]>
Commit | Line | Data |
---|---|---|
48925e37 | 1 | /* A network driver using virtio. |
296f96fc RR |
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 | |
adf8d3ff | 16 | * along with this program; if not, see <http://www.gnu.org/licenses/>. |
296f96fc RR |
17 | */ |
18 | //#define DEBUG | |
19 | #include <linux/netdevice.h> | |
20 | #include <linux/etherdevice.h> | |
a9ea3fc6 | 21 | #include <linux/ethtool.h> |
296f96fc RR |
22 | #include <linux/module.h> |
23 | #include <linux/virtio.h> | |
24 | #include <linux/virtio_net.h> | |
f600b690 | 25 | #include <linux/bpf.h> |
a67edbf4 | 26 | #include <linux/bpf_trace.h> |
296f96fc | 27 | #include <linux/scatterlist.h> |
e918085a | 28 | #include <linux/if_vlan.h> |
5a0e3ad6 | 29 | #include <linux/slab.h> |
8de4b2f3 | 30 | #include <linux/cpu.h> |
ab7db917 | 31 | #include <linux/average.h> |
d85b758f | 32 | #include <net/route.h> |
296f96fc | 33 | |
d34710e3 | 34 | static int napi_weight = NAPI_POLL_WEIGHT; |
6c0cd7c0 DL |
35 | module_param(napi_weight, int, 0444); |
36 | ||
b92f1e67 | 37 | static bool csum = true, gso = true, napi_tx; |
34a48579 RR |
38 | module_param(csum, bool, 0444); |
39 | module_param(gso, bool, 0444); | |
b92f1e67 | 40 | module_param(napi_tx, bool, 0644); |
34a48579 | 41 | |
296f96fc | 42 | /* FIXME: MTU in config. */ |
5061de36 | 43 | #define GOOD_PACKET_LEN (ETH_HLEN + VLAN_HLEN + ETH_DATA_LEN) |
3f2c31d9 | 44 | #define GOOD_COPY_LEN 128 |
296f96fc | 45 | |
f6b10209 JW |
46 | #define VIRTNET_RX_PAD (NET_IP_ALIGN + NET_SKB_PAD) |
47 | ||
2de2f7f4 JF |
48 | /* Amount of XDP headroom to prepend to packets for use by xdp_adjust_head */ |
49 | #define VIRTIO_XDP_HEADROOM 256 | |
50 | ||
5377d758 JB |
51 | /* RX packet size EWMA. The average packet size is used to determine the packet |
52 | * buffer size when refilling RX rings. As the entire RX ring may be refilled | |
53 | * at once, the weight is chosen so that the EWMA will be insensitive to short- | |
54 | * term, transient changes in packet size. | |
ab7db917 | 55 | */ |
eb1e011a | 56 | DECLARE_EWMA(pkt_len, 0, 64) |
ab7db917 | 57 | |
66846048 | 58 | #define VIRTNET_DRIVER_VERSION "1.0.0" |
2a41f71d | 59 | |
7acd4329 CIK |
60 | static const unsigned long guest_offloads[] = { |
61 | VIRTIO_NET_F_GUEST_TSO4, | |
62 | VIRTIO_NET_F_GUEST_TSO6, | |
63 | VIRTIO_NET_F_GUEST_ECN, | |
64 | VIRTIO_NET_F_GUEST_UFO | |
65 | }; | |
3f93522f | 66 | |
3fa2a1df | 67 | struct virtnet_stats { |
83a27052 ED |
68 | struct u64_stats_sync tx_syncp; |
69 | struct u64_stats_sync rx_syncp; | |
3fa2a1df | 70 | u64 tx_bytes; |
71 | u64 tx_packets; | |
72 | ||
73 | u64 rx_bytes; | |
74 | u64 rx_packets; | |
75 | }; | |
76 | ||
e9d7417b JW |
77 | /* Internal representation of a send virtqueue */ |
78 | struct send_queue { | |
79 | /* Virtqueue associated with this send _queue */ | |
80 | struct virtqueue *vq; | |
81 | ||
82 | /* TX: fragments + linear part + virtio header */ | |
83 | struct scatterlist sg[MAX_SKB_FRAGS + 2]; | |
986a4f4d JW |
84 | |
85 | /* Name of the send queue: output.$index */ | |
86 | char name[40]; | |
b92f1e67 WB |
87 | |
88 | struct napi_struct napi; | |
e9d7417b JW |
89 | }; |
90 | ||
91 | /* Internal representation of a receive virtqueue */ | |
92 | struct receive_queue { | |
93 | /* Virtqueue associated with this receive_queue */ | |
94 | struct virtqueue *vq; | |
95 | ||
296f96fc RR |
96 | struct napi_struct napi; |
97 | ||
f600b690 JF |
98 | struct bpf_prog __rcu *xdp_prog; |
99 | ||
e9d7417b JW |
100 | /* Chain pages by the private ptr. */ |
101 | struct page *pages; | |
102 | ||
ab7db917 | 103 | /* Average packet length for mergeable receive buffers. */ |
5377d758 | 104 | struct ewma_pkt_len mrg_avg_pkt_len; |
ab7db917 | 105 | |
fb51879d MD |
106 | /* Page frag for packet buffer allocation. */ |
107 | struct page_frag alloc_frag; | |
108 | ||
e9d7417b JW |
109 | /* RX: fragments + linear part + virtio header */ |
110 | struct scatterlist sg[MAX_SKB_FRAGS + 2]; | |
986a4f4d | 111 | |
d85b758f MT |
112 | /* Min single buffer size for mergeable buffers case. */ |
113 | unsigned int min_buf_len; | |
114 | ||
986a4f4d JW |
115 | /* Name of this receive queue: input.$index */ |
116 | char name[40]; | |
e9d7417b JW |
117 | }; |
118 | ||
119 | struct virtnet_info { | |
120 | struct virtio_device *vdev; | |
121 | struct virtqueue *cvq; | |
122 | struct net_device *dev; | |
986a4f4d JW |
123 | struct send_queue *sq; |
124 | struct receive_queue *rq; | |
e9d7417b JW |
125 | unsigned int status; |
126 | ||
986a4f4d JW |
127 | /* Max # of queue pairs supported by the device */ |
128 | u16 max_queue_pairs; | |
129 | ||
130 | /* # of queue pairs currently used by the driver */ | |
131 | u16 curr_queue_pairs; | |
132 | ||
672aafd5 JF |
133 | /* # of XDP queue pairs currently used by the driver */ |
134 | u16 xdp_queue_pairs; | |
135 | ||
97402b96 HX |
136 | /* I like... big packets and I cannot lie! */ |
137 | bool big_packets; | |
138 | ||
3f2c31d9 MM |
139 | /* Host will merge rx buffers for big packets (shake it! shake it!) */ |
140 | bool mergeable_rx_bufs; | |
141 | ||
986a4f4d JW |
142 | /* Has control virtqueue */ |
143 | bool has_cvq; | |
144 | ||
e7428e95 MT |
145 | /* Host can handle any s/g split between our header and packet data */ |
146 | bool any_header_sg; | |
147 | ||
012873d0 MT |
148 | /* Packet virtio header size */ |
149 | u8 hdr_len; | |
150 | ||
3fa2a1df | 151 | /* Active statistics */ |
152 | struct virtnet_stats __percpu *stats; | |
153 | ||
3161e453 RR |
154 | /* Work struct for refilling if we run low on memory. */ |
155 | struct delayed_work refill; | |
156 | ||
586d17c5 JW |
157 | /* Work struct for config space updates */ |
158 | struct work_struct config_work; | |
159 | ||
986a4f4d JW |
160 | /* Does the affinity hint is set for virtqueues? */ |
161 | bool affinity_hint_set; | |
47be2479 | 162 | |
8017c279 SAS |
163 | /* CPU hotplug instances for online & dead */ |
164 | struct hlist_node node; | |
165 | struct hlist_node node_dead; | |
2ac46030 MT |
166 | |
167 | /* Control VQ buffers: protected by the rtnl lock */ | |
168 | struct virtio_net_ctrl_hdr ctrl_hdr; | |
169 | virtio_net_ctrl_ack ctrl_status; | |
a725ee3e | 170 | struct virtio_net_ctrl_mq ctrl_mq; |
2ac46030 MT |
171 | u8 ctrl_promisc; |
172 | u8 ctrl_allmulti; | |
a725ee3e | 173 | u16 ctrl_vid; |
3f93522f | 174 | u64 ctrl_offloads; |
16032be5 NA |
175 | |
176 | /* Ethtool settings */ | |
177 | u8 duplex; | |
178 | u32 speed; | |
3f93522f JW |
179 | |
180 | unsigned long guest_offloads; | |
296f96fc RR |
181 | }; |
182 | ||
9ab86bbc | 183 | struct padded_vnet_hdr { |
012873d0 | 184 | struct virtio_net_hdr_mrg_rxbuf hdr; |
9ab86bbc | 185 | /* |
012873d0 MT |
186 | * hdr is in a separate sg buffer, and data sg buffer shares same page |
187 | * with this header sg. This padding makes next sg 16 byte aligned | |
188 | * after the header. | |
9ab86bbc | 189 | */ |
012873d0 | 190 | char padding[4]; |
9ab86bbc SM |
191 | }; |
192 | ||
986a4f4d JW |
193 | /* Converting between virtqueue no. and kernel tx/rx queue no. |
194 | * 0:rx0 1:tx0 2:rx1 3:tx1 ... 2N:rxN 2N+1:txN 2N+2:cvq | |
195 | */ | |
196 | static int vq2txq(struct virtqueue *vq) | |
197 | { | |
9d0ca6ed | 198 | return (vq->index - 1) / 2; |
986a4f4d JW |
199 | } |
200 | ||
201 | static int txq2vq(int txq) | |
202 | { | |
203 | return txq * 2 + 1; | |
204 | } | |
205 | ||
206 | static int vq2rxq(struct virtqueue *vq) | |
207 | { | |
9d0ca6ed | 208 | return vq->index / 2; |
986a4f4d JW |
209 | } |
210 | ||
211 | static int rxq2vq(int rxq) | |
212 | { | |
213 | return rxq * 2; | |
214 | } | |
215 | ||
012873d0 | 216 | static inline struct virtio_net_hdr_mrg_rxbuf *skb_vnet_hdr(struct sk_buff *skb) |
296f96fc | 217 | { |
012873d0 | 218 | return (struct virtio_net_hdr_mrg_rxbuf *)skb->cb; |
296f96fc RR |
219 | } |
220 | ||
9ab86bbc SM |
221 | /* |
222 | * private is used to chain pages for big packets, put the whole | |
223 | * most recent used list in the beginning for reuse | |
224 | */ | |
e9d7417b | 225 | static void give_pages(struct receive_queue *rq, struct page *page) |
0a888fd1 | 226 | { |
9ab86bbc | 227 | struct page *end; |
0a888fd1 | 228 | |
e9d7417b | 229 | /* Find end of list, sew whole thing into vi->rq.pages. */ |
9ab86bbc | 230 | for (end = page; end->private; end = (struct page *)end->private); |
e9d7417b JW |
231 | end->private = (unsigned long)rq->pages; |
232 | rq->pages = page; | |
0a888fd1 MM |
233 | } |
234 | ||
e9d7417b | 235 | static struct page *get_a_page(struct receive_queue *rq, gfp_t gfp_mask) |
fb6813f4 | 236 | { |
e9d7417b | 237 | struct page *p = rq->pages; |
fb6813f4 | 238 | |
9ab86bbc | 239 | if (p) { |
e9d7417b | 240 | rq->pages = (struct page *)p->private; |
9ab86bbc SM |
241 | /* clear private here, it is used to chain pages */ |
242 | p->private = 0; | |
243 | } else | |
fb6813f4 RR |
244 | p = alloc_page(gfp_mask); |
245 | return p; | |
246 | } | |
247 | ||
e4e8452a WB |
248 | static void virtqueue_napi_schedule(struct napi_struct *napi, |
249 | struct virtqueue *vq) | |
250 | { | |
251 | if (napi_schedule_prep(napi)) { | |
252 | virtqueue_disable_cb(vq); | |
253 | __napi_schedule(napi); | |
254 | } | |
255 | } | |
256 | ||
257 | static void virtqueue_napi_complete(struct napi_struct *napi, | |
258 | struct virtqueue *vq, int processed) | |
259 | { | |
260 | int opaque; | |
261 | ||
262 | opaque = virtqueue_enable_cb_prepare(vq); | |
263 | if (napi_complete_done(napi, processed) && | |
264 | unlikely(virtqueue_poll(vq, opaque))) | |
265 | virtqueue_napi_schedule(napi, vq); | |
266 | } | |
267 | ||
e9d7417b | 268 | static void skb_xmit_done(struct virtqueue *vq) |
296f96fc | 269 | { |
e9d7417b | 270 | struct virtnet_info *vi = vq->vdev->priv; |
b92f1e67 | 271 | struct napi_struct *napi = &vi->sq[vq2txq(vq)].napi; |
296f96fc | 272 | |
2cb9c6ba | 273 | /* Suppress further interrupts. */ |
e9d7417b | 274 | virtqueue_disable_cb(vq); |
11a3a154 | 275 | |
b92f1e67 WB |
276 | if (napi->weight) |
277 | virtqueue_napi_schedule(napi, vq); | |
278 | else | |
279 | /* We were probably waiting for more output buffers. */ | |
280 | netif_wake_subqueue(vi->dev, vq2txq(vq)); | |
296f96fc RR |
281 | } |
282 | ||
28b39bc7 JW |
283 | #define MRG_CTX_HEADER_SHIFT 22 |
284 | static void *mergeable_len_to_ctx(unsigned int truesize, | |
285 | unsigned int headroom) | |
286 | { | |
287 | return (void *)(unsigned long)((headroom << MRG_CTX_HEADER_SHIFT) | truesize); | |
288 | } | |
289 | ||
290 | static unsigned int mergeable_ctx_to_headroom(void *mrg_ctx) | |
291 | { | |
292 | return (unsigned long)mrg_ctx >> MRG_CTX_HEADER_SHIFT; | |
293 | } | |
294 | ||
295 | static unsigned int mergeable_ctx_to_truesize(void *mrg_ctx) | |
296 | { | |
297 | return (unsigned long)mrg_ctx & ((1 << MRG_CTX_HEADER_SHIFT) - 1); | |
298 | } | |
299 | ||
3464645a | 300 | /* Called from bottom half context */ |
946fa564 MT |
301 | static struct sk_buff *page_to_skb(struct virtnet_info *vi, |
302 | struct receive_queue *rq, | |
2613af0e MD |
303 | struct page *page, unsigned int offset, |
304 | unsigned int len, unsigned int truesize) | |
9ab86bbc SM |
305 | { |
306 | struct sk_buff *skb; | |
012873d0 | 307 | struct virtio_net_hdr_mrg_rxbuf *hdr; |
2613af0e | 308 | unsigned int copy, hdr_len, hdr_padded_len; |
9ab86bbc | 309 | char *p; |
fb6813f4 | 310 | |
2613af0e | 311 | p = page_address(page) + offset; |
3f2c31d9 | 312 | |
9ab86bbc | 313 | /* copy small packet so we can reuse these pages for small data */ |
c67f5db8 | 314 | skb = napi_alloc_skb(&rq->napi, GOOD_COPY_LEN); |
9ab86bbc SM |
315 | if (unlikely(!skb)) |
316 | return NULL; | |
3f2c31d9 | 317 | |
9ab86bbc | 318 | hdr = skb_vnet_hdr(skb); |
3f2c31d9 | 319 | |
012873d0 MT |
320 | hdr_len = vi->hdr_len; |
321 | if (vi->mergeable_rx_bufs) | |
a4a76503 | 322 | hdr_padded_len = sizeof(*hdr); |
012873d0 | 323 | else |
2613af0e | 324 | hdr_padded_len = sizeof(struct padded_vnet_hdr); |
3f2c31d9 | 325 | |
9ab86bbc | 326 | memcpy(hdr, p, hdr_len); |
3f2c31d9 | 327 | |
9ab86bbc | 328 | len -= hdr_len; |
2613af0e MD |
329 | offset += hdr_padded_len; |
330 | p += hdr_padded_len; | |
3f2c31d9 | 331 | |
9ab86bbc SM |
332 | copy = len; |
333 | if (copy > skb_tailroom(skb)) | |
334 | copy = skb_tailroom(skb); | |
59ae1d12 | 335 | skb_put_data(skb, p, copy); |
3f2c31d9 | 336 | |
9ab86bbc SM |
337 | len -= copy; |
338 | offset += copy; | |
3f2c31d9 | 339 | |
2613af0e MD |
340 | if (vi->mergeable_rx_bufs) { |
341 | if (len) | |
342 | skb_add_rx_frag(skb, 0, page, offset, len, truesize); | |
343 | else | |
344 | put_page(page); | |
345 | return skb; | |
346 | } | |
347 | ||
e878d78b SL |
348 | /* |
349 | * Verify that we can indeed put this data into a skb. | |
350 | * This is here to handle cases when the device erroneously | |
351 | * tries to receive more than is possible. This is usually | |
352 | * the case of a broken device. | |
353 | */ | |
354 | if (unlikely(len > MAX_SKB_FRAGS * PAGE_SIZE)) { | |
be443899 | 355 | net_dbg_ratelimited("%s: too much data\n", skb->dev->name); |
e878d78b SL |
356 | dev_kfree_skb(skb); |
357 | return NULL; | |
358 | } | |
2613af0e | 359 | BUG_ON(offset >= PAGE_SIZE); |
9ab86bbc | 360 | while (len) { |
2613af0e MD |
361 | unsigned int frag_size = min((unsigned)PAGE_SIZE - offset, len); |
362 | skb_add_rx_frag(skb, skb_shinfo(skb)->nr_frags, page, offset, | |
363 | frag_size, truesize); | |
364 | len -= frag_size; | |
9ab86bbc SM |
365 | page = (struct page *)page->private; |
366 | offset = 0; | |
367 | } | |
3f2c31d9 | 368 | |
9ab86bbc | 369 | if (page) |
e9d7417b | 370 | give_pages(rq, page); |
3f2c31d9 | 371 | |
9ab86bbc SM |
372 | return skb; |
373 | } | |
3f2c31d9 | 374 | |
a67edbf4 | 375 | static bool virtnet_xdp_xmit(struct virtnet_info *vi, |
f6b10209 | 376 | struct xdp_buff *xdp) |
56434a01 | 377 | { |
56434a01 | 378 | struct virtio_net_hdr_mrg_rxbuf *hdr; |
