]>
Commit | Line | Data |
---|---|---|
48925e37 | 1 | /* A network driver using virtio. |
296f96fc RR |
2 | * |
3 | * Copyright 2007 Rusty Russell <rusty@rustcorp.com.au> IBM Corporation | |
4 | * | |
5 | * This program is free software; you can redistribute it and/or modify | |
6 | * it under the terms of the GNU General Public License as published by | |
7 | * the Free Software Foundation; either version 2 of the License, or | |
8 | * (at your option) any later version. | |
9 | * | |
10 | * This program is distributed in the hope that it will be useful, | |
11 | * but WITHOUT ANY WARRANTY; without even the implied warranty of | |
12 | * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the | |
13 | * GNU General Public License for more details. | |
14 | * | |
15 | * You should have received a copy of the GNU General Public License | |
adf8d3ff | 16 | * along with this program; if not, see <http://www.gnu.org/licenses/>. |
296f96fc RR |
17 | */ |
18 | //#define DEBUG | |
19 | #include <linux/netdevice.h> | |
20 | #include <linux/etherdevice.h> | |
a9ea3fc6 | 21 | #include <linux/ethtool.h> |
296f96fc RR |
22 | #include <linux/module.h> |
23 | #include <linux/virtio.h> | |
24 | #include <linux/virtio_net.h> | |
f600b690 | 25 | #include <linux/bpf.h> |
a67edbf4 | 26 | #include <linux/bpf_trace.h> |
296f96fc | 27 | #include <linux/scatterlist.h> |
e918085a | 28 | #include <linux/if_vlan.h> |
5a0e3ad6 | 29 | #include <linux/slab.h> |
8de4b2f3 | 30 | #include <linux/cpu.h> |
ab7db917 | 31 | #include <linux/average.h> |
d85b758f | 32 | #include <net/route.h> |
296f96fc | 33 | |
d34710e3 | 34 | static int napi_weight = NAPI_POLL_WEIGHT; |
6c0cd7c0 DL |
35 | module_param(napi_weight, int, 0444); |
36 | ||
b92f1e67 | 37 | static bool csum = true, gso = true, napi_tx; |
34a48579 RR |
38 | module_param(csum, bool, 0444); |
39 | module_param(gso, bool, 0444); | |
b92f1e67 | 40 | module_param(napi_tx, bool, 0644); |
34a48579 | 41 | |
296f96fc | 42 | /* FIXME: MTU in config. */ |
5061de36 | 43 | #define GOOD_PACKET_LEN (ETH_HLEN + VLAN_HLEN + ETH_DATA_LEN) |
3f2c31d9 | 44 | #define GOOD_COPY_LEN 128 |
296f96fc | 45 | |
f6b10209 JW |
46 | #define VIRTNET_RX_PAD (NET_IP_ALIGN + NET_SKB_PAD) |
47 | ||
2de2f7f4 JF |
48 | /* Amount of XDP headroom to prepend to packets for use by xdp_adjust_head */ |
49 | #define VIRTIO_XDP_HEADROOM 256 | |
50 | ||
5377d758 JB |
51 | /* RX packet size EWMA. The average packet size is used to determine the packet |
52 | * buffer size when refilling RX rings. As the entire RX ring may be refilled | |
53 | * at once, the weight is chosen so that the EWMA will be insensitive to short- | |
54 | * term, transient changes in packet size. | |
ab7db917 | 55 | */ |
eb1e011a | 56 | DECLARE_EWMA(pkt_len, 0, 64) |
ab7db917 | 57 | |
66846048 | 58 | #define VIRTNET_DRIVER_VERSION "1.0.0" |
2a41f71d | 59 | |
3fa2a1df | 60 | struct virtnet_stats { |
83a27052 ED |
61 | struct u64_stats_sync tx_syncp; |
62 | struct u64_stats_sync rx_syncp; | |
3fa2a1df | 63 | u64 tx_bytes; |
64 | u64 tx_packets; | |
65 | ||
66 | u64 rx_bytes; | |
67 | u64 rx_packets; | |
68 | }; | |
69 | ||
e9d7417b JW |
70 | /* Internal representation of a send virtqueue */ |
71 | struct send_queue { | |
72 | /* Virtqueue associated with this send _queue */ | |
73 | struct virtqueue *vq; | |
74 | ||
75 | /* TX: fragments + linear part + virtio header */ | |
76 | struct scatterlist sg[MAX_SKB_FRAGS + 2]; | |
986a4f4d JW |
77 | |
78 | /* Name of the send queue: output.$index */ | |
79 | char name[40]; | |
b92f1e67 WB |
80 | |
81 | struct napi_struct napi; | |
e9d7417b JW |
82 | }; |
83 | ||
84 | /* Internal representation of a receive virtqueue */ | |
85 | struct receive_queue { | |
86 | /* Virtqueue associated with this receive_queue */ | |
87 | struct virtqueue *vq; | |
88 | ||
296f96fc RR |
89 | struct napi_struct napi; |
90 | ||
f600b690 JF |
91 | struct bpf_prog __rcu *xdp_prog; |
92 | ||
e9d7417b JW |
93 | /* Chain pages by the private ptr. */ |
94 | struct page *pages; | |
95 | ||
ab7db917 | 96 | /* Average packet length for mergeable receive buffers. */ |
5377d758 | 97 | struct ewma_pkt_len mrg_avg_pkt_len; |
ab7db917 | 98 | |
fb51879d MD |
99 | /* Page frag for packet buffer allocation. */ |
100 | struct page_frag alloc_frag; | |
101 | ||
e9d7417b JW |
102 | /* RX: fragments + linear part + virtio header */ |
103 | struct scatterlist sg[MAX_SKB_FRAGS + 2]; | |
986a4f4d | 104 | |
d85b758f MT |
105 | /* Min single buffer size for mergeable buffers case. */ |
106 | unsigned int min_buf_len; | |
107 | ||
986a4f4d JW |
108 | /* Name of this receive queue: input.$index */ |
109 | char name[40]; | |
e9d7417b JW |
110 | }; |
111 | ||
112 | struct virtnet_info { | |
113 | struct virtio_device *vdev; | |
114 | struct virtqueue *cvq; | |
115 | struct net_device *dev; | |
986a4f4d JW |
116 | struct send_queue *sq; |
117 | struct receive_queue *rq; | |
e9d7417b JW |
118 | unsigned int status; |
119 | ||
986a4f4d JW |
120 | /* Max # of queue pairs supported by the device */ |
121 | u16 max_queue_pairs; | |
122 | ||
123 | /* # of queue pairs currently used by the driver */ | |
124 | u16 curr_queue_pairs; | |
125 | ||
672aafd5 JF |
126 | /* # of XDP queue pairs currently used by the driver */ |
127 | u16 xdp_queue_pairs; | |
128 | ||
97402b96 HX |
129 | /* I like... big packets and I cannot lie! */ |
130 | bool big_packets; | |
131 | ||
3f2c31d9 MM |
132 | /* Host will merge rx buffers for big packets (shake it! shake it!) */ |
133 | bool mergeable_rx_bufs; | |
134 | ||
