len;
461
462 sg_init_one(sq->sg, xdpf->data, xdpf->len);
463
464 err = virtqueue_add_outbuf(sq->vq, sq->sg, 1, xdpf, GFP_ATOMIC);
465 if (unlikely(err))
466 return -ENOSPC; /* Caller handle free/refcnt */
467
468 return 0;
469 }
470
471 static struct send_queue *virtnet_xdp_sq(struct virtnet_info *vi)
472 {
473 unsigned int qp;
474
475 qp = vi->curr_queue_pairs - vi->xdp_queue_pairs + smp_processor_id();
476 return &vi->sq[qp];
477 }
478
479 static int virtnet_xdp_xmit(struct net_device *dev,
480 int n, struct xdp_frame **frames, u32 flags)
481 {
482 struct virtnet_info *vi = netdev_priv(dev);
483 struct receive_queue *rq = vi->rq;
484 struct xdp_frame *xdpf_sent;
485 struct bpf_prog *xdp_prog;
486 struct send_queue *sq;
487 unsigned int len;
488 int drops = 0;
489 int kicks = 0;
490 int ret, err;
491 int i;
492
493 sq = virtnet_xdp_sq(vi);
494
495 if (unlikely(flags & ~XDP_XMIT_FLAGS_MASK)) {
496 ret = -EINVAL;
497 drops = n;
498 goto out;
499 }
500
501 /* Only allow ndo_xdp_xmit if XDP is loaded on dev, as this
502 * indicate XDP resources have been successfully allocated.
503 */
504 xdp_prog = rcu_dereference(rq->xdp_prog);
505 if (!xdp_prog) {
506 ret = -ENXIO;
507 drops = n;
508 goto out;
509 }
510
511 /* Free up any pending old buffers before queueing new ones. */
512 while ((xdpf_sent = virtqueue_get_buf(sq->vq, &len)) != NULL)
513 xdp_return_frame(xdpf_sent);
514
515 for (i = 0; i < n; i++) {
516 struct xdp_frame *xdpf = frames[i];
517
518 err = __virtnet_xdp_xmit_one(vi, sq, xdpf);
519 if (err) {
520 xdp_return_frame_rx_napi(xdpf);
521 drops++;
522 }
523 }
524 ret = n - drops;
525
526 if (flags & XDP_XMIT_FLUSH) {
527 if (virtqueue_kick_prepare(sq->vq) && virtqueue_notify(sq->vq))
528 kicks = 1;
529 }
530 out:
531 u64_stats_update_begin(&sq->stats.syncp);
532 sq->stats.xdp_tx += n;
533 sq->stats.xdp_tx_drops += drops;
534 sq->stats.kicks += kicks;
535 u64_stats_update_end(&sq->stats.syncp);
536
537 return ret;
538 }
539
540 static unsigned int virtnet_get_headroom(struct virtnet_info *vi)
541 {
542 return vi->xdp_queue_pairs ? VIRTIO_XDP_HEADROOM : 0;
543 }
544
545 /* We copy the packet for XDP in the following cases:
546 *
547 * 1) Packet is scattered across multiple rx buffers.
548 * 2) Headroom space is insufficient.
549 *
550 * This is inefficient but it's a temporary condition that
551 * we hit right after XDP is enabled and until queue is refilled
552 * with large buffers with sufficient headroom - so it should affect
553 * at most queue size packets.
554 * Afterwards, the conditions to enable
555 * XDP should preclude the underlying device from sending packets
556 * across multiple buffers (num_buf > 1), and we make sure buffers
557 * have enough headroom.
558 */
559 static struct page *xdp_linearize_page(struct receive_queue *rq,
560 u16 *num_buf,
561 struct page *p,
562 int offset,
563 int page_off,
564 unsigned int *len)
565 {
566 struct page *page = alloc_page(GFP_ATOMIC);
567
568 if (!page)
569 return NULL;
570
571 memcpy(page_address(page) + page_off, page_address(p) + offset, *len);
572 page_off += *len;
573
574 while (--*num_buf) {
575 int tailroom = SKB_DATA_ALIGN(sizeof(struct skb_shared_info));
576 unsigned int buflen;
577 void *buf;
578 int off;
579
580 buf = virtqueue_get_buf(rq->vq, &buflen);
581 if (unlikely(!buf))
582 goto err_buf;
583
584 p = virt_to_head_page(buf);
585 off = buf - page_address(p);
586
587 /* guard against a misconfigured or uncooperative backend that
588 * is sending packet larger than the MTU.
589 */
590 if ((page_off + buflen + tailroom) > PAGE_SIZE) {
591 put_page(p);
592 goto err_buf;
593 }
594
595 memcpy(page_address(page) + page_off,
596 page_address(p) + off, buflen);
597 page_off += buflen;
598 put_page(p);
599 }
600
601 /* Headroom does not contribute to packet length */
602 *len = page_off - VIRTIO_XDP_HEADROOM;
603 return page;
604 err_buf:
605 __free_pages(page, 0);
606 return NULL;
607 }
608
609 static struct sk_buff *receive_small(struct net_device *dev,
610 struct virtnet_info *vi,
611 struct receive_queue *rq,
612 void *buf, void *ctx,
613 unsigned int len,
614 unsigned int *xdp_xmit,
615 struct virtnet_rq_stats *stats)
616 {
617 struct sk_buff *skb;
618 struct bpf_prog *xdp_prog;
619 unsigned int xdp_headroom = (unsigned long)ctx;
620 unsigned int header_offset = VIRTNET_RX_PAD + xdp_headroom;
621 unsigned int headroom = vi->hdr_len + header_offset;
622 unsigned int buflen = SKB_DATA_ALIGN(GOOD_PACKET_LEN + headroom) +
623 SKB_DATA_ALIGN(sizeof(struct skb_shared_info));
624 struct page *page = virt_to_head_page(buf);
625 unsigned int delta = 0;
626 struct page *xdp_page;
627 int err;
628
629 len -= vi->hdr_len;
630 stats->bytes += len;
631
632 rcu_read_lock();
633 xdp_prog = rcu_dereference(rq->xdp_prog);
634 if (xdp_prog) {
635 struct virtio_net_hdr_mrg_rxbuf *hdr = buf + header_offset;
636 struct xdp_frame *xdpf;
637 struct xdp_buff xdp;
638 void *orig_data;
639 u32 act;
640
641 if (unlikely(hdr->hdr.gso_type))
642 goto err_xdp;
643
644 if (unlikely(xdp_headroom < virtnet_get_headroom(vi))) {
645 int offset = buf - page_address(page) + header_offset;
646 unsigned int tlen = len + vi->hdr_len;
647 u16 num_buf = 1;
648
649 xdp_headroom = virtnet_get_headroom(vi);
650 header_offset = VIRTNET_RX_PAD + xdp_headroom;
651 headroom = vi->hdr_len + header_offset;
652 buflen = SKB_DATA_ALIGN(GOOD_PACKET_LEN + headroom) +
653 SKB_DATA_ALIGN(sizeof(struct skb_shared_info));
654 xdp_page = xdp_linearize_page(rq, &num_buf, page,
655 offset, header_offset,
656 &tlen);
657 if (!xdp_page)
658 goto err_xdp;
659
660 buf = page_address(xdp_page);
661 put_page(page);
662 page = xdp_page;
663 }
664
665 xdp.data_hard_start = buf + VIRTNET_RX_PAD + vi->hdr_len;
666 xdp.data = xdp.data_hard_start + xdp_headroom;
667 xdp_set_data_meta_invalid(&xdp);
668 xdp.data_end = xdp.data + len;
669 xdp.rxq = &rq->xdp_rxq;
670 orig_data = xdp.data;
671 act = bpf_prog_run_xdp(xdp_prog, &xdp);
672 stats->xdp_packets++;
673
674 switch (act) {
675 case XDP_PASS:
676 /* Recalculate length in case bpf program changed it */
677 delta = orig_data - xdp.data;
678 len = xdp.data_end - xdp.data;
679 break;
680 case XDP_TX:
681 stats->xdp_tx++;
682 xdpf = convert_to_xdp_frame(&xdp);
683 if (unlikely(!xdpf))
684 goto err_xdp;
685 err = virtnet_xdp_xmit(dev, 1, &xdpf, 0);
686 if (unlikely(err < 0)) {
687 trace_xdp_exception(vi->dev, xdp_prog, act);
688 goto err_xdp;
689 }
690 *xdp_xmit |= VIRTIO_XDP_TX;
691 rcu_read_unlock();
692 goto xdp_xmit;
693 case XDP_REDIRECT:
694 stats->xdp_redirects++;
695 err = xdp_do_redirect(dev, &xdp, xdp_prog);
696 if (err)
697 goto err_xdp;
698 *xdp_xmit |= VIRTIO_XDP_REDIR;
699 rcu_read_unlock();
700 goto xdp_xmit;
701 default:
702 bpf_warn_invalid_xdp_action(act);
703 /* fall through */
704 case XDP_ABORTED:
705 trace_xdp_exception(vi->dev, xdp_prog, act);
706 case XDP_DROP:
707 goto err_xdp;
708 }
709 }
710 rcu_read_unlock();
711
712 skb = build_skb(buf, buflen);
713 if (!skb) {
714 put_page(page);
715 goto err;
716 }
717 skb_reserve(skb, headroom - delta);
718 skb_put(skb, len);
719 if (!delta) {
720 buf += header_offset;
721 memcpy(skb_vnet_hdr(skb), buf, vi->hdr_len);
722 } /* keep zeroed vnet hdr since packet was changed by bpf */
723
724 err:
725 return skb;
726
727 err_xdp:
728 rcu_read_unlock();
729 stats->xdp_drops++;
730 stats->drops++;
731 put_page(page);
732 xdp_xmit:
733 return NULL;
734 }
735
736 static struct sk_buff *receive_big(struct net_device *dev,
737 struct virtnet_info *vi,
738 struct receive_queue *rq,
739 void *buf,
740 unsigned int len,
741 struct virtnet_rq_stats *stats)
742 {
743 struct page *page = buf;
744 struct sk_buff *skb = page_to_skb(vi, rq, page, 0, len,
745 PAGE_SIZE, true);
746
747 stats->bytes += len - vi->hdr_len;
748 if (unlikely(!skb))
749 goto err;
750
751 return skb;
752
753 err:
754 stats->drops++;
755 give_pages(rq, page);
756 return NULL;
757 }
758
759 static struct sk_buff *receive_mergeable(struct net_device *dev,
760 struct virtnet_info *vi,
761 struct receive_queue *rq,
762 void *buf,
763 void *ctx,
764 unsigned int len,
765 unsigned int *xdp_xmit,
766 struct virtnet_rq_stats *stats)
767 {
768 struct virtio_net_hdr_mrg_rxbuf *hdr = buf;
769 u16 num_buf = virtio16_to_cpu(vi->vdev, hdr->num_buffers);
770 struct page *page = virt_to_head_page(buf);
771 int offset = buf - page_address(page);
772 struct sk_buff *head_skb, *curr_skb;
773 struct bpf_prog *xdp_prog;
774 unsigned int truesize;
775 unsigned int headroom = mergeable_ctx_to_headroom(ctx);
776 int err;
777
778 head_skb = NULL;
779 stats->bytes += len - vi->hdr_len;
780
781 rcu_read_lock();
782 xdp_prog = rcu_dereference(rq->xdp_prog);
783 if (xdp_prog) {
784 struct xdp_frame *xdpf;
785 struct page *xdp_page;
786 struct xdp_buff xdp;
787 void *data;
788 u32 act;
789
790 /* Transient failure which in theory could occur if
791 * in-flight packets from before XDP was enabled reach
792 * the receive path after XDP is loaded.
793 */
794 if (unlikely(hdr->hdr.gso_type))
795 goto err_xdp;
796
797 /* This happens when rx buffer size is underestimated
798 * or headroom is not enough because of the buffer
799 * was refilled before XDP is set. This should only
800 * happen for the first several packets, so we don't
801 * care much about its performance.
802 */
803 if (unlikely(num_buf > 1 ||
804 headroom < virtnet_get_headroom(vi))) {
805 /* linearize data for XDP */
806 xdp_page = xdp_linearize_page(rq, &num_buf,
807 page, offset,
808 VIRTIO_XDP_HEADROOM,
809 &len);
810 if (!xdp_page)
811 goto err_xdp;
812 offset = VIRTIO_XDP_HEADROOM;
813 } else {
814 xdp_page = page;
815 }
816
817 /* Allow consuming headroom but reserve enough space to push
818 * the descriptor on if we get an XDP_TX return code.