f6b10209 | 379 | unsigned int len; |
722d8283 JF |
380 | struct send_queue *sq; |
381 | unsigned int qp; | |
56434a01 JF |
382 | void *xdp_sent; |
383 | int err; | |
384 | ||
722d8283 JF |
385 | qp = vi->curr_queue_pairs - vi->xdp_queue_pairs + smp_processor_id(); |
386 | sq = &vi->sq[qp]; | |
387 | ||
56434a01 JF |
388 | /* Free up any pending old buffers before queueing new ones. */ |
389 | while ((xdp_sent = virtqueue_get_buf(sq->vq, &len)) != NULL) { | |
f6b10209 | 390 | struct page *sent_page = virt_to_head_page(xdp_sent); |
bb91accf | 391 | |
f6b10209 | 392 | put_page(sent_page); |
bb91accf | 393 | } |
56434a01 | 394 | |
f6b10209 JW |
395 | xdp->data -= vi->hdr_len; |
396 | /* Zero header and leave csum up to XDP layers */ | |
397 | hdr = xdp->data; | |
398 | memset(hdr, 0, vi->hdr_len); | |
bb91accf | 399 | |
f6b10209 | 400 | sg_init_one(sq->sg, xdp->data, xdp->data_end - xdp->data); |
bb91accf | 401 | |
f6b10209 | 402 | err = virtqueue_add_outbuf(sq->vq, sq->sg, 1, xdp->data, GFP_ATOMIC); |
56434a01 | 403 | if (unlikely(err)) { |
f6b10209 | 404 | struct page *page = virt_to_head_page(xdp->data); |
bb91accf | 405 | |
f6b10209 | 406 | put_page(page); |
a67edbf4 | 407 | return false; |
56434a01 JF |
408 | } |
409 | ||
410 | virtqueue_kick(sq->vq); | |
a67edbf4 | 411 | return true; |
56434a01 JF |
412 | } |
413 | ||
f6b10209 JW |
414 | static unsigned int virtnet_get_headroom(struct virtnet_info *vi) |
415 | { | |
416 | return vi->xdp_queue_pairs ? VIRTIO_XDP_HEADROOM : 0; | |
417 | } | |
418 | ||
4941d472 JW |
419 | /* We copy the packet for XDP in the following cases: |
420 | * | |
421 | * 1) Packet is scattered across multiple rx buffers. | |
422 | * 2) Headroom space is insufficient. | |
423 | * | |
424 | * This is inefficient but it's a temporary condition that | |
425 | * we hit right after XDP is enabled and until queue is refilled | |
426 | * with large buffers with sufficient headroom - so it should affect | |
427 | * at most queue size packets. | |
428 | * Afterwards, the conditions to enable | |
429 | * XDP should preclude the underlying device from sending packets | |
430 | * across multiple buffers (num_buf > 1), and we make sure buffers | |
431 | * have enough headroom. | |
432 | */ | |
433 | static struct page *xdp_linearize_page(struct receive_queue *rq, | |
434 | u16 *num_buf, | |
435 | struct page *p, | |
436 | int offset, | |
437 | int page_off, | |
438 | unsigned int *len) | |
439 | { | |
440 | struct page *page = alloc_page(GFP_ATOMIC); | |
441 | ||
442 | if (!page) | |
443 | return NULL; | |
444 | ||
445 | memcpy(page_address(page) + page_off, page_address(p) + offset, *len); | |
446 | page_off += *len; | |
447 | ||
448 | while (--*num_buf) { | |
449 | unsigned int buflen; | |
450 | void *buf; | |
451 | int off; | |
452 | ||
453 | buf = virtqueue_get_buf(rq->vq, &buflen); | |
454 | if (unlikely(!buf)) | |
455 | goto err_buf; | |
456 | ||
457 | p = virt_to_head_page(buf); | |
458 | off = buf - page_address(p); | |
459 | ||
460 | /* guard against a misconfigured or uncooperative backend that | |
461 | * is sending packet larger than the MTU. | |
462 | */ | |
463 | if ((page_off + buflen) > PAGE_SIZE) { | |
464 | put_page(p); | |
465 | goto err_buf; | |
466 | } | |
467 | ||
468 | memcpy(page_address(page) + page_off, | |
469 | page_address(p) + off, buflen); | |
470 | page_off += buflen; | |
471 | put_page(p); | |
472 | } | |
473 | ||
474 | /* Headroom does not contribute to packet length */ | |
475 | *len = page_off - VIRTIO_XDP_HEADROOM; | |
476 | return page; | |
477 | err_buf: | |
478 | __free_pages(page, 0); | |
479 | return NULL; | |
480 | } | |
481 | ||
bb91accf JW |
482 | static struct sk_buff *receive_small(struct net_device *dev, |
483 | struct virtnet_info *vi, | |
484 | struct receive_queue *rq, | |
192f68cf JW |
485 | void *buf, void *ctx, |
486 | unsigned int len) | |
f121159d | 487 | { |
f6b10209 | 488 | struct sk_buff *skb; |
bb91accf | 489 | struct bpf_prog *xdp_prog; |
4941d472 | 490 | unsigned int xdp_headroom = (unsigned long)ctx; |
f6b10209 JW |
491 | unsigned int header_offset = VIRTNET_RX_PAD + xdp_headroom; |
492 | unsigned int headroom = vi->hdr_len + header_offset; | |
493 | unsigned int buflen = SKB_DATA_ALIGN(GOOD_PACKET_LEN + headroom) + | |
494 | SKB_DATA_ALIGN(sizeof(struct skb_shared_info)); | |
4941d472 | 495 | struct page *page = virt_to_head_page(buf); |
f6b10209 | 496 | unsigned int delta = 0; |
4941d472 | 497 | struct page *xdp_page; |
012873d0 | 498 | len -= vi->hdr_len; |
f121159d | 499 | |
bb91accf JW |
500 | rcu_read_lock(); |
501 | xdp_prog = rcu_dereference(rq->xdp_prog); | |
502 | if (xdp_prog) { | |
f6b10209 | 503 | struct virtio_net_hdr_mrg_rxbuf *hdr = buf + header_offset; |
0354e4d1 | 504 | struct xdp_buff xdp; |
f6b10209 | 505 | void *orig_data; |
bb91accf JW |
506 | u32 act; |
507 | ||
508 | if (unlikely(hdr->hdr.gso_type || hdr->hdr.flags)) | |
509 | goto err_xdp; | |
0354e4d1 | 510 | |
4941d472 JW |
511 | if (unlikely(xdp_headroom < virtnet_get_headroom(vi))) { |
512 | int offset = buf - page_address(page) + header_offset; | |
513 | unsigned int tlen = len + vi->hdr_len; | |
514 | u16 num_buf = 1; | |
515 | ||
516 | xdp_headroom = virtnet_get_headroom(vi); | |
517 | header_offset = VIRTNET_RX_PAD + xdp_headroom; | |
518 | headroom = vi->hdr_len + header_offset; | |
519 | buflen = SKB_DATA_ALIGN(GOOD_PACKET_LEN + headroom) + | |
520 | SKB_DATA_ALIGN(sizeof(struct skb_shared_info)); | |
521 | xdp_page = xdp_linearize_page(rq, &num_buf, page, | |
522 | offset, header_offset, | |
523 | &tlen); | |
524 | if (!xdp_page) | |
525 | goto err_xdp; | |
526 | ||
527 | buf = page_address(xdp_page); | |
528 | put_page(page); | |
529 | page = xdp_page; | |
530 | } | |
531 | ||
f6b10209 JW |
532 | xdp.data_hard_start = buf + VIRTNET_RX_PAD + vi->hdr_len; |
533 | xdp.data = xdp.data_hard_start + xdp_headroom; | |
0354e4d1 | 534 | xdp.data_end = xdp.data + len; |
f6b10209 | 535 | orig_data = xdp.data; |
0354e4d1 JF |
536 | act = bpf_prog_run_xdp(xdp_prog, &xdp); |
537 | ||
bb91accf JW |
538 | switch (act) { |
539 | case XDP_PASS: | |
2de2f7f4 | 540 | /* Recalculate length in case bpf program changed it */ |
f6b10209 | 541 | delta = orig_data - xdp.data; |
bb91accf JW |
542 | break; |
543 | case XDP_TX: | |
9457642a | 544 | if (unlikely(!virtnet_xdp_xmit(vi, &xdp))) |
0354e4d1 | 545 | trace_xdp_exception(vi->dev, xdp_prog, act); |
bb91accf JW |
546 | rcu_read_unlock(); |
547 | goto xdp_xmit; | |
bb91accf | 548 | default: |
0354e4d1 JF |
549 | bpf_warn_invalid_xdp_action(act); |
550 | case XDP_ABORTED: | |
551 | trace_xdp_exception(vi->dev, xdp_prog, act); | |
552 | case XDP_DROP: | |
bb91accf JW |
553 | goto err_xdp; |
554 | } | |
555 | } | |
556 | rcu_read_unlock(); | |
557 | ||
f6b10209 JW |
558 | skb = build_skb(buf, buflen); |
559 | if (!skb) { | |
4941d472 | 560 | put_page(page); |
f6b10209 JW |
561 | goto err; |
562 | } | |
563 | skb_reserve(skb, headroom - delta); | |
564 | skb_put(skb, len + delta); | |
565 | if (!delta) { | |
566 | buf += header_offset; | |
567 | memcpy(skb_vnet_hdr(skb), buf, vi->hdr_len); | |
568 | } /* keep zeroed vnet hdr since packet was changed by bpf */ | |
569 | ||
570 | err: | |
f121159d | 571 | return skb; |
bb91accf JW |
572 | |
573 | err_xdp: | |
574 | rcu_read_unlock(); | |
575 | dev->stats.rx_dropped++; | |
4941d472 | 576 | put_page(page); |
bb91accf JW |
577 | xdp_xmit: |
578 | return NULL; | |
f121159d MT |
579 | } |
580 | ||
581 | static struct sk_buff *receive_big(struct net_device *dev, | |
946fa564 | 582 | struct virtnet_info *vi, |
f121159d MT |
583 | struct receive_queue *rq, |
584 | void *buf, | |
585 | unsigned int len) | |
586 | { | |
587 | struct page *page = buf; | |
c47a43d3 | 588 | struct sk_buff *skb = page_to_skb(vi, rq, page, 0, len, PAGE_SIZE); |
f600b690 | 589 | |
f121159d MT |
590 | if (unlikely(!skb)) |
591 | goto err; | |
592 | ||
593 | return skb; | |
594 | ||
595 | err: | |
596 | dev->stats.rx_dropped++; | |
597 | give_pages(rq, page); | |
598 | return NULL; | |
599 | } | |
600 | ||
8fc3b9e9 | 601 | static struct sk_buff *receive_mergeable(struct net_device *dev, |
fdd819b2 | 602 | struct virtnet_info *vi, |
8fc3b9e9 | 603 | struct receive_queue *rq, |
680557cf MT |
604 | void *buf, |
605 | void *ctx, | |
8fc3b9e9 | 606 | unsigned int len) |
9ab86bbc | 607 | { |
012873d0 MT |
608 | struct virtio_net_hdr_mrg_rxbuf *hdr = buf; |
609 | u16 num_buf = virtio16_to_cpu(vi->vdev, hdr->num_buffers); | |
8fc3b9e9 MT |
610 | struct page *page = virt_to_head_page(buf); |
611 | int offset = buf - page_address(page); | |
f600b690 JF |
612 | struct sk_buff *head_skb, *curr_skb; |
613 | struct bpf_prog *xdp_prog; | |
614 | unsigned int truesize; | |
4941d472 | 615 | unsigned int headroom = mergeable_ctx_to_headroom(ctx); |
f600b690 | 616 | |
56434a01 JF |
617 | head_skb = NULL; |
618 | ||
f600b690 JF |
619 | rcu_read_lock(); |
620 | xdp_prog = rcu_dereference(rq->xdp_prog); | |
621 | if (xdp_prog) { | |
72979a6c | 622 | struct page *xdp_page; |
0354e4d1 | 623 | struct xdp_buff xdp; |
0354e4d1 | 624 | void *data; |
f600b690 JF |
625 | u32 act; |
626 | ||
73b62bd0 | 627 | /* This happens when rx buffer size is underestimated */ |
4941d472 JW |
628 | if (unlikely(num_buf > 1 || |
629 | headroom < virtnet_get_headroom(vi))) { | |
72979a6c | 630 | /* linearize data for XDP */ |
56a86f84 | 631 | xdp_page = xdp_linearize_page(rq, &num_buf, |
4941d472 JW |
632 | page, offset, |
633 | VIRTIO_XDP_HEADROOM, | |
634 | &len); | |
72979a6c JF |
635 | if (!xdp_page) |
636 | goto err_xdp; | |
2de2f7f4 | 637 | offset = VIRTIO_XDP_HEADROOM; |
72979a6c JF |
638 | } else { |
639 | xdp_page = page; | |
f600b690 JF |
640 | } |
641 | ||
642 | /* Transient failure which in theory could occur if | |
643 | * in-flight packets from before XDP was enabled reach | |
644 | * the receive path after XDP is loaded. In practice I | |
645 | * was not able to create this condition. | |
646 | */ | |
b00f70b0 | 647 | if (unlikely(hdr->hdr.gso_type)) |
f600b690 JF |
648 | goto err_xdp; |
649 | ||
2de2f7f4 JF |
650 | /* Allow consuming headroom but reserve enough space to push |
651 | * the descriptor on if we get an XDP_TX return code. | |
652 | */ | |
0354e4d1 | 653 | data = page_address(xdp_page) + offset; |
2de2f7f4 | 654 | xdp.data_hard_start = data - VIRTIO_XDP_HEADROOM + vi->hdr_len; |
0354e4d1 JF |
655 | xdp.data = data + vi->hdr_len; |
656 | xdp.data_end = xdp.data + (len - vi->hdr_len); | |
657 | act = bpf_prog_run_xdp(xdp_prog, &xdp); | |
658 | ||
56434a01 JF |
659 | switch (act) { |
660 | case XDP_PASS: | |
2de2f7f4 JF |
661 | /* recalculate offset to account for any header |
662 | * adjustments. Note other cases do not build an | |
663 | * skb and avoid using offset | |
664 | */ | |
665 | offset = xdp.data - | |
666 | page_address(xdp_page) - vi->hdr_len; | |
667 | ||
1830f893 JW |
668 | /* We can only create skb based on xdp_page. */ |
669 | if (unlikely(xdp_page != page)) { | |
670 | rcu_read_unlock(); | |
671 | put_page(page); | |
672 | head_skb = page_to_skb(vi, rq, xdp_page, | |
2de2f7f4 | 673 | offset, len, PAGE_SIZE); |
5c33474d | 674 | ewma_pkt_len_add(&rq->mrg_avg_pkt_len, len); |
1830f893 JW |
675 | return head_skb; |
676 | } | |
56434a01 JF |
677 | break; |
678 | case XDP_TX: | |
9457642a | 679 | if (unlikely(!virtnet_xdp_xmit(vi, &xdp))) |
0354e4d1 | 680 | trace_xdp_exception(vi->dev, xdp_prog, act); |
5c33474d | 681 | ewma_pkt_len_add(&rq->mrg_avg_pkt_len, len); |
72979a6c JF |
682 | if (unlikely(xdp_page != page)) |
683 | goto err_xdp; | |
56434a01 JF |
684 | rcu_read_unlock(); |
685 | goto xdp_xmit; | |
56434a01 | 686 | default: |
0354e4d1 JF |
687 | bpf_warn_invalid_xdp_action(act); |
688 | case XDP_ABORTED: | |
689 | trace_xdp_exception(vi->dev, xdp_prog, act); | |
690 | case XDP_DROP: | |
72979a6c JF |
691 | if (unlikely(xdp_page != page)) |
692 | __free_pages(xdp_page, 0); | |
5c33474d | 693 | ewma_pkt_len_add(&rq->mrg_avg_pkt_len, len); |
f600b690 | 694 | goto err_xdp; |
56434a01 | 695 | } |
f600b690 JF |
696 | } |
697 | rcu_read_unlock(); | |
ab7db917 | 698 | |
28b39bc7 JW |
699 | truesize = mergeable_ctx_to_truesize(ctx); |
700 | if (unlikely(len > truesize)) { | |
56da5fd0 | 701 | pr_debug("%s: rx error: len %u exceeds truesize %lu\n", |
680557cf MT |
702 | dev->name, len, (unsigned long)ctx); |
703 | dev->stats.rx_length_errors++; | |
704 | goto err_skb; | |
705 | } | |
28b39bc7 | 706 | |
f600b690 JF |
707 | head_skb = page_to_skb(vi, rq, page, offset, len, truesize); |
708 | curr_skb = head_skb; | |
9ab86bbc | 709 | |
8fc3b9e9 MT |
710 | if (unlikely(!curr_skb)) |
711 | goto err_skb; | |
9ab86bbc | 712 | while (--num_buf) { |
8fc3b9e9 MT |
713 | int num_skb_frags; |
714 | ||
680557cf | 715 | buf = virtqueue_get_buf_ctx(rq->vq, &len, &ctx); |
ab7db917 | 716 | if (unlikely(!ctx)) { |
8fc3b9e9 | 717 | pr_debug("%s: rx error: %d buffers out of %d missing\n", |
fdd819b2 | 718 | dev->name, num_buf, |
012873d0 MT |
719 | virtio16_to_cpu(vi->vdev, |
720 | hdr->num_buffers)); | |
8fc3b9e9 MT |
721 | dev->stats.rx_length_errors++; |
722 | goto err_buf; | |
3f2c31d9 | 723 | } |
8fc3b9e9 MT |
724 | |
725 | page = virt_to_head_page(buf); | |
28b39bc7 JW |
726 | |
727 | truesize = mergeable_ctx_to_truesize(ctx); | |
728 | if (unlikely(len > truesize)) { | |
56da5fd0 | 729 | pr_debug("%s: rx error: len %u exceeds truesize %lu\n", |
680557cf MT |
730 | dev->name, len, (unsigned long)ctx); |
731 | dev->stats.rx_length_errors++; | |
732 | goto err_skb; | |
733 | } | |
8fc3b9e9 MT |
734 | |
735 | num_skb_frags = skb_shinfo(curr_skb)->nr_frags; | |
2613af0e MD |
736 | if (unlikely(num_skb_frags == MAX_SKB_FRAGS)) { |
737 | struct sk_buff *nskb = alloc_skb(0, GFP_ATOMIC); | |
8fc3b9e9 MT |
738 | |
739 | if (unlikely(!nskb)) | |
740 | goto err_skb; | |
2613af0e MD |
