Version:  2.0.40 2.2.26 2.4.37 3.12 3.13 3.14 3.15 3.16 3.17 3.18 3.19 4.0 4.1 4.2 4.3 4.4 4.5 4.6 4.7 4.8 4.9

Linux/drivers/net/virtio_net.c

  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/scatterlist.h>
 26 #include <linux/if_vlan.h>
 27 #include <linux/slab.h>
 28 #include <linux/cpu.h>
 29 #include <linux/average.h>
 30 #include <net/busy_poll.h>
 31 
 32 static int napi_weight = NAPI_POLL_WEIGHT;
 33 module_param(napi_weight, int, 0444);
 34 
 35 static bool csum = true, gso = true;
 36 module_param(csum, bool, 0444);
 37 module_param(gso, bool, 0444);
 38 
 39 /* FIXME: MTU in config. */
 40 #define GOOD_PACKET_LEN (ETH_HLEN + VLAN_HLEN + ETH_DATA_LEN)
 41 #define GOOD_COPY_LEN   128
 42 
 43 /* RX packet size EWMA. The average packet size is used to determine the packet
 44  * buffer size when refilling RX rings. As the entire RX ring may be refilled
 45  * at once, the weight is chosen so that the EWMA will be insensitive to short-
 46  * term, transient changes in packet size.
 47  */
 48 DECLARE_EWMA(pkt_len, 1, 64)
 49 
 50 /* Minimum alignment for mergeable packet buffers. */
 51 #define MERGEABLE_BUFFER_ALIGN max(L1_CACHE_BYTES, 256)
 52 
 53 #define VIRTNET_DRIVER_VERSION "1.0.0"
 54 
 55 struct virtnet_stats {
 56         struct u64_stats_sync tx_syncp;
 57         struct u64_stats_sync rx_syncp;
 58         u64 tx_bytes;
 59         u64 tx_packets;
 60 
 61         u64 rx_bytes;
 62         u64 rx_packets;
 63 };
 64 
 65 /* Internal representation of a send virtqueue */
 66 struct send_queue {
 67         /* Virtqueue associated with this send _queue */
 68         struct virtqueue *vq;
 69 
 70         /* TX: fragments + linear part + virtio header */
 71         struct scatterlist sg[MAX_SKB_FRAGS + 2];
 72 
 73         /* Name of the send queue: output.$index */
 74         char name[40];
 75 };
 76 
 77 /* Internal representation of a receive virtqueue */
 78 struct receive_queue {
 79         /* Virtqueue associated with this receive_queue */
 80         struct virtqueue *vq;
 81 
 82         struct napi_struct napi;
 83 
 84         /* Chain pages by the private ptr. */
 85         struct page *pages;
 86 
 87         /* Average packet length for mergeable receive buffers. */
 88         struct ewma_pkt_len mrg_avg_pkt_len;
 89 
 90         /* Page frag for packet buffer allocation. */
 91         struct page_frag alloc_frag;
 92 
 93         /* RX: fragments + linear part + virtio header */
 94         struct scatterlist sg[MAX_SKB_FRAGS + 2];
 95 
 96         /* Name of this receive queue: input.$index */
 97         char name[40];
 98 };
 99 
100 struct virtnet_info {
101         struct virtio_device *vdev;
102         struct virtqueue *cvq;
103         struct net_device *dev;
104         struct send_queue *sq;
105         struct receive_queue *rq;
106         unsigned int status;
107 
108         /* Max # of queue pairs supported by the device */
109         u16 max_queue_pairs;
110 
111         /* # of queue pairs currently used by the driver */
112         u16 curr_queue_pairs;
113 
114         /* I like... big packets and I cannot lie! */
115         bool big_packets;
116 
117         /* Host will merge rx buffers for big packets (shake it! shake it!) */
118         bool mergeable_rx_bufs;
119 
120         /* Has control virtqueue */
121         bool has_cvq;
122 
123         /* Host can handle any s/g split between our header and packet data */
124         bool any_header_sg;
125 
126         /* Packet virtio header size */
127         u8 hdr_len;
128 
129         /* Active statistics */
130         struct virtnet_stats __percpu *stats;
131 
132         /* Work struct for refilling if we run low on memory. */
133         struct delayed_work refill;
134 
135         /* Work struct for config space updates */
136         struct work_struct config_work;
137 
138         /* Does the affinity hint is set for virtqueues? */
139         bool affinity_hint_set;
140 
141         /* CPU hotplug instances for online & dead */
142         struct hlist_node node;
143         struct hlist_node node_dead;
144 
145         /* Control VQ buffers: protected by the rtnl lock */
146         struct virtio_net_ctrl_hdr ctrl_hdr;
147         virtio_net_ctrl_ack ctrl_status;
148         struct virtio_net_ctrl_mq ctrl_mq;
149         u8 ctrl_promisc;
150         u8 ctrl_allmulti;
151         u16 ctrl_vid;
152 
153         /* Ethtool settings */
154         u8 duplex;
155         u32 speed;
156 };
157 
158 struct padded_vnet_hdr {
159         struct virtio_net_hdr_mrg_rxbuf hdr;
160         /*
161          * hdr is in a separate sg buffer, and data sg buffer shares same page
162          * with this header sg. This padding makes next sg 16 byte aligned
163          * after the header.
164          */
165         char padding[4];
166 };
167 
168 /* Converting between virtqueue no. and kernel tx/rx queue no.
169  * 0:rx0 1:tx0 2:rx1 3:tx1 ... 2N:rxN 2N+1:txN 2N+2:cvq
170  */
171 static int vq2txq(struct virtqueue *vq)
172 {
173         return (vq->index - 1) / 2;
174 }
175 
176 static int txq2vq(int txq)
177 {
178         return txq * 2 + 1;
179 }
180 
181 static int vq2rxq(struct virtqueue *vq)
182 {
183         return vq->index / 2;
184 }
185 
186 static int rxq2vq(int rxq)
187 {
188         return rxq * 2;
189 }
190 
191 static inline struct virtio_net_hdr_mrg_rxbuf *skb_vnet_hdr(struct sk_buff *skb)
192 {
193         return (struct virtio_net_hdr_mrg_rxbuf *)skb->cb;
194 }
195 
196 /*
197  * private is used to chain pages for big packets, put the whole
198  * most recent used list in the beginning for reuse
199  */
200 static void give_pages(struct receive_queue *rq, struct page *page)
201 {
202         struct page *end;
203 
204         /* Find end of list, sew whole thing into vi->rq.pages. */
205         for (end = page; end->private; end = (struct page *)end->private);
206         end->private = (unsigned long)rq->pages;
207         rq->pages = page;
208 }
209 
210 static struct page *get_a_page(struct receive_queue *rq, gfp_t gfp_mask)
211 {
212         struct page *p = rq->pages;
213 
214         if (p) {
215                 rq->pages = (struct page *)p->private;
216                 /* clear private here, it is used to chain pages */
217                 p->private = 0;
218         } else
219                 p = alloc_page(gfp_mask);
220         return p;
221 }
222 
223 static void skb_xmit_done(struct virtqueue *vq)
224 {
225         struct virtnet_info *vi = vq->vdev->priv;
226 
227         /* Suppress further interrupts. */
228         virtqueue_disable_cb(vq);
229 
230         /* We were probably waiting for more output buffers. */
231         netif_wake_subqueue(vi->dev, vq2txq(vq));
232 }
233 
234 static unsigned int mergeable_ctx_to_buf_truesize(unsigned long mrg_ctx)
235 {
236         unsigned int truesize = mrg_ctx & (MERGEABLE_BUFFER_ALIGN - 1);
237         return (truesize + 1) * MERGEABLE_BUFFER_ALIGN;
238 }
239 
240 static void *mergeable_ctx_to_buf_address(unsigned long mrg_ctx)
241 {
242         return (void *)(mrg_ctx & -MERGEABLE_BUFFER_ALIGN);
243 
244 }
245 
246 static unsigned long mergeable_buf_to_ctx(void *buf, unsigned int truesize)
247 {
248         unsigned int size = truesize / MERGEABLE_BUFFER_ALIGN;
249         return (unsigned long)buf | (size - 1);
250 }
251 
252 /* Called from bottom half context */
253 static struct sk_buff *page_to_skb(struct virtnet_info *vi,
254                                    struct receive_queue *rq,
255                                    struct page *page, unsigned int offset,
256                                    unsigned int len, unsigned int truesize)
257 {
258         struct sk_buff *skb;
259         struct virtio_net_hdr_mrg_rxbuf *hdr;
260         unsigned int copy, hdr_len, hdr_padded_len;
261         char *p;
262 
263         p = page_address(page) + offset;
264 
265         /* copy small packet so we can reuse these pages for small data */
266         skb = napi_alloc_skb(&rq->napi, GOOD_COPY_LEN);
267         if (unlikely(!skb))
268                 return NULL;
269 
270         hdr = skb_vnet_hdr(skb);
271 
272         hdr_len = vi->hdr_len;
273         if (vi->mergeable_rx_bufs)
274                 hdr_padded_len = sizeof *hdr;
275         else
276                 hdr_padded_len = sizeof(struct padded_vnet_hdr);
277 
278         memcpy(hdr, p, hdr_len);
279 
280         len -= hdr_len;
281         offset += hdr_padded_len;
282         p += hdr_padded_len;
283 
284         copy = len;
285         if (copy > skb_tailroom(skb))
286                 copy = skb_tailroom(skb);
287         memcpy(skb_put(skb, copy), p, copy);
288 
289         len -= copy;
290         offset += copy;
291 
292         if (vi->mergeable_rx_bufs) {
293                 if (len)
294                         skb_add_rx_frag(skb, 0, page, offset, len, truesize);
295                 else
296                         put_page(page);
297                 return skb;
298         }
299 
300         /*
301          * Verify that we can indeed put this data into a skb.
302          * This is here to handle cases when the device erroneously
303          * tries to receive more than is possible. This is usually
304          * the case of a broken device.