986a4f4d JW |
135 | /* Has control virtqueue */ |
136 | bool has_cvq; | |
137 | ||
e7428e95 MT |
138 | /* Host can handle any s/g split between our header and packet data */ |
139 | bool any_header_sg; | |
140 | ||
012873d0 MT |
141 | /* Packet virtio header size */ |
142 | u8 hdr_len; | |
143 | ||
3fa2a1df | 144 | /* Active statistics */ |
145 | struct virtnet_stats __percpu *stats; | |
146 | ||
3161e453 RR |
147 | /* Work struct for refilling if we run low on memory. */ |
148 | struct delayed_work refill; | |
149 | ||
586d17c5 JW |
150 | /* Work struct for config space updates */ |
151 | struct work_struct config_work; | |
152 | ||
986a4f4d JW |
153 | /* Does the affinity hint is set for virtqueues? */ |
154 | bool affinity_hint_set; | |
47be2479 | 155 | |
8017c279 SAS |
156 | /* CPU hotplug instances for online & dead */ |
157 | struct hlist_node node; | |
158 | struct hlist_node node_dead; | |
2ac46030 MT |
159 | |
160 | /* Control VQ buffers: protected by the rtnl lock */ | |
161 | struct virtio_net_ctrl_hdr ctrl_hdr; | |
162 | virtio_net_ctrl_ack ctrl_status; | |
a725ee3e | 163 | struct virtio_net_ctrl_mq ctrl_mq; |
2ac46030 MT |
164 | u8 ctrl_promisc; |
165 | u8 ctrl_allmulti; | |
a725ee3e | 166 | u16 ctrl_vid; |
16032be5 NA |
167 | |
168 | /* Ethtool settings */ | |
169 | u8 duplex; | |
170 | u32 speed; | |
296f96fc RR |
171 | }; |
172 | ||
9ab86bbc | 173 | struct padded_vnet_hdr { |
012873d0 | 174 | struct virtio_net_hdr_mrg_rxbuf hdr; |
9ab86bbc | 175 | /* |
012873d0 MT |
176 | * hdr is in a separate sg buffer, and data sg buffer shares same page |
177 | * with this header sg. This padding makes next sg 16 byte aligned | |
178 | * after the header. | |
9ab86bbc | 179 | */ |
012873d0 | 180 | char padding[4]; |
9ab86bbc SM |
181 | }; |
182 | ||
986a4f4d JW |
183 | /* Converting between virtqueue no. and kernel tx/rx queue no. |
184 | * 0:rx0 1:tx0 2:rx1 3:tx1 ... 2N:rxN 2N+1:txN 2N+2:cvq | |
185 | */ | |
186 | static int vq2txq(struct virtqueue *vq) | |
187 | { | |
9d0ca6ed | 188 | return (vq->index - 1) / 2; |
986a4f4d JW |
189 | } |
190 | ||
191 | static int txq2vq(int txq) | |
192 | { | |
193 | return txq * 2 + 1; | |
194 | } | |
195 | ||
196 | static int vq2rxq(struct virtqueue *vq) | |
197 | { | |
9d0ca6ed | 198 | return vq->index / 2; |
986a4f4d JW |
199 | } |
200 | ||
201 | static int rxq2vq(int rxq) | |
202 | { | |
203 | return rxq * 2; | |
204 | } | |
205 | ||
012873d0 | 206 | static inline struct virtio_net_hdr_mrg_rxbuf *skb_vnet_hdr(struct sk_buff *skb) |
296f96fc | 207 | { |
012873d0 | 208 | return (struct virtio_net_hdr_mrg_rxbuf *)skb->cb; |
296f96fc RR |
209 | } |
210 | ||
9ab86bbc SM |
211 | /* |
212 | * private is used to chain pages for big packets, put the whole | |
213 | * most recent used list in the beginning for reuse | |
214 | */ | |
e9d7417b | 215 | static void give_pages(struct receive_queue *rq, struct page *page) |
0a888fd1 | 216 | { |
9ab86bbc | 217 | struct page *end; |
0a888fd1 | 218 | |
e9d7417b | 219 | /* Find end of list, sew whole thing into vi->rq.pages. */ |
9ab86bbc | 220 | for (end = page; end->private; end = (struct page *)end->private); |
e9d7417b JW |
221 | end->private = (unsigned long)rq->pages; |
222 | rq->pages = page; | |
0a888fd1 MM |
223 | } |
224 | ||
e9d7417b | 225 | static struct page *get_a_page(struct receive_queue *rq, gfp_t gfp_mask) |
fb6813f4 | 226 | { |
e9d7417b | 227 | struct page *p = rq->pages; |
fb6813f4 | 228 | |
9ab86bbc | 229 | if (p) { |
e9d7417b | 230 | rq->pages = (struct page *)p->private; |
9ab86bbc SM |
231 | /* clear private here, it is used to chain pages */ |
232 | p->private = 0; | |
233 | } else | |
fb6813f4 RR |
234 | p = alloc_page(gfp_mask); |
235 | return p; | |
236 | } | |
237 | ||
e4e8452a WB |
238 | static void virtqueue_napi_schedule(struct napi_struct *napi, |
239 | struct virtqueue *vq) | |
240 | { | |
241 | if (napi_schedule_prep(napi)) { | |
242 | virtqueue_disable_cb(vq); | |
243 | __napi_schedule(napi); | |
244 | } | |
245 | } | |
246 | ||
247 | static void virtqueue_napi_complete(struct napi_struct *napi, | |
248 | struct virtqueue *vq, int processed) | |
249 | { | |
250 | int opaque; | |
251 | ||
252 | opaque = virtqueue_enable_cb_prepare(vq); | |
253 | if (napi_complete_done(napi, processed) && | |
254 | unlikely(virtqueue_poll(vq, opaque))) | |
255 | virtqueue_napi_schedule(napi, vq); | |
256 | } | |
257 | ||
e9d7417b | 258 | static void skb_xmit_done(struct virtqueue *vq) |
296f96fc | 259 | { |
e9d7417b | 260 | struct virtnet_info *vi = vq->vdev->priv; |
b92f1e67 | 261 | struct napi_struct *napi = &vi->sq[vq2txq(vq)].napi; |
296f96fc | 262 | |
2cb9c6ba | 263 | /* Suppress further interrupts. */ |
e9d7417b | 264 | virtqueue_disable_cb(vq); |
11a3a154 | 265 | |
b92f1e67 WB |
266 | if (napi->weight) |
267 | virtqueue_napi_schedule(napi, vq); | |
268 | else | |
269 | /* We were probably waiting for more output buffers. */ | |
270 | netif_wake_subqueue(vi->dev, vq2txq(vq)); | |
296f96fc RR |
271 | } |
272 | ||
3464645a | 273 | /* Called from bottom half context */ |
946fa564 MT |
274 | static struct sk_buff *page_to_skb(struct virtnet_info *vi, |
275 | struct receive_queue *rq, | |
2613af0e MD |
276 | struct page *page, unsigned int offset, |
277 | unsigned int len, unsigned int truesize) | |
9ab86bbc SM |
278 | { |
279 | struct sk_buff *skb; | |
012873d0 | 280 | struct virtio_net_hdr_mrg_rxbuf *hdr; |
2613af0e | 281 | unsigned int copy, hdr_len, hdr_padded_len; |
9ab86bbc | 282 | char *p; |
fb6813f4 | 283 | |