819 */
820 data = page_address(xdp_page) + offset;
821 xdp.data_hard_start = data - VIRTIO_XDP_HEADROOM + vi->hdr_len;
822 xdp.data = data + vi->hdr_len;
823 xdp_set_data_meta_invalid(&xdp);
824 xdp.data_end = xdp.data + (len - vi->hdr_len);
825 xdp.rxq = &rq->xdp_rxq;
826
827 act = bpf_prog_run_xdp(xdp_prog, &xdp);
828 stats->xdp_packets++;
829
830 switch (act) {
831 case XDP_PASS:
832 /* recalculate offset to account for any header
833 * adjustments. Note other cases do not build an
834 * skb and avoid using offset
835 */
836 offset = xdp.data -
837 page_address(xdp_page) - vi->hdr_len;
838
839 /* recalculate len if xdp.data or xdp.data_end were
840 * adjusted
841 */
842 len = xdp.data_end - xdp.data + vi->hdr_len;
843 /* We can only create skb based on xdp_page. */
844 if (unlikely(xdp_page != page)) {
845 rcu_read_unlock();
846 put_page(page);
847 head_skb = page_to_skb(vi, rq, xdp_page,
848 offset, len,
849 PAGE_SIZE, false);
850 return head_skb;
851 }
852 break;
853 case XDP_TX:
854 stats->xdp_tx++;
855 xdpf = convert_to_xdp_frame(&xdp);
856 if (unlikely(!xdpf))
857 goto err_xdp;
858 err = virtnet_xdp_xmit(dev, 1, &xdpf, 0);
859 if (unlikely(err < 0)) {
860 trace_xdp_exception(vi->dev, xdp_prog, act);
861 if (unlikely(xdp_page != page))
862 put_page(xdp_page);
863 goto err_xdp;
864 }
865 *xdp_xmit |= VIRTIO_XDP_TX;
866 if (unlikely(xdp_page != page))
867 put_page(page);
868 rcu_read_unlock();
869 goto xdp_xmit;
870 case XDP_REDIRECT:
871 stats->xdp_redirects++;
872 err = xdp_do_redirect(dev, &xdp, xdp_prog);
873 if (err) {
874 if (unlikely(xdp_page != page))
875 put_page(xdp_page);
876 goto err_xdp;
877 }
878 *xdp_xmit |= VIRTIO_XDP_REDIR;
879 if (unlikely(xdp_page != page))
880 put_page(page);
881 rcu_read_unlock();
882 goto xdp_xmit;
883 default:
884 bpf_warn_invalid_xdp_action(act);
885 /* fall through */
886 case XDP_ABORTED:
887 trace_xdp_exception(vi->dev, xdp_prog, act);
888 /* fall through */
889 case XDP_DROP:
890 if (unlikely(xdp_page != page))
891 __free_pages(xdp_page, 0);
892 goto err_xdp;
893 }
894 }
895 rcu_read_unlock();
896
897 truesize = mergeable_ctx_to_truesize(ctx);
898 if (unlikely(len > truesize)) {
899 pr_debug("%s: rx error: len %u exceeds truesize %lu\n",
900 dev->name, len, (unsigned long)ctx);
901 dev->stats.rx_length_errors++;
902 goto err_skb;
903 }
904
905 head_skb = page_to_skb(vi, rq, page, offset, len, truesize, !xdp_prog);
906 curr_skb = head_skb;
907
908 if (unlikely(!curr_skb))
909 goto err_skb;
910 while (--num_buf) {
911 int num_skb_frags;
912
913 buf = virtqueue_get_buf_ctx(rq->vq, &len, &ctx);
914 if (unlikely(!buf)) {
915 pr_debug("%s: rx error: %d buffers out of %d missing\n",
916 dev->name, num_buf,
917 virtio16_to_cpu(vi->vdev,
918 hdr->num_buffers));
919 dev->stats.rx_length_errors++;
920 goto err_buf;
921 }
922
923 stats->bytes += len;
924 page = virt_to_head_page(buf);
925
926 truesize = mergeable_ctx_to_truesize(ctx);
927 if (unlikely(len > truesize)) {
928 pr_debug("%s: rx error: len %u exceeds truesize %lu\n",
929 dev->name, len, (unsigned long)ctx);
930 dev->stats.rx_length_errors++;
931 goto err_skb;
932 }
933
934 num_skb_frags = skb_shinfo(curr_skb)->nr_frags;
935 if (unlikely(num_skb_frags == MAX_SKB_FRAGS)) {
936 struct sk_buff *nskb = alloc_skb(0, GFP_ATOMIC);
937
938 if (unlikely(!nskb))
939 goto err_skb;
940 if (curr_skb == head_skb)
941 skb_shinfo(curr_skb)->frag_list = nskb;
942 else
943 curr_skb->next = nskb;
944 curr_skb = nskb;
945 head_skb->truesize += nskb->truesize;
946 num_skb_frags = 0;
947 }
948 if (curr_skb != head_skb) {
949 head_skb->data_len += len;
950 head_skb->len += len;
951 head_skb->truesize += truesize;
952 }
953 offset = buf - page_address(page);
954 if (skb_can_coalesce(curr_skb, num_skb_frags, page, offset)) {
955 put_page(page);
956 skb_coalesce_rx_frag(curr_skb, num_skb_frags - 1,
957 len, truesize);
958 } else {
959 skb_add_rx_frag(curr_skb, num_skb_frags, page,
960 offset, len, truesize);
961 }
962 }
963
964 ewma_pkt_len_add(&rq->mrg_avg_pkt_len, head_skb->len);
965 return head_skb;
966
967 err_xdp:
968 rcu_read_unlock();
969 stats->xdp_drops++;
970 err_skb:
971 put_page(page);
972 while (num_buf-- > 1) {
973 buf = virtqueue_get_buf(rq->vq, &len);
974 if (unlikely(!buf)) {
975 pr_debug("%s: rx error: %d buffers missing\n",
976 dev->name, num_buf);
977 dev->stats.rx_length_errors++;
978 break;
979 }
980 stats->bytes += len;
981 page = virt_to_head_page(buf);
982 put_page(page);
983 }
984 err_buf:
985 stats->drops++;
986 dev_kfree_skb(head_skb);
987 xdp_xmit:
988 return NULL;
989 }
990
991 static void receive_buf(struct virtnet_info *vi, struct receive_queue *rq,
992 void *buf, unsigned int len, void **ctx,
993 unsigned int *xdp_xmit,
994 struct virtnet_rq_stats *stats)
995 {
996 struct net_device *dev = vi->dev;
997 struct sk_buff *skb;
998 struct virtio_net_hdr_mrg_rxbuf *hdr;
999
1000 if (unlikely(len < vi->hdr_len + ETH_HLEN)) {
1001 pr_debug("%s: short packet %i\n", dev->name, len);
1002 dev->stats.rx_length_errors++;
1003 if (vi->mergeable_rx_bufs) {
1004 put_page(virt_to_head_page(buf));
1005 } else if (vi->big_packets) {
1006 give_pages(rq, buf);
1007 } else {
1008 put_page(virt_to_head_page(buf));
1009 }
1010 return;
1011 }
1012
1013 if (vi->mergeable_rx_bufs)
1014 skb = receive_mergeable(dev, vi, rq, buf, ctx, len, xdp_xmit,
1015 stats);
1016 else if (vi->big_packets)
1017 skb = receive_big(dev, vi, rq, buf, len, stats);
1018 else
1019 skb = receive_small(dev, vi, rq, buf, ctx, len, xdp_xmit, stats);
1020
1021 if (unlikely(!skb))
1022 return;
1023
1024 hdr = skb_vnet_hdr(skb);
1025
1026 if (hdr->hdr.flags & VIRTIO_NET_HDR_F_DATA_VALID)
1027 skb->ip_summed = CHECKSUM_UNNECESSARY;
1028
1029 if (virtio_net_hdr_to_skb(skb, &hdr->hdr,
1030 virtio_is_little_endian(vi->vdev))) {
1031 net_warn_ratelimited("%s: bad gso: type: %u, size: %u\n",
1032 dev->name, hdr->hdr.gso_type,
1033 hdr->hdr.gso_size);
1034 goto frame_err;
1035 }
1036
1037 skb->protocol = eth_type_trans(skb, dev);
1038 pr_debug("Receiving skb proto 0x%04x len %i type %i\n",
1039 ntohs(skb->protocol), skb->len, skb->pkt_type);
1040
1041 napi_gro_receive(&rq->napi, skb);
1042 return;
1043
1044 frame_err:
1045 dev->stats.rx_frame_errors++;
1046 dev_kfree_skb(skb);
1047 }
1048
1049 /* Unlike mergeable buffers, all buffers are allocated to the
1050 * same size, except for the headroom. For this reason we do
1051 * not need to use mergeable_len_to_ctx here - it is enough
1052 * to store the headroom as the context ignoring the truesize.
1053 */
1054 static int add_recvbuf_small(struct virtnet_info *vi, struct receive_queue *rq,
1055 gfp_t gfp)
1056 {
1057 struct page_frag *alloc_frag = &rq->alloc_frag;
1058 char *buf;
1059 unsigned int xdp_headroom = virtnet_get_headroom(vi);
1060 void *ctx = (void *)(unsigned long)xdp_headroom;
1061 int len = vi->hdr_len + VIRTNET_RX_PAD + GOOD_PACKET_LEN + xdp_headroom;
1062 int err;
1063
1064 len = SKB_DATA_ALIGN(len) +
1065 SKB_DATA_ALIGN(sizeof(struct skb_shared_info));
1066 if (unlikely(!skb_page_frag_refill(len, alloc_frag, gfp)))
1067 return -ENOMEM;
1068
1069 buf = (char *)page_address(alloc_frag->page) + alloc_frag->offset;
1070 get_page(alloc_frag->page);
1071 alloc_frag->offset += len;
1072 sg_init_one(rq->sg, buf + VIRTNET_RX_PAD + xdp_headroom,
1073 vi->hdr_len + GOOD_PACKET_LEN);
1074 err = virtqueue_add_inbuf_ctx(rq->vq, rq->sg, 1, buf, ctx, gfp);
1075 if (err < 0)
1076 put_page(virt_to_head_page(buf));
1077 return err;
1078 }
1079
1080 static int add_recvbuf_big(struct virtnet_info *vi, struct receive_queue *rq,
1081 gfp_t gfp)
1082 {
1083 struct page *first, *list = NULL;
1084 char *p;
1085 int i, err, offset;
1086
1087 sg_init_table(rq->sg, MAX_SKB_FRAGS + 2);
1088
1089 /* page in rq->sg[MAX_SKB_FRAGS + 1] is list tail */
1090 for (i = MAX_SKB_FRAGS + 1; i > 1; --i) {
1091 first = get_a_page(rq, gfp);
1092 if (!first) {
1093 if (list)
1094 give_pages(rq, list);
1095 return -ENOMEM;
1096 }
1097 sg_set_buf(&rq->sg[i], page_address(first), PAGE_SIZE);
1098
1099 /* chain new page in list head to match sg */
1100 first->private = (unsigned long)list;
1101 list = first;
1102 }
1103
1104 first = get_a_page(rq, gfp);
1105 if (!first) {
1106 give_pages(rq, list);
1107 return -ENOMEM;
1108 }
1109 p = page_address(first);
1110
1111 /* rq->sg[0], rq->sg[1] share the same page */
1112 /* a separated rq->sg[0] for header - required in case !any_header_sg */
1113 sg_set_buf(&rq->sg[0], p, vi->hdr_len);
1114
1115 /* rq->sg[1] for data packet, from offset */
1116 offset = sizeof(struct padded_vnet_hdr);
1117 sg_set_buf(&rq->sg[1], p + offset, PAGE_SIZE - offset);
1118
1119 /* chain first in list head */
1120 first->private = (unsigned long)list;
1121 err = virtqueue_add_inbuf(rq->vq, rq->sg, MAX_SKB_FRAGS + 2,
1122 first, gfp);
1123 if (err < 0)
1124 give_pages(rq, first);
1125
1126 return err;
1127 }
1128
1129 static unsigned int get_mergeable_buf_len(struct receive_queue *rq,
1130 struct ewma_pkt_len *avg_pkt_len,
1131 unsigned int room)
1132 {
1133 const size_t hdr_len = sizeof(struct virtio_net_hdr_mrg_rxbuf);
1134 unsigned int len;
1135
1136 if (room)
1137 return PAGE_SIZE - room;
1138
1139 len = hdr_len + clamp_t(unsigned int, ewma_pkt_len_read(avg_pkt_len),
1140 rq->min_buf_len, PAGE_SIZE - hdr_len);
1141
1142 return ALIGN(len, L1_CACHE_BYTES);
1143 }
1144
1145 static int add_recvbuf_mergeable(struct virtnet_info *vi,
1146 struct receive_queue *rq, gfp_t gfp)
1147 {
1148 struct page_frag *alloc_frag = &rq->alloc_frag;
1149 unsigned int headroom = virtnet_get_headroom(vi);
1150 unsigned int tailroom = headroom ? sizeof(struct skb_shared_info) : 0;
1151 unsigned int room = SKB_DATA_ALIGN(headroom + tailroom);
1152 char *buf;
1153 void *ctx;
1154 int err;
1155 unsigned int len, hole;
1156
1157 /* Extra tailroom is needed to satisfy XDP's assumption. This
1158 * means rx frags coalescing won't work, but consider we've
1159 * disabled GSO for XDP, it won't be a big issue.
1160 */
1161 len = get_mergeable_buf_len(rq, &rq->mrg_avg_pkt_len, room);
1162 if (unlikely(!skb_page_frag_refill(len + room, alloc_frag, gfp)))
1163 return -ENOMEM;
1164
1165 buf = (char *)page_address(alloc_frag->page) + alloc_frag->offset;
1166 buf += headroom; /* advance address leaving hole at front of pkt */
1167 get_page(alloc_frag->page);
1168 alloc_frag->offset += len + room;
1169 hole = alloc_frag->size - alloc_frag->offset;
1170 if (hole < len + room) {
1171 /* To avoid internal fragmentation, if there is very likely not
1172 * enough space for another buffer, add the remaining space to
1173 * the current buffer.
1174 */
1175 len += hole;
1176 alloc_frag->offset += hole;
1177 }
1178
1179 sg_init_one(rq->sg, buf, len);
1180 ctx = mergeable_len_to_ctx(len, headroom);
1181 err = virtqueue_add_inbuf_ctx(rq->vq, rq->sg, 1, buf, ctx, gfp);
1182 if (err < 0)
1183 put_page(virt_to_head_page(buf));
1184
1185 return err;
1186 }
1187
1188 /*
1189 * Returns false if we couldn't fill entirely (OOM).
1190 *
1191 * Normally run in the receive path, but can also be run from ndo_open
1192 * before we're receiving packets, or from refill_work which is
1193 * careful to disable receiving (using napi_disable).
1194 */
1195 static bool try_fill_recv(struct virtnet_info *vi, struct receive_queue *rq,
1196 gfp_t gfp)
1197 {
1198 int err;
1199 bool oom;
1200
1201 do {
1202 if (vi->mergeable_rx_bufs)
1203 err = add_recvbuf_mergeable(vi, rq, gfp);
1204 else if (vi->big_packets)
1205 err = add_recvbuf_big(vi, rq, gfp);
1206 else
1207 err = add_recvbuf_small(vi, rq, gfp);
1208
1209 oom = err == -ENOMEM;
1210 if (err)
1211 break;
1212 } while (rq->vq->num_free);
1213 if (virtqueue_kick_prepare(rq->vq) && virtqueue_notify(rq->vq)) {
1214 u64_stats_update_begin(&rq->stats.syncp);
1215 rq->stats.kicks++;
1216 u64_stats_update_end(&rq->stats.syncp);
1217 }
1218
1219 return !oom;
1220 }
1221
1222 static void skb_recv_done(struct virtqueue *rvq)
1223 {
1224 struct virtnet_info *vi = rvq->vdev->priv;
1225 struct receive_queue *rq = &vi->rq[vq2rxq(rvq)];
1226
1227 virtqueue_napi_schedule(&rq->napi, rvq);
1228 }
1229
1230 static void virtnet_napi_enable(struct virtqueue *vq, struct napi_struct *napi)
1231 {
1232 napi_enable(napi);
1233
1234 /* If all buffers were filled by other side before we napi_enabled, we
1235 * won't get another interrupt, so process any outstanding packets now.