305          */
306         if (unlikely(len > MAX_SKB_FRAGS * PAGE_SIZE)) {
307                 net_dbg_ratelimited("%s: too much data\n", skb->dev->name);
308                 dev_kfree_skb(skb);
309                 return NULL;
310         }
311         BUG_ON(offset >= PAGE_SIZE);
312         while (len) {
313                 unsigned int frag_size = min((unsigned)PAGE_SIZE - offset, len);
314                 skb_add_rx_frag(skb, skb_shinfo(skb)->nr_frags, page, offset,
315                                 frag_size, truesize);
316                 len -= frag_size;
317                 page = (struct page *)page->private;
318                 offset = 0;
319         }
320 
321         if (page)
322                 give_pages(rq, page);
323 
324         return skb;
325 }
326 
327 static struct sk_buff *receive_small(struct virtnet_info *vi, void *buf, unsigned int len)
328 {
329         struct sk_buff * skb = buf;
330 
331         len -= vi->hdr_len;
332         skb_trim(skb, len);
333 
334         return skb;
335 }
336 
337 static struct sk_buff *receive_big(struct net_device *dev,
338                                    struct virtnet_info *vi,
339                                    struct receive_queue *rq,
340                                    void *buf,
341                                    unsigned int len)
342 {
343         struct page *page = buf;
344         struct sk_buff *skb = page_to_skb(vi, rq, page, 0, len, PAGE_SIZE);
345 
346         if (unlikely(!skb))
347                 goto err;
348 
349         return skb;
350 
351 err:
352         dev->stats.rx_dropped++;
353         give_pages(rq, page);
354         return NULL;
355 }
356 
357 static struct sk_buff *receive_mergeable(struct net_device *dev,
358                                          struct virtnet_info *vi,
359                                          struct receive_queue *rq,
360                                          unsigned long ctx,
361                                          unsigned int len)
362 {
363         void *buf = mergeable_ctx_to_buf_address(ctx);
364         struct virtio_net_hdr_mrg_rxbuf *hdr = buf;
365         u16 num_buf = virtio16_to_cpu(vi->vdev, hdr->num_buffers);
366         struct page *page = virt_to_head_page(buf);
367         int offset = buf - page_address(page);
368         unsigned int truesize = max(len, mergeable_ctx_to_buf_truesize(ctx));
369 
370         struct sk_buff *head_skb = page_to_skb(vi, rq, page, offset, len,
371                                                truesize);
372         struct sk_buff *curr_skb = head_skb;
373 
374         if (unlikely(!curr_skb))
375                 goto err_skb;
376         while (--num_buf) {
377                 int num_skb_frags;
378 
379                 ctx = (unsigned long)virtqueue_get_buf(rq->vq, &len);
380                 if (unlikely(!ctx)) {
381                         pr_debug("%s: rx error: %d buffers out of %d missing\n",
382                                  dev->name, num_buf,
383                                  virtio16_to_cpu(vi->vdev,
384                                                  hdr->num_buffers));
385                         dev->stats.rx_length_errors++;
386                         goto err_buf;
387                 }
388 
389                 buf = mergeable_ctx_to_buf_address(ctx);
390                 page = virt_to_head_page(buf);
391 
392                 num_skb_frags = skb_shinfo(curr_skb)->nr_frags;
393                 if (unlikely(num_skb_frags == MAX_SKB_FRAGS)) {
394                         struct sk_buff *nskb = alloc_skb(0, GFP_ATOMIC);
395 
396                         if (unlikely(!nskb))
397                                 goto err_skb;
398                         if (curr_skb == head_skb)
399                                 skb_shinfo(curr_skb)->frag_list = nskb;
400                         else
401                                 curr_skb->next = nskb;
402                         curr_skb = nskb;
403                         head_skb->truesize += nskb->truesize;
404                         num_skb_frags = 0;
405                 }
406                 truesize = max(len, mergeable_ctx_to_buf_truesize(ctx));
407                 if (curr_skb != head_skb) {
408                         head_skb->data_len += len;
409                         head_skb->len += len;
410                         head_skb->truesize += truesize;
411                 }
412                 offset = buf - page_address(page);
413                 if (skb_can_coalesce(curr_skb, num_skb_frags, page, offset)) {
414                         put_page(page);
415                         skb_coalesce_rx_frag(curr_skb, num_skb_frags - 1,
416                                              len, truesize);
417                 } else {
418                         skb_add_rx_frag(curr_skb, num_skb_frags, page,
419                                         offset, len, truesize);
420                 }
421         }
422 
423         ewma_pkt_len_add(&rq->mrg_avg_pkt_len, head_skb->len);
424         return head_skb;
425 
426 err_skb:
427         put_page(page);
428         while (--num_buf) {
429                 ctx = (unsigned long)virtqueue_get_buf(rq->vq, &len);
430                 if (unlikely(!ctx)) {
431                         pr_debug("%s: rx error: %d buffers missing\n",
432                                  dev->name, num_buf);
433                         dev->stats.rx_length_errors++;
434                         break;
435                 }
436                 page = virt_to_head_page(mergeable_ctx_to_buf_address(ctx));
437                 put_page(page);
438         }
439 err_buf:
440         dev->stats.rx_dropped++;
441         dev_kfree_skb(head_skb);
442         return NULL;
443 }
444 
445 static void receive_buf(struct virtnet_info *vi, struct receive_queue *rq,
446                         void *buf, unsigned int len)
447 {
448         struct net_device *dev = vi->dev;
449         struct virtnet_stats *stats = this_cpu_ptr(vi->stats);
450         struct sk_buff *skb;
451         struct virtio_net_hdr_mrg_rxbuf *hdr;
452 
453         if (unlikely(len < vi->hdr_len + ETH_HLEN)) {
454                 pr_debug("%s: short packet %i\n", dev->name, len);
455                 dev->stats.rx_length_errors++;
456                 if (vi->mergeable_rx_bufs) {
457                         unsigned long ctx = (unsigned long)buf;
458                         void *base = mergeable_ctx_to_buf_address(ctx);
459                         put_page(virt_to_head_page(base));
460                 } else if (vi->big_packets) {
461                         give_pages(rq, buf);
462                 } else {
463                         dev_kfree_skb(buf);
464                 }
465                 return;
466         }
467 
468         if (vi->mergeable_rx_bufs)
469                 skb = receive_mergeable(dev, vi, rq, (unsigned long)buf, len);
470         else if (vi->big_packets)
471                 skb = receive_big(dev, vi, rq, buf, len);
472         else
473                 skb = receive_small(vi, buf, len);
474 
475         if (unlikely(!skb))
476                 return;
477 
478         hdr = skb_vnet_hdr(skb);
479 
480         u64_stats_update_begin(&stats->rx_syncp);
481         stats->rx_bytes += skb->len;
482         stats->rx_packets++;
483         u64_stats_update_end(&stats->rx_syncp);
484 
485         if (hdr->hdr.flags & VIRTIO_NET_HDR_F_DATA_VALID)
486                 skb->ip_summed = CHECKSUM_UNNECESSARY;
487 
488         if (virtio_net_hdr_to_skb(skb, &hdr->hdr,
489                                   virtio_is_little_endian(vi->vdev))) {
490                 net_warn_ratelimited("%s: bad gso: type: %u, size: %u\n",
491                                      dev->name, hdr->hdr.gso_type,
492                                      hdr->hdr.gso_size);
493                 goto frame_err;
494         }
495 
496         skb->protocol = eth_type_trans(skb, dev);
497         pr_debug("Receiving skb proto 0x%04x len %i type %i\n",
498                  ntohs(skb->protocol), skb->len, skb->pkt_type);
499 
500         napi_gro_receive(&rq->napi, skb);
501         return;
502 
503 frame_err:
504         dev->stats.rx_frame_errors++;
505         dev_kfree_skb(skb);
506 }
507 
508 static int add_recvbuf_small(struct virtnet_info *vi, struct receive_queue *rq,
509                              gfp_t gfp)
510 {
511         struct sk_buff *skb;
512         struct virtio_net_hdr_mrg_rxbuf *hdr;
513         int err;
514 
515         skb = __netdev_alloc_skb_ip_align(vi->dev, GOOD_PACKET_LEN, gfp);
516         if (unlikely(!skb))
517                 return -ENOMEM;
518 
519         skb_put(skb, GOOD_PACKET_LEN);
520 
521         hdr = skb_vnet_hdr(skb);
522         sg_init_table(rq->sg, 2);
523         sg_set_buf(rq->sg, hdr, vi->hdr_len);
524         skb_to_sgvec(skb, rq->sg + 1, 0, skb->len);
525 
526         err = virtqueue_add_inbuf(rq->vq, rq->sg, 2, skb, gfp);
527         if (err < 0)
528                 dev_kfree_skb(skb);
529 
530         return err;
531 }
532 
533 static int add_recvbuf_big(struct virtnet_info *vi, struct receive_queue *rq,
534                            gfp_t gfp)
535 {
536         struct page *first, *list = NULL;
537         char *p;
538         int i, err, offset;
539 
540         sg_init_table(rq->sg, MAX_SKB_FRAGS + 2);
541 
542         /* page in rq->sg[MAX_SKB_FRAGS + 1] is list tail */
543         for (i = MAX_SKB_FRAGS + 1; i > 1; --i) {
544                 first = get_a_page(rq, gfp);
545                 if (!first) {
546                         if (list)
547                                 give_pages(rq, list);
548                         return -ENOMEM;
549                 }
550                 sg_set_buf(&rq->sg[i], page_address(first), PAGE_SIZE);
551 
552                 /* chain new page in list head to match sg */
553                 first->private = (unsigned long)list;
554                 list = first;
555         }
556 
557         first = get_a_page(rq, gfp);
558         if (!first) {
559                 give_pages(rq, list);
560                 return -ENOMEM;
561         }
562         p = page_address(first);
563 
564         /* rq->sg[0], rq->sg[1] share the same page */
565         /* a separated rq->sg[0] for header - required in case !any_header_sg */
566         sg_set_buf(&rq->sg[0], p, vi->hdr_len);
567 
568         /* rq->sg[1] for data packet, from offset */
569         offset = sizeof(struct padded_vnet_hdr);
570         sg_set_buf(&rq->sg[1], p + offset, PAGE_SIZE - offset);
571 
572         /* chain first in list head */
573         first->private = (unsigned long)list;
574         err = virtqueue_add_inbuf(rq->vq, rq->sg, MAX_SKB_FRAGS + 2,
575                                   first, gfp);
576         if (err < 0)
577                 give_pages(rq, first);
578 
579         return err;
580 }
581 
582 static unsigned int get_mergeable_buf_len(struct ewma_pkt_len *avg_pkt_len)
583 {
584         const size_t hdr_len = sizeof(struct virtio_net_hdr_mrg_rxbuf);
585         unsigned int len;
586 
587         len = hdr_len + clamp_t(unsigned int, ewma_pkt_len_read(avg_pkt_len),
588                         GOOD_PACKET_LEN, PAGE_SIZE - hdr_len);
589         return ALIGN(len, MERGEABLE_BUFFER_ALIGN);
590 }
591 
592 static int add_recvbuf_mergeable(struct receive_queue *rq, gfp_t gfp)
593 {
594         struct page_frag *alloc_frag = &rq->alloc_frag;
595         char *buf;
596         unsigned long ctx;
597         int err;
598         unsigned int len, hole;
599 
600         len = get_mergeable_buf_len(&rq->mrg_avg_pkt_len);
601         if (unlikely(!skb_page_frag_refill(len, alloc_frag, gfp)))
602                 return -ENOMEM;
603 
604         buf = (char *)page_address(alloc_frag->page) + alloc_frag->offset;
605         ctx = mergeable_buf_to_ctx(buf, len);
606         get_page(alloc_frag->page);
607         alloc_frag->offset += len;
608         hole = alloc_frag->size - alloc_frag->offset;
609         if (hole < len) {
610                 /* To avoid internal fragmentation, if there is very likely not
611                  * enough space for another buffer, add the remaining space to
612                  * the current buffer. This extra space is not included in
613                  * the truesize stored in ctx.
