]>
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 | ||
120 | struct virtnet_info { | |
121 | struct virtio_device *vdev; | |
122 | struct virtqueue *cvq; | |
123 | struct net_device *dev; | |
986a4f4d JW |
124 | struct send_queue *sq; |
125 | struct receive_queue *rq; | |
e9d7417b JW |
126 | unsigned int status; |
127 | ||
986a4f4d JW |
128 | /* Max # of queue pairs supported by the device */ |
129 | u16 max_queue_pairs; | |
130 | ||
131 | /* # of queue pairs currently used by the driver */ | |
132 | u16 curr_queue_pairs; | |
133 | ||
672aafd5 JF |
134 | /* # of XDP queue pairs currently used by the driver */ |
135 | u16 xdp_queue_pairs; | |
136 | ||
97402b96 HX |
137 | /* I like... big packets and I cannot lie! */ |
138 | bool big_packets; | |
139 | ||
3f2c31d9 MM |
140 | /* Host will merge rx buffers for big packets (shake it! shake it!) */ |
141 | bool mergeable_rx_bufs; | |
142 | ||
986a4f4d JW |
143 | /* Has control virtqueue */ |
144 | bool has_cvq; | |
145 | ||
e7428e95 MT |
146 | /* Host can handle any s/g split between our header and packet data */ |
147 | bool any_header_sg; | |
148 | ||
012873d0 MT |
149 | /* Packet virtio header size */ |
150 | u8 hdr_len; | |
151 | ||
3fa2a1df | 152 | /* Active statistics */ |
153 | struct virtnet_stats __percpu *stats; | |
154 | ||
3161e453 RR |
155 | /* Work struct for refilling if we run low on memory. */ |
156 | struct delayed_work refill; | |
157 | ||
586d17c5 JW |
158 | /* Work struct for config space updates */ |
159 | struct work_struct config_work; | |
160 | ||
986a4f4d JW |
161 | /* Does the affinity hint is set for virtqueues? */ |
162 | bool affinity_hint_set; | |
47be2479 | 163 | |
8017c279 SAS |
164 | /* CPU hotplug instances for online & dead */ |
165 | struct hlist_node node; | |
166 | struct hlist_node node_dead; | |
2ac46030 MT |
167 | |
168 | /* Control VQ buffers: protected by the rtnl lock */ | |
169 | struct virtio_net_ctrl_hdr ctrl_hdr; | |
170 | virtio_net_ctrl_ack ctrl_status; | |
a725ee3e | 171 | struct virtio_net_ctrl_mq ctrl_mq; |
2ac46030 MT |
172 | u8 ctrl_promisc; |
173 | u8 ctrl_allmulti; | |
a725ee3e | 174 | u16 ctrl_vid; |
3f93522f | 175 | u64 ctrl_offloads; |
16032be5 NA |
176 | |
177 | /* Ethtool settings */ | |
178 | u8 duplex; | |
179 | u32 speed; | |
3f93522f JW |
180 | |
181 | unsigned long guest_offloads; | |
296f96fc RR |
182 | }; |
183 | ||
9ab86bbc | 184 | struct padded_vnet_hdr { |
012873d0 | 185 | struct virtio_net_hdr_mrg_rxbuf hdr; |
9ab86bbc | 186 | /* |
012873d0 MT |
187 | * hdr is in a separate sg buffer, and data sg buffer shares same page |
188 | * with this header sg. This padding makes next sg 16 byte aligned | |
189 | * after the header. | |
9ab86bbc | 190 | */ |
012873d0 | 191 | char padding[4]; |
9ab86bbc SM |
192 | }; |
193 | ||
986a4f4d JW |
194 | /* Converting between virtqueue no. and kernel tx/rx queue no. |
195 | * 0:rx0 1:tx0 2:rx1 3:tx1 ... 2N:rxN 2N+1:txN 2N+2:cvq | |
196 | */ | |
197 | static int vq2txq(struct virtqueue *vq) | |
198 | { | |
9d0ca6ed | 199 | return (vq->index - 1) / 2; |
986a4f4d JW |
200 | } |
201 | ||
202 | static int txq2vq(int txq) | |
203 | { | |
204 | return txq * 2 + 1; | |
205 | } | |
206 | ||
207 | static int vq2rxq(struct virtqueue *vq) | |
208 | { | |
9d0ca6ed | 209 | return vq->index / 2; |
986a4f4d JW |
210 | } |
211 | ||
212 | static int rxq2vq(int rxq) | |
213 | { | |
214 | return rxq * 2; | |
215 | } | |
216 | ||
012873d0 | 217 | static inline struct virtio_net_hdr_mrg_rxbuf *skb_vnet_hdr(struct sk_buff *skb) |
296f96fc | 218 | { |
012873d0 | 219 | return (struct virtio_net_hdr_mrg_rxbuf *)skb->cb; |
296f96fc RR |
220 | } |
221 | ||
9ab86bbc SM |
222 | /* |
223 | * private is used to chain pages for big packets, put the whole | |
224 | * most recent used list in the beginning for reuse | |
225 | */ | |
e9d7417b | 226 | static void give_pages(struct receive_queue *rq, struct page *page) |
0a888fd1 | 227 | { |
9ab86bbc | 228 | struct page *end; |
0a888fd1 | 229 | |
e9d7417b | 230 | /* Find end of list, sew whole thing into vi->rq.pages. */ |
9ab86bbc | 231 | for (end = page; end->private; end = (struct page *)end->private); |
e9d7417b JW |
232 | end->private = (unsigned long)rq->pages; |
233 | rq->pages = page; | |
0a888fd1 MM |
234 | } |
235 | ||
e9d7417b | 236 | static struct page *get_a_page(struct receive_queue *rq, gfp_t gfp_mask) |
fb6813f4 | 237 | { |
e9d7417b | 238 | struct page *p = rq->pages; |
fb6813f4 | 239 | |
9ab86bbc | 240 | if (p) { |
e9d7417b | 241 | rq->pages = (struct page *)p->private; |
9ab86bbc SM |
242 | /* clear private here, it is used to chain pages */ |
243 | p->private = 0; | |
244 | } else | |
fb6813f4 RR |
245 | p = alloc_page(gfp_mask); |
246 | return p; | |
247 | } | |
248 | ||
e4e8452a WB |
249 | static void virtqueue_napi_schedule(struct napi_struct *napi, |
250 | struct virtqueue *vq) | |
251 | { | |
252 | if (napi_schedule_prep(napi)) { | |
253 | virtqueue_disable_cb(vq); | |
254 | __napi_schedule(napi); | |
255 | } | |
256 | } | |
257 | ||
258 | static void virtqueue_napi_complete(struct napi_struct *napi, | |
259 | struct virtqueue *vq, int processed) | |
260 | { | |
261 | int opaque; | |
262 | ||
263 | opaque = virtqueue_enable_cb_prepare(vq); | |
264 | if (napi_complete_done(napi, processed) && | |
265 | unlikely(virtqueue_poll(vq, opaque))) | |
266 | virtqueue_napi_schedule(napi, vq); | |
267 | } | |
268 | ||
e9d7417b | 269 | static void skb_xmit_done(struct virtqueue *vq) |
296f96fc | 270 | { |
e9d7417b | 271 | struct virtnet_info *vi = vq->vdev->priv; |
b92f1e67 | 272 | struct napi_struct *napi = &vi->sq[vq2txq(vq)].napi; |
296f96fc | 273 | |
2cb9c6ba | 274 | /* Suppress further interrupts. */ |
e9d7417b | 275 | virtqueue_disable_cb(vq); |
11a3a154 | 276 | |
b92f1e67 WB |
277 | if (napi->weight) |
278 | virtqueue_napi_schedule(napi, vq); | |
279 | else | |
280 | /* We were probably waiting for more output buffers. */ | |
281 | netif_wake_subqueue(vi->dev, vq2txq(vq)); | |
296f96fc RR |
282 | } |
283 | ||
28b39bc7 JW |
284 | #define MRG_CTX_HEADER_SHIFT 22 |
285 | static void *mergeable_len_to_ctx(unsigned int truesize, | |
286 | unsigned int headroom) | |
287 | { | |
288 | return (void *)(unsigned long)((headroom << MRG_CTX_HEADER_SHIFT) | truesize); | |
289 | } | |
290 | ||
291 | static unsigned int mergeable_ctx_to_headroom(void *mrg_ctx) | |
292 | { | |
293 | return (unsigned long)mrg_ctx >> MRG_CTX_HEADER_SHIFT; | |
294 | } | |
295 | ||
296 | static unsigned int mergeable_ctx_to_truesize(void *mrg_ctx) | |
297 | { | |
298 | return (unsigned long)mrg_ctx & ((1 << MRG_CTX_HEADER_SHIFT) - 1); | |
299 | } | |
300 | ||
3464645a | 301 | /* Called from bottom half context */ |
946fa564 MT |
302 | static struct sk_buff *page_to_skb(struct virtnet_info *vi, |
303 | struct receive_queue *rq, | |
2613af0e MD |
304 | struct page *page, unsigned int offset, |
305 | unsigned int len, unsigned int truesize) | |
9ab86bbc SM |
306 | { |
307 | struct sk_buff *skb; | |
012873d0 | 308 | struct virtio_net_hdr_mrg_rxbuf *hdr; |
2613af0e | 309 | unsigned int copy, hdr_len, hdr_padded_len; |
9ab86bbc | 310 | char *p; |
fb6813f4 | 311 | |
2613af0e | 312 | p = page_address(page) + offset; |
3f2c31d9 | 313 | |
9ab86bbc | 314 | /* copy small packet so we can reuse these pages for small data */ |
c67f5db8 | 315 | skb = napi_alloc_skb(&rq->napi, GOOD_COPY_LEN); |
9ab86bbc SM |
316 | if (unlikely(!skb)) |
317 | return NULL; | |
3f2c31d9 | 318 | |
9ab86bbc | 319 | hdr = skb_vnet_hdr(skb); |
3f2c31d9 | 320 | |
012873d0 MT |
321 | hdr_len = vi->hdr_len; |
322 | if (vi->mergeable_rx_bufs) | |
a4a76503 | 323 | hdr_padded_len = sizeof(*hdr); |
012873d0 | 324 | else |
2613af0e | 325 | hdr_padded_len = sizeof(struct padded_vnet_hdr); |
3f2c31d9 | 326 | |
9ab86bbc | 327 | memcpy(hdr, p, hdr_len); |
3f2c31d9 | 328 | |
9ab86bbc | 329 | len -= hdr_len; |
2613af0e MD |
330 | offset += hdr_padded_len; |
331 | p += hdr_padded_len; | |
3f2c31d9 | 332 | |
9ab86bbc SM |
333 | copy = len; |
334 | if (copy > skb_tailroom(skb)) | |
335 | copy = skb_tailroom(skb); | |
59ae1d12 | 336 | skb_put_data(skb, p, copy); |
3f2c31d9 | 337 | |
9ab86bbc SM |
338 | len -= copy; |
339 | offset += copy; | |
3f2c31d9 | 340 | |
2613af0e MD |
341 | if (vi->mergeable_rx_bufs) { |
342 | if (len) | |
343 | skb_add_rx_frag(skb, 0, page, offset, len, truesize); | |
344 | else | |
345 | put_page(page); | |
346 | return skb; | |
347 | } | |
348 | ||
e878d78b SL |
349 | /* |
350 | * Verify that we can indeed put this data into a skb. | |
351 | * This is here to handle cases when the device erroneously | |
352 | * tries to receive more than is possible. This is usually | |
353 | * the case of a broken device. | |
354 | */ | |
355 | if (unlikely(len > MAX_SKB_FRAGS * PAGE_SIZE)) { | |
be443899 | 356 | net_dbg_ratelimited("%s: too much data\n", skb->dev->name); |
e878d78b SL |
357 | dev_kfree_skb(skb); |
358 | return NULL; | |
359 | } | |
2613af0e | 360 | BUG_ON(offset >= PAGE_SIZE); |
9ab86bbc | 361 | while (len) { |
2613af0e MD |
362 | unsigned int frag_size = min((unsigned)PAGE_SIZE - offset, len); |
363 | skb_add_rx_frag(skb, skb_shinfo(skb)->nr_frags, page, offset, | |
364 | frag_size, truesize); | |
365 | len -= frag_size; | |
9ab86bbc SM |
366 | page = (struct page *)page->private; |
367 | offset = 0; | |
368 | } | |
3f2c31d9 | 369 | |
9ab86bbc | 370 | if (page) |
e9d7417b | 371 | give_pages(rq, page); |
3f2c31d9 | 372 | |
9ab86bbc SM |
373 | return skb; |
374 | } | |
3f2c31d9 | 375 | |
186b3c99 JW |
376 | static void virtnet_xdp_flush(struct net_device *dev) |
377 | { | |
378 | struct virtnet_info *vi = netdev_priv(dev); | |
379 | struct send_queue *sq; | |
380 | unsigned int qp; | |
381 | ||
382 | qp = vi->curr_queue_pairs - vi->xdp_queue_pairs + smp_processor_id(); | |
383 | sq = &vi->sq[qp]; | |
384 | ||
385 | virtqueue_kick(sq->vq); | |
386 | } | |
387 | ||
388 | static bool __virtnet_xdp_xmit(struct virtnet_info *vi, | |
389 | struct xdp_buff *xdp) | |
56434a01 | 390 | { |
56434a01 | 391 | struct virtio_net_hdr_mrg_rxbuf *hdr; |
f6b10209 | 392 | unsigned int len; |
722d8283 JF |
393 | struct send_queue *sq; |
394 | unsigned int qp; | |
56434a01 JF |
395 | void *xdp_sent; |
396 | int err; | |
397 | ||
722d8283 JF |
398 | qp = vi->curr_queue_pairs - vi->xdp_queue_pairs + smp_processor_id(); |
399 | sq = &vi->sq[qp]; | |
400 | ||
56434a01 JF |
401 | /* Free up any pending old buffers before queueing new ones. */ |
402 | while ((xdp_sent = virtqueue_get_buf(sq->vq, &len)) != NULL) { | |
f6b10209 | 403 | struct page *sent_page = virt_to_head_page(xdp_sent); |
bb91accf | 404 | |
f6b10209 | 405 | put_page(sent_page); |
bb91accf | 406 | } |
56434a01 | 407 | |
f6b10209 JW |
408 | xdp->data -= vi->hdr_len; |
409 | /* Zero header and leave csum up to XDP layers */ | |
410 | hdr = xdp->data; | |
411 | memset(hdr, 0, vi->hdr_len); | |
bb91accf | 412 | |
f6b10209 | 413 | sg_init_one(sq->sg, xdp->data, xdp->data_end - xdp->data); |
bb91accf | 414 | |
f6b10209 | 415 | err = virtqueue_add_outbuf(sq->vq, sq->sg, 1, xdp->data, GFP_ATOMIC); |
56434a01 | 416 | if (unlikely(err)) { |
f6b10209 | 417 | struct page *page = virt_to_head_page(xdp->data); |
bb91accf | 418 | |
f6b10209 | 419 | put_page(page); |
a67edbf4 | 420 | return false; |
56434a01 JF |
421 | } |
422 | ||
a67edbf4 | 423 | return true; |
56434a01 JF |
424 | } |
425 | ||
186b3c99 JW |
426 | static int virtnet_xdp_xmit(struct net_device *dev, struct xdp_buff *xdp) |
427 | { | |
428 | struct virtnet_info *vi = netdev_priv(dev); | |
429 | bool sent = __virtnet_xdp_xmit(vi, xdp); | |
430 | ||
431 | if (!sent) | |
432 | return -ENOSPC; | |
433 | return 0; | |
434 | } | |
435 | ||
f6b10209 JW |
436 | static unsigned int virtnet_get_headroom(struct virtnet_info *vi) |
437 | { | |
438 | return vi->xdp_queue_pairs ? VIRTIO_XDP_HEADROOM : 0; | |
439 | } | |
440 | ||
4941d472 JW |
441 | /* We copy the packet for XDP in the following cases: |
442 | * | |
443 | * 1) Packet is scattered across multiple rx buffers. | |
444 | * 2) Headroom space is insufficient. | |
445 | * | |
446 | * This is inefficient but it's a temporary condition that | |
447 | * we hit right after XDP is enabled and until queue is refilled | |
448 | * with large buffers with sufficient headroom - so it should affect | |
449 | * at most queue size packets. | |
450 | * Afterwards, the conditions to enable | |
451 | * XDP should preclude the underlying device from sending packets | |
452 | * across multiple buffers (num_buf > 1), and we make sure buffers | |
453 | * have enough headroom. | |
454 | */ | |
455 | static struct page *xdp_linearize_page(struct receive_queue *rq, | |
456 | u16 *num_buf, | |
457 | struct page *p, | |
458 | int offset, | |
459 | int page_off, | |
460 | unsigned int *len) | |
461 | { | |
462 | struct page *page = alloc_page(GFP_ATOMIC); | |
463 | ||
464 | if (!page) | |
465 | return NULL; | |
466 | ||
467 | memcpy(page_address(page) + page_off, page_address(p) + offset, *len); | |
468 | page_off += *len; | |
469 | ||
470 | while (--*num_buf) { | |
471 | unsigned int buflen; | |
472 | void *buf; | |
473 | int off; | |
474 | ||
475 | buf = virtqueue_get_buf(rq->vq, &buflen); | |
476 | if (unlikely(!buf)) | |
477 | goto err_buf; | |
478 | ||
479 | p = virt_to_head_page(buf); | |
480 | off = buf - page_address(p); | |
481 | ||
482 | /* guard against a misconfigured or uncooperative backend that | |
483 | * is sending packet larger than the MTU. | |
484 | */ | |
485 | if ((page_off + buflen) > PAGE_SIZE) { | |
486 | put_page(p); | |