2613af0e | 284 | p = page_address(page) + offset; |
3f2c31d9 | 285 | |
9ab86bbc | 286 | /* copy small packet so we can reuse these pages for small data */ |
c67f5db8 | 287 | skb = napi_alloc_skb(&rq->napi, GOOD_COPY_LEN); |
9ab86bbc SM |
288 | if (unlikely(!skb)) |
289 | return NULL; | |
3f2c31d9 | 290 | |
9ab86bbc | 291 | hdr = skb_vnet_hdr(skb); |
3f2c31d9 | 292 | |
012873d0 MT |
293 | hdr_len = vi->hdr_len; |
294 | if (vi->mergeable_rx_bufs) | |
295 | hdr_padded_len = sizeof *hdr; | |
296 | else | |
2613af0e | 297 | hdr_padded_len = sizeof(struct padded_vnet_hdr); |
3f2c31d9 | 298 | |
9ab86bbc | 299 | memcpy(hdr, p, hdr_len); |
3f2c31d9 | 300 | |
9ab86bbc | 301 | len -= hdr_len; |
2613af0e MD |
302 | offset += hdr_padded_len; |
303 | p += hdr_padded_len; | |
3f2c31d9 | 304 | |
9ab86bbc SM |
305 | copy = len; |
306 | if (copy > skb_tailroom(skb)) | |
307 | copy = skb_tailroom(skb); | |
308 | memcpy(skb_put(skb, copy), p, copy); | |
3f2c31d9 | 309 | |
9ab86bbc SM |
310 | len -= copy; |
311 | offset += copy; | |
3f2c31d9 | 312 | |
2613af0e MD |
313 | if (vi->mergeable_rx_bufs) { |
314 | if (len) | |
315 | skb_add_rx_frag(skb, 0, page, offset, len, truesize); | |
316 | else | |
317 | put_page(page); | |
318 | return skb; | |
319 | } | |
320 | ||
e878d78b SL |
321 | /* |
322 | * Verify that we can indeed put this data into a skb. | |
323 | * This is here to handle cases when the device erroneously | |
324 | * tries to receive more than is possible. This is usually | |
325 | * the case of a broken device. | |
326 | */ | |
327 | if (unlikely(len > MAX_SKB_FRAGS * PAGE_SIZE)) { | |
be443899 | 328 | net_dbg_ratelimited("%s: too much data\n", skb->dev->name); |
e878d78b SL |
329 | dev_kfree_skb(skb); |
330 | return NULL; | |
331 | } | |
2613af0e | 332 | BUG_ON(offset >= PAGE_SIZE); |
9ab86bbc | 333 | while (len) { |
2613af0e MD |
334 | unsigned int frag_size = min((unsigned)PAGE_SIZE - offset, len); |
335 | skb_add_rx_frag(skb, skb_shinfo(skb)->nr_frags, page, offset, | |
336 | frag_size, truesize); | |
337 | len -= frag_size; | |
9ab86bbc SM |
338 | page = (struct page *)page->private; |
339 | offset = 0; | |
340 | } | |
3f2c31d9 | 341 | |
9ab86bbc | 342 | if (page) |
e9d7417b | 343 | give_pages(rq, page); |
3f2c31d9 | 344 | |
9ab86bbc SM |
345 | return skb; |
346 | } | |
3f2c31d9 | 347 | |
a67edbf4 | 348 | static bool virtnet_xdp_xmit(struct virtnet_info *vi, |
56434a01 | 349 | struct receive_queue *rq, |
f6b10209 | 350 | struct xdp_buff *xdp) |
56434a01 | 351 | { |
56434a01 | 352 | struct virtio_net_hdr_mrg_rxbuf *hdr; |
f6b10209 | 353 | unsigned int len; |
722d8283 JF |
354 | struct send_queue *sq; |
355 | unsigned int qp; | |
56434a01 JF |
356 | void *xdp_sent; |
357 | int err; | |
358 | ||
722d8283 JF |
359 | qp = vi->curr_queue_pairs - vi->xdp_queue_pairs + smp_processor_id(); |
360 | sq = &vi->sq[qp]; | |
361 | ||
56434a01 JF |
362 | /* Free up any pending old buffers before queueing new ones. */ |
363 | while ((xdp_sent = virtqueue_get_buf(sq->vq, &len)) != NULL) { | |
f6b10209 | 364 | struct page *sent_page = virt_to_head_page(xdp_sent); |
bb91accf | 365 | |
f6b10209 | 366 | put_page(sent_page); |
bb91accf | 367 | } |
56434a01 | 368 | |
f6b10209 JW |
369 | xdp->data -= vi->hdr_len; |
370 | /* Zero header and leave csum up to XDP layers */ | |
371 | hdr = xdp->data; | |
372 | memset(hdr, 0, vi->hdr_len); | |
bb91accf | 373 | |
f6b10209 | 374 | sg_init_one(sq->sg, xdp->data, xdp->data_end - xdp->data); |
bb91accf | 375 | |
f6b10209 | 376 | err = virtqueue_add_outbuf(sq->vq, sq->sg, 1, xdp->data, GFP_ATOMIC); |
56434a01 | 377 | if (unlikely(err)) { |
f6b10209 | 378 | struct page *page = virt_to_head_page(xdp->data); |
bb91accf | 379 | |
f6b10209 | 380 | put_page(page); |
a67edbf4 | 381 | return false; |
56434a01 JF |
382 | } |
383 | ||
384 | virtqueue_kick(sq->vq); | |
a67edbf4 | 385 | return true; |
56434a01 JF |
386 | } |
387 | ||
f6b10209 JW |
388 | static unsigned int virtnet_get_headroom(struct virtnet_info *vi) |
389 | { | |
390 | return vi->xdp_queue_pairs ? VIRTIO_XDP_HEADROOM : 0; | |
391 | } | |
392 | ||
bb91accf JW |
393 | static struct sk_buff *receive_small(struct net_device *dev, |
394 | struct virtnet_info *vi, | |
395 | struct receive_queue *rq, | |
396 | void *buf, unsigned int len) | |
f121159d | 397 | { |
f6b10209 | 398 | struct sk_buff *skb; |
bb91accf | 399 | struct bpf_prog *xdp_prog; |
f6b10209 JW |
400 | unsigned int xdp_headroom = virtnet_get_headroom(vi); |
401 | unsigned int header_offset = VIRTNET_RX_PAD + xdp_headroom; | |
402 | unsigned int headroom = vi->hdr_len + header_offset; | |
403 | unsigned int buflen = SKB_DATA_ALIGN(GOOD_PACKET_LEN + headroom) + | |
404 | SKB_DATA_ALIGN(sizeof(struct skb_shared_info)); | |
405 | unsigned int delta = 0; | |
012873d0 | 406 | len -= vi->hdr_len; |
f121159d | 407 | |
bb91accf JW |
408 | rcu_read_lock(); |
409 | xdp_prog = rcu_dereference(rq->xdp_prog); | |
410 | if (xdp_prog) { | |
f6b10209 | 411 | struct virtio_net_hdr_mrg_rxbuf *hdr = buf + header_offset; |
0354e4d1 | 412 | struct xdp_buff xdp; |
f6b10209 | 413 | void *orig_data; |
bb91accf JW |
414 | u32 act; |
415 | ||
416 | if (unlikely(hdr->hdr.gso_type || hdr->hdr.flags)) | |
417 | goto err_xdp; | |
0354e4d1 | 418 | |
f6b10209 JW |
419 | xdp.data_hard_start = buf + VIRTNET_RX_PAD + vi->hdr_len; |
420 | xdp.data = xdp.data_hard_start + xdp_headroom; | |
0354e4d1 | 421 | xdp.data_end = xdp.data + len; |
f6b10209 | 422 | orig_data = xdp.data; |
0354e4d1 JF |
423 | act = bpf_prog_run_xdp(xdp_prog, &xdp); |
424 | ||