741 | if (curr_skb == head_skb) |
742 | skb_shinfo(curr_skb)->frag_list = nskb; | |
743 | else | |
744 | curr_skb->next = nskb; | |
745 | curr_skb = nskb; | |
746 | head_skb->truesize += nskb->truesize; | |
747 | num_skb_frags = 0; | |
748 | } | |
749 | if (curr_skb != head_skb) { | |
750 | head_skb->data_len += len; | |
751 | head_skb->len += len; | |
fb51879d | 752 | head_skb->truesize += truesize; |
2613af0e | 753 | } |
8fc3b9e9 | 754 | offset = buf - page_address(page); |
ba275241 JW |
755 | if (skb_can_coalesce(curr_skb, num_skb_frags, page, offset)) { |
756 | put_page(page); | |
757 | skb_coalesce_rx_frag(curr_skb, num_skb_frags - 1, | |
fb51879d | 758 | len, truesize); |
ba275241 JW |
759 | } else { |
760 | skb_add_rx_frag(curr_skb, num_skb_frags, page, | |
fb51879d | 761 | offset, len, truesize); |
ba275241 | 762 | } |
8fc3b9e9 MT |
763 | } |
764 | ||
5377d758 | 765 | ewma_pkt_len_add(&rq->mrg_avg_pkt_len, head_skb->len); |
8fc3b9e9 MT |
766 | return head_skb; |
767 | ||
f600b690 JF |
768 | err_xdp: |
769 | rcu_read_unlock(); | |
8fc3b9e9 MT |
770 | err_skb: |
771 | put_page(page); | |
772 | while (--num_buf) { | |
680557cf MT |
773 | buf = virtqueue_get_buf(rq->vq, &len); |
774 | if (unlikely(!buf)) { | |
8fc3b9e9 MT |
775 | pr_debug("%s: rx error: %d buffers missing\n", |
776 | dev->name, num_buf); | |
777 | dev->stats.rx_length_errors++; | |
778 | break; | |
779 | } | |
680557cf | 780 | page = virt_to_head_page(buf); |
8fc3b9e9 | 781 | put_page(page); |
9ab86bbc | 782 | } |
8fc3b9e9 MT |
783 | err_buf: |
784 | dev->stats.rx_dropped++; | |
785 | dev_kfree_skb(head_skb); | |
56434a01 | 786 | xdp_xmit: |
8fc3b9e9 | 787 | return NULL; |
9ab86bbc SM |
788 | } |
789 | ||
61845d20 | 790 | static int receive_buf(struct virtnet_info *vi, struct receive_queue *rq, |
680557cf | 791 | void *buf, unsigned int len, void **ctx) |
9ab86bbc | 792 | { |
e9d7417b | 793 | struct net_device *dev = vi->dev; |
9ab86bbc | 794 | struct sk_buff *skb; |
012873d0 | 795 | struct virtio_net_hdr_mrg_rxbuf *hdr; |
61845d20 | 796 | int ret; |
3f2c31d9 | 797 | |
bcff3162 | 798 | if (unlikely(len < vi->hdr_len + ETH_HLEN)) { |
9ab86bbc SM |
799 | pr_debug("%s: short packet %i\n", dev->name, len); |
800 | dev->stats.rx_length_errors++; | |
ab7db917 | 801 | if (vi->mergeable_rx_bufs) { |
680557cf | 802 | put_page(virt_to_head_page(buf)); |
ab7db917 | 803 | } else if (vi->big_packets) { |
98bfd23c | 804 | give_pages(rq, buf); |
ab7db917 | 805 | } else { |
f6b10209 | 806 | put_page(virt_to_head_page(buf)); |
ab7db917 | 807 | } |
61845d20 | 808 | return 0; |
9ab86bbc | 809 | } |
3f2c31d9 | 810 | |
f121159d | 811 | if (vi->mergeable_rx_bufs) |
680557cf | 812 | skb = receive_mergeable(dev, vi, rq, buf, ctx, len); |
f121159d | 813 | else if (vi->big_packets) |
946fa564 | 814 | skb = receive_big(dev, vi, rq, buf, len); |
f121159d | 815 | else |
192f68cf | 816 | skb = receive_small(dev, vi, rq, buf, ctx, len); |
f121159d MT |
817 | |
818 | if (unlikely(!skb)) | |
61845d20 | 819 | return 0; |
3f2c31d9 | 820 | |
9ab86bbc | 821 | hdr = skb_vnet_hdr(skb); |
3fa2a1df | 822 | |
61845d20 | 823 | ret = skb->len; |
296f96fc | 824 | |
e858fae2 | 825 | if (hdr->hdr.flags & VIRTIO_NET_HDR_F_DATA_VALID) |
10a8d94a | 826 | skb->ip_summed = CHECKSUM_UNNECESSARY; |
296f96fc | 827 | |
e858fae2 MR |
828 | if (virtio_net_hdr_to_skb(skb, &hdr->hdr, |
829 | virtio_is_little_endian(vi->vdev))) { | |
830 | net_warn_ratelimited("%s: bad gso: type: %u, size: %u\n", | |
831 | dev->name, hdr->hdr.gso_type, | |
832 | hdr->hdr.gso_size); | |
833 | goto frame_err; | |
296f96fc RR |
834 | } |
835 | ||
d1dc06dc MR |
836 | skb->protocol = eth_type_trans(skb, dev); |
837 | pr_debug("Receiving skb proto 0x%04x len %i type %i\n", | |
838 | ntohs(skb->protocol), skb->len, skb->pkt_type); | |
839 | ||
0fbd050a | 840 | napi_gro_receive(&rq->napi, skb); |
61845d20 | 841 | return ret; |
296f96fc RR |
842 | |
843 | frame_err: | |
844 | dev->stats.rx_frame_errors++; | |
296f96fc | 845 | dev_kfree_skb(skb); |
61845d20 | 846 | return 0; |
296f96fc RR |
847 | } |
848 | ||
192f68cf JW |
849 | /* Unlike mergeable buffers, all buffers are allocated to the |
850 | * same size, except for the headroom. For this reason we do | |
851 | * not need to use mergeable_len_to_ctx here - it is enough | |
852 | * to store the headroom as the context ignoring the truesize. | |
853 | */ | |
946fa564 MT |
854 | static int add_recvbuf_small(struct virtnet_info *vi, struct receive_queue *rq, |
855 | gfp_t gfp) | |
296f96fc | 856 | { |
f6b10209 JW |
857 | struct page_frag *alloc_frag = &rq->alloc_frag; |
858 | char *buf; | |
2de2f7f4 | 859 | unsigned int xdp_headroom = virtnet_get_headroom(vi); |
192f68cf | 860 | void *ctx = (void *)(unsigned long)xdp_headroom; |
f6b10209 | 861 | int len = vi->hdr_len + VIRTNET_RX_PAD + GOOD_PACKET_LEN + xdp_headroom; |
9ab86bbc | 862 | int err; |
3f2c31d9 | 863 | |
f6b10209 JW |
864 | len = SKB_DATA_ALIGN(len) + |
865 | SKB_DATA_ALIGN(sizeof(struct skb_shared_info)); | |
866 | if (unlikely(!skb_page_frag_refill(len, alloc_frag, gfp))) | |
9ab86bbc | 867 | return -ENOMEM; |
296f96fc | 868 | |
f6b10209 JW |
869 | buf = (char *)page_address(alloc_frag->page) + alloc_frag->offset; |
870 | get_page(alloc_frag->page); | |
871 | alloc_frag->offset += len; | |
872 | sg_init_one(rq->sg, buf + VIRTNET_RX_PAD + xdp_headroom, | |
873 | vi->hdr_len + GOOD_PACKET_LEN); | |
192f68cf | 874 | err = virtqueue_add_inbuf_ctx(rq->vq, rq->sg, 1, buf, ctx, gfp); |
9ab86bbc | 875 | if (err < 0) |
f6b10209 | 876 | put_page(virt_to_head_page(buf)); |
9ab86bbc SM |
877 | return err; |
878 | } | |
97402b96 | 879 | |
012873d0 MT |
880 | static int add_recvbuf_big(struct virtnet_info *vi, struct receive_queue *rq, |
881 | gfp_t gfp) | |
9ab86bbc | 882 | { |
9ab86bbc SM |
883 | struct page *first, *list = NULL; |
884 | char *p; | |
885 | int i, err, offset; | |
886 | ||
a5835440 RR |
887 | sg_init_table(rq->sg, MAX_SKB_FRAGS + 2); |
888 | ||
e9d7417b | 889 | /* page in rq->sg[MAX_SKB_FRAGS + 1] is list tail */ |
9ab86bbc | 890 | for (i = MAX_SKB_FRAGS + 1; i > 1; --i) { |
e9d7417b | 891 | first = get_a_page(rq, gfp); |
9ab86bbc SM |
892 | if (!first) { |
893 | if (list) | |
e9d7417b | 894 | give_pages(rq, list); |
9ab86bbc | 895 | return -ENOMEM; |
97402b96 | 896 | } |
e9d7417b | 897 | sg_set_buf(&rq->sg[i], page_address(first), PAGE_SIZE); |
97402b96 | 898 | |
9ab86bbc SM |
899 | /* chain new page in list head to match sg */ |
900 | first->private = (unsigned long)list; | |
901 | list = first; | |
902 | } | |
296f96fc | 903 | |
e9d7417b | 904 | first = get_a_page(rq, gfp); |
9ab86bbc | 905 | if (!first) { |
e9d7417b | 906 | give_pages(rq, list); |
9ab86bbc SM |
907 | return -ENOMEM; |
908 | } | |
909 | p = page_address(first); | |
910 | ||
e9d7417b | 911 | /* rq->sg[0], rq->sg[1] share the same page */ |
012873d0 MT |
912 | /* a separated rq->sg[0] for header - required in case !any_header_sg */ |
913 | sg_set_buf(&rq->sg[0], p, vi->hdr_len); | |
9ab86bbc | 914 | |
e9d7417b | 915 | /* rq->sg[1] for data packet, from offset */ |
9ab86bbc | 916 | offset = sizeof(struct padded_vnet_hdr); |
e9d7417b | 917 | sg_set_buf(&rq->sg[1], p + offset, PAGE_SIZE - offset); |
9ab86bbc SM |
918 | |
919 | /* chain first in list head */ | |
920 | first->private = (unsigned long)list; | |
9dc7b9e4 RR |
921 | err = virtqueue_add_inbuf(rq->vq, rq->sg, MAX_SKB_FRAGS + 2, |
922 | first, gfp); | |
9ab86bbc | 923 | if (err < 0) |
e9d7417b | 924 | give_pages(rq, first); |
9ab86bbc SM |
925 | |
926 | return err; | |
296f96fc RR |
927 | } |
928 | ||
d85b758f MT |
929 | static unsigned int get_mergeable_buf_len(struct receive_queue *rq, |
930 | struct ewma_pkt_len *avg_pkt_len) | |
3f2c31d9 | 931 | { |
ab7db917 | 932 | const size_t hdr_len = sizeof(struct virtio_net_hdr_mrg_rxbuf); |
fbf28d78 MD |
933 | unsigned int len; |
934 | ||
5377d758 | 935 | len = hdr_len + clamp_t(unsigned int, ewma_pkt_len_read(avg_pkt_len), |
f0c3192c | 936 | rq->min_buf_len, PAGE_SIZE - hdr_len); |
e377fcc8 | 937 | return ALIGN(len, L1_CACHE_BYTES); |
fbf28d78 MD |
938 | } |
939 | ||
2de2f7f4 JF |
940 | static int add_recvbuf_mergeable(struct virtnet_info *vi, |
941 | struct receive_queue *rq, gfp_t gfp) | |
fbf28d78 | 942 | { |
fb51879d | 943 | struct page_frag *alloc_frag = &rq->alloc_frag; |
2de2f7f4 | 944 | unsigned int headroom = virtnet_get_headroom(vi); |
fb51879d | 945 | char *buf; |
680557cf | 946 | void *ctx; |
3f2c31d9 | 947 | int err; |
fb51879d | 948 | unsigned int len, hole; |
3f2c31d9 | 949 | |
d85b758f | 950 | len = get_mergeable_buf_len(rq, &rq->mrg_avg_pkt_len); |
2de2f7f4 | 951 | if (unlikely(!skb_page_frag_refill(len + headroom, alloc_frag, gfp))) |
9ab86bbc | 952 | return -ENOMEM; |
ab7db917 | 953 | |
fb51879d | 954 | buf = (char *)page_address(alloc_frag->page) + alloc_frag->offset; |
2de2f7f4 | 955 | buf += headroom; /* advance address leaving hole at front of pkt */ |
fb51879d | 956 | get_page(alloc_frag->page); |
2de2f7f4 | 957 | alloc_frag->offset += len + headroom; |
fb51879d | 958 | hole = alloc_frag->size - alloc_frag->offset; |
2de2f7f4 | 959 | if (hole < len + headroom) { |
ab7db917 MD |
960 | /* To avoid internal fragmentation, if there is very likely not |
961 | * enough space for another buffer, add the remaining space to | |
1daa8790 | 962 | * the current buffer. |
ab7db917 | 963 | */ |
fb51879d MD |
964 | len += hole; |
965 | alloc_frag->offset += hole; | |
966 | } | |
3f2c31d9 | 967 | |
fb51879d | 968 | sg_init_one(rq->sg, buf, len); |
29fda25a | 969 | ctx = mergeable_len_to_ctx(len, headroom); |
680557cf | 970 | err = virtqueue_add_inbuf_ctx(rq->vq, rq->sg, 1, buf, ctx, gfp); |
9ab86bbc | 971 | if (err < 0) |
2613af0e | 972 | put_page(virt_to_head_page(buf)); |
3f2c31d9 | 973 | |
9ab86bbc SM |
974 | return err; |
975 | } | |
3f2c31d9 | 976 | |
b2baed69 RR |
977 | /* |
978 | * Returns false if we couldn't fill entirely (OOM). | |
979 | * | |
980 | * Normally run in the receive path, but can also be run from ndo_open | |
981 | * before we're receiving packets, or from refill_work which is | |
982 | * careful to disable receiving (using napi_disable). | |
983 | */ | |
946fa564 MT |
984 | static bool try_fill_recv(struct virtnet_info *vi, struct receive_queue *rq, |
985 | gfp_t gfp) | |
9ab86bbc SM |
986 | { |
987 | int err; | |
1788f495 | 988 | bool oom; |
3f2c31d9 | 989 | |
fb51879d | 990 | gfp |= __GFP_COLD; |
9ab86bbc SM |
991 | do { |
992 | if (vi->mergeable_rx_bufs) | |
2de2f7f4 | 993 | err = add_recvbuf_mergeable(vi, rq, gfp); |
9ab86bbc | 994 | else if (vi->big_packets) |
012873d0 | 995 | err = add_recvbuf_big(vi, rq, gfp); |
9ab86bbc | 996 | else |
946fa564 | 997 | err = add_recvbuf_small(vi, rq, gfp); |
3f2c31d9 | 998 | |
1788f495 | 999 | oom = err == -ENOMEM; |
9ed4cb07 | 1000 | if (err) |
3f2c31d9 | 1001 | break; |
b7dfde95 | 1002 | } while (rq->vq->num_free); |
681daee2 | 1003 | virtqueue_kick(rq->vq); |
3161e453 | 1004 | return !oom; |
3f2c31d9 MM |
1005 | } |
1006 | ||
18445c4d | 1007 | static void skb_recv_done(struct virtqueue *rvq) |
296f96fc RR |
1008 | { |
1009 | struct virtnet_info *vi = rvq->vdev->priv; | |
986a4f4d | 1010 | struct receive_queue *rq = &vi->rq[vq2rxq(rvq)]; |
e9d7417b | 1011 | |
e4e8452a | 1012 | virtqueue_napi_schedule(&rq->napi, rvq); |
296f96fc RR |
1013 | } |
1014 | ||
e4e8452a | 1015 | static void virtnet_napi_enable(struct virtqueue *vq, struct napi_struct *napi) |
3e9d08ec | 1016 | { |
e4e8452a | 1017 | napi_enable(napi); |
3e9d08ec BR |
1018 | |
1019 | /* If all buffers were filled by other side before we napi_enabled, we | |
e4e8452a WB |
1020 | * won't get another interrupt, so process any outstanding packets now. |
1021 | * Call local_bh_enable after to trigger softIRQ processing. | |
1022 | */ | |
1023 | local_bh_disable(); | |
1024 | virtqueue_napi_schedule(napi, vq); | |
1025 | local_bh_enable(); | |
3e9d08ec BR |
1026 | } |
1027 | ||
b92f1e67 WB |
1028 | static void virtnet_napi_tx_enable(struct virtnet_info *vi, |
1029 | struct virtqueue *vq, | |
1030 | struct napi_struct *napi) | |
1031 | { | |
1032 | if (!napi->weight) | |
1033 | return; | |
1034 | ||
1035 | /* Tx napi touches cachelines on the cpu handling tx interrupts. Only | |
1036 | * enable the feature if this is likely affine with the transmit path. | |
1037 | */ | |
1038 | if (!vi->affinity_hint_set) { | |
1039 | napi->weight = 0; | |
1040 | return; | |
1041 | } | |
1042 | ||
1043 | return virtnet_napi_enable(vq, napi); | |
1044 | } | |
1045 | ||
78a57b48 WB |
1046 | static void virtnet_napi_tx_disable(struct napi_struct *napi) |
1047 | { | |
1048 | if (napi->weight) | |
1049 | napi_disable(napi); | |
1050 | } | |
1051 | ||
3161e453 RR |
1052 | static void refill_work(struct work_struct *work) |
1053 | { | |
e9d7417b JW |
1054 | struct virtnet_info *vi = |
1055 | container_of(work, struct virtnet_info, refill.work); | |
3161e453 | 1056 | bool still_empty; |
986a4f4d JW |
1057 | int i; |
1058 | ||
55257d72 | 1059 | for (i = 0; i < vi->curr_queue_pairs; i++) { |
986a4f4d | 1060 | struct receive_queue *rq = &vi->rq[i]; |
3161e453 | 1061 | |
986a4f4d | 1062 | napi_disable(&rq->napi); |
946fa564 | 1063 | still_empty = !try_fill_recv(vi, rq, GFP_KERNEL); |
e4e8452a | 1064 | virtnet_napi_enable(rq->vq, &rq->napi); |
3161e453 | 1065 | |
986a4f4d JW |
1066 | /* In theory, this can happen: if we don't get any buffers in |
1067 | * we will *never* try to fill again. | |
1068 | */ | |
1069 | if (still_empty) | |
1070 | schedule_delayed_work(&vi->refill, HZ/2); | |
1071 | } | |
3161e453 RR |
1072 | } |
1073 | ||
2ffa7598 | 1074 | static int virtnet_receive(struct receive_queue *rq, int budget) |
296f96fc | 1075 | { |
e9d7417b | 1076 | struct virtnet_info *vi = rq->vq->vdev->priv; |
61845d20 | 1077 | unsigned int len, received = 0, bytes = 0; |
9ab86bbc | 1078 | void *buf; |