487 | goto err_buf; | |
488 | } | |
489 | ||
490 | memcpy(page_address(page) + page_off, | |
491 | page_address(p) + off, buflen); | |
492 | page_off += buflen; | |
493 | put_page(p); | |
494 | } | |
495 | ||
496 | /* Headroom does not contribute to packet length */ | |
497 | *len = page_off - VIRTIO_XDP_HEADROOM; | |
498 | return page; | |
499 | err_buf: | |
500 | __free_pages(page, 0); | |
501 | return NULL; | |
502 | } | |
503 | ||
bb91accf JW |
504 | static struct sk_buff *receive_small(struct net_device *dev, |
505 | struct virtnet_info *vi, | |
506 | struct receive_queue *rq, | |
192f68cf | 507 | void *buf, void *ctx, |
186b3c99 JW |
508 | unsigned int len, |
509 | bool *xdp_xmit) | |
f121159d | 510 | { |
f6b10209 | 511 | struct sk_buff *skb; |
bb91accf | 512 | struct bpf_prog *xdp_prog; |
4941d472 | 513 | unsigned int xdp_headroom = (unsigned long)ctx; |
f6b10209 JW |
514 | unsigned int header_offset = VIRTNET_RX_PAD + xdp_headroom; |
515 | unsigned int headroom = vi->hdr_len + header_offset; | |
516 | unsigned int buflen = SKB_DATA_ALIGN(GOOD_PACKET_LEN + headroom) + | |
517 | SKB_DATA_ALIGN(sizeof(struct skb_shared_info)); | |
4941d472 | 518 | struct page *page = virt_to_head_page(buf); |
186b3c99 | 519 | unsigned int delta = 0, err; |
4941d472 | 520 | struct page *xdp_page; |
012873d0 | 521 | len -= vi->hdr_len; |
f121159d | 522 | |
bb91accf JW |
523 | rcu_read_lock(); |
524 | xdp_prog = rcu_dereference(rq->xdp_prog); | |
525 | if (xdp_prog) { | |
f6b10209 | 526 | struct virtio_net_hdr_mrg_rxbuf *hdr = buf + header_offset; |
0354e4d1 | 527 | struct xdp_buff xdp; |
f6b10209 | 528 | void *orig_data; |
bb91accf JW |
529 | u32 act; |
530 | ||
531 | if (unlikely(hdr->hdr.gso_type || hdr->hdr.flags)) | |
532 | goto err_xdp; | |
0354e4d1 | 533 | |
4941d472 JW |
534 | if (unlikely(xdp_headroom < virtnet_get_headroom(vi))) { |
535 | int offset = buf - page_address(page) + header_offset; | |
536 | unsigned int tlen = len + vi->hdr_len; | |
537 | u16 num_buf = 1; | |
538 | ||
539 | xdp_headroom = virtnet_get_headroom(vi); | |
540 | header_offset = VIRTNET_RX_PAD + xdp_headroom; | |
541 | headroom = vi->hdr_len + header_offset; | |
542 | buflen = SKB_DATA_ALIGN(GOOD_PACKET_LEN + headroom) + | |
543 | SKB_DATA_ALIGN(sizeof(struct skb_shared_info)); | |
544 | xdp_page = xdp_linearize_page(rq, &num_buf, page, | |
545 | offset, header_offset, | |
546 | &tlen); | |
547 | if (!xdp_page) | |
548 | goto err_xdp; | |
549 | ||
550 | buf = page_address(xdp_page); | |
551 | put_page(page); | |
552 | page = xdp_page; | |
553 | } | |
554 | ||
f6b10209 JW |
555 | xdp.data_hard_start = buf + VIRTNET_RX_PAD + vi->hdr_len; |
556 | xdp.data = xdp.data_hard_start + xdp_headroom; | |
de8f3a83 | 557 | xdp_set_data_meta_invalid(&xdp); |
0354e4d1 | 558 | xdp.data_end = xdp.data + len; |
f6b10209 | 559 | orig_data = xdp.data; |
0354e4d1 JF |
560 | act = bpf_prog_run_xdp(xdp_prog, &xdp); |
561 | ||
bb91accf JW |
562 | switch (act) { |
563 | case XDP_PASS: | |
2de2f7f4 | 564 | /* Recalculate length in case bpf program changed it */ |
f6b10209 | 565 | delta = orig_data - xdp.data; |
bb91accf JW |
566 | break; |
567 | case XDP_TX: | |
186b3c99 | 568 | if (unlikely(!__virtnet_xdp_xmit(vi, &xdp))) |
0354e4d1 | 569 | trace_xdp_exception(vi->dev, xdp_prog, act); |
186b3c99 JW |
570 | else |
571 | *xdp_xmit = true; | |
572 | rcu_read_unlock(); | |
573 | goto xdp_xmit; | |
574 | case XDP_REDIRECT: | |
575 | err = xdp_do_redirect(dev, &xdp, xdp_prog); | |
576 | if (!err) | |
577 | *xdp_xmit = true; | |
bb91accf JW |
578 | rcu_read_unlock(); |
579 | goto xdp_xmit; | |
bb91accf | 580 | default: |
0354e4d1 JF |
581 | bpf_warn_invalid_xdp_action(act); |
582 | case XDP_ABORTED: | |
583 | trace_xdp_exception(vi->dev, xdp_prog, act); | |
584 | case XDP_DROP: | |
bb91accf JW |
585 | goto err_xdp; |
586 | } | |
587 | } | |
588 | rcu_read_unlock(); | |
589 | ||
f6b10209 JW |
590 | skb = build_skb(buf, buflen); |
591 | if (!skb) { | |
4941d472 | 592 | put_page(page); |
f6b10209 JW |
593 | goto err; |
594 | } | |
595 | skb_reserve(skb, headroom - delta); | |
596 | skb_put(skb, len + delta); | |
597 | if (!delta) { | |
598 | buf += header_offset; | |
599 | memcpy(skb_vnet_hdr(skb), buf, vi->hdr_len); | |
600 | } /* keep zeroed vnet hdr since packet was changed by bpf */ | |
601 | ||
602 | err: | |
f121159d | 603 | return skb; |
bb91accf JW |
604 | |
605 | err_xdp: | |
606 | rcu_read_unlock(); | |
607 | dev->stats.rx_dropped++; | |
4941d472 | 608 | put_page(page); |
bb91accf JW |
609 | xdp_xmit: |
610 | return NULL; | |
f121159d MT |
611 | } |
612 | ||
613 | static struct sk_buff *receive_big(struct net_device *dev, | |
946fa564 | 614 | struct virtnet_info *vi, |
f121159d MT |
615 | struct receive_queue *rq, |
616 | void *buf, | |
617 | unsigned int len) | |
618 | { | |
619 | struct page *page = buf; | |
c47a43d3 | 620 | struct sk_buff *skb = page_to_skb(vi, rq, page, 0, len, PAGE_SIZE); |
f600b690 | 621 | |
f121159d MT |
622 | if (unlikely(!skb)) |
623 | goto err; | |
624 | ||
625 | return skb; | |
626 | ||
627 | err: | |
628 | dev->stats.rx_dropped++; | |
629 | give_pages(rq, page); | |
630 | return NULL; | |
631 | } | |
632 | ||
8fc3b9e9 | 633 | static struct sk_buff *receive_mergeable(struct net_device *dev, |
fdd819b2 | 634 | struct virtnet_info *vi, |
8fc3b9e9 | 635 | struct receive_queue *rq, |
680557cf MT |
636 | void *buf, |
637 | void *ctx, | |
186b3c99 JW |
638 | unsigned int len, |
639 | bool *xdp_xmit) | |
9ab86bbc | 640 | { |
012873d0 MT |
641 | struct virtio_net_hdr_mrg_rxbuf *hdr = buf; |
642 | u16 num_buf = virtio16_to_cpu(vi->vdev, hdr->num_buffers); | |
8fc3b9e9 MT |
643 | struct page *page = virt_to_head_page(buf); |
644 | int offset = buf - page_address(page); | |
f600b690 JF |
645 | struct sk_buff *head_skb, *curr_skb; |
646 | struct bpf_prog *xdp_prog; | |
647 | unsigned int truesize; | |
4941d472 | 648 | unsigned int headroom = mergeable_ctx_to_headroom(ctx); |
186b3c99 | 649 | int err; |
f600b690 | 650 | |
56434a01 JF |
651 | head_skb = NULL; |
652 | ||
f600b690 JF |
653 | rcu_read_lock(); |
654 | xdp_prog = rcu_dereference(rq->xdp_prog); | |
655 | if (xdp_prog) { | |
72979a6c | 656 | struct page *xdp_page; |
0354e4d1 | 657 | struct xdp_buff xdp; |
0354e4d1 | 658 | void *data; |
f600b690 JF |
659 | u32 act; |
660 | ||
73b62bd0 | 661 | /* This happens when rx buffer size is underestimated */ |
4941d472 JW |
662 | if (unlikely(num_buf > 1 || |
663 | headroom < virtnet_get_headroom(vi))) { | |
72979a6c | 664 | /* linearize data for XDP */ |
56a86f84 | 665 | xdp_page = xdp_linearize_page(rq, &num_buf, |
4941d472 JW |
666 | page, offset, |
667 | VIRTIO_XDP_HEADROOM, | |
668 | &len); | |
72979a6c JF |
669 | if (!xdp_page) |
670 | goto err_xdp; | |
2de2f7f4 | 671 | offset = VIRTIO_XDP_HEADROOM; |
72979a6c JF |
672 | } else { |
673 | xdp_page = page; | |
f600b690 JF |
674 | } |
675 | ||
676 | /* Transient failure which in theory could occur if | |
677 | * in-flight packets from before XDP was enabled reach | |
678 | * the receive path after XDP is loaded. In practice I | |
679 | * was not able to create this condition. | |
680 | */ | |
b00f70b0 | 681 | if (unlikely(hdr->hdr.gso_type)) |
f600b690 JF |
682 | goto err_xdp; |
683 | ||
2de2f7f4 JF |
684 | /* Allow consuming headroom but reserve enough space to push |
685 | * the descriptor on if we get an XDP_TX return code. | |
686 | */ | |
0354e4d1 | 687 | data = page_address(xdp_page) + offset; |
2de2f7f4 | 688 | xdp.data_hard_start = data - VIRTIO_XDP_HEADROOM + vi->hdr_len; |
0354e4d1 | 689 | xdp.data = data + vi->hdr_len; |
de8f3a83 | 690 | xdp_set_data_meta_invalid(&xdp); |
0354e4d1 JF |
691 | xdp.data_end = xdp.data + (len - vi->hdr_len); |
692 | act = bpf_prog_run_xdp(xdp_prog, &xdp); | |
693 | ||
31240345 JW |
694 | if (act != XDP_PASS) |
695 | ewma_pkt_len_add(&rq->mrg_avg_pkt_len, len); | |
696 | ||
56434a01 JF |
697 | switch (act) { |
698 | case XDP_PASS: | |
2de2f7f4 JF |
699 | /* recalculate offset to account for any header |
700 | * adjustments. Note other cases do not build an | |
701 | * skb and avoid using offset | |
702 | */ | |
703 | offset = xdp.data - | |
704 | page_address(xdp_page) - vi->hdr_len; | |
705 | ||
1830f893 JW |
706 | /* We can only create skb based on xdp_page. */ |
707 | if (unlikely(xdp_page != page)) { | |
708 | rcu_read_unlock(); | |
709 | put_page(page); | |
710 | head_skb = page_to_skb(vi, rq, xdp_page, | |
2de2f7f4 | 711 | offset, len, PAGE_SIZE); |
1830f893 JW |
712 | return head_skb; |
713 | } | |
56434a01 JF |
714 | break; |
715 | case XDP_TX: | |
186b3c99 | 716 | if (unlikely(!__virtnet_xdp_xmit(vi, &xdp))) |
0354e4d1 | 717 | trace_xdp_exception(vi->dev, xdp_prog, act); |
186b3c99 JW |
718 | else |
719 | *xdp_xmit = true; | |
72979a6c JF |
720 | if (unlikely(xdp_page != page)) |
721 | goto err_xdp; | |
56434a01 JF |
722 | rcu_read_unlock(); |
723 | goto xdp_xmit; | |
186b3c99 JW |
724 | case XDP_REDIRECT: |
725 | err = xdp_do_redirect(dev, &xdp, xdp_prog); | |
dd543797 | 726 | if (!err) |
186b3c99 JW |
727 | *xdp_xmit = true; |
728 | rcu_read_unlock(); | |
729 | goto xdp_xmit; | |
56434a01 | 730 | default: |
0354e4d1 JF |
731 | bpf_warn_invalid_xdp_action(act); |
732 | case XDP_ABORTED: | |
733 | trace_xdp_exception(vi->dev, xdp_prog, act); | |
734 | case XDP_DROP: | |
72979a6c JF |
735 | if (unlikely(xdp_page != page)) |
736 | __free_pages(xdp_page, 0); | |
f600b690 | 737 | goto err_xdp; |
56434a01 | 738 | } |
f600b690 JF |
739 | } |
740 | rcu_read_unlock(); | |
ab7db917 | 741 | |
28b39bc7 JW |
742 | truesize = mergeable_ctx_to_truesize(ctx); |
743 | if (unlikely(len > truesize)) { | |
56da5fd0 | 744 | pr_debug("%s: rx error: len %u exceeds truesize %lu\n", |
680557cf MT |
745 | dev->name, len, (unsigned long)ctx); |
746 | dev->stats.rx_length_errors++; | |
747 | goto err_skb; | |
748 | } | |
28b39bc7 | 749 | |
f600b690 JF |
750 | head_skb = page_to_skb(vi, rq, page, offset, len, truesize); |
751 | curr_skb = head_skb; | |
9ab86bbc | 752 | |
8fc3b9e9 MT |
753 | if (unlikely(!curr_skb)) |
754 | goto err_skb; | |
9ab86bbc | 755 | while (--num_buf) { |
8fc3b9e9 MT |
756 | int num_skb_frags; |
757 | ||
680557cf | 758 | buf = virtqueue_get_buf_ctx(rq->vq, &len, &ctx); |
ab7db917 | 759 | if (unlikely(!ctx)) { |
8fc3b9e9 | 760 | pr_debug("%s: rx error: %d buffers out of %d missing\n", |
fdd819b2 | 761 | dev->name, num_buf, |
012873d0 MT |
762 | virtio16_to_cpu(vi->vdev, |
763 | hdr->num_buffers)); | |
8fc3b9e9 MT |
764 | dev->stats.rx_length_errors++; |
765 | goto err_buf; | |
3f2c31d9 | 766 | } |
8fc3b9e9 MT |
767 | |
768 | page = virt_to_head_page(buf); | |
28b39bc7 JW |
769 | |
770 | truesize = mergeable_ctx_to_truesize(ctx); | |
771 | if (unlikely(len > truesize)) { | |
56da5fd0 | 772 | pr_debug("%s: rx error: len %u exceeds truesize %lu\n", |
680557cf MT |
773 | dev->name, len, (unsigned long)ctx); |
774 | dev->stats.rx_length_errors++; | |
775 | goto err_skb; | |
776 | } | |
8fc3b9e9 MT |
777 | |
778 | num_skb_frags = skb_shinfo(curr_skb)->nr_frags; | |
2613af0e MD |
779 | if (unlikely(num_skb_frags == MAX_SKB_FRAGS)) { |
780 | struct sk_buff *nskb = alloc_skb(0, GFP_ATOMIC); | |
8fc3b9e9 MT |
781 | |
782 | if (unlikely(!nskb)) | |
783 | goto err_skb; | |
2613af0e MD |
784 | if (curr_skb == head_skb) |
785 | skb_shinfo(curr_skb)->frag_list = nskb; | |
786 | else | |
787 | curr_skb->next = nskb; | |
788 | curr_skb = nskb; | |
789 | head_skb->truesize += nskb->truesize; | |
790 | num_skb_frags = 0; | |
791 | } | |
792 | if (curr_skb != head_skb) { | |
793 | head_skb->data_len += len; | |
794 | head_skb->len += len; | |
fb51879d | 795 | head_skb->truesize += truesize; |
2613af0e | 796 | } |
8fc3b9e9 | 797 | offset = buf - page_address(page); |
ba275241 JW |
798 | if (skb_can_coalesce(curr_skb, num_skb_frags, page, offset)) { |
799 | put_page(page); | |
800 | skb_coalesce_rx_frag(curr_skb, num_skb_frags - 1, | |
fb51879d | 801 | len, truesize); |
ba275241 JW |
802 | } else { |
803 | skb_add_rx_frag(curr_skb, num_skb_frags, page, | |
fb51879d | 804 | offset, len, truesize); |
ba275241 | 805 | } |
8fc3b9e9 MT |
806 | } |
807 | ||
5377d758 | 808 | ewma_pkt_len_add(&rq->mrg_avg_pkt_len, head_skb->len); |
8fc3b9e9 MT |
809 | return head_skb; |
810 | ||
f600b690 JF |
811 | err_xdp: |
812 | rcu_read_unlock(); | |
8fc3b9e9 MT |
813 | err_skb: |
814 | put_page(page); | |
815 | while (--num_buf) { | |
680557cf MT |
816 | buf = virtqueue_get_buf(rq->vq, &len); |
817 | if (unlikely(!buf)) { | |
8fc3b9e9 MT |
818 | pr_debug("%s: rx error: %d buffers missing\n", |
819 | dev->name, num_buf); | |
820 | dev->stats.rx_length_errors++; | |
821 | break; | |
822 | } | |
680557cf | 823 | page = virt_to_head_page(buf); |
8fc3b9e9 | 824 | put_page(page); |
9ab86bbc | 825 | } |
8fc3b9e9 MT |
826 | err_buf: |
827 | dev->stats.rx_dropped++; | |
828 | dev_kfree_skb(head_skb); | |
56434a01 | 829 | xdp_xmit: |
8fc3b9e9 | 830 | return NULL; |
9ab86bbc SM |
831 | } |
832 | ||
61845d20 | 833 | static int receive_buf(struct virtnet_info *vi, struct receive_queue *rq, |
186b3c99 | 834 | void *buf, unsigned int len, void **ctx, bool *xdp_xmit) |
9ab86bbc | 835 | { |
e9d7417b | 836 | struct net_device *dev = vi->dev; |
9ab86bbc | 837 | struct sk_buff *skb; |
012873d0 | 838 | struct virtio_net_hdr_mrg_rxbuf *hdr; |
61845d20 | 839 | int ret; |
3f2c31d9 | 840 | |
bcff3162 | 841 | if (unlikely(len < vi->hdr_len + ETH_HLEN)) { |
9ab86bbc SM |
842 | pr_debug("%s: short packet %i\n", dev->name, len); |
843 | dev->stats.rx_length_errors++; | |