1236 * Call local_bh_enable after to trigger softIRQ processing.
1237 */
1238 local_bh_disable();
1239 virtqueue_napi_schedule(napi, vq);
1240 local_bh_enable();
1241 }
1242
1243 static void virtnet_napi_tx_enable(struct virtnet_info *vi,
1244 struct virtqueue *vq,
1245 struct napi_struct *napi)
1246 {
1247 if (!napi->weight)
1248 return;
1249
1250 /* Tx napi touches cachelines on the cpu handling tx interrupts. Only
1251 * enable the feature if this is likely affine with the transmit path.
1252 */
1253 if (!vi->affinity_hint_set) {
1254 napi->weight = 0;
1255 return;
1256 }
1257
1258 return virtnet_napi_enable(vq, napi);
1259 }
1260
1261 static void virtnet_napi_tx_disable(struct napi_struct *napi)
1262 {
1263 if (napi->weight)
1264 napi_disable(napi);
1265 }
1266
1267 static void refill_work(struct work_struct *work)
1268 {
1269 struct virtnet_info *vi =
1270 container_of(work, struct virtnet_info, refill.work);
1271 bool still_empty;
1272 int i;
1273
1274 for (i = 0; i < vi->curr_queue_pairs; i++) {
1275 struct receive_queue *rq = &vi->rq[i];
1276
1277 napi_disable(&rq->napi);
1278 still_empty = !try_fill_recv(vi, rq, GFP_KERNEL);
1279 virtnet_napi_enable(rq->vq, &rq->napi);
1280
1281 /* In theory, this can happen: if we don't get any buffers in
1282 * we will *never* try to fill again.
1283 */
1284 if (still_empty)
1285 schedule_delayed_work(&vi->refill, HZ/2);
1286 }
1287 }
1288
1289 static int virtnet_receive(struct receive_queue *rq, int budget,
1290 unsigned int *xdp_xmit)
1291 {
1292 struct virtnet_info *vi = rq->vq->vdev->priv;
1293 struct virtnet_rq_stats stats = {};
1294 unsigned int len;
1295 void *buf;
1296 int i;
1297
1298 if (!vi->big_packets || vi->mergeable_rx_bufs) {
1299 void *ctx;
1300
1301 while (stats.packets < budget &&
1302 (buf = virtqueue_get_buf_ctx(rq->vq, &len, &ctx))) {
1303 receive_buf(vi, rq, buf, len, ctx, xdp_xmit, &stats);
1304 stats.packets++;
1305 }
1306 } else {
1307 while (stats.packets < budget &&
1308 (buf = virtqueue_get_buf(rq->vq, &len)) != NULL) {
1309 receive_buf(vi, rq, buf, len, NULL, xdp_xmit, &stats);
1310 stats.packets++;
1311 }
1312 }
1313
1314 if (rq->vq->num_free > virtqueue_get_vring_size(rq->vq) / 2) {
1315 if (!try_fill_recv(vi, rq, GFP_ATOMIC))
1316 schedule_delayed_work(&vi->refill, 0);
1317 }
1318
1319 u64_stats_update_begin(&rq->stats.syncp);
1320 for (i = 0; i < VIRTNET_RQ_STATS_LEN; i++) {
1321 size_t offset = virtnet_rq_stats_desc[i].offset;
1322 u64 *item;
1323
1324 item = (u64 *)((u8 *)&rq->stats + offset);
1325 *item += *(u64 *)((u8 *)&stats + offset);
1326 }
1327 u64_stats_update_end(&rq->stats.syncp);
1328
1329 return stats.packets;
1330 }
1331
1332 static void free_old_xmit_skbs(struct send_queue *sq)
1333 {
1334 struct sk_buff *skb;
1335 unsigned int len;
1336 unsigned int packets = 0;
1337 unsigned int bytes = 0;
1338
1339 while ((skb = virtqueue_get_buf(sq->vq, &len)) != NULL) {
1340 pr_debug("Sent skb %p\n", skb);
1341
1342 bytes += skb->len;
1343 packets++;
1344
1345 dev_consume_skb_any(skb);
1346 }
1347
1348 /* Avoid overhead when no packets have been processed
1349 * happens when called speculatively from start_xmit.
1350 */
1351 if (!packets)
1352 return;
1353
1354 u64_stats_update_begin(&sq->stats.syncp);
1355 sq->stats.bytes += bytes;
1356 sq->stats.packets += packets;
1357 u64_stats_update_end(&sq->stats.syncp);
1358 }
1359
1360 static void virtnet_poll_cleantx(struct receive_queue *rq)
1361 {
1362 struct virtnet_info *vi = rq->vq->vdev->priv;
1363 unsigned int index = vq2rxq(rq->vq);
1364 struct send_queue *sq = &vi->sq[index];
1365 struct netdev_queue *txq = netdev_get_tx_queue(vi->dev, index);
1366
1367 if (!sq->napi.weight)
1368 return;
1369
1370 if (__netif_tx_trylock(txq)) {
1371 free_old_xmit_skbs(sq);
1372 __netif_tx_unlock(txq);
1373 }
1374
1375 if (sq->vq->num_free >= 2 + MAX_SKB_FRAGS)
1376 netif_tx_wake_queue(txq);
1377 }
1378
1379 static int virtnet_poll(struct napi_struct *napi, int budget)
1380 {
1381 struct receive_queue *rq =
1382 container_of(napi, struct receive_queue, napi);
1383 struct virtnet_info *vi = rq->vq->vdev->priv;
1384 struct send_queue *sq;
1385 unsigned int received;
1386 unsigned int xdp_xmit = 0;
1387
1388 virtnet_poll_cleantx(rq);
1389
1390 received = virtnet_receive(rq, budget, &xdp_xmit);
1391
1392 /* Out of packets? */
1393 if (received < budget)
1394 virtqueue_napi_complete(napi, rq->vq, received);
1395
1396 if (xdp_xmit & VIRTIO_XDP_REDIR)
1397 xdp_do_flush_map();
1398
1399 if (xdp_xmit & VIRTIO_XDP_TX) {
1400 sq = virtnet_xdp_sq(vi);
1401 if (virtqueue_kick_prepare(sq->vq) && virtqueue_notify(sq->vq)) {
1402 u64_stats_update_begin(&sq->stats.syncp);
1403 sq->stats.kicks++;
1404 u64_stats_update_end(&sq->stats.syncp);
1405 }
1406 }
1407
1408 return received;
1409 }
1410
1411 static int virtnet_open(struct net_device *dev)
1412 {
1413 struct virtnet_info *vi = netdev_priv(dev);
1414 int i, err;
1415
1416 for (i = 0; i < vi->max_queue_pairs; i++) {
1417 if (i < vi->curr_queue_pairs)
1418 /* Make sure we have some buffers: if oom use wq. */
1419 if (!try_fill_recv(vi, &vi->rq[i], GFP_KERNEL))
1420 schedule_delayed_work(&vi->refill, 0);
1421
1422 err = xdp_rxq_info_reg(&vi->rq[i].xdp_rxq, dev, i);
1423 if (err < 0)
1424 return err;
1425
1426 err = xdp_rxq_info_reg_mem_model(&vi->rq[i].xdp_rxq,
1427 MEM_TYPE_PAGE_SHARED, NULL);
1428 if (err < 0) {
1429 xdp_rxq_info_unreg(&vi->rq[i].xdp_rxq);
1430 return err;
1431 }
1432
1433 virtnet_napi_enable(vi->rq[i].vq, &vi->rq[i].napi);
1434 virtnet_napi_tx_enable(vi, vi->sq[i].vq, &vi->sq[i].napi);
1435 }
1436
1437 return 0;
1438 }
1439
1440 static int virtnet_poll_tx(struct napi_struct *napi, int budget)
1441 {
1442 struct send_queue *sq = container_of(napi, struct send_queue, napi);
1443 struct virtnet_info *vi = sq->vq->vdev->priv;
1444 struct netdev_queue *txq = netdev_get_tx_queue(vi->dev, vq2txq(sq->vq));
1445
1446 __netif_tx_lock(txq, raw_smp_processor_id());
1447 free_old_xmit_skbs(sq);
1448 __netif_tx_unlock(txq);
1449
1450 virtqueue_napi_complete(napi, sq->vq, 0);
1451
1452 if (sq->vq->num_free >= 2 + MAX_SKB_FRAGS)
1453 netif_tx_wake_queue(txq);
1454
1455 return 0;
1456 }
1457
1458 static int xmit_skb(struct send_queue *sq, struct sk_buff *skb)
1459 {
1460 struct virtio_net_hdr_mrg_rxbuf *hdr;
1461 const unsigned char *dest = ((struct ethhdr *)skb->data)->h_dest;
1462 struct virtnet_info *vi = sq->vq->vdev->priv;
1463 int num_sg;
1464 unsigned hdr_len = vi->hdr_len;
1465 bool can_push;
1466
1467 pr_debug("%s: xmit %p %pM\n", vi->dev->name, skb, dest);
1468
1469 can_push = vi->any_header_sg &&
1470 !((unsigned long)skb->data & (__alignof__(*hdr) - 1)) &&
1471 !skb_header_cloned(skb) && skb_headroom(skb) >= hdr_len;
1472 /* Even if we can, don't push here yet as this would skew
1473 * csum_start offset below. */
1474 if (can_push)
1475 hdr = (struct virtio_net_hdr_mrg_rxbuf *)(skb->data - hdr_len);
1476 else
1477 hdr = skb_vnet_hdr(skb);
1478
1479 if (virtio_net_hdr_from_skb(skb, &hdr->hdr,
1480 virtio_is_little_endian(vi->vdev), false,
1481 0))
1482 BUG();
1483
1484 if (vi->mergeable_rx_bufs)
1485 hdr->num_buffers = 0;
1486
1487 sg_init_table(sq->sg, skb_shinfo(skb)->nr_frags + (can_push ? 1 : 2));
1488 if (can_push) {
1489 __skb_push(skb, hdr_len);
1490 num_sg = skb_to_sgvec(skb, sq->sg, 0, skb->len);
1491 if (unlikely(num_sg < 0))
1492 return num_sg;
1493 /* Pull header back to avoid skew in tx bytes calculations. */
1494 __skb_pull(skb, hdr_len);
1495 } else {
1496 sg_set_buf(sq->sg, hdr, hdr_len);
1497 num_sg = skb_to_sgvec(skb, sq->sg + 1, 0, skb->len);
1498 if (unlikely(num_sg < 0))
1499 return num_sg;
1500 num_sg++;
1501 }
1502 return virtqueue_add_outbuf(sq->vq, sq->sg, num_sg, skb, GFP_ATOMIC);
1503 }
1504
1505 static netdev_tx_t start_xmit(struct sk_buff *skb, struct net_device *dev)
1506 {
1507 struct virtnet_info *vi = netdev_priv(dev);
1508 int qnum = skb_get_queue_mapping(skb);
1509 struct send_queue *sq = &vi->sq[qnum];
1510 int err;
1511 struct netdev_queue *txq = netdev_get_tx_queue(dev, qnum);
1512 bool kick = !skb->xmit_more;
1513 bool use_napi = sq->napi.weight;
1514
1515 /* Free up any pending old buffers before queueing new ones. */
1516 free_old_xmit_skbs(sq);
1517
1518 if (use_napi && kick)
1519 virtqueue_enable_cb_delayed(sq->vq);
1520
1521 /* timestamp packet in software */
1522 skb_tx_timestamp(skb);
1523
1524 /* Try to transmit */
1525 err = xmit_skb(sq, skb);
1526
1527 /* This should not happen! */
1528 if (unlikely(err)) {
1529 dev->stats.tx_fifo_errors++;
1530 if (net_ratelimit())
1531 dev_warn(&dev->dev,
1532 "Unexpected TXQ (%d) queue failure: %d\n", qnum, err);
1533 dev->stats.tx_dropped++;
1534 dev_kfree_skb_any(skb);
1535 return NETDEV_TX_OK;
1536 }
1537
1538 /* Don't wait up for transmitted skbs to be freed. */
1539 if (!use_napi) {
1540 skb_orphan(skb);
1541 nf_reset(skb);
1542 }
1543
1544 /* If running out of space, stop queue to avoid getting packets that we
1545 * are then unable to transmit.
1546 * An alternative would be to force queuing layer to requeue the skb by
1547 * returning NETDEV_TX_BUSY. However, NETDEV_TX_BUSY should not be
1548 * returned in a normal path of operation: it means that driver is not
1549 * maintaining the TX queue stop/start state properly, and causes
1550 * the stack to do a non-trivial amount of useless work.
1551 * Since most packets only take 1 or 2 ring slots, stopping the queue
1552 * early means 16 slots are typically wasted.
1553 */
1554 if (sq->vq->num_free < 2+MAX_SKB_FRAGS) {
1555 netif_stop_subqueue(dev, qnum);
1556 if (!use_napi &&
1557 unlikely(!virtqueue_enable_cb_delayed(sq->vq))) {
1558 /* More just got used, free them then recheck. */
1559 free_old_xmit_skbs(sq);
1560 if (sq->vq->num_free >= 2+MAX_SKB_FRAGS) {
1561 netif_start_subqueue(dev, qnum);
1562 virtqueue_disable_cb(sq->vq);
1563 }
1564 }
1565 }
1566
1567 if (kick || netif_xmit_stopped(txq)) {
1568 if (virtqueue_kick_prepare(sq->vq) && virtqueue_notify(sq->vq)) {
1569 u64_stats_update_begin(&sq->stats.syncp);
1570 sq->stats.kicks++;
1571 u64_stats_update_end(&sq->stats.syncp);
1572 }
1573 }
1574
1575 return NETDEV_TX_OK;
1576 }
1577
1578 /*
1579 * Send command via the control virtqueue and check status. Commands
1580 * supported by the hypervisor, as indicated by feature bits, should
1581 * never fail unless improperly formatted.