bb91accf JW |
425 | switch (act) { |
426 | case XDP_PASS: | |
2de2f7f4 | 427 | /* Recalculate length in case bpf program changed it */ |
f6b10209 | 428 | delta = orig_data - xdp.data; |
bb91accf JW |
429 | break; |
430 | case XDP_TX: | |
f6b10209 | 431 | if (unlikely(!virtnet_xdp_xmit(vi, rq, &xdp))) |
0354e4d1 | 432 | trace_xdp_exception(vi->dev, xdp_prog, act); |
bb91accf JW |
433 | rcu_read_unlock(); |
434 | goto xdp_xmit; | |
bb91accf | 435 | default: |
0354e4d1 JF |
436 | bpf_warn_invalid_xdp_action(act); |
437 | case XDP_ABORTED: | |
438 | trace_xdp_exception(vi->dev, xdp_prog, act); | |
439 | case XDP_DROP: | |
bb91accf JW |
440 | goto err_xdp; |
441 | } | |
442 | } | |
443 | rcu_read_unlock(); | |
444 | ||
f6b10209 JW |
445 | skb = build_skb(buf, buflen); |
446 | if (!skb) { | |
447 | put_page(virt_to_head_page(buf)); | |
448 | goto err; | |
449 | } | |
450 | skb_reserve(skb, headroom - delta); | |
451 | skb_put(skb, len + delta); | |
452 | if (!delta) { | |
453 | buf += header_offset; | |
454 | memcpy(skb_vnet_hdr(skb), buf, vi->hdr_len); | |
455 | } /* keep zeroed vnet hdr since packet was changed by bpf */ | |
456 | ||
457 | err: | |
f121159d | 458 | return skb; |
bb91accf JW |
459 | |
460 | err_xdp: | |
461 | rcu_read_unlock(); | |
462 | dev->stats.rx_dropped++; | |
f6b10209 | 463 | put_page(virt_to_head_page(buf)); |
bb91accf JW |
464 | xdp_xmit: |
465 | return NULL; | |
f121159d MT |
466 | } |
467 | ||
468 | static struct sk_buff *receive_big(struct net_device *dev, | |
946fa564 | 469 | struct virtnet_info *vi, |
f121159d MT |
470 | struct receive_queue *rq, |
471 | void *buf, | |
472 | unsigned int len) | |
473 | { | |
474 | struct page *page = buf; | |
c47a43d3 | 475 | struct sk_buff *skb = page_to_skb(vi, rq, page, 0, len, PAGE_SIZE); |
f600b690 | 476 | |
f121159d MT |
477 | if (unlikely(!skb)) |
478 | goto err; | |
479 | ||
480 | return skb; | |
481 | ||
482 | err: | |
483 | dev->stats.rx_dropped++; | |
484 | give_pages(rq, page); | |
485 | return NULL; | |
486 | } | |
487 | ||
72979a6c JF |
488 | /* The conditions to enable XDP should preclude the underlying device from |
489 | * sending packets across multiple buffers (num_buf > 1). However per spec | |
490 | * it does not appear to be illegal to do so but rather just against convention. | |
491 | * So in order to avoid making a system unresponsive the packets are pushed | |
492 | * into a page and the XDP program is run. This will be extremely slow and we | |
493 | * push a warning to the user to fix this as soon as possible. Fixing this may | |
494 | * require resolving the underlying hardware to determine why multiple buffers | |
495 | * are being received or simply loading the XDP program in the ingress stack | |
496 | * after the skb is built because there is no advantage to running it here | |
497 | * anymore. | |
498 | */ | |
499 | static struct page *xdp_linearize_page(struct receive_queue *rq, | |
56a86f84 | 500 | u16 *num_buf, |
72979a6c JF |
501 | struct page *p, |
502 | int offset, | |
503 | unsigned int *len) | |
504 | { | |
505 | struct page *page = alloc_page(GFP_ATOMIC); | |
2de2f7f4 | 506 | unsigned int page_off = VIRTIO_XDP_HEADROOM; |
72979a6c JF |
507 | |
508 | if (!page) | |
509 | return NULL; | |
510 | ||
511 | memcpy(page_address(page) + page_off, page_address(p) + offset, *len); | |
512 | page_off += *len; | |
513 | ||
56a86f84 | 514 | while (--*num_buf) { |
72979a6c | 515 | unsigned int buflen; |
72979a6c JF |
516 | void *buf; |
517 | int off; | |
518 | ||
680557cf MT |
519 | buf = virtqueue_get_buf(rq->vq, &buflen); |
520 | if (unlikely(!buf)) | |
72979a6c JF |
521 | goto err_buf; |
522 | ||
56a86f84 JW |
523 | p = virt_to_head_page(buf); |
524 | off = buf - page_address(p); | |
525 | ||
72979a6c JF |
526 | /* guard against a misconfigured or uncooperative backend that |
527 | * is sending packet larger than the MTU. | |
528 | */ | |
56a86f84 JW |
529 | if ((page_off + buflen) > PAGE_SIZE) { |
530 | put_page(p); | |
72979a6c | 531 | goto err_buf; |
56a86f84 | 532 | } |
72979a6c JF |
533 | |
534 | memcpy(page_address(page) + page_off, | |
535 | page_address(p) + off, buflen); | |
536 | page_off += buflen; | |
56a86f84 | 537 | put_page(p); |
72979a6c JF |
538 | } |
539 | ||
2de2f7f4 JF |
540 | /* Headroom does not contribute to packet length */ |
541 | *len = page_off - VIRTIO_XDP_HEADROOM; | |
72979a6c JF |
542 | return page; |
543 | err_buf: | |
544 | __free_pages(page, 0); | |
545 | return NULL; | |
546 | } | |
547 | ||
8fc3b9e9 | 548 | static struct sk_buff *receive_mergeable(struct net_device *dev, |
fdd819b2 | 549 | struct virtnet_info *vi, |
8fc3b9e9 | 550 | struct receive_queue *rq, |
680557cf MT |
551 | void *buf, |
552 | void *ctx, | |
8fc3b9e9 | 553 | unsigned int len) |
9ab86bbc | 554 | { |
012873d0 MT |
555 | struct virtio_net_hdr_mrg_rxbuf *hdr = buf; |
556 | u16 num_buf = virtio16_to_cpu(vi->vdev, hdr->num_buffers); | |
8fc3b9e9 MT |
557 | struct page *page = virt_to_head_page(buf); |
558 | int offset = buf - page_address(page); | |
f600b690 JF |
559 | struct sk_buff *head_skb, *curr_skb; |
560 | struct bpf_prog *xdp_prog; | |
561 | unsigned int truesize; | |
562 | ||
56434a01 JF |
563 | head_skb = NULL; |
564 | ||
f600b690 JF |
565 | rcu_read_lock(); |
566 | xdp_prog = rcu_dereference(rq->xdp_prog); | |
567 | if (xdp_prog) { | |
72979a6c | 568 | struct page *xdp_page; |
0354e4d1 | 569 | struct xdp_buff xdp; |
0354e4d1 | 570 | void *data; |
f600b690 JF |
571 | u32 act; |
572 | ||