ab7db917 | 844 | if (vi->mergeable_rx_bufs) { |
680557cf | 845 | put_page(virt_to_head_page(buf)); |
ab7db917 | 846 | } else if (vi->big_packets) { |
98bfd23c | 847 | give_pages(rq, buf); |
ab7db917 | 848 | } else { |
f6b10209 | 849 | put_page(virt_to_head_page(buf)); |
ab7db917 | 850 | } |
61845d20 | 851 | return 0; |
9ab86bbc | 852 | } |
3f2c31d9 | 853 | |
f121159d | 854 | if (vi->mergeable_rx_bufs) |
186b3c99 | 855 | skb = receive_mergeable(dev, vi, rq, buf, ctx, len, xdp_xmit); |
f121159d | 856 | else if (vi->big_packets) |
946fa564 | 857 | skb = receive_big(dev, vi, rq, buf, len); |
f121159d | 858 | else |
186b3c99 | 859 | skb = receive_small(dev, vi, rq, buf, ctx, len, xdp_xmit); |
f121159d MT |
860 | |
861 | if (unlikely(!skb)) | |
61845d20 | 862 | return 0; |
3f2c31d9 | 863 | |
9ab86bbc | 864 | hdr = skb_vnet_hdr(skb); |
3fa2a1df | 865 | |
61845d20 | 866 | ret = skb->len; |
296f96fc | 867 | |
e858fae2 | 868 | if (hdr->hdr.flags & VIRTIO_NET_HDR_F_DATA_VALID) |
10a8d94a | 869 | skb->ip_summed = CHECKSUM_UNNECESSARY; |
296f96fc | 870 | |
e858fae2 MR |
871 | if (virtio_net_hdr_to_skb(skb, &hdr->hdr, |
872 | virtio_is_little_endian(vi->vdev))) { | |
873 | net_warn_ratelimited("%s: bad gso: type: %u, size: %u\n", | |
874 | dev->name, hdr->hdr.gso_type, | |
875 | hdr->hdr.gso_size); | |
876 | goto frame_err; | |
296f96fc RR |
877 | } |
878 | ||
d1dc06dc MR |
879 | skb->protocol = eth_type_trans(skb, dev); |
880 | pr_debug("Receiving skb proto 0x%04x len %i type %i\n", | |
881 | ntohs(skb->protocol), skb->len, skb->pkt_type); | |
882 | ||
0fbd050a | 883 | napi_gro_receive(&rq->napi, skb); |
61845d20 | 884 | return ret; |
296f96fc RR |
885 | |
886 | frame_err: | |
887 | dev->stats.rx_frame_errors++; | |
296f96fc | 888 | dev_kfree_skb(skb); |
61845d20 | 889 | return 0; |
296f96fc RR |
890 | } |
891 | ||
192f68cf JW |
892 | /* Unlike mergeable buffers, all buffers are allocated to the |
893 | * same size, except for the headroom. For this reason we do | |
894 | * not need to use mergeable_len_to_ctx here - it is enough | |
895 | * to store the headroom as the context ignoring the truesize. | |
896 | */ | |
946fa564 MT |
897 | static int add_recvbuf_small(struct virtnet_info *vi, struct receive_queue *rq, |
898 | gfp_t gfp) | |
296f96fc | 899 | { |
f6b10209 JW |
900 | struct page_frag *alloc_frag = &rq->alloc_frag; |
901 | char *buf; | |
2de2f7f4 | 902 | unsigned int xdp_headroom = virtnet_get_headroom(vi); |
192f68cf | 903 | void *ctx = (void *)(unsigned long)xdp_headroom; |
f6b10209 | 904 | int len = vi->hdr_len + VIRTNET_RX_PAD + GOOD_PACKET_LEN + xdp_headroom; |
9ab86bbc | 905 | int err; |
3f2c31d9 | 906 | |
f6b10209 JW |
907 | len = SKB_DATA_ALIGN(len) + |
908 | SKB_DATA_ALIGN(sizeof(struct skb_shared_info)); | |
909 | if (unlikely(!skb_page_frag_refill(len, alloc_frag, gfp))) | |
9ab86bbc | 910 | return -ENOMEM; |
296f96fc | 911 | |
f6b10209 JW |
912 | buf = (char *)page_address(alloc_frag->page) + alloc_frag->offset; |
913 | get_page(alloc_frag->page); | |
914 | alloc_frag->offset += len; | |
915 | sg_init_one(rq->sg, buf + VIRTNET_RX_PAD + xdp_headroom, | |
916 | vi->hdr_len + GOOD_PACKET_LEN); | |
192f68cf | 917 | err = virtqueue_add_inbuf_ctx(rq->vq, rq->sg, 1, buf, ctx, gfp); |
9ab86bbc | 918 | if (err < 0) |
f6b10209 | 919 | put_page(virt_to_head_page(buf)); |
9ab86bbc SM |
920 | return err; |
921 | } | |
97402b96 | 922 | |
012873d0 MT |
923 | static int add_recvbuf_big(struct virtnet_info *vi, struct receive_queue *rq, |
924 | gfp_t gfp) | |
9ab86bbc | 925 | { |
9ab86bbc SM |
926 | struct page *first, *list = NULL; |
927 | char *p; | |
928 | int i, err, offset; | |
929 | ||
a5835440 RR |
930 | sg_init_table(rq->sg, MAX_SKB_FRAGS + 2); |
931 | ||
e9d7417b | 932 | /* page in rq->sg[MAX_SKB_FRAGS + 1] is list tail */ |
9ab86bbc | 933 | for (i = MAX_SKB_FRAGS + 1; i > 1; --i) { |
e9d7417b | 934 | first = get_a_page(rq, gfp); |
9ab86bbc SM |
935 | if (!first) { |
936 | if (list) | |
e9d7417b | 937 | give_pages(rq, list); |
9ab86bbc | 938 | return -ENOMEM; |
97402b96 | 939 | } |
e9d7417b | 940 | sg_set_buf(&rq->sg[i], page_address(first), PAGE_SIZE); |
97402b96 | 941 | |
9ab86bbc SM |
942 | /* chain new page in list head to match sg */ |
943 | first->private = (unsigned long)list; | |
944 | list = first; | |
945 | } | |
296f96fc | 946 | |
e9d7417b | 947 | first = get_a_page(rq, gfp); |
9ab86bbc | 948 | if (!first) { |
e9d7417b | 949 | give_pages(rq, list); |
9ab86bbc SM |
950 | return -ENOMEM; |
951 | } | |
952 | p = page_address(first); | |
953 | ||
e9d7417b | 954 | /* rq->sg[0], rq->sg[1] share the same page */ |
012873d0 MT |
955 | /* a separated rq->sg[0] for header - required in case !any_header_sg */ |
956 | sg_set_buf(&rq->sg[0], p, vi->hdr_len); | |
9ab86bbc | 957 | |
e9d7417b | 958 | /* rq->sg[1] for data packet, from offset */ |
9ab86bbc | 959 | offset = sizeof(struct padded_vnet_hdr); |
e9d7417b | 960 | sg_set_buf(&rq->sg[1], p + offset, PAGE_SIZE - offset); |
9ab86bbc SM |
961 | |
962 | /* chain first in list head */ | |
963 | first->private = (unsigned long)list; | |
9dc7b9e4 RR |
964 | err = virtqueue_add_inbuf(rq->vq, rq->sg, MAX_SKB_FRAGS + 2, |
965 | first, gfp); | |
9ab86bbc | 966 | if (err < 0) |
e9d7417b | 967 | give_pages(rq, first); |
9ab86bbc SM |
968 | |
969 | return err; | |
296f96fc RR |
970 | } |
971 | ||
d85b758f MT |
972 | static unsigned int get_mergeable_buf_len(struct receive_queue *rq, |
973 | struct ewma_pkt_len *avg_pkt_len) | |
3f2c31d9 | 974 | { |
ab7db917 | 975 | const size_t hdr_len = sizeof(struct virtio_net_hdr_mrg_rxbuf); |
fbf28d78 MD |
976 | unsigned int len; |
977 | ||
5377d758 | 978 | len = hdr_len + clamp_t(unsigned int, ewma_pkt_len_read(avg_pkt_len), |
f0c3192c | 979 | rq->min_buf_len, PAGE_SIZE - hdr_len); |
e377fcc8 | 980 | return ALIGN(len, L1_CACHE_BYTES); |
fbf28d78 MD |
981 | } |
982 | ||
2de2f7f4 JF |
983 | static int add_recvbuf_mergeable(struct virtnet_info *vi, |
984 | struct receive_queue *rq, gfp_t gfp) | |
fbf28d78 | 985 | { |
fb51879d | 986 | struct page_frag *alloc_frag = &rq->alloc_frag; |
2de2f7f4 | 987 | unsigned int headroom = virtnet_get_headroom(vi); |
fb51879d | 988 | char *buf; |
680557cf | 989 | void *ctx; |
3f2c31d9 | 990 | int err; |
fb51879d | 991 | unsigned int len, hole; |
3f2c31d9 | 992 | |
d85b758f | 993 | len = get_mergeable_buf_len(rq, &rq->mrg_avg_pkt_len); |
2de2f7f4 | 994 | if (unlikely(!skb_page_frag_refill(len + headroom, alloc_frag, gfp))) |
9ab86bbc | 995 | return -ENOMEM; |
ab7db917 | 996 | |
fb51879d | 997 | buf = (char *)page_address(alloc_frag->page) + alloc_frag->offset; |
2de2f7f4 | 998 | buf += headroom; /* advance address leaving hole at front of pkt */ |
fb51879d | 999 | get_page(alloc_frag->page); |
2de2f7f4 | 1000 | alloc_frag->offset += len + headroom; |
fb51879d | 1001 | hole = alloc_frag->size - alloc_frag->offset; |
2de2f7f4 | 1002 | if (hole < len + headroom) { |
ab7db917 MD |
1003 | /* To avoid internal fragmentation, if there is very likely not |
1004 | * enough space for another buffer, add the remaining space to | |
1daa8790 | 1005 | * the current buffer. |
ab7db917 | 1006 | */ |
fb51879d MD |
1007 | len += hole; |
1008 | alloc_frag->offset += hole; | |
1009 | } | |
3f2c31d9 | 1010 | |
fb51879d | 1011 | sg_init_one(rq->sg, buf, len); |
29fda25a | 1012 | ctx = mergeable_len_to_ctx(len, headroom); |
680557cf | 1013 | err = virtqueue_add_inbuf_ctx(rq->vq, rq->sg, 1, buf, ctx, gfp); |
9ab86bbc | 1014 | if (err < 0) |
2613af0e | 1015 | put_page(virt_to_head_page(buf)); |
3f2c31d9 | 1016 | |
9ab86bbc SM |
1017 | return err; |
1018 | } | |
3f2c31d9 | 1019 | |
b2baed69 RR |
1020 | /* |
1021 | * Returns false if we couldn't fill entirely (OOM). | |
1022 | * | |
1023 | * Normally run in the receive path, but can also be run from ndo_open | |
1024 | * before we're receiving packets, or from refill_work which is | |
1025 | * careful to disable receiving (using napi_disable). | |
1026 | */ | |
946fa564 MT |
1027 | static bool try_fill_recv(struct virtnet_info *vi, struct receive_queue *rq, |
1028 | gfp_t gfp) | |
9ab86bbc SM |
1029 | { |
1030 | int err; | |
1788f495 | 1031 | bool oom; |
3f2c31d9 | 1032 | |
fb51879d | 1033 | gfp |= __GFP_COLD; |
9ab86bbc SM |
1034 | do { |
1035 | if (vi->mergeable_rx_bufs) | |
2de2f7f4 | 1036 | err = add_recvbuf_mergeable(vi, rq, gfp); |
9ab86bbc | 1037 | else if (vi->big_packets) |
012873d0 | 1038 | err = add_recvbuf_big(vi, rq, gfp); |
9ab86bbc | 1039 | else |
946fa564 | 1040 | err = add_recvbuf_small(vi, rq, gfp); |
3f2c31d9 | 1041 | |
1788f495 | 1042 | oom = err == -ENOMEM; |
9ed4cb07 | 1043 | if (err) |
3f2c31d9 | 1044 | break; |
b7dfde95 | 1045 | } while (rq->vq->num_free); |
681daee2 | 1046 | virtqueue_kick(rq->vq); |
3161e453 | 1047 | return !oom; |
3f2c31d9 MM |
1048 | } |
1049 | ||
18445c4d | 1050 | static void skb_recv_done(struct virtqueue *rvq) |
296f96fc RR |
1051 | { |
1052 | struct virtnet_info *vi = rvq->vdev->priv; | |
986a4f4d | 1053 | struct receive_queue *rq = &vi->rq[vq2rxq(rvq)]; |
e9d7417b | 1054 | |
e4e8452a | 1055 | virtqueue_napi_schedule(&rq->napi, rvq); |
296f96fc RR |
1056 | } |
1057 | ||
e4e8452a | 1058 | static void virtnet_napi_enable(struct virtqueue *vq, struct napi_struct *napi) |
3e9d08ec | 1059 | { |
e4e8452a | 1060 | napi_enable(napi); |
3e9d08ec BR |
1061 | |
1062 | /* If all buffers were filled by other side before we napi_enabled, we | |
e4e8452a WB |
1063 | * won't get another interrupt, so process any outstanding packets now. |
1064 | * Call local_bh_enable after to trigger softIRQ processing. | |
1065 | */ | |
1066 | local_bh_disable(); | |
1067 | virtqueue_napi_schedule(napi, vq); | |
1068 | local_bh_enable(); | |
3e9d08ec BR |
1069 | } |
1070 | ||
b92f1e67 WB |
1071 | static void virtnet_napi_tx_enable(struct virtnet_info *vi, |
1072 | struct virtqueue *vq, | |
1073 | struct napi_struct *napi) | |
1074 | { | |
1075 | if (!napi->weight) | |
1076 | return; | |
1077 | ||
1078 | /* Tx napi touches cachelines on the cpu handling tx interrupts. Only | |
1079 | * enable the feature if this is likely affine with the transmit path. | |
1080 | */ | |
1081 | if (!vi->affinity_hint_set) { | |
1082 | napi->weight = 0; | |
1083 | return; | |
1084 | } | |
1085 | ||
1086 | return virtnet_napi_enable(vq, napi); | |
1087 | } | |
1088 | ||
78a57b48 WB |
1089 | static void virtnet_napi_tx_disable(struct napi_struct *napi) |
1090 | { | |
1091 | if (napi->weight) | |
1092 | napi_disable(napi); | |
1093 | } | |
1094 | ||
3161e453 RR |
1095 | static void refill_work(struct work_struct *work) |
1096 | { | |
e9d7417b JW |
1097 | struct virtnet_info *vi = |
1098 | container_of(work, struct virtnet_info, refill.work); | |
3161e453 | 1099 | bool still_empty; |
986a4f4d JW |
1100 | int i; |
1101 | ||
55257d72 | 1102 | for (i = 0; i < vi->curr_queue_pairs; i++) { |
986a4f4d | 1103 | struct receive_queue *rq = &vi->rq[i]; |
3161e453 | 1104 | |
986a4f4d | 1105 | napi_disable(&rq->napi); |
946fa564 | 1106 | still_empty = !try_fill_recv(vi, rq, GFP_KERNEL); |
e4e8452a | 1107 | virtnet_napi_enable(rq->vq, &rq->napi); |
3161e453 | 1108 | |
986a4f4d JW |
1109 | /* In theory, this can happen: if we don't get any buffers in |
1110 | * we will *never* try to fill again. | |
1111 | */ | |
1112 | if (still_empty) | |
1113 | schedule_delayed_work(&vi->refill, HZ/2); | |
1114 | } | |
3161e453 RR |
1115 | } |
1116 | ||
186b3c99 | 1117 | static int virtnet_receive(struct receive_queue *rq, int budget, bool *xdp_xmit) |
296f96fc | 1118 | { |
e9d7417b | 1119 | struct virtnet_info *vi = rq->vq->vdev->priv; |
61845d20 | 1120 | unsigned int len, received = 0, bytes = 0; |
9ab86bbc | 1121 | void *buf; |
61845d20 | 1122 | struct virtnet_stats *stats = this_cpu_ptr(vi->stats); |
296f96fc | 1123 | |
192f68cf | 1124 | if (!vi->big_packets || vi->mergeable_rx_bufs) { |
680557cf MT |
1125 | void *ctx; |
1126 | ||
1127 | while (received < budget && | |
1128 | (buf = virtqueue_get_buf_ctx(rq->vq, &len, &ctx))) { | |
186b3c99 | 1129 | bytes += receive_buf(vi, rq, buf, len, ctx, xdp_xmit); |
680557cf MT |
1130 | received++; |
1131 | } | |
1132 | } else { | |
1133 | while (received < budget && | |
1134 | (buf = virtqueue_get_buf(rq->vq, &len)) != NULL) { | |
186b3c99 | 1135 | bytes += receive_buf(vi, rq, buf, len, NULL, xdp_xmit); |
680557cf MT |
1136 | received++; |
1137 | } | |
296f96fc RR |
1138 | } |
1139 | ||
be121f46 | 1140 | if (rq->vq->num_free > virtqueue_get_vring_size(rq->vq) / 2) { |
946fa564 | 1141 | if (!try_fill_recv(vi, rq, GFP_ATOMIC)) |
3b07e9ca | 1142 | schedule_delayed_work(&vi->refill, 0); |
3161e453 | 1143 | } |
296f96fc | 1144 | |
61845d20 JW |
1145 | u64_stats_update_begin(&stats->rx_syncp); |
1146 | stats->rx_bytes += bytes; | |
1147 | stats->rx_packets += received; | |
1148 | u64_stats_update_end(&stats->rx_syncp); | |
1149 | ||
2ffa7598 JW |
1150 | return received; |
1151 | } | |
1152 | ||
ea7735d9 WB |
1153 | static void free_old_xmit_skbs(struct send_queue *sq) |
1154 | { | |
1155 | struct sk_buff *skb; | |
1156 | unsigned int len; | |
1157 | struct virtnet_info *vi = sq->vq->vdev->priv; | |
1158 | struct virtnet_stats *stats = this_cpu_ptr(vi->stats); | |
1159 | unsigned int packets = 0; | |
1160 | unsigned int bytes = 0; | |
1161 | ||
1162 | while ((skb = virtqueue_get_buf(sq->vq, &len)) != NULL) { | |
1163 | pr_debug("Sent skb %p\n", skb); | |
1164 | ||
1165 | bytes += skb->len; | |
1166 | packets++; | |
1167 | ||
dadc0736 | 1168 | dev_consume_skb_any(skb); |
ea7735d9 WB |
1169 | } |
1170 | ||
1171 | /* Avoid overhead when no packets have been processed | |