1582 */
1583 static bool virtnet_send_command(struct virtnet_info *vi, u8 class, u8 cmd,
1584 struct scatterlist *out)
1585 {
1586 struct scatterlist *sgs[4], hdr, stat;
1587 unsigned out_num = 0, tmp;
1588
1589 /* Caller should know better */
1590 BUG_ON(!virtio_has_feature(vi->vdev, VIRTIO_NET_F_CTRL_VQ));
1591
1592 vi->ctrl->status = ~0;
1593 vi->ctrl->hdr.class = class;
1594 vi->ctrl->hdr.cmd = cmd;
1595 /* Add header */
1596 sg_init_one(&hdr, &vi->ctrl->hdr, sizeof(vi->ctrl->hdr));
1597 sgs[out_num++] = &hdr;
1598
1599 if (out)
1600 sgs[out_num++] = out;
1601
1602 /* Add return status. */
1603 sg_init_one(&stat, &vi->ctrl->status, sizeof(vi->ctrl->status));
1604 sgs[out_num] = &stat;
1605
1606 BUG_ON(out_num + 1 > ARRAY_SIZE(sgs));
1607 virtqueue_add_sgs(vi->cvq, sgs, out_num, 1, vi, GFP_ATOMIC);
1608
1609 if (unlikely(!virtqueue_kick(vi->cvq)))
1610 return vi->ctrl->status == VIRTIO_NET_OK;
1611
1612 /* Spin for a response, the kick causes an ioport write, trapping
1613 * into the hypervisor, so the request should be handled immediately.
1614 */
1615 while (!virtqueue_get_buf(vi->cvq, &tmp) &&
1616 !virtqueue_is_broken(vi->cvq))
1617 cpu_relax();
1618
1619 return vi->ctrl->status == VIRTIO_NET_OK;
1620 }
1621
1622 static int virtnet_set_mac_address(struct net_device *dev, void *p)
1623 {
1624 struct virtnet_info *vi = netdev_priv(dev);
1625 struct virtio_device *vdev = vi->vdev;
1626 int ret;
1627 struct sockaddr *addr;
1628 struct scatterlist sg;
1629
1630 if (virtio_has_feature(vi->vdev, VIRTIO_NET_F_STANDBY))
1631 return -EOPNOTSUPP;
1632
1633 addr = kmemdup(p, sizeof(*addr), GFP_KERNEL);
1634 if (!addr)
1635 return -ENOMEM;
1636
1637 ret = eth_prepare_mac_addr_change(dev, addr);
1638 if (ret)
1639 goto out;
1640
1641 if (virtio_has_feature(vdev, VIRTIO_NET_F_CTRL_MAC_ADDR)) {
1642 sg_init_one(&sg, addr->sa_data, dev->addr_len);
1643 if (!virtnet_send_command(vi, VIRTIO_NET_CTRL_MAC,
1644 VIRTIO_NET_CTRL_MAC_ADDR_SET, &sg)) {
1645 dev_warn(&vdev->dev,
1646 "Failed to set mac address by vq command.\n");
1647 ret = -EINVAL;
1648 goto out;
1649 }
1650 } else if (virtio_has_feature(vdev, VIRTIO_NET_F_MAC) &&
1651 !virtio_has_feature(vdev, VIRTIO_F_VERSION_1)) {
1652 unsigned int i;
1653
1654 /* Naturally, this has an atomicity problem. */
1655 for (i = 0; i < dev->addr_len; i++)
1656 virtio_cwrite8(vdev,
1657 offsetof(struct virtio_net_config, mac) +
1658 i, addr->sa_data[i]);
1659 }
1660
1661 eth_commit_mac_addr_change(dev, p);
1662 ret = 0;
1663
1664 out:
1665 kfree(addr);
1666 return ret;
1667 }
1668
1669 static void virtnet_stats(struct net_device *dev,
1670 struct rtnl_link_stats64 *tot)
1671 {
1672 struct virtnet_info *vi = netdev_priv(dev);
1673 unsigned int start;
1674 int i;
1675
1676 for (i = 0; i < vi->max_queue_pairs; i++) {
1677 u64 tpackets, tbytes, rpackets, rbytes, rdrops;
1678 struct receive_queue *rq = &vi->rq[i];
1679 struct send_queue *sq = &vi->sq[i];
1680
1681 do {
1682 start = u64_stats_fetch_begin_irq(&sq->stats.syncp);
1683 tpackets = sq->stats.packets;
1684 tbytes = sq->stats.bytes;
1685 } while (u64_stats_fetch_retry_irq(&sq->stats.syncp, start));
1686
1687 do {
1688 start = u64_stats_fetch_begin_irq(&rq->stats.syncp);
1689 rpackets = rq->stats.packets;
1690 rbytes = rq->stats.bytes;
1691 rdrops = rq->stats.drops;
1692 } while (u64_stats_fetch_retry_irq(&rq->stats.syncp, start));
1693
1694 tot->rx_packets += rpackets;
1695 tot->tx_packets += tpackets;
1696 tot->rx_bytes += rbytes;
1697 tot->tx_bytes += tbytes;
1698 tot->rx_dropped += rdrops;
1699 }
1700
1701 tot->tx_dropped = dev->stats.tx_dropped;
1702 tot->tx_fifo_errors = dev->stats.tx_fifo_errors;
1703 tot->rx_length_errors = dev->stats.rx_length_errors;
1704 tot->rx_frame_errors = dev->stats.rx_frame_errors;
1705 }
1706
1707 static void virtnet_ack_link_announce(struct virtnet_info *vi)
1708 {
1709 rtnl_lock();
1710 if (!virtnet_send_command(vi, VIRTIO_NET_CTRL_ANNOUNCE,
1711 VIRTIO_NET_CTRL_ANNOUNCE_ACK, NULL))
1712 dev_warn(&vi->dev->dev, "Failed to ack link announce.\n");
1713 rtnl_unlock();
1714 }
1715
1716 static int _virtnet_set_queues(struct virtnet_info *vi, u16 queue_pairs)
1717 {
1718 struct scatterlist sg;
1719 struct net_device *dev = vi->dev;
1720
1721 if (!vi->has_cvq || !virtio_has_feature(vi->vdev, VIRTIO_NET_F_MQ))
1722 return 0;
1723
1724 vi->ctrl->mq.virtqueue_pairs = cpu_to_virtio16(vi->vdev, queue_pairs);
1725 sg_init_one(&sg, &vi->ctrl->mq, sizeof(vi->ctrl->mq));
1726
1727 if (!virtnet_send_command(vi, VIRTIO_NET_CTRL_MQ,
1728 VIRTIO_NET_CTRL_MQ_VQ_PAIRS_SET, &sg)) {
1729 dev_warn(&dev->dev, "Fail to set num of queue pairs to %d\n",
1730 queue_pairs);
1731 return -EINVAL;
1732 } else {
1733 vi->curr_queue_pairs = queue_pairs;
1734 /* virtnet_open() will refill when device is going to up. */
1735 if (dev->flags & IFF_UP)
1736 schedule_delayed_work(&vi->refill, 0);
1737 }
1738
1739 return 0;
1740 }
1741
1742 static int virtnet_set_queues(struct virtnet_info *vi, u16 queue_pairs)
1743 {
1744 int err;
1745
1746 rtnl_lock();
1747 err = _virtnet_set_queues(vi, queue_pairs);
1748 rtnl_unlock();
1749 return err;
1750 }
1751
1752 static int virtnet_close(struct net_device *dev)
1753 {
1754 struct virtnet_info *vi = netdev_priv(dev);
1755 int i;
1756
1757 /* Make sure refill_work doesn't re-enable napi! */
1758 cancel_delayed_work_sync(&vi->refill);
1759
1760 for (i = 0; i < vi->max_queue_pairs; i++) {
1761 xdp_rxq_info_unreg(&vi->rq[i].xdp_rxq);
1762 napi_disable(&vi->rq[i].napi);
1763 virtnet_napi_tx_disable(&vi->sq[i].napi);
1764 }
1765
1766 return 0;
1767 }
1768
1769 static void virtnet_set_rx_mode(struct net_device *dev)
1770 {
1771 struct virtnet_info *vi = netdev_priv(dev);
1772 struct scatterlist sg[2];
1773 struct virtio_net_ctrl_mac *mac_data;
1774 struct netdev_hw_addr *ha;
1775 int uc_count;
1776 int mc_count;
1777 void *buf;
1778 int i;
1779
1780 /* We can't dynamically set ndo_set_rx_mode, so return gracefully */
1781 if (!virtio_has_feature(vi->vdev, VIRTIO_NET_F_CTRL_RX))
1782 return;
1783
1784 vi->ctrl->promisc = ((dev->flags & IFF_PROMISC) != 0);
1785 vi->ctrl->allmulti = ((dev->flags & IFF_ALLMULTI) != 0);
1786
1787 sg_init_one(sg, &vi->ctrl->promisc, sizeof(vi->ctrl->promisc));
1788
1789 if (!virtnet_send_command(vi, VIRTIO_NET_CTRL_RX,
1790 VIRTIO_NET_CTRL_RX_PROMISC, sg))
1791 dev_warn(&dev->dev, "Failed to %sable promisc mode.\n",
1792 vi->ctrl->promisc ? "en" : "dis");
1793
1794 sg_init_one(sg, &vi->ctrl->allmulti, sizeof(vi->ctrl->allmulti));
1795
1796 if (!virtnet_send_command(vi, VIRTIO_NET_CTRL_RX,
1797 VIRTIO_NET_CTRL_RX_ALLMULTI, sg))
1798 dev_warn(&dev->dev, "Failed to %sable allmulti mode.\n",
1799 vi->ctrl->allmulti ? "en" : "dis");
1800
1801 uc_count = netdev_uc_count(dev);
1802 mc_count = netdev_mc_count(dev);
1803 /* MAC filter - use one buffer for both lists */
1804 buf = kzalloc(((uc_count + mc_count) * ETH_ALEN) +
1805 (2 * sizeof(mac_data->entries)), GFP_ATOMIC);
1806 mac_data = buf;
1807 if (!buf)
1808 return;
1809
1810 sg_init_table(sg, 2);
1811
1812 /* Store the unicast list and count in the front of the buffer */
1813 mac_data->entries = cpu_to_virtio32(vi->vdev, uc_count);
1814 i = 0;
1815 netdev_for_each_uc_addr(ha, dev)
1816 memcpy(&mac_data->macs[i++][0], ha->addr, ETH_ALEN);
1817
1818 sg_set_buf(&sg[0], mac_data,
1819 sizeof(mac_data->entries) + (uc_count * ETH_ALEN));
1820
1821 /* multicast list and count fill the end */
1822 mac_data = (void *)&mac_data->macs[uc_count][0];
1823
1824 mac_data->entries = cpu_to_virtio32(vi->vdev, mc_count);
1825 i = 0;
1826 netdev_for_each_mc_addr(ha, dev)
1827 memcpy(&mac_data->macs[i++][0], ha->addr, ETH_ALEN);
1828
1829 sg_set_buf(&sg[1], mac_data,
1830 sizeof(mac_data->entries) + (mc_count * ETH_ALEN));
1831
1832 if (!virtnet_send_command(vi, VIRTIO_NET_CTRL_MAC,
1833 VIRTIO_NET_CTRL_MAC_TABLE_SET, sg))
1834 dev_warn(&dev->dev, "Failed to set MAC filter table.\n");
1835
1836 kfree(buf);
1837 }
1838
1839 static int virtnet_vlan_rx_add_vid(struct net_device *dev,
1840 __be16 proto, u16 vid)
1841 {
1842 struct virtnet_info *vi = netdev_priv(dev);
1843 struct scatterlist sg;
1844
1845 vi->ctrl->vid = cpu_to_virtio16(vi->vdev, vid);
1846 sg_init_one(&sg, &vi->ctrl->vid, sizeof(vi->ctrl->vid));
1847
1848 if (!virtnet_send_command(vi, VIRTIO_NET_CTRL_VLAN,
1849 VIRTIO_NET_CTRL_VLAN_ADD, &sg))
1850 dev_warn(&dev->dev, "Failed to add VLAN ID %d.\n", vid);
1851 return 0;
1852 }
1853
1854 static int virtnet_vlan_rx_kill_vid(struct net_device *dev,
1855 __be16 proto, u16 vid)
1856 {
1857 struct virtnet_info *vi = netdev_priv(dev);
1858 struct scatterlist sg;
1859
1860 vi->ctrl->vid = cpu_to_virtio16(vi->vdev, vid);
1861 sg_init_one(&sg, &vi->ctrl->vid, sizeof(vi->ctrl->vid));
1862
1863 if (!virtnet_send_command(vi, VIRTIO_NET_CTRL_VLAN,
1864 VIRTIO_NET_CTRL_VLAN_DEL, &sg))
1865 dev_warn(&dev->dev, "Failed to kill VLAN ID %d.\n", vid);
1866 return 0;
1867 }
1868
1869 static void virtnet_clean_affinity(struct virtnet_info *vi, long hcpu)
1870 {
1871 int i;
1872
1873 if (vi->affinity_hint_set) {
1874 for (i = 0; i < vi->max_queue_pairs; i++) {
1875 virtqueue_set_affinity(vi->rq[i].vq, NULL);
1876 virtqueue_set_affinity(vi->sq[i].vq, NULL);
1877 }
1878
1879 vi->affinity_hint_set = false;
1880 }
1881 }
1882
1883 static void virtnet_set_affinity(struct virtnet_info *vi)
1884 {
1885 cpumask_var_t mask;
1886 int stragglers;
1887 int group_size;
1888 int i, j, cpu;
1889 int num_cpu;
1890 int stride;
1891
1892 if (!zalloc_cpumask_var(&mask, GFP_KERNEL)) {
1893 virtnet_clean_affinity(vi, -1);
1894 return;
1895 }
1896
1897 num_cpu = num_online_cpus();
1898 stride = max_t(int, num_cpu / vi->curr_queue_pairs, 1);
1899 stragglers = num_cpu >= vi->curr_queue_pairs ?