61845d20 | 1079 | struct virtnet_stats *stats = this_cpu_ptr(vi->stats); |
296f96fc | 1080 | |
192f68cf | 1081 | if (!vi->big_packets || vi->mergeable_rx_bufs) { |
680557cf MT |
1082 | void *ctx; |
1083 | ||
1084 | while (received < budget && | |
1085 | (buf = virtqueue_get_buf_ctx(rq->vq, &len, &ctx))) { | |
1086 | bytes += receive_buf(vi, rq, buf, len, ctx); | |
1087 | received++; | |
1088 | } | |
1089 | } else { | |
1090 | while (received < budget && | |
1091 | (buf = virtqueue_get_buf(rq->vq, &len)) != NULL) { | |
1092 | bytes += receive_buf(vi, rq, buf, len, NULL); | |
1093 | received++; | |
1094 | } | |
296f96fc RR |
1095 | } |
1096 | ||
be121f46 | 1097 | if (rq->vq->num_free > virtqueue_get_vring_size(rq->vq) / 2) { |
946fa564 | 1098 | if (!try_fill_recv(vi, rq, GFP_ATOMIC)) |
3b07e9ca | 1099 | schedule_delayed_work(&vi->refill, 0); |
3161e453 | 1100 | } |
296f96fc | 1101 | |
61845d20 JW |
1102 | u64_stats_update_begin(&stats->rx_syncp); |
1103 | stats->rx_bytes += bytes; | |
1104 | stats->rx_packets += received; | |
1105 | u64_stats_update_end(&stats->rx_syncp); | |
1106 | ||
2ffa7598 JW |
1107 | return received; |
1108 | } | |
1109 | ||
ea7735d9 WB |
1110 | static void free_old_xmit_skbs(struct send_queue *sq) |
1111 | { | |
1112 | struct sk_buff *skb; | |
1113 | unsigned int len; | |
1114 | struct virtnet_info *vi = sq->vq->vdev->priv; | |
1115 | struct virtnet_stats *stats = this_cpu_ptr(vi->stats); | |
1116 | unsigned int packets = 0; | |
1117 | unsigned int bytes = 0; | |
1118 | ||
1119 | while ((skb = virtqueue_get_buf(sq->vq, &len)) != NULL) { | |
1120 | pr_debug("Sent skb %p\n", skb); | |
1121 | ||
1122 | bytes += skb->len; | |
1123 | packets++; | |
1124 | ||
dadc0736 | 1125 | dev_consume_skb_any(skb); |
ea7735d9 WB |
1126 | } |
1127 | ||
1128 | /* Avoid overhead when no packets have been processed | |
1129 | * happens when called speculatively from start_xmit. | |
1130 | */ | |
1131 | if (!packets) | |
1132 | return; | |
1133 | ||
1134 | u64_stats_update_begin(&stats->tx_syncp); | |
1135 | stats->tx_bytes += bytes; | |
1136 | stats->tx_packets += packets; | |
1137 | u64_stats_update_end(&stats->tx_syncp); | |
1138 | } | |
1139 | ||
7b0411ef WB |
1140 | static void virtnet_poll_cleantx(struct receive_queue *rq) |
1141 | { | |
1142 | struct virtnet_info *vi = rq->vq->vdev->priv; | |
1143 | unsigned int index = vq2rxq(rq->vq); | |
1144 | struct send_queue *sq = &vi->sq[index]; | |
1145 | struct netdev_queue *txq = netdev_get_tx_queue(vi->dev, index); | |
1146 | ||
1147 | if (!sq->napi.weight) | |
1148 | return; | |
1149 | ||
1150 | if (__netif_tx_trylock(txq)) { | |
1151 | free_old_xmit_skbs(sq); | |
1152 | __netif_tx_unlock(txq); | |
1153 | } | |
1154 | ||
1155 | if (sq->vq->num_free >= 2 + MAX_SKB_FRAGS) | |
1156 | netif_tx_wake_queue(txq); | |
1157 | } | |
1158 | ||
2ffa7598 JW |
1159 | static int virtnet_poll(struct napi_struct *napi, int budget) |
1160 | { | |
1161 | struct receive_queue *rq = | |
1162 | container_of(napi, struct receive_queue, napi); | |
e4e8452a | 1163 | unsigned int received; |
2ffa7598 | 1164 | |
7b0411ef WB |
1165 | virtnet_poll_cleantx(rq); |
1166 | ||
faadb05f | 1167 | received = virtnet_receive(rq, budget); |
2ffa7598 | 1168 | |
8329d98e | 1169 | /* Out of packets? */ |
e4e8452a WB |
1170 | if (received < budget) |
1171 | virtqueue_napi_complete(napi, rq->vq, received); | |
296f96fc RR |
1172 | |
1173 | return received; | |
1174 | } | |
1175 | ||
986a4f4d JW |
1176 | static int virtnet_open(struct net_device *dev) |
1177 | { | |
1178 | struct virtnet_info *vi = netdev_priv(dev); | |
1179 | int i; | |
1180 | ||
e4166625 JW |
1181 | for (i = 0; i < vi->max_queue_pairs; i++) { |
1182 | if (i < vi->curr_queue_pairs) | |
1183 | /* Make sure we have some buffers: if oom use wq. */ | |
946fa564 | 1184 | if (!try_fill_recv(vi, &vi->rq[i], GFP_KERNEL)) |
e4166625 | 1185 | schedule_delayed_work(&vi->refill, 0); |
e4e8452a | 1186 | virtnet_napi_enable(vi->rq[i].vq, &vi->rq[i].napi); |
b92f1e67 | 1187 | virtnet_napi_tx_enable(vi, vi->sq[i].vq, &vi->sq[i].napi); |
986a4f4d JW |
1188 | } |
1189 | ||
1190 | return 0; | |
1191 | } | |
1192 | ||
b92f1e67 WB |
1193 | static int virtnet_poll_tx(struct napi_struct *napi, int budget) |
1194 | { | |
1195 | struct send_queue *sq = container_of(napi, struct send_queue, napi); | |
1196 | struct virtnet_info *vi = sq->vq->vdev->priv; | |
1197 | struct netdev_queue *txq = netdev_get_tx_queue(vi->dev, vq2txq(sq->vq)); | |
1198 | ||
1199 | __netif_tx_lock(txq, raw_smp_processor_id()); | |
1200 | free_old_xmit_skbs(sq); | |
1201 | __netif_tx_unlock(txq); | |
1202 | ||
1203 | virtqueue_napi_complete(napi, sq->vq, 0); | |
1204 | ||
1205 | if (sq->vq->num_free >= 2 + MAX_SKB_FRAGS) | |
1206 | netif_tx_wake_queue(txq); | |
1207 | ||
1208 | return 0; | |
1209 | } | |
1210 | ||
e9d7417b | 1211 | static int xmit_skb(struct send_queue *sq, struct sk_buff *skb) |
296f96fc | 1212 | { |
012873d0 | 1213 | struct virtio_net_hdr_mrg_rxbuf *hdr; |
296f96fc | 1214 | const unsigned char *dest = ((struct ethhdr *)skb->data)->h_dest; |
e9d7417b | 1215 | struct virtnet_info *vi = sq->vq->vdev->priv; |
e2fcad58 | 1216 | int num_sg; |
012873d0 | 1217 | unsigned hdr_len = vi->hdr_len; |
e7428e95 | 1218 | bool can_push; |
296f96fc | 1219 | |
e174961c | 1220 | pr_debug("%s: xmit %p %pM\n", vi->dev->name, skb, dest); |
e7428e95 MT |
1221 | |
1222 | can_push = vi->any_header_sg && | |
1223 | !((unsigned long)skb->data & (__alignof__(*hdr) - 1)) && | |
1224 | !skb_header_cloned(skb) && skb_headroom(skb) >= hdr_len; | |
1225 | /* Even if we can, don't push here yet as this would skew | |
1226 | * csum_start offset below. */ | |
1227 | if (can_push) | |
012873d0 | 1228 | hdr = (struct virtio_net_hdr_mrg_rxbuf *)(skb->data - hdr_len); |
e7428e95 MT |
1229 | else |
1230 | hdr = skb_vnet_hdr(skb); | |
296f96fc | 1231 | |
e858fae2 | 1232 | if (virtio_net_hdr_from_skb(skb, &hdr->hdr, |
6391a448 | 1233 | virtio_is_little_endian(vi->vdev), false)) |
e858fae2 | 1234 | BUG(); |
296f96fc | 1235 | |
3f2c31d9 | 1236 | if (vi->mergeable_rx_bufs) |
012873d0 | 1237 | hdr->num_buffers = 0; |
3f2c31d9 | 1238 | |
547c890c | 1239 | sg_init_table(sq->sg, skb_shinfo(skb)->nr_frags + (can_push ? 1 : 2)); |
e7428e95 MT |
1240 | if (can_push) { |
1241 | __skb_push(skb, hdr_len); | |
1242 | num_sg = skb_to_sgvec(skb, sq->sg, 0, skb->len); | |
e2fcad58 JD |
1243 | if (unlikely(num_sg < 0)) |
1244 | return num_sg; | |
e7428e95 MT |
1245 | /* Pull header back to avoid skew in tx bytes calculations. */ |
1246 | __skb_pull(skb, hdr_len); | |
1247 | } else { | |
1248 | sg_set_buf(sq->sg, hdr, hdr_len); | |
e2fcad58 JD |
1249 | num_sg = skb_to_sgvec(skb, sq->sg + 1, 0, skb->len); |
1250 | if (unlikely(num_sg < 0)) | |
1251 | return num_sg; | |
1252 | num_sg++; | |
e7428e95 | 1253 | } |
9dc7b9e4 | 1254 | return virtqueue_add_outbuf(sq->vq, sq->sg, num_sg, skb, GFP_ATOMIC); |
11a3a154 RR |
1255 | } |
1256 | ||
424efe9c | 1257 | static netdev_tx_t start_xmit(struct sk_buff *skb, struct net_device *dev) |
99ffc696 RR |
1258 | { |
1259 | struct virtnet_info *vi = netdev_priv(dev); | |
986a4f4d JW |
1260 | int qnum = skb_get_queue_mapping(skb); |
1261 | struct send_queue *sq = &vi->sq[qnum]; | |
9ed4cb07 | 1262 | int err; |
4b7fd2e6 MT |
1263 | struct netdev_queue *txq = netdev_get_tx_queue(dev, qnum); |
1264 | bool kick = !skb->xmit_more; | |
b92f1e67 | 1265 | bool use_napi = sq->napi.weight; |
2cb9c6ba | 1266 | |
2cb9c6ba | 1267 | /* Free up any pending old buffers before queueing new ones. */ |
e9d7417b | 1268 | free_old_xmit_skbs(sq); |
99ffc696 | 1269 | |
bdb12e0d WB |
1270 | if (use_napi && kick) |
1271 | virtqueue_enable_cb_delayed(sq->vq); | |
1272 | ||
074c3582 JK |
1273 | /* timestamp packet in software */ |
1274 | skb_tx_timestamp(skb); | |
1275 | ||
03f191ba | 1276 | /* Try to transmit */ |
b7dfde95 | 1277 | err = xmit_skb(sq, skb); |
48925e37 | 1278 | |
9ed4cb07 | 1279 | /* This should not happen! */ |
681daee2 | 1280 | if (unlikely(err)) { |
9ed4cb07 RR |
1281 | dev->stats.tx_fifo_errors++; |
1282 | if (net_ratelimit()) | |
1283 | dev_warn(&dev->dev, | |
b7dfde95 | 1284 | "Unexpected TXQ (%d) queue failure: %d\n", qnum, err); |
58eba97d | 1285 | dev->stats.tx_dropped++; |
85e94525 | 1286 | dev_kfree_skb_any(skb); |
58eba97d | 1287 | return NETDEV_TX_OK; |
296f96fc | 1288 | } |
03f191ba | 1289 | |
48925e37 | 1290 | /* Don't wait up for transmitted skbs to be freed. */ |
b92f1e67 WB |
1291 | if (!use_napi) { |
1292 | skb_orphan(skb); | |
1293 | nf_reset(skb); | |
1294 | } | |
48925e37 | 1295 | |
60302ff6 MT |
1296 | /* If running out of space, stop queue to avoid getting packets that we |
1297 | * are then unable to transmit. | |
1298 | * An alternative would be to force queuing layer to requeue the skb by | |
1299 | * returning NETDEV_TX_BUSY. However, NETDEV_TX_BUSY should not be | |
1300 | * returned in a normal path of operation: it means that driver is not | |
1301 | * maintaining the TX queue stop/start state properly, and causes | |
1302 | * the stack to do a non-trivial amount of useless work. | |
1303 | * Since most packets only take 1 or 2 ring slots, stopping the queue | |
1304 | * early means 16 slots are typically wasted. | |
d631b94e | 1305 | */ |
b7dfde95 | 1306 | if (sq->vq->num_free < 2+MAX_SKB_FRAGS) { |
986a4f4d | 1307 | netif_stop_subqueue(dev, qnum); |
b92f1e67 WB |
1308 | if (!use_napi && |
1309 | unlikely(!virtqueue_enable_cb_delayed(sq->vq))) { | |
48925e37 | 1310 | /* More just got used, free them then recheck. */ |
b7dfde95 LT |
1311 | free_old_xmit_skbs(sq); |
1312 | if (sq->vq->num_free >= 2+MAX_SKB_FRAGS) { | |
986a4f4d | 1313 | netif_start_subqueue(dev, qnum); |
e9d7417b | 1314 | virtqueue_disable_cb(sq->vq); |
48925e37 RR |
1315 | } |
1316 | } | |
99ffc696 | 1317 | } |
48925e37 | 1318 | |
4b7fd2e6 | 1319 | if (kick || netif_xmit_stopped(txq)) |
0b725a2c | 1320 | virtqueue_kick(sq->vq); |
296f96fc | 1321 | |
0b725a2c | 1322 | return NETDEV_TX_OK; |
c223a078 DM |
1323 | } |
1324 | ||
40cbfc37 AK |
1325 | /* |
1326 | * Send command via the control virtqueue and check status. Commands | |
1327 | * supported by the hypervisor, as indicated by feature bits, should | |
788a8b6d | 1328 | * never fail unless improperly formatted. |
40cbfc37 AK |
1329 | */ |
1330 | static bool virtnet_send_command(struct virtnet_info *vi, u8 class, u8 cmd, | |
d24bae32 | 1331 | struct scatterlist *out) |
40cbfc37 | 1332 | { |
f7bc9594 | 1333 | struct scatterlist *sgs[4], hdr, stat; |
d24bae32 | 1334 | unsigned out_num = 0, tmp; |
40cbfc37 AK |
1335 | |
1336 | /* Caller should know better */ | |
f7bc9594 | 1337 | BUG_ON(!virtio_has_feature(vi->vdev, VIRTIO_NET_F_CTRL_VQ)); |
40cbfc37 | 1338 | |
2ac46030 MT |
1339 | vi->ctrl_status = ~0; |
1340 | vi->ctrl_hdr.class = class; | |
1341 | vi->ctrl_hdr.cmd = cmd; | |
f7bc9594 | 1342 | /* Add header */ |
2ac46030 | 1343 | sg_init_one(&hdr, &vi->ctrl_hdr, sizeof(vi->ctrl_hdr)); |
f7bc9594 | 1344 | sgs[out_num++] = &hdr; |
40cbfc37 | 1345 | |
f7bc9594 RR |
1346 | if (out) |
1347 | sgs[out_num++] = out; | |
40cbfc37 | 1348 | |
f7bc9594 | 1349 | /* Add return status. */ |
2ac46030 | 1350 | sg_init_one(&stat, &vi->ctrl_status, sizeof(vi->ctrl_status)); |
d24bae32 | 1351 | sgs[out_num] = &stat; |
40cbfc37 | 1352 | |
d24bae32 | 1353 | BUG_ON(out_num + 1 > ARRAY_SIZE(sgs)); |
a7c58146 | 1354 | virtqueue_add_sgs(vi->cvq, sgs, out_num, 1, vi, GFP_ATOMIC); |
40cbfc37 | 1355 | |
67975901 | 1356 | if (unlikely(!virtqueue_kick(vi->cvq))) |
2ac46030 | 1357 | return vi->ctrl_status == VIRTIO_NET_OK; |
40cbfc37 AK |
1358 | |
1359 | /* Spin for a response, the kick causes an ioport write, trapping | |
1360 | * into the hypervisor, so the request should be handled immediately. | |
1361 | */ | |
047b9b94 HG |
1362 | while (!virtqueue_get_buf(vi->cvq, &tmp) && |
1363 | !virtqueue_is_broken(vi->cvq)) | |
40cbfc37 AK |
1364 | cpu_relax(); |
1365 | ||
2ac46030 | 1366 | return vi->ctrl_status == VIRTIO_NET_OK; |
40cbfc37 AK |
1367 | } |
1368 | ||
9c46f6d4 AW |
1369 | static int virtnet_set_mac_address(struct net_device *dev, void *p) |
1370 | { | |
1371 | struct virtnet_info *vi = netdev_priv(dev); | |
1372 | struct virtio_device *vdev = vi->vdev; | |
f2f2c8b4 | 1373 | int ret; |
e37e2ff3 | 1374 | struct sockaddr *addr; |
7e58d5ae | 1375 | struct scatterlist sg; |
9c46f6d4 | 1376 | |
801822d1 | 1377 | addr = kmemdup(p, sizeof(*addr), GFP_KERNEL); |
e37e2ff3 AL |
1378 | if (!addr) |
1379 | return -ENOMEM; | |
e37e2ff3 AL |
1380 | |
1381 | ret = eth_prepare_mac_addr_change(dev, addr); | |
f2f2c8b4 | 1382 | if (ret) |
e37e2ff3 | 1383 | goto out; |
9c46f6d4 | 1384 | |
7e58d5ae AK |
1385 | if (virtio_has_feature(vdev, VIRTIO_NET_F_CTRL_MAC_ADDR)) { |
1386 | sg_init_one(&sg, addr->sa_data, dev->addr_len); | |
1387 | if (!virtnet_send_command(vi, VIRTIO_NET_CTRL_MAC, | |
d24bae32 | 1388 | VIRTIO_NET_CTRL_MAC_ADDR_SET, &sg)) { |
7e58d5ae AK |
1389 | dev_warn(&vdev->dev, |
1390 | "Failed to set mac address by vq command.\n"); | |
e37e2ff3 AL |
1391 | ret = -EINVAL; |
1392 | goto out; | |
7e58d5ae | 1393 | } |
7e93a02f MT |
1394 | } else if (virtio_has_feature(vdev, VIRTIO_NET_F_MAC) && |
1395 | !virtio_has_feature(vdev, VIRTIO_F_VERSION_1)) { | |
855e0c52 RR |
1396 | unsigned int i; |
1397 | ||
1398 | /* Naturally, this has an atomicity problem. */ | |
1399 | for (i = 0; i < dev->addr_len; i++) | |
1400 | virtio_cwrite8(vdev, | |
1401 | offsetof(struct virtio_net_config, mac) + | |
1402 | i, addr->sa_data[i]); | |
7e58d5ae AK |
1403 | } |
1404 | ||
1405 | eth_commit_mac_addr_change(dev, p); | |
e37e2ff3 | 1406 | ret = 0; |
9c46f6d4 | 1407 | |
e37e2ff3 AL |
1408 | out: |
1409 | kfree(addr); | |
1410 | return ret; | |
9c46f6d4 AW |
1411 | } |
1412 | ||
bc1f4470 | 1413 | static void virtnet_stats(struct net_device *dev, |
1414 | struct rtnl_link_stats64 *tot) | |
3fa2a1df | 1415 | { |