614                  */
615                 len += hole;
616                 alloc_frag->offset += hole;
617         }
618 
619         sg_init_one(rq->sg, buf, len);
620         err = virtqueue_add_inbuf(rq->vq, rq->sg, 1, (void *)ctx, gfp);
621         if (err < 0)
622                 put_page(virt_to_head_page(buf));
623 
624         return err;
625 }
626 
627 /*
628  * Returns false if we couldn't fill entirely (OOM).
629  *
630  * Normally run in the receive path, but can also be run from ndo_open
631  * before we're receiving packets, or from refill_work which is
632  * careful to disable receiving (using napi_disable).
633  */
634 static bool try_fill_recv(struct virtnet_info *vi, struct receive_queue *rq,
635                           gfp_t gfp)
636 {
637         int err;
638         bool oom;
639 
640         gfp |= __GFP_COLD;
641         do {
642                 if (vi->mergeable_rx_bufs)
643                         err = add_recvbuf_mergeable(rq, gfp);
644                 else if (vi->big_packets)
645                         err = add_recvbuf_big(vi, rq, gfp);
646                 else
647                         err = add_recvbuf_small(vi, rq, gfp);
648 
649                 oom = err == -ENOMEM;
650                 if (err)
651                         break;
652         } while (rq->vq->num_free);
653         virtqueue_kick(rq->vq);
654         return !oom;
655 }
656 
657 static void skb_recv_done(struct virtqueue *rvq)
658 {
659         struct virtnet_info *vi = rvq->vdev->priv;
660         struct receive_queue *rq = &vi->rq[vq2rxq(rvq)];
661 
662         /* Schedule NAPI, Suppress further interrupts if successful. */
663         if (napi_schedule_prep(&rq->napi)) {
664                 virtqueue_disable_cb(rvq);
665                 __napi_schedule(&rq->napi);
666         }
667 }
668 
669 static void virtnet_napi_enable(struct receive_queue *rq)
670 {
671         napi_enable(&rq->napi);
672 
673         /* If all buffers were filled by other side before we napi_enabled, we
674          * won't get another interrupt, so process any outstanding packets
675          * now.  virtnet_poll wants re-enable the queue, so we disable here.
676          * We synchronize against interrupts via NAPI_STATE_SCHED */
677         if (napi_schedule_prep(&rq->napi)) {
678                 virtqueue_disable_cb(rq->vq);
679                 local_bh_disable();
680                 __napi_schedule(&rq->napi);
681                 local_bh_enable();
682         }
683 }
684 
685 static void refill_work(struct work_struct *work)
686 {
687         struct virtnet_info *vi =
688                 container_of(work, struct virtnet_info, refill.work);
689         bool still_empty;
690         int i;
691 
692         for (i = 0; i < vi->curr_queue_pairs; i++) {
693                 struct receive_queue *rq = &vi->rq[i];
694 
695                 napi_disable(&rq->napi);
696                 still_empty = !try_fill_recv(vi, rq, GFP_KERNEL);
697                 virtnet_napi_enable(rq);
698 
699                 /* In theory, this can happen: if we don't get any buffers in
700                  * we will *never* try to fill again.
701                  */
702                 if (still_empty)
703                         schedule_delayed_work(&vi->refill, HZ/2);
704         }
705 }
706 
707 static int virtnet_receive(struct receive_queue *rq, int budget)
708 {
709         struct virtnet_info *vi = rq->vq->vdev->priv;
710         unsigned int len, received = 0;
711         void *buf;
712 
713         while (received < budget &&
714                (buf = virtqueue_get_buf(rq->vq, &len)) != NULL) {
715                 receive_buf(vi, rq, buf, len);
716                 received++;
717         }
718 
719         if (rq->vq->num_free > virtqueue_get_vring_size(rq->vq) / 2) {
720                 if (!try_fill_recv(vi, rq, GFP_ATOMIC))
721                         schedule_delayed_work(&vi->refill, 0);
722         }
723 
724         return received;
725 }
726 
727 static int virtnet_poll(struct napi_struct *napi, int budget)
728 {
729         struct receive_queue *rq =
730                 container_of(napi, struct receive_queue, napi);
731         unsigned int r, received;
732 
733         received = virtnet_receive(rq, budget);
734 
735         /* Out of packets? */
736         if (received < budget) {
737                 r = virtqueue_enable_cb_prepare(rq->vq);
738                 napi_complete_done(napi, received);
739                 if (unlikely(virtqueue_poll(rq->vq, r)) &&
740                     napi_schedule_prep(napi)) {
741                         virtqueue_disable_cb(rq->vq);
742                         __napi_schedule(napi);
743                 }
744         }
745 
746         return received;
747 }
748 
749 #ifdef CONFIG_NET_RX_BUSY_POLL
750 /* must be called with local_bh_disable()d */
751 static int virtnet_busy_poll(struct napi_struct *napi)
752 {
753         struct receive_queue *rq =
754                 container_of(napi, struct receive_queue, napi);
755         struct virtnet_info *vi = rq->vq->vdev->priv;
756         int r, received = 0, budget = 4;
757 
758         if (!(vi->status & VIRTIO_NET_S_LINK_UP))
759                 return LL_FLUSH_FAILED;
760 
761         if (!napi_schedule_prep(napi))
762                 return LL_FLUSH_BUSY;
763 
764         virtqueue_disable_cb(rq->vq);
765 
766 again:
767         received += virtnet_receive(rq, budget);
768 
769         r = virtqueue_enable_cb_prepare(rq->vq);
770         clear_bit(NAPI_STATE_SCHED, &napi->state);
771         if (unlikely(virtqueue_poll(rq->vq, r)) &&
772             napi_schedule_prep(napi)) {
773                 virtqueue_disable_cb(rq->vq);
774                 if (received < budget) {
775                         budget -= received;
776                         goto again;
777                 } else {
778                         __napi_schedule(napi);
779                 }
780         }
781 
782         return received;
783 }
784 #endif  /* CONFIG_NET_RX_BUSY_POLL */
785 
786 static int virtnet_open(struct net_device *dev)
787 {
788         struct virtnet_info *vi = netdev_priv(dev);
789         int i;
790 
791         for (i = 0; i < vi->max_queue_pairs; i++) {
792                 if (i < vi->curr_queue_pairs)
793                         /* Make sure we have some buffers: if oom use wq. */
794                         if (!try_fill_recv(vi, &vi->rq[i], GFP_KERNEL))
795                                 schedule_delayed_work(&vi->refill, 0);
796                 virtnet_napi_enable(&vi->rq[i]);
797         }
798 
799         return 0;
800 }
801 
802 static void free_old_xmit_skbs(struct send_queue *sq)
803 {
804         struct sk_buff *skb;
805         unsigned int len;
806         struct virtnet_info *vi = sq->vq->vdev->priv;
807         struct virtnet_stats *stats = this_cpu_ptr(vi->stats);
808 
809         while ((skb = virtqueue_get_buf(sq->vq, &len)) != NULL) {
810                 pr_debug("Sent skb %p\n", skb);
811 
812                 u64_stats_update_begin(&stats->tx_syncp);
813                 stats->tx_bytes += skb->len;
814                 stats->tx_packets++;
815                 u64_stats_update_end(&stats->tx_syncp);
816 
817                 dev_kfree_skb_any(skb);
818         }
819 }
820 
821 static int xmit_skb(struct send_queue *sq, struct sk_buff *skb)
822 {
823         struct virtio_net_hdr_mrg_rxbuf *hdr;
824         const unsigned char *dest = ((struct ethhdr *)skb->data)->h_dest;
825         struct virtnet_info *vi = sq->vq->vdev->priv;
826         unsigned num_sg;
827         unsigned hdr_len = vi->hdr_len;
828         bool can_push;
829 
830         pr_debug("%s: xmit %p %pM\n", vi->dev->name, skb, dest);
831 
832         can_push = vi->any_header_sg &&
833                 !((unsigned long)skb->data & (__alignof__(*hdr) - 1)) &&
834                 !skb_header_cloned(skb) && skb_headroom(skb) >= hdr_len;
835         /* Even if we can, don't push here yet as this would skew
836          * csum_start offset below. */
837         if (can_push)
838                 hdr = (struct virtio_net_hdr_mrg_rxbuf *)(skb->data - hdr_len);
839         else
840                 hdr = skb_vnet_hdr(skb);
841 
842         if (virtio_net_hdr_from_skb(skb, &hdr->hdr,
843                                     virtio_is_little_endian(vi->vdev)))
844                 BUG();
845 
846         if (vi->mergeable_rx_bufs)
847                 hdr->num_buffers = 0;
848 
849         sg_init_table(sq->sg, skb_shinfo(skb)->nr_frags + (can_push ? 1 : 2));
850         if (can_push) {
851                 __skb_push(skb, hdr_len);
852                 num_sg = skb_to_sgvec(skb, sq->sg, 0, skb->len);
853                 /* Pull header back to avoid skew in tx bytes calculations. */
854                 __skb_pull(skb, hdr_len);
855         } else {
856                 sg_set_buf(sq->sg, hdr, hdr_len);
857                 num_sg = skb_to_sgvec(skb, sq->sg + 1, 0, skb->len) + 1;
858         }
859         return virtqueue_add_outbuf(sq->vq, sq->sg, num_sg, skb, GFP_ATOMIC);
860 }
861 
862 static netdev_tx_t start_xmit(struct sk_buff *skb, struct net_device *dev)
863 {
864         struct virtnet_info *vi = netdev_priv(dev);
865         int qnum = skb_get_queue_mapping(skb);
866         struct send_queue *sq = &vi->sq[qnum];
867         int err;
868         struct netdev_queue *txq = netdev_get_tx_queue(dev, qnum);
869         bool kick = !skb->xmit_more;
870 
871         /* Free up any pending old buffers before queueing new ones. */
872         free_old_xmit_skbs(sq);
873 
874         /* timestamp packet in software */
875         skb_tx_timestamp(skb);
876 
877         /* Try to transmit */
878         err = xmit_skb(sq, skb);
879 
880         /* This should not happen! */
881         if (unlikely(err)) {
882                 dev->stats.tx_fifo_errors++;
883                 if (net_ratelimit())
884                         dev_warn(&dev->dev,
885                                  "Unexpected TXQ (%d) queue failure: %d\n", qnum, err);
886                 dev->stats.tx_dropped++;
887                 dev_kfree_skb_any(skb);
888                 return NETDEV_TX_OK;
889         }
890 
891         /* Don't wait up for transmitted skbs to be freed. */
892         skb_orphan(skb);
893         nf_reset(skb);
894 
895         /* If running out of space, stop queue to avoid getting packets that we
896          * are then unable to transmit.