1900 num_cpu % vi->curr_queue_pairs :
1901 0;
1902 cpu = cpumask_next(-1, cpu_online_mask);
1903
1904 for (i = 0; i < vi->curr_queue_pairs; i++) {
1905 group_size = stride + (i < stragglers ? 1 : 0);
1906
1907 for (j = 0; j < group_size; j++) {
1908 cpumask_set_cpu(cpu, mask);
1909 cpu = cpumask_next_wrap(cpu, cpu_online_mask,
1910 nr_cpu_ids, false);
1911 }
1912 virtqueue_set_affinity(vi->rq[i].vq, mask);
1913 virtqueue_set_affinity(vi->sq[i].vq, mask);
1914 __netif_set_xps_queue(vi->dev, cpumask_bits(mask), i, false);
1915 cpumask_clear(mask);
1916 }
1917
1918 vi->affinity_hint_set = true;
1919 free_cpumask_var(mask);
1920 }
1921
1922 static int virtnet_cpu_online(unsigned int cpu, struct hlist_node *node)
1923 {
1924 struct virtnet_info *vi = hlist_entry_safe(node, struct virtnet_info,
1925 node);
1926 virtnet_set_affinity(vi);
1927 return 0;
1928 }
1929
1930 static int virtnet_cpu_dead(unsigned int cpu, struct hlist_node *node)
1931 {
1932 struct virtnet_info *vi = hlist_entry_safe(node, struct virtnet_info,
1933 node_dead);
1934 virtnet_set_affinity(vi);
1935 return 0;
1936 }
1937
1938 static int virtnet_cpu_down_prep(unsigned int cpu, struct hlist_node *node)
1939 {
1940 struct virtnet_info *vi = hlist_entry_safe(node, struct virtnet_info,
1941 node);
1942
1943 virtnet_clean_affinity(vi, cpu);
1944 return 0;
1945 }
1946
1947 static enum cpuhp_state virtionet_online;
1948
1949 static int virtnet_cpu_notif_add(struct virtnet_info *vi)
1950 {
1951 int ret;
1952
1953 ret = cpuhp_state_add_instance_nocalls(virtionet_online, &vi->node);
1954 if (ret)
1955 return ret;
1956 ret = cpuhp_state_add_instance_nocalls(CPUHP_VIRT_NET_DEAD,
1957 &vi->node_dead);
1958 if (!ret)
1959 return ret;
1960 cpuhp_state_remove_instance_nocalls(virtionet_online, &vi->node);
1961 return ret;
1962 }
1963
1964 static void virtnet_cpu_notif_remove(struct virtnet_info *vi)
1965 {
1966 cpuhp_state_remove_instance_nocalls(virtionet_online, &vi->node);
1967 cpuhp_state_remove_instance_nocalls(CPUHP_VIRT_NET_DEAD,
1968 &vi->node_dead);
1969 }
1970
1971 static void virtnet_get_ringparam(struct net_device *dev,
1972 struct ethtool_ringparam *ring)
1973 {
1974 struct virtnet_info *vi = netdev_priv(dev);
1975
1976 ring->rx_max_pending = virtqueue_get_vring_size(vi->rq[0].vq);
1977 ring->tx_max_pending = virtqueue_get_vring_size(vi->sq[0].vq);
1978 ring->rx_pending = ring->rx_max_pending;
1979 ring->tx_pending = ring->tx_max_pending;
1980 }
1981
1982
1983 static void virtnet_get_drvinfo(struct net_device *dev,
1984 struct ethtool_drvinfo *info)
1985 {
1986 struct virtnet_info *vi = netdev_priv(dev);
1987 struct virtio_device *vdev = vi->vdev;
1988
1989 strlcpy(info->driver, KBUILD_MODNAME, sizeof(info->driver));
1990 strlcpy(info->version, VIRTNET_DRIVER_VERSION, sizeof(info->version));
1991 strlcpy(info->bus_info, virtio_bus_name(vdev), sizeof(info->bus_info));
1992
1993 }
1994
1995 /* TODO: Eliminate OOO packets during switching */
1996 static int virtnet_set_channels(struct net_device *dev,
1997 struct ethtool_channels *channels)
1998 {
1999 struct virtnet_info *vi = netdev_priv(dev);
2000 u16 queue_pairs = channels->combined_count;
2001 int err;
2002
2003 /* We don't support separate rx/tx channels.
2004 * We don't allow setting 'other' channels.
2005 */
2006 if (channels->rx_count || channels->tx_count || channels->other_count)
2007 return -EINVAL;
2008
2009 if (queue_pairs > vi->max_queue_pairs || queue_pairs == 0)
2010 return -EINVAL;
2011
2012 /* For now we don't support modifying channels while XDP is loaded
2013 * also when XDP is loaded all RX queues have XDP programs so we only
2014 * need to check a single RX queue.
2015 */
2016 if (vi->rq[0].xdp_prog)
2017 return -EINVAL;
2018
2019 get_online_cpus();
2020 err = _virtnet_set_queues(vi, queue_pairs);
2021 if (!err) {
2022 netif_set_real_num_tx_queues(dev, queue_pairs);
2023 netif_set_real_num_rx_queues(dev, queue_pairs);
2024
2025 virtnet_set_affinity(vi);
2026 }
2027 put_online_cpus();
2028
2029 return err;
2030 }
2031
2032 static void virtnet_get_strings(struct net_device *dev, u32 stringset, u8 *data)
2033 {
2034 struct virtnet_info *vi = netdev_priv(dev);
2035 char *p = (char *)data;
2036 unsigned int i, j;
2037
2038 switch (stringset) {
2039 case ETH_SS_STATS:
2040 for (i = 0; i < vi->curr_queue_pairs; i++) {
2041 for (j = 0; j < VIRTNET_RQ_STATS_LEN; j++) {
2042 snprintf(p, ETH_GSTRING_LEN, "rx_queue_%u_%s",
2043 i, virtnet_rq_stats_desc[j].desc);
2044 p += ETH_GSTRING_LEN;
2045 }
2046 }
2047
2048 for (i = 0; i < vi->curr_queue_pairs; i++) {
2049 for (j = 0; j < VIRTNET_SQ_STATS_LEN; j++) {
2050 snprintf(p, ETH_GSTRING_LEN, "tx_queue_%u_%s",
2051 i, virtnet_sq_stats_desc[j].desc);
2052 p += ETH_GSTRING_LEN;
2053 }
2054 }
2055 break;
2056 }
2057 }
2058
2059 static int virtnet_get_sset_count(struct net_device *dev, int sset)
2060 {
2061 struct virtnet_info *vi = netdev_priv(dev);
2062
2063 switch (sset) {
2064 case ETH_SS_STATS:
2065 return vi->curr_queue_pairs * (VIRTNET_RQ_STATS_LEN +
2066 VIRTNET_SQ_STATS_LEN);
2067 default:
2068 return -EOPNOTSUPP;
2069 }
2070 }
2071
2072 static void virtnet_get_ethtool_stats(struct net_device *dev,
2073 struct ethtool_stats *stats, u64 *data)
2074 {
2075 struct virtnet_info *vi = netdev_priv(dev);
2076 unsigned int idx = 0, start, i, j;
2077 const u8 *stats_base;
2078 size_t offset;
2079
2080 for (i = 0; i < vi->curr_queue_pairs; i++) {
2081 struct receive_queue *rq = &vi->rq[i];
2082
2083 stats_base = (u8 *)&rq->stats;
2084 do {
2085 start = u64_stats_fetch_begin_irq(&rq->stats.syncp);
2086 for (j = 0; j < VIRTNET_RQ_STATS_LEN; j++) {
2087 offset = virtnet_rq_stats_desc[j].offset;
2088 data[idx + j] = *(u64 *)(stats_base + offset);
2089 }
2090 } while (u64_stats_fetch_retry_irq(&rq->stats.syncp, start));
2091 idx += VIRTNET_RQ_STATS_LEN;
2092 }
2093
2094 for (i = 0; i < vi->curr_queue_pairs; i++) {
2095 struct send_queue *sq = &vi->sq[i];
2096
2097 stats_base = (u8 *)&sq->stats;
2098 do {
2099 start = u64_stats_fetch_begin_irq(&sq->stats.syncp);
2100 for (j = 0; j < VIRTNET_SQ_STATS_LEN; j++) {
2101 offset = virtnet_sq_stats_desc[j].offset;
2102 data[idx + j] = *(u64 *)(stats_base + offset);
2103 }
2104 } while (u64_stats_fetch_retry_irq(&sq->stats.syncp, start));
2105 idx += VIRTNET_SQ_STATS_LEN;
2106 }
2107 }
2108
2109 static void virtnet_get_channels(struct net_device *dev,
2110 struct ethtool_channels *channels)
2111 {
2112 struct virtnet_info *vi = netdev_priv(dev);
2113
2114 channels->combined_count = vi->curr_queue_pairs;
2115 channels->max_combined = vi->max_queue_pairs;
2116 channels->max_other = 0;
2117 channels->rx_count = 0;
2118 channels->tx_count = 0;
2119 channels->other_count = 0;
2120 }
2121
2122 /* Check if the user is trying to change anything besides speed/duplex */
2123 static bool
2124 virtnet_validate_ethtool_cmd(const struct ethtool_link_ksettings *cmd)
2125 {
2126 struct ethtool_link_ksettings diff1 = *cmd;
2127 struct ethtool_link_ksettings diff2 = {};
2128
2129 /* cmd is always set so we need to clear it, validate the port type
2130 * and also without autonegotiation we can ignore advertising
2131 */
2132 diff1.base.speed = 0;
2133 diff2.base.port = PORT_OTHER;
2134 ethtool_link_ksettings_zero_link_mode(&diff1, advertising);
2135 diff1.base.duplex = 0;
2136 diff1.base.cmd = 0;
2137 diff1.base.link_mode_masks_nwords = 0;
2138
2139 return !memcmp(&diff1.base, &diff2.base, sizeof(diff1.base)) &&
2140 bitmap_empty(diff1.link_modes.supported,
2141 __ETHTOOL_LINK_MODE_MASK_NBITS) &&
2142 bitmap_empty(diff1.link_modes.advertising,
2143 __ETHTOOL_LINK_MODE_MASK_NBITS) &&
2144 bitmap_empty(diff1.link_modes.lp_advertising,
2145 __ETHTOOL_LINK_MODE_MASK_NBITS);
2146 }
2147
2148 static int virtnet_set_link_ksettings(struct net_device *dev,
2149 const struct ethtool_link_ksettings *cmd)
2150 {
2151 struct virtnet_info *vi = netdev_priv(dev);
2152 u32 speed;
2153
2154 speed = cmd->base.speed;
2155 /* don't allow custom speed and duplex */
2156 if (!ethtool_validate_speed(speed) ||
2157 !ethtool_validate_duplex(cmd->base.duplex) ||
2158 !virtnet_validate_ethtool_cmd(cmd))
2159 return -EINVAL;
2160 vi->speed = speed;
2161 vi->duplex = cmd->base.duplex;
2162
2163 return 0;
2164 }
2165
2166 static int virtnet_get_link_ksettings(struct net_device *dev,
2167 struct ethtool_link_ksettings *cmd)
2168 {
2169 struct virtnet_info *vi = netdev_priv(dev);
2170
2171 cmd->base.speed = vi->speed;
2172 cmd->base.duplex = vi->duplex;
2173 cmd->base.port = PORT_OTHER;
2174
2175 return 0;
2176 }
2177
2178 static int virtnet_set_coalesce(struct net_device *dev,
2179 struct ethtool_coalesce *ec)
2180 {
2181 struct ethtool_coalesce ec_default = {
2182 .cmd = ETHTOOL_SCOALESCE,
2183 .rx_max_coalesced_frames = 1,
2184 };
2185 struct virtnet_info *vi = netdev_priv(dev);
2186 int i, napi_weight;
2187
2188 if (ec->tx_max_coalesced_frames > 1)
2189 return -EINVAL;
2190
2191 ec_default.tx_max_coalesced_frames = ec->tx_max_coalesced_frames;
2192 napi_weight = ec->tx_max_coalesced_frames ? NAPI_POLL_WEIGHT : 0;
2193
2194 /* disallow changes to fields not explicitly tested above */
2195 if (memcmp(ec, &ec_default, sizeof(ec_default)))
2196 return -EINVAL;
2197
2198 if (napi_weight ^ vi->sq[0].napi.weight) {
2199 if (dev->flags & IFF_UP)
2200 return -EBUSY;
2201 for (i = 0; i < vi->max_queue_pairs; i++)
2202 vi->sq[i].napi.weight = napi_weight;
2203 }
2204
2205 return 0;
2206 }
2207
2208 static int virtnet_get_coalesce(struct net_device *dev,
2209 struct ethtool_coalesce *ec)
2210 {
2211 struct ethtool_coalesce ec_default = {
2212 .cmd = ETHTOOL_GCOALESCE,
2213 .rx_max_coalesced_frames = 1,
2214 };
2215 struct virtnet_info *vi = netdev_priv(dev);
2216
2217 memcpy(ec, &ec_default, sizeof(ec_default));
2218
2219 if (vi->sq[0].napi.weight)
2220 ec->tx_max_coalesced_frames = 1;
2221
2222 return 0;
2223 }
2224
2225 static void virtnet_init_settings(struct net_device *dev)
2226 {
2227 struct virtnet_info *vi = netdev_priv(dev);
2228
2229 vi->speed = SPEED_UNKNOWN;
2230 vi->duplex = DUPLEX_UNKNOWN;
2231 }
2232
2233 static void virtnet_update_settings(struct virtnet_info *vi)
2234 {
2235 u32 speed;
2236 u8 duplex;
2237
2238 if (!virtio_has_feature(vi->vdev, VIRTIO_NET_F_SPEED_DUPLEX))
2239 return;
2240
2241 speed = virtio_cread32(vi->vdev, offsetof(struct virtio_net_config,
2242 speed));
2243 if (ethtool_validate_speed(speed))
2244 vi->speed = speed;
2245 duplex = virtio_cread8(vi->vdev, offsetof(struct virtio_net_config,
2246 duplex));
2247 if (ethtool_validate_duplex(duplex))
2248 vi->duplex = duplex;
2249 }
2250
2251 static const struct ethtool_ops virtnet_ethtool_ops = {
2252 .get_drvinfo = virtnet_get_drvinfo,
2253 .get_link = ethtool_op_get_link,
2254 .get_ringparam = virtnet_get_ringparam,
2255 .get_strings = virtnet_get_strings,
2256 .get_sset_count = virtnet_get_sset_count,
2257 .get_ethtool_stats = virtnet_get_ethtool_stats,
