]>
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> |
186b3c99 | 32 | #include <linux/filter.h> |
d85b758f | 33 | #include <net/route.h> |
296f96fc | 34 | |
d34710e3 | 35 | static int napi_weight = NAPI_POLL_WEIGHT; |
6c0cd7c0 DL |
36 | module_param(napi_weight, int, 0444); |
37 | ||
b92f1e67 | 38 | static bool csum = true, gso = true, napi_tx; |
34a48579 RR |
39 | module_param(csum, bool, 0444); |
40 | module_param(gso, bool, 0444); | |
b92f1e67 | 41 | module_param(napi_tx, bool, 0644); |
34a48579 | 42 | |
296f96fc | 43 | /* FIXME: MTU in config. */ |
5061de36 | 44 | #define GOOD_PACKET_LEN (ETH_HLEN + VLAN_HLEN + ETH_DATA_LEN) |
3f2c31d9 | 45 | #define GOOD_COPY_LEN 128 |
296f96fc | 46 | |
f6b10209 JW |
47 | #define VIRTNET_RX_PAD (NET_IP_ALIGN + NET_SKB_PAD) |
48 | ||
2de2f7f4 JF |
49 | /* Amount of XDP headroom to prepend to packets for use by xdp_adjust_head */ |
50 | #define VIRTIO_XDP_HEADROOM 256 | |
51 | ||
5377d758 JB |
52 | /* RX packet size EWMA. The average packet size is used to determine the packet |
53 | * buffer size when refilling RX rings. As the entire RX ring may be refilled | |
54 | * at once, the weight is chosen so that the EWMA will be insensitive to short- | |
55 | * term, transient changes in packet size. | |
ab7db917 | 56 | */ |
eb1e011a | 57 | DECLARE_EWMA(pkt_len, 0, 64) |
ab7db917 | 58 | |
66846048 | 59 | #define VIRTNET_DRIVER_VERSION "1.0.0" |
2a41f71d | 60 | |
7acd4329 CIK |
61 | static const unsigned long guest_offloads[] = { |
62 | VIRTIO_NET_F_GUEST_TSO4, | |
63 | VIRTIO_NET_F_GUEST_TSO6, | |
64 | VIRTIO_NET_F_GUEST_ECN, | |
65 | VIRTIO_NET_F_GUEST_UFO | |
66 | }; | |
3f93522f | 67 | |
3fa2a1df | 68 | struct virtnet_stats { |
83a27052 ED |
69 | struct u64_stats_sync tx_syncp; |
70 | struct u64_stats_sync rx_syncp; | |
3fa2a1df | 71 | u64 tx_bytes; |
72 | u64 tx_packets; | |
73 | ||
74 | u64 rx_bytes; | |
75 | u64 rx_packets; | |
76 | }; | |
77 | ||
e9d7417b JW |
78 | /* Internal representation of a send virtqueue */ |
79 | struct send_queue { | |
80 | /* Virtqueue associated with this send _queue */ | |
81 | struct virtqueue *vq; | |
82 | ||
83 | /* TX: fragments + linear part + virtio header */ | |
84 | struct scatterlist sg[MAX_SKB_FRAGS + 2]; | |
986a4f4d JW |
85 | |
86 | /* Name of the send queue: output.$index */ | |
87 | char name[40]; | |
b92f1e67 WB |
88 | |
89 | struct napi_struct napi; | |
e9d7417b JW |
90 | }; |
91 | ||
92 | /* Internal representation of a receive virtqueue */ | |
93 | struct receive_queue { | |
94 | /* Virtqueue associated with this receive_queue */ | |
95 | struct virtqueue *vq; | |
96 | ||
296f96fc RR |
97 | struct napi_struct napi; |
98 | ||
f600b690 JF |
99 | struct bpf_prog __rcu *xdp_prog; |
100 | ||
e9d7417b JW |
101 | /* Chain pages by the private ptr. */ |
102 | struct page *pages; | |
103 | ||
ab7db917 | 104 | /* Average packet length for mergeable receive buffers. */ |
5377d758 | 105 | struct ewma_pkt_len mrg_avg_pkt_len; |
ab7db917 | 106 | |
fb51879d MD |
107 | /* Page frag for packet buffer allocation. */ |
108 | struct page_frag alloc_frag; | |
109 | ||
e9d7417b JW |
110 | /* RX: fragments + linear part + virtio header */ |
111 | struct scatterlist sg[MAX_SKB_FRAGS + 2]; | |
986a4f4d | 112 | |
d85b758f MT |
113 | /* Min single buffer size for mergeable buffers case. */ |
114 | unsigned int min_buf_len; | |
115 | ||
986a4f4d JW |
116 | /* Name of this receive queue: input.$index */ |
117 | char name[40]; | |
e9d7417b JW |
118 | }; |
119 | ||
603343b9 MT |
120 | /* Control VQ buffers: protected by the rtnl lock */ |
121 | struct control_buf { | |
122 | struct virtio_net_ctrl_hdr hdr; | |
123 | virtio_net_ctrl_ack status; | |
124 | struct virtio_net_ctrl_mq mq; | |
125 | u8 promisc; | |
126 | u8 allmulti; | |
c285a7d5 | 127 | __virtio16 vid; |
603343b9 MT |
128 | u64 offloads; |
129 | }; | |
130 | ||
e9d7417b JW |
131 | struct virtnet_info { |
132 | struct virtio_device *vdev; | |
133 | struct virtqueue *cvq; | |
134 | struct net_device *dev; | |
986a4f4d JW |
135 | struct send_queue *sq; |
136 | struct receive_queue *rq; | |
e9d7417b JW |
137 | unsigned int status; |
138 | ||
986a4f4d JW |
139 | /* Max # of queue pairs supported by the device */ |
140 | u16 max_queue_pairs; | |
141 | ||
142 | /* # of queue pairs currently used by the driver */ | |
143 | u16 curr_queue_pairs; | |
144 | ||
672aafd5 JF |
145 | /* # of XDP queue pairs currently used by the driver */ |
146 | u16 xdp_queue_pairs; | |
147 | ||
97402b96 HX |
148 | /* I like... big packets and I cannot lie! */ |
149 | bool big_packets; | |
150 | ||
3f2c31d9 MM |
151 | /* Host will merge rx buffers for big packets (shake it! shake it!) */ |
152 | bool mergeable_rx_bufs; | |
153 | ||
986a4f4d JW |
154 | /* Has control virtqueue */ |
155 | bool has_cvq; | |
156 | ||
e7428e95 MT |
157 | /* Host can handle any s/g split between our header and packet data */ |
158 | bool any_header_sg; | |
159 | ||
012873d0 MT |
160 | /* Packet virtio header size */ |
161 | u8 hdr_len; | |
162 | ||
3fa2a1df | 163 | /* Active statistics */ |
164 | struct virtnet_stats __percpu *stats; | |
165 | ||
3161e453 RR |
166 | /* Work struct for refilling if we run low on memory. */ |
167 | struct delayed_work refill; | |
168 | ||
586d17c5 JW |
169 | /* Work struct for config space updates */ |
170 | struct work_struct config_work; | |
171 | ||
986a4f4d JW |
172 | /* Does the affinity hint is set for virtqueues? */ |
173 | bool affinity_hint_set; | |
47be2479 | 174 | |
8017c279 SAS |
175 | /* CPU hotplug instances for online & dead */ |
176 | struct hlist_node node; | |
177 | struct hlist_node node_dead; | |
2ac46030 | 178 | |
603343b9 | 179 | struct control_buf *ctrl; |
16032be5 NA |
180 | |
181 | /* Ethtool settings */ | |
182 | u8 duplex; | |
183 | u32 speed; | |
3f93522f JW |
184 | |
185 | unsigned long guest_offloads; | |
296f96fc RR |
186 | }; |
187 | ||
9ab86bbc | 188 | struct padded_vnet_hdr { |
012873d0 | 189 | struct virtio_net_hdr_mrg_rxbuf hdr; |
9ab86bbc | 190 | /* |
012873d0 MT |
191 | * hdr is in a separate sg buffer, and data sg buffer shares same page |
192 | * with this header sg. This padding makes next sg 16 byte aligned | |
193 | * after the header. | |
9ab86bbc | 194 | */ |
012873d0 | 195 | char padding[4]; |
9ab86bbc SM |
196 | }; |
197 | ||
986a4f4d JW |
198 | /* Converting between virtqueue no. and kernel tx/rx queue no. |
199 | * 0:rx0 1:tx0 2:rx1 3:tx1 ... 2N:rxN 2N+1:txN 2N+2:cvq | |
200 | */ | |
201 | static int vq2txq(struct virtqueue *vq) | |
202 | { | |
9d0ca6ed | 203 | return (vq->index - 1) / 2; |
986a4f4d JW |
204 | } |
205 | ||
206 | static int txq2vq(int txq) | |
207 | { | |
208 | return txq * 2 + 1; | |
209 | } | |
210 | ||
211 | static int vq2rxq(struct virtqueue *vq) | |
212 | { | |
9d0ca6ed | 213 | return vq->index / 2; |
986a4f4d JW |
214 | } |
215 | ||
216 | static int rxq2vq(int rxq) | |
217 | { | |
218 | return rxq * 2; | |
219 | } | |
220 | ||
012873d0 | 221 | static inline struct virtio_net_hdr_mrg_rxbuf *skb_vnet_hdr(struct sk_buff *skb) |
296f96fc | 222 | { |
012873d0 | 223 | return (struct virtio_net_hdr_mrg_rxbuf *)skb->cb; |
296f96fc RR |
224 | } |
225 | ||
9ab86bbc SM |
226 | /* |
227 | * private is used to chain pages for big packets, put the whole | |
228 | * most recent used list in the beginning for reuse | |
229 | */ | |
e9d7417b | 230 | static void give_pages(struct receive_queue *rq, struct page *page) |
0a888fd1 | 231 | { |
9ab86bbc | 232 | struct page *end; |
0a888fd1 | 233 | |
e9d7417b | 234 | /* Find end of list, sew whole thing into vi->rq.pages. */ |
9ab86bbc | 235 | for (end = page; end->private; end = (struct page *)end->private); |
e9d7417b JW |
236 | end->private = (unsigned long)rq->pages; |
237 | rq->pages = page; | |
0a888fd1 MM |
238 | } |
239 | ||
e9d7417b | 240 | static struct page *get_a_page(struct receive_queue *rq, gfp_t gfp_mask) |
fb6813f4 | 241 | { |
e9d7417b | 242 | struct page *p = rq->pages; |
fb6813f4 | 243 | |
9ab86bbc | 244 | if (p) { |
e9d7417b | 245 | rq->pages = (struct page *)p->private; |
9ab86bbc SM |
246 | /* clear private here, it is used to chain pages */ |
247 | p->private = 0; | |
248 | } else | |
fb6813f4 RR |
249 | p = alloc_page(gfp_mask); |
250 | return p; | |
251 | } | |
252 | ||
e4e8452a WB |
253 | static void virtqueue_napi_schedule(struct napi_struct *napi, |
254 | struct virtqueue *vq) | |
255 | { | |
256 | if (napi_schedule_prep(napi)) { | |
257 | virtqueue_disable_cb(vq); | |
258 | __napi_schedule(napi); | |
259 | } | |
260 | } | |
261 | ||
262 | static void virtqueue_napi_complete(struct napi_struct *napi, | |
263 | struct virtqueue *vq, int processed) | |
264 | { | |
265 | int opaque; | |
266 | ||
267 | opaque = virtqueue_enable_cb_prepare(vq); | |
e3eb8fe2 TM |
268 | if (napi_complete_done(napi, processed)) { |
269 | if (unlikely(virtqueue_poll(vq, opaque))) | |
270 | virtqueue_napi_schedule(napi, vq); | |
271 | } else { | |
272 | virtqueue_disable_cb(vq); | |
273 | } | |
e4e8452a WB |
274 | } |
275 | ||
e9d7417b | 276 | static void skb_xmit_done(struct virtqueue *vq) |
296f96fc | 277 | { |
e9d7417b | 278 | struct virtnet_info *vi = vq->vdev->priv; |
b92f1e67 | 279 | struct napi_struct *napi = &vi->sq[vq2txq(vq)].napi; |
296f96fc | 280 | |
2cb9c6ba | 281 | /* Suppress further interrupts. */ |
e9d7417b | 282 | virtqueue_disable_cb(vq); |
11a3a154 | 283 | |
b92f1e67 WB |
284 | if (napi->weight) |
285 | virtqueue_napi_schedule(napi, vq); | |
286 | else | |
287 | /* We were probably waiting for more output buffers. */ | |
288 | netif_wake_subqueue(vi->dev, vq2txq(vq)); | |
296f96fc RR |
289 | } |
290 | ||
28b39bc7 JW |
291 | #define MRG_CTX_HEADER_SHIFT 22 |
292 | static void *mergeable_len_to_ctx(unsigned int truesize, | |
293 | unsigned int headroom) | |
294 | { | |
295 | return (void *)(unsigned long)((headroom << MRG_CTX_HEADER_SHIFT) | truesize); | |
296 | } | |
297 | ||
298 | static unsigned int mergeable_ctx_to_headroom(void *mrg_ctx) | |
299 | { | |
300 | return (unsigned long)mrg_ctx >> MRG_CTX_HEADER_SHIFT; | |
301 | } | |
302 | ||
303 | static unsigned int mergeable_ctx_to_truesize(void *mrg_ctx) | |
304 | { | |
305 | return (unsigned long)mrg_ctx & ((1 << MRG_CTX_HEADER_SHIFT) - 1); | |
306 | } | |
307 | ||
3464645a | 308 | /* Called from bottom half context */ |
946fa564 MT |
309 | static struct sk_buff *page_to_skb(struct virtnet_info *vi, |
310 | struct receive_queue *rq, | |
2613af0e MD |
311 | struct page *page, unsigned int offset, |
312 | unsigned int len, unsigned int truesize) | |
9ab86bbc SM |
313 | { |
314 | struct sk_buff *skb; | |
012873d0 | 315 | struct virtio_net_hdr_mrg_rxbuf *hdr; |
2613af0e | 316 | unsigned int copy, hdr_len, hdr_padded_len; |
9ab86bbc | 317 | char *p; |
fb6813f4 | 318 | |
2613af0e | 319 | p = page_address(page) + offset; |
3f2c31d9 | 320 | |
9ab86bbc | 321 | /* copy small packet so we can reuse these pages for small data */ |
c67f5db8 | 322 | skb = napi_alloc_skb(&rq->napi, GOOD_COPY_LEN); |
9ab86bbc SM |
323 | if (unlikely(!skb)) |
324 | return NULL; | |
3f2c31d9 | 325 | |
9ab86bbc | 326 | hdr = skb_vnet_hdr(skb); |
3f2c31d9 | 327 | |
012873d0 MT |
328 | hdr_len = vi->hdr_len; |
329 | if (vi->mergeable_rx_bufs) | |
a4a76503 | 330 | hdr_padded_len = sizeof(*hdr); |
012873d0 | 331 | else |
2613af0e | 332 | hdr_padded_len = sizeof(struct padded_vnet_hdr); |
3f2c31d9 | 333 | |
9ab86bbc | 334 | memcpy(hdr, p, hdr_len); |
3f2c31d9 | 335 | |
9ab86bbc | 336 | len -= hdr_len; |
2613af0e MD |
337 | offset += hdr_padded_len; |
338 | p += hdr_padded_len; | |
3f2c31d9 | 339 | |
9ab86bbc SM |
340 | copy = len; |
341 | if (copy > skb_tailroom(skb)) | |
342 | copy = skb_tailroom(skb); | |
59ae1d12 | 343 | skb_put_data(skb, p, copy); |
3f2c31d9 | 344 | |
9ab86bbc SM |
345 | len -= copy; |
346 | offset += copy; | |
3f2c31d9 | 347 | |
2613af0e MD |
348 | if (vi->mergeable_rx_bufs) { |
349 | if (len) | |
350 | skb_add_rx_frag(skb, 0, page, offset, len, truesize); | |
351 | else | |
352 | put_page(page); | |
353 | return skb; | |
354 | } | |
355 | ||
e878d78b SL |
356 | /* |
357 | * Verify that we can indeed put this data into a skb. | |
358 | * This is here to handle cases when the device erroneously | |
359 | * tries to receive more than is possible. This is usually | |
360 | * the case of a broken device. | |
361 | */ | |
362 | if (unlikely(len > MAX_SKB_FRAGS * PAGE_SIZE)) { | |
be443899 | 363 | net_dbg_ratelimited("%s: too much data\n", skb->dev->name); |
e878d78b SL |
364 | dev_kfree_skb(skb); |
365 | return NULL; | |
366 | } | |
2613af0e | 367 | BUG_ON(offset >= PAGE_SIZE); |
9ab86bbc | 368 | while (len) { |
2613af0e MD |
369 | unsigned int frag_size = min((unsigned)PAGE_SIZE - offset, len); |
370 | skb_add_rx_frag(skb, skb_shinfo(skb)->nr_frags, page, offset, | |
371 | frag_size, truesize); | |
372 | len -= frag_size; | |
9ab86bbc SM |
373 | page = (struct page *)page->private; |
374 | offset = 0; | |
375 | } | |
3f2c31d9 | 376 | |
9ab86bbc | 377 | if (page) |
e9d7417b | 378 | give_pages(rq, page); |
3f2c31d9 | 379 | |
9ab86bbc SM |
380 | return skb; |
381 | } | |
3f2c31d9 | 382 | |
186b3c99 JW |
383 | static void virtnet_xdp_flush(struct net_device *dev) |
384 | { | |
385 | struct virtnet_info *vi = netdev_priv(dev); | |
386 | struct send_queue *sq; | |
387 | unsigned int qp; | |
388 | ||
389 | qp = vi->curr_queue_pairs - vi->xdp_queue_pairs + smp_processor_id(); | |
390 | sq = &vi->sq[qp]; | |
391 | ||
392 | virtqueue_kick(sq->vq); | |
393 | } | |
394 | ||
395 | static bool __virtnet_xdp_xmit(struct virtnet_info *vi, | |
396 | struct xdp_buff *xdp) | |
56434a01 | 397 | { |
56434a01 | 398 | struct virtio_net_hdr_mrg_rxbuf *hdr; |
f6b10209 | 399 | unsigned int len; |
722d8283 JF |
400 | struct send_queue *sq; |
401 | unsigned int qp; | |
56434a01 JF |
402 | void *xdp_sent; |
403 | int err; | |
404 | ||
722d8283 JF |
405 | qp = vi->curr_queue_pairs - vi->xdp_queue_pairs + smp_processor_id(); |
406 | sq = &vi->sq[qp]; | |
407 | ||
56434a01 JF |
408 | /* Free up any pending old buffers before queueing new ones. */ |
409 | while ((xdp_sent = virtqueue_get_buf(sq->vq, &len)) != NULL) { | |