73b62bd0 | 573 | /* This happens when rx buffer size is underestimated */ |
f600b690 | 574 | if (unlikely(num_buf > 1)) { |
72979a6c | 575 | /* linearize data for XDP */ |
56a86f84 | 576 | xdp_page = xdp_linearize_page(rq, &num_buf, |
72979a6c JF |
577 | page, offset, &len); |
578 | if (!xdp_page) | |
579 | goto err_xdp; | |
2de2f7f4 | 580 | offset = VIRTIO_XDP_HEADROOM; |
72979a6c JF |
581 | } else { |
582 | xdp_page = page; | |
f600b690 JF |
583 | } |
584 | ||
585 | /* Transient failure which in theory could occur if | |
586 | * in-flight packets from before XDP was enabled reach | |
587 | * the receive path after XDP is loaded. In practice I | |
588 | * was not able to create this condition. | |
589 | */ | |
b00f70b0 | 590 | if (unlikely(hdr->hdr.gso_type)) |
f600b690 JF |
591 | goto err_xdp; |
592 | ||
2de2f7f4 JF |
593 | /* Allow consuming headroom but reserve enough space to push |
594 | * the descriptor on if we get an XDP_TX return code. | |
595 | */ | |
0354e4d1 | 596 | data = page_address(xdp_page) + offset; |
2de2f7f4 | 597 | xdp.data_hard_start = data - VIRTIO_XDP_HEADROOM + vi->hdr_len; |
0354e4d1 JF |
598 | xdp.data = data + vi->hdr_len; |
599 | xdp.data_end = xdp.data + (len - vi->hdr_len); | |
600 | act = bpf_prog_run_xdp(xdp_prog, &xdp); | |
601 | ||
56434a01 JF |
602 | switch (act) { |
603 | case XDP_PASS: | |
2de2f7f4 JF |
604 | /* recalculate offset to account for any header |
605 | * adjustments. Note other cases do not build an | |
606 | * skb and avoid using offset | |
607 | */ | |
608 | offset = xdp.data - | |
609 | page_address(xdp_page) - vi->hdr_len; | |
610 | ||
1830f893 JW |
611 | /* We can only create skb based on xdp_page. */ |
612 | if (unlikely(xdp_page != page)) { | |
613 | rcu_read_unlock(); | |
614 | put_page(page); | |
615 | head_skb = page_to_skb(vi, rq, xdp_page, | |
2de2f7f4 | 616 | offset, len, PAGE_SIZE); |
5c33474d | 617 | ewma_pkt_len_add(&rq->mrg_avg_pkt_len, len); |
1830f893 JW |
618 | return head_skb; |
619 | } | |
56434a01 JF |
620 | break; |
621 | case XDP_TX: | |
f6b10209 | 622 | if (unlikely(!virtnet_xdp_xmit(vi, rq, &xdp))) |
0354e4d1 | 623 | trace_xdp_exception(vi->dev, xdp_prog, act); |
5c33474d | 624 | ewma_pkt_len_add(&rq->mrg_avg_pkt_len, len); |
72979a6c JF |
625 | if (unlikely(xdp_page != page)) |
626 | goto err_xdp; | |
56434a01 JF |
627 | rcu_read_unlock(); |
628 | goto xdp_xmit; | |
56434a01 | 629 | default: |
0354e4d1 JF |
630 | bpf_warn_invalid_xdp_action(act); |
631 | case XDP_ABORTED: | |
632 | trace_xdp_exception(vi->dev, xdp_prog, act); | |
633 | case XDP_DROP: | |
72979a6c JF |
634 | if (unlikely(xdp_page != page)) |
635 | __free_pages(xdp_page, 0); | |
5c33474d | 636 | ewma_pkt_len_add(&rq->mrg_avg_pkt_len, len); |
f600b690 | 637 | goto err_xdp; |
56434a01 | 638 | } |
f600b690 JF |
639 | } |
640 | rcu_read_unlock(); | |
ab7db917 | 641 | |
680557cf | 642 | if (unlikely(len > (unsigned long)ctx)) { |
56da5fd0 | 643 | pr_debug("%s: rx error: len %u exceeds truesize %lu\n", |
680557cf MT |
644 | dev->name, len, (unsigned long)ctx); |
645 | dev->stats.rx_length_errors++; | |
646 | goto err_skb; | |
647 | } | |
648 | truesize = (unsigned long)ctx; | |
f600b690 JF |
649 | head_skb = page_to_skb(vi, rq, page, offset, len, truesize); |
650 | curr_skb = head_skb; | |
9ab86bbc | 651 | |
8fc3b9e9 MT |
652 | if (unlikely(!curr_skb)) |
653 | goto err_skb; | |
9ab86bbc | 654 | while (--num_buf) { |
8fc3b9e9 MT |
655 | int num_skb_frags; |
656 | ||
680557cf | 657 | buf = virtqueue_get_buf_ctx(rq->vq, &len, &ctx); |
ab7db917 | 658 | if (unlikely(!ctx)) { |
8fc3b9e9 | 659 | pr_debug("%s: rx error: %d buffers out of %d missing\n", |
fdd819b2 | 660 | dev->name, num_buf, |
012873d0 MT |
661 | virtio16_to_cpu(vi->vdev, |
662 | hdr->num_buffers)); | |
8fc3b9e9 MT |
663 | dev->stats.rx_length_errors++; |
664 | goto err_buf; | |
3f2c31d9 | 665 | } |
8fc3b9e9 MT |
666 | |
667 | page = virt_to_head_page(buf); | |
680557cf | 668 | if (unlikely(len > (unsigned long)ctx)) { |
56da5fd0 | 669 | pr_debug("%s: rx error: len %u exceeds truesize %lu\n", |
680557cf MT |
670 | dev->name, len, (unsigned long)ctx); |
671 | dev->stats.rx_length_errors++; | |
672 | goto err_skb; | |
673 | } | |
674 | truesize = (unsigned long)ctx; | |
8fc3b9e9 MT |
675 | |
676 | num_skb_frags = skb_shinfo(curr_skb)->nr_frags; | |
2613af0e MD |
677 | if (unlikely(num_skb_frags == MAX_SKB_FRAGS)) { |
678 | struct sk_buff *nskb = alloc_skb(0, GFP_ATOMIC); | |
8fc3b9e9 MT |
679 | |
680 | if (unlikely(!nskb)) | |
681 | goto err_skb; | |
2613af0e MD |
682 | if (curr_skb == head_skb) |
683 | skb_shinfo(curr_skb)->frag_list = nskb; | |
684 | else | |
685 | curr_skb->next = nskb; | |
686 | curr_skb = nskb; | |
687 | head_skb->truesize += nskb->truesize; | |
688 | num_skb_frags = 0; | |
689 | } | |
690 | if (curr_skb != head_skb) { | |
691 | head_skb->data_len += len; | |
692 | head_skb->len += len; | |
fb51879d | 693 | head_skb->truesize += truesize; |
2613af0e | 694 | } |
8fc3b9e9 | 695 | offset = buf - page_address(page); |
ba275241 JW |
696 | if (skb_can_coalesce(curr_skb, num_skb_frags, page, offset)) { |
697 | put_page(page); | |
698 | skb_coalesce_rx_frag(curr_skb, num_skb_frags - 1, | |
fb51879d | 699 | len, truesize); |
ba275241 JW |
700 | } else { |
701 | skb_add_rx_frag(curr_skb, num_skb_frags, page, | |
fb51879d | 702 | offset, len, truesize); |
ba275241 | 703 | } |
8fc3b9e9 MT |
704 | } |
705 | ||
5377d758 | 706 | ewma_pkt_len_add(&rq->mrg_avg_pkt_len, head_skb->len); |
8fc3b9e9 MT |
707 | return head_skb; |
708 | ||
f600b690 JF |
709 | err_xdp: |