2258 .set_channels = virtnet_set_channels,
2259 .get_channels = virtnet_get_channels,
2260 .get_ts_info = ethtool_op_get_ts_info,
2261 .get_link_ksettings = virtnet_get_link_ksettings,
2262 .set_link_ksettings = virtnet_set_link_ksettings,
2263 .set_coalesce = virtnet_set_coalesce,
2264 .get_coalesce = virtnet_get_coalesce,
2265 };
2266
2267 static void virtnet_freeze_down(struct virtio_device *vdev)
2268 {
2269 struct virtnet_info *vi = vdev->priv;
2270 int i;
2271
2272 /* Make sure no work handler is accessing the device */
2273 flush_work(&vi->config_work);
2274
2275 netif_tx_lock_bh(vi->dev);
2276 netif_device_detach(vi->dev);
2277 netif_tx_unlock_bh(vi->dev);
2278 cancel_delayed_work_sync(&vi->refill);
2279
2280 if (netif_running(vi->dev)) {
2281 for (i = 0; i < vi->max_queue_pairs; i++) {
2282 napi_disable(&vi->rq[i].napi);
2283 virtnet_napi_tx_disable(&vi->sq[i].napi);
2284 }
2285 }
2286 }
2287
2288 static int init_vqs(struct virtnet_info *vi);
2289
2290 static int virtnet_restore_up(struct virtio_device *vdev)
2291 {
2292 struct virtnet_info *vi = vdev->priv;
2293 int err, i;
2294
2295 err = init_vqs(vi);
2296 if (err)
2297 return err;
2298
2299 virtio_device_ready(vdev);
2300
2301 if (netif_running(vi->dev)) {
2302 for (i = 0; i < vi->curr_queue_pairs; i++)
2303 if (!try_fill_recv(vi, &vi->rq[i], GFP_KERNEL))
2304 schedule_delayed_work(&vi->refill, 0);
2305
2306 for (i = 0; i < vi->max_queue_pairs; i++) {
2307 virtnet_napi_enable(vi->rq[i].vq, &vi->rq[i].napi);
2308 virtnet_napi_tx_enable(vi, vi->sq[i].vq,
2309 &vi->sq[i].napi);
2310 }
2311 }
2312
2313 netif_tx_lock_bh(vi->dev);
2314 netif_device_attach(vi->dev);
2315 netif_tx_unlock_bh(vi->dev);
2316 return err;
2317 }
2318
2319 static int virtnet_set_guest_offloads(struct virtnet_info *vi, u64 offloads)
2320 {
2321 struct scatterlist sg;
2322 vi->ctrl->offloads = cpu_to_virtio64(vi->vdev, offloads);
2323
2324 sg_init_one(&sg, &vi->ctrl->offloads, sizeof(vi->ctrl->offloads));
2325
2326 if (!virtnet_send_command(vi, VIRTIO_NET_CTRL_GUEST_OFFLOADS,
2327 VIRTIO_NET_CTRL_GUEST_OFFLOADS_SET, &sg)) {
2328 dev_warn(&vi->dev->dev, "Fail to set guest offload. \n");
2329 return -EINVAL;
2330 }
2331
2332 return 0;
2333 }
2334
2335 static int virtnet_clear_guest_offloads(struct virtnet_info *vi)
2336 {
2337 u64 offloads = 0;
2338
2339 if (!vi->guest_offloads)
2340 return 0;
2341
2342 return virtnet_set_guest_offloads(vi, offloads);
2343 }
2344
2345 static int virtnet_restore_guest_offloads(struct virtnet_info *vi)
2346 {
2347 u64 offloads = vi->guest_offloads;
2348
2349 if (!vi->guest_offloads)
2350 return 0;
2351
2352 return virtnet_set_guest_offloads(vi, offloads);
2353 }
2354
2355 static int virtnet_xdp_set(struct net_device *dev, struct bpf_prog *prog,
2356 struct netlink_ext_ack *extack)
2357 {
2358 unsigned long int max_sz = PAGE_SIZE - sizeof(struct padded_vnet_hdr);
2359 struct virtnet_info *vi = netdev_priv(dev);
2360 struct bpf_prog *old_prog;
2361 u16 xdp_qp = 0, curr_qp;
2362 int i, err;
2363
2364 if (!virtio_has_feature(vi->vdev, VIRTIO_NET_F_CTRL_GUEST_OFFLOADS)
2365 && (virtio_has_feature(vi->vdev, VIRTIO_NET_F_GUEST_TSO4) ||
2366 virtio_has_feature(vi->vdev, VIRTIO_NET_F_GUEST_TSO6) ||
2367 virtio_has_feature(vi->vdev, VIRTIO_NET_F_GUEST_ECN) ||
2368 virtio_has_feature(vi->vdev, VIRTIO_NET_F_GUEST_UFO) ||
2369 virtio_has_feature(vi->vdev, VIRTIO_NET_F_GUEST_CSUM))) {
2370 NL_SET_ERR_MSG_MOD(extack, "Can't set XDP while host is implementing LRO/CSUM, disable LRO/CSUM first");
2371 return -EOPNOTSUPP;
2372 }
2373
2374 if (vi->mergeable_rx_bufs && !vi->any_header_sg) {
2375 NL_SET_ERR_MSG_MOD(extack, "XDP expects header/data in single page, any_header_sg required");
2376 return -EINVAL;
2377 }
2378
2379 if (dev->mtu > max_sz) {
2380 NL_SET_ERR_MSG_MOD(extack, "MTU too large to enable XDP");
2381 netdev_warn(dev, "XDP requires MTU less than %lu\n", max_sz);
2382 return -EINVAL;
2383 }
2384
2385 curr_qp = vi->curr_queue_pairs - vi->xdp_queue_pairs;
2386 if (prog)
2387 xdp_qp = nr_cpu_ids;
2388
2389 /* XDP requires extra queues for XDP_TX */
2390 if (curr_qp + xdp_qp > vi->max_queue_pairs) {
2391 NL_SET_ERR_MSG_MOD(extack, "Too few free TX rings available");
2392 netdev_warn(dev, "request %i queues but max is %i\n",
2393 curr_qp + xdp_qp, vi->max_queue_pairs);
2394 return -ENOMEM;
2395 }
2396
2397 if (prog) {
2398 prog = bpf_prog_add(prog, vi->max_queue_pairs - 1);
2399 if (IS_ERR(prog))
2400 return PTR_ERR(prog);
2401 }
2402
2403 /* Make sure NAPI is not using any XDP TX queues for RX. */
2404 if (netif_running(dev))
2405 for (i = 0; i < vi->max_queue_pairs; i++)
2406 napi_disable(&vi->rq[i].napi);
2407
2408 netif_set_real_num_rx_queues(dev, curr_qp + xdp_qp);
2409 err = _virtnet_set_queues(vi, curr_qp + xdp_qp);
2410 if (err)
2411 goto err;
2412 vi->xdp_queue_pairs = xdp_qp;
2413
2414 for (i = 0; i < vi->max_queue_pairs; i++) {
2415 old_prog = rtnl_dereference(vi->rq[i].xdp_prog);
2416 rcu_assign_pointer(vi->rq[i].xdp_prog, prog);
2417 if (i == 0) {
2418 if (!old_prog)
2419 virtnet_clear_guest_offloads(vi);
2420 if (!prog)
2421 virtnet_restore_guest_offloads(vi);
2422 }
2423 if (old_prog)
2424 bpf_prog_put(old_prog);
2425 if (netif_running(dev))
2426 virtnet_napi_enable(vi->rq[i].vq, &vi->rq[i].napi);
2427 }
2428
2429 return 0;
2430
2431 err:
2432 for (i = 0; i < vi->max_queue_pairs; i++)
2433 virtnet_napi_enable(vi->rq[i].vq, &vi->rq[i].napi);
2434 if (prog)
2435 bpf_prog_sub(prog, vi->max_queue_pairs - 1);
2436 return err;
2437 }
2438
2439 static u32 virtnet_xdp_query(struct net_device *dev)
2440 {
2441 struct virtnet_info *vi = netdev_priv(dev);
2442 const struct bpf_prog *xdp_prog;
2443 int i;
2444
2445 for (i = 0; i < vi->max_queue_pairs; i++) {
2446 xdp_prog = rtnl_dereference(vi->rq[i].xdp_prog);
2447 if (xdp_prog)
2448 return xdp_prog->aux->id;
2449 }
2450 return 0;
2451 }
2452
2453 static int virtnet_xdp(struct net_device *dev, struct netdev_bpf *xdp)
2454 {
2455 switch (xdp->command) {
2456 case XDP_SETUP_PROG:
2457 return virtnet_xdp_set(dev, xdp->prog, xdp->extack);
2458 case XDP_QUERY_PROG:
2459 xdp->prog_id = virtnet_xdp_query(dev);
2460 return 0;
2461 default:
2462 return -EINVAL;
2463 }
2464 }
2465
2466 static int virtnet_get_phys_port_name(struct net_device *dev, char *buf,
2467 size_t len)
2468 {
2469 struct virtnet_info *vi = netdev_priv(dev);
2470 int ret;
2471
2472 if (!virtio_has_feature(vi->vdev, VIRTIO_NET_F_STANDBY))
2473 return -EOPNOTSUPP;
2474
2475 ret = snprintf(buf, len, "sby");
2476 if (ret >= len)
2477 return -EOPNOTSUPP;
2478
2479 return 0;
2480 }
2481
2482 static const struct net_device_ops virtnet_netdev = {
2483 .ndo_open = virtnet_open,
2484 .ndo_stop = virtnet_close,
2485 .ndo_start_xmit = start_xmit,
2486 .ndo_validate_addr = eth_validate_addr,
2487 .ndo_set_mac_address = virtnet_set_mac_address,
2488 .ndo_set_rx_mode = virtnet_set_rx_mode,
2489 .ndo_get_stats64 = virtnet_stats,
2490 .ndo_vlan_rx_add_vid = virtnet_vlan_rx_add_vid,
2491 .ndo_vlan_rx_kill_vid = virtnet_vlan_rx_kill_vid,
2492 .ndo_bpf = virtnet_xdp,
2493 .ndo_xdp_xmit = virtnet_xdp_xmit,
2494 .ndo_features_check = passthru_features_check,
2495 .ndo_get_phys_port_name = virtnet_get_phys_port_name,
2496 };
2497
2498 static void virtnet_config_changed_work(struct work_struct *work)
2499 {
2500 struct virtnet_info *vi =
2501 container_of(work, struct virtnet_info, config_work);
2502 u16 v;
2503
2504 if (virtio_cread_feature(vi->vdev, VIRTIO_NET_F_STATUS,
2505 struct virtio_net_config, status, &v) < 0)
2506 return;
2507
2508 if (v & VIRTIO_NET_S_ANNOUNCE) {
2509 netdev_notify_peers(vi->dev);
2510 virtnet_ack_link_announce(vi);
2511 }
2512
2513 /* Ignore unknown (future) status bits */
2514 v &= VIRTIO_NET_S_LINK_UP;
2515
2516 if (vi->status == v)
2517 return;
2518
2519 vi->status = v;
2520
2521 if (vi->status & VIRTIO_NET_S_LINK_UP) {
2522 virtnet_update_settings(vi);
2523 netif_carrier_on(vi->dev);
2524 netif_tx_wake_all_queues(vi->dev);
2525 } else {
2526 netif_carrier_off(vi->dev);
2527 netif_tx_stop_all_queues(vi->dev);
2528 }
2529 }
2530
2531 static void virtnet_config_changed(struct virtio_device *vdev)
2532 {
2533 struct virtnet_info *vi = vdev->priv;
2534
2535 schedule_work(&vi->config_work);
2536 }
2537
2538 static void virtnet_free_queues(struct virtnet_info *vi)
2539 {
2540 int i;
2541
2542 for (i = 0; i < vi->max_queue_pairs; i++) {
2543 napi_hash_del(&vi->rq[i].napi);
2544 netif_napi_del(&vi->rq[i].napi);
2545 netif_napi_del(&vi->sq[i].napi);
2546 }
2547
2548 /* We called napi_hash_del() before netif_napi_del(),
2549 * we need to respect an RCU grace period before freeing vi->rq
2550 */
2551 synchronize_net();
2552
2553 kfree(vi->rq);
2554 kfree(vi->sq);
2555 kfree(vi->ctrl);
2556 }
2557
2558 static void _free_receive_bufs(struct virtnet_info *vi)
2559 {
2560 struct bpf_prog *old_prog;
2561 int i;
2562
2563 for (i = 0; i < vi->max_queue_pairs; i++) {
2564 while (vi->rq[i].pages)
2565 __free_pages(get_a_page(&vi->rq[i], GFP_KERNEL), 0);
2566
2567 old_prog = rtnl_dereference(vi->rq[i].xdp_prog);
2568 RCU_INIT_POINTER(vi->rq[i].xdp_prog, NULL);
2569 if (old_prog)
2570 bpf_prog_put(old_prog);
2571 }
2572 }
2573
2574 static void free_receive_bufs(struct virtnet_info *vi)
2575 {
2576 rtnl_lock();
2577 _free_receive_bufs(vi);
2578 rtnl_unlock();
2579 }
2580
2581 static void free_receive_page_frags(struct virtnet_info *vi)
2582 {
2583 int i;
2584 for (i = 0; i < vi->max_queue_pairs; i++)
2585 if (vi->rq[i].alloc_frag.page)
2586 put_page(vi->rq[i].alloc_frag.page);
2587 }
2588
2589 static bool is_xdp_raw_buffer_queue(struct virtnet_info *vi, int q)
2590 {
2591 if (q < (vi->curr_queue_pairs - vi->xdp_queue_pairs))
2592 return false;
2593 else if (q < vi->curr_queue_pairs)
2594 return true;
2595 else
2596 return false;
2597 }
2598
2599 static void free_unused_bufs(struct virtnet_info *vi)
2600 {
2601 void *buf;
2602 int i;
2603
2604 for (i = 0; i < vi->max_queue_pairs; i++) {
2605 struct virtqueue *vq = vi->sq[i].vq;
2606 while ((buf = virtqueue_detach_unused_buf(vq)) != NULL) {
2607 if (!is_xdp_raw_buffer_queue(vi, i))
2608 dev_kfree_skb(buf);
2609 else
2610 put_page(virt_to_head_page(buf));
2611 }
2612 }
2613
2614 for (i = 0; i < vi->max_queue_pairs; i++) {
2615 struct virtqueue *vq = vi->rq[i].vq;
2616
2617 while ((buf = virtqueue_detach_unused_buf(vq)) != NULL) {
2618 if (vi->mergeable_rx_bufs) {
2619 put_page(virt_to_head_page(buf));
2620 } else if (vi->big_packets) {
2621 give_pages(&vi->rq[i], buf);
2622 } else {
2623 put_page(virt_to_head_page(buf));
2624 }
2625 }
2626 }
2627 }
2628
2629 static void virtnet_del_vqs(struct virtnet_info *vi)
2630 {
2631 struct virtio_device *vdev = vi->vdev;
2632
2633 virtnet_clean_affinity(vi, -1);
2634
2635 vdev->config->del_vqs(vdev);
2636
2637 virtnet_free_queues(vi);
2638 }
2639
2640 /* How large should a single buffer be so a queue full of these can fit at
2641 * least one full packet?