f6b10209 | 410 | struct page *sent_page = virt_to_head_page(xdp_sent); |
bb91accf | 411 | |
f6b10209 | 412 | put_page(sent_page); |
bb91accf | 413 | } |
56434a01 | 414 | |
f6b10209 JW |
415 | xdp->data -= vi->hdr_len; |
416 | /* Zero header and leave csum up to XDP layers */ | |
417 | hdr = xdp->data; | |
418 | memset(hdr, 0, vi->hdr_len); | |
bb91accf | 419 | |
f6b10209 | 420 | sg_init_one(sq->sg, xdp->data, xdp->data_end - xdp->data); |
bb91accf | 421 | |
f6b10209 | 422 | err = virtqueue_add_outbuf(sq->vq, sq->sg, 1, xdp->data, GFP_ATOMIC); |
6e49f219 JDB |
423 | if (unlikely(err)) |
424 | return false; /* Caller handle free/refcnt */ | |
56434a01 | 425 | |
a67edbf4 | 426 | return true; |
56434a01 JF |
427 | } |
428 | ||
186b3c99 JW |
429 | static int virtnet_xdp_xmit(struct net_device *dev, struct xdp_buff *xdp) |
430 | { | |
431 | struct virtnet_info *vi = netdev_priv(dev); | |
432 | bool sent = __virtnet_xdp_xmit(vi, xdp); | |
433 | ||
434 | if (!sent) | |
435 | return -ENOSPC; | |
436 | return 0; | |
437 | } | |
438 | ||
f6b10209 JW |
439 | static unsigned int virtnet_get_headroom(struct virtnet_info *vi) |
440 | { | |
441 | return vi->xdp_queue_pairs ? VIRTIO_XDP_HEADROOM : 0; | |
442 | } | |
443 | ||
4941d472 JW |
444 | /* We copy the packet for XDP in the following cases: |
445 | * | |
446 | * 1) Packet is scattered across multiple rx buffers. | |
447 | * 2) Headroom space is insufficient. | |
448 | * | |
449 | * This is inefficient but it's a temporary condition that | |
450 | * we hit right after XDP is enabled and until queue is refilled | |
451 | * with large buffers with sufficient headroom - so it should affect | |
452 | * at most queue size packets. | |
453 | * Afterwards, the conditions to enable | |
454 | * XDP should preclude the underlying device from sending packets | |
455 | * across multiple buffers (num_buf > 1), and we make sure buffers | |
456 | * have enough headroom. | |
457 | */ | |
458 | static struct page *xdp_linearize_page(struct receive_queue *rq, | |
459 | u16 *num_buf, | |
460 | struct page *p, | |
461 | int offset, | |
462 | int page_off, | |
463 | unsigned int *len) | |
464 | { | |
465 | struct page *page = alloc_page(GFP_ATOMIC); | |
466 | ||
467 | if (!page) | |
468 | return NULL; | |
469 | ||
470 | memcpy(page_address(page) + page_off, page_address(p) + offset, *len); | |
471 | page_off += *len; | |
472 | ||
473 | while (--*num_buf) { | |
474 | unsigned int buflen; | |
475 | void *buf; | |
476 | int off; | |
477 | ||
478 | buf = virtqueue_get_buf(rq->vq, &buflen); | |
479 | if (unlikely(!buf)) | |
480 | goto err_buf; | |
481 | ||
482 | p = virt_to_head_page(buf); | |
483 | off = buf - page_address(p); | |
484 | ||
485 | /* guard against a misconfigured or uncooperative backend that | |
486 | * is sending packet larger than the MTU. | |
487 | */ | |
488 | if ((page_off + buflen) > PAGE_SIZE) { | |
489 | put_page(p); | |
490 | goto err_buf; | |
491 | } | |
492 | ||
493 | memcpy(page_address(page) + page_off, | |
494 | page_address(p) + off, buflen); | |
495 | page_off += buflen; | |
496 | put_page(p); | |
497 | } | |
498 | ||
499 | /* Headroom does not contribute to packet length */ | |
500 | *len = page_off - VIRTIO_XDP_HEADROOM; | |
501 | return page; | |
502 | err_buf: | |
503 | __free_pages(page, 0); | |
504 | return NULL; | |
505 | } | |
506 | ||
bb91accf JW |
507 | static struct sk_buff *receive_small(struct net_device *dev, |
508 | struct virtnet_info *vi, | |
509 | struct receive_queue *rq, | |
192f68cf | 510 | void *buf, void *ctx, |
186b3c99 JW |
511 | unsigned int len, |
512 | bool *xdp_xmit) | |
f121159d | 513 | { |
f6b10209 | 514 | struct sk_buff *skb; |
bb91accf | 515 | struct bpf_prog *xdp_prog; |
4941d472 | 516 | unsigned int xdp_headroom = (unsigned long)ctx; |
f6b10209 JW |
517 | unsigned int header_offset = VIRTNET_RX_PAD + xdp_headroom; |
518 | unsigned int headroom = vi->hdr_len + header_offset; | |
519 | unsigned int buflen = SKB_DATA_ALIGN(GOOD_PACKET_LEN + headroom) + | |
520 | SKB_DATA_ALIGN(sizeof(struct skb_shared_info)); | |
4941d472 | 521 | struct page *page = virt_to_head_page(buf); |
6e49f219 | 522 | unsigned int delta = 0; |
4941d472 | 523 | struct page *xdp_page; |
6e49f219 JDB |
524 | bool sent; |
525 | int err; | |
526 | ||
012873d0 | 527 | len -= vi->hdr_len; |
f121159d | 528 | |
bb91accf JW |
529 | rcu_read_lock(); |
530 | xdp_prog = rcu_dereference(rq->xdp_prog); | |
531 | if (xdp_prog) { | |
f6b10209 | 532 | struct virtio_net_hdr_mrg_rxbuf *hdr = buf + header_offset; |
0354e4d1 | 533 | struct xdp_buff xdp; |
f6b10209 | 534 | void *orig_data; |
bb91accf JW |
535 | u32 act; |
536 | ||
36c3dce0 | 537 | if (unlikely(hdr->hdr.gso_type)) |
bb91accf | 538 | goto err_xdp; |
0354e4d1 | 539 | |
4941d472 JW |
540 | if (unlikely(xdp_headroom < virtnet_get_headroom(vi))) { |
541 | int offset = buf - page_address(page) + header_offset; | |
542 | unsigned int tlen = len + vi->hdr_len; | |
543 | u16 num_buf = 1; | |
544 | ||
545 | xdp_headroom = virtnet_get_headroom(vi); | |
546 | header_offset = VIRTNET_RX_PAD + xdp_headroom; | |
547 | headroom = vi->hdr_len + header_offset; | |
548 | buflen = SKB_DATA_ALIGN(GOOD_PACKET_LEN + headroom) + | |
549 | SKB_DATA_ALIGN(sizeof(struct skb_shared_info)); | |
550 | xdp_page = xdp_linearize_page(rq, &num_buf, page, | |
551 | offset, header_offset, | |
552 | &tlen); | |
553 | if (!xdp_page) | |
554 | goto err_xdp; | |
555 | ||
556 | buf = page_address(xdp_page); | |
557 | put_page(page); | |
558 | page = xdp_page; | |
559 | } | |
560 | ||
f6b10209 JW |
561 | xdp.data_hard_start = buf + VIRTNET_RX_PAD + vi->hdr_len; |
562 | xdp.data = xdp.data_hard_start + xdp_headroom; | |
de8f3a83 | 563 | xdp_set_data_meta_invalid(&xdp); |
0354e4d1 | 564 | xdp.data_end = xdp.data + len; |
f6b10209 | 565 | orig_data = xdp.data; |
0354e4d1 JF |
566 | act = bpf_prog_run_xdp(xdp_prog, &xdp); |
567 | ||
bb91accf JW |
568 | switch (act) { |
569 | case XDP_PASS: | |
2de2f7f4 | 570 | /* Recalculate length in case bpf program changed it */ |
f6b10209 | 571 | delta = orig_data - xdp.data; |
bb91accf JW |
572 | break; |
573 | case XDP_TX: | |
6e49f219 JDB |
574 | sent = __virtnet_xdp_xmit(vi, &xdp); |
575 | if (unlikely(!sent)) { | |
0354e4d1 | 576 | trace_xdp_exception(vi->dev, xdp_prog, act); |
6e49f219 JDB |
577 | goto err_xdp; |
578 | } | |
579 | *xdp_xmit = true; | |
186b3c99 JW |
580 | rcu_read_unlock(); |
581 | goto xdp_xmit; | |
582 | case XDP_REDIRECT: | |
583 | err = xdp_do_redirect(dev, &xdp, xdp_prog); | |
6e49f219 JDB |
584 | if (err) |
585 | goto err_xdp; | |
586 | *xdp_xmit = true; | |
bb91accf JW |
587 | rcu_read_unlock(); |
588 | goto xdp_xmit; | |
bb91accf | 589 | default: |
0354e4d1 JF |
590 | bpf_warn_invalid_xdp_action(act); |
591 | case XDP_ABORTED: | |
592 | trace_xdp_exception(vi->dev, xdp_prog, act); | |
593 | case XDP_DROP: | |
bb91accf JW |
594 | goto err_xdp; |
595 | } | |
596 | } | |
597 | rcu_read_unlock(); | |
598 | ||
f6b10209 JW |
599 | skb = build_skb(buf, buflen); |
600 | if (!skb) { | |
4941d472 | 601 | put_page(page); |
f6b10209 JW |
602 | goto err; |
603 | } | |
604 | skb_reserve(skb, headroom - delta); | |
605 | skb_put(skb, len + delta); | |
606 | if (!delta) { | |
607 | buf += header_offset; | |
608 | memcpy(skb_vnet_hdr(skb), buf, vi->hdr_len); | |
609 | } /* keep zeroed vnet hdr since packet was changed by bpf */ | |
610 | ||
611 | err: | |
f121159d | 612 | return skb; |
bb91accf JW |
613 | |
614 | err_xdp: | |
615 | rcu_read_unlock(); | |
616 | dev->stats.rx_dropped++; | |
4941d472 | 617 | put_page(page); |
bb91accf JW |
618 | xdp_xmit: |
619 | return NULL; | |
f121159d MT |
620 | } |
621 | ||
622 | static struct sk_buff *receive_big(struct net_device *dev, | |
946fa564 | 623 | struct virtnet_info *vi, |
f121159d MT |
624 | struct receive_queue *rq, |
625 | void *buf, | |
626 | unsigned int len) | |
627 | { | |
628 | struct page *page = buf; | |
c47a43d3 | 629 | struct sk_buff *skb = page_to_skb(vi, rq, page, 0, len, PAGE_SIZE); |
f600b690 | 630 | |
f121159d MT |
631 | if (unlikely(!skb)) |
632 | goto err; | |
633 | ||
634 | return skb; | |
635 | ||
636 | err: | |
637 | dev->stats.rx_dropped++; | |
638 | give_pages(rq, page); | |
639 | return NULL; | |
640 | } | |
641 | ||
8fc3b9e9 | 642 | static struct sk_buff *receive_mergeable(struct net_device *dev, |
fdd819b2 | 643 | struct virtnet_info *vi, |
8fc3b9e9 | 644 | struct receive_queue *rq, |
680557cf MT |
645 | void *buf, |
646 | void *ctx, | |
186b3c99 JW |
647 | unsigned int len, |
648 | bool *xdp_xmit) | |
9ab86bbc | 649 | { |
012873d0 MT |
650 | struct virtio_net_hdr_mrg_rxbuf *hdr = buf; |
651 | u16 num_buf = virtio16_to_cpu(vi->vdev, hdr->num_buffers); | |
8fc3b9e9 MT |
652 | struct page *page = virt_to_head_page(buf); |
653 | int offset = buf - page_address(page); | |
f600b690 JF |
654 | struct sk_buff *head_skb, *curr_skb; |
655 | struct bpf_prog *xdp_prog; | |
656 | unsigned int truesize; | |
4941d472 | 657 | unsigned int headroom = mergeable_ctx_to_headroom(ctx); |
186b3c99 | 658 | int err; |
6e49f219 | 659 | bool sent; |
f600b690 | 660 | |
56434a01 JF |
661 | head_skb = NULL; |
662 | ||
f600b690 JF |
663 | rcu_read_lock(); |
664 | xdp_prog = rcu_dereference(rq->xdp_prog); | |
665 | if (xdp_prog) { | |
72979a6c | 666 | struct page *xdp_page; |
0354e4d1 | 667 | struct xdp_buff xdp; |
0354e4d1 | 668 | void *data; |
f600b690 JF |
669 | u32 act; |
670 | ||
5cf0b8a6 JW |
671 | /* Transient failure which in theory could occur if |
672 | * in-flight packets from before XDP was enabled reach | |
673 | * the receive path after XDP is loaded. | |
674 | */ | |
675 | if (unlikely(hdr->hdr.gso_type)) | |
676 | goto err_xdp; | |
677 | ||
73b62bd0 | 678 | /* This happens when rx buffer size is underestimated */ |
4941d472 JW |
679 | if (unlikely(num_buf > 1 || |
680 | headroom < virtnet_get_headroom(vi))) { | |
72979a6c | 681 | /* linearize data for XDP */ |
56a86f84 | 682 | xdp_page = xdp_linearize_page(rq, &num_buf, |
4941d472 JW |
683 | page, offset, |
684 | VIRTIO_XDP_HEADROOM, | |
685 | &len); | |
72979a6c JF |
686 | if (!xdp_page) |
687 | goto err_xdp; | |
2de2f7f4 | 688 | offset = VIRTIO_XDP_HEADROOM; |
72979a6c JF |
689 | } else { |
690 | xdp_page = page; | |
f600b690 JF |
691 | } |
692 | ||
2de2f7f4 JF |
693 | /* Allow consuming headroom but reserve enough space to push |
694 | * the descriptor on if we get an XDP_TX return code. | |
695 | */ | |
0354e4d1 | 696 | data = page_address(xdp_page) + offset; |
2de2f7f4 | 697 | xdp.data_hard_start = data - VIRTIO_XDP_HEADROOM + vi->hdr_len; |
0354e4d1 | 698 | xdp.data = data + vi->hdr_len; |
de8f3a83 | 699 | xdp_set_data_meta_invalid(&xdp); |
0354e4d1 JF |
700 | xdp.data_end = xdp.data + (len - vi->hdr_len); |
701 | act = bpf_prog_run_xdp(xdp_prog, &xdp); | |
702 | ||
31240345 JW |
703 | if (act != XDP_PASS) |
704 | ewma_pkt_len_add(&rq->mrg_avg_pkt_len, len); | |
705 | ||
56434a01 JF |
706 | switch (act) { |
707 | case XDP_PASS: | |
2de2f7f4 JF |
708 | /* recalculate offset to account for any header |
709 | * adjustments. Note other cases do not build an | |
710 | * skb and avoid using offset | |
711 | */ | |
712 | offset = xdp.data - | |
713 | page_address(xdp_page) - vi->hdr_len; | |
714 | ||
1830f893 JW |
715 | /* We can only create skb based on xdp_page. */ |
716 | if (unlikely(xdp_page != page)) { | |
717 | rcu_read_unlock(); | |
718 | put_page(page); | |
719 | head_skb = page_to_skb(vi, rq, xdp_page, | |
2de2f7f4 | 720 | offset, len, PAGE_SIZE); |
1830f893 JW |
721 | return head_skb; |
722 | } | |
56434a01 JF |
723 | break; |
724 | case XDP_TX: | |
6e49f219 JDB |
725 | sent = __virtnet_xdp_xmit(vi, &xdp); |
726 | if (unlikely(!sent)) { | |
0354e4d1 | 727 | trace_xdp_exception(vi->dev, xdp_prog, act); |
6e49f219 JDB |
728 | if (unlikely(xdp_page != page)) |
729 | put_page(xdp_page); | |
730 | goto err_xdp; | |
731 | } | |
732 | *xdp_xmit = true; | |
72979a6c | 733 | if (unlikely(xdp_page != page)) |
62b92b48 | 734 | put_page(page); |
56434a01 JF |
735 | rcu_read_unlock(); |
736 | goto xdp_xmit; | |
186b3c99 JW |
737 | case XDP_REDIRECT: |
738 | err = xdp_do_redirect(dev, &xdp, xdp_prog); | |
dd543797 | 739 | if (!err) |
186b3c99 JW |
740 | *xdp_xmit = true; |
741 | rcu_read_unlock(); | |
742 | goto xdp_xmit; | |
56434a01 | 743 | default: |
0354e4d1 JF |
744 | bpf_warn_invalid_xdp_action(act); |
745 | case XDP_ABORTED: | |
746 | trace_xdp_exception(vi->dev, xdp_prog, act); | |
747 | case XDP_DROP: | |
72979a6c JF |
748 | if (unlikely(xdp_page != page)) |
749 | __free_pages(xdp_page, 0); | |
f600b690 | 750 | goto err_xdp; |
56434a01 | 751 | } |
f600b690 JF |
752 | } |
753 | rcu_read_unlock(); | |
ab7db917 | 754 | |
28b39bc7 JW |
755 | truesize = mergeable_ctx_to_truesize(ctx); |
756 | if (unlikely(len > truesize)) { | |
56da5fd0 | 757 | pr_debug("%s: rx error: len %u exceeds truesize %lu\n", |
680557cf MT |
758 | dev->name, len, (unsigned long)ctx); |
759 | dev->stats.rx_length_errors++; | |
760 | goto err_skb; | |
761 | } | |
28b39bc7 | 762 | |
f600b690 JF |
763 | head_skb = page_to_skb(vi, rq, page, offset, len, truesize); |
764 | curr_skb = head_skb; | |
9ab86bbc | 765 | |
8fc3b9e9 MT |
766 | if (unlikely(!curr_skb)) |
767 | goto err_skb; | |
9ab86bbc | 768 | while (--num_buf) { |
8fc3b9e9 MT |
769 | int num_skb_frags; |
770 | ||
680557cf | 771 | buf = virtqueue_get_buf_ctx(rq->vq, &len, &ctx); |
03e9f8a0 | 772 | if (unlikely(!buf)) { |
8fc3b9e9 | 773 | pr_debug("%s: rx error: %d buffers out of %d missing\n", |
fdd819b2 | 774 | dev->name, num_buf, |
012873d0 MT |
775 | virtio16_to_cpu(vi->vdev, |
776 | hdr->num_buffers)); | |
8fc3b9e9 MT |
777 | dev->stats.rx_length_errors++; |
778 | goto err_buf; | |
3f2c31d9 | 779 | } |
8fc3b9e9 MT |
780 | |
781 | page = virt_to_head_page(buf); | |
28b39bc7 JW |
782 | |
783 | truesize = mergeable_ctx_to_truesize(ctx); | |
784 | if (unlikely(len > truesize)) { | |
56da5fd0 | 785 | pr_debug("%s: rx error: len %u exceeds truesize %lu\n", |
680557cf MT |
786 | dev->name, len, (unsigned long)ctx); |
787 | dev->stats.rx_length_errors++; | |
788 | goto err_skb; | |