710 | rcu_read_unlock(); | |
8fc3b9e9 MT |
711 | err_skb: |
712 | put_page(page); | |
713 | while (--num_buf) { | |
680557cf MT |
714 | buf = virtqueue_get_buf(rq->vq, &len); |
715 | if (unlikely(!buf)) { | |
8fc3b9e9 MT |
716 | pr_debug("%s: rx error: %d buffers missing\n", |
717 | dev->name, num_buf); | |
718 | dev->stats.rx_length_errors++; | |
719 | break; | |
720 | } | |
680557cf | 721 | page = virt_to_head_page(buf); |
8fc3b9e9 | 722 | put_page(page); |
9ab86bbc | 723 | } |
8fc3b9e9 MT |
724 | err_buf: |
725 | dev->stats.rx_dropped++; | |
726 | dev_kfree_skb(head_skb); | |
56434a01 | 727 | xdp_xmit: |
8fc3b9e9 | 728 | return NULL; |
9ab86bbc SM |
729 | } |
730 | ||
61845d20 | 731 | static int receive_buf(struct virtnet_info *vi, struct receive_queue *rq, |
680557cf | 732 | void *buf, unsigned int len, void **ctx) |
9ab86bbc | 733 | { |
e9d7417b | 734 | struct net_device *dev = vi->dev; |
9ab86bbc | 735 | struct sk_buff *skb; |
012873d0 | 736 | struct virtio_net_hdr_mrg_rxbuf *hdr; |
61845d20 | 737 | int ret; |
3f2c31d9 | 738 | |
bcff3162 | 739 | if (unlikely(len < vi->hdr_len + ETH_HLEN)) { |
9ab86bbc SM |
740 | pr_debug("%s: short packet %i\n", dev->name, len); |
741 | dev->stats.rx_length_errors++; | |
ab7db917 | 742 | if (vi->mergeable_rx_bufs) { |
680557cf | 743 | put_page(virt_to_head_page(buf)); |
ab7db917 | 744 | } else if (vi->big_packets) { |
98bfd23c | 745 | give_pages(rq, buf); |
ab7db917 | 746 | } else { |
f6b10209 | 747 | put_page(virt_to_head_page(buf)); |
ab7db917 | 748 | } |
61845d20 | 749 | return 0; |
9ab86bbc | 750 | } |
3f2c31d9 | 751 | |
f121159d | 752 | if (vi->mergeable_rx_bufs) |
680557cf | 753 | skb = receive_mergeable(dev, vi, rq, buf, ctx, len); |
f121159d | 754 | else if (vi->big_packets) |
946fa564 | 755 | skb = receive_big(dev, vi, rq, buf, len); |
f121159d | 756 | else |
bb91accf | 757 | skb = receive_small(dev, vi, rq, buf, len); |
f121159d MT |
758 | |
759 | if (unlikely(!skb)) | |
61845d20 | 760 | return 0; |
3f2c31d9 | 761 | |
9ab86bbc | 762 | hdr = skb_vnet_hdr(skb); |
3fa2a1df | 763 | |
61845d20 | 764 | ret = skb->len; |
296f96fc | 765 | |
e858fae2 | 766 | if (hdr->hdr.flags & VIRTIO_NET_HDR_F_DATA_VALID) |
10a8d94a | 767 | skb->ip_summed = CHECKSUM_UNNECESSARY; |
296f96fc | 768 | |
e858fae2 MR |
769 | if (virtio_net_hdr_to_skb(skb, &hdr->hdr, |
770 | virtio_is_little_endian(vi->vdev))) { | |
771 | net_warn_ratelimited("%s: bad gso: type: %u, size: %u\n", | |
772 | dev->name, hdr->hdr.gso_type, | |
773 | hdr->hdr.gso_size); | |
774 | goto frame_err; | |
296f96fc RR |
775 | } |
776 | ||
d1dc06dc MR |
777 | skb->protocol = eth_type_trans(skb, dev); |
778 | pr_debug("Receiving skb proto 0x%04x len %i type %i\n", | |
779 | ntohs(skb->protocol), skb->len, skb->pkt_type); | |
780 | ||
0fbd050a | 781 | napi_gro_receive(&rq->napi, skb); |
61845d20 | 782 | return ret; |
296f96fc RR |
783 | |
784 | frame_err: | |
785 | dev->stats.rx_frame_errors++; | |
296f96fc | 786 | dev_kfree_skb(skb); |
61845d20 | 787 | return 0; |
296f96fc RR |
788 | } |
789 | ||
946fa564 MT |
790 | static int add_recvbuf_small(struct virtnet_info *vi, struct receive_queue *rq, |
791 | gfp_t gfp) | |
296f96fc | 792 | { |
f6b10209 JW |
793 | struct page_frag *alloc_frag = &rq->alloc_frag; |
794 | char *buf; | |
2de2f7f4 | 795 | unsigned int xdp_headroom = virtnet_get_headroom(vi); |
f6b10209 | 796 | int len = vi->hdr_len + VIRTNET_RX_PAD + GOOD_PACKET_LEN + xdp_headroom; |
9ab86bbc | 797 | int err; |
3f2c31d9 | 798 | |
f6b10209 JW |
799 | len = SKB_DATA_ALIGN(len) + |
800 | SKB_DATA_ALIGN(sizeof(struct skb_shared_info)); | |
801 | if (unlikely(!skb_page_frag_refill(len, alloc_frag, gfp))) | |
9ab86bbc | 802 | return -ENOMEM; |
296f96fc | 803 | |
f6b10209 JW |
804 | buf = (char *)page_address(alloc_frag->page) + alloc_frag->offset; |
805 | get_page(alloc_frag->page); | |
806 | alloc_frag->offset += len; | |
807 | sg_init_one(rq->sg, buf + VIRTNET_RX_PAD + xdp_headroom, | |
808 | vi->hdr_len + GOOD_PACKET_LEN); | |
809 | err = virtqueue_add_inbuf(rq->vq, rq->sg, 1, buf, gfp); | |
9ab86bbc | 810 | if (err < 0) |
f6b10209 | 811 | put_page(virt_to_head_page(buf)); |
97402b96 | 812 | |
9ab86bbc SM |
813 | return err; |
814 | } | |
97402b96 | 815 | |
012873d0 MT |
816 | static int add_recvbuf_big(struct virtnet_info *vi, struct receive_queue *rq, |
817 | gfp_t gfp) | |
9ab86bbc | 818 | { |
9ab86bbc SM |
819 | struct page *first, *list = NULL; |
820 | char *p; | |
821 | int i, err, offset; | |
822 | ||
a5835440 RR |
823 | sg_init_table(rq->sg, MAX_SKB_FRAGS + 2); |
824 | ||
e9d7417b | 825 | /* page in rq->sg[MAX_SKB_FRAGS + 1] is list tail */ |
9ab86bbc | 826 | for (i = MAX_SKB_FRAGS + 1; i > 1; --i) { |
e9d7417b | 827 | first = get_a_page(rq, gfp); |
9ab86bbc SM |
828 | if (!first) { |
829 | if (list) | |
e9d7417b | 830 | give_pages(rq, list); |
9ab86bbc | 831 | return -ENOMEM; |
97402b96 | 832 | } |
e9d7417b | 833 | sg_set_buf(&rq->sg[i], page_address(first), PAGE_SIZE); |
97402b96 | 834 | |
9ab86bbc SM |
835 | /* chain new page in list head to match sg */ |
836 | first->private = (unsigned long)list; | |
837 | list = first; | |
838 | } | |
296f96fc | 839 | |
e9d7417b | 840 | first = get_a_page(rq, gfp); |
9ab86bbc | 841 | if (!first) { |
e9d7417b | 842 | give_pages(rq, list); |
9ab86bbc SM |
843 | return -ENOMEM; |
844 | } | |
845 | p = page_address(first); | |
846 | ||
e9d7417b | 847 | /* rq->sg[0], rq->sg[1] share the same page */ |
012873d0 MT |
848 | /* a separated rq->sg[0] for header - required in case !any_header_sg */ |