2642 * Logic below assumes the mergeable buffer header is used.
2643 */
2644 static unsigned int mergeable_min_buf_len(struct virtnet_info *vi, struct virtqueue *vq)
2645 {
2646 const unsigned int hdr_len = sizeof(struct virtio_net_hdr_mrg_rxbuf);
2647 unsigned int rq_size = virtqueue_get_vring_size(vq);
2648 unsigned int packet_len = vi->big_packets ? IP_MAX_MTU : vi->dev->max_mtu;
2649 unsigned int buf_len = hdr_len + ETH_HLEN + VLAN_HLEN + packet_len;
2650 unsigned int min_buf_len = DIV_ROUND_UP(buf_len, rq_size);
2651
2652 return max(max(min_buf_len, hdr_len) - hdr_len,
2653 (unsigned int)GOOD_PACKET_LEN);
2654 }
2655
2656 static int virtnet_find_vqs(struct virtnet_info *vi)
2657 {
2658 vq_callback_t **callbacks;
2659 struct virtqueue **vqs;
2660 int ret = -ENOMEM;
2661 int i, total_vqs;
2662 const char **names;
2663 bool *ctx;
2664
2665 /* We expect 1 RX virtqueue followed by 1 TX virtqueue, followed by
2666 * possible N-1 RX/TX queue pairs used in multiqueue mode, followed by
2667 * possible control vq.
2668 */
2669 total_vqs = vi->max_queue_pairs * 2 +
2670 virtio_has_feature(vi->vdev, VIRTIO_NET_F_CTRL_VQ);
2671
2672 /* Allocate space for find_vqs parameters */
2673 vqs = kcalloc(total_vqs, sizeof(*vqs), GFP_KERNEL);
2674 if (!vqs)
2675 goto err_vq;
2676 callbacks = kmalloc_array(total_vqs, sizeof(*callbacks), GFP_KERNEL);
2677 if (!callbacks)
2678 goto err_callback;
2679 names = kmalloc_array(total_vqs, sizeof(*names), GFP_KERNEL);
2680 if (!names)
2681 goto err_names;
2682 if (!vi->big_packets || vi->mergeable_rx_bufs) {
2683 ctx = kcalloc(total_vqs, sizeof(*ctx), GFP_KERNEL);
2684 if (!ctx)
2685 goto err_ctx;
2686 } else {
2687 ctx = NULL;
2688 }
2689
2690 /* Parameters for control virtqueue, if any */
2691 if (vi->has_cvq) {
2692 callbacks[total_vqs - 1] = NULL;
2693 names[total_vqs - 1] = "control";
2694 }
2695
2696 /* Allocate/initialize parameters for send/receive virtqueues */
2697 for (i = 0; i < vi->max_queue_pairs; i++) {
2698 callbacks[rxq2vq(i)] = skb_recv_done;
2699 callbacks[txq2vq(i)] = skb_xmit_done;
2700 sprintf(vi->rq[i].name, "input.%d", i);
2701 sprintf(vi->sq[i].name, "output.%d", i);
2702 names[rxq2vq(i)] = vi->rq[i].name;
2703 names[txq2vq(i)] = vi->sq[i].name;
2704 if (ctx)
2705 ctx[rxq2vq(i)] = true;
2706 }
2707
2708 ret = vi->vdev->config->find_vqs(vi->vdev, total_vqs, vqs, callbacks,
2709 names, ctx, NULL);
2710 if (ret)
2711 goto err_find;
2712
2713 if (vi->has_cvq) {
2714 vi->cvq = vqs[total_vqs - 1];
2715 if (virtio_has_feature(vi->vdev, VIRTIO_NET_F_CTRL_VLAN))
2716 vi->dev->features |= NETIF_F_HW_VLAN_CTAG_FILTER;
2717 }
2718
2719 for (i = 0; i < vi->max_queue_pairs; i++) {
2720 vi->rq[i].vq = vqs[rxq2vq(i)];
2721 vi->rq[i].min_buf_len = mergeable_min_buf_len(vi, vi->rq[i].vq);
2722 vi->sq[i].vq = vqs[txq2vq(i)];
2723 }
2724
2725 /* run here: ret == 0. */
2726
2727
2728 err_find:
2729 kfree(ctx);
2730 err_ctx:
2731 kfree(names);
2732 err_names:
2733 kfree(callbacks);
2734 err_callback:
2735 kfree(vqs);
2736 err_vq:
2737 return ret;
2738 }
2739
2740 static int virtnet_alloc_queues(struct virtnet_info *vi)
2741 {
2742 int i;
2743
2744 vi->ctrl = kzalloc(sizeof(*vi->ctrl), GFP_KERNEL);
2745 if (!vi->ctrl)
2746 goto err_ctrl;
2747 vi->sq = kcalloc(vi->max_queue_pairs, sizeof(*vi->sq), GFP_KERNEL);
2748 if (!vi->sq)
2749 goto err_sq;
2750 vi->rq = kcalloc(vi->max_queue_pairs, sizeof(*vi->rq), GFP_KERNEL);
2751 if (!vi->rq)
2752 goto err_rq;
2753
2754 INIT_DELAYED_WORK(&vi->refill, refill_work);
2755 for (i = 0; i < vi->max_queue_pairs; i++) {
2756 vi->rq[i].pages = NULL;
2757 netif_napi_add(vi->dev, &vi->rq[i].napi, virtnet_poll,
2758 napi_weight);
2759 netif_tx_napi_add(vi->dev, &vi->sq[i].napi, virtnet_poll_tx,
2760 napi_tx ? napi_weight : 0);
2761
2762 sg_init_table(vi->rq[i].sg, ARRAY_SIZE(vi->rq[i].sg));
2763 ewma_pkt_len_init(&vi->rq[i].mrg_avg_pkt_len);
2764 sg_init_table(vi->sq[i].sg, ARRAY_SIZE(vi->sq[i].sg));
2765
2766 u64_stats_init(&vi->rq[i].stats.syncp);
2767 u64_stats_init(&vi->sq[i].stats.syncp);
2768 }
2769
2770 return 0;
2771
2772 err_rq:
2773 kfree(vi->sq);
2774 err_sq:
2775 kfree(vi->ctrl);
2776 err_ctrl:
2777 return -ENOMEM;
2778 }
2779
2780 static int init_vqs(struct virtnet_info *vi)
2781 {
2782 int ret;
2783
2784 /* Allocate send & receive queues */
2785 ret = virtnet_alloc_queues(vi);
2786 if (ret)
2787 goto err;
2788
2789 ret = virtnet_find_vqs(vi);
2790 if (ret)
2791 goto err_free;
2792
2793 get_online_cpus();
2794 virtnet_set_affinity(vi);
2795 put_online_cpus();
2796
2797 return 0;
2798
2799 err_free:
2800 virtnet_free_queues(vi);
2801 err:
2802 return ret;
2803 }
2804
2805 #ifdef CONFIG_SYSFS
2806 static ssize_t mergeable_rx_buffer_size_show(struct netdev_rx_queue *queue,
2807 char *buf)
2808 {
2809 struct virtnet_info *vi = netdev_priv(queue->dev);
2810 unsigned int queue_index = get_netdev_rx_queue_index(queue);
2811 unsigned int headroom = virtnet_get_headroom(vi);
2812 unsigned int tailroom = headroom ? sizeof(struct skb_shared_info) : 0;
2813 struct ewma_pkt_len *avg;
2814
2815 BUG_ON(queue_index >= vi->max_queue_pairs);
2816 avg = &vi->rq[queue_index].mrg_avg_pkt_len;
2817 return sprintf(buf, "%u\n",
2818 get_mergeable_buf_len(&vi->rq[queue_index], avg,
2819 SKB_DATA_ALIGN(headroom + tailroom)));
2820 }
2821
2822 static struct rx_queue_attribute mergeable_rx_buffer_size_attribute =
2823 __ATTR_RO(mergeable_rx_buffer_size);
2824
2825 static struct attribute *virtio_net_mrg_rx_attrs[] = {
2826 &mergeable_rx_buffer_size_attribute.attr,
2827 NULL
2828 };
2829
2830 static const struct attribute_group virtio_net_mrg_rx_group = {
2831 .name = "virtio_net",
2832 .attrs = virtio_net_mrg_rx_attrs
2833 };
2834 #endif
2835
2836 static bool virtnet_fail_on_feature(struct virtio_device *vdev,
2837 unsigned int fbit,
2838 const char *fname, const char *dname)
2839 {
2840 if (!virtio_has_feature(vdev, fbit))
2841 return false;
2842
2843 dev_err(&vdev->dev, "device advertises feature %s but not %s",
2844 fname, dname);
2845
2846 return true;
2847 }
2848
2849 #define VIRTNET_FAIL_ON(vdev, fbit, dbit) \
2850 virtnet_fail_on_feature(vdev, fbit, #fbit, dbit)
2851
2852 static bool virtnet_validate_features(struct virtio_device *vdev)
2853 {
2854 if (!virtio_has_feature(vdev, VIRTIO_NET_F_CTRL_VQ) &&
2855 (VIRTNET_FAIL_ON(vdev, VIRTIO_NET_F_CTRL_RX,
2856 "VIRTIO_NET_F_CTRL_VQ") ||
2857 VIRTNET_FAIL_ON(vdev, VIRTIO_NET_F_CTRL_VLAN,
2858 "VIRTIO_NET_F_CTRL_VQ") ||
2859 VIRTNET_FAIL_ON(vdev, VIRTIO_NET_F_GUEST_ANNOUNCE,
2860 "VIRTIO_NET_F_CTRL_VQ") ||
2861 VIRTNET_FAIL_ON(vdev, VIRTIO_NET_F_MQ, "VIRTIO_NET_F_CTRL_VQ") ||
2862 VIRTNET_FAIL_ON(vdev, VIRTIO_NET_F_CTRL_MAC_ADDR,
2863 "VIRTIO_NET_F_CTRL_VQ"))) {
2864 return false;
2865 }
2866
2867 return true;
2868 }
2869
2870 #define MIN_MTU ETH_MIN_MTU
2871 #define MAX_MTU ETH_MAX_MTU
2872
2873 static int virtnet_validate(struct virtio_device *vdev)
2874 {
2875 if (!vdev->config->get) {
2876 dev_err(&vdev->dev, "%s failure: config access disabled\n",
2877 __func__);
2878 return -EINVAL;
2879 }
2880
2881 if (!virtnet_validate_features(vdev))
2882 return -EINVAL;
2883
2884 if (virtio_has_feature(vdev, VIRTIO_NET_F_MTU)) {
2885 int mtu = virtio_cread16(vdev,
2886 offsetof(struct virtio_net_config,
2887 mtu));
2888 if (mtu < MIN_MTU)
2889 __virtio_clear_bit(vdev, VIRTIO_NET_F_MTU);
2890 }
2891
2892 return 0;
2893 }
2894
2895 static int virtnet_probe(struct virtio_device *vdev)
2896 {
2897 int i, err = -ENOMEM;
2898 struct net_device *dev;
2899 struct virtnet_info *vi;
2900 u16 max_queue_pairs;
2901 int mtu;
2902
2903 /* Find if host supports multiqueue virtio_net device */
2904 err = virtio_cread_feature(vdev, VIRTIO_NET_F_MQ,
2905 struct virtio_net_config,
2906 max_virtqueue_pairs, &max_queue_pairs);
2907
2908 /* We need at least 2 queue's */
2909 if (err || max_queue_pairs < VIRTIO_NET_CTRL_MQ_VQ_PAIRS_MIN ||
2910 max_queue_pairs > VIRTIO_NET_CTRL_MQ_VQ_PAIRS_MAX ||
2911 !virtio_has_feature(vdev, VIRTIO_NET_F_CTRL_VQ))
2912 max_queue_pairs = 1;
2913
2914 /* Allocate ourselves a network device with room for our info */
2915 dev = alloc_etherdev_mq(sizeof(struct virtnet_info), max_queue_pairs);
2916 if (!dev)
2917 return -ENOMEM;
2918
2919 /* Set up network device as normal. */
2920 dev->priv_flags |= IFF_UNICAST_FLT | IFF_LIVE_ADDR_CHANGE;
2921 dev->netdev_ops = &virtnet_netdev;
2922 dev->features = NETIF_F_HIGHDMA;
2923
2924 dev->ethtool_ops = &virtnet_ethtool_ops;
2925 SET_NETDEV_DEV(dev, &vdev->dev);
2926
2927 /* Do we support "hardware" checksums? */
2928 if (virtio_has_feature(vdev, VIRTIO_NET_F_CSUM)) {
2929 /* This opens up the world of extra features. */
2930 dev->hw_features |= NETIF_F_HW_CSUM | NETIF_F_SG;
2931 if (csum)
2932 dev->features |= NETIF_F_HW_CSUM | NETIF_F_SG;
2933
2934 if (virtio_has_feature(vdev, VIRTIO_NET_F_GSO)) {
2935 dev->hw_features |= NETIF_F_TSO
2936 | NETIF_F_TSO_ECN | NETIF_F_TSO6;
2937 }
2938 /* Individual feature bits: what can host handle? */
2939 if (virtio_has_feature(vdev, VIRTIO_NET_F_HOST_TSO4))
2940 dev->hw_features |= NETIF_F_TSO;