1172 | * happens when called speculatively from start_xmit. | |
1173 | */ | |
1174 | if (!packets) | |
1175 | return; | |
1176 | ||
1177 | u64_stats_update_begin(&stats->tx_syncp); | |
1178 | stats->tx_bytes += bytes; | |
1179 | stats->tx_packets += packets; | |
1180 | u64_stats_update_end(&stats->tx_syncp); | |
1181 | } | |
1182 | ||
7b0411ef WB |
1183 | static void virtnet_poll_cleantx(struct receive_queue *rq) |
1184 | { | |
1185 | struct virtnet_info *vi = rq->vq->vdev->priv; | |
1186 | unsigned int index = vq2rxq(rq->vq); | |
1187 | struct send_queue *sq = &vi->sq[index]; | |
1188 | struct netdev_queue *txq = netdev_get_tx_queue(vi->dev, index); | |
1189 | ||
1190 | if (!sq->napi.weight) | |
1191 | return; | |
1192 | ||
1193 | if (__netif_tx_trylock(txq)) { | |
1194 | free_old_xmit_skbs(sq); | |
1195 | __netif_tx_unlock(txq); | |
1196 | } | |
1197 | ||
1198 | if (sq->vq->num_free >= 2 + MAX_SKB_FRAGS) | |
1199 | netif_tx_wake_queue(txq); | |
1200 | } | |
1201 | ||
2ffa7598 JW |
1202 | static int virtnet_poll(struct napi_struct *napi, int budget) |
1203 | { | |
1204 | struct receive_queue *rq = | |
1205 | container_of(napi, struct receive_queue, napi); | |
e4e8452a | 1206 | unsigned int received; |
186b3c99 | 1207 | bool xdp_xmit = false; |
2ffa7598 | 1208 | |
7b0411ef WB |
1209 | virtnet_poll_cleantx(rq); |
1210 | ||
186b3c99 | 1211 | received = virtnet_receive(rq, budget, &xdp_xmit); |
2ffa7598 | 1212 | |
8329d98e | 1213 | /* Out of packets? */ |
e4e8452a WB |
1214 | if (received < budget) |
1215 | virtqueue_napi_complete(napi, rq->vq, received); | |
296f96fc | 1216 | |
186b3c99 JW |
1217 | if (xdp_xmit) |
1218 | xdp_do_flush_map(); | |
1219 | ||
296f96fc RR |
1220 | return received; |
1221 | } | |
1222 | ||
986a4f4d JW |
1223 | static int virtnet_open(struct net_device *dev) |
1224 | { | |
1225 | struct virtnet_info *vi = netdev_priv(dev); | |
1226 | int i; | |
1227 | ||
e4166625 JW |
1228 | for (i = 0; i < vi->max_queue_pairs; i++) { |
1229 | if (i < vi->curr_queue_pairs) | |
1230 | /* Make sure we have some buffers: if oom use wq. */ | |
946fa564 | 1231 | if (!try_fill_recv(vi, &vi->rq[i], GFP_KERNEL)) |
e4166625 | 1232 | schedule_delayed_work(&vi->refill, 0); |
e4e8452a | 1233 | virtnet_napi_enable(vi->rq[i].vq, &vi->rq[i].napi); |
b92f1e67 | 1234 | virtnet_napi_tx_enable(vi, vi->sq[i].vq, &vi->sq[i].napi); |
986a4f4d JW |
1235 | } |
1236 | ||
1237 | return 0; | |
1238 | } | |
1239 | ||
b92f1e67 WB |
1240 | static int virtnet_poll_tx(struct napi_struct *napi, int budget) |
1241 | { | |
1242 | struct send_queue *sq = container_of(napi, struct send_queue, napi); | |
1243 | struct virtnet_info *vi = sq->vq->vdev->priv; | |
1244 | struct netdev_queue *txq = netdev_get_tx_queue(vi->dev, vq2txq(sq->vq)); | |
1245 | ||
1246 | __netif_tx_lock(txq, raw_smp_processor_id()); | |
1247 | free_old_xmit_skbs(sq); | |
1248 | __netif_tx_unlock(txq); | |
1249 | ||
1250 | virtqueue_napi_complete(napi, sq->vq, 0); | |
1251 | ||
1252 | if (sq->vq->num_free >= 2 + MAX_SKB_FRAGS) | |
1253 | netif_tx_wake_queue(txq); | |
1254 | ||
1255 | return 0; | |
1256 | } | |
1257 | ||
e9d7417b | 1258 | static int xmit_skb(struct send_queue *sq, struct sk_buff *skb) |
296f96fc | 1259 | { |
012873d0 | 1260 | struct virtio_net_hdr_mrg_rxbuf *hdr; |
296f96fc | 1261 | const unsigned char *dest = ((struct ethhdr *)skb->data)->h_dest; |
e9d7417b | 1262 | struct virtnet_info *vi = sq->vq->vdev->priv; |
e2fcad58 | 1263 | int num_sg; |
012873d0 | 1264 | unsigned hdr_len = vi->hdr_len; |
e7428e95 | 1265 | bool can_push; |
296f96fc | 1266 | |
e174961c | 1267 | pr_debug("%s: xmit %p %pM\n", vi->dev->name, skb, dest); |
e7428e95 MT |
1268 | |
1269 | can_push = vi->any_header_sg && | |
1270 | !((unsigned long)skb->data & (__alignof__(*hdr) - 1)) && | |
1271 | !skb_header_cloned(skb) && skb_headroom(skb) >= hdr_len; | |
1272 | /* Even if we can, don't push here yet as this would skew | |
1273 | * csum_start offset below. */ | |
1274 | if (can_push) | |
012873d0 | 1275 | hdr = (struct virtio_net_hdr_mrg_rxbuf *)(skb->data - hdr_len); |
e7428e95 MT |
1276 | else |
1277 | hdr = skb_vnet_hdr(skb); | |
296f96fc | 1278 | |
e858fae2 | 1279 | if (virtio_net_hdr_from_skb(skb, &hdr->hdr, |
6391a448 | 1280 | virtio_is_little_endian(vi->vdev), false)) |
e858fae2 | 1281 | BUG(); |
296f96fc | 1282 | |
3f2c31d9 | 1283 | if (vi->mergeable_rx_bufs) |
012873d0 | 1284 | hdr->num_buffers = 0; |
3f2c31d9 | 1285 | |
547c890c | 1286 | sg_init_table(sq->sg, skb_shinfo(skb)->nr_frags + (can_push ? 1 : 2)); |
e7428e95 MT |
1287 | if (can_push) { |
1288 | __skb_push(skb, hdr_len); | |
1289 | num_sg = skb_to_sgvec(skb, sq->sg, 0, skb->len); | |
e2fcad58 JD |
1290 | if (unlikely(num_sg < 0)) |
1291 | return num_sg; | |
e7428e95 MT |
1292 | /* Pull header back to avoid skew in tx bytes calculations. */ |
1293 | __skb_pull(skb, hdr_len); | |
1294 | } else { | |
1295 | sg_set_buf(sq->sg, hdr, hdr_len); | |
e2fcad58 JD |
1296 | num_sg = skb_to_sgvec(skb, sq->sg + 1, 0, skb->len); |
1297 | if (unlikely(num_sg < 0)) | |
1298 | return num_sg; | |
1299 | num_sg++; | |
e7428e95 | 1300 | } |
9dc7b9e4 | 1301 | return virtqueue_add_outbuf(sq->vq, sq->sg, num_sg, skb, GFP_ATOMIC); |
11a3a154 RR |
1302 | } |
1303 | ||
424efe9c | 1304 | static netdev_tx_t start_xmit(struct sk_buff *skb, struct net_device *dev) |
99ffc696 RR |
1305 | { |
1306 | struct virtnet_info *vi = netdev_priv(dev); | |
986a4f4d JW |
1307 | int qnum = skb_get_queue_mapping(skb); |
1308 | struct send_queue *sq = &vi->sq[qnum]; | |
9ed4cb07 | 1309 | int err; |
4b7fd2e6 MT |
1310 | struct netdev_queue *txq = netdev_get_tx_queue(dev, qnum); |
1311 | bool kick = !skb->xmit_more; | |
b92f1e67 | 1312 | bool use_napi = sq->napi.weight; |
2cb9c6ba | 1313 | |
2cb9c6ba | 1314 | /* Free up any pending old buffers before queueing new ones. */ |
e9d7417b | 1315 | free_old_xmit_skbs(sq); |
99ffc696 | 1316 | |
bdb12e0d WB |
1317 | if (use_napi && kick) |
1318 | virtqueue_enable_cb_delayed(sq->vq); | |
1319 | ||
074c3582 JK |
1320 | /* timestamp packet in software */ |
1321 | skb_tx_timestamp(skb); | |
1322 | ||
03f191ba | 1323 | /* Try to transmit */ |
b7dfde95 | 1324 | err = xmit_skb(sq, skb); |
48925e37 | 1325 | |
9ed4cb07 | 1326 | /* This should not happen! */ |
681daee2 | 1327 | if (unlikely(err)) { |
9ed4cb07 RR |
1328 | dev->stats.tx_fifo_errors++; |
1329 | if (net_ratelimit()) | |
1330 | dev_warn(&dev->dev, | |
b7dfde95 | 1331 | "Unexpected TXQ (%d) queue failure: %d\n", qnum, err); |
58eba97d | 1332 | dev->stats.tx_dropped++; |
85e94525 | 1333 | dev_kfree_skb_any(skb); |
58eba97d | 1334 | return NETDEV_TX_OK; |
296f96fc | 1335 | } |
03f191ba | 1336 | |
48925e37 | 1337 | /* Don't wait up for transmitted skbs to be freed. */ |
b92f1e67 WB |
1338 | if (!use_napi) { |
1339 | skb_orphan(skb); | |
1340 | nf_reset(skb); | |
1341 | } | |
48925e37 | 1342 | |
60302ff6 MT |
1343 | /* If running out of space, stop queue to avoid getting packets that we |
1344 | * are then unable to transmit. | |
1345 | * An alternative would be to force queuing layer to requeue the skb by | |
1346 | * returning NETDEV_TX_BUSY. However, NETDEV_TX_BUSY should not be | |
1347 | * returned in a normal path of operation: it means that driver is not | |
1348 | * maintaining the TX queue stop/start state properly, and causes | |
1349 | * the stack to do a non-trivial amount of useless work. | |
1350 | * Since most packets only take 1 or 2 ring slots, stopping the queue | |
1351 | * early means 16 slots are typically wasted. | |
d631b94e | 1352 | */ |
b7dfde95 | 1353 | if (sq->vq->num_free < 2+MAX_SKB_FRAGS) { |
986a4f4d | 1354 | netif_stop_subqueue(dev, qnum); |
b92f1e67 WB |
1355 | if (!use_napi && |
1356 | unlikely(!virtqueue_enable_cb_delayed(sq->vq))) { | |
48925e37 | 1357 | /* More just got used, free them then recheck. */ |
b7dfde95 LT |
1358 | free_old_xmit_skbs(sq); |
1359 | if (sq->vq->num_free >= 2+MAX_SKB_FRAGS) { | |
986a4f4d | 1360 | netif_start_subqueue(dev, qnum); |
e9d7417b | 1361 | virtqueue_disable_cb(sq->vq); |
48925e37 RR |
1362 | } |
1363 | } | |
99ffc696 | 1364 | } |
48925e37 | 1365 | |
4b7fd2e6 | 1366 | if (kick || netif_xmit_stopped(txq)) |
0b725a2c | 1367 | virtqueue_kick(sq->vq); |
296f96fc | 1368 | |
0b725a2c | 1369 | return NETDEV_TX_OK; |
c223a078 DM |
1370 | } |
1371 | ||
40cbfc37 AK |
1372 | /* |
1373 | * Send command via the control virtqueue and check status. Commands | |
1374 | * supported by the hypervisor, as indicated by feature bits, should | |
788a8b6d | 1375 | * never fail unless improperly formatted. |
40cbfc37 AK |
1376 | */ |
1377 | static bool virtnet_send_command(struct virtnet_info *vi, u8 class, u8 cmd, | |
d24bae32 | 1378 | struct scatterlist *out) |
40cbfc37 | 1379 | { |
f7bc9594 | 1380 | struct scatterlist *sgs[4], hdr, stat; |
d24bae32 | 1381 | unsigned out_num = 0, tmp; |
40cbfc37 AK |
1382 | |
1383 | /* Caller should know better */ | |
f7bc9594 | 1384 | BUG_ON(!virtio_has_feature(vi->vdev, VIRTIO_NET_F_CTRL_VQ)); |
40cbfc37 | 1385 | |
2ac46030 MT |
1386 | vi->ctrl_status = ~0; |
1387 | vi->ctrl_hdr.class = class; | |
1388 | vi->ctrl_hdr.cmd = cmd; | |
f7bc9594 | 1389 | /* Add header */ |
2ac46030 | 1390 | sg_init_one(&hdr, &vi->ctrl_hdr, sizeof(vi->ctrl_hdr)); |
f7bc9594 | 1391 | sgs[out_num++] = &hdr; |
40cbfc37 | 1392 | |
f7bc9594 RR |
1393 | if (out) |
1394 | sgs[out_num++] = out; | |
40cbfc37 | 1395 | |
f7bc9594 | 1396 | /* Add return status. */ |
2ac46030 | 1397 | sg_init_one(&stat, &vi->ctrl_status, sizeof(vi->ctrl_status)); |
d24bae32 | 1398 | sgs[out_num] = &stat; |
40cbfc37 | 1399 | |
d24bae32 | 1400 | BUG_ON(out_num + 1 > ARRAY_SIZE(sgs)); |
a7c58146 | 1401 | virtqueue_add_sgs(vi->cvq, sgs, out_num, 1, vi, GFP_ATOMIC); |
40cbfc37 | 1402 | |
67975901 | 1403 | if (unlikely(!virtqueue_kick(vi->cvq))) |
2ac46030 | 1404 | return vi->ctrl_status == VIRTIO_NET_OK; |
40cbfc37 AK |
1405 | |
1406 | /* Spin for a response, the kick causes an ioport write, trapping | |
1407 | * into the hypervisor, so the request should be handled immediately. | |
1408 | */ | |
047b9b94 HG |
1409 | while (!virtqueue_get_buf(vi->cvq, &tmp) && |
1410 | !virtqueue_is_broken(vi->cvq)) | |
40cbfc37 AK |
1411 | cpu_relax(); |
1412 | ||
2ac46030 | 1413 | return vi->ctrl_status == VIRTIO_NET_OK; |
40cbfc37 AK |
1414 | } |
1415 | ||
9c46f6d4 AW |
1416 | static int virtnet_set_mac_address(struct net_device *dev, void *p) |
1417 | { | |
1418 | struct virtnet_info *vi = netdev_priv(dev); | |
1419 | struct virtio_device *vdev = vi->vdev; | |
f2f2c8b4 | 1420 | int ret; |
e37e2ff3 | 1421 | struct sockaddr *addr; |
7e58d5ae | 1422 | struct scatterlist sg; |
9c46f6d4 | 1423 | |
801822d1 | 1424 | addr = kmemdup(p, sizeof(*addr), GFP_KERNEL); |
e37e2ff3 AL |
1425 | if (!addr) |
1426 | return -ENOMEM; | |
e37e2ff3 AL |
1427 | |
1428 | ret = eth_prepare_mac_addr_change(dev, addr); | |
f2f2c8b4 | 1429 | if (ret) |
e37e2ff3 | 1430 | goto out; |
9c46f6d4 | 1431 | |
7e58d5ae AK |
1432 | if (virtio_has_feature(vdev, VIRTIO_NET_F_CTRL_MAC_ADDR)) { |
1433 | sg_init_one(&sg, addr->sa_data, dev->addr_len); | |
1434 | if (!virtnet_send_command(vi, VIRTIO_NET_CTRL_MAC, | |
d24bae32 | 1435 | VIRTIO_NET_CTRL_MAC_ADDR_SET, &sg)) { |
7e58d5ae AK |
1436 | dev_warn(&vdev->dev, |
1437 | "Failed to set mac address by vq command.\n"); | |
e37e2ff3 AL |
1438 | ret = -EINVAL; |
1439 | goto out; | |
7e58d5ae | 1440 | } |
7e93a02f MT |
1441 | } else if (virtio_has_feature(vdev, VIRTIO_NET_F_MAC) && |
1442 | !virtio_has_feature(vdev, VIRTIO_F_VERSION_1)) { | |
855e0c52 RR |
1443 | unsigned int i; |
1444 | ||
1445 | /* Naturally, this has an atomicity problem. */ | |
1446 | for (i = 0; i < dev->addr_len; i++) | |
1447 | virtio_cwrite8(vdev, | |
1448 | offsetof(struct virtio_net_config, mac) + | |
1449 | i, addr->sa_data[i]); | |
7e58d5ae AK |
1450 | } |
1451 | ||
1452 | eth_commit_mac_addr_change(dev, p); | |
e37e2ff3 | 1453 | ret = 0; |
9c46f6d4 | 1454 | |
e37e2ff3 AL |
1455 | out: |
1456 | kfree(addr); | |
1457 | return ret; | |
9c46f6d4 AW |
1458 | } |
1459 | ||
bc1f4470 | 1460 | static void virtnet_stats(struct net_device *dev, |
1461 | struct rtnl_link_stats64 *tot) | |
3fa2a1df | 1462 | { |
1463 | struct virtnet_info *vi = netdev_priv(dev); | |
1464 | int cpu; | |
1465 | unsigned int start; | |
1466 | ||
1467 | for_each_possible_cpu(cpu) { | |
58472a76 | 1468 | struct virtnet_stats *stats = per_cpu_ptr(vi->stats, cpu); |
3fa2a1df | 1469 | u64 tpackets, tbytes, rpackets, rbytes; |
1470 | ||
1471 | do { | |
57a7744e | 1472 | start = u64_stats_fetch_begin_irq(&stats->tx_syncp); |
3fa2a1df | 1473 | tpackets = stats->tx_packets; |
1474 | tbytes = stats->tx_bytes; | |
57a7744e | 1475 | } while (u64_stats_fetch_retry_irq(&stats->tx_syncp, start)); |
83a27052 ED |
1476 | |
1477 | do { | |
57a7744e | 1478 | start = u64_stats_fetch_begin_irq(&stats->rx_syncp); |
3fa2a1df | 1479 | rpackets = stats->rx_packets; |
1480 | rbytes = stats->rx_bytes; | |
57a7744e | 1481 | } while (u64_stats_fetch_retry_irq(&stats->rx_syncp, start)); |
3fa2a1df | 1482 | |
1483 | tot->rx_packets += rpackets; | |
1484 | tot->tx_packets += tpackets; | |
1485 | tot->rx_bytes += rbytes; | |
1486 | tot->tx_bytes += tbytes; | |
1487 | } | |
1488 | ||
1489 | tot->tx_dropped = dev->stats.tx_dropped; | |
021ac8d3 | 1490 | tot->tx_fifo_errors = dev->stats.tx_fifo_errors; |
3fa2a1df | 1491 | tot->rx_dropped = dev->stats.rx_dropped; |
1492 | tot->rx_length_errors = dev->stats.rx_length_errors; | |
1493 | tot->rx_frame_errors = dev->stats.rx_frame_errors; | |