897          * An alternative would be to force queuing layer to requeue the skb by
898          * returning NETDEV_TX_BUSY. However, NETDEV_TX_BUSY should not be
899          * returned in a normal path of operation: it means that driver is not
900          * maintaining the TX queue stop/start state properly, and causes
901          * the stack to do a non-trivial amount of useless work.
902          * Since most packets only take 1 or 2 ring slots, stopping the queue
903          * early means 16 slots are typically wasted.
904          */
905         if (sq->vq->num_free < 2+MAX_SKB_FRAGS) {
906                 netif_stop_subqueue(dev, qnum);
907                 if (unlikely(!virtqueue_enable_cb_delayed(sq->vq))) {
908                         /* More just got used, free them then recheck. */
909                         free_old_xmit_skbs(sq);
910                         if (sq->vq->num_free >= 2+MAX_SKB_FRAGS) {
911                                 netif_start_subqueue(dev, qnum);
912                                 virtqueue_disable_cb(sq->vq);
913                         }
914                 }
915         }
916 
917         if (kick || netif_xmit_stopped(txq))
918                 virtqueue_kick(sq->vq);
919 
920         return NETDEV_TX_OK;
921 }
922 
923 /*
924  * Send command via the control virtqueue and check status.  Commands
925  * supported by the hypervisor, as indicated by feature bits, should
926  * never fail unless improperly formatted.
927  */
928 static bool virtnet_send_command(struct virtnet_info *vi, u8 class, u8 cmd,
929                                  struct scatterlist *out)
930 {
931         struct scatterlist *sgs[4], hdr, stat;
932         unsigned out_num = 0, tmp;
933 
934         /* Caller should know better */
935         BUG_ON(!virtio_has_feature(vi->vdev, VIRTIO_NET_F_CTRL_VQ));
936 
937         vi->ctrl_status = ~0;
938         vi->ctrl_hdr.class = class;
939         vi->ctrl_hdr.cmd = cmd;
940         /* Add header */
941         sg_init_one(&hdr, &vi->ctrl_hdr, sizeof(vi->ctrl_hdr));
942         sgs[out_num++] = &hdr;
943 
944         if (out)
945                 sgs[out_num++] = out;
946 
947         /* Add return status. */
948         sg_init_one(&stat, &vi->ctrl_status, sizeof(vi->ctrl_status));
949         sgs[out_num] = &stat;
950 
951         BUG_ON(out_num + 1 > ARRAY_SIZE(sgs));
952         virtqueue_add_sgs(vi->cvq, sgs, out_num, 1, vi, GFP_ATOMIC);
953 
954         if (unlikely(!virtqueue_kick(vi->cvq)))
955                 return vi->ctrl_status == VIRTIO_NET_OK;
956 
957         /* Spin for a response, the kick causes an ioport write, trapping
958          * into the hypervisor, so the request should be handled immediately.
959          */
960         while (!virtqueue_get_buf(vi->cvq, &tmp) &&
961                !virtqueue_is_broken(vi->cvq))
962                 cpu_relax();
963 
964         return vi->ctrl_status == VIRTIO_NET_OK;
965 }
966 
967 static int virtnet_set_mac_address(struct net_device *dev, void *p)
968 {
969         struct virtnet_info *vi = netdev_priv(dev);
970         struct virtio_device *vdev = vi->vdev;
971         int ret;
972         struct sockaddr *addr;
973         struct scatterlist sg;
974 
975         addr = kmalloc(sizeof(*addr), GFP_KERNEL);
976         if (!addr)
977                 return -ENOMEM;
978         memcpy(addr, p, sizeof(*addr));
979 
980         ret = eth_prepare_mac_addr_change(dev, addr);
981         if (ret)
982                 goto out;
983 
984         if (virtio_has_feature(vdev, VIRTIO_NET_F_CTRL_MAC_ADDR)) {
985                 sg_init_one(&sg, addr->sa_data, dev->addr_len);
986                 if (!virtnet_send_command(vi, VIRTIO_NET_CTRL_MAC,
987                                           VIRTIO_NET_CTRL_MAC_ADDR_SET, &sg)) {
988                         dev_warn(&vdev->dev,
989                                  "Failed to set mac address by vq command.\n");
990                         ret = -EINVAL;
991                         goto out;
992                 }
993         } else if (virtio_has_feature(vdev, VIRTIO_NET_F_MAC) &&
994                    !virtio_has_feature(vdev, VIRTIO_F_VERSION_1)) {
995                 unsigned int i;
996 
997                 /* Naturally, this has an atomicity problem. */
998                 for (i = 0; i < dev->addr_len; i++)
999                         virtio_cwrite8(vdev,
1000                                        offsetof(struct virtio_net_config, mac) +
1001                                        i, addr->sa_data[i]);
1002         }
1003 
1004         eth_commit_mac_addr_change(dev, p);
1005         ret = 0;
1006 
1007 out:
1008         kfree(addr);
1009         return ret;
1010 }
1011 
1012 static struct rtnl_link_stats64 *virtnet_stats(struct net_device *dev,
1013                                                struct rtnl_link_stats64 *tot)
1014 {
1015         struct virtnet_info *vi = netdev_priv(dev);
1016         int cpu;
1017         unsigned int start;
1018 
1019         for_each_possible_cpu(cpu) {
1020                 struct virtnet_stats *stats = per_cpu_ptr(vi->stats, cpu);
1021                 u64 tpackets, tbytes, rpackets, rbytes;
1022 
1023                 do {
1024                         start = u64_stats_fetch_begin_irq(&stats->tx_syncp);
1025                         tpackets = stats->tx_packets;
1026                         tbytes   = stats->tx_bytes;
1027                 } while (u64_stats_fetch_retry_irq(&stats->tx_syncp, start));
1028 
1029                 do {
1030                         start = u64_stats_fetch_begin_irq(&stats->rx_syncp);
1031                         rpackets = stats->rx_packets;
1032                         rbytes   = stats->rx_bytes;
1033                 } while (u64_stats_fetch_retry_irq(&stats->rx_syncp, start));
1034 
1035                 tot->rx_packets += rpackets;
1036                 tot->tx_packets += tpackets;
1037                 tot->rx_bytes   += rbytes;
1038                 tot->tx_bytes   += tbytes;
1039         }
1040 
1041         tot->tx_dropped = dev->stats.tx_dropped;
1042         tot->tx_fifo_errors = dev->stats.tx_fifo_errors;
1043         tot->rx_dropped = dev->stats.rx_dropped;
1044         tot->rx_length_errors = dev->stats.rx_length_errors;
1045         tot->rx_frame_errors = dev->stats.rx_frame_errors;
1046 
1047         return tot;
1048 }
1049 
1050 #ifdef CONFIG_NET_POLL_CONTROLLER
1051 static void virtnet_netpoll(struct net_device *dev)
1052 {
1053         struct virtnet_info *vi = netdev_priv(dev);
1054         int i;
1055 
1056         for (i = 0; i < vi->curr_queue_pairs; i++)
1057                 napi_schedule(&vi->rq[i].napi);
1058 }
1059 #endif
1060 
1061 static void virtnet_ack_link_announce(struct virtnet_info *vi)
1062 {
1063         rtnl_lock();
1064         if (!virtnet_send_command(vi, VIRTIO_NET_CTRL_ANNOUNCE,
1065                                   VIRTIO_NET_CTRL_ANNOUNCE_ACK, NULL))
1066                 dev_warn(&vi->dev->dev, "Failed to ack link announce.\n");
1067         rtnl_unlock();
1068 }
1069 
1070 static int virtnet_set_queues(struct virtnet_info *vi, u16 queue_pairs)
1071 {
1072         struct scatterlist sg;
1073         struct net_device *dev = vi->dev;
1074 
1075         if (!vi->has_cvq || !virtio_has_feature(vi->vdev, VIRTIO_NET_F_MQ))
1076                 return 0;
1077 
1078         vi->ctrl_mq.virtqueue_pairs = cpu_to_virtio16(vi->vdev, queue_pairs);
1079         sg_init_one(&sg, &vi->ctrl_mq, sizeof(vi->ctrl_mq));
1080 
1081         if (!virtnet_send_command(vi, VIRTIO_NET_CTRL_MQ,
1082                                   VIRTIO_NET_CTRL_MQ_VQ_PAIRS_SET, &sg)) {
1083                 dev_warn(&dev->dev, "Fail to set num of queue pairs to %d\n",
1084                          queue_pairs);
1085                 return -EINVAL;
1086         } else {
1087                 vi->curr_queue_pairs = queue_pairs;
1088                 /* virtnet_open() will refill when device is going to up. */
1089                 if (dev->flags & IFF_UP)
1090                         schedule_delayed_work(&vi->refill, 0);
1091         }
1092 
1093         return 0;
1094 }
1095 
1096 static int virtnet_close(struct net_device *dev)
1097 {
1098         struct virtnet_info *vi = netdev_priv(dev);
1099         int i;
1100 
1101         /* Make sure refill_work doesn't re-enable napi! */
1102         cancel_delayed_work_sync(&vi->refill);
1103 
1104         for (i = 0; i < vi->max_queue_pairs; i++)
1105                 napi_disable(&vi->rq[i].napi);
1106 
1107         return 0;
1108 }
1109 
1110 static void virtnet_set_rx_mode(struct net_device *dev)
1111 {
1112         struct virtnet_info *vi = netdev_priv(dev);
1113         struct scatterlist sg[2];