849 | sg_set_buf(&rq->sg[0], p, vi->hdr_len); | |
9ab86bbc | 850 | |
e9d7417b | 851 | /* rq->sg[1] for data packet, from offset */ |
9ab86bbc | 852 | offset = sizeof(struct padded_vnet_hdr); |
e9d7417b | 853 | sg_set_buf(&rq->sg[1], p + offset, PAGE_SIZE - offset); |
9ab86bbc SM |
854 | |
855 | /* chain first in list head */ | |
856 | first->private = (unsigned long)list; | |
9dc7b9e4 RR |
857 | err = virtqueue_add_inbuf(rq->vq, rq->sg, MAX_SKB_FRAGS + 2, |
858 | first, gfp); | |
9ab86bbc | 859 | if (err < 0) |
e9d7417b | 860 | give_pages(rq, first); |
9ab86bbc SM |
861 | |
862 | return err; | |
296f96fc RR |
863 | } |
864 | ||
d85b758f MT |
865 | static unsigned int get_mergeable_buf_len(struct receive_queue *rq, |
866 | struct ewma_pkt_len *avg_pkt_len) | |
3f2c31d9 | 867 | { |
ab7db917 | 868 | const size_t hdr_len = sizeof(struct virtio_net_hdr_mrg_rxbuf); |
fbf28d78 MD |
869 | unsigned int len; |
870 | ||
5377d758 | 871 | len = hdr_len + clamp_t(unsigned int, ewma_pkt_len_read(avg_pkt_len), |
f0c3192c | 872 | rq->min_buf_len, PAGE_SIZE - hdr_len); |
e377fcc8 | 873 | return ALIGN(len, L1_CACHE_BYTES); |
fbf28d78 MD |
874 | } |
875 | ||
2de2f7f4 JF |
876 | static int add_recvbuf_mergeable(struct virtnet_info *vi, |
877 | struct receive_queue *rq, gfp_t gfp) | |
fbf28d78 | 878 | { |
fb51879d | 879 | struct page_frag *alloc_frag = &rq->alloc_frag; |
2de2f7f4 | 880 | unsigned int headroom = virtnet_get_headroom(vi); |
fb51879d | 881 | char *buf; |
680557cf | 882 | void *ctx; |
3f2c31d9 | 883 | int err; |
fb51879d | 884 | unsigned int len, hole; |
3f2c31d9 | 885 | |
d85b758f | 886 | len = get_mergeable_buf_len(rq, &rq->mrg_avg_pkt_len); |
2de2f7f4 | 887 | if (unlikely(!skb_page_frag_refill(len + headroom, alloc_frag, gfp))) |
9ab86bbc | 888 | return -ENOMEM; |
ab7db917 | 889 | |
fb51879d | 890 | buf = (char *)page_address(alloc_frag->page) + alloc_frag->offset; |
2de2f7f4 | 891 | buf += headroom; /* advance address leaving hole at front of pkt */ |
680557cf | 892 | ctx = (void *)(unsigned long)len; |
fb51879d | 893 | get_page(alloc_frag->page); |
2de2f7f4 | 894 | alloc_frag->offset += len + headroom; |
fb51879d | 895 | hole = alloc_frag->size - alloc_frag->offset; |
2de2f7f4 | 896 | if (hole < len + headroom) { |
ab7db917 MD |
897 | /* To avoid internal fragmentation, if there is very likely not |
898 | * enough space for another buffer, add the remaining space to | |
899 | * the current buffer. This extra space is not included in | |
900 | * the truesize stored in ctx. | |
901 | */ | |
fb51879d MD |
902 | len += hole; |
903 | alloc_frag->offset += hole; | |
904 | } | |
3f2c31d9 | 905 | |
fb51879d | 906 | sg_init_one(rq->sg, buf, len); |
680557cf | 907 | err = virtqueue_add_inbuf_ctx(rq->vq, rq->sg, 1, buf, ctx, gfp); |
9ab86bbc | 908 | if (err < 0) |
2613af0e | 909 | put_page(virt_to_head_page(buf)); |
3f2c31d9 | 910 | |
9ab86bbc SM |
911 | return err; |
912 | } | |
3f2c31d9 | 913 | |
b2baed69 RR |
914 | /* |
915 | * Returns false if we couldn't fill entirely (OOM). | |
916 | * | |
917 | * Normally run in the receive path, but can also be run from ndo_open | |
918 | * before we're receiving packets, or from refill_work which is | |
919 | * careful to disable receiving (using napi_disable). | |
920 | */ | |
946fa564 MT |
921 | static bool try_fill_recv(struct virtnet_info *vi, struct receive_queue *rq, |
922 | gfp_t gfp) | |
9ab86bbc SM |
923 | { |
924 | int err; | |
1788f495 | 925 | bool oom; |
3f2c31d9 | 926 | |
fb51879d | 927 | gfp |= __GFP_COLD; |
9ab86bbc SM |
928 | do { |
929 | if (vi->mergeable_rx_bufs) | |
2de2f7f4 | 930 | err = add_recvbuf_mergeable(vi, rq, gfp); |
9ab86bbc | 931 | else if (vi->big_packets) |
012873d0 | 932 | err = add_recvbuf_big(vi, rq, gfp); |
9ab86bbc | 933 | else |
946fa564 | 934 | err = add_recvbuf_small(vi, rq, gfp); |
3f2c31d9 | 935 | |
1788f495 | 936 | oom = err == -ENOMEM; |
9ed4cb07 | 937 | if (err) |
3f2c31d9 | 938 | break; |
b7dfde95 | 939 | } while (rq->vq->num_free); |
681daee2 | 940 | virtqueue_kick(rq->vq); |
3161e453 | 941 | return !oom; |
3f2c31d9 MM |
942 | } |
943 | ||
18445c4d | 944 | static void skb_recv_done(struct virtqueue *rvq) |
296f96fc RR |
945 | { |
946 | struct virtnet_info *vi = rvq->vdev->priv; | |
986a4f4d | 947 | struct receive_queue *rq = &vi->rq[vq2rxq(rvq)]; |
e9d7417b | 948 | |
e4e8452a | 949 | virtqueue_napi_schedule(&rq->napi, rvq); |
296f96fc RR |
950 | } |
951 | ||
e4e8452a | 952 | static void virtnet_napi_enable(struct virtqueue *vq, struct napi_struct *napi) |
3e9d08ec | 953 | { |
e4e8452a | 954 | napi_enable(napi); |
3e9d08ec BR |
955 | |
956 | /* If all buffers were filled by other side before we napi_enabled, we | |
e4e8452a WB |
957 | * won't get another interrupt, so process any outstanding packets now. |
958 | * Call local_bh_enable after to trigger softIRQ processing. | |
959 | */ | |
960 | local_bh_disable(); | |
961 | virtqueue_napi_schedule(napi, vq); | |
962 | local_bh_enable(); | |
3e9d08ec BR |
963 | } |
964 | ||
b92f1e67 WB |
965 | static void virtnet_napi_tx_enable(struct virtnet_info *vi, |
966 | struct virtqueue *vq, | |
967 | struct napi_struct *napi) | |
968 | { | |
969 | if (!napi->weight) | |
970 | return; | |
971 | ||
972 | /* Tx napi touches cachelines on the cpu handling tx interrupts. Only | |
973 | * enable the feature if this is likely affine with the transmit path. | |
974 | */ | |
975 | if (!vi->affinity_hint_set) { | |
976 | napi->weight = 0; | |
977 | return; | |
978 | } | |
979 | ||
980 | return virtnet_napi_enable(vq, napi); | |