3fa2a1df | 1494 | } |
1495 | ||
da74e89d AS |
1496 | #ifdef CONFIG_NET_POLL_CONTROLLER |
1497 | static void virtnet_netpoll(struct net_device *dev) | |
1498 | { | |
1499 | struct virtnet_info *vi = netdev_priv(dev); | |
986a4f4d | 1500 | int i; |
da74e89d | 1501 | |
986a4f4d JW |
1502 | for (i = 0; i < vi->curr_queue_pairs; i++) |
1503 | napi_schedule(&vi->rq[i].napi); | |
da74e89d AS |
1504 | } |
1505 | #endif | |
1506 | ||
586d17c5 JW |
1507 | static void virtnet_ack_link_announce(struct virtnet_info *vi) |
1508 | { | |
1509 | rtnl_lock(); | |
1510 | if (!virtnet_send_command(vi, VIRTIO_NET_CTRL_ANNOUNCE, | |
d24bae32 | 1511 | VIRTIO_NET_CTRL_ANNOUNCE_ACK, NULL)) |
586d17c5 JW |
1512 | dev_warn(&vi->dev->dev, "Failed to ack link announce.\n"); |
1513 | rtnl_unlock(); | |
1514 | } | |
1515 | ||
47315329 | 1516 | static int _virtnet_set_queues(struct virtnet_info *vi, u16 queue_pairs) |
986a4f4d JW |
1517 | { |
1518 | struct scatterlist sg; | |
986a4f4d JW |
1519 | struct net_device *dev = vi->dev; |
1520 | ||
1521 | if (!vi->has_cvq || !virtio_has_feature(vi->vdev, VIRTIO_NET_F_MQ)) | |
1522 | return 0; | |
1523 | ||
a725ee3e AL |
1524 | vi->ctrl_mq.virtqueue_pairs = cpu_to_virtio16(vi->vdev, queue_pairs); |
1525 | sg_init_one(&sg, &vi->ctrl_mq, sizeof(vi->ctrl_mq)); | |
986a4f4d JW |
1526 | |
1527 | if (!virtnet_send_command(vi, VIRTIO_NET_CTRL_MQ, | |
d24bae32 | 1528 | VIRTIO_NET_CTRL_MQ_VQ_PAIRS_SET, &sg)) { |
986a4f4d JW |
1529 | dev_warn(&dev->dev, "Fail to set num of queue pairs to %d\n", |
1530 | queue_pairs); | |
1531 | return -EINVAL; | |
55257d72 | 1532 | } else { |
986a4f4d | 1533 | vi->curr_queue_pairs = queue_pairs; |
35ed159b JW |
1534 | /* virtnet_open() will refill when device is going to up. */ |
1535 | if (dev->flags & IFF_UP) | |
1536 | schedule_delayed_work(&vi->refill, 0); | |
55257d72 | 1537 | } |
986a4f4d JW |
1538 | |
1539 | return 0; | |
1540 | } | |
1541 | ||
47315329 JF |
1542 | static int virtnet_set_queues(struct virtnet_info *vi, u16 queue_pairs) |
1543 | { | |
1544 | int err; | |
1545 | ||
1546 | rtnl_lock(); | |
1547 | err = _virtnet_set_queues(vi, queue_pairs); | |
1548 | rtnl_unlock(); | |
1549 | return err; | |
1550 | } | |
1551 | ||
296f96fc RR |
1552 | static int virtnet_close(struct net_device *dev) |
1553 | { | |
1554 | struct virtnet_info *vi = netdev_priv(dev); | |
986a4f4d | 1555 | int i; |
296f96fc | 1556 | |
b2baed69 RR |
1557 | /* Make sure refill_work doesn't re-enable napi! */ |
1558 | cancel_delayed_work_sync(&vi->refill); | |
986a4f4d | 1559 | |
b92f1e67 | 1560 | for (i = 0; i < vi->max_queue_pairs; i++) { |
986a4f4d | 1561 | napi_disable(&vi->rq[i].napi); |
78a57b48 | 1562 | virtnet_napi_tx_disable(&vi->sq[i].napi); |
b92f1e67 | 1563 | } |
296f96fc | 1564 | |
296f96fc RR |
1565 | return 0; |
1566 | } | |
1567 | ||
2af7698e AW |
1568 | static void virtnet_set_rx_mode(struct net_device *dev) |
1569 | { | |
1570 | struct virtnet_info *vi = netdev_priv(dev); | |
f565a7c2 | 1571 | struct scatterlist sg[2]; |
f565a7c2 | 1572 | struct virtio_net_ctrl_mac *mac_data; |
ccffad25 | 1573 | struct netdev_hw_addr *ha; |
32e7bfc4 | 1574 | int uc_count; |
4cd24eaf | 1575 | int mc_count; |
f565a7c2 AW |
1576 | void *buf; |
1577 | int i; | |
2af7698e | 1578 | |
788a8b6d | 1579 | /* We can't dynamically set ndo_set_rx_mode, so return gracefully */ |
2af7698e AW |
1580 | if (!virtio_has_feature(vi->vdev, VIRTIO_NET_F_CTRL_RX)) |
1581 | return; | |
1582 | ||
2ac46030 MT |
1583 | vi->ctrl_promisc = ((dev->flags & IFF_PROMISC) != 0); |
1584 | vi->ctrl_allmulti = ((dev->flags & IFF_ALLMULTI) != 0); | |
2af7698e | 1585 | |
2ac46030 | 1586 | sg_init_one(sg, &vi->ctrl_promisc, sizeof(vi->ctrl_promisc)); |
2af7698e AW |
1587 | |
1588 | if (!virtnet_send_command(vi, VIRTIO_NET_CTRL_RX, | |
d24bae32 | 1589 | VIRTIO_NET_CTRL_RX_PROMISC, sg)) |
2af7698e | 1590 | dev_warn(&dev->dev, "Failed to %sable promisc mode.\n", |
2ac46030 | 1591 | vi->ctrl_promisc ? "en" : "dis"); |
2af7698e | 1592 | |
2ac46030 | 1593 | sg_init_one(sg, &vi->ctrl_allmulti, sizeof(vi->ctrl_allmulti)); |
2af7698e AW |
1594 | |
1595 | if (!virtnet_send_command(vi, VIRTIO_NET_CTRL_RX, | |
d24bae32 | 1596 | VIRTIO_NET_CTRL_RX_ALLMULTI, sg)) |
2af7698e | 1597 | dev_warn(&dev->dev, "Failed to %sable allmulti mode.\n", |
2ac46030 | 1598 | vi->ctrl_allmulti ? "en" : "dis"); |
f565a7c2 | 1599 | |
32e7bfc4 | 1600 | uc_count = netdev_uc_count(dev); |
4cd24eaf | 1601 | mc_count = netdev_mc_count(dev); |
f565a7c2 | 1602 | /* MAC filter - use one buffer for both lists */ |
4cd24eaf JP |
1603 | buf = kzalloc(((uc_count + mc_count) * ETH_ALEN) + |
1604 | (2 * sizeof(mac_data->entries)), GFP_ATOMIC); | |
1605 | mac_data = buf; | |
e68ed8f0 | 1606 | if (!buf) |
f565a7c2 | 1607 | return; |
f565a7c2 | 1608 | |
23e258e1 AW |
1609 | sg_init_table(sg, 2); |
1610 | ||
f565a7c2 | 1611 | /* Store the unicast list and count in the front of the buffer */ |
fdd819b2 | 1612 | mac_data->entries = cpu_to_virtio32(vi->vdev, uc_count); |
ccffad25 | 1613 | i = 0; |
32e7bfc4 | 1614 | netdev_for_each_uc_addr(ha, dev) |
ccffad25 | 1615 | memcpy(&mac_data->macs[i++][0], ha->addr, ETH_ALEN); |
f565a7c2 AW |
1616 | |
1617 | sg_set_buf(&sg[0], mac_data, | |
32e7bfc4 | 1618 | sizeof(mac_data->entries) + (uc_count * ETH_ALEN)); |
f565a7c2 AW |
1619 | |
1620 | /* multicast list and count fill the end */ | |
32e7bfc4 | 1621 | mac_data = (void *)&mac_data->macs[uc_count][0]; |
f565a7c2 | 1622 | |
fdd819b2 | 1623 | mac_data->entries = cpu_to_virtio32(vi->vdev, mc_count); |
567ec874 | 1624 | i = 0; |
22bedad3 JP |
1625 | netdev_for_each_mc_addr(ha, dev) |
1626 | memcpy(&mac_data->macs[i++][0], ha->addr, ETH_ALEN); | |
f565a7c2 AW |
1627 | |
1628 | sg_set_buf(&sg[1], mac_data, | |
4cd24eaf | 1629 | sizeof(mac_data->entries) + (mc_count * ETH_ALEN)); |
f565a7c2 AW |
1630 | |
1631 | if (!virtnet_send_command(vi, VIRTIO_NET_CTRL_MAC, | |
d24bae32 | 1632 | VIRTIO_NET_CTRL_MAC_TABLE_SET, sg)) |
99e872ae | 1633 | dev_warn(&dev->dev, "Failed to set MAC filter table.\n"); |
f565a7c2 AW |
1634 | |
1635 | kfree(buf); | |
2af7698e AW |
1636 | } |
1637 | ||
80d5c368 PM |
1638 | static int virtnet_vlan_rx_add_vid(struct net_device *dev, |
1639 | __be16 proto, u16 vid) | |
0bde9569 AW |
1640 | { |
1641 | struct virtnet_info *vi = netdev_priv(dev); | |
1642 | struct scatterlist sg; | |
1643 | ||
a725ee3e AL |
1644 | vi->ctrl_vid = vid; |
1645 | sg_init_one(&sg, &vi->ctrl_vid, sizeof(vi->ctrl_vid)); | |
0bde9569 AW |
1646 | |
1647 | if (!virtnet_send_command(vi, VIRTIO_NET_CTRL_VLAN, | |
d24bae32 | 1648 | VIRTIO_NET_CTRL_VLAN_ADD, &sg)) |
0bde9569 | 1649 | dev_warn(&dev->dev, "Failed to add VLAN ID %d.\n", vid); |
8e586137 | 1650 | return 0; |
0bde9569 AW |
1651 | } |
1652 | ||
80d5c368 PM |
1653 | static int virtnet_vlan_rx_kill_vid(struct net_device *dev, |
1654 | __be16 proto, u16 vid) | |
0bde9569 AW |
1655 | { |
1656 | struct virtnet_info *vi = netdev_priv(dev); | |
1657 | struct scatterlist sg; | |
1658 | ||
a725ee3e AL |
1659 | vi->ctrl_vid = vid; |
1660 | sg_init_one(&sg, &vi->ctrl_vid, sizeof(vi->ctrl_vid)); | |
0bde9569 AW |
1661 | |
1662 | if (!virtnet_send_command(vi, VIRTIO_NET_CTRL_VLAN, | |
d24bae32 | 1663 | VIRTIO_NET_CTRL_VLAN_DEL, &sg)) |
0bde9569 | 1664 | dev_warn(&dev->dev, "Failed to kill VLAN ID %d.\n", vid); |
8e586137 | 1665 | return 0; |
0bde9569 AW |
1666 | } |
1667 | ||
8898c21c | 1668 | static void virtnet_clean_affinity(struct virtnet_info *vi, long hcpu) |
986a4f4d JW |
1669 | { |
1670 | int i; | |
1671 | ||
8898c21c WG |
1672 | if (vi->affinity_hint_set) { |
1673 | for (i = 0; i < vi->max_queue_pairs; i++) { | |
47be2479 WG |
1674 | virtqueue_set_affinity(vi->rq[i].vq, -1); |
1675 | virtqueue_set_affinity(vi->sq[i].vq, -1); | |
1676 | } | |
1677 | ||
8898c21c WG |
1678 | vi->affinity_hint_set = false; |
1679 | } | |
8898c21c | 1680 | } |
47be2479 | 1681 | |
8898c21c WG |
1682 | static void virtnet_set_affinity(struct virtnet_info *vi) |
1683 | { | |
1684 | int i; | |
1685 | int cpu; | |
986a4f4d JW |
1686 | |
1687 | /* In multiqueue mode, when the number of cpu is equal to the number of | |
1688 | * queue pairs, we let the queue pairs to be private to one cpu by | |
1689 | * setting the affinity hint to eliminate the contention. | |
1690 | */ | |
8898c21c WG |
1691 | if (vi->curr_queue_pairs == 1 || |
1692 | vi->max_queue_pairs != num_online_cpus()) { | |
1693 | virtnet_clean_affinity(vi, -1); | |
1694 | return; | |
986a4f4d JW |
1695 | } |
1696 | ||
8898c21c WG |
1697 | i = 0; |
1698 | for_each_online_cpu(cpu) { | |
986a4f4d JW |
1699 | virtqueue_set_affinity(vi->rq[i].vq, cpu); |
1700 | virtqueue_set_affinity(vi->sq[i].vq, cpu); | |
9bb8ca86 | 1701 | netif_set_xps_queue(vi->dev, cpumask_of(cpu), i); |
8898c21c | 1702 | i++; |
986a4f4d JW |
1703 | } |
1704 | ||
8898c21c | 1705 | vi->affinity_hint_set = true; |
986a4f4d JW |
1706 | } |
1707 | ||
8017c279 | 1708 | static int virtnet_cpu_online(unsigned int cpu, struct hlist_node *node) |
8de4b2f3 | 1709 | { |
8017c279 SAS |
1710 | struct virtnet_info *vi = hlist_entry_safe(node, struct virtnet_info, |
1711 | node); | |
1712 | virtnet_set_affinity(vi); | |
1713 | return 0; | |
1714 | } | |
8de4b2f3 | 1715 | |
8017c279 SAS |
1716 | static int virtnet_cpu_dead(unsigned int cpu, struct hlist_node *node) |
1717 | { | |
1718 | struct virtnet_info *vi = hlist_entry_safe(node, struct virtnet_info, | |
1719 | node_dead); | |
1720 | virtnet_set_affinity(vi); | |
1721 | return 0; | |
1722 | } | |
3ab098df | 1723 | |
8017c279 SAS |
1724 | static int virtnet_cpu_down_prep(unsigned int cpu, struct hlist_node *node) |
1725 | { | |
1726 | struct virtnet_info *vi = hlist_entry_safe(node, struct virtnet_info, | |
1727 | node); | |
1728 | ||
1729 | virtnet_clean_affinity(vi, cpu); | |
1730 | return 0; | |
1731 | } | |
1732 | ||
1733 | static enum cpuhp_state virtionet_online; | |
1734 | ||
1735 | static int virtnet_cpu_notif_add(struct virtnet_info *vi) | |
1736 | { | |
1737 | int ret; | |
1738 | ||
1739 | ret = cpuhp_state_add_instance_nocalls(virtionet_online, &vi->node); | |
1740 | if (ret) | |
1741 | return ret; | |
1742 | ret = cpuhp_state_add_instance_nocalls(CPUHP_VIRT_NET_DEAD, | |
1743 | &vi->node_dead); | |
1744 | if (!ret) | |
1745 | return ret; | |
1746 | cpuhp_state_remove_instance_nocalls(virtionet_online, &vi->node); | |
1747 | return ret; | |
1748 | } | |
1749 | ||
1750 | static void virtnet_cpu_notif_remove(struct virtnet_info *vi) | |
1751 | { | |
1752 | cpuhp_state_remove_instance_nocalls(virtionet_online, &vi->node); | |
1753 | cpuhp_state_remove_instance_nocalls(CPUHP_VIRT_NET_DEAD, | |
1754 | &vi->node_dead); | |
986a4f4d JW |
1755 | } |
1756 | ||
8f9f4668 RJ |
1757 | static void virtnet_get_ringparam(struct net_device *dev, |
1758 | struct ethtool_ringparam *ring) | |
1759 | { | |
1760 | struct virtnet_info *vi = netdev_priv(dev); | |
1761 | ||
986a4f4d JW |
1762 | ring->rx_max_pending = virtqueue_get_vring_size(vi->rq[0].vq); |
1763 | ring->tx_max_pending = virtqueue_get_vring_size(vi->sq[0].vq); | |
8f9f4668 RJ |
1764 | ring->rx_pending = ring->rx_max_pending; |
1765 | ring->tx_pending = ring->tx_max_pending; | |
8f9f4668 RJ |
1766 | } |
1767 | ||
66846048 RJ |
1768 | |
1769 | static void virtnet_get_drvinfo(struct net_device *dev, | |
1770 | struct ethtool_drvinfo *info) | |
1771 | { | |
1772 | struct virtnet_info *vi = netdev_priv(dev); | |
1773 | struct virtio_device *vdev = vi->vdev; | |
1774 | ||
1775 | strlcpy(info->driver, KBUILD_MODNAME, sizeof(info->driver)); | |
1776 | strlcpy(info->version, VIRTNET_DRIVER_VERSION, sizeof(info->version)); | |
1777 | strlcpy(info->bus_info, virtio_bus_name(vdev), sizeof(info->bus_info)); | |
1778 | ||
1779 | } | |
1780 | ||
d73bcd2c JW |
1781 | /* TODO: Eliminate OOO packets during switching */ |
1782 | static int virtnet_set_channels(struct net_device *dev, | |
1783 | struct ethtool_channels *channels) | |
1784 | { | |
1785 | struct virtnet_info *vi = netdev_priv(dev); | |
1786 | u16 queue_pairs = channels->combined_count; | |
1787 | int err; | |
1788 | ||
1789 | /* We don't support separate rx/tx channels. | |
1790 | * We don't allow setting 'other' channels. | |
1791 | */ | |
1792 | if (channels->rx_count || channels->tx_count || channels->other_count) | |
1793 | return -EINVAL; | |
1794 | ||
c18e9cd6 | 1795 | if (queue_pairs > vi->max_queue_pairs || queue_pairs == 0) |
d73bcd2c JW |
1796 | return -EINVAL; |
1797 | ||
f600b690 JF |
1798 | /* For now we don't support modifying channels while XDP is loaded |
1799 | * also when XDP is loaded all RX queues have XDP programs so we only | |
1800 | * need to check a single RX queue. | |
1801 | */ | |
1802 | if (vi->rq[0].xdp_prog) | |
1803 | return -EINVAL; | |
1804 | ||
47be2479 | 1805 | get_online_cpus(); |
47315329 | 1806 | err = _virtnet_set_queues(vi, queue_pairs); |
d73bcd2c JW |
1807 | if (!err) { |
1808 | netif_set_real_num_tx_queues(dev, queue_pairs); | |
1809 | netif_set_real_num_rx_queues(dev, queue_pairs); | |
1810 | ||
8898c21c | 1811 | virtnet_set_affinity(vi); |
d73bcd2c | 1812 | } |
47be2479 | 1813 | put_online_cpus(); |
d73bcd2c JW |
1814 | |
1815 | return err; | |
1816 | } | |
1817 | ||
1818 | static void virtnet_get_channels(struct net_device *dev, | |
1819 | struct ethtool_channels *channels) | |
1820 | { | |
1821 | struct virtnet_info *vi = netdev_priv(dev); | |
1822 | ||
1823 | channels->combined_count = vi->curr_queue_pairs; | |
1824 | channels->max_combined = vi->max_queue_pairs; | |
1825 | channels->max_other = 0; | |
1826 | channels->rx_count = 0; | |