1114         struct virtio_net_ctrl_mac *mac_data;
1115         struct netdev_hw_addr *ha;
1116         int uc_count;
1117         int mc_count;
1118         void *buf;
1119         int i;
1120 
1121         /* We can't dynamically set ndo_set_rx_mode, so return gracefully */
1122         if (!virtio_has_feature(vi->vdev, VIRTIO_NET_F_CTRL_RX))
1123                 return;
1124 
1125         vi->ctrl_promisc = ((dev->flags & IFF_PROMISC) != 0);
1126         vi->ctrl_allmulti = ((dev->flags & IFF_ALLMULTI) != 0);
1127 
1128         sg_init_one(sg, &vi->ctrl_promisc, sizeof(vi->ctrl_promisc));
1129 
1130         if (!virtnet_send_command(vi, VIRTIO_NET_CTRL_RX,
1131                                   VIRTIO_NET_CTRL_RX_PROMISC, sg))
1132                 dev_warn(&dev->dev, "Failed to %sable promisc mode.\n",
1133                          vi->ctrl_promisc ? "en" : "dis");
1134 
1135         sg_init_one(sg, &vi->ctrl_allmulti, sizeof(vi->ctrl_allmulti));
1136 
1137         if (!virtnet_send_command(vi, VIRTIO_NET_CTRL_RX,
1138                                   VIRTIO_NET_CTRL_RX_ALLMULTI, sg))
1139                 dev_warn(&dev->dev, "Failed to %sable allmulti mode.\n",
1140                          vi->ctrl_allmulti ? "en" : "dis");
1141 
1142         uc_count = netdev_uc_count(dev);
1143         mc_count = netdev_mc_count(dev);
1144         /* MAC filter - use one buffer for both lists */
1145         buf = kzalloc(((uc_count + mc_count) * ETH_ALEN) +
1146                       (2 * sizeof(mac_data->entries)), GFP_ATOMIC);
1147         mac_data = buf;
1148         if (!buf)
1149                 return;
1150 
1151         sg_init_table(sg, 2);
1152 
1153         /* Store the unicast list and count in the front of the buffer */
1154         mac_data->entries = cpu_to_virtio32(vi->vdev, uc_count);
1155         i = 0;
1156         netdev_for_each_uc_addr(ha, dev)
1157                 memcpy(&mac_data->macs[i++][0], ha->addr, ETH_ALEN);
1158 
1159         sg_set_buf(&sg[0], mac_data,
1160                    sizeof(mac_data->entries) + (uc_count * ETH_ALEN));
1161 
1162         /* multicast list and count fill the end */
1163         mac_data = (void *)&mac_data->macs[uc_count][0];
1164 
1165         mac_data->entries = cpu_to_virtio32(vi->vdev, mc_count);
1166         i = 0;
1167         netdev_for_each_mc_addr(ha, dev)
1168                 memcpy(&mac_data->macs[i++][0], ha->addr, ETH_ALEN);
1169 
1170         sg_set_buf(&sg[1], mac_data,
1171                    sizeof(mac_data->entries) + (mc_count * ETH_ALEN));
1172 
1173         if (!virtnet_send_command(vi, VIRTIO_NET_CTRL_MAC,
1174                                   VIRTIO_NET_CTRL_MAC_TABLE_SET, sg))
1175                 dev_warn(&dev->dev, "Failed to set MAC filter table.\n");
1176 
1177         kfree(buf);
1178 }
1179 
1180 static int virtnet_vlan_rx_add_vid(struct net_device *dev,
1181                                    __be16 proto, u16 vid)
1182 {
1183         struct virtnet_info *vi = netdev_priv(dev);
1184         struct scatterlist sg;
1185 
1186         vi->ctrl_vid = vid;
1187         sg_init_one(&sg, &vi->ctrl_vid, sizeof(vi->ctrl_vid));
1188 
1189         if (!virtnet_send_command(vi, VIRTIO_NET_CTRL_VLAN,
1190                                   VIRTIO_NET_CTRL_VLAN_ADD, &sg))
1191                 dev_warn(&dev->dev, "Failed to add VLAN ID %d.\n", vid);
1192         return 0;
1193 }
1194 
1195 static int virtnet_vlan_rx_kill_vid(struct net_device *dev,
1196                                     __be16 proto, u16 vid)
1197 {
1198         struct virtnet_info *vi = netdev_priv(dev);
1199         struct scatterlist sg;
1200 
1201         vi->ctrl_vid = vid;
1202         sg_init_one(&sg, &vi->ctrl_vid, sizeof(vi->ctrl_vid));
1203 
1204         if (!virtnet_send_command(vi, VIRTIO_NET_CTRL_VLAN,
1205                                   VIRTIO_NET_CTRL_VLAN_DEL, &sg))
1206                 dev_warn(&dev->dev, "Failed to kill VLAN ID %d.\n", vid);
1207         return 0;
1208 }
1209 
1210 static void virtnet_clean_affinity(struct virtnet_info *vi, long hcpu)
1211 {
1212         int i;
1213 
1214         if (vi->affinity_hint_set) {
1215                 for (i = 0; i < vi->max_queue_pairs; i++) {
1216                         virtqueue_set_affinity(vi->rq[i].vq, -1);
1217                         virtqueue_set_affinity(vi->sq[i].vq, -1);
1218                 }
1219 
1220                 vi->affinity_hint_set = false;
1221         }
1222 }
1223 
1224 static void virtnet_set_affinity(struct virtnet_info *vi)
1225 {
1226         int i;
1227         int cpu;
1228 
1229         /* In multiqueue mode, when the number of cpu is equal to the number of
1230          * queue pairs, we let the queue pairs to be private to one cpu by
1231          * setting the affinity hint to eliminate the contention.
1232          */
1233         if (vi->curr_queue_pairs == 1 ||
1234             vi->max_queue_pairs != num_online_cpus()) {
1235                 virtnet_clean_affinity(vi, -1);
1236                 return;
1237         }
1238 
1239         i = 0;
1240         for_each_online_cpu(cpu) {
1241                 virtqueue_set_affinity(vi->rq[i].vq, cpu);
1242                 virtqueue_set_affinity(vi->sq[i].vq, cpu);
1243                 netif_set_xps_queue(vi->dev, cpumask_of(cpu), i);
1244                 i++;
1245         }
1246 
1247         vi->affinity_hint_set = true;
1248 }
1249 
1250 static int virtnet_cpu_online(unsigned int cpu, struct hlist_node *node)
1251 {
1252         struct virtnet_info *vi = hlist_entry_safe(node, struct virtnet_info,
1253                                                    node);
1254         virtnet_set_affinity(vi);
1255         return 0;
1256 }
1257 
1258 static int virtnet_cpu_dead(unsigned int cpu, struct hlist_node *node)
1259 {
1260         struct virtnet_info *vi = hlist_entry_safe(node, struct virtnet_info,
1261                                                    node_dead);
1262         virtnet_set_affinity(vi);
1263         return 0;
1264 }
1265 
1266 static int virtnet_cpu_down_prep(unsigned int cpu, struct hlist_node *node)
1267 {
1268         struct virtnet_info *vi = hlist_entry_safe(node, struct virtnet_info,
1269                                                    node);
1270 
1271         virtnet_clean_affinity(vi, cpu);
1272         return 0;
1273 }
1274 
1275 static enum cpuhp_state virtionet_online;
1276 
1277 static int virtnet_cpu_notif_add(struct virtnet_info *vi)
1278 {
1279         int ret;
1280 
1281         ret = cpuhp_state_add_instance_nocalls(virtionet_online, &vi->node);
1282         if (ret)
1283                 return ret;
1284         ret = cpuhp_state_add_instance_nocalls(CPUHP_VIRT_NET_DEAD,
1285                                                &vi->node_dead);
1286         if (!ret)
1287                 return ret;
1288         cpuhp_state_remove_instance_nocalls(virtionet_online, &vi->node);
1289         return ret;
1290 }
1291 
1292 static void virtnet_cpu_notif_remove(struct virtnet_info *vi)
1293 {
1294         cpuhp_state_remove_instance_nocalls(virtionet_online, &vi->node);
1295         cpuhp_state_remove_instance_nocalls(CPUHP_VIRT_NET_DEAD,
1296                                             &vi->node_dead);
1297 }
1298 
1299 static void virtnet_get_ringparam(struct net_device *dev,
1300                                 struct ethtool_ringparam *ring)
1301 {
1302         struct virtnet_info *vi = netdev_priv(dev);
1303 
1304         ring->rx_max_pending = virtqueue_get_vring_size(vi->rq[0].vq);
1305         ring->tx_max_pending = virtqueue_get_vring_size(vi->sq[0].vq);
1306         ring->rx_pending = ring->rx_max_pending;
1307         ring->tx_pending = ring->tx_max_pending;
1308 }
1309 
1310 
1311 static void virtnet_get_drvinfo(struct net_device *dev,
1312                                 struct ethtool_drvinfo *info)
1313 {
1314         struct virtnet_info *vi = netdev_priv(dev);
1315         struct virtio_device *vdev = vi->vdev;
1316 
1317         strlcpy(info->driver, KBUILD_MODNAME, sizeof(info->driver));
1318         strlcpy(info->version, VIRTNET_DRIVER_VERSION, sizeof(info->version));
1319         strlcpy(info->bus_info, virtio_bus_name(vdev), sizeof(info->bus_info));
1320 
1321 }
1322 
1323 /* TODO: Eliminate OOO packets during switching */
1324 static int virtnet_set_channels(struct net_device *dev,
1325                                 struct ethtool_channels *channels)
1326 {
1327         struct virtnet_info *vi = netdev_priv(dev);
1328         u16 queue_pairs = channels->combined_count;
1329         int err;
1330 
1331         /* We don't support separate rx/tx channels.
1332          * We don't allow setting 'other' channels.