1416 | struct virtnet_info *vi = netdev_priv(dev); | |
1417 | int cpu; | |
1418 | unsigned int start; | |
1419 | ||
1420 | for_each_possible_cpu(cpu) { | |
58472a76 | 1421 | struct virtnet_stats *stats = per_cpu_ptr(vi->stats, cpu); |
3fa2a1df | 1422 | u64 tpackets, tbytes, rpackets, rbytes; |
1423 | ||
1424 | do { | |
57a7744e | 1425 | start = u64_stats_fetch_begin_irq(&stats->tx_syncp); |
3fa2a1df | 1426 | tpackets = stats->tx_packets; |
1427 | tbytes = stats->tx_bytes; | |
57a7744e | 1428 | } while (u64_stats_fetch_retry_irq(&stats->tx_syncp, start)); |
83a27052 ED |
1429 | |
1430 | do { | |
57a7744e | 1431 | start = u64_stats_fetch_begin_irq(&stats->rx_syncp); |
3fa2a1df | 1432 | rpackets = stats->rx_packets; |
1433 | rbytes = stats->rx_bytes; | |
57a7744e | 1434 | } while (u64_stats_fetch_retry_irq(&stats->rx_syncp, start)); |
3fa2a1df | 1435 | |
1436 | tot->rx_packets += rpackets; | |
1437 | tot->tx_packets += tpackets; | |
1438 | tot->rx_bytes += rbytes; | |
1439 | tot->tx_bytes += tbytes; | |
1440 | } | |
1441 | ||
1442 | tot->tx_dropped = dev->stats.tx_dropped; | |
021ac8d3 | 1443 | tot->tx_fifo_errors = dev->stats.tx_fifo_errors; |
3fa2a1df | 1444 | tot->rx_dropped = dev->stats.rx_dropped; |
1445 | tot->rx_length_errors = dev->stats.rx_length_errors; | |
1446 | tot->rx_frame_errors = dev->stats.rx_frame_errors; | |
3fa2a1df | 1447 | } |
1448 | ||
da74e89d AS |
1449 | #ifdef CONFIG_NET_POLL_CONTROLLER |
1450 | static void virtnet_netpoll(struct net_device *dev) | |
1451 | { | |
1452 | struct virtnet_info *vi = netdev_priv(dev); | |
986a4f4d | 1453 | int i; |
da74e89d | 1454 | |
986a4f4d JW |
1455 | for (i = 0; i < vi->curr_queue_pairs; i++) |
1456 | napi_schedule(&vi->rq[i].napi); | |
da74e89d AS |
1457 | } |
1458 | #endif | |
1459 | ||
586d17c5 JW |
1460 | static void virtnet_ack_link_announce(struct virtnet_info *vi) |
1461 | { | |
1462 | rtnl_lock(); | |
1463 | if (!virtnet_send_command(vi, VIRTIO_NET_CTRL_ANNOUNCE, | |
d24bae32 | 1464 | VIRTIO_NET_CTRL_ANNOUNCE_ACK, NULL)) |
586d17c5 JW |
1465 | dev_warn(&vi->dev->dev, "Failed to ack link announce.\n"); |
1466 | rtnl_unlock(); | |
1467 | } | |
1468 | ||
47315329 | 1469 | static int _virtnet_set_queues(struct virtnet_info *vi, u16 queue_pairs) |
986a4f4d JW |
1470 | { |
1471 | struct scatterlist sg; | |
986a4f4d JW |
1472 | struct net_device *dev = vi->dev; |
1473 | ||
1474 | if (!vi->has_cvq || !virtio_has_feature(vi->vdev, VIRTIO_NET_F_MQ)) | |
1475 | return 0; | |
1476 | ||
a725ee3e AL |
1477 | vi->ctrl_mq.virtqueue_pairs = cpu_to_virtio16(vi->vdev, queue_pairs); |
1478 | sg_init_one(&sg, &vi->ctrl_mq, sizeof(vi->ctrl_mq)); | |
986a4f4d JW |
1479 | |
1480 | if (!virtnet_send_command(vi, VIRTIO_NET_CTRL_MQ, | |
d24bae32 | 1481 | VIRTIO_NET_CTRL_MQ_VQ_PAIRS_SET, &sg)) { |
986a4f4d JW |
1482 | dev_warn(&dev->dev, "Fail to set num of queue pairs to %d\n", |
1483 | queue_pairs); | |
1484 | return -EINVAL; | |
55257d72 | 1485 | } else { |
986a4f4d | 1486 | vi->curr_queue_pairs = queue_pairs; |
35ed159b JW |
1487 | /* virtnet_open() will refill when device is going to up. */ |
1488 | if (dev->flags & IFF_UP) | |
1489 | schedule_delayed_work(&vi->refill, 0); | |
55257d72 | 1490 | } |
986a4f4d JW |
1491 | |
1492 | return 0; | |
1493 | } | |
1494 | ||
47315329 JF |
1495 | static int virtnet_set_queues(struct virtnet_info *vi, u16 queue_pairs) |
1496 | { | |
1497 | int err; | |
1498 | ||
1499 | rtnl_lock(); | |
1500 | err = _virtnet_set_queues(vi, queue_pairs); | |
1501 | rtnl_unlock(); | |
1502 | return err; | |
1503 | } | |
1504 | ||
296f96fc RR |
1505 | static int virtnet_close(struct net_device *dev) |
1506 | { | |
1507 | struct virtnet_info *vi = netdev_priv(dev); | |
986a4f4d | 1508 | int i; |
296f96fc | 1509 | |
b2baed69 RR |
1510 | /* Make sure refill_work doesn't re-enable napi! */ |
1511 | cancel_delayed_work_sync(&vi->refill); | |
986a4f4d | 1512 | |
b92f1e67 | 1513 | for (i = 0; i < vi->max_queue_pairs; i++) { |
986a4f4d | 1514 | napi_disable(&vi->rq[i].napi); |
78a57b48 | 1515 | virtnet_napi_tx_disable(&vi->sq[i].napi); |
b92f1e67 | 1516 | } |
296f96fc | 1517 | |
296f96fc RR |
1518 | return 0; |
1519 | } | |
1520 | ||
2af7698e AW |
1521 | static void virtnet_set_rx_mode(struct net_device *dev) |
1522 | { | |
1523 | struct virtnet_info *vi = netdev_priv(dev); | |
f565a7c2 | 1524 | struct scatterlist sg[2]; |
f565a7c2 | 1525 | struct virtio_net_ctrl_mac *mac_data; |
ccffad25 | 1526 | struct netdev_hw_addr *ha; |
32e7bfc4 | 1527 | int uc_count; |
4cd24eaf | 1528 | int mc_count; |
f565a7c2 AW |
1529 | void *buf; |
1530 | int i; | |
2af7698e | 1531 | |
788a8b6d | 1532 | /* We can't dynamically set ndo_set_rx_mode, so return gracefully */ |
2af7698e AW |
1533 | if (!virtio_has_feature(vi->vdev, VIRTIO_NET_F_CTRL_RX)) |
1534 | return; | |
1535 | ||
2ac46030 MT |
1536 | vi->ctrl_promisc = ((dev->flags & IFF_PROMISC) != 0); |
1537 | vi->ctrl_allmulti = ((dev->flags & IFF_ALLMULTI) != 0); | |
2af7698e | 1538 | |
2ac46030 | 1539 | sg_init_one(sg, &vi->ctrl_promisc, sizeof(vi->ctrl_promisc)); |
2af7698e AW |
1540 | |
1541 | if (!virtnet_send_command(vi, VIRTIO_NET_CTRL_RX, | |
d24bae32 | 1542 | VIRTIO_NET_CTRL_RX_PROMISC, sg)) |
2af7698e | 1543 | dev_warn(&dev->dev, "Failed to %sable promisc mode.\n", |
2ac46030 | 1544 | vi->ctrl_promisc ? "en" : "dis"); |
2af7698e | 1545 | |
2ac46030 | 1546 | sg_init_one(sg, &vi->ctrl_allmulti, sizeof(vi->ctrl_allmulti)); |
2af7698e AW |
1547 | |
1548 | if (!virtnet_send_command(vi, VIRTIO_NET_CTRL_RX, | |
d24bae32 | 1549 | VIRTIO_NET_CTRL_RX_ALLMULTI, sg)) |
2af7698e | 1550 | dev_warn(&dev->dev, "Failed to %sable allmulti mode.\n", |
2ac46030 | 1551 | vi->ctrl_allmulti ? "en" : "dis"); |
f565a7c2 | 1552 | |
32e7bfc4 | 1553 | uc_count = netdev_uc_count(dev); |
4cd24eaf | 1554 | mc_count = netdev_mc_count(dev); |
f565a7c2 | 1555 | /* MAC filter - use one buffer for both lists */ |
4cd24eaf JP |
1556 | buf = kzalloc(((uc_count + mc_count) * ETH_ALEN) + |
1557 | (2 * sizeof(mac_data->entries)), GFP_ATOMIC); | |
1558 | mac_data = buf; | |
e68ed8f0 | 1559 | if (!buf) |
f565a7c2 | 1560 | return; |
f565a7c2 | 1561 | |
23e258e1 AW |
1562 | sg_init_table(sg, 2); |
1563 | ||
f565a7c2 | 1564 | /* Store the unicast list and count in the front of the buffer */ |
fdd819b2 | 1565 | mac_data->entries = cpu_to_virtio32(vi->vdev, uc_count); |
ccffad25 | 1566 | i = 0; |
32e7bfc4 | 1567 | netdev_for_each_uc_addr(ha, dev) |
ccffad25 | 1568 | memcpy(&mac_data->macs[i++][0], ha->addr, ETH_ALEN); |
f565a7c2 AW |
1569 | |
1570 | sg_set_buf(&sg[0], mac_data, | |
32e7bfc4 | 1571 | sizeof(mac_data->entries) + (uc_count * ETH_ALEN)); |
f565a7c2 AW |
1572 | |
1573 | /* multicast list and count fill the end */ | |
32e7bfc4 | 1574 | mac_data = (void *)&mac_data->macs[uc_count][0]; |
f565a7c2 | 1575 | |
fdd819b2 | 1576 | mac_data->entries = cpu_to_virtio32(vi->vdev, mc_count); |
567ec874 | 1577 | i = 0; |
22bedad3 JP |
1578 | netdev_for_each_mc_addr(ha, dev) |
1579 | memcpy(&mac_data->macs[i++][0], ha->addr, ETH_ALEN); | |
f565a7c2 AW |
1580 | |
1581 | sg_set_buf(&sg[1], mac_data, | |
4cd24eaf | 1582 | sizeof(mac_data->entries) + (mc_count * ETH_ALEN)); |
f565a7c2 AW |
1583 | |
1584 | if (!virtnet_send_command(vi, VIRTIO_NET_CTRL_MAC, | |
d24bae32 | 1585 | VIRTIO_NET_CTRL_MAC_TABLE_SET, sg)) |
99e872ae | 1586 | dev_warn(&dev->dev, "Failed to set MAC filter table.\n"); |
f565a7c2 AW |
1587 | |
1588 | kfree(buf); | |
2af7698e AW |
1589 | } |
1590 | ||
80d5c368 PM |
1591 | static int virtnet_vlan_rx_add_vid(struct net_device *dev, |
1592 | __be16 proto, u16 vid) | |
0bde9569 AW |
1593 | { |
1594 | struct virtnet_info *vi = netdev_priv(dev); | |
1595 | struct scatterlist sg; | |
1596 | ||
a725ee3e AL |
1597 | vi->ctrl_vid = vid; |
1598 | sg_init_one(&sg, &vi->ctrl_vid, sizeof(vi->ctrl_vid)); | |
0bde9569 AW |
1599 | |
1600 | if (!virtnet_send_command(vi, VIRTIO_NET_CTRL_VLAN, | |
d24bae32 | 1601 | VIRTIO_NET_CTRL_VLAN_ADD, &sg)) |
0bde9569 | 1602 | dev_warn(&dev->dev, "Failed to add VLAN ID %d.\n", vid); |
8e586137 | 1603 | return 0; |
0bde9569 AW |
1604 | } |
1605 | ||
80d5c368 PM |
1606 | static int virtnet_vlan_rx_kill_vid(struct net_device *dev, |
1607 | __be16 proto, u16 vid) | |
0bde9569 AW |
1608 | { |
1609 | struct virtnet_info *vi = netdev_priv(dev); | |
1610 | struct scatterlist sg; | |
1611 | ||
a725ee3e AL |
1612 | vi->ctrl_vid = vid; |
1613 | sg_init_one(&sg, &vi->ctrl_vid, sizeof(vi->ctrl_vid)); | |
0bde9569 AW |
1614 | |
1615 | if (!virtnet_send_command(vi, VIRTIO_NET_CTRL_VLAN, | |
d24bae32 | 1616 | VIRTIO_NET_CTRL_VLAN_DEL, &sg)) |
0bde9569 | 1617 | dev_warn(&dev->dev, "Failed to kill VLAN ID %d.\n", vid); |
8e586137 | 1618 | return 0; |
0bde9569 AW |
1619 | } |
1620 | ||
8898c21c | 1621 | static void virtnet_clean_affinity(struct virtnet_info *vi, long hcpu) |
986a4f4d JW |
1622 | { |
1623 | int i; | |
1624 | ||
8898c21c WG |
1625 | if (vi->affinity_hint_set) { |
1626 | for (i = 0; i < vi->max_queue_pairs; i++) { | |
47be2479 WG |
1627 | virtqueue_set_affinity(vi->rq[i].vq, -1); |
1628 | virtqueue_set_affinity(vi->sq[i].vq, -1); | |
1629 | } | |
1630 | ||
8898c21c WG |
1631 | vi->affinity_hint_set = false; |
1632 | } | |
8898c21c | 1633 | } |
47be2479 | 1634 | |
8898c21c WG |
1635 | static void virtnet_set_affinity(struct virtnet_info *vi) |
1636 | { | |
1637 | int i; | |
1638 | int cpu; | |
986a4f4d JW |
1639 | |
1640 | /* In multiqueue mode, when the number of cpu is equal to the number of | |
1641 | * queue pairs, we let the queue pairs to be private to one cpu by | |
1642 | * setting the affinity hint to eliminate the contention. | |
1643 | */ | |
8898c21c WG |
1644 | if (vi->curr_queue_pairs == 1 || |
1645 | vi->max_queue_pairs != num_online_cpus()) { | |
1646 | virtnet_clean_affinity(vi, -1); | |
1647 | return; | |
986a4f4d JW |
1648 | } |
1649 | ||
8898c21c WG |
1650 | i = 0; |
1651 | for_each_online_cpu(cpu) { | |
986a4f4d JW |
1652 | virtqueue_set_affinity(vi->rq[i].vq, cpu); |
1653 | virtqueue_set_affinity(vi->sq[i].vq, cpu); | |
9bb8ca86 | 1654 | netif_set_xps_queue(vi->dev, cpumask_of(cpu), i); |
8898c21c | 1655 | i++; |
986a4f4d JW |
1656 | } |
1657 | ||
8898c21c | 1658 | vi->affinity_hint_set = true; |
986a4f4d JW |
1659 | } |
1660 | ||
8017c279 | 1661 | static int virtnet_cpu_online(unsigned int cpu, struct hlist_node *node) |
8de4b2f3 | 1662 | { |
8017c279 SAS |
1663 | struct virtnet_info *vi = hlist_entry_safe(node, struct virtnet_info, |
1664 | node); | |
1665 | virtnet_set_affinity(vi); | |
1666 | return 0; | |
1667 | } | |
8de4b2f3 | 1668 | |
8017c279 SAS |
1669 | static int virtnet_cpu_dead(unsigned int cpu, struct hlist_node *node) |
1670 | { | |
1671 | struct virtnet_info *vi = hlist_entry_safe(node, struct virtnet_info, | |
1672 | node_dead); | |
1673 | virtnet_set_affinity(vi); | |
1674 | return 0; | |
1675 | } | |
3ab098df | 1676 | |
8017c279 SAS |
1677 | static int virtnet_cpu_down_prep(unsigned int cpu, struct hlist_node *node) |
1678 | { | |
1679 | struct virtnet_info *vi = hlist_entry_safe(node, struct virtnet_info, | |
1680 | node); | |
1681 | ||
1682 | virtnet_clean_affinity(vi, cpu); | |
1683 | return 0; | |
1684 | } | |
1685 | ||
1686 | static enum cpuhp_state virtionet_online; | |
1687 | ||
1688 | static int virtnet_cpu_notif_add(struct virtnet_info *vi) | |
1689 | { | |
1690 | int ret; | |
1691 | ||
1692 | ret = cpuhp_state_add_instance_nocalls(virtionet_online, &vi->node); | |
1693 | if (ret) | |
1694 | return ret; | |
1695 | ret = cpuhp_state_add_instance_nocalls(CPUHP_VIRT_NET_DEAD, | |
1696 | &vi->node_dead); | |
1697 | if (!ret) | |
1698 | return ret; | |
1699 | cpuhp_state_remove_instance_nocalls(virtionet_online, &vi->node); | |
1700 | return ret; | |
1701 | } | |
1702 | ||
1703 | static void virtnet_cpu_notif_remove(struct virtnet_info *vi) | |
1704 | { | |
1705 | cpuhp_state_remove_instance_nocalls(virtionet_online, &vi->node); | |
1706 | cpuhp_state_remove_instance_nocalls(CPUHP_VIRT_NET_DEAD, | |
1707 | &vi->node_dead); | |
986a4f4d JW |
1708 | } |
1709 | ||
8f9f4668 RJ |
1710 | static void virtnet_get_ringparam(struct net_device *dev, |
1711 | struct ethtool_ringparam *ring) | |
1712 | { | |
1713 | struct virtnet_info *vi = netdev_priv(dev); | |
1714 | ||
986a4f4d JW |
1715 | ring->rx_max_pending = virtqueue_get_vring_size(vi->rq[0].vq); |
1716 | ring->tx_max_pending = virtqueue_get_vring_size(vi->sq[0].vq); | |
8f9f4668 RJ |
1717 | ring->rx_pending = ring->rx_max_pending; |
1718 | ring->tx_pending = ring->tx_max_pending; | |
8f9f4668 RJ |
1719 | } |
1720 | ||
66846048 RJ |
1721 | |
1722 | static void virtnet_get_drvinfo(struct net_device *dev, | |
1723 | struct ethtool_drvinfo *info) | |
1724 | { | |
1725 | struct virtnet_info *vi = netdev_priv(dev); | |
1726 | struct virtio_device *vdev = vi->vdev; | |
1727 | ||
1728 | strlcpy(info->driver, KBUILD_MODNAME, sizeof(info->driver)); | |
1729 | strlcpy(info->version, VIRTNET_DRIVER_VERSION, sizeof(info->version)); | |
1730 | strlcpy(info->bus_info, virtio_bus_name(vdev), sizeof(info->bus_info)); | |
1731 | ||
1732 | } | |
1733 | ||
d73bcd2c JW |
1734 | /* TODO: Eliminate OOO packets during switching */ |
1735 | static int virtnet_set_channels(struct net_device *dev, | |
1736 | struct ethtool_channels *channels) | |
1737 | { | |
1738 | struct virtnet_info *vi = netdev_priv(dev); | |
1739 | u16 queue_pairs = channels->combined_count; | |
1740 | int err; | |
1741 | ||
1742 | /* We don't support separate rx/tx channels. | |
1743 | * We don't allow setting 'other' channels. | |
1744 | */ | |
1745 | if (channels->rx_count || channels->tx_count || channels->other_count) | |
1746 | return -EINVAL; | |
1747 | ||
c18e9cd6 | 1748 | if (queue_pairs > vi->max_queue_pairs || queue_pairs == 0) |
d73bcd2c JW |
1749 | return -EINVAL; |
1750 | ||
f600b690 JF |
1751 | /* For now we don't support modifying channels while XDP is loaded |