1827 | channels->tx_count = 0; | |
1828 | channels->other_count = 0; | |
1829 | } | |
1830 | ||
16032be5 | 1831 | /* Check if the user is trying to change anything besides speed/duplex */ |
ebb6b4b1 PR |
1832 | static bool |
1833 | virtnet_validate_ethtool_cmd(const struct ethtool_link_ksettings *cmd) | |
16032be5 | 1834 | { |
ebb6b4b1 PR |
1835 | struct ethtool_link_ksettings diff1 = *cmd; |
1836 | struct ethtool_link_ksettings diff2 = {}; | |
16032be5 | 1837 | |
0cf3ace9 NA |
1838 | /* cmd is always set so we need to clear it, validate the port type |
1839 | * and also without autonegotiation we can ignore advertising | |
1840 | */ | |
ebb6b4b1 PR |
1841 | diff1.base.speed = 0; |
1842 | diff2.base.port = PORT_OTHER; | |
1843 | ethtool_link_ksettings_zero_link_mode(&diff1, advertising); | |
1844 | diff1.base.duplex = 0; | |
1845 | diff1.base.cmd = 0; | |
1846 | diff1.base.link_mode_masks_nwords = 0; | |
1847 | ||
1848 | return !memcmp(&diff1.base, &diff2.base, sizeof(diff1.base)) && | |
1849 | bitmap_empty(diff1.link_modes.supported, | |
1850 | __ETHTOOL_LINK_MODE_MASK_NBITS) && | |
1851 | bitmap_empty(diff1.link_modes.advertising, | |
1852 | __ETHTOOL_LINK_MODE_MASK_NBITS) && | |
1853 | bitmap_empty(diff1.link_modes.lp_advertising, | |
1854 | __ETHTOOL_LINK_MODE_MASK_NBITS); | |
16032be5 NA |
1855 | } |
1856 | ||
ebb6b4b1 PR |
1857 | static int virtnet_set_link_ksettings(struct net_device *dev, |
1858 | const struct ethtool_link_ksettings *cmd) | |
16032be5 NA |
1859 | { |
1860 | struct virtnet_info *vi = netdev_priv(dev); | |
1861 | u32 speed; | |
1862 | ||
ebb6b4b1 | 1863 | speed = cmd->base.speed; |
16032be5 NA |
1864 | /* don't allow custom speed and duplex */ |
1865 | if (!ethtool_validate_speed(speed) || | |
ebb6b4b1 | 1866 | !ethtool_validate_duplex(cmd->base.duplex) || |
16032be5 NA |
1867 | !virtnet_validate_ethtool_cmd(cmd)) |
1868 | return -EINVAL; | |
1869 | vi->speed = speed; | |
ebb6b4b1 | 1870 | vi->duplex = cmd->base.duplex; |
16032be5 NA |
1871 | |
1872 | return 0; | |
1873 | } | |
1874 | ||
ebb6b4b1 PR |
1875 | static int virtnet_get_link_ksettings(struct net_device *dev, |
1876 | struct ethtool_link_ksettings *cmd) | |
16032be5 NA |
1877 | { |
1878 | struct virtnet_info *vi = netdev_priv(dev); | |
1879 | ||
ebb6b4b1 PR |
1880 | cmd->base.speed = vi->speed; |
1881 | cmd->base.duplex = vi->duplex; | |
1882 | cmd->base.port = PORT_OTHER; | |
16032be5 NA |
1883 | |
1884 | return 0; | |
1885 | } | |
1886 | ||
1887 | static void virtnet_init_settings(struct net_device *dev) | |
1888 | { | |
1889 | struct virtnet_info *vi = netdev_priv(dev); | |
1890 | ||
1891 | vi->speed = SPEED_UNKNOWN; | |
1892 | vi->duplex = DUPLEX_UNKNOWN; | |
1893 | } | |
1894 | ||
0fc0b732 | 1895 | static const struct ethtool_ops virtnet_ethtool_ops = { |
66846048 | 1896 | .get_drvinfo = virtnet_get_drvinfo, |
9f4d26d0 | 1897 | .get_link = ethtool_op_get_link, |
8f9f4668 | 1898 | .get_ringparam = virtnet_get_ringparam, |
d73bcd2c JW |
1899 | .set_channels = virtnet_set_channels, |
1900 | .get_channels = virtnet_get_channels, | |
074c3582 | 1901 | .get_ts_info = ethtool_op_get_ts_info, |
ebb6b4b1 PR |
1902 | .get_link_ksettings = virtnet_get_link_ksettings, |
1903 | .set_link_ksettings = virtnet_set_link_ksettings, | |
a9ea3fc6 HX |
1904 | }; |
1905 | ||
9fe7bfce JF |
1906 | static void virtnet_freeze_down(struct virtio_device *vdev) |
1907 | { | |
1908 | struct virtnet_info *vi = vdev->priv; | |
1909 | int i; | |
1910 | ||
1911 | /* Make sure no work handler is accessing the device */ | |
1912 | flush_work(&vi->config_work); | |
1913 | ||
1914 | netif_device_detach(vi->dev); | |
713a98d9 | 1915 | netif_tx_disable(vi->dev); |
9fe7bfce JF |
1916 | cancel_delayed_work_sync(&vi->refill); |
1917 | ||
1918 | if (netif_running(vi->dev)) { | |
b92f1e67 | 1919 | for (i = 0; i < vi->max_queue_pairs; i++) { |
9fe7bfce | 1920 | napi_disable(&vi->rq[i].napi); |
78a57b48 | 1921 | virtnet_napi_tx_disable(&vi->sq[i].napi); |
b92f1e67 | 1922 | } |
9fe7bfce JF |
1923 | } |
1924 | } | |
1925 | ||
1926 | static int init_vqs(struct virtnet_info *vi); | |
1927 | ||
1928 | static int virtnet_restore_up(struct virtio_device *vdev) | |
1929 | { | |
1930 | struct virtnet_info *vi = vdev->priv; | |
1931 | int err, i; | |
1932 | ||
1933 | err = init_vqs(vi); | |
1934 | if (err) | |
1935 | return err; | |
1936 | ||
1937 | virtio_device_ready(vdev); | |
1938 | ||
1939 | if (netif_running(vi->dev)) { | |
1940 | for (i = 0; i < vi->curr_queue_pairs; i++) | |
1941 | if (!try_fill_recv(vi, &vi->rq[i], GFP_KERNEL)) | |
1942 | schedule_delayed_work(&vi->refill, 0); | |
1943 | ||
b92f1e67 | 1944 | for (i = 0; i < vi->max_queue_pairs; i++) { |
e4e8452a | 1945 | virtnet_napi_enable(vi->rq[i].vq, &vi->rq[i].napi); |
b92f1e67 WB |
1946 | virtnet_napi_tx_enable(vi, vi->sq[i].vq, |
1947 | &vi->sq[i].napi); | |
1948 | } | |
9fe7bfce JF |
1949 | } |
1950 | ||
1951 | netif_device_attach(vi->dev); | |
1952 | return err; | |
1953 | } | |
1954 | ||
3f93522f JW |
1955 | static int virtnet_set_guest_offloads(struct virtnet_info *vi, u64 offloads) |
1956 | { | |
1957 | struct scatterlist sg; | |
1958 | vi->ctrl_offloads = cpu_to_virtio64(vi->vdev, offloads); | |
1959 | ||
1960 | sg_init_one(&sg, &vi->ctrl_offloads, sizeof(vi->ctrl_offloads)); | |
1961 | ||
1962 | if (!virtnet_send_command(vi, VIRTIO_NET_CTRL_GUEST_OFFLOADS, | |
1963 | VIRTIO_NET_CTRL_GUEST_OFFLOADS_SET, &sg)) { | |
1964 | dev_warn(&vi->dev->dev, "Fail to set guest offload. \n"); | |
1965 | return -EINVAL; | |
1966 | } | |
1967 | ||
1968 | return 0; | |
1969 | } | |
1970 | ||
1971 | static int virtnet_clear_guest_offloads(struct virtnet_info *vi) | |
1972 | { | |
1973 | u64 offloads = 0; | |
1974 | ||
1975 | if (!vi->guest_offloads) | |
1976 | return 0; | |
1977 | ||
1978 | if (virtio_has_feature(vi->vdev, VIRTIO_NET_F_GUEST_CSUM)) | |
1979 | offloads = 1ULL << VIRTIO_NET_F_GUEST_CSUM; | |
1980 | ||
1981 | return virtnet_set_guest_offloads(vi, offloads); | |
1982 | } | |
1983 | ||
1984 | static int virtnet_restore_guest_offloads(struct virtnet_info *vi) | |
1985 | { | |
1986 | u64 offloads = vi->guest_offloads; | |
1987 | ||
1988 | if (!vi->guest_offloads) | |
1989 | return 0; | |
1990 | if (virtio_has_feature(vi->vdev, VIRTIO_NET_F_GUEST_CSUM)) | |
1991 | offloads |= 1ULL << VIRTIO_NET_F_GUEST_CSUM; | |
1992 | ||
1993 | return virtnet_set_guest_offloads(vi, offloads); | |
1994 | } | |
1995 | ||
9861ce03 JK |
1996 | static int virtnet_xdp_set(struct net_device *dev, struct bpf_prog *prog, |
1997 | struct netlink_ext_ack *extack) | |
f600b690 JF |
1998 | { |
1999 | unsigned long int max_sz = PAGE_SIZE - sizeof(struct padded_vnet_hdr); | |
2000 | struct virtnet_info *vi = netdev_priv(dev); | |
2001 | struct bpf_prog *old_prog; | |
017b29c3 | 2002 | u16 xdp_qp = 0, curr_qp; |
672aafd5 | 2003 | int i, err; |
f600b690 | 2004 | |
3f93522f JW |
2005 | if (!virtio_has_feature(vi->vdev, VIRTIO_NET_F_CTRL_GUEST_OFFLOADS) |
2006 | && (virtio_has_feature(vi->vdev, VIRTIO_NET_F_GUEST_TSO4) || | |
2007 | virtio_has_feature(vi->vdev, VIRTIO_NET_F_GUEST_TSO6) || | |
2008 | virtio_has_feature(vi->vdev, VIRTIO_NET_F_GUEST_ECN) || | |
2009 | virtio_has_feature(vi->vdev, VIRTIO_NET_F_GUEST_UFO))) { | |
4d463c4d | 2010 | NL_SET_ERR_MSG_MOD(extack, "Can't set XDP while host is implementing LRO, disable LRO first"); |
f600b690 JF |
2011 | return -EOPNOTSUPP; |
2012 | } | |
2013 | ||
2014 | if (vi->mergeable_rx_bufs && !vi->any_header_sg) { | |
4d463c4d | 2015 | NL_SET_ERR_MSG_MOD(extack, "XDP expects header/data in single page, any_header_sg required"); |
f600b690 JF |
2016 | return -EINVAL; |
2017 | } | |
2018 | ||
2019 | if (dev->mtu > max_sz) { | |
4d463c4d | 2020 | NL_SET_ERR_MSG_MOD(extack, "MTU too large to enable XDP"); |
f600b690 JF |
2021 | netdev_warn(dev, "XDP requires MTU less than %lu\n", max_sz); |
2022 | return -EINVAL; | |
2023 | } | |
2024 | ||
672aafd5 JF |
2025 | curr_qp = vi->curr_queue_pairs - vi->xdp_queue_pairs; |
2026 | if (prog) | |
2027 | xdp_qp = nr_cpu_ids; | |
2028 | ||
2029 | /* XDP requires extra queues for XDP_TX */ | |
2030 | if (curr_qp + xdp_qp > vi->max_queue_pairs) { | |
4d463c4d | 2031 | NL_SET_ERR_MSG_MOD(extack, "Too few free TX rings available"); |
672aafd5 JF |
2032 | netdev_warn(dev, "request %i queues but max is %i\n", |
2033 | curr_qp + xdp_qp, vi->max_queue_pairs); | |
2034 | return -ENOMEM; | |
2035 | } | |
2036 | ||
2de2f7f4 JF |
2037 | if (prog) { |
2038 | prog = bpf_prog_add(prog, vi->max_queue_pairs - 1); | |
2039 | if (IS_ERR(prog)) | |
2040 | return PTR_ERR(prog); | |
2041 | } | |
2042 | ||
4941d472 JW |
2043 | /* Make sure NAPI is not using any XDP TX queues for RX. */ |
2044 | for (i = 0; i < vi->max_queue_pairs; i++) | |
2045 | napi_disable(&vi->rq[i].napi); | |
f600b690 | 2046 | |
672aafd5 | 2047 | netif_set_real_num_rx_queues(dev, curr_qp + xdp_qp); |
4941d472 JW |
2048 | err = _virtnet_set_queues(vi, curr_qp + xdp_qp); |
2049 | if (err) | |
2050 | goto err; | |
2051 | vi->xdp_queue_pairs = xdp_qp; | |
672aafd5 | 2052 | |
f600b690 JF |
2053 | for (i = 0; i < vi->max_queue_pairs; i++) { |
2054 | old_prog = rtnl_dereference(vi->rq[i].xdp_prog); | |
2055 | rcu_assign_pointer(vi->rq[i].xdp_prog, prog); | |
3f93522f JW |
2056 | if (i == 0) { |
2057 | if (!old_prog) | |
2058 | virtnet_clear_guest_offloads(vi); | |
2059 | if (!prog) | |
2060 | virtnet_restore_guest_offloads(vi); | |
2061 | } | |
f600b690 JF |
2062 | if (old_prog) |
2063 | bpf_prog_put(old_prog); | |
4941d472 | 2064 | virtnet_napi_enable(vi->rq[i].vq, &vi->rq[i].napi); |
f600b690 JF |
2065 | } |
2066 | ||
2067 | return 0; | |
2de2f7f4 | 2068 | |
4941d472 JW |
2069 | err: |
2070 | for (i = 0; i < vi->max_queue_pairs; i++) | |
2071 | virtnet_napi_enable(vi->rq[i].vq, &vi->rq[i].napi); | |
2de2f7f4 JF |
2072 | if (prog) |
2073 | bpf_prog_sub(prog, vi->max_queue_pairs - 1); | |
2074 | return err; | |
f600b690 JF |
2075 | } |
2076 | ||
5b0e6629 | 2077 | static u32 virtnet_xdp_query(struct net_device *dev) |
f600b690 JF |
2078 | { |
2079 | struct virtnet_info *vi = netdev_priv(dev); | |
5b0e6629 | 2080 | const struct bpf_prog *xdp_prog; |
f600b690 JF |
2081 | int i; |
2082 | ||
2083 | for (i = 0; i < vi->max_queue_pairs; i++) { | |
5b0e6629 MKL |
2084 | xdp_prog = rtnl_dereference(vi->rq[i].xdp_prog); |
2085 | if (xdp_prog) | |
2086 | return xdp_prog->aux->id; | |
f600b690 | 2087 | } |
5b0e6629 | 2088 | return 0; |
f600b690 JF |
2089 | } |
2090 | ||
2091 | static int virtnet_xdp(struct net_device *dev, struct netdev_xdp *xdp) | |
2092 | { | |
2093 | switch (xdp->command) { | |
2094 | case XDP_SETUP_PROG: | |
9861ce03 | 2095 | return virtnet_xdp_set(dev, xdp->prog, xdp->extack); |
f600b690 | 2096 | case XDP_QUERY_PROG: |
5b0e6629 MKL |
2097 | xdp->prog_id = virtnet_xdp_query(dev); |
2098 | xdp->prog_attached = !!xdp->prog_id; | |
f600b690 JF |
2099 | return 0; |
2100 | default: | |
2101 | return -EINVAL; | |
2102 | } | |
2103 | } | |
2104 | ||
76288b4e SH |
2105 | static const struct net_device_ops virtnet_netdev = { |
2106 | .ndo_open = virtnet_open, | |
2107 | .ndo_stop = virtnet_close, | |
2108 | .ndo_start_xmit = start_xmit, | |
2109 | .ndo_validate_addr = eth_validate_addr, | |
9c46f6d4 | 2110 | .ndo_set_mac_address = virtnet_set_mac_address, |
2af7698e | 2111 | .ndo_set_rx_mode = virtnet_set_rx_mode, |
3fa2a1df | 2112 | .ndo_get_stats64 = virtnet_stats, |
1824a989 AW |
2113 | .ndo_vlan_rx_add_vid = virtnet_vlan_rx_add_vid, |
2114 | .ndo_vlan_rx_kill_vid = virtnet_vlan_rx_kill_vid, | |
76288b4e SH |
2115 | #ifdef CONFIG_NET_POLL_CONTROLLER |
2116 | .ndo_poll_controller = virtnet_netpoll, | |
91815639 | 2117 | #endif |
f600b690 | 2118 | .ndo_xdp = virtnet_xdp, |
186b3c99 JW |
2119 | .ndo_xdp_xmit = virtnet_xdp_xmit, |
2120 | .ndo_xdp_flush = virtnet_xdp_flush, | |
2836b4f2 | 2121 | .ndo_features_check = passthru_features_check, |
76288b4e SH |
2122 | }; |
2123 | ||
586d17c5 | 2124 | static void virtnet_config_changed_work(struct work_struct *work) |
9f4d26d0 | 2125 | { |
586d17c5 JW |
2126 | struct virtnet_info *vi = |
2127 | container_of(work, struct virtnet_info, config_work); | |
9f4d26d0 MM |
2128 | u16 v; |
2129 | ||
855e0c52 RR |
2130 | if (virtio_cread_feature(vi->vdev, VIRTIO_NET_F_STATUS, |
2131 | struct virtio_net_config, status, &v) < 0) | |
507613bf | 2132 | return; |
586d17c5 JW |
2133 | |
2134 | if (v & VIRTIO_NET_S_ANNOUNCE) { | |
ee89bab1 | 2135 | netdev_notify_peers(vi->dev); |
586d17c5 JW |
2136 | virtnet_ack_link_announce(vi); |
2137 | } | |
9f4d26d0 MM |
2138 | |
2139 | /* Ignore unknown (future) status bits */ | |
2140 | v &= VIRTIO_NET_S_LINK_UP; | |
2141 | ||
2142 | if (vi->status == v) | |
507613bf | 2143 | return; |
9f4d26d0 MM |
2144 | |
2145 | vi->status = v; | |
2146 | ||
2147 | if (vi->status & VIRTIO_NET_S_LINK_UP) { | |
2148 | netif_carrier_on(vi->dev); | |
986a4f4d | 2149 | netif_tx_wake_all_queues(vi->dev); |
9f4d26d0 MM |
2150 | } else { |
2151 | netif_carrier_off(vi->dev); | |
986a4f4d | 2152 | netif_tx_stop_all_queues(vi->dev); |
9f4d26d0 MM |
2153 | } |
2154 | } | |
2155 | ||
2156 | static void virtnet_config_changed(struct virtio_device *vdev) | |
2157 | { | |
2158 | struct virtnet_info *vi = vdev->priv; | |
2159 | ||
3b07e9ca | 2160 | schedule_work(&vi->config_work); |
9f4d26d0 MM |
2161 | } |
2162 | ||
986a4f4d JW |
2163 | static void virtnet_free_queues(struct virtnet_info *vi) |
2164 | { | |
d4fb84ee AV |
2165 | int i; |
2166 | ||
ab3971b1 JW |
2167 | for (i = 0; i < vi->max_queue_pairs; i++) { |
2168 | napi_hash_del(&vi->rq[i].napi); | |