1333          */
1334         if (channels->rx_count || channels->tx_count || channels->other_count)
1335                 return -EINVAL;
1336 
1337         if (queue_pairs > vi->max_queue_pairs || queue_pairs == 0)
1338                 return -EINVAL;
1339 
1340         get_online_cpus();
1341         err = virtnet_set_queues(vi, queue_pairs);
1342         if (!err) {
1343                 netif_set_real_num_tx_queues(dev, queue_pairs);
1344                 netif_set_real_num_rx_queues(dev, queue_pairs);
1345 
1346                 virtnet_set_affinity(vi);
1347         }
1348         put_online_cpus();
1349 
1350         return err;
1351 }
1352 
1353 static void virtnet_get_channels(struct net_device *dev,
1354                                  struct ethtool_channels *channels)
1355 {
1356         struct virtnet_info *vi = netdev_priv(dev);
1357 
1358         channels->combined_count = vi->curr_queue_pairs;
1359         channels->max_combined = vi->max_queue_pairs;
1360         channels->max_other = 0;
1361         channels->rx_count = 0;
1362         channels->tx_count = 0;
1363         channels->other_count = 0;
1364 }
1365 
1366 /* Check if the user is trying to change anything besides speed/duplex */
1367 static bool virtnet_validate_ethtool_cmd(const struct ethtool_cmd *cmd)
1368 {
1369         struct ethtool_cmd diff1 = *cmd;
1370         struct ethtool_cmd diff2 = {};
1371 
1372         /* cmd is always set so we need to clear it, validate the port type
1373          * and also without autonegotiation we can ignore advertising
1374          */
1375         ethtool_cmd_speed_set(&diff1, 0);
1376         diff2.port = PORT_OTHER;
1377         diff1.advertising = 0;
1378         diff1.duplex = 0;
1379         diff1.cmd = 0;
1380 
1381         return !memcmp(&diff1, &diff2, sizeof(diff1));
1382 }
1383 
1384 static int virtnet_set_settings(struct net_device *dev, struct ethtool_cmd *cmd)
1385 {
1386         struct virtnet_info *vi = netdev_priv(dev);
1387         u32 speed;
1388 
1389         speed = ethtool_cmd_speed(cmd);
1390         /* don't allow custom speed and duplex */
1391         if (!ethtool_validate_speed(speed) ||
1392             !ethtool_validate_duplex(cmd->duplex) ||
1393             !virtnet_validate_ethtool_cmd(cmd))
1394                 return -EINVAL;
1395         vi->speed = speed;
1396         vi->duplex = cmd->duplex;
1397 
1398         return 0;
1399 }
1400 
1401 static int virtnet_get_settings(struct net_device *dev, struct ethtool_cmd *cmd)
1402 {
1403         struct virtnet_info *vi = netdev_priv(dev);
1404 
1405         ethtool_cmd_speed_set(cmd, vi->speed);
1406         cmd->duplex = vi->duplex;
1407         cmd->port = PORT_OTHER;
1408 
1409         return 0;
1410 }
1411 
1412 static void virtnet_init_settings(struct net_device *dev)
1413 {
1414         struct virtnet_info *vi = netdev_priv(dev);
1415 
1416         vi->speed = SPEED_UNKNOWN;
1417         vi->duplex = DUPLEX_UNKNOWN;
1418 }
1419 
1420 static const struct ethtool_ops virtnet_ethtool_ops = {
1421         .get_drvinfo = virtnet_get_drvinfo,
1422         .get_link = ethtool_op_get_link,
1423         .get_ringparam = virtnet_get_ringparam,
1424         .set_channels = virtnet_set_channels,
1425         .get_channels = virtnet_get_channels,
1426         .get_ts_info = ethtool_op_get_ts_info,
1427         .get_settings = virtnet_get_settings,
1428         .set_settings = virtnet_set_settings,
1429 };
1430 
1431 #define MIN_MTU 68
1432 #define MAX_MTU 65535
1433 
1434 static int virtnet_change_mtu(struct net_device *dev, int new_mtu)
1435 {
1436         if (new_mtu < MIN_MTU || new_mtu > MAX_MTU)
1437                 return -EINVAL;
1438         dev->mtu = new_mtu;
1439         return 0;
1440 }
1441 
1442 static const struct net_device_ops virtnet_netdev = {
1443         .ndo_open            = virtnet_open,
1444         .ndo_stop            = virtnet_close,
1445         .ndo_start_xmit      = start_xmit,
1446         .ndo_validate_addr   = eth_validate_addr,
1447         .ndo_set_mac_address = virtnet_set_mac_address,
1448         .ndo_set_rx_mode     = virtnet_set_rx_mode,
1449         .ndo_change_mtu      = virtnet_change_mtu,
1450         .ndo_get_stats64     = virtnet_stats,
1451         .ndo_vlan_rx_add_vid = virtnet_vlan_rx_add_vid,
1452         .ndo_vlan_rx_kill_vid = virtnet_vlan_rx_kill_vid,
1453 #ifdef CONFIG_NET_POLL_CONTROLLER
1454         .ndo_poll_controller = virtnet_netpoll,
1455 #endif
1456 #ifdef CONFIG_NET_RX_BUSY_POLL
1457         .ndo_busy_poll          = virtnet_busy_poll,
1458 #endif
1459 };
1460 
1461 static void virtnet_config_changed_work(struct work_struct *work)
1462 {
1463         struct virtnet_info *vi =
1464                 container_of(work, struct virtnet_info, config_work);
1465         u16 v;
1466 
1467         if (virtio_cread_feature(vi->vdev, VIRTIO_NET_F_STATUS,
1468                                  struct virtio_net_config, status, &v) < 0)
1469                 return;
1470 
1471         if (v & VIRTIO_NET_S_ANNOUNCE) {
1472                 netdev_notify_peers(vi->dev);
1473                 virtnet_ack_link_announce(vi);
1474         }
1475 
1476         /* Ignore unknown (future) status bits */
1477         v &= VIRTIO_NET_S_LINK_UP;
1478 
1479         if (vi->status == v)
1480                 return;
1481 
1482         vi->status = v;
1483 
1484         if (vi->status & VIRTIO_NET_S_LINK_UP) {
1485                 netif_carrier_on(vi->dev);
1486                 netif_tx_wake_all_queues(vi->dev);
1487         } else {
1488                 netif_carrier_off(vi->dev);
1489                 netif_tx_stop_all_queues(vi->dev);
1490         }
1491 }
1492 
1493 static void virtnet_config_changed(struct virtio_device *vdev)
1494 {
1495         struct virtnet_info *vi = vdev->priv;
1496 
1497         schedule_work(&vi->config_work);
1498 }
1499 
1500 static void virtnet_free_queues(struct virtnet_info *vi)
1501 {
1502         int i;
1503 
1504         for (i = 0; i < vi->max_queue_pairs; i++) {
1505                 napi_hash_del(&vi->rq[i].napi);
1506                 netif_napi_del(&vi->rq[i].napi);
1507         }
1508 
1509         /* We called napi_hash_del() before netif_napi_del(),
1510          * we need to respect an RCU grace period before freeing vi->rq
1511          */
1512         synchronize_net();
1513 
1514         kfree(vi->rq);
1515         kfree(vi->sq);
1516 }
1517 
1518 static void free_receive_bufs(struct virtnet_info *vi)
1519 {
1520         int i;
1521 
1522         for (i = 0; i < vi->max_queue_pairs; i++) {
1523                 while (vi->rq[i].pages)
1524                         __free_pages(get_a_page(&vi->rq[i], GFP_KERNEL), 0);
1525         }
1526 }
1527 
1528 static void free_receive_page_frags(struct virtnet_info *vi)
1529 {
1530         int i;
1531         for (i = 0; i < vi->max_queue_pairs; i++)
1532                 if (vi->rq[i].alloc_frag.page)
1533                         put_page(vi->rq[i].alloc_frag.page);
1534 }
1535 
1536 static void free_unused_bufs(struct virtnet_info *vi)
1537 {
1538         void *buf;
1539         int i;
1540 
1541         for (i = 0; i < vi->max_queue_pairs; i++) {
1542                 struct virtqueue *vq = vi->sq[i].vq;
1543                 while ((buf = virtqueue_detach_unused_buf(vq)) != NULL)
1544                         dev_kfree_skb(buf);
1545         }
1546 
1547         for (i = 0; i < vi->max_queue_pairs; i++) {
1548                 struct virtqueue *vq = vi->rq[i].vq;
1549 
1550                 while ((buf = virtqueue_detach_unused_buf(vq)) != NULL) {
1551                         if (vi->mergeable_rx_bufs) {
1552                                 unsigned long ctx = (unsigned long)buf;
1553                                 void *base = mergeable_ctx_to_buf_address(ctx);
1554                                 put_page(virt_to_head_page(base));
1555                         } else if (vi->big_packets) {
1556                                 give_pages(&vi->rq[i], buf);
1557                         } else {
1558                                 dev_kfree_skb(buf);
1559                         }
1560                 }
1561         }
1562 }
1563 
1564 static void virtnet_del_vqs(struct virtnet_info *vi)
1565 {
1566         struct virtio_device *vdev = vi->vdev;
1567 
1568         virtnet_clean_affinity(vi, -1);
1569 
1570         vdev->config->del_vqs(vdev);
1571 
1572         virtnet_free_queues(vi);
1573 }
1574 
1575 static int virtnet_find_vqs(struct virtnet_info *vi)
1576 {
1577         vq_callback_t **callbacks;
1578         struct virtqueue **vqs;
1579         int ret = -ENOMEM;
1580         int i, total_vqs;
1581         const char **names;
1582 
1583         /* We expect 1 RX virtqueue followed by 1 TX virtqueue, followed by
1584          * possible N-1 RX/TX queue pairs used in multiqueue mode, followed by
1585          * possible control vq.