1752 | * also when XDP is loaded all RX queues have XDP programs so we only | |
1753 | * need to check a single RX queue. | |
1754 | */ | |
1755 | if (vi->rq[0].xdp_prog) | |
1756 | return -EINVAL; | |
1757 | ||
47be2479 | 1758 | get_online_cpus(); |
47315329 | 1759 | err = _virtnet_set_queues(vi, queue_pairs); |
d73bcd2c JW |
1760 | if (!err) { |
1761 | netif_set_real_num_tx_queues(dev, queue_pairs); | |
1762 | netif_set_real_num_rx_queues(dev, queue_pairs); | |
1763 | ||
8898c21c | 1764 | virtnet_set_affinity(vi); |
d73bcd2c | 1765 | } |
47be2479 | 1766 | put_online_cpus(); |
d73bcd2c JW |
1767 | |
1768 | return err; | |
1769 | } | |
1770 | ||
1771 | static void virtnet_get_channels(struct net_device *dev, | |
1772 | struct ethtool_channels *channels) | |
1773 | { | |
1774 | struct virtnet_info *vi = netdev_priv(dev); | |
1775 | ||
1776 | channels->combined_count = vi->curr_queue_pairs; | |
1777 | channels->max_combined = vi->max_queue_pairs; | |
1778 | channels->max_other = 0; | |
1779 | channels->rx_count = 0; | |
1780 | channels->tx_count = 0; | |
1781 | channels->other_count = 0; | |
1782 | } | |
1783 | ||
16032be5 | 1784 | /* Check if the user is trying to change anything besides speed/duplex */ |
ebb6b4b1 PR |
1785 | static bool |
1786 | virtnet_validate_ethtool_cmd(const struct ethtool_link_ksettings *cmd) | |
16032be5 | 1787 | { |
ebb6b4b1 PR |
1788 | struct ethtool_link_ksettings diff1 = *cmd; |
1789 | struct ethtool_link_ksettings diff2 = {}; | |
16032be5 | 1790 | |
0cf3ace9 NA |
1791 | /* cmd is always set so we need to clear it, validate the port type |
1792 | * and also without autonegotiation we can ignore advertising | |
1793 | */ | |
ebb6b4b1 PR |
1794 | diff1.base.speed = 0; |
1795 | diff2.base.port = PORT_OTHER; | |
1796 | ethtool_link_ksettings_zero_link_mode(&diff1, advertising); | |
1797 | diff1.base.duplex = 0; | |
1798 | diff1.base.cmd = 0; | |
1799 | diff1.base.link_mode_masks_nwords = 0; | |
1800 | ||
1801 | return !memcmp(&diff1.base, &diff2.base, sizeof(diff1.base)) && | |
1802 | bitmap_empty(diff1.link_modes.supported, | |
1803 | __ETHTOOL_LINK_MODE_MASK_NBITS) && | |
1804 | bitmap_empty(diff1.link_modes.advertising, | |
1805 | __ETHTOOL_LINK_MODE_MASK_NBITS) && | |
1806 | bitmap_empty(diff1.link_modes.lp_advertising, | |
1807 | __ETHTOOL_LINK_MODE_MASK_NBITS); | |
16032be5 NA |
1808 | } |
1809 | ||
ebb6b4b1 PR |
1810 | static int virtnet_set_link_ksettings(struct net_device *dev, |
1811 | const struct ethtool_link_ksettings *cmd) | |
16032be5 NA |
1812 | { |
1813 | struct virtnet_info *vi = netdev_priv(dev); | |
1814 | u32 speed; | |
1815 | ||
ebb6b4b1 | 1816 | speed = cmd->base.speed; |
16032be5 NA |
1817 | /* don't allow custom speed and duplex */ |
1818 | if (!ethtool_validate_speed(speed) || | |
ebb6b4b1 | 1819 | !ethtool_validate_duplex(cmd->base.duplex) || |
16032be5 NA |
1820 | !virtnet_validate_ethtool_cmd(cmd)) |
1821 | return -EINVAL; | |
1822 | vi->speed = speed; | |
ebb6b4b1 | 1823 | vi->duplex = cmd->base.duplex; |
16032be5 NA |
1824 | |
1825 | return 0; | |
1826 | } | |
1827 | ||
ebb6b4b1 PR |
1828 | static int virtnet_get_link_ksettings(struct net_device *dev, |
1829 | struct ethtool_link_ksettings *cmd) | |
16032be5 NA |
1830 | { |
1831 | struct virtnet_info *vi = netdev_priv(dev); | |
1832 | ||
ebb6b4b1 PR |
1833 | cmd->base.speed = vi->speed; |
1834 | cmd->base.duplex = vi->duplex; | |
1835 | cmd->base.port = PORT_OTHER; | |
16032be5 NA |
1836 | |
1837 | return 0; | |
1838 | } | |
1839 | ||
1840 | static void virtnet_init_settings(struct net_device *dev) | |
1841 | { | |
1842 | struct virtnet_info *vi = netdev_priv(dev); | |
1843 | ||
1844 | vi->speed = SPEED_UNKNOWN; | |
1845 | vi->duplex = DUPLEX_UNKNOWN; | |
1846 | } | |
1847 | ||
0fc0b732 | 1848 | static const struct ethtool_ops virtnet_ethtool_ops = { |
66846048 | 1849 | .get_drvinfo = virtnet_get_drvinfo, |
9f4d26d0 | 1850 | .get_link = ethtool_op_get_link, |
8f9f4668 | 1851 | .get_ringparam = virtnet_get_ringparam, |
d73bcd2c JW |
1852 | .set_channels = virtnet_set_channels, |
1853 | .get_channels = virtnet_get_channels, | |
074c3582 | 1854 | .get_ts_info = ethtool_op_get_ts_info, |
ebb6b4b1 PR |
1855 | .get_link_ksettings = virtnet_get_link_ksettings, |
1856 | .set_link_ksettings = virtnet_set_link_ksettings, | |
a9ea3fc6 HX |
1857 | }; |
1858 | ||
9fe7bfce JF |
1859 | static void virtnet_freeze_down(struct virtio_device *vdev) |
1860 | { | |
1861 | struct virtnet_info *vi = vdev->priv; | |
1862 | int i; | |
1863 | ||
1864 | /* Make sure no work handler is accessing the device */ | |
1865 | flush_work(&vi->config_work); | |
1866 | ||
1867 | netif_device_detach(vi->dev); | |
713a98d9 | 1868 | netif_tx_disable(vi->dev); |
9fe7bfce JF |
1869 | cancel_delayed_work_sync(&vi->refill); |
1870 | ||
1871 | if (netif_running(vi->dev)) { | |
b92f1e67 | 1872 | for (i = 0; i < vi->max_queue_pairs; i++) { |
9fe7bfce | 1873 | napi_disable(&vi->rq[i].napi); |
78a57b48 | 1874 | virtnet_napi_tx_disable(&vi->sq[i].napi); |
b92f1e67 | 1875 | } |
9fe7bfce JF |
1876 | } |
1877 | } | |
1878 | ||
1879 | static int init_vqs(struct virtnet_info *vi); | |
1880 | ||
1881 | static int virtnet_restore_up(struct virtio_device *vdev) | |
1882 | { | |
1883 | struct virtnet_info *vi = vdev->priv; | |
1884 | int err, i; | |
1885 | ||
1886 | err = init_vqs(vi); | |
1887 | if (err) | |
1888 | return err; | |
1889 | ||
1890 | virtio_device_ready(vdev); | |
1891 | ||
1892 | if (netif_running(vi->dev)) { | |
1893 | for (i = 0; i < vi->curr_queue_pairs; i++) | |
1894 | if (!try_fill_recv(vi, &vi->rq[i], GFP_KERNEL)) | |
1895 | schedule_delayed_work(&vi->refill, 0); | |
1896 | ||
b92f1e67 | 1897 | for (i = 0; i < vi->max_queue_pairs; i++) { |
e4e8452a | 1898 | virtnet_napi_enable(vi->rq[i].vq, &vi->rq[i].napi); |
b92f1e67 WB |
1899 | virtnet_napi_tx_enable(vi, vi->sq[i].vq, |
1900 | &vi->sq[i].napi); | |
1901 | } | |
9fe7bfce JF |
1902 | } |
1903 | ||
1904 | netif_device_attach(vi->dev); | |
1905 | return err; | |
1906 | } | |
1907 | ||
3f93522f JW |
1908 | static int virtnet_set_guest_offloads(struct virtnet_info *vi, u64 offloads) |
1909 | { | |
1910 | struct scatterlist sg; | |
1911 | vi->ctrl_offloads = cpu_to_virtio64(vi->vdev, offloads); | |
1912 | ||
1913 | sg_init_one(&sg, &vi->ctrl_offloads, sizeof(vi->ctrl_offloads)); | |
1914 | ||
1915 | if (!virtnet_send_command(vi, VIRTIO_NET_CTRL_GUEST_OFFLOADS, | |
1916 | VIRTIO_NET_CTRL_GUEST_OFFLOADS_SET, &sg)) { | |
1917 | dev_warn(&vi->dev->dev, "Fail to set guest offload. \n"); | |
1918 | return -EINVAL; | |
1919 | } | |
1920 | ||
1921 | return 0; | |
1922 | } | |
1923 | ||
1924 | static int virtnet_clear_guest_offloads(struct virtnet_info *vi) | |
1925 | { | |
1926 | u64 offloads = 0; | |
1927 | ||
1928 | if (!vi->guest_offloads) | |
1929 | return 0; | |
1930 | ||
1931 | if (virtio_has_feature(vi->vdev, VIRTIO_NET_F_GUEST_CSUM)) | |
1932 | offloads = 1ULL << VIRTIO_NET_F_GUEST_CSUM; | |
1933 | ||
1934 | return virtnet_set_guest_offloads(vi, offloads); | |
1935 | } | |
1936 | ||
1937 | static int virtnet_restore_guest_offloads(struct virtnet_info *vi) | |
1938 | { | |
1939 | u64 offloads = vi->guest_offloads; | |
1940 | ||
1941 | if (!vi->guest_offloads) | |
1942 | return 0; | |
1943 | if (virtio_has_feature(vi->vdev, VIRTIO_NET_F_GUEST_CSUM)) | |
1944 | offloads |= 1ULL << VIRTIO_NET_F_GUEST_CSUM; | |
1945 | ||
1946 | return virtnet_set_guest_offloads(vi, offloads); | |
1947 | } | |
1948 | ||
9861ce03 JK |
1949 | static int virtnet_xdp_set(struct net_device *dev, struct bpf_prog *prog, |
1950 | struct netlink_ext_ack *extack) | |
f600b690 JF |
1951 | { |
1952 | unsigned long int max_sz = PAGE_SIZE - sizeof(struct padded_vnet_hdr); | |
1953 | struct virtnet_info *vi = netdev_priv(dev); | |
1954 | struct bpf_prog *old_prog; | |
017b29c3 | 1955 | u16 xdp_qp = 0, curr_qp; |
672aafd5 | 1956 | int i, err; |
f600b690 | 1957 | |
3f93522f JW |
1958 | if (!virtio_has_feature(vi->vdev, VIRTIO_NET_F_CTRL_GUEST_OFFLOADS) |
1959 | && (virtio_has_feature(vi->vdev, VIRTIO_NET_F_GUEST_TSO4) || | |
1960 | virtio_has_feature(vi->vdev, VIRTIO_NET_F_GUEST_TSO6) || | |
1961 | virtio_has_feature(vi->vdev, VIRTIO_NET_F_GUEST_ECN) || | |
1962 | virtio_has_feature(vi->vdev, VIRTIO_NET_F_GUEST_UFO))) { | |
4d463c4d | 1963 | NL_SET_ERR_MSG_MOD(extack, "Can't set XDP while host is implementing LRO, disable LRO first"); |
f600b690 JF |
1964 | return -EOPNOTSUPP; |
1965 | } | |
1966 | ||
1967 | if (vi->mergeable_rx_bufs && !vi->any_header_sg) { | |
4d463c4d | 1968 | NL_SET_ERR_MSG_MOD(extack, "XDP expects header/data in single page, any_header_sg required"); |
f600b690 JF |
1969 | return -EINVAL; |
1970 | } | |
1971 | ||
1972 | if (dev->mtu > max_sz) { | |
4d463c4d | 1973 | NL_SET_ERR_MSG_MOD(extack, "MTU too large to enable XDP"); |
f600b690 JF |
1974 | netdev_warn(dev, "XDP requires MTU less than %lu\n", max_sz); |
1975 | return -EINVAL; | |
1976 | } | |
1977 | ||
672aafd5 JF |
1978 | curr_qp = vi->curr_queue_pairs - vi->xdp_queue_pairs; |
1979 | if (prog) | |
1980 | xdp_qp = nr_cpu_ids; | |
1981 | ||
1982 | /* XDP requires extra queues for XDP_TX */ | |
1983 | if (curr_qp + xdp_qp > vi->max_queue_pairs) { | |
4d463c4d | 1984 | NL_SET_ERR_MSG_MOD(extack, "Too few free TX rings available"); |
672aafd5 JF |
1985 | netdev_warn(dev, "request %i queues but max is %i\n", |
1986 | curr_qp + xdp_qp, vi->max_queue_pairs); | |
1987 | return -ENOMEM; | |
1988 | } | |
1989 | ||
2de2f7f4 JF |
1990 | if (prog) { |
1991 | prog = bpf_prog_add(prog, vi->max_queue_pairs - 1); | |
1992 | if (IS_ERR(prog)) | |
1993 | return PTR_ERR(prog); | |
1994 | } | |
1995 | ||
4941d472 JW |
1996 | /* Make sure NAPI is not using any XDP TX queues for RX. */ |
1997 | for (i = 0; i < vi->max_queue_pairs; i++) | |
1998 | napi_disable(&vi->rq[i].napi); | |
f600b690 | 1999 | |
672aafd5 | 2000 | netif_set_real_num_rx_queues(dev, curr_qp + xdp_qp); |
4941d472 JW |
2001 | err = _virtnet_set_queues(vi, curr_qp + xdp_qp); |
2002 | if (err) | |
2003 | goto err; | |
2004 | vi->xdp_queue_pairs = xdp_qp; | |
672aafd5 | 2005 | |
f600b690 JF |
2006 | for (i = 0; i < vi->max_queue_pairs; i++) { |
2007 | old_prog = rtnl_dereference(vi->rq[i].xdp_prog); | |
2008 | rcu_assign_pointer(vi->rq[i].xdp_prog, prog); | |
3f93522f JW |
2009 | if (i == 0) { |
2010 | if (!old_prog) | |
2011 | virtnet_clear_guest_offloads(vi); | |
2012 | if (!prog) | |
2013 | virtnet_restore_guest_offloads(vi); | |
2014 | } | |
f600b690 JF |
2015 | if (old_prog) |
2016 | bpf_prog_put(old_prog); | |
4941d472 | 2017 | virtnet_napi_enable(vi->rq[i].vq, &vi->rq[i].napi); |
f600b690 JF |
2018 | } |
2019 | ||
2020 | return 0; | |
2de2f7f4 | 2021 | |
4941d472 JW |
2022 | err: |
2023 | for (i = 0; i < vi->max_queue_pairs; i++) | |
2024 | virtnet_napi_enable(vi->rq[i].vq, &vi->rq[i].napi); | |
2de2f7f4 JF |
2025 | if (prog) |
2026 | bpf_prog_sub(prog, vi->max_queue_pairs - 1); | |
2027 | return err; | |
f600b690 JF |
2028 | } |
2029 | ||
5b0e6629 | 2030 | static u32 virtnet_xdp_query(struct net_device *dev) |
f600b690 JF |
2031 | { |
2032 | struct virtnet_info *vi = netdev_priv(dev); | |
5b0e6629 | 2033 | const struct bpf_prog *xdp_prog; |
f600b690 JF |
2034 | int i; |
2035 | ||
2036 | for (i = 0; i < vi->max_queue_pairs; i++) { | |
5b0e6629 MKL |
2037 | xdp_prog = rtnl_dereference(vi->rq[i].xdp_prog); |
2038 | if (xdp_prog) | |
2039 | return xdp_prog->aux->id; | |
f600b690 | 2040 | } |
5b0e6629 | 2041 | return 0; |
f600b690 JF |
2042 | } |
2043 | ||
2044 | static int virtnet_xdp(struct net_device *dev, struct netdev_xdp *xdp) | |
2045 | { | |
2046 | switch (xdp->command) { | |
2047 | case XDP_SETUP_PROG: | |
9861ce03 | 2048 | return virtnet_xdp_set(dev, xdp->prog, xdp->extack); |
f600b690 | 2049 | case XDP_QUERY_PROG: |
5b0e6629 MKL |
2050 | xdp->prog_id = virtnet_xdp_query(dev); |
2051 | xdp->prog_attached = !!xdp->prog_id; | |
f600b690 JF |
2052 | return 0; |
2053 | default: | |
2054 | return -EINVAL; | |
2055 | } | |
2056 | } | |
2057 | ||
76288b4e SH |
2058 | static const struct net_device_ops virtnet_netdev = { |
2059 | .ndo_open = virtnet_open, | |
2060 | .ndo_stop = virtnet_close, | |
2061 | .ndo_start_xmit = start_xmit, | |
2062 | .ndo_validate_addr = eth_validate_addr, | |
9c46f6d4 | 2063 | .ndo_set_mac_address = virtnet_set_mac_address, |
2af7698e | 2064 | .ndo_set_rx_mode = virtnet_set_rx_mode, |
3fa2a1df | 2065 | .ndo_get_stats64 = virtnet_stats, |
1824a989 AW |
2066 | .ndo_vlan_rx_add_vid = virtnet_vlan_rx_add_vid, |
2067 | .ndo_vlan_rx_kill_vid = virtnet_vlan_rx_kill_vid, | |
76288b4e SH |
2068 | #ifdef CONFIG_NET_POLL_CONTROLLER |
2069 | .ndo_poll_controller = virtnet_netpoll, | |
91815639 | 2070 | #endif |
f600b690 | 2071 | .ndo_xdp = virtnet_xdp, |
2836b4f2 | 2072 | .ndo_features_check = passthru_features_check, |
76288b4e SH |
2073 | }; |
2074 | ||
586d17c5 | 2075 | static void virtnet_config_changed_work(struct work_struct *work) |
9f4d26d0 | 2076 | { |
586d17c5 JW |
2077 | struct virtnet_info *vi = |
2078 | container_of(work, struct virtnet_info, config_work); | |
9f4d26d0 MM |
2079 | u16 v; |
2080 | ||
855e0c52 RR |
2081 | if (virtio_cread_feature(vi->vdev, VIRTIO_NET_F_STATUS, |
2082 | struct virtio_net_config, status, &v) < 0) | |
507613bf | 2083 | return; |
586d17c5 JW |
2084 | |
2085 | if (v & VIRTIO_NET_S_ANNOUNCE) { | |
ee89bab1 | 2086 | netdev_notify_peers(vi->dev); |
586d17c5 JW |
2087 | virtnet_ack_link_announce(vi); |
2088 | } | |
9f4d26d0 MM |
2089 | |
2090 | /* Ignore unknown (future) status bits */ | |
2091 | v &= VIRTIO_NET_S_LINK_UP; | |
2092 | ||
2093 | if (vi->status == v) | |
507613bf | 2094 | return; |
9f4d26d0 MM |
2095 | |
2096 | vi->status = v; | |
2097 | ||
2098 | if (vi->status & VIRTIO_NET_S_LINK_UP) { | |
2099 | netif_carrier_on(vi->dev); | |
986a4f4d | 2100 | netif_tx_wake_all_queues(vi->dev); |
9f4d26d0 MM |