d4fb84ee | 2169 | netif_napi_del(&vi->rq[i].napi); |
b92f1e67 | 2170 | netif_napi_del(&vi->sq[i].napi); |
ab3971b1 | 2171 | } |
d4fb84ee | 2172 | |
963abe5c ED |
2173 | /* We called napi_hash_del() before netif_napi_del(), |
2174 | * we need to respect an RCU grace period before freeing vi->rq | |
2175 | */ | |
2176 | synchronize_net(); | |
2177 | ||
986a4f4d JW |
2178 | kfree(vi->rq); |
2179 | kfree(vi->sq); | |
2180 | } | |
2181 | ||
47315329 | 2182 | static void _free_receive_bufs(struct virtnet_info *vi) |
986a4f4d | 2183 | { |
f600b690 | 2184 | struct bpf_prog *old_prog; |
986a4f4d JW |
2185 | int i; |
2186 | ||
2187 | for (i = 0; i < vi->max_queue_pairs; i++) { | |
2188 | while (vi->rq[i].pages) | |
2189 | __free_pages(get_a_page(&vi->rq[i], GFP_KERNEL), 0); | |
f600b690 JF |
2190 | |
2191 | old_prog = rtnl_dereference(vi->rq[i].xdp_prog); | |
2192 | RCU_INIT_POINTER(vi->rq[i].xdp_prog, NULL); | |
2193 | if (old_prog) | |
2194 | bpf_prog_put(old_prog); | |
986a4f4d | 2195 | } |
47315329 JF |
2196 | } |
2197 | ||
2198 | static void free_receive_bufs(struct virtnet_info *vi) | |
2199 | { | |
2200 | rtnl_lock(); | |
2201 | _free_receive_bufs(vi); | |
f600b690 | 2202 | rtnl_unlock(); |
986a4f4d JW |
2203 | } |
2204 | ||
fb51879d MD |
2205 | static void free_receive_page_frags(struct virtnet_info *vi) |
2206 | { | |
2207 | int i; | |
2208 | for (i = 0; i < vi->max_queue_pairs; i++) | |
2209 | if (vi->rq[i].alloc_frag.page) | |
2210 | put_page(vi->rq[i].alloc_frag.page); | |
2211 | } | |
2212 | ||
b68df015 | 2213 | static bool is_xdp_raw_buffer_queue(struct virtnet_info *vi, int q) |
56434a01 JF |
2214 | { |
2215 | if (q < (vi->curr_queue_pairs - vi->xdp_queue_pairs)) | |
2216 | return false; | |
2217 | else if (q < vi->curr_queue_pairs) | |
2218 | return true; | |
2219 | else | |
2220 | return false; | |
2221 | } | |
2222 | ||
986a4f4d JW |
2223 | static void free_unused_bufs(struct virtnet_info *vi) |
2224 | { | |
2225 | void *buf; | |
2226 | int i; | |
2227 | ||
2228 | for (i = 0; i < vi->max_queue_pairs; i++) { | |
2229 | struct virtqueue *vq = vi->sq[i].vq; | |
56434a01 | 2230 | while ((buf = virtqueue_detach_unused_buf(vq)) != NULL) { |
b68df015 | 2231 | if (!is_xdp_raw_buffer_queue(vi, i)) |
56434a01 JF |
2232 | dev_kfree_skb(buf); |
2233 | else | |
2234 | put_page(virt_to_head_page(buf)); | |
2235 | } | |
986a4f4d JW |
2236 | } |
2237 | ||
2238 | for (i = 0; i < vi->max_queue_pairs; i++) { | |
2239 | struct virtqueue *vq = vi->rq[i].vq; | |
2240 | ||
2241 | while ((buf = virtqueue_detach_unused_buf(vq)) != NULL) { | |
ab7db917 | 2242 | if (vi->mergeable_rx_bufs) { |
680557cf | 2243 | put_page(virt_to_head_page(buf)); |
ab7db917 | 2244 | } else if (vi->big_packets) { |
fa9fac17 | 2245 | give_pages(&vi->rq[i], buf); |
ab7db917 | 2246 | } else { |
f6b10209 | 2247 | put_page(virt_to_head_page(buf)); |
ab7db917 | 2248 | } |
986a4f4d | 2249 | } |
986a4f4d JW |
2250 | } |
2251 | } | |
2252 | ||
e9d7417b JW |
2253 | static void virtnet_del_vqs(struct virtnet_info *vi) |
2254 | { | |
2255 | struct virtio_device *vdev = vi->vdev; | |
2256 | ||
8898c21c | 2257 | virtnet_clean_affinity(vi, -1); |
986a4f4d | 2258 | |
e9d7417b | 2259 | vdev->config->del_vqs(vdev); |
986a4f4d JW |
2260 | |
2261 | virtnet_free_queues(vi); | |
e9d7417b JW |
2262 | } |
2263 | ||
d85b758f MT |
2264 | /* How large should a single buffer be so a queue full of these can fit at |
2265 | * least one full packet? | |
2266 | * Logic below assumes the mergeable buffer header is used. | |
2267 | */ | |
2268 | static unsigned int mergeable_min_buf_len(struct virtnet_info *vi, struct virtqueue *vq) | |
2269 | { | |
2270 | const unsigned int hdr_len = sizeof(struct virtio_net_hdr_mrg_rxbuf); | |
2271 | unsigned int rq_size = virtqueue_get_vring_size(vq); | |
2272 | unsigned int packet_len = vi->big_packets ? IP_MAX_MTU : vi->dev->max_mtu; | |
2273 | unsigned int buf_len = hdr_len + ETH_HLEN + VLAN_HLEN + packet_len; | |
2274 | unsigned int min_buf_len = DIV_ROUND_UP(buf_len, rq_size); | |
2275 | ||
f0c3192c MT |
2276 | return max(max(min_buf_len, hdr_len) - hdr_len, |
2277 | (unsigned int)GOOD_PACKET_LEN); | |
d85b758f MT |
2278 | } |
2279 | ||
986a4f4d | 2280 | static int virtnet_find_vqs(struct virtnet_info *vi) |
3f9c10b0 | 2281 | { |
986a4f4d JW |
2282 | vq_callback_t **callbacks; |
2283 | struct virtqueue **vqs; | |
2284 | int ret = -ENOMEM; | |
2285 | int i, total_vqs; | |
2286 | const char **names; | |
d45b897b | 2287 | bool *ctx; |
986a4f4d JW |
2288 | |
2289 | /* We expect 1 RX virtqueue followed by 1 TX virtqueue, followed by | |
2290 | * possible N-1 RX/TX queue pairs used in multiqueue mode, followed by | |
2291 | * possible control vq. | |
2292 | */ | |
2293 | total_vqs = vi->max_queue_pairs * 2 + | |
2294 | virtio_has_feature(vi->vdev, VIRTIO_NET_F_CTRL_VQ); | |
2295 | ||
2296 | /* Allocate space for find_vqs parameters */ | |
2297 | vqs = kzalloc(total_vqs * sizeof(*vqs), GFP_KERNEL); | |
2298 | if (!vqs) | |
2299 | goto err_vq; | |
2300 | callbacks = kmalloc(total_vqs * sizeof(*callbacks), GFP_KERNEL); | |
2301 | if (!callbacks) | |
2302 | goto err_callback; | |
2303 | names = kmalloc(total_vqs * sizeof(*names), GFP_KERNEL); | |
2304 | if (!names) | |
2305 | goto err_names; | |
192f68cf | 2306 | if (!vi->big_packets || vi->mergeable_rx_bufs) { |
d45b897b MT |
2307 | ctx = kzalloc(total_vqs * sizeof(*ctx), GFP_KERNEL); |
2308 | if (!ctx) | |
2309 | goto err_ctx; | |
2310 | } else { | |
2311 | ctx = NULL; | |
2312 | } | |
986a4f4d JW |
2313 | |
2314 | /* Parameters for control virtqueue, if any */ | |
2315 | if (vi->has_cvq) { | |
2316 | callbacks[total_vqs - 1] = NULL; | |
2317 | names[total_vqs - 1] = "control"; | |
2318 | } | |
3f9c10b0 | 2319 | |
986a4f4d JW |
2320 | /* Allocate/initialize parameters for send/receive virtqueues */ |
2321 | for (i = 0; i < vi->max_queue_pairs; i++) { | |
2322 | callbacks[rxq2vq(i)] = skb_recv_done; | |
2323 | callbacks[txq2vq(i)] = skb_xmit_done; | |
2324 | sprintf(vi->rq[i].name, "input.%d", i); | |
2325 | sprintf(vi->sq[i].name, "output.%d", i); | |
2326 | names[rxq2vq(i)] = vi->rq[i].name; | |
2327 | names[txq2vq(i)] = vi->sq[i].name; | |
d45b897b MT |
2328 | if (ctx) |
2329 | ctx[rxq2vq(i)] = true; | |
986a4f4d | 2330 | } |
3f9c10b0 | 2331 | |
986a4f4d | 2332 | ret = vi->vdev->config->find_vqs(vi->vdev, total_vqs, vqs, callbacks, |
d45b897b | 2333 | names, ctx, NULL); |
986a4f4d JW |
2334 | if (ret) |
2335 | goto err_find; | |
3f9c10b0 | 2336 | |
986a4f4d JW |
2337 | if (vi->has_cvq) { |
2338 | vi->cvq = vqs[total_vqs - 1]; | |
3f9c10b0 | 2339 | if (virtio_has_feature(vi->vdev, VIRTIO_NET_F_CTRL_VLAN)) |
f646968f | 2340 | vi->dev->features |= NETIF_F_HW_VLAN_CTAG_FILTER; |
3f9c10b0 | 2341 | } |
986a4f4d JW |
2342 | |
2343 | for (i = 0; i < vi->max_queue_pairs; i++) { | |
2344 | vi->rq[i].vq = vqs[rxq2vq(i)]; | |
d85b758f | 2345 | vi->rq[i].min_buf_len = mergeable_min_buf_len(vi, vi->rq[i].vq); |
986a4f4d JW |
2346 | vi->sq[i].vq = vqs[txq2vq(i)]; |
2347 | } | |
2348 | ||
2349 | kfree(names); | |
2350 | kfree(callbacks); | |
2351 | kfree(vqs); | |
55281621 | 2352 | kfree(ctx); |
986a4f4d | 2353 | |
3f9c10b0 | 2354 | return 0; |
986a4f4d JW |
2355 | |
2356 | err_find: | |
d45b897b MT |
2357 | kfree(ctx); |
2358 | err_ctx: | |
986a4f4d JW |
2359 | kfree(names); |
2360 | err_names: | |
2361 | kfree(callbacks); | |
2362 | err_callback: | |
2363 | kfree(vqs); | |
2364 | err_vq: | |
2365 | return ret; | |
2366 | } | |
2367 | ||
2368 | static int virtnet_alloc_queues(struct virtnet_info *vi) | |
2369 | { | |
2370 | int i; | |
2371 | ||
2372 | vi->sq = kzalloc(sizeof(*vi->sq) * vi->max_queue_pairs, GFP_KERNEL); | |
2373 | if (!vi->sq) | |
2374 | goto err_sq; | |
2375 | vi->rq = kzalloc(sizeof(*vi->rq) * vi->max_queue_pairs, GFP_KERNEL); | |
008d4278 | 2376 | if (!vi->rq) |
986a4f4d JW |
2377 | goto err_rq; |
2378 | ||
2379 | INIT_DELAYED_WORK(&vi->refill, refill_work); | |
2380 | for (i = 0; i < vi->max_queue_pairs; i++) { | |
2381 | vi->rq[i].pages = NULL; | |
2382 | netif_napi_add(vi->dev, &vi->rq[i].napi, virtnet_poll, | |
2383 | napi_weight); | |
1d11e732 WB |
2384 | netif_tx_napi_add(vi->dev, &vi->sq[i].napi, virtnet_poll_tx, |
2385 | napi_tx ? napi_weight : 0); | |
986a4f4d JW |
2386 | |
2387 | sg_init_table(vi->rq[i].sg, ARRAY_SIZE(vi->rq[i].sg)); | |
5377d758 | 2388 | ewma_pkt_len_init(&vi->rq[i].mrg_avg_pkt_len); |
986a4f4d JW |
2389 | sg_init_table(vi->sq[i].sg, ARRAY_SIZE(vi->sq[i].sg)); |
2390 | } | |
2391 | ||
2392 | return 0; | |
2393 | ||
2394 | err_rq: | |
2395 | kfree(vi->sq); | |
2396 | err_sq: | |
2397 | return -ENOMEM; | |
2398 | } | |
2399 | ||
2400 | static int init_vqs(struct virtnet_info *vi) | |
2401 | { | |
2402 | int ret; | |
2403 | ||
2404 | /* Allocate send & receive queues */ | |
2405 | ret = virtnet_alloc_queues(vi); | |
2406 | if (ret) | |
2407 | goto err; | |
2408 | ||
2409 | ret = virtnet_find_vqs(vi); | |
2410 | if (ret) | |
2411 | goto err_free; | |
2412 | ||
47be2479 | 2413 | get_online_cpus(); |
8898c21c | 2414 | virtnet_set_affinity(vi); |
47be2479 WG |
2415 | put_online_cpus(); |
2416 | ||
986a4f4d JW |
2417 | return 0; |
2418 | ||
2419 | err_free: | |
2420 | virtnet_free_queues(vi); | |
2421 | err: | |
2422 | return ret; | |
3f9c10b0 AS |
2423 | } |
2424 | ||
fbf28d78 MD |
2425 | #ifdef CONFIG_SYSFS |
2426 | static ssize_t mergeable_rx_buffer_size_show(struct netdev_rx_queue *queue, | |
718ad681 | 2427 | char *buf) |
fbf28d78 MD |
2428 | { |
2429 | struct virtnet_info *vi = netdev_priv(queue->dev); | |
2430 | unsigned int queue_index = get_netdev_rx_queue_index(queue); | |
5377d758 | 2431 | struct ewma_pkt_len *avg; |
fbf28d78 MD |
2432 | |
2433 | BUG_ON(queue_index >= vi->max_queue_pairs); | |
2434 | avg = &vi->rq[queue_index].mrg_avg_pkt_len; | |
d85b758f MT |
2435 | return sprintf(buf, "%u\n", |
2436 | get_mergeable_buf_len(&vi->rq[queue_index], avg)); | |
fbf28d78 MD |
2437 | } |
2438 | ||
2439 | static struct rx_queue_attribute mergeable_rx_buffer_size_attribute = | |
2440 | __ATTR_RO(mergeable_rx_buffer_size); | |
2441 | ||
2442 | static struct attribute *virtio_net_mrg_rx_attrs[] = { | |
2443 | &mergeable_rx_buffer_size_attribute.attr, | |
2444 | NULL | |
2445 | }; | |
2446 | ||
2447 | static const struct attribute_group virtio_net_mrg_rx_group = { | |
2448 | .name = "virtio_net", | |
2449 | .attrs = virtio_net_mrg_rx_attrs | |
2450 | }; | |
2451 | #endif | |
2452 | ||
892d6eb1 JW |
2453 | static bool virtnet_fail_on_feature(struct virtio_device *vdev, |
2454 | unsigned int fbit, | |
2455 | const char *fname, const char *dname) | |
2456 | { | |
2457 | if (!virtio_has_feature(vdev, fbit)) | |
2458 | return false; | |
2459 | ||
2460 | dev_err(&vdev->dev, "device advertises feature %s but not %s", | |
2461 | fname, dname); | |
2462 | ||
2463 | return true; | |
2464 | } | |
2465 | ||
2466 | #define VIRTNET_FAIL_ON(vdev, fbit, dbit) \ | |
2467 | virtnet_fail_on_feature(vdev, fbit, #fbit, dbit) | |
2468 | ||
2469 | static bool virtnet_validate_features(struct virtio_device *vdev) | |
2470 | { | |
2471 | if (!virtio_has_feature(vdev, VIRTIO_NET_F_CTRL_VQ) && | |
2472 | (VIRTNET_FAIL_ON(vdev, VIRTIO_NET_F_CTRL_RX, | |
2473 | "VIRTIO_NET_F_CTRL_VQ") || | |
2474 | VIRTNET_FAIL_ON(vdev, VIRTIO_NET_F_CTRL_VLAN, | |
2475 | "VIRTIO_NET_F_CTRL_VQ") || | |
2476 | VIRTNET_FAIL_ON(vdev, VIRTIO_NET_F_GUEST_ANNOUNCE, | |
2477 | "VIRTIO_NET_F_CTRL_VQ") || | |
2478 | VIRTNET_FAIL_ON(vdev, VIRTIO_NET_F_MQ, "VIRTIO_NET_F_CTRL_VQ") || | |
2479 | VIRTNET_FAIL_ON(vdev, VIRTIO_NET_F_CTRL_MAC_ADDR, | |
2480 | "VIRTIO_NET_F_CTRL_VQ"))) { | |
2481 | return false; | |
2482 | } | |
2483 | ||
2484 | return true; | |
2485 | } | |
2486 | ||
d0c2c997 JW |
2487 | #define MIN_MTU ETH_MIN_MTU |
2488 | #define MAX_MTU ETH_MAX_MTU | |
2489 | ||
fe36cbe0 | 2490 | static int virtnet_validate(struct virtio_device *vdev) |
296f96fc | 2491 | { |
6ba42248 MT |
2492 | if (!vdev->config->get) { |
2493 | dev_err(&vdev->dev, "%s failure: config access disabled\n", | |
2494 | __func__); | |
2495 | return -EINVAL; | |
2496 | } | |
2497 | ||
892d6eb1 JW |
2498 | if (!virtnet_validate_features(vdev)) |
2499 | return -EINVAL; | |
2500 | ||
fe36cbe0 MT |
2501 | if (virtio_has_feature(vdev, VIRTIO_NET_F_MTU)) { |
2502 | int mtu = virtio_cread16(vdev, | |
2503 | offsetof(struct virtio_net_config, | |
2504 | mtu)); | |
2505 | if (mtu < MIN_MTU) | |
2506 | __virtio_clear_bit(vdev, VIRTIO_NET_F_MTU); | |
2507 | } | |
2508 | ||
2509 | return 0; | |
2510 | } | |
2511 | ||
2512 | static int virtnet_probe(struct virtio_device *vdev) | |
2513 | { | |
2514 | int i, err; | |
2515 | struct net_device *dev; | |
2516 | struct virtnet_info *vi; | |
2517 | u16 max_queue_pairs; | |
2518 | int mtu; | |
2519 | ||
986a4f4d | 2520 | /* Find if host supports multiqueue virtio_net device */ |
855e0c52 RR |
2521 | err = virtio_cread_feature(vdev, VIRTIO_NET_F_MQ, |
2522 | struct virtio_net_config, | |
2523 | max_virtqueue_pairs, &max_queue_pairs); | |
986a4f4d JW |
2524 | |
2525 | /* We need at least 2 queue's */ | |
2526 | if (err || max_queue_pairs < VIRTIO_NET_CTRL_MQ_VQ_PAIRS_MIN || | |
2527 | max_queue_pairs > VIRTIO_NET_CTRL_MQ_VQ_PAIRS_MAX || | |
2528 | !virtio_has_feature(vdev, VIRTIO_NET_F_CTRL_VQ)) | |
2529 | max_queue_pairs = 1; | |
296f96fc RR |
2530 | |
2531 | /* Allocate ourselves a network device with room for our info */ | |
986a4f4d | 2532 | dev = alloc_etherdev_mq(sizeof(struct virtnet_info), max_queue_pairs); |
296f96fc RR |
2533 | if (!dev) |