1586          */
1587         total_vqs = vi->max_queue_pairs * 2 +
1588                     virtio_has_feature(vi->vdev, VIRTIO_NET_F_CTRL_VQ);
1589 
1590         /* Allocate space for find_vqs parameters */
1591         vqs = kzalloc(total_vqs * sizeof(*vqs), GFP_KERNEL);
1592         if (!vqs)
1593                 goto err_vq;
1594         callbacks = kmalloc(total_vqs * sizeof(*callbacks), GFP_KERNEL);
1595         if (!callbacks)
1596                 goto err_callback;
1597         names = kmalloc(total_vqs * sizeof(*names), GFP_KERNEL);
1598         if (!names)
1599                 goto err_names;
1600 
1601         /* Parameters for control virtqueue, if any */
1602         if (vi->has_cvq) {
1603                 callbacks[total_vqs - 1] = NULL;
1604                 names[total_vqs - 1] = "control";
1605         }
1606 
1607         /* Allocate/initialize parameters for send/receive virtqueues */
1608         for (i = 0; i < vi->max_queue_pairs; i++) {
1609                 callbacks[rxq2vq(i)] = skb_recv_done;
1610                 callbacks[txq2vq(i)] = skb_xmit_done;
1611                 sprintf(vi->rq[i].name, "input.%d", i);
1612                 sprintf(vi->sq[i].name, "output.%d", i);
1613                 names[rxq2vq(i)] = vi->rq[i].name;
1614                 names[txq2vq(i)] = vi->sq[i].name;
1615         }
1616 
1617         ret = vi->vdev->config->find_vqs(vi->vdev, total_vqs, vqs, callbacks,
1618                                          names);
1619         if (ret)
1620                 goto err_find;
1621 
1622         if (vi->has_cvq) {
1623                 vi->cvq = vqs[total_vqs - 1];
1624                 if (virtio_has_feature(vi->vdev, VIRTIO_NET_F_CTRL_VLAN))
1625                         vi->dev->features |= NETIF_F_HW_VLAN_CTAG_FILTER;
1626         }
1627 
1628         for (i = 0; i < vi->max_queue_pairs; i++) {
1629                 vi->rq[i].vq = vqs[rxq2vq(i)];
1630                 vi->sq[i].vq = vqs[txq2vq(i)];
1631         }
1632 
1633         kfree(names);
1634         kfree(callbacks);
1635         kfree(vqs);
1636 
1637         return 0;
1638 
1639 err_find:
1640         kfree(names);
1641 err_names:
1642         kfree(callbacks);
1643 err_callback:
1644         kfree(vqs);
1645 err_vq:
1646         return ret;
1647 }
1648 
1649 static int virtnet_alloc_queues(struct virtnet_info *vi)
1650 {
1651         int i;
1652 
1653         vi->sq = kzalloc(sizeof(*vi->sq) * vi->max_queue_pairs, GFP_KERNEL);
1654         if (!vi->sq)
1655                 goto err_sq;
1656         vi->rq = kzalloc(sizeof(*vi->rq) * vi->max_queue_pairs, GFP_KERNEL);
1657         if (!vi->rq)
1658                 goto err_rq;
1659 
1660         INIT_DELAYED_WORK(&vi->refill, refill_work);
1661         for (i = 0; i < vi->max_queue_pairs; i++) {
1662                 vi->rq[i].pages = NULL;
1663                 netif_napi_add(vi->dev, &vi->rq[i].napi, virtnet_poll,
1664                                napi_weight);
1665 
1666                 sg_init_table(vi->rq[i].sg, ARRAY_SIZE(vi->rq[i].sg));
1667                 ewma_pkt_len_init(&vi->rq[i].mrg_avg_pkt_len);
1668                 sg_init_table(vi->sq[i].sg, ARRAY_SIZE(vi->sq[i].sg));
1669         }
1670 
1671         return 0;
1672 
1673 err_rq:
1674         kfree(vi->sq);
1675 err_sq:
1676         return -ENOMEM;
1677 }
1678 
1679 static int init_vqs(struct virtnet_info *vi)
1680 {
1681         int ret;
1682 
1683         /* Allocate send & receive queues */
1684         ret = virtnet_alloc_queues(vi);
1685         if (ret)
1686                 goto err;
1687 
1688         ret = virtnet_find_vqs(vi);
1689         if (ret)
1690                 goto err_free;
1691 
1692         get_online_cpus();
1693         virtnet_set_affinity(vi);
1694         put_online_cpus();
1695 
1696         return 0;
1697 
1698 err_free:
1699         virtnet_free_queues(vi);
1700 err:
1701         return ret;
1702 }
1703 
1704 #ifdef CONFIG_SYSFS
1705 static ssize_t mergeable_rx_buffer_size_show(struct netdev_rx_queue *queue,
1706                 struct rx_queue_attribute *attribute, char *buf)
1707 {
1708         struct virtnet_info *vi = netdev_priv(queue->dev);
1709         unsigned int queue_index = get_netdev_rx_queue_index(queue);
1710         struct ewma_pkt_len *avg;
1711 
1712         BUG_ON(queue_index >= vi->max_queue_pairs);
1713         avg = &vi->rq[queue_index].mrg_avg_pkt_len;
1714         return sprintf(buf, "%u\n", get_mergeable_buf_len(avg));
1715 }
1716 
1717 static struct rx_queue_attribute mergeable_rx_buffer_size_attribute =
1718         __ATTR_RO(mergeable_rx_buffer_size);
1719 
1720 static struct attribute *virtio_net_mrg_rx_attrs[] = {
1721         &mergeable_rx_buffer_size_attribute.attr,
1722         NULL
1723 };
1724 
1725 static const struct attribute_group virtio_net_mrg_rx_group = {
1726         .name = "virtio_net",
1727         .attrs = virtio_net_mrg_rx_attrs
1728 };
1729 #endif
1730 
1731 static bool virtnet_fail_on_feature(struct virtio_device *vdev,
1732                                     unsigned int fbit,
1733                                     const char *fname, const char *dname)
1734 {
1735         if (!virtio_has_feature(vdev, fbit))
1736                 return false;
1737 
1738         dev_err(&vdev->dev, "device advertises feature %s but not %s",
1739                 fname, dname);
1740 
1741         return true;
1742 }
1743 
1744 #define VIRTNET_FAIL_ON(vdev, fbit, dbit)                       \
1745         virtnet_fail_on_feature(vdev, fbit, #fbit, dbit)
1746 
1747 static bool virtnet_validate_features(struct virtio_device *vdev)
1748 {
1749         if (!virtio_has_feature(vdev, VIRTIO_NET_F_CTRL_VQ) &&
1750             (VIRTNET_FAIL_ON(vdev, VIRTIO_NET_F_CTRL_RX,
1751                              "VIRTIO_NET_F_CTRL_VQ") ||
1752              VIRTNET_FAIL_ON(vdev, VIRTIO_NET_F_CTRL_VLAN,
1753                              "VIRTIO_NET_F_CTRL_VQ") ||
1754              VIRTNET_FAIL_ON(vdev, VIRTIO_NET_F_GUEST_ANNOUNCE,
1755                              "VIRTIO_NET_F_CTRL_VQ") ||
1756              VIRTNET_FAIL_ON(vdev, VIRTIO_NET_F_MQ, "VIRTIO_NET_F_CTRL_VQ") ||
1757              VIRTNET_FAIL_ON(vdev, VIRTIO_NET_F_CTRL_MAC_ADDR,
1758                              "VIRTIO_NET_F_CTRL_VQ"))) {
1759                 return false;
1760         }
1761 
1762         return true;
1763 }
1764 
1765 static int virtnet_probe(struct virtio_device *vdev)
1766 {
1767         int i, err;
1768         struct net_device *dev;
1769         struct virtnet_info *vi;
1770         u16 max_queue_pairs;
1771         int mtu;
1772 
1773         if (!vdev->config->get) {
1774                 dev_err(&vdev->dev, "%s failure: config access disabled\n",
1775                         __func__);
1776                 return -EINVAL;
1777         }
1778 
1779         if (!virtnet_validate_features(vdev))
1780                 return -EINVAL;
1781 
1782         /* Find if host supports multiqueue virtio_net device */
1783         err = virtio_cread_feature(vdev, VIRTIO_NET_F_MQ,
1784                                    struct virtio_net_config,
1785                                    max_virtqueue_pairs, &max_queue_pairs);
1786 
1787         /* We need at least 2 queue's */
1788         if (err || max_queue_pairs < VIRTIO_NET_CTRL_MQ_VQ_PAIRS_MIN ||
1789             max_queue_pairs > VIRTIO_NET_CTRL_MQ_VQ_PAIRS_MAX ||
1790             !virtio_has_feature(vdev, VIRTIO_NET_F_CTRL_VQ))
1791                 max_queue_pairs = 1;
1792 
1793         /* Allocate ourselves a network device with room for our info */
1794         dev = alloc_etherdev_mq(sizeof(struct virtnet_info), max_queue_pairs);
1795         if (!dev)
1796                 return -ENOMEM;
1797 
1798         /* Set up network device as normal. */
1799         dev->priv_flags |= IFF_UNICAST_FLT | IFF_LIVE_ADDR_CHANGE;
1800         dev->netdev_ops = &virtnet_netdev;
1801         dev->features = NETIF_F_HIGHDMA;
1802 
1803         dev->ethtool_ops = &virtnet_ethtool_ops;
1804         SET_NETDEV_DEV(dev, &vdev->dev);
1805 
1806         /* Do we support "hardware" checksums? */
1807         if (virtio_has_feature(vdev, VIRTIO_NET_F_CSUM)) {
1808                 /* This opens up the world of extra features. */
1809                 dev->hw_features |= NETIF_F_HW_CSUM | NETIF_F_SG;
1810                 if (csum)
1811                         dev->features |= NETIF_F_HW_CSUM | NETIF_F_SG;
1812 
1813                 if (virtio_has_feature(vdev, VIRTIO_NET_F_GSO)) {
1814                         dev->hw_features |= NETIF_F_TSO | NETIF_F_UFO
1815                                 | NETIF_F_TSO_ECN | NETIF_F_TSO6;
1816                 }
1817                 /* Individual feature bits: what can host handle? */
1818                 if (virtio_has_feature(vdev, VIRTIO_NET_F_HOST_TSO4))
1819                         dev->hw_features |= NETIF_F_TSO;
1820                 if (virtio_has_feature(vdev, VIRTIO_NET_F_HOST_TSO6))
1821                         dev->hw_features |= NETIF_F_TSO6;
1822                 if (virtio_has_feature(vdev, VIRTIO_NET_F_HOST_ECN))
1823                         dev->hw_features |= NETIF_F_TSO_ECN;
1824                 if (virtio_has_feature(vdev, VIRTIO_NET_F_HOST_UFO))
1825                         dev->hw_features |= NETIF_F_UFO;
1826 
1827                 dev->features |= NETIF_F_GSO_ROBUST;
1828 
1829                 if (gso)
1830                         dev->features |= dev->hw_features & (NETIF_F_ALL_TSO|NETIF_F_UFO);
1831                 /* (!csum && gso) case will be fixed by register_netdev() */
1832         }
1833         if (virtio_has_feature(vdev, VIRTIO_NET_F_GUEST_CSUM))
1834                 dev->features |= NETIF_F_RXCSUM;
1835 
1836         dev->vlan_features = dev->features;
1837 
1838         /* Configuration may specify what MAC to use.  Otherwise random. */
1839         if (virtio_has_feature(vdev, VIRTIO_NET_F_MAC))
1840                 virtio_cread_bytes(vdev,
1841                                    offsetof(struct virtio_net_config, mac),
1842                                    dev->dev_addr, dev->addr_len);
1843         else
1844                 eth_hw_addr_random(dev);
1845 
1846         /* Set up our device-specific information */