2941 if (virtio_has_feature(vdev, VIRTIO_NET_F_HOST_TSO6))
2942 dev->hw_features |= NETIF_F_TSO6;
2943 if (virtio_has_feature(vdev, VIRTIO_NET_F_HOST_ECN))
2944 dev->hw_features |= NETIF_F_TSO_ECN;
2945
2946 dev->features |= NETIF_F_GSO_ROBUST;
2947
2948 if (gso)
2949 dev->features |= dev->hw_features & NETIF_F_ALL_TSO;
2950 /* (!csum && gso) case will be fixed by register_netdev() */
2951 }
2952 if (virtio_has_feature(vdev, VIRTIO_NET_F_GUEST_CSUM))
2953 dev->features |= NETIF_F_RXCSUM;
2954
2955 dev->vlan_features = dev->features;
2956
2957 /* MTU range: 68 - 65535 */
2958 dev->min_mtu = MIN_MTU;
2959 dev->max_mtu = MAX_MTU;
2960
2961 /* Configuration may specify what MAC to use. Otherwise random. */
2962 if (virtio_has_feature(vdev, VIRTIO_NET_F_MAC))
2963 virtio_cread_bytes(vdev,
2964 offsetof(struct virtio_net_config, mac),
2965 dev->dev_addr, dev->addr_len);
2966 else
2967 eth_hw_addr_random(dev);
2968
2969 /* Set up our device-specific information */
2970 vi = netdev_priv(dev);
2971 vi->dev = dev;
2972 vi->vdev = vdev;
2973 vdev->priv = vi;
2974
2975 INIT_WORK(&vi->config_work, virtnet_config_changed_work);
2976
2977 /* If we can receive ANY GSO packets, we must allocate large ones. */
2978 if (virtio_has_feature(vdev, VIRTIO_NET_F_GUEST_TSO4) ||
2979 virtio_has_feature(vdev, VIRTIO_NET_F_GUEST_TSO6) ||
2980 virtio_has_feature(vdev, VIRTIO_NET_F_GUEST_ECN) ||
2981 virtio_has_feature(vdev, VIRTIO_NET_F_GUEST_UFO))
2982 vi->big_packets = true;
2983
2984 if (virtio_has_feature(vdev, VIRTIO_NET_F_MRG_RXBUF))
2985 vi->mergeable_rx_bufs = true;
2986
2987 if (virtio_has_feature(vdev, VIRTIO_NET_F_MRG_RXBUF) ||
2988 virtio_has_feature(vdev, VIRTIO_F_VERSION_1))
2989 vi->hdr_len = sizeof(struct virtio_net_hdr_mrg_rxbuf);
2990 else
2991 vi->hdr_len = sizeof(struct virtio_net_hdr);
2992
2993 if (virtio_has_feature(vdev, VIRTIO_F_ANY_LAYOUT) ||
2994 virtio_has_feature(vdev, VIRTIO_F_VERSION_1))
2995 vi->any_header_sg = true;
2996
2997 if (virtio_has_feature(vdev, VIRTIO_NET_F_CTRL_VQ))
2998 vi->has_cvq = true;
2999
3000 if (virtio_has_feature(vdev, VIRTIO_NET_F_MTU)) {
3001 mtu = virtio_cread16(vdev,
3002 offsetof(struct virtio_net_config,
3003 mtu));
3004 if (mtu < dev->min_mtu) {
3005 /* Should never trigger: MTU was previously validated
3006 * in virtnet_validate.
3007 */
3008 dev_err(&vdev->dev, "device MTU appears to have changed "
3009 "it is now %d < %d", mtu, dev->min_mtu);
3010 goto free;
3011 }
3012
3013 dev->mtu = mtu;
3014 dev->max_mtu = mtu;
3015
3016 /* TODO: size buffers correctly in this case. */
3017 if (dev->mtu > ETH_DATA_LEN)
3018 vi->big_packets = true;
3019 }
3020
3021 if (vi->any_header_sg)
3022 dev->needed_headroom = vi->hdr_len;
3023
3024 /* Enable multiqueue by default */
3025 if (num_online_cpus() >= max_queue_pairs)
3026 vi->curr_queue_pairs = max_queue_pairs;
3027 else
3028 vi->curr_queue_pairs = num_online_cpus();
3029 vi->max_queue_pairs = max_queue_pairs;
3030
3031 /* Allocate/initialize the rx/tx queues, and invoke find_vqs */
3032 err = init_vqs(vi);
3033 if (err)
3034 goto free;
3035
3036 #ifdef CONFIG_SYSFS
3037 if (vi->mergeable_rx_bufs)
3038 dev->sysfs_rx_queue_group = &virtio_net_mrg_rx_group;
3039 #endif
3040 netif_set_real_num_tx_queues(dev, vi->curr_queue_pairs);
3041 netif_set_real_num_rx_queues(dev, vi->curr_queue_pairs);
3042
3043 virtnet_init_settings(dev);
3044
3045 if (virtio_has_feature(vdev, VIRTIO_NET_F_STANDBY)) {
3046 vi->failover = net_failover_create(vi->dev);
3047 if (IS_ERR(vi->failover)) {
3048 err = PTR_ERR(vi->failover);
3049 goto free_vqs;
3050 }
3051 }
3052
3053 err = register_netdev(dev);
3054 if (err) {
3055 pr_debug("virtio_net: registering device failed\n");
3056 goto free_failover;
3057 }
3058
3059 virtio_device_ready(vdev);
3060
3061 err = virtnet_cpu_notif_add(vi);
3062 if (err) {
3063 pr_debug("virtio_net: registering cpu notifier failed\n");
3064 goto free_unregister_netdev;
3065 }
3066
3067 virtnet_set_queues(vi, vi->curr_queue_pairs);
3068
3069 /* Assume link up if device can't report link status,
3070 otherwise get link status from config. */
3071 netif_carrier_off(dev);
3072 if (virtio_has_feature(vi->vdev, VIRTIO_NET_F_STATUS)) {
3073 schedule_work(&vi->config_work);
3074 } else {
3075 vi->status = VIRTIO_NET_S_LINK_UP;
3076 virtnet_update_settings(vi);
3077 netif_carrier_on(dev);
3078 }
3079
3080 for (i = 0; i < ARRAY_SIZE(guest_offloads); i++)
3081 if (virtio_has_feature(vi->vdev, guest_offloads[i]))
3082 set_bit(guest_offloads[i], &vi->guest_offloads);
3083
3084 pr_debug("virtnet: registered device %s with %d RX and TX vq's\n",
3085 dev->name, max_queue_pairs);
3086
3087 return 0;
3088
3089 free_unregister_netdev:
3090 vi->vdev->config->reset(vdev);
3091
3092 unregister_netdev(dev);
3093 free_failover:
3094 net_failover_destroy(vi->failover);
3095 free_vqs:
3096 cancel_delayed_work_sync(&vi->refill);
3097 free_receive_page_frags(vi);
3098 virtnet_del_vqs(vi);
3099 free:
3100 free_netdev(dev);
3101 return err;
3102 }
3103
3104 static void remove_vq_common(struct virtnet_info *vi)
3105 {
3106 vi->vdev->config->reset(vi->vdev);
3107
3108 /* Free unused buffers in both send and recv, if any. */
3109 free_unused_bufs(vi);
3110
3111 free_receive_bufs(vi);
3112
3113 free_receive_page_frags(vi);
3114
3115 virtnet_del_vqs(vi);
3116 }
3117
3118 static void virtnet_remove(struct virtio_device *vdev)
3119 {
3120 struct virtnet_info *vi = vdev->priv;
3121
3122 virtnet_cpu_notif_remove(vi);
3123
3124 /* Make sure no work handler is accessing the device. */
3125 flush_work(&vi->config_work);
3126
3127 unregister_netdev(vi->dev);
3128
3129 net_failover_destroy(vi->failover);
3130
3131 remove_vq_common(vi);
3132
3133 free_netdev(vi->dev);
3134 }
3135
3136 static __maybe_unused int virtnet_freeze(struct virtio_device *vdev)
3137 {
3138 struct virtnet_info *vi = vdev->priv;
3139
3140 virtnet_cpu_notif_remove(vi);
3141 virtnet_freeze_down(vdev);
3142 remove_vq_common(vi);
3143
3144 return 0;
3145 }
3146
3147 static __maybe_unused int virtnet_restore(struct virtio_device *vdev)
3148 {
3149 struct virtnet_info *vi = vdev->priv;
3150 int err;
3151
3152 err = virtnet_restore_up(vdev);
3153 if (err)
3154 return err;
3155 virtnet_set_queues(vi, vi->curr_queue_pairs);
3156
3157 err = virtnet_cpu_notif_add(vi);
3158 if (err)
3159 return err;
3160
3161 return 0;
3162 }
3163
3164 static struct virtio_device_id id_table[] = {
3165 { VIRTIO_ID_NET, VIRTIO_DEV_ANY_ID },
3166 { 0 },
3167 };
3168
3169 #define VIRTNET_FEATURES \
3170 VIRTIO_NET_F_CSUM, VIRTIO_NET_F_GUEST_CSUM, \
3171 VIRTIO_NET_F_MAC, \
3172 VIRTIO_NET_F_HOST_TSO4, VIRTIO_NET_F_HOST_UFO, VIRTIO_NET_F_HOST_TSO6, \
3173 VIRTIO_NET_F_HOST_ECN, VIRTIO_NET_F_GUEST_TSO4, VIRTIO_NET_F_GUEST_TSO6, \
3174 VIRTIO_NET_F_GUEST_ECN, VIRTIO_NET_F_GUEST_UFO, \
3175 VIRTIO_NET_F_MRG_RXBUF, VIRTIO_NET_F_STATUS, VIRTIO_NET_F_CTRL_VQ, \
3176 VIRTIO_NET_F_CTRL_RX, VIRTIO_NET_F_CTRL_VLAN, \
3177 VIRTIO_NET_F_GUEST_ANNOUNCE, VIRTIO_NET_F_MQ, \
3178 VIRTIO_NET_F_CTRL_MAC_ADDR, \
3179 VIRTIO_NET_F_MTU, VIRTIO_NET_F_CTRL_GUEST_OFFLOADS, \
3180 VIRTIO_NET_F_SPEED_DUPLEX, VIRTIO_NET_F_STANDBY
3181
3182 static unsigned int features[] = {
3183 VIRTNET_FEATURES,
3184 };
3185
3186 static unsigned int features_legacy[] = {
3187 VIRTNET_FEATURES,
3188 VIRTIO_NET_F_GSO,
3189 VIRTIO_F_ANY_LAYOUT,
3190 };
3191
3192 static struct virtio_driver virtio_net_driver = {
3193 .feature_table = features,
3194 .feature_table_size = ARRAY_SIZE(features),
3195 .feature_table_legacy = features_legacy,
3196 .feature_table_size_legacy = ARRAY_SIZE(features_legacy),
3197 .driver.name = KBUILD_MODNAME,
3198 .driver.owner = THIS_MODULE,
3199 .id_table = id_table,
3200 .validate = virtnet_validate,
3201 .probe = virtnet_probe,
3202 .remove = virtnet_remove,
3203 .config_changed = virtnet_config_changed,
3204 #ifdef CONFIG_PM_SLEEP
3205 .freeze = virtnet_freeze,
3206 .restore = virtnet_restore,
3207 #endif
3208 };
3209
3210 static __init int virtio_net_driver_init(void)
3211 {
3212 int ret;
3213
3214 ret = cpuhp_setup_state_multi(CPUHP_AP_ONLINE_DYN, "virtio/net:online",
3215 virtnet_cpu_online,
3216 virtnet_cpu_down_prep);
3217 if (ret < 0)
3218 goto out;
3219 virtionet_online = ret;
3220 ret = cpuhp_setup_state_multi(CPUHP_VIRT_NET_DEAD, "virtio/net:dead",
3221 NULL, virtnet_cpu_dead);
3222 if (ret)
3223 goto err_dead;
3224
3225 ret = register_virtio_driver(&virtio_net_driver);
3226 if (ret)
3227 goto err_virtio;
3228 return 0;
3229 err_virtio:
3230 cpuhp_remove_multi_state(CPUHP_VIRT_NET_DEAD);
3231 err_dead:
3232 cpuhp_remove_multi_state(virtionet_online);
3233 out:
3234 return ret;
3235 }
3236 module_init(virtio_net_driver_init);
3237
3238 static __exit void virtio_net_driver_exit(void)
3239 {
3240 unregister_virtio_driver(&virtio_net_driver);
3241 cpuhp_remove_multi_state(CPUHP_VIRT_NET_DEAD);
3242 cpuhp_remove_multi_state(virtionet_online);
3243 }
3244 module_exit(virtio_net_driver_exit);
3245
3246 MODULE_DEVICE_TABLE(virtio, id_table);
3247 MODULE_DESCRIPTION("Virtio network driver");
3248 MODULE_LICENSE("GPL");
Linux with added FPC-III support