2534 | return -ENOMEM; | |
2535 | ||
2536 | /* Set up network device as normal. */ | |
f2f2c8b4 | 2537 | dev->priv_flags |= IFF_UNICAST_FLT | IFF_LIVE_ADDR_CHANGE; |
76288b4e | 2538 | dev->netdev_ops = &virtnet_netdev; |
296f96fc | 2539 | dev->features = NETIF_F_HIGHDMA; |
3fa2a1df | 2540 | |
7ad24ea4 | 2541 | dev->ethtool_ops = &virtnet_ethtool_ops; |
296f96fc RR |
2542 | SET_NETDEV_DEV(dev, &vdev->dev); |
2543 | ||
2544 | /* Do we support "hardware" checksums? */ | |
98e778c9 | 2545 | if (virtio_has_feature(vdev, VIRTIO_NET_F_CSUM)) { |
296f96fc | 2546 | /* This opens up the world of extra features. */ |
48900cb6 | 2547 | dev->hw_features |= NETIF_F_HW_CSUM | NETIF_F_SG; |
98e778c9 | 2548 | if (csum) |
48900cb6 | 2549 | dev->features |= NETIF_F_HW_CSUM | NETIF_F_SG; |
98e778c9 MM |
2550 | |
2551 | if (virtio_has_feature(vdev, VIRTIO_NET_F_GSO)) { | |
e078de03 | 2552 | dev->hw_features |= NETIF_F_TSO |
34a48579 RR |
2553 | | NETIF_F_TSO_ECN | NETIF_F_TSO6; |
2554 | } | |
5539ae96 | 2555 | /* Individual feature bits: what can host handle? */ |
98e778c9 MM |
2556 | if (virtio_has_feature(vdev, VIRTIO_NET_F_HOST_TSO4)) |
2557 | dev->hw_features |= NETIF_F_TSO; | |
2558 | if (virtio_has_feature(vdev, VIRTIO_NET_F_HOST_TSO6)) | |
2559 | dev->hw_features |= NETIF_F_TSO6; | |
2560 | if (virtio_has_feature(vdev, VIRTIO_NET_F_HOST_ECN)) | |
2561 | dev->hw_features |= NETIF_F_TSO_ECN; | |
98e778c9 | 2562 | |
41f2f127 JW |
2563 | dev->features |= NETIF_F_GSO_ROBUST; |
2564 | ||
98e778c9 | 2565 | if (gso) |
e078de03 | 2566 | dev->features |= dev->hw_features & NETIF_F_ALL_TSO; |
98e778c9 | 2567 | /* (!csum && gso) case will be fixed by register_netdev() */ |
296f96fc | 2568 | } |
4f49129b TH |
2569 | if (virtio_has_feature(vdev, VIRTIO_NET_F_GUEST_CSUM)) |
2570 | dev->features |= NETIF_F_RXCSUM; | |
296f96fc | 2571 | |
4fda8302 JW |
2572 | dev->vlan_features = dev->features; |
2573 | ||
d0c2c997 JW |
2574 | /* MTU range: 68 - 65535 */ |
2575 | dev->min_mtu = MIN_MTU; | |
2576 | dev->max_mtu = MAX_MTU; | |
2577 | ||
296f96fc | 2578 | /* Configuration may specify what MAC to use. Otherwise random. */ |
855e0c52 RR |
2579 | if (virtio_has_feature(vdev, VIRTIO_NET_F_MAC)) |
2580 | virtio_cread_bytes(vdev, | |
2581 | offsetof(struct virtio_net_config, mac), | |
2582 | dev->dev_addr, dev->addr_len); | |
2583 | else | |
f2cedb63 | 2584 | eth_hw_addr_random(dev); |
296f96fc RR |
2585 | |
2586 | /* Set up our device-specific information */ | |
2587 | vi = netdev_priv(dev); | |
296f96fc RR |
2588 | vi->dev = dev; |
2589 | vi->vdev = vdev; | |
d9d5dcc8 | 2590 | vdev->priv = vi; |
3fa2a1df | 2591 | vi->stats = alloc_percpu(struct virtnet_stats); |
2592 | err = -ENOMEM; | |
2593 | if (vi->stats == NULL) | |
2594 | goto free; | |
2595 | ||
827da44c JS |
2596 | for_each_possible_cpu(i) { |
2597 | struct virtnet_stats *virtnet_stats; | |
2598 | virtnet_stats = per_cpu_ptr(vi->stats, i); | |
2599 | u64_stats_init(&virtnet_stats->tx_syncp); | |
2600 | u64_stats_init(&virtnet_stats->rx_syncp); | |
2601 | } | |
2602 | ||
586d17c5 | 2603 | INIT_WORK(&vi->config_work, virtnet_config_changed_work); |
296f96fc | 2604 | |
97402b96 | 2605 | /* If we can receive ANY GSO packets, we must allocate large ones. */ |
8e95a202 JP |
2606 | if (virtio_has_feature(vdev, VIRTIO_NET_F_GUEST_TSO4) || |
2607 | virtio_has_feature(vdev, VIRTIO_NET_F_GUEST_TSO6) || | |
e3e3c423 VY |
2608 | virtio_has_feature(vdev, VIRTIO_NET_F_GUEST_ECN) || |
2609 | virtio_has_feature(vdev, VIRTIO_NET_F_GUEST_UFO)) | |
97402b96 HX |
2610 | vi->big_packets = true; |
2611 | ||
3f2c31d9 MM |
2612 | if (virtio_has_feature(vdev, VIRTIO_NET_F_MRG_RXBUF)) |
2613 | vi->mergeable_rx_bufs = true; | |
2614 | ||
d04302b3 MT |
2615 | if (virtio_has_feature(vdev, VIRTIO_NET_F_MRG_RXBUF) || |
2616 | virtio_has_feature(vdev, VIRTIO_F_VERSION_1)) | |
012873d0 MT |
2617 | vi->hdr_len = sizeof(struct virtio_net_hdr_mrg_rxbuf); |
2618 | else | |
2619 | vi->hdr_len = sizeof(struct virtio_net_hdr); | |
2620 | ||
75993300 MT |
2621 | if (virtio_has_feature(vdev, VIRTIO_F_ANY_LAYOUT) || |
2622 | virtio_has_feature(vdev, VIRTIO_F_VERSION_1)) | |
e7428e95 MT |
2623 | vi->any_header_sg = true; |
2624 | ||
986a4f4d JW |
2625 | if (virtio_has_feature(vdev, VIRTIO_NET_F_CTRL_VQ)) |
2626 | vi->has_cvq = true; | |
2627 | ||
14de9d11 AC |
2628 | if (virtio_has_feature(vdev, VIRTIO_NET_F_MTU)) { |
2629 | mtu = virtio_cread16(vdev, | |
2630 | offsetof(struct virtio_net_config, | |
2631 | mtu)); | |
93a205ee | 2632 | if (mtu < dev->min_mtu) { |
fe36cbe0 MT |
2633 | /* Should never trigger: MTU was previously validated |
2634 | * in virtnet_validate. | |
2635 | */ | |
2636 | dev_err(&vdev->dev, "device MTU appears to have changed " | |
2637 | "it is now %d < %d", mtu, dev->min_mtu); | |
2638 | goto free_stats; | |
93a205ee | 2639 | } |
2e123b44 | 2640 | |
fe36cbe0 MT |
2641 | dev->mtu = mtu; |
2642 | dev->max_mtu = mtu; | |
2643 | ||
2e123b44 MT |
2644 | /* TODO: size buffers correctly in this case. */ |
2645 | if (dev->mtu > ETH_DATA_LEN) | |
2646 | vi->big_packets = true; | |
14de9d11 AC |
2647 | } |
2648 | ||
012873d0 MT |
2649 | if (vi->any_header_sg) |
2650 | dev->needed_headroom = vi->hdr_len; | |
6ebbc1a6 | 2651 | |
44900010 JW |
2652 | /* Enable multiqueue by default */ |
2653 | if (num_online_cpus() >= max_queue_pairs) | |
2654 | vi->curr_queue_pairs = max_queue_pairs; | |
2655 | else | |
2656 | vi->curr_queue_pairs = num_online_cpus(); | |
986a4f4d JW |
2657 | vi->max_queue_pairs = max_queue_pairs; |
2658 | ||
2659 | /* Allocate/initialize the rx/tx queues, and invoke find_vqs */ | |
3f9c10b0 | 2660 | err = init_vqs(vi); |
d2a7ddda | 2661 | if (err) |
9bb8ca86 | 2662 | goto free_stats; |
296f96fc | 2663 | |
fbf28d78 MD |
2664 | #ifdef CONFIG_SYSFS |
2665 | if (vi->mergeable_rx_bufs) | |
2666 | dev->sysfs_rx_queue_group = &virtio_net_mrg_rx_group; | |
2667 | #endif | |
0f13b66b ZYW |
2668 | netif_set_real_num_tx_queues(dev, vi->curr_queue_pairs); |
2669 | netif_set_real_num_rx_queues(dev, vi->curr_queue_pairs); | |
986a4f4d | 2670 | |
16032be5 NA |
2671 | virtnet_init_settings(dev); |
2672 | ||
296f96fc RR |
2673 | err = register_netdev(dev); |
2674 | if (err) { | |
2675 | pr_debug("virtio_net: registering device failed\n"); | |
d2a7ddda | 2676 | goto free_vqs; |
296f96fc | 2677 | } |
b3369c1f | 2678 | |
4baf1e33 MT |
2679 | virtio_device_ready(vdev); |
2680 | ||
8017c279 | 2681 | err = virtnet_cpu_notif_add(vi); |
8de4b2f3 WG |
2682 | if (err) { |
2683 | pr_debug("virtio_net: registering cpu notifier failed\n"); | |
f00e35e2 | 2684 | goto free_unregister_netdev; |
8de4b2f3 WG |
2685 | } |
2686 | ||
a220871b | 2687 | virtnet_set_queues(vi, vi->curr_queue_pairs); |
44900010 | 2688 | |
167c25e4 JW |
2689 | /* Assume link up if device can't report link status, |
2690 | otherwise get link status from config. */ | |
2691 | if (virtio_has_feature(vi->vdev, VIRTIO_NET_F_STATUS)) { | |
2692 | netif_carrier_off(dev); | |
3b07e9ca | 2693 | schedule_work(&vi->config_work); |
167c25e4 JW |
2694 | } else { |
2695 | vi->status = VIRTIO_NET_S_LINK_UP; | |
2696 | netif_carrier_on(dev); | |
2697 | } | |
9f4d26d0 | 2698 | |
3f93522f JW |
2699 | for (i = 0; i < ARRAY_SIZE(guest_offloads); i++) |
2700 | if (virtio_has_feature(vi->vdev, guest_offloads[i])) | |
2701 | set_bit(guest_offloads[i], &vi->guest_offloads); | |
2702 | ||
986a4f4d JW |
2703 | pr_debug("virtnet: registered device %s with %d RX and TX vq's\n", |
2704 | dev->name, max_queue_pairs); | |
2705 | ||
296f96fc RR |
2706 | return 0; |
2707 | ||
f00e35e2 | 2708 | free_unregister_netdev: |
02465555 MT |
2709 | vi->vdev->config->reset(vdev); |
2710 | ||
b3369c1f | 2711 | unregister_netdev(dev); |
d2a7ddda | 2712 | free_vqs: |
986a4f4d | 2713 | cancel_delayed_work_sync(&vi->refill); |
fb51879d | 2714 | free_receive_page_frags(vi); |
e9d7417b | 2715 | virtnet_del_vqs(vi); |
3fa2a1df | 2716 | free_stats: |
2717 | free_percpu(vi->stats); | |
296f96fc RR |
2718 | free: |
2719 | free_netdev(dev); | |
2720 | return err; | |
2721 | } | |
2722 | ||
04486ed0 | 2723 | static void remove_vq_common(struct virtnet_info *vi) |
296f96fc | 2724 | { |
04486ed0 | 2725 | vi->vdev->config->reset(vi->vdev); |
830a8a97 SM |
2726 | |
2727 | /* Free unused buffers in both send and recv, if any. */ | |
9ab86bbc | 2728 | free_unused_bufs(vi); |
fb6813f4 | 2729 | |
986a4f4d | 2730 | free_receive_bufs(vi); |
d2a7ddda | 2731 | |
fb51879d MD |
2732 | free_receive_page_frags(vi); |
2733 | ||
986a4f4d | 2734 | virtnet_del_vqs(vi); |
04486ed0 AS |
2735 | } |
2736 | ||
8cc085d6 | 2737 | static void virtnet_remove(struct virtio_device *vdev) |
04486ed0 AS |
2738 | { |
2739 | struct virtnet_info *vi = vdev->priv; | |
2740 | ||
8017c279 | 2741 | virtnet_cpu_notif_remove(vi); |
8de4b2f3 | 2742 | |
102a2786 MT |
2743 | /* Make sure no work handler is accessing the device. */ |
2744 | flush_work(&vi->config_work); | |
586d17c5 | 2745 | |
04486ed0 AS |
2746 | unregister_netdev(vi->dev); |
2747 | ||
2748 | remove_vq_common(vi); | |
fb6813f4 | 2749 | |
2e66f55b | 2750 | free_percpu(vi->stats); |
74b2553f | 2751 | free_netdev(vi->dev); |
296f96fc RR |
2752 | } |
2753 | ||
67a75194 | 2754 | static __maybe_unused int virtnet_freeze(struct virtio_device *vdev) |
0741bcb5 AS |
2755 | { |
2756 | struct virtnet_info *vi = vdev->priv; | |
2757 | ||
8017c279 | 2758 | virtnet_cpu_notif_remove(vi); |
9fe7bfce | 2759 | virtnet_freeze_down(vdev); |
0741bcb5 AS |
2760 | remove_vq_common(vi); |
2761 | ||
2762 | return 0; | |
2763 | } | |
2764 | ||
67a75194 | 2765 | static __maybe_unused int virtnet_restore(struct virtio_device *vdev) |
0741bcb5 AS |
2766 | { |
2767 | struct virtnet_info *vi = vdev->priv; | |
9fe7bfce | 2768 | int err; |
0741bcb5 | 2769 | |
9fe7bfce | 2770 | err = virtnet_restore_up(vdev); |
0741bcb5 AS |
2771 | if (err) |
2772 | return err; | |
986a4f4d JW |
2773 | virtnet_set_queues(vi, vi->curr_queue_pairs); |
2774 | ||
8017c279 | 2775 | err = virtnet_cpu_notif_add(vi); |
ec9debbd JW |
2776 | if (err) |
2777 | return err; | |
2778 | ||
0741bcb5 AS |
2779 | return 0; |
2780 | } | |
0741bcb5 | 2781 | |
296f96fc RR |
2782 | static struct virtio_device_id id_table[] = { |
2783 | { VIRTIO_ID_NET, VIRTIO_DEV_ANY_ID }, | |
2784 | { 0 }, | |
2785 | }; | |
2786 | ||
f3358507 MT |
2787 | #define VIRTNET_FEATURES \ |
2788 | VIRTIO_NET_F_CSUM, VIRTIO_NET_F_GUEST_CSUM, \ | |
2789 | VIRTIO_NET_F_MAC, \ | |
2790 | VIRTIO_NET_F_HOST_TSO4, VIRTIO_NET_F_HOST_UFO, VIRTIO_NET_F_HOST_TSO6, \ | |
2791 | VIRTIO_NET_F_HOST_ECN, VIRTIO_NET_F_GUEST_TSO4, VIRTIO_NET_F_GUEST_TSO6, \ | |
2792 | VIRTIO_NET_F_GUEST_ECN, VIRTIO_NET_F_GUEST_UFO, \ | |
2793 | VIRTIO_NET_F_MRG_RXBUF, VIRTIO_NET_F_STATUS, VIRTIO_NET_F_CTRL_VQ, \ | |
2794 | VIRTIO_NET_F_CTRL_RX, VIRTIO_NET_F_CTRL_VLAN, \ | |
2795 | VIRTIO_NET_F_GUEST_ANNOUNCE, VIRTIO_NET_F_MQ, \ | |
2796 | VIRTIO_NET_F_CTRL_MAC_ADDR, \ | |
3f93522f | 2797 | VIRTIO_NET_F_MTU, VIRTIO_NET_F_CTRL_GUEST_OFFLOADS |
f3358507 | 2798 | |
c45a6816 | 2799 | static unsigned int features[] = { |
f3358507 MT |
2800 | VIRTNET_FEATURES, |
2801 | }; | |
2802 | ||
2803 | static unsigned int features_legacy[] = { | |
2804 | VIRTNET_FEATURES, | |
2805 | VIRTIO_NET_F_GSO, | |
e7428e95 | 2806 | VIRTIO_F_ANY_LAYOUT, |
c45a6816 RR |
2807 | }; |
2808 | ||
22402529 | 2809 | static struct virtio_driver virtio_net_driver = { |
c45a6816 RR |
2810 | .feature_table = features, |
2811 | .feature_table_size = ARRAY_SIZE(features), | |
f3358507 MT |
2812 | .feature_table_legacy = features_legacy, |
2813 | .feature_table_size_legacy = ARRAY_SIZE(features_legacy), | |
296f96fc RR |
2814 | .driver.name = KBUILD_MODNAME, |
2815 | .driver.owner = THIS_MODULE, | |
2816 | .id_table = id_table, | |
fe36cbe0 | 2817 | .validate = virtnet_validate, |
296f96fc | 2818 | .probe = virtnet_probe, |
8cc085d6 | 2819 | .remove = virtnet_remove, |
9f4d26d0 | 2820 | .config_changed = virtnet_config_changed, |
89107000 | 2821 | #ifdef CONFIG_PM_SLEEP |
0741bcb5 AS |
2822 | .freeze = virtnet_freeze, |
2823 | .restore = virtnet_restore, | |
2824 | #endif | |
296f96fc RR |
2825 | }; |
2826 | ||
8017c279 SAS |
2827 | static __init int virtio_net_driver_init(void) |
2828 | { | |
2829 | int ret; | |
2830 | ||
73c1b41e | 2831 | ret = cpuhp_setup_state_multi(CPUHP_AP_ONLINE_DYN, "virtio/net:online", |
8017c279 SAS |
2832 | virtnet_cpu_online, |
2833 | virtnet_cpu_down_prep); | |
2834 | if (ret < 0) | |
2835 | goto out; | |
2836 | virtionet_online = ret; | |
73c1b41e | 2837 | ret = cpuhp_setup_state_multi(CPUHP_VIRT_NET_DEAD, "virtio/net:dead", |
8017c279 SAS |
2838 | NULL, virtnet_cpu_dead); |
2839 | if (ret) | |
2840 | goto err_dead; | |
2841 | ||
2842 | ret = register_virtio_driver(&virtio_net_driver); | |
2843 | if (ret) | |
2844 | goto err_virtio; | |
2845 | return 0; | |
2846 | err_virtio: | |
2847 | cpuhp_remove_multi_state(CPUHP_VIRT_NET_DEAD); | |
2848 | err_dead: | |
2849 | cpuhp_remove_multi_state(virtionet_online); | |
2850 | out: | |
2851 | return ret; | |
2852 | } | |
2853 | module_init(virtio_net_driver_init); | |
2854 | ||
2855 | static __exit void virtio_net_driver_exit(void) | |
2856 | { | |
cfa0ebc9 | 2857 | unregister_virtio_driver(&virtio_net_driver); |
8017c279 SAS |
2858 | cpuhp_remove_multi_state(CPUHP_VIRT_NET_DEAD); |
2859 | cpuhp_remove_multi_state(virtionet_online); | |
8017c279 SAS |
2860 | } |
2861 | module_exit(virtio_net_driver_exit); | |
296f96fc RR |
2862 | |
2863 | MODULE_DEVICE_TABLE(virtio, id_table); | |
2864 | MODULE_DESCRIPTION("Virtio network driver"); | |
2865 | MODULE_LICENSE("GPL"); |