1847         vi = netdev_priv(dev);
1848         vi->dev = dev;
1849         vi->vdev = vdev;
1850         vdev->priv = vi;
1851         vi->stats = alloc_percpu(struct virtnet_stats);
1852         err = -ENOMEM;
1853         if (vi->stats == NULL)
1854                 goto free;
1855 
1856         for_each_possible_cpu(i) {
1857                 struct virtnet_stats *virtnet_stats;
1858                 virtnet_stats = per_cpu_ptr(vi->stats, i);
1859                 u64_stats_init(&virtnet_stats->tx_syncp);
1860                 u64_stats_init(&virtnet_stats->rx_syncp);
1861         }
1862 
1863         INIT_WORK(&vi->config_work, virtnet_config_changed_work);
1864 
1865         /* If we can receive ANY GSO packets, we must allocate large ones. */
1866         if (virtio_has_feature(vdev, VIRTIO_NET_F_GUEST_TSO4) ||
1867             virtio_has_feature(vdev, VIRTIO_NET_F_GUEST_TSO6) ||
1868             virtio_has_feature(vdev, VIRTIO_NET_F_GUEST_ECN) ||
1869             virtio_has_feature(vdev, VIRTIO_NET_F_GUEST_UFO))
1870                 vi->big_packets = true;
1871 
1872         if (virtio_has_feature(vdev, VIRTIO_NET_F_MRG_RXBUF))
1873                 vi->mergeable_rx_bufs = true;
1874 
1875         if (virtio_has_feature(vdev, VIRTIO_NET_F_MRG_RXBUF) ||
1876             virtio_has_feature(vdev, VIRTIO_F_VERSION_1))
1877                 vi->hdr_len = sizeof(struct virtio_net_hdr_mrg_rxbuf);
1878         else
1879                 vi->hdr_len = sizeof(struct virtio_net_hdr);
1880 
1881         if (virtio_has_feature(vdev, VIRTIO_F_ANY_LAYOUT) ||
1882             virtio_has_feature(vdev, VIRTIO_F_VERSION_1))
1883                 vi->any_header_sg = true;
1884 
1885         if (virtio_has_feature(vdev, VIRTIO_NET_F_CTRL_VQ))
1886                 vi->has_cvq = true;
1887 
1888         if (virtio_has_feature(vdev, VIRTIO_NET_F_MTU)) {
1889                 mtu = virtio_cread16(vdev,
1890                                      offsetof(struct virtio_net_config,
1891                                               mtu));
1892                 if (virtnet_change_mtu(dev, mtu))
1893                         __virtio_clear_bit(vdev, VIRTIO_NET_F_MTU);
1894         }
1895 
1896         if (vi->any_header_sg)
1897                 dev->needed_headroom = vi->hdr_len;
1898 
1899         /* Use single tx/rx queue pair as default */
1900         vi->curr_queue_pairs = 1;
1901         vi->max_queue_pairs = max_queue_pairs;
1902 
1903         /* Allocate/initialize the rx/tx queues, and invoke find_vqs */
1904         err = init_vqs(vi);
1905         if (err)
1906                 goto free_stats;
1907 
1908 #ifdef CONFIG_SYSFS
1909         if (vi->mergeable_rx_bufs)
1910                 dev->sysfs_rx_queue_group = &virtio_net_mrg_rx_group;
1911 #endif
1912         netif_set_real_num_tx_queues(dev, vi->curr_queue_pairs);
1913         netif_set_real_num_rx_queues(dev, vi->curr_queue_pairs);
1914 
1915         virtnet_init_settings(dev);
1916 
1917         err = register_netdev(dev);
1918         if (err) {
1919                 pr_debug("virtio_net: registering device failed\n");
1920                 goto free_vqs;
1921         }
1922 
1923         virtio_device_ready(vdev);
1924 
1925         err = virtnet_cpu_notif_add(vi);
1926         if (err) {
1927                 pr_debug("virtio_net: registering cpu notifier failed\n");
1928                 goto free_unregister_netdev;
1929         }
1930 
1931         /* Assume link up if device can't report link status,
1932            otherwise get link status from config. */
1933         if (virtio_has_feature(vi->vdev, VIRTIO_NET_F_STATUS)) {
1934                 netif_carrier_off(dev);
1935                 schedule_work(&vi->config_work);
1936         } else {
1937                 vi->status = VIRTIO_NET_S_LINK_UP;
1938                 netif_carrier_on(dev);
1939         }
1940 
1941         pr_debug("virtnet: registered device %s with %d RX and TX vq's\n",
1942                  dev->name, max_queue_pairs);
1943 
1944         return 0;
1945 
1946 free_unregister_netdev:
1947         vi->vdev->config->reset(vdev);
1948 
1949         unregister_netdev(dev);
1950 free_vqs:
1951         cancel_delayed_work_sync(&vi->refill);
1952         free_receive_page_frags(vi);
1953         virtnet_del_vqs(vi);
1954 free_stats:
1955         free_percpu(vi->stats);
1956 free:
1957         free_netdev(dev);
1958         return err;
1959 }
1960 
1961 static void remove_vq_common(struct virtnet_info *vi)
1962 {
1963         vi->vdev->config->reset(vi->vdev);
1964 
1965         /* Free unused buffers in both send and recv, if any. */
1966         free_unused_bufs(vi);
1967 
1968         free_receive_bufs(vi);
1969 
1970         free_receive_page_frags(vi);
1971 
1972         virtnet_del_vqs(vi);
1973 }
1974 
1975 static void virtnet_remove(struct virtio_device *vdev)
1976 {
1977         struct virtnet_info *vi = vdev->priv;
1978 
1979         virtnet_cpu_notif_remove(vi);
1980 
1981         /* Make sure no work handler is accessing the device. */
1982         flush_work(&vi->config_work);
1983 
1984         unregister_netdev(vi->dev);
1985 
1986         remove_vq_common(vi);
1987 
1988         free_percpu(vi->stats);
1989         free_netdev(vi->dev);
1990 }
1991 
1992 #ifdef CONFIG_PM_SLEEP
1993 static int virtnet_freeze(struct virtio_device *vdev)
1994 {
1995         struct virtnet_info *vi = vdev->priv;
1996         int i;
1997 
1998         virtnet_cpu_notif_remove(vi);
1999 
2000         /* Make sure no work handler is accessing the device */
2001         flush_work(&vi->config_work);
2002 
2003         netif_device_detach(vi->dev);
2004         cancel_delayed_work_sync(&vi->refill);
2005 
2006         if (netif_running(vi->dev)) {
2007                 for (i = 0; i < vi->max_queue_pairs; i++)
2008                         napi_disable(&vi->rq[i].napi);
2009         }
2010 
2011         remove_vq_common(vi);
2012 
2013         return 0;
2014 }
2015 
2016 static int virtnet_restore(struct virtio_device *vdev)
2017 {
2018         struct virtnet_info *vi = vdev->priv;
2019         int err, i;
2020 
2021         err = init_vqs(vi);
2022         if (err)
2023                 return err;
2024 
2025         virtio_device_ready(vdev);
2026 
2027         if (netif_running(vi->dev)) {
2028                 for (i = 0; i < vi->curr_queue_pairs; i++)
2029                         if (!try_fill_recv(vi, &vi->rq[i], GFP_KERNEL))
2030                                 schedule_delayed_work(&vi->refill, 0);
2031 
2032                 for (i = 0; i < vi->max_queue_pairs; i++)
2033                         virtnet_napi_enable(&vi->rq[i]);
2034         }
2035 
2036         netif_device_attach(vi->dev);
2037 
2038         rtnl_lock();
2039         virtnet_set_queues(vi, vi->curr_queue_pairs);
2040         rtnl_unlock();
2041 
2042         err = virtnet_cpu_notif_add(vi);
2043         if (err)
2044                 return err;
2045 
2046         return 0;
2047 }
2048 #endif
2049 
2050 static struct virtio_device_id id_table[] = {
2051         { VIRTIO_ID_NET, VIRTIO_DEV_ANY_ID },
2052         { 0 },
2053 };
2054 
2055 #define VIRTNET_FEATURES \
2056         VIRTIO_NET_F_CSUM, VIRTIO_NET_F_GUEST_CSUM, \
2057         VIRTIO_NET_F_MAC, \
2058         VIRTIO_NET_F_HOST_TSO4, VIRTIO_NET_F_HOST_UFO, VIRTIO_NET_F_HOST_TSO6, \
2059         VIRTIO_NET_F_HOST_ECN, VIRTIO_NET_F_GUEST_TSO4, VIRTIO_NET_F_GUEST_TSO6, \
2060         VIRTIO_NET_F_GUEST_ECN, VIRTIO_NET_F_GUEST_UFO, \
2061         VIRTIO_NET_F_MRG_RXBUF, VIRTIO_NET_F_STATUS, VIRTIO_NET_F_CTRL_VQ, \
2062         VIRTIO_NET_F_CTRL_RX, VIRTIO_NET_F_CTRL_VLAN, \
2063         VIRTIO_NET_F_GUEST_ANNOUNCE, VIRTIO_NET_F_MQ, \
2064         VIRTIO_NET_F_CTRL_MAC_ADDR, \
2065         VIRTIO_NET_F_MTU
2066 
2067 static unsigned int features[] = {
2068         VIRTNET_FEATURES,
2069 };
2070 
2071 static unsigned int features_legacy[] = {
2072         VIRTNET_FEATURES,
2073         VIRTIO_NET_F_GSO,
2074         VIRTIO_F_ANY_LAYOUT,
2075 };
2076 
2077 static struct virtio_driver virtio_net_driver = {
2078         .feature_table = features,
2079         .feature_table_size = ARRAY_SIZE(features),
2080         .feature_table_legacy = features_legacy,
2081         .feature_table_size_legacy = ARRAY_SIZE(features_legacy),
2082         .driver.name =  KBUILD_MODNAME,
2083         .driver.owner = THIS_MODULE,
2084         .id_table =     id_table,
2085         .probe =        virtnet_probe,
2086         .remove =       virtnet_remove,
2087         .config_changed = virtnet_config_changed,
2088 #ifdef CONFIG_PM_SLEEP
2089         .freeze =       virtnet_freeze,
2090         .restore =      virtnet_restore,
2091 #endif
2092 };
2093 
2094 static __init int virtio_net_driver_init(void)
2095 {
2096         int ret;
2097 
2098         ret = cpuhp_setup_state_multi(CPUHP_AP_ONLINE_DYN, "AP_VIRT_NET_ONLINE",
2099                                       virtnet_cpu_online,
2100                                       virtnet_cpu_down_prep);
2101         if (ret < 0)
2102                 goto out;
2103         virtionet_online = ret;
2104         ret = cpuhp_setup_state_multi(CPUHP_VIRT_NET_DEAD, "VIRT_NET_DEAD",
2105                                       NULL, virtnet_cpu_dead);
2106         if (ret)
2107                 goto err_dead;
2108 
2109         ret = register_virtio_driver(&virtio_net_driver);
2110         if (ret)
2111                 goto err_virtio;
2112         return 0;
2113 err_virtio:
2114         cpuhp_remove_multi_state(CPUHP_VIRT_NET_DEAD);
2115 err_dead:
2116         cpuhp_remove_multi_state(virtionet_online);
2117 out:
2118         return ret;
2119 }
2120 module_init(virtio_net_driver_init);
2121 
2122 static __exit void virtio_net_driver_exit(void)
2123 {
2124         cpuhp_remove_multi_state(CPUHP_VIRT_NET_DEAD);
2125         cpuhp_remove_multi_state(virtionet_online);
2126         unregister_virtio_driver(&virtio_net_driver);
2127 }
2128 module_exit(virtio_net_driver_exit);
2129 
2130 MODULE_DEVICE_TABLE(virtio, id_table);
2131 MODULE_DESCRIPTION("Virtio network driver");
2132 MODULE_LICENSE("GPL");
2133 

This page was automatically generated by LXR 0.3.1 (source).  •  Linux is a registered trademark of Linus Torvalds  •  Contact us