2101 | } else { |
2102 | netif_carrier_off(vi->dev); | |
986a4f4d | 2103 | netif_tx_stop_all_queues(vi->dev); |
9f4d26d0 MM |
2104 | } |
2105 | } | |
2106 | ||
2107 | static void virtnet_config_changed(struct virtio_device *vdev) | |
2108 | { | |
2109 | struct virtnet_info *vi = vdev->priv; | |
2110 | ||
3b07e9ca | 2111 | schedule_work(&vi->config_work); |
9f4d26d0 MM |
2112 | } |
2113 | ||
986a4f4d JW |
2114 | static void virtnet_free_queues(struct virtnet_info *vi) |
2115 | { | |
d4fb84ee AV |
2116 | int i; |
2117 | ||
ab3971b1 JW |
2118 | for (i = 0; i < vi->max_queue_pairs; i++) { |
2119 | napi_hash_del(&vi->rq[i].napi); | |
d4fb84ee | 2120 | netif_napi_del(&vi->rq[i].napi); |
b92f1e67 | 2121 | netif_napi_del(&vi->sq[i].napi); |
ab3971b1 | 2122 | } |
d4fb84ee | 2123 | |
963abe5c ED |
2124 | /* We called napi_hash_del() before netif_napi_del(), |
2125 | * we need to respect an RCU grace period before freeing vi->rq | |
2126 | */ | |
2127 | synchronize_net(); | |
2128 | ||
986a4f4d JW |
2129 | kfree(vi->rq); |
2130 | kfree(vi->sq); | |
2131 | } | |
2132 | ||
47315329 | 2133 | static void _free_receive_bufs(struct virtnet_info *vi) |
986a4f4d | 2134 | { |
f600b690 | 2135 | struct bpf_prog *old_prog; |
986a4f4d JW |
2136 | int i; |
2137 | ||
2138 | for (i = 0; i < vi->max_queue_pairs; i++) { | |
2139 | while (vi->rq[i].pages) | |
2140 | __free_pages(get_a_page(&vi->rq[i], GFP_KERNEL), 0); | |
f600b690 JF |
2141 | |
2142 | old_prog = rtnl_dereference(vi->rq[i].xdp_prog); | |
2143 | RCU_INIT_POINTER(vi->rq[i].xdp_prog, NULL); | |
2144 | if (old_prog) | |
2145 | bpf_prog_put(old_prog); | |
986a4f4d | 2146 | } |
47315329 JF |
2147 | } |
2148 | ||
2149 | static void free_receive_bufs(struct virtnet_info *vi) | |
2150 | { | |
2151 | rtnl_lock(); | |
2152 | _free_receive_bufs(vi); | |
f600b690 | 2153 | rtnl_unlock(); |
986a4f4d JW |
2154 | } |
2155 | ||
fb51879d MD |
2156 | static void free_receive_page_frags(struct virtnet_info *vi) |
2157 | { | |
2158 | int i; | |
2159 | for (i = 0; i < vi->max_queue_pairs; i++) | |
2160 | if (vi->rq[i].alloc_frag.page) | |
2161 | put_page(vi->rq[i].alloc_frag.page); | |
2162 | } | |
2163 | ||
b68df015 | 2164 | static bool is_xdp_raw_buffer_queue(struct virtnet_info *vi, int q) |
56434a01 JF |
2165 | { |
2166 | if (q < (vi->curr_queue_pairs - vi->xdp_queue_pairs)) | |
2167 | return false; | |
2168 | else if (q < vi->curr_queue_pairs) | |
2169 | return true; | |
2170 | else | |
2171 | return false; | |
2172 | } | |
2173 | ||
986a4f4d JW |
2174 | static void free_unused_bufs(struct virtnet_info *vi) |
2175 | { | |
2176 | void *buf; | |
2177 | int i; | |
2178 | ||
2179 | for (i = 0; i < vi->max_queue_pairs; i++) { | |
2180 | struct virtqueue *vq = vi->sq[i].vq; | |
56434a01 | 2181 | while ((buf = virtqueue_detach_unused_buf(vq)) != NULL) { |
b68df015 | 2182 | if (!is_xdp_raw_buffer_queue(vi, i)) |
56434a01 JF |
2183 | dev_kfree_skb(buf); |
2184 | else | |
2185 | put_page(virt_to_head_page(buf)); | |
2186 | } | |
986a4f4d JW |
2187 | } |
2188 | ||
2189 | for (i = 0; i < vi->max_queue_pairs; i++) { | |
2190 | struct virtqueue *vq = vi->rq[i].vq; | |
2191 | ||
2192 | while ((buf = virtqueue_detach_unused_buf(vq)) != NULL) { | |
ab7db917 | 2193 | if (vi->mergeable_rx_bufs) { |
680557cf | 2194 | put_page(virt_to_head_page(buf)); |
ab7db917 | 2195 | } else if (vi->big_packets) { |
fa9fac17 | 2196 | give_pages(&vi->rq[i], buf); |
ab7db917 | 2197 | } else { |
f6b10209 | 2198 | put_page(virt_to_head_page(buf)); |
ab7db917 | 2199 | } |
986a4f4d | 2200 | } |
986a4f4d JW |
2201 | } |
2202 | } | |
2203 | ||
e9d7417b JW |
2204 | static void virtnet_del_vqs(struct virtnet_info *vi) |
2205 | { | |
2206 | struct virtio_device *vdev = vi->vdev; | |
2207 | ||
8898c21c | 2208 | virtnet_clean_affinity(vi, -1); |
986a4f4d | 2209 | |
e9d7417b | 2210 | vdev->config->del_vqs(vdev); |
986a4f4d JW |
2211 | |
2212 | virtnet_free_queues(vi); | |
e9d7417b JW |
2213 | } |
2214 | ||
d85b758f MT |
2215 | /* How large should a single buffer be so a queue full of these can fit at |
2216 | * least one full packet? | |
2217 | * Logic below assumes the mergeable buffer header is used. | |
2218 | */ | |
2219 | static unsigned int mergeable_min_buf_len(struct virtnet_info *vi, struct virtqueue *vq) | |
2220 | { | |
2221 | const unsigned int hdr_len = sizeof(struct virtio_net_hdr_mrg_rxbuf); | |
2222 | unsigned int rq_size = virtqueue_get_vring_size(vq); | |
2223 | unsigned int packet_len = vi->big_packets ? IP_MAX_MTU : vi->dev->max_mtu; | |
2224 | unsigned int buf_len = hdr_len + ETH_HLEN + VLAN_HLEN + packet_len; | |
2225 | unsigned int min_buf_len = DIV_ROUND_UP(buf_len, rq_size); | |
2226 | ||
f0c3192c MT |
2227 | return max(max(min_buf_len, hdr_len) - hdr_len, |
2228 | (unsigned int)GOOD_PACKET_LEN); | |
d85b758f MT |
2229 | } |
2230 | ||
986a4f4d | 2231 | static int virtnet_find_vqs(struct virtnet_info *vi) |
3f9c10b0 | 2232 | { |
986a4f4d JW |
2233 | vq_callback_t **callbacks; |
2234 | struct virtqueue **vqs; | |
2235 | int ret = -ENOMEM; | |
2236 | int i, total_vqs; | |
2237 | const char **names; | |
d45b897b | 2238 | bool *ctx; |
986a4f4d JW |
2239 | |
2240 | /* We expect 1 RX virtqueue followed by 1 TX virtqueue, followed by | |
2241 | * possible N-1 RX/TX queue pairs used in multiqueue mode, followed by | |
2242 | * possible control vq. | |
2243 | */ | |
2244 | total_vqs = vi->max_queue_pairs * 2 + | |
2245 | virtio_has_feature(vi->vdev, VIRTIO_NET_F_CTRL_VQ); | |
2246 | ||
2247 | /* Allocate space for find_vqs parameters */ | |
2248 | vqs = kzalloc(total_vqs * sizeof(*vqs), GFP_KERNEL); | |
2249 | if (!vqs) | |
2250 | goto err_vq; | |
2251 | callbacks = kmalloc(total_vqs * sizeof(*callbacks), GFP_KERNEL); | |
2252 | if (!callbacks) | |
2253 | goto err_callback; | |
2254 | names = kmalloc(total_vqs * sizeof(*names), GFP_KERNEL); | |
2255 | if (!names) | |
2256 | goto err_names; | |
192f68cf | 2257 | if (!vi->big_packets || vi->mergeable_rx_bufs) { |
d45b897b MT |
2258 | ctx = kzalloc(total_vqs * sizeof(*ctx), GFP_KERNEL); |
2259 | if (!ctx) | |
2260 | goto err_ctx; | |
2261 | } else { | |
2262 | ctx = NULL; | |
2263 | } | |
986a4f4d JW |
2264 | |
2265 | /* Parameters for control virtqueue, if any */ | |
2266 | if (vi->has_cvq) { | |
2267 | callbacks[total_vqs - 1] = NULL; | |
2268 | names[total_vqs - 1] = "control"; | |
2269 | } | |
3f9c10b0 | 2270 | |
986a4f4d JW |
2271 | /* Allocate/initialize parameters for send/receive virtqueues */ |
2272 | for (i = 0; i < vi->max_queue_pairs; i++) { | |
2273 | callbacks[rxq2vq(i)] = skb_recv_done; | |
2274 | callbacks[txq2vq(i)] = skb_xmit_done; | |
2275 | sprintf(vi->rq[i].name, "input.%d", i); | |
2276 | sprintf(vi->sq[i].name, "output.%d", i); | |
2277 | names[rxq2vq(i)] = vi->rq[i].name; | |
2278 | names[txq2vq(i)] = vi->sq[i].name; | |
d45b897b MT |
2279 | if (ctx) |
2280 | ctx[rxq2vq(i)] = true; | |
986a4f4d | 2281 | } |
3f9c10b0 | 2282 | |
986a4f4d | 2283 | ret = vi->vdev->config->find_vqs(vi->vdev, total_vqs, vqs, callbacks, |
d45b897b | 2284 | names, ctx, NULL); |
986a4f4d JW |
2285 | if (ret) |
2286 | goto err_find; | |
3f9c10b0 | 2287 | |
986a4f4d JW |
2288 | if (vi->has_cvq) { |
2289 | vi->cvq = vqs[total_vqs - 1]; | |
3f9c10b0 | 2290 | if (virtio_has_feature(vi->vdev, VIRTIO_NET_F_CTRL_VLAN)) |
f646968f | 2291 | vi->dev->features |= NETIF_F_HW_VLAN_CTAG_FILTER; |
3f9c10b0 | 2292 | } |
986a4f4d JW |
2293 | |
2294 | for (i = 0; i < vi->max_queue_pairs; i++) { | |
2295 | vi->rq[i].vq = vqs[rxq2vq(i)]; | |
d85b758f | 2296 | vi->rq[i].min_buf_len = mergeable_min_buf_len(vi, vi->rq[i].vq); |
986a4f4d JW |
2297 | vi->sq[i].vq = vqs[txq2vq(i)]; |
2298 | } | |
2299 | ||
2300 | kfree(names); | |
2301 | kfree(callbacks); | |
2302 | kfree(vqs); | |
55281621 | 2303 | kfree(ctx); |
986a4f4d | 2304 | |
3f9c10b0 | 2305 | return 0; |
986a4f4d JW |
2306 | |
2307 | err_find: | |
d45b897b MT |
2308 | kfree(ctx); |
2309 | err_ctx: | |
986a4f4d JW |
2310 | kfree(names); |
2311 | err_names: | |
2312 | kfree(callbacks); | |
2313 | err_callback: | |
2314 | kfree(vqs); | |
2315 | err_vq: | |
2316 | return ret; | |
2317 | } | |
2318 | ||
2319 | static int virtnet_alloc_queues(struct virtnet_info *vi) | |
2320 | { | |
2321 | int i; | |
2322 | ||
2323 | vi->sq = kzalloc(sizeof(*vi->sq) * vi->max_queue_pairs, GFP_KERNEL); | |
2324 | if (!vi->sq) | |
2325 | goto err_sq; | |
2326 | vi->rq = kzalloc(sizeof(*vi->rq) * vi->max_queue_pairs, GFP_KERNEL); | |
008d4278 | 2327 | if (!vi->rq) |
986a4f4d JW |
2328 | goto err_rq; |
2329 | ||
2330 | INIT_DELAYED_WORK(&vi->refill, refill_work); | |
2331 | for (i = 0; i < vi->max_queue_pairs; i++) { | |
2332 | vi->rq[i].pages = NULL; | |
2333 | netif_napi_add(vi->dev, &vi->rq[i].napi, virtnet_poll, | |
2334 | napi_weight); | |
1d11e732 WB |
2335 | netif_tx_napi_add(vi->dev, &vi->sq[i].napi, virtnet_poll_tx, |
2336 | napi_tx ? napi_weight : 0); | |
986a4f4d JW |
2337 | |
2338 | sg_init_table(vi->rq[i].sg, ARRAY_SIZE(vi->rq[i].sg)); | |
5377d758 | 2339 | ewma_pkt_len_init(&vi->rq[i].mrg_avg_pkt_len); |
986a4f4d JW |
2340 | sg_init_table(vi->sq[i].sg, ARRAY_SIZE(vi->sq[i].sg)); |
2341 | } | |
2342 | ||
2343 | return 0; | |
2344 | ||
2345 | err_rq: | |
2346 | kfree(vi->sq); | |
2347 | err_sq: | |
2348 | return -ENOMEM; | |
2349 | } | |
2350 | ||
2351 | static int init_vqs(struct virtnet_info *vi) | |
2352 | { | |
2353 | int ret; | |
2354 | ||
2355 | /* Allocate send & receive queues */ | |
2356 | ret = virtnet_alloc_queues(vi); | |
2357 | if (ret) | |
2358 | goto err; | |
2359 | ||
2360 | ret = virtnet_find_vqs(vi); | |
2361 | if (ret) | |
2362 | goto err_free; | |
2363 | ||
47be2479 | 2364 | get_online_cpus(); |
8898c21c | 2365 | virtnet_set_affinity(vi); |
47be2479 WG |
2366 | put_online_cpus(); |
2367 | ||
986a4f4d JW |
2368 | return 0; |
2369 | ||
2370 | err_free: | |
2371 | virtnet_free_queues(vi); | |
2372 | err: | |
2373 | return ret; | |
3f9c10b0 AS |
2374 | } |
2375 | ||
fbf28d78 MD |
2376 | #ifdef CONFIG_SYSFS |
2377 | static ssize_t mergeable_rx_buffer_size_show(struct netdev_rx_queue *queue, | |
718ad681 | 2378 | char *buf) |
fbf28d78 MD |
2379 | { |
2380 | struct virtnet_info *vi = netdev_priv(queue->dev); | |
2381 | unsigned int queue_index = get_netdev_rx_queue_index(queue); | |
5377d758 | 2382 | struct ewma_pkt_len *avg; |
fbf28d78 MD |
2383 | |
2384 | BUG_ON(queue_index >= vi->max_queue_pairs); | |
2385 | avg = &vi->rq[queue_index].mrg_avg_pkt_len; | |
d85b758f MT |
2386 | return sprintf(buf, "%u\n", |
2387 | get_mergeable_buf_len(&vi->rq[queue_index], avg)); | |
fbf28d78 MD |
2388 | } |
2389 | ||
2390 | static struct rx_queue_attribute mergeable_rx_buffer_size_attribute = | |
2391 | __ATTR_RO(mergeable_rx_buffer_size); | |
2392 | ||
2393 | static struct attribute *virtio_net_mrg_rx_attrs[] = { | |
2394 | &mergeable_rx_buffer_size_attribute.attr, | |
2395 | NULL | |
2396 | }; | |
2397 | ||
2398 | static const struct attribute_group virtio_net_mrg_rx_group = { | |
2399 | .name = "virtio_net", | |
2400 | .attrs = virtio_net_mrg_rx_attrs | |
2401 | }; | |
2402 | #endif | |
2403 | ||
892d6eb1 JW |
2404 | static bool virtnet_fail_on_feature(struct virtio_device *vdev, |
2405 | unsigned int fbit, | |
2406 | const char *fname, const char *dname) | |
2407 | { | |
2408 | if (!virtio_has_feature(vdev, fbit)) | |
2409 | return false; | |
2410 | ||
2411 | dev_err(&vdev->dev, "device advertises feature %s but not %s", | |
2412 | fname, dname); | |
2413 | ||
2414 | return true; | |
2415 | } | |
2416 | ||
2417 | #define VIRTNET_FAIL_ON(vdev, fbit, dbit) \ | |
2418 | virtnet_fail_on_feature(vdev, fbit, #fbit, dbit) | |
2419 | ||
2420 | static bool virtnet_validate_features(struct virtio_device *vdev) | |
2421 | { | |
2422 | if (!virtio_has_feature(vdev, VIRTIO_NET_F_CTRL_VQ) && | |
2423 | (VIRTNET_FAIL_ON(vdev, VIRTIO_NET_F_CTRL_RX, | |
2424 | "VIRTIO_NET_F_CTRL_VQ") || | |
2425 | VIRTNET_FAIL_ON(vdev, VIRTIO_NET_F_CTRL_VLAN, | |
2426 | "VIRTIO_NET_F_CTRL_VQ") || | |
2427 | VIRTNET_FAIL_ON(vdev, VIRTIO_NET_F_GUEST_ANNOUNCE, | |
2428 | "VIRTIO_NET_F_CTRL_VQ") || | |
2429 | VIRTNET_FAIL_ON(vdev, VIRTIO_NET_F_MQ, "VIRTIO_NET_F_CTRL_VQ") || | |
2430 | VIRTNET_FAIL_ON(vdev, VIRTIO_NET_F_CTRL_MAC_ADDR, | |
2431 | "VIRTIO_NET_F_CTRL_VQ"))) { | |
2432 | return false; | |
2433 | } | |
2434 | ||
2435 | return true; | |
2436 | } | |
2437 | ||
d0c2c997 JW |
2438 | #define MIN_MTU ETH_MIN_MTU |
2439 | #define MAX_MTU ETH_MAX_MTU | |
2440 | ||
fe36cbe0 | 2441 | static int virtnet_validate(struct virtio_device *vdev) |
296f96fc | 2442 | { |
6ba42248 MT |
2443 | if (!vdev->config->get) { |
2444 | dev_err(&vdev->dev, "%s failure: config access disabled\n", | |
2445 | __func__); | |
2446 | return -EINVAL; | |
2447 | } | |
2448 | ||
892d6eb1 JW |
2449 | if (!virtnet_validate_features(vdev)) |
2450 | return -EINVAL; | |
2451 | ||
fe36cbe0 MT |
2452 | if (virtio_has_feature(vdev, VIRTIO_NET_F_MTU)) { |
2453 | int mtu = virtio_cread16(vdev, | |
2454 | offsetof(struct virtio_net_config, | |
2455 | mtu)); | |
2456 | if (mtu < MIN_MTU) | |
2457 | __virtio_clear_bit(vdev, VIRTIO_NET_F_MTU); | |
2458 | } | |
2459 | ||
2460 | return 0; | |
2461 | } | |
2462 | ||
2463 | static int virtnet_probe(struct virtio_device *vdev) | |
2464 | { | |
2465 | int i, err; | |
2466 | struct net_device *dev; | |
2467 | struct virtnet_info *vi; | |
2468 | u16 max_queue_pairs; | |
2469 | int mtu; | |
2470 | ||
986a4f4d | 2471 | /* Find if host supports multiqueue virtio_net device */ |
855e0c52 RR |
2472 | err = virtio_cread_feature(vdev, VIRTIO_NET_F_MQ, |
2473 | struct virtio_net_config, | |
2474 | max_virtqueue_pairs, &max_queue_pairs); | |
986a4f4d JW |
2475 | |
2476 | /* We need at least 2 queue's */ | |
2477 | if (err || max_queue_pairs < VIRTIO_NET_CTRL_MQ_VQ_PAIRS_MIN || | |
2478 | max_queue_pairs > VIRTIO_NET_CTRL_MQ_VQ_PAIRS_MAX || | |