789 | } | |
8fc3b9e9 MT |
790 | |
791 | num_skb_frags = skb_shinfo(curr_skb)->nr_frags; | |
2613af0e MD |
792 | if (unlikely(num_skb_frags == MAX_SKB_FRAGS)) { |
793 | struct sk_buff *nskb = alloc_skb(0, GFP_ATOMIC); | |
8fc3b9e9 MT |
794 | |
795 | if (unlikely(!nskb)) | |
796 | goto err_skb; | |
2613af0e MD |
797 | if (curr_skb == head_skb) |
798 | skb_shinfo(curr_skb)->frag_list = nskb; | |
799 | else | |
800 | curr_skb->next = nskb; | |
801 | curr_skb = nskb; | |
802 | head_skb->truesize += nskb->truesize; | |
803 | num_skb_frags = 0; | |
804 | } | |
805 | if (curr_skb != head_skb) { | |
806 | head_skb->data_len += len; | |
807 | head_skb->len += len; | |
fb51879d | 808 | head_skb->truesize += truesize; |
2613af0e | 809 | } |
8fc3b9e9 | 810 | offset = buf - page_address(page); |
ba275241 JW |
811 | if (skb_can_coalesce(curr_skb, num_skb_frags, page, offset)) { |
812 | put_page(page); | |
813 | skb_coalesce_rx_frag(curr_skb, num_skb_frags - 1, | |
fb51879d | 814 | len, truesize); |
ba275241 JW |
815 | } else { |
816 | skb_add_rx_frag(curr_skb, num_skb_frags, page, | |
fb51879d | 817 | offset, len, truesize); |
ba275241 | 818 | } |
8fc3b9e9 MT |
819 | } |
820 | ||
5377d758 | 821 | ewma_pkt_len_add(&rq->mrg_avg_pkt_len, head_skb->len); |
8fc3b9e9 MT |
822 | return head_skb; |
823 | ||
f600b690 JF |
824 | err_xdp: |
825 | rcu_read_unlock(); | |
8fc3b9e9 MT |
826 | err_skb: |
827 | put_page(page); | |
26f1676b | 828 | while (num_buf-- > 1) { |
680557cf MT |
829 | buf = virtqueue_get_buf(rq->vq, &len); |
830 | if (unlikely(!buf)) { | |
8fc3b9e9 MT |
831 | pr_debug("%s: rx error: %d buffers missing\n", |
832 | dev->name, num_buf); | |
833 | dev->stats.rx_length_errors++; | |
834 | break; | |
835 | } | |
680557cf | 836 | page = virt_to_head_page(buf); |
8fc3b9e9 | 837 | put_page(page); |
9ab86bbc | 838 | } |
8fc3b9e9 MT |
839 | err_buf: |
840 | dev->stats.rx_dropped++; | |
841 | dev_kfree_skb(head_skb); | |
56434a01 | 842 | xdp_xmit: |
8fc3b9e9 | 843 | return NULL; |
9ab86bbc SM |
844 | } |
845 | ||
61845d20 | 846 | static int receive_buf(struct virtnet_info *vi, struct receive_queue *rq, |
186b3c99 | 847 | void *buf, unsigned int len, void **ctx, bool *xdp_xmit) |
9ab86bbc | 848 | { |
e9d7417b | 849 | struct net_device *dev = vi->dev; |
9ab86bbc | 850 | struct sk_buff *skb; |
012873d0 | 851 | struct virtio_net_hdr_mrg_rxbuf *hdr; |
61845d20 | 852 | int ret; |
3f2c31d9 | 853 | |
bcff3162 | 854 | if (unlikely(len < vi->hdr_len + ETH_HLEN)) { |
9ab86bbc SM |
855 | pr_debug("%s: short packet %i\n", dev->name, len); |
856 | dev->stats.rx_length_errors++; | |
ab7db917 | 857 | if (vi->mergeable_rx_bufs) { |
680557cf | 858 | put_page(virt_to_head_page(buf)); |
ab7db917 | 859 | } else if (vi->big_packets) { |
98bfd23c | 860 | give_pages(rq, buf); |
ab7db917 | 861 | } else { |
f6b10209 | 862 | put_page(virt_to_head_page(buf)); |
ab7db917 | 863 | } |
61845d20 | 864 | return 0; |
9ab86bbc | 865 | } |
3f2c31d9 | 866 | |
f121159d | 867 | if (vi->mergeable_rx_bufs) |
186b3c99 | 868 | skb = receive_mergeable(dev, vi, rq, buf, ctx, len, xdp_xmit); |
f121159d | 869 | else if (vi->big_packets) |
946fa564 | 870 | skb = receive_big(dev, vi, rq, buf, len); |
f121159d | 871 | else |
186b3c99 | 872 | skb = receive_small(dev, vi, rq, buf, ctx, len, xdp_xmit); |
f121159d MT |
873 | |
874 | if (unlikely(!skb)) | |
61845d20 | 875 | return 0; |
3f2c31d9 | 876 | |
9ab86bbc | 877 | hdr = skb_vnet_hdr(skb); |
3fa2a1df | 878 | |
61845d20 | 879 | ret = skb->len; |
296f96fc | 880 | |
e858fae2 | 881 | if (hdr->hdr.flags & VIRTIO_NET_HDR_F_DATA_VALID) |
10a8d94a | 882 | skb->ip_summed = CHECKSUM_UNNECESSARY; |
296f96fc | 883 | |
e858fae2 MR |
884 | if (virtio_net_hdr_to_skb(skb, &hdr->hdr, |
885 | virtio_is_little_endian(vi->vdev))) { | |
886 | net_warn_ratelimited("%s: bad gso: type: %u, size: %u\n", | |
887 | dev->name, hdr->hdr.gso_type, | |
888 | hdr->hdr.gso_size); | |
889 | goto frame_err; | |
296f96fc RR |
890 | } |
891 | ||
d1dc06dc MR |
892 | skb->protocol = eth_type_trans(skb, dev); |
893 | pr_debug("Receiving skb proto 0x%04x len %i type %i\n", | |
894 | ntohs(skb->protocol), skb->len, skb->pkt_type); | |
895 | ||
0fbd050a | 896 | napi_gro_receive(&rq->napi, skb); |
61845d20 | 897 | return ret; |
296f96fc RR |
898 | |
899 | frame_err: | |
900 | dev->stats.rx_frame_errors++; | |
296f96fc | 901 | dev_kfree_skb(skb); |
61845d20 | 902 | return 0; |
296f96fc RR |
903 | } |
904 | ||
192f68cf JW |
905 | /* Unlike mergeable buffers, all buffers are allocated to the |
906 | * same size, except for the headroom. For this reason we do | |
907 | * not need to use mergeable_len_to_ctx here - it is enough | |
908 | * to store the headroom as the context ignoring the truesize. | |
909 | */ | |
946fa564 MT |
910 | static int add_recvbuf_small(struct virtnet_info *vi, struct receive_queue *rq, |
911 | gfp_t gfp) | |
296f96fc | 912 | { |
f6b10209 JW |
913 | struct page_frag *alloc_frag = &rq->alloc_frag; |
914 | char *buf; | |
2de2f7f4 | 915 | unsigned int xdp_headroom = virtnet_get_headroom(vi); |
192f68cf | 916 | void *ctx = (void *)(unsigned long)xdp_headroom; |
f6b10209 | 917 | int len = vi->hdr_len + VIRTNET_RX_PAD + GOOD_PACKET_LEN + xdp_headroom; |
9ab86bbc | 918 | int err; |
3f2c31d9 | 919 | |
f6b10209 JW |
920 | len = SKB_DATA_ALIGN(len) + |
921 | SKB_DATA_ALIGN(sizeof(struct skb_shared_info)); | |
922 | if (unlikely(!skb_page_frag_refill(len, alloc_frag, gfp))) | |
9ab86bbc | 923 | return -ENOMEM; |
296f96fc | 924 | |
f6b10209 JW |
925 | buf = (char *)page_address(alloc_frag->page) + alloc_frag->offset; |
926 | get_page(alloc_frag->page); | |
927 | alloc_frag->offset += len; | |
928 | sg_init_one(rq->sg, buf + VIRTNET_RX_PAD + xdp_headroom, | |
929 | vi->hdr_len + GOOD_PACKET_LEN); | |
192f68cf | 930 | err = virtqueue_add_inbuf_ctx(rq->vq, rq->sg, 1, buf, ctx, gfp); |
9ab86bbc | 931 | if (err < 0) |
f6b10209 | 932 | put_page(virt_to_head_page(buf)); |
9ab86bbc SM |
933 | return err; |
934 | } | |
97402b96 | 935 | |
012873d0 MT |
936 | static int add_recvbuf_big(struct virtnet_info *vi, struct receive_queue *rq, |
937 | gfp_t gfp) | |
9ab86bbc | 938 | { |
9ab86bbc SM |
939 | struct page *first, *list = NULL; |
940 | char *p; | |
941 | int i, err, offset; | |
942 | ||
a5835440 RR |
943 | sg_init_table(rq->sg, MAX_SKB_FRAGS + 2); |
944 | ||
e9d7417b | 945 | /* page in rq->sg[MAX_SKB_FRAGS + 1] is list tail */ |
9ab86bbc | 946 | for (i = MAX_SKB_FRAGS + 1; i > 1; --i) { |
e9d7417b | 947 | first = get_a_page(rq, gfp); |
9ab86bbc SM |
948 | if (!first) { |
949 | if (list) | |
e9d7417b | 950 | give_pages(rq, list); |
9ab86bbc | 951 | return -ENOMEM; |
97402b96 | 952 | } |
e9d7417b | 953 | sg_set_buf(&rq->sg[i], page_address(first), PAGE_SIZE); |
97402b96 | 954 | |
9ab86bbc SM |
955 | /* chain new page in list head to match sg */ |
956 | first->private = (unsigned long)list; | |
957 | list = first; | |
958 | } | |
296f96fc | 959 | |
e9d7417b | 960 | first = get_a_page(rq, gfp); |
9ab86bbc | 961 | if (!first) { |
e9d7417b | 962 | give_pages(rq, list); |
9ab86bbc SM |
963 | return -ENOMEM; |
964 | } | |
965 | p = page_address(first); | |
966 | ||
e9d7417b | 967 | /* rq->sg[0], rq->sg[1] share the same page */ |
012873d0 MT |
968 | /* a separated rq->sg[0] for header - required in case !any_header_sg */ |
969 | sg_set_buf(&rq->sg[0], p, vi->hdr_len); | |
9ab86bbc | 970 | |
e9d7417b | 971 | /* rq->sg[1] for data packet, from offset */ |
9ab86bbc | 972 | offset = sizeof(struct padded_vnet_hdr); |
e9d7417b | 973 | sg_set_buf(&rq->sg[1], p + offset, PAGE_SIZE - offset); |
9ab86bbc SM |
974 | |
975 | /* chain first in list head */ | |
976 | first->private = (unsigned long)list; | |
9dc7b9e4 RR |
977 | err = virtqueue_add_inbuf(rq->vq, rq->sg, MAX_SKB_FRAGS + 2, |
978 | first, gfp); | |
9ab86bbc | 979 | if (err < 0) |
e9d7417b | 980 | give_pages(rq, first); |
9ab86bbc SM |
981 | |
982 | return err; | |
296f96fc RR |
983 | } |
984 | ||
d85b758f MT |
985 | static unsigned int get_mergeable_buf_len(struct receive_queue *rq, |
986 | struct ewma_pkt_len *avg_pkt_len) | |
3f2c31d9 | 987 | { |
ab7db917 | 988 | const size_t hdr_len = sizeof(struct virtio_net_hdr_mrg_rxbuf); |
fbf28d78 MD |
989 | unsigned int len; |
990 | ||
5377d758 | 991 | len = hdr_len + clamp_t(unsigned int, ewma_pkt_len_read(avg_pkt_len), |
f0c3192c | 992 | rq->min_buf_len, PAGE_SIZE - hdr_len); |
e377fcc8 | 993 | return ALIGN(len, L1_CACHE_BYTES); |
fbf28d78 MD |
994 | } |
995 | ||
2de2f7f4 JF |
996 | static int add_recvbuf_mergeable(struct virtnet_info *vi, |
997 | struct receive_queue *rq, gfp_t gfp) | |
fbf28d78 | 998 | { |
fb51879d | 999 | struct page_frag *alloc_frag = &rq->alloc_frag; |
2de2f7f4 | 1000 | unsigned int headroom = virtnet_get_headroom(vi); |
fb51879d | 1001 | char *buf; |
680557cf | 1002 | void *ctx; |
3f2c31d9 | 1003 | int err; |
fb51879d | 1004 | unsigned int len, hole; |
3f2c31d9 | 1005 | |
d85b758f | 1006 | len = get_mergeable_buf_len(rq, &rq->mrg_avg_pkt_len); |
2de2f7f4 | 1007 | if (unlikely(!skb_page_frag_refill(len + headroom, alloc_frag, gfp))) |
9ab86bbc | 1008 | return -ENOMEM; |
ab7db917 | 1009 | |
fb51879d | 1010 | buf = (char *)page_address(alloc_frag->page) + alloc_frag->offset; |
2de2f7f4 | 1011 | buf += headroom; /* advance address leaving hole at front of pkt */ |
fb51879d | 1012 | get_page(alloc_frag->page); |
2de2f7f4 | 1013 | alloc_frag->offset += len + headroom; |
fb51879d | 1014 | hole = alloc_frag->size - alloc_frag->offset; |
2de2f7f4 | 1015 | if (hole < len + headroom) { |
ab7db917 MD |
1016 | /* To avoid internal fragmentation, if there is very likely not |
1017 | * enough space for another buffer, add the remaining space to | |
1daa8790 | 1018 | * the current buffer. |
ab7db917 | 1019 | */ |
fb51879d MD |
1020 | len += hole; |
1021 | alloc_frag->offset += hole; | |
1022 | } | |
3f2c31d9 | 1023 | |
fb51879d | 1024 | sg_init_one(rq->sg, buf, len); |
29fda25a | 1025 | ctx = mergeable_len_to_ctx(len, headroom); |
680557cf | 1026 | err = virtqueue_add_inbuf_ctx(rq->vq, rq->sg, 1, buf, ctx, gfp); |
9ab86bbc | 1027 | if (err < 0) |
2613af0e | 1028 | put_page(virt_to_head_page(buf)); |
3f2c31d9 | 1029 | |
9ab86bbc SM |
1030 | return err; |
1031 | } | |
3f2c31d9 | 1032 | |
b2baed69 RR |
1033 | /* |
1034 | * Returns false if we couldn't fill entirely (OOM). | |
1035 | * | |
1036 | * Normally run in the receive path, but can also be run from ndo_open | |
1037 | * before we're receiving packets, or from refill_work which is | |
1038 | * careful to disable receiving (using napi_disable). | |
1039 | */ | |
946fa564 MT |
1040 | static bool try_fill_recv(struct virtnet_info *vi, struct receive_queue *rq, |
1041 | gfp_t gfp) | |
9ab86bbc SM |
1042 | { |
1043 | int err; | |
1788f495 | 1044 | bool oom; |
3f2c31d9 | 1045 | |
9ab86bbc SM |
1046 | do { |
1047 | if (vi->mergeable_rx_bufs) | |
2de2f7f4 | 1048 | err = add_recvbuf_mergeable(vi, rq, gfp); |
9ab86bbc | 1049 | else if (vi->big_packets) |
012873d0 | 1050 | err = add_recvbuf_big(vi, rq, gfp); |
9ab86bbc | 1051 | else |
946fa564 | 1052 | err = add_recvbuf_small(vi, rq, gfp); |
3f2c31d9 | 1053 | |
1788f495 | 1054 | oom = err == -ENOMEM; |
9ed4cb07 | 1055 | if (err) |
3f2c31d9 | 1056 | break; |
b7dfde95 | 1057 | } while (rq->vq->num_free); |
681daee2 | 1058 | virtqueue_kick(rq->vq); |
3161e453 | 1059 | return !oom; |
3f2c31d9 MM |
1060 | } |
1061 | ||
18445c4d | 1062 | static void skb_recv_done(struct virtqueue *rvq) |
296f96fc RR |
1063 | { |
1064 | struct virtnet_info *vi = rvq->vdev->priv; | |
986a4f4d | 1065 | struct receive_queue *rq = &vi->rq[vq2rxq(rvq)]; |
e9d7417b | 1066 | |
e4e8452a | 1067 | virtqueue_napi_schedule(&rq->napi, rvq); |
296f96fc RR |
1068 | } |
1069 | ||
e4e8452a | 1070 | static void virtnet_napi_enable(struct virtqueue *vq, struct napi_struct *napi) |
3e9d08ec | 1071 | { |
e4e8452a | 1072 | napi_enable(napi); |
3e9d08ec BR |
1073 | |
1074 | /* If all buffers were filled by other side before we napi_enabled, we | |
e4e8452a WB |
1075 | * won't get another interrupt, so process any outstanding packets now. |
1076 | * Call local_bh_enable after to trigger softIRQ processing. | |
1077 | */ | |
1078 | local_bh_disable(); | |
1079 | virtqueue_napi_schedule(napi, vq); | |
1080 | local_bh_enable(); | |
3e9d08ec BR |
1081 | } |
1082 | ||
b92f1e67 WB |
1083 | static void virtnet_napi_tx_enable(struct virtnet_info *vi, |
1084 | struct virtqueue *vq, | |
1085 | struct napi_struct *napi) | |
1086 | { | |
1087 | if (!napi->weight) | |
1088 | return; | |
1089 | ||
1090 | /* Tx napi touches cachelines on the cpu handling tx interrupts. Only | |
1091 | * enable the feature if this is likely affine with the transmit path. | |
1092 | */ | |
1093 | if (!vi->affinity_hint_set) { | |
1094 | napi->weight = 0; | |
1095 | return; | |
1096 | } | |
1097 | ||
1098 | return virtnet_napi_enable(vq, napi); | |
1099 | } | |
1100 | ||
78a57b48 WB |
1101 | static void virtnet_napi_tx_disable(struct napi_struct *napi) |
1102 | { | |
1103 | if (napi->weight) | |
1104 | napi_disable(napi); | |
1105 | } | |
1106 | ||
3161e453 RR |
1107 | static void refill_work(struct work_struct *work) |
1108 | { | |
e9d7417b JW |
1109 | struct virtnet_info *vi = |
1110 | container_of(work, struct virtnet_info, refill.work); | |
3161e453 | 1111 | bool still_empty; |
986a4f4d JW |
1112 | int i; |
1113 | ||
55257d72 | 1114 | for (i = 0; i < vi->curr_queue_pairs; i++) { |
986a4f4d | 1115 | struct receive_queue *rq = &vi->rq[i]; |
3161e453 | 1116 | |
986a4f4d | 1117 | napi_disable(&rq->napi); |
946fa564 | 1118 | still_empty = !try_fill_recv(vi, rq, GFP_KERNEL); |
e4e8452a | 1119 | virtnet_napi_enable(rq->vq, &rq->napi); |
3161e453 | 1120 | |
986a4f4d JW |
1121 | /* In theory, this can happen: if we don't get any buffers in |
1122 | * we will *never* try to fill again. | |
1123 | */ | |
1124 | if (still_empty) | |
1125 | schedule_delayed_work(&vi->refill, HZ/2); | |
1126 | } | |
3161e453 RR |
1127 | } |
1128 | ||
186b3c99 | 1129 | static int virtnet_receive(struct receive_queue *rq, int budget, bool *xdp_xmit) |
296f96fc | 1130 | { |
e9d7417b | 1131 | struct virtnet_info *vi = rq->vq->vdev->priv; |