2479 | !virtio_has_feature(vdev, VIRTIO_NET_F_CTRL_VQ)) | |
2480 | max_queue_pairs = 1; | |
296f96fc RR |
2481 | |
2482 | /* Allocate ourselves a network device with room for our info */ | |
986a4f4d | 2483 | dev = alloc_etherdev_mq(sizeof(struct virtnet_info), max_queue_pairs); |
296f96fc RR |
2484 | if (!dev) |
2485 | return -ENOMEM; | |
2486 | ||
2487 | /* Set up network device as normal. */ | |
f2f2c8b4 | 2488 | dev->priv_flags |= IFF_UNICAST_FLT | IFF_LIVE_ADDR_CHANGE; |
76288b4e | 2489 | dev->netdev_ops = &virtnet_netdev; |
296f96fc | 2490 | dev->features = NETIF_F_HIGHDMA; |
3fa2a1df | 2491 | |
7ad24ea4 | 2492 | dev->ethtool_ops = &virtnet_ethtool_ops; |
296f96fc RR |
2493 | SET_NETDEV_DEV(dev, &vdev->dev); |
2494 | ||
2495 | /* Do we support "hardware" checksums? */ | |
98e778c9 | 2496 | if (virtio_has_feature(vdev, VIRTIO_NET_F_CSUM)) { |
296f96fc | 2497 | /* This opens up the world of extra features. */ |
48900cb6 | 2498 | dev->hw_features |= NETIF_F_HW_CSUM | NETIF_F_SG; |
98e778c9 | 2499 | if (csum) |
48900cb6 | 2500 | dev->features |= NETIF_F_HW_CSUM | NETIF_F_SG; |
98e778c9 MM |
2501 | |
2502 | if (virtio_has_feature(vdev, VIRTIO_NET_F_GSO)) { | |
e078de03 | 2503 | dev->hw_features |= NETIF_F_TSO |
34a48579 RR |
2504 | | NETIF_F_TSO_ECN | NETIF_F_TSO6; |
2505 | } | |
5539ae96 | 2506 | /* Individual feature bits: what can host handle? */ |
98e778c9 MM |
2507 | if (virtio_has_feature(vdev, VIRTIO_NET_F_HOST_TSO4)) |
2508 | dev->hw_features |= NETIF_F_TSO; | |
2509 | if (virtio_has_feature(vdev, VIRTIO_NET_F_HOST_TSO6)) | |
2510 | dev->hw_features |= NETIF_F_TSO6; | |
2511 | if (virtio_has_feature(vdev, VIRTIO_NET_F_HOST_ECN)) | |
2512 | dev->hw_features |= NETIF_F_TSO_ECN; | |
98e778c9 | 2513 | |
41f2f127 JW |
2514 | dev->features |= NETIF_F_GSO_ROBUST; |
2515 | ||
98e778c9 | 2516 | if (gso) |
e078de03 | 2517 | dev->features |= dev->hw_features & NETIF_F_ALL_TSO; |
98e778c9 | 2518 | /* (!csum && gso) case will be fixed by register_netdev() */ |
296f96fc | 2519 | } |
4f49129b TH |
2520 | if (virtio_has_feature(vdev, VIRTIO_NET_F_GUEST_CSUM)) |
2521 | dev->features |= NETIF_F_RXCSUM; | |
296f96fc | 2522 | |
4fda8302 JW |
2523 | dev->vlan_features = dev->features; |
2524 | ||
d0c2c997 JW |
2525 | /* MTU range: 68 - 65535 */ |
2526 | dev->min_mtu = MIN_MTU; | |
2527 | dev->max_mtu = MAX_MTU; | |
2528 | ||
296f96fc | 2529 | /* Configuration may specify what MAC to use. Otherwise random. */ |
855e0c52 RR |
2530 | if (virtio_has_feature(vdev, VIRTIO_NET_F_MAC)) |
2531 | virtio_cread_bytes(vdev, | |
2532 | offsetof(struct virtio_net_config, mac), | |
2533 | dev->dev_addr, dev->addr_len); | |
2534 | else | |
f2cedb63 | 2535 | eth_hw_addr_random(dev); |
296f96fc RR |
2536 | |
2537 | /* Set up our device-specific information */ | |
2538 | vi = netdev_priv(dev); | |
296f96fc RR |
2539 | vi->dev = dev; |
2540 | vi->vdev = vdev; | |
d9d5dcc8 | 2541 | vdev->priv = vi; |
3fa2a1df | 2542 | vi->stats = alloc_percpu(struct virtnet_stats); |
2543 | err = -ENOMEM; | |
2544 | if (vi->stats == NULL) | |
2545 | goto free; | |
2546 | ||
827da44c JS |
2547 | for_each_possible_cpu(i) { |
2548 | struct virtnet_stats *virtnet_stats; | |
2549 | virtnet_stats = per_cpu_ptr(vi->stats, i); | |
2550 | u64_stats_init(&virtnet_stats->tx_syncp); | |
2551 | u64_stats_init(&virtnet_stats->rx_syncp); | |
2552 | } | |
2553 | ||
586d17c5 | 2554 | INIT_WORK(&vi->config_work, virtnet_config_changed_work); |
296f96fc | 2555 | |
97402b96 | 2556 | /* If we can receive ANY GSO packets, we must allocate large ones. */ |
8e95a202 JP |
2557 | if (virtio_has_feature(vdev, VIRTIO_NET_F_GUEST_TSO4) || |
2558 | virtio_has_feature(vdev, VIRTIO_NET_F_GUEST_TSO6) || | |
e3e3c423 VY |
2559 | virtio_has_feature(vdev, VIRTIO_NET_F_GUEST_ECN) || |
2560 | virtio_has_feature(vdev, VIRTIO_NET_F_GUEST_UFO)) | |
97402b96 HX |
2561 | vi->big_packets = true; |
2562 | ||
3f2c31d9 MM |
2563 | if (virtio_has_feature(vdev, VIRTIO_NET_F_MRG_RXBUF)) |
2564 | vi->mergeable_rx_bufs = true; | |
2565 | ||
d04302b3 MT |
2566 | if (virtio_has_feature(vdev, VIRTIO_NET_F_MRG_RXBUF) || |
2567 | virtio_has_feature(vdev, VIRTIO_F_VERSION_1)) | |
012873d0 MT |
2568 | vi->hdr_len = sizeof(struct virtio_net_hdr_mrg_rxbuf); |
2569 | else | |
2570 | vi->hdr_len = sizeof(struct virtio_net_hdr); | |
2571 | ||
75993300 MT |
2572 | if (virtio_has_feature(vdev, VIRTIO_F_ANY_LAYOUT) || |
2573 | virtio_has_feature(vdev, VIRTIO_F_VERSION_1)) | |
e7428e95 MT |
2574 | vi->any_header_sg = true; |
2575 | ||
986a4f4d JW |
2576 | if (virtio_has_feature(vdev, VIRTIO_NET_F_CTRL_VQ)) |
2577 | vi->has_cvq = true; | |
2578 | ||
14de9d11 AC |
2579 | if (virtio_has_feature(vdev, VIRTIO_NET_F_MTU)) { |
2580 | mtu = virtio_cread16(vdev, | |
2581 | offsetof(struct virtio_net_config, | |
2582 | mtu)); | |
93a205ee | 2583 | if (mtu < dev->min_mtu) { |
fe36cbe0 MT |
2584 | /* Should never trigger: MTU was previously validated |
2585 | * in virtnet_validate. | |
2586 | */ | |
2587 | dev_err(&vdev->dev, "device MTU appears to have changed " | |
2588 | "it is now %d < %d", mtu, dev->min_mtu); | |
2589 | goto free_stats; | |
93a205ee | 2590 | } |
2e123b44 | 2591 | |
fe36cbe0 MT |
2592 | dev->mtu = mtu; |
2593 | dev->max_mtu = mtu; | |
2594 | ||
2e123b44 MT |
2595 | /* TODO: size buffers correctly in this case. */ |
2596 | if (dev->mtu > ETH_DATA_LEN) | |
2597 | vi->big_packets = true; | |
14de9d11 AC |
2598 | } |
2599 | ||
012873d0 MT |
2600 | if (vi->any_header_sg) |
2601 | dev->needed_headroom = vi->hdr_len; | |
6ebbc1a6 | 2602 | |
44900010 JW |
2603 | /* Enable multiqueue by default */ |
2604 | if (num_online_cpus() >= max_queue_pairs) | |
2605 | vi->curr_queue_pairs = max_queue_pairs; | |
2606 | else | |
2607 | vi->curr_queue_pairs = num_online_cpus(); | |
986a4f4d JW |
2608 | vi->max_queue_pairs = max_queue_pairs; |
2609 | ||
2610 | /* Allocate/initialize the rx/tx queues, and invoke find_vqs */ | |
3f9c10b0 | 2611 | err = init_vqs(vi); |
d2a7ddda | 2612 | if (err) |
9bb8ca86 | 2613 | goto free_stats; |
296f96fc | 2614 | |
fbf28d78 MD |
2615 | #ifdef CONFIG_SYSFS |
2616 | if (vi->mergeable_rx_bufs) | |
2617 | dev->sysfs_rx_queue_group = &virtio_net_mrg_rx_group; | |
2618 | #endif | |
0f13b66b ZYW |
2619 | netif_set_real_num_tx_queues(dev, vi->curr_queue_pairs); |
2620 | netif_set_real_num_rx_queues(dev, vi->curr_queue_pairs); | |
986a4f4d | 2621 | |
16032be5 NA |
2622 | virtnet_init_settings(dev); |
2623 | ||
296f96fc RR |
2624 | err = register_netdev(dev); |
2625 | if (err) { | |
2626 | pr_debug("virtio_net: registering device failed\n"); | |
d2a7ddda | 2627 | goto free_vqs; |
296f96fc | 2628 | } |
b3369c1f | 2629 | |
4baf1e33 MT |
2630 | virtio_device_ready(vdev); |
2631 | ||
8017c279 | 2632 | err = virtnet_cpu_notif_add(vi); |
8de4b2f3 WG |
2633 | if (err) { |
2634 | pr_debug("virtio_net: registering cpu notifier failed\n"); | |
f00e35e2 | 2635 | goto free_unregister_netdev; |
8de4b2f3 WG |
2636 | } |
2637 | ||
a220871b | 2638 | virtnet_set_queues(vi, vi->curr_queue_pairs); |
44900010 | 2639 | |
167c25e4 JW |
2640 | /* Assume link up if device can't report link status, |
2641 | otherwise get link status from config. */ | |
2642 | if (virtio_has_feature(vi->vdev, VIRTIO_NET_F_STATUS)) { | |
2643 | netif_carrier_off(dev); | |
3b07e9ca | 2644 | schedule_work(&vi->config_work); |
167c25e4 JW |
2645 | } else { |
2646 | vi->status = VIRTIO_NET_S_LINK_UP; | |
2647 | netif_carrier_on(dev); | |
2648 | } | |
9f4d26d0 | 2649 | |
3f93522f JW |
2650 | for (i = 0; i < ARRAY_SIZE(guest_offloads); i++) |
2651 | if (virtio_has_feature(vi->vdev, guest_offloads[i])) | |
2652 | set_bit(guest_offloads[i], &vi->guest_offloads); | |
2653 | ||
986a4f4d JW |
2654 | pr_debug("virtnet: registered device %s with %d RX and TX vq's\n", |
2655 | dev->name, max_queue_pairs); | |
2656 | ||
296f96fc RR |
2657 | return 0; |
2658 | ||
f00e35e2 | 2659 | free_unregister_netdev: |
02465555 MT |
2660 | vi->vdev->config->reset(vdev); |
2661 | ||
b3369c1f | 2662 | unregister_netdev(dev); |
d2a7ddda | 2663 | free_vqs: |
986a4f4d | 2664 | cancel_delayed_work_sync(&vi->refill); |
fb51879d | 2665 | free_receive_page_frags(vi); |
e9d7417b | 2666 | virtnet_del_vqs(vi); |
3fa2a1df | 2667 | free_stats: |
2668 | free_percpu(vi->stats); | |
296f96fc RR |
2669 | free: |
2670 | free_netdev(dev); | |
2671 | return err; | |
2672 | } | |
2673 | ||
04486ed0 | 2674 | static void remove_vq_common(struct virtnet_info *vi) |
296f96fc | 2675 | { |
04486ed0 | 2676 | vi->vdev->config->reset(vi->vdev); |
830a8a97 SM |
2677 | |
2678 | /* Free unused buffers in both send and recv, if any. */ | |
9ab86bbc | 2679 | free_unused_bufs(vi); |
fb6813f4 | 2680 | |
986a4f4d | 2681 | free_receive_bufs(vi); |
d2a7ddda | 2682 | |
fb51879d MD |
2683 | free_receive_page_frags(vi); |
2684 | ||
986a4f4d | 2685 | virtnet_del_vqs(vi); |
04486ed0 AS |
2686 | } |
2687 | ||
8cc085d6 | 2688 | static void virtnet_remove(struct virtio_device *vdev) |
04486ed0 AS |
2689 | { |
2690 | struct virtnet_info *vi = vdev->priv; | |
2691 | ||
8017c279 | 2692 | virtnet_cpu_notif_remove(vi); |
8de4b2f3 | 2693 | |
102a2786 MT |
2694 | /* Make sure no work handler is accessing the device. */ |
2695 | flush_work(&vi->config_work); | |
586d17c5 | 2696 | |
04486ed0 AS |
2697 | unregister_netdev(vi->dev); |
2698 | ||
2699 | remove_vq_common(vi); | |
fb6813f4 | 2700 | |
2e66f55b | 2701 | free_percpu(vi->stats); |
74b2553f | 2702 | free_netdev(vi->dev); |
296f96fc RR |
2703 | } |
2704 | ||
67a75194 | 2705 | static __maybe_unused int virtnet_freeze(struct virtio_device *vdev) |
0741bcb5 AS |
2706 | { |
2707 | struct virtnet_info *vi = vdev->priv; | |
2708 | ||
8017c279 | 2709 | virtnet_cpu_notif_remove(vi); |
9fe7bfce | 2710 | virtnet_freeze_down(vdev); |
0741bcb5 AS |
2711 | remove_vq_common(vi); |
2712 | ||
2713 | return 0; | |
2714 | } | |
2715 | ||
67a75194 | 2716 | static __maybe_unused int virtnet_restore(struct virtio_device *vdev) |
0741bcb5 AS |
2717 | { |
2718 | struct virtnet_info *vi = vdev->priv; | |
9fe7bfce | 2719 | int err; |
0741bcb5 | 2720 | |
9fe7bfce | 2721 | err = virtnet_restore_up(vdev); |
0741bcb5 AS |
2722 | if (err) |
2723 | return err; | |
986a4f4d JW |
2724 | virtnet_set_queues(vi, vi->curr_queue_pairs); |
2725 | ||
8017c279 | 2726 | err = virtnet_cpu_notif_add(vi); |
ec9debbd JW |
2727 | if (err) |
2728 | return err; | |
2729 | ||
0741bcb5 AS |
2730 | return 0; |
2731 | } | |
0741bcb5 | 2732 | |
296f96fc RR |
2733 | static struct virtio_device_id id_table[] = { |
2734 | { VIRTIO_ID_NET, VIRTIO_DEV_ANY_ID }, | |
2735 | { 0 }, | |
2736 | }; | |
2737 | ||
f3358507 MT |
2738 | #define VIRTNET_FEATURES \ |
2739 | VIRTIO_NET_F_CSUM, VIRTIO_NET_F_GUEST_CSUM, \ | |
2740 | VIRTIO_NET_F_MAC, \ | |
2741 | VIRTIO_NET_F_HOST_TSO4, VIRTIO_NET_F_HOST_UFO, VIRTIO_NET_F_HOST_TSO6, \ | |
2742 | VIRTIO_NET_F_HOST_ECN, VIRTIO_NET_F_GUEST_TSO4, VIRTIO_NET_F_GUEST_TSO6, \ | |
2743 | VIRTIO_NET_F_GUEST_ECN, VIRTIO_NET_F_GUEST_UFO, \ | |
2744 | VIRTIO_NET_F_MRG_RXBUF, VIRTIO_NET_F_STATUS, VIRTIO_NET_F_CTRL_VQ, \ | |
2745 | VIRTIO_NET_F_CTRL_RX, VIRTIO_NET_F_CTRL_VLAN, \ | |
2746 | VIRTIO_NET_F_GUEST_ANNOUNCE, VIRTIO_NET_F_MQ, \ | |
2747 | VIRTIO_NET_F_CTRL_MAC_ADDR, \ | |
3f93522f | 2748 | VIRTIO_NET_F_MTU, VIRTIO_NET_F_CTRL_GUEST_OFFLOADS |
f3358507 | 2749 | |
c45a6816 | 2750 | static unsigned int features[] = { |
f3358507 MT |
2751 | VIRTNET_FEATURES, |
2752 | }; | |
2753 | ||
2754 | static unsigned int features_legacy[] = { | |
2755 | VIRTNET_FEATURES, | |
2756 | VIRTIO_NET_F_GSO, | |
e7428e95 | 2757 | VIRTIO_F_ANY_LAYOUT, |
c45a6816 RR |
2758 | }; |
2759 | ||
22402529 | 2760 | static struct virtio_driver virtio_net_driver = { |
c45a6816 RR |
2761 | .feature_table = features, |
2762 | .feature_table_size = ARRAY_SIZE(features), | |
f3358507 MT |
2763 | .feature_table_legacy = features_legacy, |
2764 | .feature_table_size_legacy = ARRAY_SIZE(features_legacy), | |
296f96fc RR |
2765 | .driver.name = KBUILD_MODNAME, |
2766 | .driver.owner = THIS_MODULE, | |
2767 | .id_table = id_table, | |
fe36cbe0 | 2768 | .validate = virtnet_validate, |
296f96fc | 2769 | .probe = virtnet_probe, |
8cc085d6 | 2770 | .remove = virtnet_remove, |
9f4d26d0 | 2771 | .config_changed = virtnet_config_changed, |
89107000 | 2772 | #ifdef CONFIG_PM_SLEEP |
0741bcb5 AS |
2773 | .freeze = virtnet_freeze, |
2774 | .restore = virtnet_restore, | |
2775 | #endif | |
296f96fc RR |
2776 | }; |
2777 | ||
8017c279 SAS |
2778 | static __init int virtio_net_driver_init(void) |
2779 | { | |
2780 | int ret; | |
2781 | ||
73c1b41e | 2782 | ret = cpuhp_setup_state_multi(CPUHP_AP_ONLINE_DYN, "virtio/net:online", |
8017c279 SAS |
2783 | virtnet_cpu_online, |
2784 | virtnet_cpu_down_prep); | |
2785 | if (ret < 0) | |
2786 | goto out; | |
2787 | virtionet_online = ret; | |
73c1b41e | 2788 | ret = cpuhp_setup_state_multi(CPUHP_VIRT_NET_DEAD, "virtio/net:dead", |
8017c279 SAS |
2789 | NULL, virtnet_cpu_dead); |
2790 | if (ret) | |
2791 | goto err_dead; | |
2792 | ||
2793 | ret = register_virtio_driver(&virtio_net_driver); | |
2794 | if (ret) | |
2795 | goto err_virtio; | |
2796 | return 0; | |
2797 | err_virtio: | |
2798 | cpuhp_remove_multi_state(CPUHP_VIRT_NET_DEAD); | |
2799 | err_dead: | |
2800 | cpuhp_remove_multi_state(virtionet_online); | |
2801 | out: | |
2802 | return ret; | |
2803 | } | |
2804 | module_init(virtio_net_driver_init); | |
2805 | ||
2806 | static __exit void virtio_net_driver_exit(void) | |
2807 | { | |
cfa0ebc9 | 2808 | unregister_virtio_driver(&virtio_net_driver); |
8017c279 SAS |
2809 | cpuhp_remove_multi_state(CPUHP_VIRT_NET_DEAD); |
2810 | cpuhp_remove_multi_state(virtionet_online); | |
8017c279 SAS |
2811 | } |
2812 | module_exit(virtio_net_driver_exit); | |
296f96fc RR |
2813 | |
2814 | MODULE_DEVICE_TABLE(virtio, id_table); | |
2815 | MODULE_DESCRIPTION("Virtio network driver"); | |
2816 | MODULE_LICENSE("GPL"); |