61845d20 | 1132 | unsigned int len, received = 0, bytes = 0; |
9ab86bbc | 1133 | void *buf; |
61845d20 | 1134 | struct virtnet_stats *stats = this_cpu_ptr(vi->stats); |
296f96fc | 1135 | |
192f68cf | 1136 | if (!vi->big_packets || vi->mergeable_rx_bufs) { |
680557cf MT |
1137 | void *ctx; |
1138 | ||
1139 | while (received < budget && | |
1140 | (buf = virtqueue_get_buf_ctx(rq->vq, &len, &ctx))) { | |
186b3c99 | 1141 | bytes += receive_buf(vi, rq, buf, len, ctx, xdp_xmit); |
680557cf MT |
1142 | received++; |
1143 | } | |
1144 | } else { | |
1145 | while (received < budget && | |
1146 | (buf = virtqueue_get_buf(rq->vq, &len)) != NULL) { | |
186b3c99 | 1147 | bytes += receive_buf(vi, rq, buf, len, NULL, xdp_xmit); |
680557cf MT |
1148 | received++; |
1149 | } | |
296f96fc RR |
1150 | } |
1151 | ||
be121f46 | 1152 | if (rq->vq->num_free > virtqueue_get_vring_size(rq->vq) / 2) { |
946fa564 | 1153 | if (!try_fill_recv(vi, rq, GFP_ATOMIC)) |
3b07e9ca | 1154 | schedule_delayed_work(&vi->refill, 0); |
3161e453 | 1155 | } |
296f96fc | 1156 | |
61845d20 JW |
1157 | u64_stats_update_begin(&stats->rx_syncp); |
1158 | stats->rx_bytes += bytes; | |
1159 | stats->rx_packets += received; | |
1160 | u64_stats_update_end(&stats->rx_syncp); | |
1161 | ||
2ffa7598 JW |
1162 | return received; |
1163 | } | |
1164 | ||
ea7735d9 WB |
1165 | static void free_old_xmit_skbs(struct send_queue *sq) |
1166 | { | |
1167 | struct sk_buff *skb; | |
1168 | unsigned int len; | |
1169 | struct virtnet_info *vi = sq->vq->vdev->priv; | |
1170 | struct virtnet_stats *stats = this_cpu_ptr(vi->stats); | |
1171 | unsigned int packets = 0; | |
1172 | unsigned int bytes = 0; | |
1173 | ||
1174 | while ((skb = virtqueue_get_buf(sq->vq, &len)) != NULL) { | |
1175 | pr_debug("Sent skb %p\n", skb); | |
1176 | ||
1177 | bytes += skb->len; | |
1178 | packets++; | |
1179 | ||
dadc0736 | 1180 | dev_consume_skb_any(skb); |
ea7735d9 WB |
1181 | } |
1182 | ||
1183 | /* Avoid overhead when no packets have been processed | |
1184 | * happens when called speculatively from start_xmit. | |
1185 | */ | |
1186 | if (!packets) | |
1187 | return; | |
1188 | ||
1189 | u64_stats_update_begin(&stats->tx_syncp); | |
1190 | stats->tx_bytes += bytes; | |
1191 | stats->tx_packets += packets; | |
1192 | u64_stats_update_end(&stats->tx_syncp); | |
1193 | } | |
1194 | ||
7b0411ef WB |
1195 | static void virtnet_poll_cleantx(struct receive_queue *rq) |
1196 | { | |
1197 | struct virtnet_info *vi = rq->vq->vdev->priv; | |
1198 | unsigned int index = vq2rxq(rq->vq); | |
1199 | struct send_queue *sq = &vi->sq[index]; | |
1200 | struct netdev_queue *txq = netdev_get_tx_queue(vi->dev, index); | |
1201 | ||
1202 | if (!sq->napi.weight) | |
1203 | return; | |
1204 | ||
1205 | if (__netif_tx_trylock(txq)) { | |
1206 | free_old_xmit_skbs(sq); | |
1207 | __netif_tx_unlock(txq); | |
1208 | } | |
1209 | ||
1210 | if (sq->vq->num_free >= 2 + MAX_SKB_FRAGS) | |
1211 | netif_tx_wake_queue(txq); | |
1212 | } | |
1213 | ||
2ffa7598 JW |
1214 | static int virtnet_poll(struct napi_struct *napi, int budget) |
1215 | { | |
1216 | struct receive_queue *rq = | |
1217 | container_of(napi, struct receive_queue, napi); | |
3c93c06c JW |
1218 | struct virtnet_info *vi = rq->vq->vdev->priv; |
1219 | struct send_queue *sq; | |
1220 | unsigned int received, qp; | |
186b3c99 | 1221 | bool xdp_xmit = false; |
2ffa7598 | 1222 | |
7b0411ef WB |
1223 | virtnet_poll_cleantx(rq); |
1224 | ||
186b3c99 | 1225 | received = virtnet_receive(rq, budget, &xdp_xmit); |
2ffa7598 | 1226 | |
8329d98e | 1227 | /* Out of packets? */ |
e4e8452a WB |
1228 | if (received < budget) |
1229 | virtqueue_napi_complete(napi, rq->vq, received); | |
296f96fc | 1230 | |
3c93c06c JW |
1231 | if (xdp_xmit) { |
1232 | qp = vi->curr_queue_pairs - vi->xdp_queue_pairs + | |
1233 | smp_processor_id(); | |
1234 | sq = &vi->sq[qp]; | |
1235 | virtqueue_kick(sq->vq); | |
186b3c99 | 1236 | xdp_do_flush_map(); |
3c93c06c | 1237 | } |
186b3c99 | 1238 | |
296f96fc RR |
1239 | return received; |
1240 | } | |
1241 | ||
986a4f4d JW |
1242 | static int virtnet_open(struct net_device *dev) |
1243 | { | |
1244 | struct virtnet_info *vi = netdev_priv(dev); | |
1245 | int i; | |
1246 | ||
e4166625 JW |
1247 | for (i = 0; i < vi->max_queue_pairs; i++) { |
1248 | if (i < vi->curr_queue_pairs) | |
1249 | /* Make sure we have some buffers: if oom use wq. */ | |
946fa564 | 1250 | if (!try_fill_recv(vi, &vi->rq[i], GFP_KERNEL)) |
e4166625 | 1251 | schedule_delayed_work(&vi->refill, 0); |
e4e8452a | 1252 | virtnet_napi_enable(vi->rq[i].vq, &vi->rq[i].napi); |
b92f1e67 | 1253 | virtnet_napi_tx_enable(vi, vi->sq[i].vq, &vi->sq[i].napi); |
986a4f4d JW |
1254 | } |
1255 | ||
1256 | return 0; | |
1257 | } | |
1258 | ||
b92f1e67 WB |
1259 | static int virtnet_poll_tx(struct napi_struct *napi, int budget) |
1260 | { | |
1261 | struct send_queue *sq = container_of(napi, struct send_queue, napi); | |
1262 | struct virtnet_info *vi = sq->vq->vdev->priv; | |
1263 | struct netdev_queue *txq = netdev_get_tx_queue(vi->dev, vq2txq(sq->vq)); | |
1264 | ||
1265 | __netif_tx_lock(txq, raw_smp_processor_id()); | |
1266 | free_old_xmit_skbs(sq); | |
1267 | __netif_tx_unlock(txq); | |
1268 | ||
1269 | virtqueue_napi_complete(napi, sq->vq, 0); | |
1270 | ||
1271 | if (sq->vq->num_free >= 2 + MAX_SKB_FRAGS) | |
1272 | netif_tx_wake_queue(txq); | |
1273 | ||
1274 | return 0; | |
1275 | } | |
1276 | ||
e9d7417b | 1277 | static int xmit_skb(struct send_queue *sq, struct sk_buff *skb) |
296f96fc | 1278 | { |
012873d0 | 1279 | struct virtio_net_hdr_mrg_rxbuf *hdr; |
296f96fc | 1280 | const unsigned char *dest = ((struct ethhdr *)skb->data)->h_dest; |
e9d7417b | 1281 | struct virtnet_info *vi = sq->vq->vdev->priv; |
e2fcad58 | 1282 | int num_sg; |
012873d0 | 1283 | unsigned hdr_len = vi->hdr_len; |
e7428e95 | 1284 | bool can_push; |
296f96fc | 1285 | |
e174961c | 1286 | pr_debug("%s: xmit %p %pM\n", vi->dev->name, skb, dest); |
e7428e95 MT |
1287 | |
1288 | can_push = vi->any_header_sg && | |
1289 | !((unsigned long)skb->data & (__alignof__(*hdr) - 1)) && | |
1290 | !skb_header_cloned(skb) && skb_headroom(skb) >= hdr_len; | |
1291 | /* Even if we can, don't push here yet as this would skew | |
1292 | * csum_start offset below. */ | |
1293 | if (can_push) | |
012873d0 | 1294 | hdr = (struct virtio_net_hdr_mrg_rxbuf *)(skb->data - hdr_len); |
e7428e95 MT |
1295 | else |
1296 | hdr = skb_vnet_hdr(skb); | |
296f96fc | 1297 | |
e858fae2 | 1298 | if (virtio_net_hdr_from_skb(skb, &hdr->hdr, |
34d736aa WB |
1299 | virtio_is_little_endian(vi->vdev), false, |
1300 | 0)) | |
e858fae2 | 1301 | BUG(); |
296f96fc | 1302 | |
3f2c31d9 | 1303 | if (vi->mergeable_rx_bufs) |
012873d0 | 1304 | hdr->num_buffers = 0; |
3f2c31d9 | 1305 | |
547c890c | 1306 | sg_init_table(sq->sg, skb_shinfo(skb)->nr_frags + (can_push ? 1 : 2)); |
e7428e95 MT |
1307 | if (can_push) { |
1308 | __skb_push(skb, hdr_len); | |
1309 | num_sg = skb_to_sgvec(skb, sq->sg, 0, skb->len); | |
e2fcad58 JD |
1310 | if (unlikely(num_sg < 0)) |
1311 | return num_sg; | |
e7428e95 MT |
1312 | /* Pull header back to avoid skew in tx bytes calculations. */ |
1313 | __skb_pull(skb, hdr_len); | |
1314 | } else { | |
1315 | sg_set_buf(sq->sg, hdr, hdr_len); | |
e2fcad58 JD |
1316 | num_sg = skb_to_sgvec(skb, sq->sg + 1, 0, skb->len); |
1317 | if (unlikely(num_sg < 0)) | |
1318 | return num_sg; | |
1319 | num_sg++; | |
e7428e95 | 1320 | } |
9dc7b9e4 | 1321 | return virtqueue_add_outbuf(sq->vq, sq->sg, num_sg, skb, GFP_ATOMIC); |
11a3a154 RR |
1322 | } |
1323 | ||
424efe9c | 1324 | static netdev_tx_t start_xmit(struct sk_buff *skb, struct net_device *dev) |
99ffc696 RR |
1325 | { |
1326 | struct virtnet_info *vi = netdev_priv(dev); | |
986a4f4d JW |
1327 | int qnum = skb_get_queue_mapping(skb); |
1328 | struct send_queue *sq = &vi->sq[qnum]; | |
9ed4cb07 | 1329 | int err; |
4b7fd2e6 MT |
1330 | struct netdev_queue *txq = netdev_get_tx_queue(dev, qnum); |
1331 | bool kick = !skb->xmit_more; | |
b92f1e67 | 1332 | bool use_napi = sq->napi.weight; |
2cb9c6ba | 1333 | |
2cb9c6ba | 1334 | /* Free up any pending old buffers before queueing new ones. */ |
e9d7417b | 1335 | free_old_xmit_skbs(sq); |
99ffc696 | 1336 | |
bdb12e0d WB |
1337 | if (use_napi && kick) |
1338 | virtqueue_enable_cb_delayed(sq->vq); | |
1339 | ||
074c3582 JK |
1340 | /* timestamp packet in software */ |
1341 | skb_tx_timestamp(skb); | |
1342 | ||
03f191ba | 1343 | /* Try to transmit */ |
b7dfde95 | 1344 | err = xmit_skb(sq, skb); |
48925e37 | 1345 | |
9ed4cb07 | 1346 | /* This should not happen! */ |
681daee2 | 1347 | if (unlikely(err)) { |
9ed4cb07 RR |
1348 | dev->stats.tx_fifo_errors++; |
1349 | if (net_ratelimit()) | |
1350 | dev_warn(&dev->dev, | |
b7dfde95 | 1351 | "Unexpected TXQ (%d) queue failure: %d\n", qnum, err); |
58eba97d | 1352 | dev->stats.tx_dropped++; |
85e94525 | 1353 | dev_kfree_skb_any(skb); |
58eba97d | 1354 | return NETDEV_TX_OK; |
296f96fc | 1355 | } |
03f191ba | 1356 | |
48925e37 | 1357 | /* Don't wait up for transmitted skbs to be freed. */ |
b92f1e67 WB |
1358 | if (!use_napi) { |
1359 | skb_orphan(skb); | |
1360 | nf_reset(skb); | |
1361 | } | |
48925e37 | 1362 | |
60302ff6 MT |
1363 | /* If running out of space, stop queue to avoid getting packets that we |
1364 | * are then unable to transmit. | |
1365 | * An alternative would be to force queuing layer to requeue the skb by | |
1366 | * returning NETDEV_TX_BUSY. However, NETDEV_TX_BUSY should not be | |
1367 | * returned in a normal path of operation: it means that driver is not | |
1368 | * maintaining the TX queue stop/start state properly, and causes | |
1369 | * the stack to do a non-trivial amount of useless work. | |
1370 | * Since most packets only take 1 or 2 ring slots, stopping the queue | |
1371 | * early means 16 slots are typically wasted. | |
d631b94e | 1372 | */ |
b7dfde95 | 1373 | if (sq->vq->num_free < 2+MAX_SKB_FRAGS) { |
986a4f4d | 1374 | netif_stop_subqueue(dev, qnum); |
b92f1e67 WB |
1375 | if (!use_napi && |
1376 | unlikely(!virtqueue_enable_cb_delayed(sq->vq))) { | |
48925e37 | 1377 | /* More just got used, free them then recheck. */ |
b7dfde95 LT |
1378 | free_old_xmit_skbs(sq); |
1379 | if (sq->vq->num_free >= 2+MAX_SKB_FRAGS) { | |
986a4f4d | 1380 | netif_start_subqueue(dev, qnum); |
e9d7417b | 1381 | virtqueue_disable_cb(sq->vq); |
48925e37 RR |
1382 | } |
1383 | } | |
99ffc696 | 1384 | } |
48925e37 | 1385 | |
4b7fd2e6 | 1386 | if (kick || netif_xmit_stopped(txq)) |
0b725a2c | 1387 | virtqueue_kick(sq->vq); |
296f96fc | 1388 | |
0b725a2c | 1389 | return NETDEV_TX_OK; |
c223a078 DM |
1390 | } |
1391 | ||
40cbfc37 AK |
1392 | /* |
1393 | * Send command via the control virtqueue and check status. Commands | |
1394 | * supported by the hypervisor, as indicated by feature bits, should | |
788a8b6d | 1395 | * never fail unless improperly formatted. |
40cbfc37 AK |
1396 | */ |
1397 | static bool virtnet_send_command(struct virtnet_info *vi, u8 class, u8 cmd, | |
d24bae32 | 1398 | struct scatterlist *out) |
40cbfc37 | 1399 | { |
f7bc9594 | 1400 | struct scatterlist *sgs[4], hdr, stat; |
d24bae32 | 1401 | unsigned out_num = 0, tmp; |
40cbfc37 AK |
1402 | |
1403 | /* Caller should know better */ | |
f7bc9594 | 1404 | BUG_ON(!virtio_has_feature(vi->vdev, VIRTIO_NET_F_CTRL_VQ)); |
40cbfc37 | 1405 | |
603343b9 MT |
1406 | vi->ctrl->status = ~0; |
1407 | vi->ctrl->hdr.class = class; | |
1408 | vi->ctrl->hdr.cmd = cmd; | |
f7bc9594 | 1409 | /* Add header */ |
603343b9 | 1410 | sg_init_one(&hdr, &vi->ctrl->hdr, sizeof(vi->ctrl->hdr)); |
f7bc9594 | 1411 | sgs[out_num++] = &hdr; |
40cbfc37 | 1412 | |
f7bc9594 RR |
1413 | if (out) |
1414 | sgs[out_num++] = out; | |
40cbfc37 | 1415 | |
f7bc9594 | 1416 | /* Add return status. */ |
603343b9 | 1417 | sg_init_one(&stat, &vi->ctrl->status, sizeof(vi->ctrl->status)); |
d24bae32 | 1418 | sgs[out_num] = &stat; |
40cbfc37 | 1419 | |
d24bae32 | 1420 | BUG_ON(out_num + 1 > ARRAY_SIZE(sgs)); |
a7c58146 | 1421 | virtqueue_add_sgs(vi->cvq, sgs, out_num, 1, vi, GFP_ATOMIC); |
40cbfc37 | 1422 | |
67975901 | 1423 | if (unlikely(!virtqueue_kick(vi->cvq))) |
603343b9 | 1424 | return vi->ctrl->status == VIRTIO_NET_OK; |
40cbfc37 AK |
1425 | |
1426 | /* Spin for a response, the kick causes an ioport write, trapping | |
1427 | * into the hypervisor, so the request should be handled immediately. | |
1428 | */ | |
047b9b94 HG |
1429 | while (!virtqueue_get_buf(vi->cvq, &tmp) && |
1430 | !virtqueue_is_broken(vi->cvq)) | |
40cbfc37 AK |
1431 | cpu_relax(); |
1432 | ||
603343b9 | 1433 | return vi->ctrl->status == VIRTIO_NET_OK; |
40cbfc37 AK |
1434 | } |
1435 | ||
9c46f6d4 AW |
1436 | static int virtnet_set_mac_address(struct net_device *dev, void *p) |
1437 | { | |
1438 | struct virtnet_info *vi = netdev_priv(dev); | |
1439 | struct virtio_device *vdev = vi->vdev; | |
f2f2c8b4 | 1440 | int ret; |
e37e2ff3 | 1441 | struct sockaddr *addr; |
7e58d5ae | 1442 | struct scatterlist sg; |
9c46f6d4 | 1443 | |
801822d1 | 1444 | addr = kmemdup(p, sizeof(*addr), GFP_KERNEL); |
e37e2ff3 AL |
1445 | if (!addr) |
1446 | return -ENOMEM; | |
e37e2ff3 AL |
1447 | |
1448 | ret = eth_prepare_mac_addr_change(dev, addr); | |
f2f2c8b4 | 1449 | if (ret) |
e37e2ff3 | 1450 | goto out; |
9c46f6d4 | 1451 | |
7e58d5ae AK |
1452 | if (virtio_has_feature(vdev, VIRTIO_NET_F_CTRL_MAC_ADDR)) { |
1453 | sg_init_one(&sg, addr->sa_data, dev->addr_len); | |
1454 | if (!virtnet_send_command(vi, VIRTIO_NET_CTRL_MAC, | |
d24bae32 | 1455 | VIRTIO_NET_CTRL_MAC_ADDR_SET, &sg)) { |
7e58d5ae AK |
1456 | dev_warn(&vdev->dev, |
1457 | "Failed to set mac address by vq command.\n"); | |
e37e2ff3 AL |
1458 | ret = -EINVAL; |
1459 | goto out; | |
7e58d5ae | 1460 | } |
7e93a02f MT |
1461 | } else if (virtio_has_feature(vdev, VIRTIO_NET_F_MAC) && |
1462 | !virtio_has_feature(vdev, VIRTIO_F_VERSION_1)) { | |
855e0c52 RR |
1463 | unsigned int i; |
1464 | ||
1465 | /* Naturally, this has an atomicity problem. */ | |
1466 | for (i = 0; i < dev->addr_len; i++) | |
1467 | virtio_cwrite8(vdev, | |
1468 | offsetof(struct virtio_net_config, mac) + | |
1469 | i, addr->sa_data[i]); | |
7e58d5ae AK |
1470 | } |
1471 | ||
1472 | eth_commit_mac_addr_change(dev, p); | |
e37e2ff3 | 1473 | ret = 0; |
9c46f6d4 | 1474 | |
e37e2ff3 AL |
1475 | out: |
1476 | kfree(addr); | |
1477 | return ret; | |
9c46f6d4 AW |
1478 | } |
1479 | ||
bc1f4470 | 1480 | static void virtnet_stats(struct net_device *dev, |
1481 | struct rtnl_link_stats64 *tot) | |
3fa2a1df | 1482 | { |
1483 | struct virtnet_info *vi = netdev_priv(dev); | |
1484 | int cpu; | |
1485 | unsigned int start; | |
1486 | ||
1487 | for_each_possible_cpu(cpu) { | |
58472a76 | 1488 | struct virtnet_stats *stats = per_cpu_ptr(vi->stats, cpu); |
3fa2a1df | 1489 | u64 tpackets, tbytes, rpackets, rbytes; |
1490 | ||
1491 | do { | |
57a7744e | 1492 | start = u64_stats_fetch_begin_irq(&stats->tx_syncp); |
3fa2a1df | 1493 | tpackets = stats->tx_packets; |
1494 | tbytes = stats->tx_bytes; | |
57a7744e | 1495 | } while (u64_stats_fetch_retry_irq(&stats->tx_syncp, start)); |
83a27052 ED |
1496 | |
1497 | do { | |
57a7744e | 1498 | start = u64_stats_fetch_begin_irq(&stats->rx_syncp); |
3fa2a1df | 1499 | rpackets = stats->rx_packets; |
1500 | rbytes = stats->rx_bytes; | |
57a7744e | 1501 | } while (u64_stats_fetch_retry_irq(&stats->rx_syncp, start)); |
3fa2a1df | 1502 | |
1503 | tot->rx_packets += rpackets; | |
1504 | tot->tx_packets += tpackets; | |
1505 | tot->rx_bytes += rbytes; | |
1506 | tot->tx_bytes += tbytes; | |
1507 | } | |
1508 | ||
1509 | tot->tx_dropped = dev->stats.tx_dropped; | |
021ac8d3 | 1510 | tot->tx_fifo_errors = dev->stats.tx_fifo_errors; |
3fa2a1df | 1511 | tot->rx_dropped = dev->stats.rx_dropped; |
1512 | tot->rx_length_errors = dev->stats.rx_length_errors; | |
1513 | tot->rx_frame_errors = dev->stats.rx_frame_errors; | |
3fa2a1df | 1514 | } |
1515 | ||
da74e89d AS |
1516 | #ifdef CONFIG_NET_POLL_CONTROLLER |
1517 | static void virtnet_netpoll(struct net_device *dev) | |
1518 | { | |
1519 | struct virtnet_info *vi = netdev_priv(dev); | |
986a4f4d | 1520 | int i; |
da74e89d | 1521 | |
986a4f4d JW |
1522 | for (i = 0; i < vi->curr_queue_pairs; i++) |
1523 | napi_schedule(&vi->rq[i].napi); | |
da74e89d AS |
1524 | } |
1525 | #endif | |
1526 | ||
586d17c5 JW |
1527 | static void virtnet_ack_link_announce(struct virtnet_info *vi) |
1528 | { | |
1529 | rtnl_lock(); | |
1530 | if (!virtnet_send_command(vi, VIRTIO_NET_CTRL_ANNOUNCE, | |
d24bae32 | 1531 | VIRTIO_NET_CTRL_ANNOUNCE_ACK, NULL)) |
586d17c5 JW |
1532 | dev_warn(&vi->dev->dev, "Failed to ack link announce.\n"); |
1533 | rtnl_unlock(); | |
1534 | } | |
1535 | ||
47315329 | 1536 | static int _virtnet_set_queues(struct virtnet_info *vi, u16 queue_pairs) |
986a4f4d JW |
1537 | { |
1538 | struct scatterlist sg; | |
986a4f4d JW |
1539 | struct net_device *dev = vi->dev; |
1540 | ||
1541 | if (!vi->has_cvq || !virtio_has_feature(vi->vdev, VIRTIO_NET_F_MQ)) | |
1542 | return 0; | |
1543 | ||
603343b9 MT |
1544 | vi->ctrl->mq.virtqueue_pairs = cpu_to_virtio16(vi->vdev, queue_pairs); |
1545 | sg_init_one(&sg, &vi->ctrl->mq, sizeof(vi->ctrl->mq)); | |
986a4f4d JW |
1546 | |
1547 | if (!virtnet_send_command(vi, VIRTIO_NET_CTRL_MQ, | |
d24bae32 | 1548 | VIRTIO_NET_CTRL_MQ_VQ_PAIRS_SET, &sg)) { |
986a4f4d JW |
1549 | dev_warn(&dev->dev, "Fail to set num of queue pairs to %d\n", |
1550 | queue_pairs); | |
1551 | return -EINVAL; | |
55257d72 | 1552 | } else { |
986a4f4d | 1553 | vi->curr_queue_pairs = queue_pairs; |
35ed159b JW |
1554 | /* virtnet_open() will refill when device is going to up. */ |
1555 | if (dev->flags & IFF_UP) | |
1556 | schedule_delayed_work(&vi->refill, 0); | |
55257d72 | 1557 | } |
986a4f4d JW |
1558 | |
1559 | return 0; | |
1560 | } | |
1561 | ||
47315329 JF |
1562 | static int virtnet_set_queues(struct virtnet_info *vi, u16 queue_pairs) |
1563 | { | |
1564 | int err; | |
1565 | ||
1566 | rtnl_lock(); | |
1567 | err = _virtnet_set_queues(vi, queue_pairs); | |
1568 | rtnl_unlock(); | |
1569 | return err; | |
1570 | } | |
1571 | ||
296f96fc RR |
1572 | static int virtnet_close(struct net_device *dev) |
1573 | { | |
1574 | struct virtnet_info *vi = netdev_priv(dev); | |
986a4f4d | 1575 | int i; |
296f96fc | 1576 | |
b2baed69 RR |
1577 | /* Make sure refill_work doesn't re-enable napi! */ |
1578 | cancel_delayed_work_sync(&vi->refill); | |
986a4f4d | 1579 | |
b92f1e67 | 1580 | for (i = 0; i < vi->max_queue_pairs; i++) { |
986a4f4d | 1581 | napi_disable(&vi->rq[i].napi); |
78a57b48 | 1582 | virtnet_napi_tx_disable(&vi->sq[i].napi); |
b92f1e67 | 1583 | } |
296f96fc | 1584 | |
296f96fc RR |
1585 | return 0; |
1586 | } | |
1587 | ||
2af7698e AW |
1588 | static void virtnet_set_rx_mode(struct net_device *dev) |
1589 | { | |
1590 | struct virtnet_info *vi = netdev_priv(dev); | |
f565a7c2 | 1591 | struct scatterlist sg[2]; |
f565a7c2 | 1592 | struct virtio_net_ctrl_mac *mac_data; |
ccffad25 | 1593 | struct netdev_hw_addr *ha; |
32e7bfc4 | 1594 | int uc_count; |
4cd24eaf | 1595 | int mc_count; |
f565a7c2 AW |
1596 | void *buf; |
1597 | int i; | |
2af7698e | 1598 | |
788a8b6d | 1599 | /* We can't dynamically set ndo_set_rx_mode, so return gracefully */ |
2af7698e AW |
1600 | if (!virtio_has_feature(vi->vdev, VIRTIO_NET_F_CTRL_RX)) |
1601 | return; | |
1602 | ||
603343b9 MT |
1603 | vi->ctrl->promisc = ((dev->flags & IFF_PROMISC) != 0); |
1604 | vi->ctrl->allmulti = ((dev->flags & IFF_ALLMULTI) != 0); | |
2af7698e | 1605 | |
603343b9 | 1606 | sg_init_one(sg, &vi->ctrl->promisc, sizeof(vi->ctrl->promisc)); |
2af7698e AW |
1607 | |
1608 | if (!virtnet_send_command(vi, VIRTIO_NET_CTRL_RX, | |
d24bae32 | 1609 | VIRTIO_NET_CTRL_RX_PROMISC, sg)) |
2af7698e | 1610 | dev_warn(&dev->dev, "Failed to %sable promisc mode.\n", |
603343b9 | 1611 | vi->ctrl->promisc ? "en" : "dis"); |
2af7698e | 1612 | |
603343b9 | 1613 | sg_init_one(sg, &vi->ctrl->allmulti, sizeof(vi->ctrl->allmulti)); |
2af7698e AW |
1614 | |
1615 | if (!virtnet_send_command(vi, VIRTIO_NET_CTRL_RX, | |
d24bae32 | 1616 | VIRTIO_NET_CTRL_RX_ALLMULTI, sg)) |
2af7698e | 1617 | dev_warn(&dev->dev, "Failed to %sable allmulti mode.\n", |
603343b9 | 1618 | vi->ctrl->allmulti ? "en" : "dis"); |
f565a7c2 | 1619 | |
32e7bfc4 | 1620 | uc_count = netdev_uc_count(dev); |
4cd24eaf | 1621 | mc_count = netdev_mc_count(dev); |
f565a7c2 | 1622 | /* MAC filter - use one buffer for both lists */ |
4cd24eaf JP |
1623 | buf = kzalloc(((uc_count + mc_count) * ETH_ALEN) + |
1624 | (2 * sizeof(mac_data->entries)), GFP_ATOMIC); | |
1625 | mac_data = buf; | |
e68ed8f0 | 1626 | if (!buf) |
f565a7c2 | 1627 | return; |
f565a7c2 | 1628 | |
23e258e1 AW |
1629 | sg_init_table(sg, 2); |
1630 | ||
f565a7c2 | 1631 | /* Store the unicast list and count in the front of the buffer */ |
fdd819b2 | 1632 | mac_data->entries = cpu_to_virtio32(vi->vdev, uc_count); |
ccffad25 | 1633 | i = 0; |
32e7bfc4 | 1634 | netdev_for_each_uc_addr(ha, dev) |
ccffad25 | 1635 | memcpy(&mac_data->macs[i++][0], ha->addr, ETH_ALEN); |
f565a7c2 AW |
1636 | |
1637 | sg_set_buf(&sg[0], mac_data, | |
32e7bfc4 | 1638 | sizeof(mac_data->entries) + (uc_count * ETH_ALEN)); |
f565a7c2 AW |
1639 | |
1640 | /* multicast list and count fill the end */ | |
32e7bfc4 | 1641 | mac_data = (void *)&mac_data->macs[uc_count][0]; |
f565a7c2 | 1642 | |
fdd819b2 | 1643 | mac_data->entries = cpu_to_virtio32(vi->vdev, mc_count); |
567ec874 | 1644 | i = 0; |
22bedad3 JP |
1645 | netdev_for_each_mc_addr(ha, dev) |
1646 | memcpy(&mac_data->macs[i++][0], ha->addr, ETH_ALEN); | |
f565a7c2 AW |
1647 | |
1648 | sg_set_buf(&sg[1], mac_data, | |
4cd24eaf | 1649 | sizeof(mac_data->entries) + (mc_count * ETH_ALEN)); |
f565a7c2 AW |
1650 | |
1651 | if (!virtnet_send_command(vi, VIRTIO_NET_CTRL_MAC, | |
d24bae32 | 1652 | VIRTIO_NET_CTRL_MAC_TABLE_SET, sg)) |
99e872ae | 1653 | dev_warn(&dev->dev, "Failed to set MAC filter table.\n"); |
f565a7c2 AW |
1654 | |
1655 | kfree(buf); | |
2af7698e AW |
1656 | } |
1657 | ||
80d5c368 PM |
1658 | static int virtnet_vlan_rx_add_vid(struct net_device *dev, |
1659 | __be16 proto, u16 vid) | |
0bde9569 AW |
1660 | { |
1661 | struct virtnet_info *vi = netdev_priv(dev); | |
1662 | struct scatterlist sg; | |
1663 | ||
c285a7d5 | 1664 | vi->ctrl->vid = cpu_to_virtio16(vi->vdev, vid); |
603343b9 | 1665 | sg_init_one(&sg, &vi->ctrl->vid, sizeof(vi->ctrl->vid)); |
0bde9569 AW |
1666 | |
1667 | if (!virtnet_send_command(vi, VIRTIO_NET_CTRL_VLAN, | |
d24bae32 | 1668 | VIRTIO_NET_CTRL_VLAN_ADD, &sg)) |
0bde9569 | 1669 | dev_warn(&dev->dev, "Failed to add VLAN ID %d.\n", vid); |
8e586137 | 1670 | return 0; |
0bde9569 AW |
1671 | } |
1672 | ||
80d5c368 PM |
1673 | static int virtnet_vlan_rx_kill_vid(struct net_device *dev, |
1674 | __be16 proto, u16 vid) | |
0bde9569 AW |
1675 | { |
1676 | struct virtnet_info *vi = netdev_priv(dev); | |
1677 | struct scatterlist sg; | |
1678 | ||
c285a7d5 | 1679 | vi->ctrl->vid = cpu_to_virtio16(vi->vdev, vid); |
603343b9 | 1680 | sg_init_one(&sg, &vi->ctrl->vid, sizeof(vi->ctrl->vid)); |
0bde9569 AW |
1681 | |
1682 | if (!virtnet_send_command(vi, VIRTIO_NET_CTRL_VLAN, | |
d24bae32 | 1683 | VIRTIO_NET_CTRL_VLAN_DEL, &sg)) |
0bde9569 | 1684 | dev_warn(&dev->dev, "Failed to kill VLAN ID %d.\n", vid); |
8e586137 | 1685 | return 0; |
0bde9569 AW |
1686 | } |
1687 | ||
8898c21c | 1688 | static void virtnet_clean_affinity(struct virtnet_info *vi, long hcpu) |
986a4f4d JW |
1689 | { |
1690 | int i; | |
1691 | ||
8898c21c WG |
1692 | if (vi->affinity_hint_set) { |
1693 | for (i = 0; i < vi->max_queue_pairs; i++) { | |
47be2479 WG |
1694 | virtqueue_set_affinity(vi->rq[i].vq, -1); |
1695 | virtqueue_set_affinity(vi->sq[i].vq, -1); | |
1696 | } | |
1697 | ||
8898c21c WG |
1698 | vi->affinity_hint_set = false; |
1699 | } | |
8898c21c | 1700 | } |
47be2479 | 1701 | |
8898c21c WG |
1702 | static void virtnet_set_affinity(struct virtnet_info *vi) |
1703 | { | |
1704 | int i; | |
1705 | int cpu; | |
986a4f4d JW |
1706 | |
1707 | /* In multiqueue mode, when the number of cpu is equal to the number of | |
1708 | * queue pairs, we let the queue pairs to be private to one cpu by | |
1709 | * setting the affinity hint to eliminate the contention. | |
1710 | */ | |
8898c21c WG |
1711 | if (vi->curr_queue_pairs == 1 || |
1712 | vi->max_queue_pairs != num_online_cpus()) { | |
1713 | virtnet_clean_affinity(vi, -1); | |
1714 | return; | |
986a4f4d JW |
1715 | } |
1716 | ||
8898c21c WG |
1717 | i = 0; |
1718 | for_each_online_cpu(cpu) { | |
986a4f4d JW |
1719 | virtqueue_set_affinity(vi->rq[i].vq, cpu); |
1720 | virtqueue_set_affinity(vi->sq[i].vq, cpu); | |
9bb8ca86 | 1721 | netif_set_xps_queue(vi->dev, cpumask_of(cpu), i); |
8898c21c | 1722 | i++; |
986a4f4d JW |
1723 | } |
1724 | ||
8898c21c | 1725 | vi->affinity_hint_set = true; |
986a4f4d JW |
1726 | } |
1727 | ||
8017c279 | 1728 | static int virtnet_cpu_online(unsigned int cpu, struct hlist_node *node) |
8de4b2f3 | 1729 | { |
8017c279 SAS |
1730 | struct virtnet_info *vi = hlist_entry_safe(node, struct virtnet_info, |
1731 | node); | |
1732 | virtnet_set_affinity(vi); | |
1733 | return 0; | |
1734 | } | |
8de4b2f3 | 1735 | |
8017c279 SAS |
1736 | static int virtnet_cpu_dead(unsigned int cpu, struct hlist_node *node) |
1737 | { | |
1738 | struct virtnet_info *vi = hlist_entry_safe(node, struct virtnet_info, | |
1739 | node_dead); | |
1740 | virtnet_set_affinity(vi); | |
1741 | return 0; | |
1742 | } | |
3ab098df | 1743 | |
8017c279 SAS |
1744 | static int virtnet_cpu_down_prep(unsigned int cpu, struct hlist_node *node) |
1745 | { | |
1746 | struct virtnet_info *vi = hlist_entry_safe(node, struct virtnet_info, | |
1747 | node); | |
1748 | ||
1749 | virtnet_clean_affinity(vi, cpu); | |
1750 | return 0; | |
1751 | } | |
1752 | ||
1753 | static enum cpuhp_state virtionet_online; | |
1754 | ||
1755 | static int virtnet_cpu_notif_add(struct virtnet_info *vi) | |
1756 | { | |
1757 | int ret; | |
1758 | ||
1759 | ret = cpuhp_state_add_instance_nocalls(virtionet_online, &vi->node); | |
1760 | if (ret) | |
1761 | return ret; | |
1762 | ret = cpuhp_state_add_instance_nocalls(CPUHP_VIRT_NET_DEAD, | |
1763 | &vi->node_dead); | |
1764 | if (!ret) | |
1765 | return ret; | |
1766 | cpuhp_state_remove_instance_nocalls(virtionet_online, &vi->node); | |
1767 | return ret; | |
1768 | } | |
1769 | ||
1770 | static void virtnet_cpu_notif_remove(struct virtnet_info *vi) | |
1771 | { | |
1772 | cpuhp_state_remove_instance_nocalls(virtionet_online, &vi->node); | |
1773 | cpuhp_state_remove_instance_nocalls(CPUHP_VIRT_NET_DEAD, | |
1774 | &vi->node_dead); | |
986a4f4d JW |
1775 | } |
1776 | ||
8f9f4668 RJ |
1777 | static void virtnet_get_ringparam(struct net_device *dev, |
1778 | struct ethtool_ringparam *ring) | |
1779 | { | |
1780 | struct virtnet_info *vi = netdev_priv(dev); | |
1781 | ||
986a4f4d JW |
1782 | ring->rx_max_pending = virtqueue_get_vring_size(vi->rq[0].vq); |
1783 | ring->tx_max_pending = virtqueue_get_vring_size(vi->sq[0].vq); | |
8f9f4668 RJ |
1784 | ring->rx_pending = ring->rx_max_pending; |
1785 | ring->tx_pending = ring->tx_max_pending; | |
8f9f4668 RJ |
1786 | } |
1787 | ||
66846048 RJ |
1788 | |
1789 | static void virtnet_get_drvinfo(struct net_device *dev, | |
1790 | struct ethtool_drvinfo *info) | |
1791 | { | |
1792 | struct virtnet_info *vi = netdev_priv(dev); | |
1793 | struct virtio_device *vdev = vi->vdev; | |
1794 | ||
1795 | strlcpy(info->driver, KBUILD_MODNAME, sizeof(info->driver)); | |
1796 | strlcpy(info->version, VIRTNET_DRIVER_VERSION, sizeof(info->version)); | |
1797 | strlcpy(info->bus_info, virtio_bus_name(vdev), sizeof(info->bus_info)); | |
1798 | ||
1799 | } | |
1800 | ||
d73bcd2c JW |
1801 | /* TODO: Eliminate OOO packets during switching */ |
1802 | static int virtnet_set_channels(struct net_device *dev, | |
1803 | struct ethtool_channels *channels) | |
1804 | { | |
1805 | struct virtnet_info *vi = netdev_priv(dev); | |
1806 | u16 queue_pairs = channels->combined_count; | |
1807 | int err; | |
1808 | ||
1809 | /* We don't support separate rx/tx channels. | |
1810 | * We don't allow setting 'other' channels. | |
1811 | */ | |
1812 | if (channels->rx_count || channels->tx_count || channels->other_count) | |
1813 | return -EINVAL; | |
1814 | ||
c18e9cd6 | 1815 | if (queue_pairs > vi->max_queue_pairs || queue_pairs == 0) |
d73bcd2c JW |
1816 | return -EINVAL; |
1817 | ||
f600b690 JF |
1818 | /* For now we don't support modifying channels while XDP is loaded |
1819 | * also when XDP is loaded all RX queues have XDP programs so we only | |
1820 | * need to check a single RX queue. | |
1821 | */ | |
1822 | if (vi->rq[0].xdp_prog) | |
1823 | return -EINVAL; | |
1824 | ||
47be2479 | 1825 | get_online_cpus(); |
47315329 | 1826 | err = _virtnet_set_queues(vi, queue_pairs); |
d73bcd2c JW |
1827 | if (!err) { |
1828 | netif_set_real_num_tx_queues(dev, queue_pairs); | |
1829 | netif_set_real_num_rx_queues(dev, queue_pairs); | |
1830 | ||
8898c21c | 1831 | virtnet_set_affinity(vi); |
d73bcd2c | 1832 | } |
47be2479 | 1833 | put_online_cpus(); |
d73bcd2c JW |
1834 | |
1835 | return err; | |
1836 | } | |
1837 | ||
1838 | static void virtnet_get_channels(struct net_device *dev, | |
1839 | struct ethtool_channels *channels) | |
1840 | { | |
1841 | struct virtnet_info *vi = netdev_priv(dev); | |
1842 | ||
1843 | channels->combined_count = vi->curr_queue_pairs; | |
1844 | channels->max_combined = vi->max_queue_pairs; | |
1845 | channels->max_other = 0; | |
1846 | channels->rx_count = 0; | |
1847 | channels->tx_count = 0; | |
1848 | channels->other_count = 0; | |
1849 | } | |
1850 | ||
16032be5 | 1851 | /* Check if the user is trying to change anything besides speed/duplex */ |
ebb6b4b1 PR |
1852 | static bool |
1853 | virtnet_validate_ethtool_cmd(const struct ethtool_link_ksettings *cmd) | |
16032be5 | 1854 | { |
ebb6b4b1 PR |
1855 | struct ethtool_link_ksettings diff1 = *cmd; |
1856 | struct ethtool_link_ksettings diff2 = {}; | |
16032be5 | 1857 | |
0cf3ace9 NA |
1858 | /* cmd is always set so we need to clear it, validate the port type |
1859 | * and also without autonegotiation we can ignore advertising | |
1860 | */ | |
ebb6b4b1 PR |
1861 | diff1.base.speed = 0; |
1862 | diff2.base.port = PORT_OTHER; | |
1863 | ethtool_link_ksettings_zero_link_mode(&diff1, advertising); | |
1864 | diff1.base.duplex = 0; | |
1865 | diff1.base.cmd = 0; | |
1866 | diff1.base.link_mode_masks_nwords = 0; | |
1867 | ||
1868 | return !memcmp(&diff1.base, &diff2.base, sizeof(diff1.base)) && | |
1869 | bitmap_empty(diff1.link_modes.supported, | |
1870 | __ETHTOOL_LINK_MODE_MASK_NBITS) && | |
1871 | bitmap_empty(diff1.link_modes.advertising, | |
1872 | __ETHTOOL_LINK_MODE_MASK_NBITS) && | |
1873 | bitmap_empty(diff1.link_modes.lp_advertising, | |
1874 | __ETHTOOL_LINK_MODE_MASK_NBITS); | |
16032be5 NA |
1875 | } |
1876 | ||
ebb6b4b1 PR |
1877 | static int virtnet_set_link_ksettings(struct net_device *dev, |
1878 | const struct ethtool_link_ksettings *cmd) | |
16032be5 NA |
1879 | { |
1880 | struct virtnet_info *vi = netdev_priv(dev); | |
1881 | u32 speed; | |
1882 | ||
ebb6b4b1 | 1883 | speed = cmd->base.speed; |
16032be5 NA |
1884 | /* don't allow custom speed and duplex */ |
1885 | if (!ethtool_validate_speed(speed) || | |
ebb6b4b1 | 1886 | !ethtool_validate_duplex(cmd->base.duplex) || |
16032be5 NA |
1887 | !virtnet_validate_ethtool_cmd(cmd)) |
1888 | return -EINVAL; | |
1889 | vi->speed = speed; | |
ebb6b4b1 | 1890 | vi->duplex = cmd->base.duplex; |
16032be5 NA |
1891 | |
1892 | return 0; | |
1893 | } | |
1894 | ||
ebb6b4b1 PR |
1895 | static int virtnet_get_link_ksettings(struct net_device *dev, |
1896 | struct ethtool_link_ksettings *cmd) | |
16032be5 NA |
1897 | { |
1898 | struct virtnet_info *vi = netdev_priv(dev); | |
1899 | ||
ebb6b4b1 PR |
1900 | cmd->base.speed = vi->speed; |
1901 | cmd->base.duplex = vi->duplex; | |
1902 | cmd->base.port = PORT_OTHER; | |
16032be5 NA |
1903 | |
1904 | return 0; | |
1905 | } | |
1906 | ||
1907 | static void virtnet_init_settings(struct net_device *dev) | |
1908 | { | |
1909 | struct virtnet_info *vi = netdev_priv(dev); | |
1910 | ||
1911 | vi->speed = SPEED_UNKNOWN; | |
1912 | vi->duplex = DUPLEX_UNKNOWN; | |
1913 | } | |
1914 | ||
0fc0b732 | 1915 | static const struct ethtool_ops virtnet_ethtool_ops = { |
66846048 | 1916 | .get_drvinfo = virtnet_get_drvinfo, |
9f4d26d0 | 1917 | .get_link = ethtool_op_get_link, |
8f9f4668 | 1918 | .get_ringparam = virtnet_get_ringparam, |
d73bcd2c JW |
1919 | .set_channels = virtnet_set_channels, |
1920 | .get_channels = virtnet_get_channels, | |
074c3582 | 1921 | .get_ts_info = ethtool_op_get_ts_info, |
ebb6b4b1 PR |
1922 | .get_link_ksettings = virtnet_get_link_ksettings, |
1923 | .set_link_ksettings = virtnet_set_link_ksettings, | |
a9ea3fc6 HX |
1924 | }; |
1925 | ||
9fe7bfce JF |
1926 | static void virtnet_freeze_down(struct virtio_device *vdev) |
1927 | { | |
1928 | struct virtnet_info *vi = vdev->priv; | |
1929 | int i; | |
1930 | ||
1931 | /* Make sure no work handler is accessing the device */ | |
1932 | flush_work(&vi->config_work); | |
1933 | ||
1934 | netif_device_detach(vi->dev); | |
713a98d9 | 1935 | netif_tx_disable(vi->dev); |
9fe7bfce JF |
1936 | cancel_delayed_work_sync(&vi->refill); |
1937 | ||
1938 | if (netif_running(vi->dev)) { | |
b92f1e67 | 1939 | for (i = 0; i < vi->max_queue_pairs; i++) { |
9fe7bfce | 1940 | napi_disable(&vi->rq[i].napi); |
78a57b48 | 1941 | virtnet_napi_tx_disable(&vi->sq[i].napi); |
b92f1e67 | 1942 | } |
9fe7bfce JF |
1943 | } |
1944 | } | |
1945 | ||
1946 | static int init_vqs(struct virtnet_info *vi); | |
1947 | ||
1948 | static int virtnet_restore_up(struct virtio_device *vdev) | |
1949 | { | |
1950 | struct virtnet_info *vi = vdev->priv; | |
1951 | int err, i; | |
1952 | ||
1953 | err = init_vqs(vi); | |
1954 | if (err) | |
1955 | return err; | |
1956 | ||
1957 | virtio_device_ready(vdev); | |
1958 | ||
1959 | if (netif_running(vi->dev)) { | |
1960 | for (i = 0; i < vi->curr_queue_pairs; i++) | |
1961 | if (!try_fill_recv(vi, &vi->rq[i], GFP_KERNEL)) | |
1962 | schedule_delayed_work(&vi->refill, 0); | |
1963 | ||
b92f1e67 | 1964 | for (i = 0; i < vi->max_queue_pairs; i++) { |
e4e8452a | 1965 | virtnet_napi_enable(vi->rq[i].vq, &vi->rq[i].napi); |
b92f1e67 WB |
1966 | virtnet_napi_tx_enable(vi, vi->sq[i].vq, |
1967 | &vi->sq[i].napi); | |
1968 | } | |
9fe7bfce JF |
1969 | } |
1970 | ||
1971 | netif_device_attach(vi->dev); | |
1972 | return err; | |
1973 | } | |
1974 | ||
3f93522f JW |
1975 | static int virtnet_set_guest_offloads(struct virtnet_info *vi, u64 offloads) |
1976 | { | |
1977 | struct scatterlist sg; | |
603343b9 | 1978 | vi->ctrl->offloads = cpu_to_virtio64(vi->vdev, offloads); |
3f93522f | 1979 | |
603343b9 | 1980 | sg_init_one(&sg, &vi->ctrl->offloads, sizeof(vi->ctrl->offloads)); |
3f93522f JW |
1981 | |
1982 | if (!virtnet_send_command(vi, VIRTIO_NET_CTRL_GUEST_OFFLOADS, | |
1983 | VIRTIO_NET_CTRL_GUEST_OFFLOADS_SET, &sg)) { | |
1984 | dev_warn(&vi->dev->dev, "Fail to set guest offload. \n"); | |
1985 | return -EINVAL; | |
1986 | } | |
1987 | ||
1988 | return 0; | |
1989 | } | |
1990 | ||
1991 | static int virtnet_clear_guest_offloads(struct virtnet_info *vi) | |
1992 | { | |
1993 | u64 offloads = 0; | |
1994 | ||
1995 | if (!vi->guest_offloads) | |
1996 | return 0; | |
1997 | ||
1998 | if (virtio_has_feature(vi->vdev, VIRTIO_NET_F_GUEST_CSUM)) | |
1999 | offloads = 1ULL << VIRTIO_NET_F_GUEST_CSUM; | |
2000 | ||
2001 | return virtnet_set_guest_offloads(vi, offloads); | |
2002 | } | |
2003 | ||
2004 | static int virtnet_restore_guest_offloads(struct virtnet_info *vi) | |
2005 | { | |
2006 | u64 offloads = vi->guest_offloads; | |
2007 | ||
2008 | if (!vi->guest_offloads) | |
2009 | return 0; | |
2010 | if (virtio_has_feature(vi->vdev, VIRTIO_NET_F_GUEST_CSUM)) | |
2011 | offloads |= 1ULL << VIRTIO_NET_F_GUEST_CSUM; | |
2012 | ||
2013 | return virtnet_set_guest_offloads(vi, offloads); | |
2014 | } | |
2015 | ||
9861ce03 JK |
2016 | static int virtnet_xdp_set(struct net_device *dev, struct bpf_prog *prog, |
2017 | struct netlink_ext_ack *extack) | |
f600b690 JF |
2018 | { |
2019 | unsigned long int max_sz = PAGE_SIZE - sizeof(struct padded_vnet_hdr); | |
2020 | struct virtnet_info *vi = netdev_priv(dev); | |
2021 | struct bpf_prog *old_prog; | |
017b29c3 | 2022 | u16 xdp_qp = 0, curr_qp; |
672aafd5 | 2023 | int i, err; |
f600b690 | 2024 | |
3f93522f JW |
2025 | if (!virtio_has_feature(vi->vdev, VIRTIO_NET_F_CTRL_GUEST_OFFLOADS) |
2026 | && (virtio_has_feature(vi->vdev, VIRTIO_NET_F_GUEST_TSO4) || | |
2027 | virtio_has_feature(vi->vdev, VIRTIO_NET_F_GUEST_TSO6) || | |
2028 | virtio_has_feature(vi->vdev, VIRTIO_NET_F_GUEST_ECN) || | |
2029 | virtio_has_feature(vi->vdev, VIRTIO_NET_F_GUEST_UFO))) { | |
4d463c4d | 2030 | NL_SET_ERR_MSG_MOD(extack, "Can't set XDP while host is implementing LRO, disable LRO first"); |
f600b690 JF |
2031 | return -EOPNOTSUPP; |
2032 | } | |
2033 | ||
2034 | if (vi->mergeable_rx_bufs && !vi->any_header_sg) { | |
4d463c4d | 2035 | NL_SET_ERR_MSG_MOD(extack, "XDP expects header/data in single page, any_header_sg required"); |
f600b690 JF |
2036 | return -EINVAL; |
2037 | } | |
2038 | ||
2039 | if (dev->mtu > max_sz) { | |
4d463c4d | 2040 | NL_SET_ERR_MSG_MOD(extack, "MTU too large to enable XDP"); |
f600b690 JF |
2041 | netdev_warn(dev, "XDP requires MTU less than %lu\n", max_sz); |
2042 | return -EINVAL; | |
2043 | } | |
2044 | ||
672aafd5 JF |
2045 | curr_qp = vi->curr_queue_pairs - vi->xdp_queue_pairs; |
2046 | if (prog) | |
2047 | xdp_qp = nr_cpu_ids; | |
2048 | ||
2049 | /* XDP requires extra queues for XDP_TX */ | |
2050 | if (curr_qp + xdp_qp > vi->max_queue_pairs) { | |
4d463c4d | 2051 | NL_SET_ERR_MSG_MOD(extack, "Too few free TX rings available"); |
672aafd5 JF |
2052 | netdev_warn(dev, "request %i queues but max is %i\n", |
2053 | curr_qp + xdp_qp, vi->max_queue_pairs); | |
2054 | return -ENOMEM; | |
2055 | } | |
2056 | ||
2de2f7f4 JF |
2057 | if (prog) { |
2058 | prog = bpf_prog_add(prog, vi->max_queue_pairs - 1); | |
2059 | if (IS_ERR(prog)) | |
2060 | return PTR_ERR(prog); | |
2061 | } | |
2062 | ||
4941d472 | 2063 | /* Make sure NAPI is not using any XDP TX queues for RX. */ |
e527a479 JW |
2064 | if (netif_running(dev)) |
2065 | for (i = 0; i < vi->max_queue_pairs; i++) | |
2066 | napi_disable(&vi->rq[i].napi); | |
f600b690 | 2067 | |
672aafd5 | 2068 | netif_set_real_num_rx_queues(dev, curr_qp + xdp_qp); |
4941d472 JW |
2069 | err = _virtnet_set_queues(vi, curr_qp + xdp_qp); |
2070 | if (err) | |
2071 | goto err; | |
2072 | vi->xdp_queue_pairs = xdp_qp; | |
672aafd5 | 2073 | |
f600b690 JF |
2074 | for (i = 0; i < vi->max_queue_pairs; i++) { |
2075 | old_prog = rtnl_dereference(vi->rq[i].xdp_prog); | |
2076 | rcu_assign_pointer(vi->rq[i].xdp_prog, prog); | |
3f93522f JW |
2077 | if (i == 0) { |
2078 | if (!old_prog) | |
2079 | virtnet_clear_guest_offloads(vi); | |
2080 | if (!prog) | |
2081 | virtnet_restore_guest_offloads(vi); | |
2082 | } | |
f600b690 JF |
2083 | if (old_prog) |
2084 | bpf_prog_put(old_prog); | |
e527a479 JW |
2085 | if (netif_running(dev)) |
2086 | virtnet_napi_enable(vi->rq[i].vq, &vi->rq[i].napi); | |
f600b690 JF |
2087 | } |
2088 | ||
2089 | return 0; | |
2de2f7f4 | 2090 | |
4941d472 JW |
2091 | err: |
2092 | for (i = 0; i < vi->max_queue_pairs; i++) | |
2093 | virtnet_napi_enable(vi->rq[i].vq, &vi->rq[i].napi); | |
2de2f7f4 JF |
2094 | if (prog) |
2095 | bpf_prog_sub(prog, vi->max_queue_pairs - 1); | |
2096 | return err; | |
f600b690 JF |
2097 | } |
2098 | ||
5b0e6629 | 2099 | static u32 virtnet_xdp_query(struct net_device *dev) |
f600b690 JF |
2100 | { |
2101 | struct virtnet_info *vi = netdev_priv(dev); | |
5b0e6629 | 2102 | const struct bpf_prog *xdp_prog; |
f600b690 JF |
2103 | int i; |
2104 | ||
2105 | for (i = 0; i < vi->max_queue_pairs; i++) { | |
5b0e6629 MKL |
2106 | xdp_prog = rtnl_dereference(vi->rq[i].xdp_prog); |
2107 | if (xdp_prog) | |
2108 | return xdp_prog->aux->id; | |
f600b690 | 2109 | } |
5b0e6629 | 2110 | return 0; |
f600b690 JF |
2111 | } |
2112 | ||
f4e63525 | 2113 | static int virtnet_xdp(struct net_device *dev, struct netdev_bpf *xdp) |
f600b690 JF |
2114 | { |
2115 | switch (xdp->command) { | |
2116 | case XDP_SETUP_PROG: | |
9861ce03 | 2117 | return virtnet_xdp_set(dev, xdp->prog, xdp->extack); |
f600b690 | 2118 | case XDP_QUERY_PROG: |
5b0e6629 MKL |
2119 | xdp->prog_id = virtnet_xdp_query(dev); |
2120 | xdp->prog_attached = !!xdp->prog_id; | |
f600b690 JF |
2121 | return 0; |
2122 | default: | |
2123 | return -EINVAL; | |
2124 | } | |
2125 | } | |
2126 | ||
76288b4e SH |
2127 | static const struct net_device_ops virtnet_netdev = { |
2128 | .ndo_open = virtnet_open, | |
2129 | .ndo_stop = virtnet_close, | |
2130 | .ndo_start_xmit = start_xmit, | |
2131 | .ndo_validate_addr = eth_validate_addr, | |
9c46f6d4 | 2132 | .ndo_set_mac_address = virtnet_set_mac_address, |
2af7698e | 2133 | .ndo_set_rx_mode = virtnet_set_rx_mode, |
3fa2a1df | 2134 | .ndo_get_stats64 = virtnet_stats, |
1824a989 AW |
2135 | .ndo_vlan_rx_add_vid = virtnet_vlan_rx_add_vid, |
2136 | .ndo_vlan_rx_kill_vid = virtnet_vlan_rx_kill_vid, | |
76288b4e SH |
2137 | #ifdef CONFIG_NET_POLL_CONTROLLER |
2138 | .ndo_poll_controller = virtnet_netpoll, | |
91815639 | 2139 | #endif |
f4e63525 | 2140 | .ndo_bpf = virtnet_xdp, |
186b3c99 JW |
2141 | .ndo_xdp_xmit = virtnet_xdp_xmit, |
2142 | .ndo_xdp_flush = virtnet_xdp_flush, | |
2836b4f2 | 2143 | .ndo_features_check = passthru_features_check, |
76288b4e SH |
2144 | }; |
2145 | ||
586d17c5 | 2146 | static void virtnet_config_changed_work(struct work_struct *work) |
9f4d26d0 | 2147 | { |
586d17c5 JW |
2148 | struct virtnet_info *vi = |
2149 | container_of(work, struct virtnet_info, config_work); | |
9f4d26d0 MM |
2150 | u16 v; |
2151 | ||
855e0c52 RR |
2152 | if (virtio_cread_feature(vi->vdev, VIRTIO_NET_F_STATUS, |
2153 | struct virtio_net_config, status, &v) < 0) | |
507613bf | 2154 | return; |
586d17c5 JW |
2155 | |
2156 | if (v & VIRTIO_NET_S_ANNOUNCE) { | |
ee89bab1 | 2157 | netdev_notify_peers(vi->dev); |
586d17c5 JW |
2158 | virtnet_ack_link_announce(vi); |
2159 | } | |
9f4d26d0 MM |
2160 | |
2161 | /* Ignore unknown (future) status bits */ | |
2162 | v &= VIRTIO_NET_S_LINK_UP; | |
2163 | ||
2164 | if (vi->status == v) | |
507613bf | 2165 | return; |
9f4d26d0 MM |
2166 | |
2167 | vi->status = v; | |
2168 | ||
2169 | if (vi->status & VIRTIO_NET_S_LINK_UP) { | |
2170 | netif_carrier_on(vi->dev); | |
986a4f4d | 2171 | netif_tx_wake_all_queues(vi->dev); |
9f4d26d0 MM |
2172 | } else { |
2173 | netif_carrier_off(vi->dev); | |
986a4f4d | 2174 | netif_tx_stop_all_queues(vi->dev); |
9f4d26d0 MM |
2175 | } |
2176 | } | |
2177 | ||
2178 | static void virtnet_config_changed(struct virtio_device *vdev) | |
2179 | { | |
2180 | struct virtnet_info *vi = vdev->priv; | |
2181 | ||
3b07e9ca | 2182 | schedule_work(&vi->config_work); |
9f4d26d0 MM |
2183 | } |
2184 | ||
986a4f4d JW |
2185 | static void virtnet_free_queues(struct virtnet_info *vi) |
2186 | { | |
d4fb84ee AV |
2187 | int i; |
2188 | ||
ab3971b1 JW |
2189 | for (i = 0; i < vi->max_queue_pairs; i++) { |
2190 | napi_hash_del(&vi->rq[i].napi); | |
d4fb84ee | 2191 | netif_napi_del(&vi->rq[i].napi); |
b92f1e67 | 2192 | netif_napi_del(&vi->sq[i].napi); |
ab3971b1 | 2193 | } |
d4fb84ee | 2194 | |
963abe5c ED |
2195 | /* We called napi_hash_del() before netif_napi_del(), |
2196 | * we need to respect an RCU grace period before freeing vi->rq | |
2197 | */ | |
2198 | synchronize_net(); | |
2199 | ||
986a4f4d JW |
2200 | kfree(vi->rq); |
2201 | kfree(vi->sq); | |
603343b9 | 2202 | kfree(vi->ctrl); |
986a4f4d JW |
2203 | } |
2204 | ||
47315329 | 2205 | static void _free_receive_bufs(struct virtnet_info *vi) |
986a4f4d | 2206 | { |
f600b690 | 2207 | struct bpf_prog *old_prog; |
986a4f4d JW |
2208 | int i; |
2209 | ||
2210 | for (i = 0; i < vi->max_queue_pairs; i++) { | |
2211 | while (vi->rq[i].pages) | |
2212 | __free_pages(get_a_page(&vi->rq[i], GFP_KERNEL), 0); | |
f600b690 JF |
2213 | |
2214 | old_prog = rtnl_dereference(vi->rq[i].xdp_prog); | |
2215 | RCU_INIT_POINTER(vi->rq[i].xdp_prog, NULL); | |
2216 | if (old_prog) | |
2217 | bpf_prog_put(old_prog); | |
986a4f4d | 2218 | } |
47315329 JF |
2219 | } |
2220 | ||
2221 | static void free_receive_bufs(struct virtnet_info *vi) | |
2222 | { | |
2223 | rtnl_lock(); | |
2224 | _free_receive_bufs(vi); | |
f600b690 | 2225 | rtnl_unlock(); |
986a4f4d JW |
2226 | } |
2227 | ||
fb51879d MD |
2228 | static void free_receive_page_frags(struct virtnet_info *vi) |
2229 | { | |
2230 | int i; | |
2231 | for (i = 0; i < vi->max_queue_pairs; i++) | |
2232 | if (vi->rq[i].alloc_frag.page) | |
2233 | put_page(vi->rq[i].alloc_frag.page); | |
2234 | } | |
2235 | ||
b68df015 | 2236 | static bool is_xdp_raw_buffer_queue(struct virtnet_info *vi, int q) |
56434a01 JF |
2237 | { |
2238 | if (q < (vi->curr_queue_pairs - vi->xdp_queue_pairs)) | |
2239 | return false; | |
2240 | else if (q < vi->curr_queue_pairs) | |
2241 | return true; | |
2242 | else | |
2243 | return false; | |
2244 | } | |
2245 | ||
986a4f4d JW |
2246 | static void free_unused_bufs(struct virtnet_info *vi) |
2247 | { | |
2248 | void *buf; | |
2249 | int i; | |
2250 | ||
2251 | for (i = 0; i < vi->max_queue_pairs; i++) { | |
2252 | struct virtqueue *vq = vi->sq[i].vq; | |
56434a01 | 2253 | while ((buf = virtqueue_detach_unused_buf(vq)) != NULL) { |
b68df015 | 2254 | if (!is_xdp_raw_buffer_queue(vi, i)) |
56434a01 JF |
2255 | dev_kfree_skb(buf); |
2256 | else | |
2257 | put_page(virt_to_head_page(buf)); | |
2258 | } | |
986a4f4d JW |
2259 | } |
2260 | ||
2261 | for (i = 0; i < vi->max_queue_pairs; i++) { | |
2262 | struct virtqueue *vq = vi->rq[i].vq; | |
2263 | ||
2264 | while ((buf = virtqueue_detach_unused_buf(vq)) != NULL) { | |
ab7db917 | 2265 | if (vi->mergeable_rx_bufs) { |
680557cf | 2266 | put_page(virt_to_head_page(buf)); |
ab7db917 | 2267 | } else if (vi->big_packets) { |
fa9fac17 | 2268 | give_pages(&vi->rq[i], buf); |
ab7db917 | 2269 | } else { |
f6b10209 | 2270 | put_page(virt_to_head_page(buf)); |
ab7db917 | 2271 | } |
986a4f4d | 2272 | } |
986a4f4d JW |
2273 | } |
2274 | } | |
2275 | ||
e9d7417b JW |
2276 | static void virtnet_del_vqs(struct virtnet_info *vi) |
2277 | { | |
2278 | struct virtio_device *vdev = vi->vdev; | |
2279 | ||
8898c21c | 2280 | virtnet_clean_affinity(vi, -1); |
986a4f4d | 2281 | |
e9d7417b | 2282 | vdev->config->del_vqs(vdev); |
986a4f4d JW |
2283 | |
2284 | virtnet_free_queues(vi); | |
e9d7417b JW |
2285 | } |
2286 | ||
d85b758f MT |
2287 | /* How large should a single buffer be so a queue full of these can fit at |
2288 | * least one full packet? | |
2289 | * Logic below assumes the mergeable buffer header is used. | |
2290 | */ | |
2291 | static unsigned int mergeable_min_buf_len(struct virtnet_info *vi, struct virtqueue *vq) | |
2292 | { | |
2293 | const unsigned int hdr_len = sizeof(struct virtio_net_hdr_mrg_rxbuf); | |
2294 | unsigned int rq_size = virtqueue_get_vring_size(vq); | |
2295 | unsigned int packet_len = vi->big_packets ? IP_MAX_MTU : vi->dev->max_mtu; | |
2296 | unsigned int buf_len = hdr_len + ETH_HLEN + VLAN_HLEN + packet_len; | |
2297 | unsigned int min_buf_len = DIV_ROUND_UP(buf_len, rq_size); | |
2298 | ||
f0c3192c MT |
2299 | return max(max(min_buf_len, hdr_len) - hdr_len, |
2300 | (unsigned int)GOOD_PACKET_LEN); | |
d85b758f MT |
2301 | } |
2302 | ||
986a4f4d | 2303 | static int virtnet_find_vqs(struct virtnet_info *vi) |
3f9c10b0 | 2304 | { |
986a4f4d JW |
2305 | vq_callback_t **callbacks; |
2306 | struct virtqueue **vqs; | |
2307 | int ret = -ENOMEM; | |
2308 | int i, total_vqs; | |
2309 | const char **names; | |
d45b897b | 2310 | bool *ctx; |
986a4f4d JW |
2311 | |
2312 | /* We expect 1 RX virtqueue followed by 1 TX virtqueue, followed by | |
2313 | * possible N-1 RX/TX queue pairs used in multiqueue mode, followed by | |
2314 | * possible control vq. | |
2315 | */ | |
2316 | total_vqs = vi->max_queue_pairs * 2 + | |
2317 | virtio_has_feature(vi->vdev, VIRTIO_NET_F_CTRL_VQ); | |
2318 | ||
2319 | /* Allocate space for find_vqs parameters */ | |
2320 | vqs = kzalloc(total_vqs * sizeof(*vqs), GFP_KERNEL); | |
2321 | if (!vqs) | |
2322 | goto err_vq; | |
2323 | callbacks = kmalloc(total_vqs * sizeof(*callbacks), GFP_KERNEL); | |
2324 | if (!callbacks) | |
2325 | goto err_callback; | |
2326 | names = kmalloc(total_vqs * sizeof(*names), GFP_KERNEL); | |
2327 | if (!names) | |
2328 | goto err_names; | |
192f68cf | 2329 | if (!vi->big_packets || vi->mergeable_rx_bufs) { |
d45b897b MT |
2330 | ctx = kzalloc(total_vqs * sizeof(*ctx), GFP_KERNEL); |
2331 | if (!ctx) | |
2332 | goto err_ctx; | |
2333 | } else { | |
2334 | ctx = NULL; | |
2335 | } | |
986a4f4d JW |
2336 | |
2337 | /* Parameters for control virtqueue, if any */ | |
2338 | if (vi->has_cvq) { | |
2339 | callbacks[total_vqs - 1] = NULL; | |
2340 | names[total_vqs - 1] = "control"; | |
2341 | } | |
3f9c10b0 | 2342 | |
986a4f4d JW |
2343 | /* Allocate/initialize parameters for send/receive virtqueues */ |
2344 | for (i = 0; i < vi->max_queue_pairs; i++) { | |
2345 | callbacks[rxq2vq(i)] = skb_recv_done; | |
2346 | callbacks[txq2vq(i)] = skb_xmit_done; | |
2347 | sprintf(vi->rq[i].name, "input.%d", i); | |
2348 | sprintf(vi->sq[i].name, "output.%d", i); | |
2349 | names[rxq2vq(i)] = vi->rq[i].name; | |
2350 | names[txq2vq(i)] = vi->sq[i].name; | |
d45b897b MT |
2351 | if (ctx) |
2352 | ctx[rxq2vq(i)] = true; | |
986a4f4d | 2353 | } |
3f9c10b0 | 2354 | |
986a4f4d | 2355 | ret = vi->vdev->config->find_vqs(vi->vdev, total_vqs, vqs, callbacks, |
d45b897b | 2356 | names, ctx, NULL); |
986a4f4d JW |
2357 | if (ret) |
2358 | goto err_find; | |
3f9c10b0 | 2359 | |
986a4f4d JW |
2360 | if (vi->has_cvq) { |
2361 | vi->cvq = vqs[total_vqs - 1]; | |
3f9c10b0 | 2362 | if (virtio_has_feature(vi->vdev, VIRTIO_NET_F_CTRL_VLAN)) |
f646968f | 2363 | vi->dev->features |= NETIF_F_HW_VLAN_CTAG_FILTER; |
3f9c10b0 | 2364 | } |
986a4f4d JW |
2365 | |
2366 | for (i = 0; i < vi->max_queue_pairs; i++) { | |
2367 | vi->rq[i].vq = vqs[rxq2vq(i)]; | |
d85b758f | 2368 | vi->rq[i].min_buf_len = mergeable_min_buf_len(vi, vi->rq[i].vq); |
986a4f4d JW |
2369 | vi->sq[i].vq = vqs[txq2vq(i)]; |
2370 | } | |
2371 | ||
2372 | kfree(names); | |
2373 | kfree(callbacks); | |
2374 | kfree(vqs); | |
55281621 | 2375 | kfree(ctx); |
986a4f4d | 2376 | |
3f9c10b0 | 2377 | return 0; |
986a4f4d JW |
2378 | |
2379 | err_find: | |
d45b897b MT |
2380 | kfree(ctx); |
2381 | err_ctx: | |
986a4f4d JW |
2382 | kfree(names); |
2383 | err_names: | |
2384 | kfree(callbacks); | |
2385 | err_callback: | |
2386 | kfree(vqs); | |
2387 | err_vq: | |
2388 | return ret; | |
2389 | } | |
2390 | ||
2391 | static int virtnet_alloc_queues(struct virtnet_info *vi) | |
2392 | { | |
2393 | int i; | |
2394 | ||
603343b9 MT |
2395 | vi->ctrl = kzalloc(sizeof(*vi->ctrl), GFP_KERNEL); |
2396 | if (!vi->ctrl) | |
2397 | goto err_ctrl; | |
986a4f4d JW |
2398 | vi->sq = kzalloc(sizeof(*vi->sq) * vi->max_queue_pairs, GFP_KERNEL); |
2399 | if (!vi->sq) | |
2400 | goto err_sq; | |
2401 | vi->rq = kzalloc(sizeof(*vi->rq) * vi->max_queue_pairs, GFP_KERNEL); | |
008d4278 | 2402 | if (!vi->rq) |
986a4f4d JW |
2403 | goto err_rq; |
2404 | ||
2405 | INIT_DELAYED_WORK(&vi->refill, refill_work); | |
2406 | for (i = 0; i < vi->max_queue_pairs; i++) { | |
2407 | vi->rq[i].pages = NULL; | |
2408 | netif_napi_add(vi->dev, &vi->rq[i].napi, virtnet_poll, | |
2409 | napi_weight); | |
1d11e732 WB |
2410 | netif_tx_napi_add(vi->dev, &vi->sq[i].napi, virtnet_poll_tx, |
2411 | napi_tx ? napi_weight : 0); | |
986a4f4d JW |
2412 | |
2413 | sg_init_table(vi->rq[i].sg, ARRAY_SIZE(vi->rq[i].sg)); | |
5377d758 | 2414 | ewma_pkt_len_init(&vi->rq[i].mrg_avg_pkt_len); |
986a4f4d JW |
2415 | sg_init_table(vi->sq[i].sg, ARRAY_SIZE(vi->sq[i].sg)); |
2416 | } | |
2417 | ||
2418 | return 0; | |
2419 | ||
2420 | err_rq: | |
2421 | kfree(vi->sq); | |
2422 | err_sq: | |
603343b9 MT |
2423 | kfree(vi->ctrl); |
2424 | err_ctrl: | |
986a4f4d JW |
2425 | return -ENOMEM; |
2426 | } | |
2427 | ||
2428 | static int init_vqs(struct virtnet_info *vi) | |
2429 | { | |
2430 | int ret; | |
2431 | ||
2432 | /* Allocate send & receive queues */ | |
2433 | ret = virtnet_alloc_queues(vi); | |
2434 | if (ret) | |
2435 | goto err; | |
2436 | ||
2437 | ret = virtnet_find_vqs(vi); | |
2438 | if (ret) | |
2439 | goto err_free; | |
2440 | ||
47be2479 | 2441 | get_online_cpus(); |
8898c21c | 2442 | virtnet_set_affinity(vi); |
47be2479 WG |
2443 | put_online_cpus(); |
2444 | ||
986a4f4d JW |
2445 | return 0; |
2446 | ||
2447 | err_free: | |
2448 | virtnet_free_queues(vi); | |
2449 | err: | |
2450 | return ret; | |
3f9c10b0 AS |
2451 | } |
2452 | ||
fbf28d78 MD |
2453 | #ifdef CONFIG_SYSFS |
2454 | static ssize_t mergeable_rx_buffer_size_show(struct netdev_rx_queue *queue, | |
718ad681 | 2455 | char *buf) |
fbf28d78 MD |
2456 | { |
2457 | struct virtnet_info *vi = netdev_priv(queue->dev); | |
2458 | unsigned int queue_index = get_netdev_rx_queue_index(queue); | |
5377d758 | 2459 | struct ewma_pkt_len *avg; |
fbf28d78 MD |
2460 | |
2461 | BUG_ON(queue_index >= vi->max_queue_pairs); | |
2462 | avg = &vi->rq[queue_index].mrg_avg_pkt_len; | |
d85b758f MT |
2463 | return sprintf(buf, "%u\n", |
2464 | get_mergeable_buf_len(&vi->rq[queue_index], avg)); | |
fbf28d78 MD |
2465 | } |
2466 | ||
2467 | static struct rx_queue_attribute mergeable_rx_buffer_size_attribute = | |
2468 | __ATTR_RO(mergeable_rx_buffer_size); | |
2469 | ||
2470 | static struct attribute *virtio_net_mrg_rx_attrs[] = { | |
2471 | &mergeable_rx_buffer_size_attribute.attr, | |
2472 | NULL | |
2473 | }; | |
2474 | ||
2475 | static const struct attribute_group virtio_net_mrg_rx_group = { | |
2476 | .name = "virtio_net", | |
2477 | .attrs = virtio_net_mrg_rx_attrs | |
2478 | }; | |
2479 | #endif | |
2480 | ||
892d6eb1 JW |
2481 | static bool virtnet_fail_on_feature(struct virtio_device *vdev, |
2482 | unsigned int fbit, | |
2483 | const char *fname, const char *dname) | |
2484 | { | |
2485 | if (!virtio_has_feature(vdev, fbit)) | |
2486 | return false; | |
2487 | ||
2488 | dev_err(&vdev->dev, "device advertises feature %s but not %s", | |
2489 | fname, dname); | |
2490 | ||
2491 | return true; | |
2492 | } | |
2493 | ||
2494 | #define VIRTNET_FAIL_ON(vdev, fbit, dbit) \ | |
2495 | virtnet_fail_on_feature(vdev, fbit, #fbit, dbit) | |
2496 | ||
2497 | static bool virtnet_validate_features(struct virtio_device *vdev) | |
2498 | { | |
2499 | if (!virtio_has_feature(vdev, VIRTIO_NET_F_CTRL_VQ) && | |
2500 | (VIRTNET_FAIL_ON(vdev, VIRTIO_NET_F_CTRL_RX, | |
2501 | "VIRTIO_NET_F_CTRL_VQ") || | |
2502 | VIRTNET_FAIL_ON(vdev, VIRTIO_NET_F_CTRL_VLAN, | |
2503 | "VIRTIO_NET_F_CTRL_VQ") || | |
2504 | VIRTNET_FAIL_ON(vdev, VIRTIO_NET_F_GUEST_ANNOUNCE, | |
2505 | "VIRTIO_NET_F_CTRL_VQ") || | |
2506 | VIRTNET_FAIL_ON(vdev, VIRTIO_NET_F_MQ, "VIRTIO_NET_F_CTRL_VQ") || | |
2507 | VIRTNET_FAIL_ON(vdev, VIRTIO_NET_F_CTRL_MAC_ADDR, | |
2508 | "VIRTIO_NET_F_CTRL_VQ"))) { | |
2509 | return false; | |
2510 | } | |
2511 | ||
2512 | return true; | |
2513 | } | |
2514 | ||
d0c2c997 JW |
2515 | #define MIN_MTU ETH_MIN_MTU |
2516 | #define MAX_MTU ETH_MAX_MTU | |
2517 | ||
fe36cbe0 | 2518 | static int virtnet_validate(struct virtio_device *vdev) |
296f96fc | 2519 | { |
6ba42248 MT |
2520 | if (!vdev->config->get) { |
2521 | dev_err(&vdev->dev, "%s failure: config access disabled\n", | |
2522 | __func__); | |
2523 | return -EINVAL; | |
2524 | } | |
2525 | ||
892d6eb1 JW |
2526 | if (!virtnet_validate_features(vdev)) |
2527 | return -EINVAL; | |
2528 | ||
fe36cbe0 MT |
2529 | if (virtio_has_feature(vdev, VIRTIO_NET_F_MTU)) { |
2530 | int mtu = virtio_cread16(vdev, | |
2531 | offsetof(struct virtio_net_config, | |
2532 | mtu)); | |
2533 | if (mtu < MIN_MTU) | |
2534 | __virtio_clear_bit(vdev, VIRTIO_NET_F_MTU); | |
2535 | } | |
2536 | ||
2537 | return 0; | |
2538 | } | |
2539 | ||
2540 | static int virtnet_probe(struct virtio_device *vdev) | |
2541 | { | |
2542 | int i, err; | |
2543 | struct net_device *dev; | |
2544 | struct virtnet_info *vi; | |
2545 | u16 max_queue_pairs; | |
2546 | int mtu; | |
2547 | ||
986a4f4d | 2548 | /* Find if host supports multiqueue virtio_net device */ |
855e0c52 RR |
2549 | err = virtio_cread_feature(vdev, VIRTIO_NET_F_MQ, |
2550 | struct virtio_net_config, | |
2551 | max_virtqueue_pairs, &max_queue_pairs); | |
986a4f4d JW |
2552 | |
2553 | /* We need at least 2 queue's */ | |
2554 | if (err || max_queue_pairs < VIRTIO_NET_CTRL_MQ_VQ_PAIRS_MIN || | |
2555 | max_queue_pairs > VIRTIO_NET_CTRL_MQ_VQ_PAIRS_MAX || | |
2556 | !virtio_has_feature(vdev, VIRTIO_NET_F_CTRL_VQ)) | |
2557 | max_queue_pairs = 1; | |
296f96fc RR |
2558 | |
2559 | /* Allocate ourselves a network device with room for our info */ | |
986a4f4d | 2560 | dev = alloc_etherdev_mq(sizeof(struct virtnet_info), max_queue_pairs); |
296f96fc RR |
2561 | if (!dev) |
2562 | return -ENOMEM; | |
2563 | ||
2564 | /* Set up network device as normal. */ | |
f2f2c8b4 | 2565 | dev->priv_flags |= IFF_UNICAST_FLT | IFF_LIVE_ADDR_CHANGE; |
76288b4e | 2566 | dev->netdev_ops = &virtnet_netdev; |
296f96fc | 2567 | dev->features = NETIF_F_HIGHDMA; |
3fa2a1df | 2568 | |
7ad24ea4 | 2569 | dev->ethtool_ops = &virtnet_ethtool_ops; |
296f96fc RR |
2570 | SET_NETDEV_DEV(dev, &vdev->dev); |
2571 | ||
2572 | /* Do we support "hardware" checksums? */ | |
98e778c9 | 2573 | if (virtio_has_feature(vdev, VIRTIO_NET_F_CSUM)) { |
296f96fc | 2574 | /* This opens up the world of extra features. */ |
48900cb6 | 2575 | dev->hw_features |= NETIF_F_HW_CSUM | NETIF_F_SG; |
98e778c9 | 2576 | if (csum) |
48900cb6 | 2577 | dev->features |= NETIF_F_HW_CSUM | NETIF_F_SG; |
98e778c9 MM |
2578 | |
2579 | if (virtio_has_feature(vdev, VIRTIO_NET_F_GSO)) { | |
e078de03 | 2580 | dev->hw_features |= NETIF_F_TSO |
34a48579 RR |
2581 | | NETIF_F_TSO_ECN | NETIF_F_TSO6; |
2582 | } | |
5539ae96 | 2583 | /* Individual feature bits: what can host handle? */ |
98e778c9 MM |
2584 | if (virtio_has_feature(vdev, VIRTIO_NET_F_HOST_TSO4)) |
2585 | dev->hw_features |= NETIF_F_TSO; | |
2586 | if (virtio_has_feature(vdev, VIRTIO_NET_F_HOST_TSO6)) | |
2587 | dev->hw_features |= NETIF_F_TSO6; | |
2588 | if (virtio_has_feature(vdev, VIRTIO_NET_F_HOST_ECN)) | |
2589 | dev->hw_features |= NETIF_F_TSO_ECN; | |
98e778c9 | 2590 | |
41f2f127 JW |
2591 | dev->features |= NETIF_F_GSO_ROBUST; |
2592 | ||
98e778c9 | 2593 | if (gso) |
e078de03 | 2594 | dev->features |= dev->hw_features & NETIF_F_ALL_TSO; |
98e778c9 | 2595 | /* (!csum && gso) case will be fixed by register_netdev() */ |
296f96fc | 2596 | } |
4f49129b TH |
2597 | if (virtio_has_feature(vdev, VIRTIO_NET_F_GUEST_CSUM)) |
2598 | dev->features |= NETIF_F_RXCSUM; | |
296f96fc | 2599 | |
4fda8302 JW |
2600 | dev->vlan_features = dev->features; |
2601 | ||
d0c2c997 JW |
2602 | /* MTU range: 68 - 65535 */ |
2603 | dev->min_mtu = MIN_MTU; | |
2604 | dev->max_mtu = MAX_MTU; | |
2605 | ||
296f96fc | 2606 | /* Configuration may specify what MAC to use. Otherwise random. */ |
855e0c52 RR |
2607 | if (virtio_has_feature(vdev, VIRTIO_NET_F_MAC)) |
2608 | virtio_cread_bytes(vdev, | |
2609 | offsetof(struct virtio_net_config, mac), | |
2610 | dev->dev_addr, dev->addr_len); | |
2611 | else | |
f2cedb63 | 2612 | eth_hw_addr_random(dev); |
296f96fc RR |
2613 | |
2614 | /* Set up our device-specific information */ | |
2615 | vi = netdev_priv(dev); | |
296f96fc RR |
2616 | vi->dev = dev; |
2617 | vi->vdev = vdev; | |
d9d5dcc8 | 2618 | vdev->priv = vi; |
3fa2a1df | 2619 | vi->stats = alloc_percpu(struct virtnet_stats); |
2620 | err = -ENOMEM; | |
2621 | if (vi->stats == NULL) | |
2622 | goto free; | |
2623 | ||
827da44c JS |
2624 | for_each_possible_cpu(i) { |
2625 | struct virtnet_stats *virtnet_stats; | |
2626 | virtnet_stats = per_cpu_ptr(vi->stats, i); | |
2627 | u64_stats_init(&virtnet_stats->tx_syncp); | |
2628 | u64_stats_init(&virtnet_stats->rx_syncp); | |
2629 | } | |
2630 | ||
586d17c5 | 2631 | INIT_WORK(&vi->config_work, virtnet_config_changed_work); |
296f96fc | 2632 | |
97402b96 | 2633 | /* If we can receive ANY GSO packets, we must allocate large ones. */ |
8e95a202 JP |
2634 | if (virtio_has_feature(vdev, VIRTIO_NET_F_GUEST_TSO4) || |
2635 | virtio_has_feature(vdev, VIRTIO_NET_F_GUEST_TSO6) || | |
e3e3c423 VY |
2636 | virtio_has_feature(vdev, VIRTIO_NET_F_GUEST_ECN) || |
2637 | virtio_has_feature(vdev, VIRTIO_NET_F_GUEST_UFO)) | |
97402b96 HX |
2638 | vi->big_packets = true; |
2639 | ||
3f2c31d9 MM |
2640 | if (virtio_has_feature(vdev, VIRTIO_NET_F_MRG_RXBUF)) |
2641 | vi->mergeable_rx_bufs = true; | |
2642 | ||
d04302b3 MT |
2643 | if (virtio_has_feature(vdev, VIRTIO_NET_F_MRG_RXBUF) || |
2644 | virtio_has_feature(vdev, VIRTIO_F_VERSION_1)) | |
012873d0 MT |
2645 | vi->hdr_len = sizeof(struct virtio_net_hdr_mrg_rxbuf); |
2646 | else | |
2647 | vi->hdr_len = sizeof(struct virtio_net_hdr); | |
2648 | ||
75993300 MT |
2649 | if (virtio_has_feature(vdev, VIRTIO_F_ANY_LAYOUT) || |
2650 | virtio_has_feature(vdev, VIRTIO_F_VERSION_1)) | |
e7428e95 MT |
2651 | vi->any_header_sg = true; |
2652 | ||
986a4f4d JW |
2653 | if (virtio_has_feature(vdev, VIRTIO_NET_F_CTRL_VQ)) |
2654 | vi->has_cvq = true; | |
2655 | ||
14de9d11 AC |
2656 | if (virtio_has_feature(vdev, VIRTIO_NET_F_MTU)) { |
2657 | mtu = virtio_cread16(vdev, | |
2658 | offsetof(struct virtio_net_config, | |
2659 | mtu)); | |
93a205ee | 2660 | if (mtu < dev->min_mtu) { |
fe36cbe0 MT |
2661 | /* Should never trigger: MTU was previously validated |
2662 | * in virtnet_validate. | |
2663 | */ | |
2664 | dev_err(&vdev->dev, "device MTU appears to have changed " | |
2665 | "it is now %d < %d", mtu, dev->min_mtu); | |
2666 | goto free_stats; | |
93a205ee | 2667 | } |
2e123b44 | 2668 | |
fe36cbe0 MT |
2669 | dev->mtu = mtu; |
2670 | dev->max_mtu = mtu; | |
2671 | ||
2e123b44 MT |
2672 | /* TODO: size buffers correctly in this case. */ |
2673 | if (dev->mtu > ETH_DATA_LEN) | |
2674 | vi->big_packets = true; | |
14de9d11 AC |
2675 | } |
2676 | ||
012873d0 MT |
2677 | if (vi->any_header_sg) |
2678 | dev->needed_headroom = vi->hdr_len; | |
6ebbc1a6 | 2679 | |
44900010 JW |
2680 | /* Enable multiqueue by default */ |
2681 | if (num_online_cpus() >= max_queue_pairs) | |
2682 | vi->curr_queue_pairs = max_queue_pairs; | |
2683 | else | |
2684 | vi->curr_queue_pairs = num_online_cpus(); | |
986a4f4d JW |
2685 | vi->max_queue_pairs = max_queue_pairs; |
2686 | ||
2687 | /* Allocate/initialize the rx/tx queues, and invoke find_vqs */ | |
3f9c10b0 | 2688 | err = init_vqs(vi); |
d2a7ddda | 2689 | if (err) |
9bb8ca86 | 2690 | goto free_stats; |
296f96fc | 2691 | |
fbf28d78 MD |
2692 | #ifdef CONFIG_SYSFS |
2693 | if (vi->mergeable_rx_bufs) | |
2694 | dev->sysfs_rx_queue_group = &virtio_net_mrg_rx_group; | |
2695 | #endif | |
0f13b66b ZYW |
2696 | netif_set_real_num_tx_queues(dev, vi->curr_queue_pairs); |
2697 | netif_set_real_num_rx_queues(dev, vi->curr_queue_pairs); | |
986a4f4d | 2698 | |
16032be5 NA |
2699 | virtnet_init_settings(dev); |
2700 | ||
296f96fc RR |
2701 | err = register_netdev(dev); |
2702 | if (err) { | |
2703 | pr_debug("virtio_net: registering device failed\n"); | |
d2a7ddda | 2704 | goto free_vqs; |
296f96fc | 2705 | } |
b3369c1f | 2706 | |
4baf1e33 MT |
2707 | virtio_device_ready(vdev); |
2708 | ||
8017c279 | 2709 | err = virtnet_cpu_notif_add(vi); |
8de4b2f3 WG |
2710 | if (err) { |
2711 | pr_debug("virtio_net: registering cpu notifier failed\n"); | |
f00e35e2 | 2712 | goto free_unregister_netdev; |
8de4b2f3 WG |
2713 | } |
2714 | ||
a220871b | 2715 | virtnet_set_queues(vi, vi->curr_queue_pairs); |
44900010 | 2716 | |
167c25e4 JW |
2717 | /* Assume link up if device can't report link status, |
2718 | otherwise get link status from config. */ | |
5bf6a429 | 2719 | netif_carrier_off(dev); |
167c25e4 | 2720 | if (virtio_has_feature(vi->vdev, VIRTIO_NET_F_STATUS)) { |
3b07e9ca | 2721 | schedule_work(&vi->config_work); |
167c25e4 JW |
2722 | } else { |
2723 | vi->status = VIRTIO_NET_S_LINK_UP; | |
2724 | netif_carrier_on(dev); | |
2725 | } | |
9f4d26d0 | 2726 | |
3f93522f JW |
2727 | for (i = 0; i < ARRAY_SIZE(guest_offloads); i++) |
2728 | if (virtio_has_feature(vi->vdev, guest_offloads[i])) | |
2729 | set_bit(guest_offloads[i], &vi->guest_offloads); | |
2730 | ||
986a4f4d JW |
2731 | pr_debug("virtnet: registered device %s with %d RX and TX vq's\n", |
2732 | dev->name, max_queue_pairs); | |
2733 | ||
296f96fc RR |
2734 | return 0; |
2735 | ||
f00e35e2 | 2736 | free_unregister_netdev: |
02465555 MT |
2737 | vi->vdev->config->reset(vdev); |
2738 | ||
b3369c1f | 2739 | unregister_netdev(dev); |
d2a7ddda | 2740 | free_vqs: |
986a4f4d | 2741 | cancel_delayed_work_sync(&vi->refill); |
fb51879d | 2742 | free_receive_page_frags(vi); |
e9d7417b | 2743 | virtnet_del_vqs(vi); |
3fa2a1df | 2744 | free_stats: |
2745 | free_percpu(vi->stats); | |
296f96fc RR |
2746 | free: |
2747 | free_netdev(dev); | |
2748 | return err; | |
2749 | } | |
2750 | ||
04486ed0 | 2751 | static void remove_vq_common(struct virtnet_info *vi) |
296f96fc | 2752 | { |
04486ed0 | 2753 | vi->vdev->config->reset(vi->vdev); |
830a8a97 SM |
2754 | |
2755 | /* Free unused buffers in both send and recv, if any. */ | |
9ab86bbc | 2756 | free_unused_bufs(vi); |
fb6813f4 | 2757 | |
986a4f4d | 2758 | free_receive_bufs(vi); |
d2a7ddda | 2759 | |
fb51879d MD |
2760 | free_receive_page_frags(vi); |
2761 | ||
986a4f4d | 2762 | virtnet_del_vqs(vi); |
04486ed0 AS |
2763 | } |
2764 | ||
8cc085d6 | 2765 | static void virtnet_remove(struct virtio_device *vdev) |
04486ed0 AS |
2766 | { |
2767 | struct virtnet_info *vi = vdev->priv; | |
2768 | ||
8017c279 | 2769 | virtnet_cpu_notif_remove(vi); |
8de4b2f3 | 2770 | |
102a2786 MT |
2771 | /* Make sure no work handler is accessing the device. */ |
2772 | flush_work(&vi->config_work); | |
586d17c5 | 2773 | |
04486ed0 AS |
2774 | unregister_netdev(vi->dev); |
2775 | ||
2776 | remove_vq_common(vi); | |
fb6813f4 | 2777 | |
2e66f55b | 2778 | free_percpu(vi->stats); |
74b2553f | 2779 | free_netdev(vi->dev); |
296f96fc RR |
2780 | } |
2781 | ||
67a75194 | 2782 | static __maybe_unused int virtnet_freeze(struct virtio_device *vdev) |
0741bcb5 AS |
2783 | { |
2784 | struct virtnet_info *vi = vdev->priv; | |
2785 | ||
8017c279 | 2786 | virtnet_cpu_notif_remove(vi); |
9fe7bfce | 2787 | virtnet_freeze_down(vdev); |
0741bcb5 AS |
2788 | remove_vq_common(vi); |
2789 | ||
2790 | return 0; | |
2791 | } | |
2792 | ||
67a75194 | 2793 | static __maybe_unused int virtnet_restore(struct virtio_device *vdev) |
0741bcb5 AS |
2794 | { |
2795 | struct virtnet_info *vi = vdev->priv; | |
9fe7bfce | 2796 | int err; |
0741bcb5 | 2797 | |
9fe7bfce | 2798 | err = virtnet_restore_up(vdev); |
0741bcb5 AS |
2799 | if (err) |
2800 | return err; | |
986a4f4d JW |
2801 | virtnet_set_queues(vi, vi->curr_queue_pairs); |
2802 | ||
8017c279 | 2803 | err = virtnet_cpu_notif_add(vi); |
ec9debbd JW |
2804 | if (err) |
2805 | return err; | |
2806 | ||
0741bcb5 AS |
2807 | return 0; |
2808 | } | |
0741bcb5 | 2809 | |
296f96fc RR |
2810 | static struct virtio_device_id id_table[] = { |
2811 | { VIRTIO_ID_NET, VIRTIO_DEV_ANY_ID }, | |
2812 | { 0 }, | |
2813 | }; | |
2814 | ||
f3358507 MT |
2815 | #define VIRTNET_FEATURES \ |
2816 | VIRTIO_NET_F_CSUM, VIRTIO_NET_F_GUEST_CSUM, \ | |
2817 | VIRTIO_NET_F_MAC, \ | |
2818 | VIRTIO_NET_F_HOST_TSO4, VIRTIO_NET_F_HOST_UFO, VIRTIO_NET_F_HOST_TSO6, \ | |
2819 | VIRTIO_NET_F_HOST_ECN, VIRTIO_NET_F_GUEST_TSO4, VIRTIO_NET_F_GUEST_TSO6, \ | |
2820 | VIRTIO_NET_F_GUEST_ECN, VIRTIO_NET_F_GUEST_UFO, \ | |
2821 | VIRTIO_NET_F_MRG_RXBUF, VIRTIO_NET_F_STATUS, VIRTIO_NET_F_CTRL_VQ, \ | |
2822 | VIRTIO_NET_F_CTRL_RX, VIRTIO_NET_F_CTRL_VLAN, \ | |
2823 | VIRTIO_NET_F_GUEST_ANNOUNCE, VIRTIO_NET_F_MQ, \ | |
2824 | VIRTIO_NET_F_CTRL_MAC_ADDR, \ | |
3f93522f | 2825 | VIRTIO_NET_F_MTU, VIRTIO_NET_F_CTRL_GUEST_OFFLOADS |
f3358507 | 2826 | |
c45a6816 | 2827 | static unsigned int features[] = { |
f3358507 MT |
2828 | VIRTNET_FEATURES, |
2829 | }; | |
2830 | ||
2831 | static unsigned int features_legacy[] = { | |
2832 | VIRTNET_FEATURES, | |
2833 | VIRTIO_NET_F_GSO, | |
e7428e95 | 2834 | VIRTIO_F_ANY_LAYOUT, |
c45a6816 RR |
2835 | }; |
2836 | ||
22402529 | 2837 | static struct virtio_driver virtio_net_driver = { |
c45a6816 RR |
2838 | .feature_table = features, |
2839 | .feature_table_size = ARRAY_SIZE(features), | |
f3358507 MT |
2840 | .feature_table_legacy = features_legacy, |
2841 | .feature_table_size_legacy = ARRAY_SIZE(features_legacy), | |
296f96fc RR |
2842 | .driver.name = KBUILD_MODNAME, |
2843 | .driver.owner = THIS_MODULE, | |
2844 | .id_table = id_table, | |
fe36cbe0 | 2845 | .validate = virtnet_validate, |
296f96fc | 2846 | .probe = virtnet_probe, |
8cc085d6 | 2847 | .remove = virtnet_remove, |
9f4d26d0 | 2848 | .config_changed = virtnet_config_changed, |
89107000 | 2849 | #ifdef CONFIG_PM_SLEEP |
0741bcb5 AS |
2850 | .freeze = virtnet_freeze, |
2851 | .restore = virtnet_restore, | |
2852 | #endif | |
296f96fc RR |
2853 | }; |
2854 | ||
8017c279 SAS |
2855 | static __init int virtio_net_driver_init(void) |
2856 | { | |
2857 | int ret; | |
2858 | ||
73c1b41e | 2859 | ret = cpuhp_setup_state_multi(CPUHP_AP_ONLINE_DYN, "virtio/net:online", |
8017c279 SAS |
2860 | virtnet_cpu_online, |
2861 | virtnet_cpu_down_prep); | |
2862 | if (ret < 0) | |
2863 | goto out; | |
2864 | virtionet_online = ret; | |
73c1b41e | 2865 | ret = cpuhp_setup_state_multi(CPUHP_VIRT_NET_DEAD, "virtio/net:dead", |
8017c279 SAS |
2866 | NULL, virtnet_cpu_dead); |
2867 | if (ret) | |
2868 | goto err_dead; | |
2869 | ||
2870 | ret = register_virtio_driver(&virtio_net_driver); | |
2871 | if (ret) | |
2872 | goto err_virtio; | |
2873 | return 0; | |
2874 | err_virtio: | |
2875 | cpuhp_remove_multi_state(CPUHP_VIRT_NET_DEAD); | |
2876 | err_dead: | |
2877 | cpuhp_remove_multi_state(virtionet_online); | |
2878 | out: | |
2879 | return ret; | |
2880 | } | |
2881 | module_init(virtio_net_driver_init); | |
2882 | ||
2883 | static __exit void virtio_net_driver_exit(void) | |
2884 | { | |
cfa0ebc9 | 2885 | unregister_virtio_driver(&virtio_net_driver); |
8017c279 SAS |
2886 | cpuhp_remove_multi_state(CPUHP_VIRT_NET_DEAD); |
2887 | cpuhp_remove_multi_state(virtionet_online); | |
8017c279 SAS |
2888 | } |
2889 | module_exit(virtio_net_driver_exit); | |
296f96fc RR |
2890 | |
2891 | MODULE_DEVICE_TABLE(virtio, id_table); | |
2892 | MODULE_DESCRIPTION("Virtio network driver"); | |
2893 | MODULE_LICENSE("GPL"); |