981 | } | |
982 | ||
78a57b48 WB |
983 | static void virtnet_napi_tx_disable(struct napi_struct *napi) |
984 | { | |
985 | if (napi->weight) | |
986 | napi_disable(napi); | |
987 | } | |
988 | ||
3161e453 RR |
989 | static void refill_work(struct work_struct *work) |
990 | { | |
e9d7417b JW |
991 | struct virtnet_info *vi = |
992 | container_of(work, struct virtnet_info, refill.work); | |
3161e453 | 993 | bool still_empty; |
986a4f4d JW |
994 | int i; |
995 | ||
55257d72 | 996 | for (i = 0; i < vi->curr_queue_pairs; i++) { |
986a4f4d | 997 | struct receive_queue *rq = &vi->rq[i]; |
3161e453 | 998 | |
986a4f4d | 999 | napi_disable(&rq->napi); |
946fa564 | 1000 | still_empty = !try_fill_recv(vi, rq, GFP_KERNEL); |
e4e8452a | 1001 | virtnet_napi_enable(rq->vq, &rq->napi); |
3161e453 | 1002 | |
986a4f4d JW |
1003 | /* In theory, this can happen: if we don't get any buffers in |
1004 | * we will *never* try to fill again. | |
1005 | */ | |
1006 | if (still_empty) | |
1007 | schedule_delayed_work(&vi->refill, HZ/2); | |
1008 | } | |
3161e453 RR |
1009 | } |
1010 | ||
2ffa7598 | 1011 | static int virtnet_receive(struct receive_queue *rq, int budget) |
296f96fc | 1012 | { |
e9d7417b | 1013 | struct virtnet_info *vi = rq->vq->vdev->priv; |
61845d20 | 1014 | unsigned int len, received = 0, bytes = 0; |
9ab86bbc | 1015 | void *buf; |
61845d20 | 1016 | struct virtnet_stats *stats = this_cpu_ptr(vi->stats); |
296f96fc | 1017 | |
680557cf MT |
1018 | if (vi->mergeable_rx_bufs) { |
1019 | void *ctx; | |
1020 | ||
1021 | while (received < budget && | |
1022 | (buf = virtqueue_get_buf_ctx(rq->vq, &len, &ctx))) { | |
1023 | bytes += receive_buf(vi, rq, buf, len, ctx); | |
1024 | received++; | |
1025 | } | |
1026 | } else { | |
1027 | while (received < budget && | |
1028 | (buf = virtqueue_get_buf(rq->vq, &len)) != NULL) { | |
1029 | bytes += receive_buf(vi, rq, buf, len, NULL); | |
1030 | received++; | |
1031 | } | |
296f96fc RR |
1032 | } |
1033 | ||
be121f46 | 1034 | if (rq->vq->num_free > virtqueue_get_vring_size(rq->vq) / 2) { |
946fa564 | 1035 | if (!try_fill_recv(vi, rq, GFP_ATOMIC)) |
3b07e9ca | 1036 | schedule_delayed_work(&vi->refill, 0); |
3161e453 | 1037 | } |
296f96fc | 1038 | |
61845d20 JW |
1039 | u64_stats_update_begin(&stats->rx_syncp); |
1040 | stats->rx_bytes += bytes; | |
1041 | stats->rx_packets += received; | |
1042 | u64_stats_update_end(&stats->rx_syncp); | |
1043 | ||
2ffa7598 JW |
1044 | return received; |
1045 | } | |
1046 | ||
ea7735d9 WB |
1047 | static void free_old_xmit_skbs(struct send_queue *sq) |
1048 | { | |
1049 | struct sk_buff *skb; | |
1050 | unsigned int len; | |
1051 | struct virtnet_info *vi = sq->vq->vdev->priv; | |
1052 | struct virtnet_stats *stats = this_cpu_ptr(vi->stats); | |
1053 | unsigned int packets = 0; | |
1054 | unsigned int bytes = 0; | |
1055 | ||
1056 | while ((skb = virtqueue_get_buf(sq->vq, &len)) != NULL) { | |
1057 | pr_debug("Sent skb %p\n", skb); | |
1058 | ||
1059 | bytes += skb->len; | |
1060 | packets++; | |
1061 | ||
1062 | dev_kfree_skb_any(skb); | |
1063 | } | |
1064 | ||
1065 | /* Avoid overhead when no packets have been processed | |
1066 | * happens when called speculatively from start_xmit. | |
1067 | */ | |
1068 | if (!packets) | |
1069 | return; | |
1070 | ||
1071 | u64_stats_update_begin(&stats->tx_syncp); | |
1072 | stats->tx_bytes += bytes; | |
1073 | stats->tx_packets += packets; | |
1074 | u64_stats_update_end(&stats->tx_syncp); | |
1075 | } | |
1076 | ||
7b0411ef WB |
1077 | static void virtnet_poll_cleantx(struct receive_queue *rq) |
1078 | { | |
1079 | struct virtnet_info *vi = rq->vq->vdev->priv; | |
1080 | unsigned int index = vq2rxq(rq->vq); | |
1081 | struct send_queue *sq = &vi->sq[index]; | |
1082 | struct netdev_queue *txq = netdev_get_tx_queue(vi->dev, index); | |
1083 | ||
1084 | if (!sq->napi.weight) | |
1085 | return; | |
1086 | ||
1087 | if (__netif_tx_trylock(txq)) { | |
1088 | free_old_xmit_skbs(sq); | |
1089 | __netif_tx_unlock(txq); | |
1090 | } | |
1091 | ||
1092 | if (sq->vq->num_free >= 2 + MAX_SKB_FRAGS) | |
1093 | netif_tx_wake_queue(txq); | |
1094 | } | |
1095 | ||
2ffa7598 JW |
1096 | static int virtnet_poll(struct napi_struct *napi, int budget) |
1097 | { | |
1098 | struct receive_queue *rq = | |
1099 | container_of(napi, struct receive_queue, napi); | |
e4e8452a | 1100 | unsigned int received; |
2ffa7598 | 1101 | |
7b0411ef WB |
1102 | virtnet_poll_cleantx(rq); |
1103 | ||
faadb05f | 1104 | received = virtnet_receive(rq, budget); |
2ffa7598 | 1105 | |
8329d98e | 1106 | /* Out of packets? */ |
e4e8452a WB |
1107 | if (received < budget) |
1108 | virtqueue_napi_complete(napi, rq->vq, received); | |
296f96fc RR |
1109 | |
1110 | return received; | |
1111 | } | |
1112 | ||
986a4f4d JW |
1113 | static int virtnet_open(struct net_device *dev) |
1114 | { | |
1115 | struct virtnet_info *vi = netdev_priv(dev); | |
1116 | int i; | |
1117 | ||
e4166625 JW |
1118 | for (i = 0; i < vi->max_queue_pairs; i++) { |
1119 | if (i < vi->curr_queue_pairs) | |
1120 | /* Make sure we have some buffers: if oom use wq. */ | |
946fa564 | 1121 | if (!try_fill_recv(vi, &vi->rq[i], GFP_KERNEL)) |
e4166625 | 1122 | schedule_delayed_work(&vi->refill, 0); |
e4e8452a | 1123 | virtnet_napi_enable(vi->rq[i].vq, &vi->rq[i].napi); |
b92f1e67 | 1124 | virtnet_napi_tx_enable(vi, vi->sq[i].vq, &vi->sq[i].napi); |
986a4f4d JW |
1125 | } |
1126 | ||
1127 | return 0; | |
1128 | } | |
1129 | ||
b92f1e67 WB |
1130 | static int virtnet_poll_tx(struct napi_struct *napi, int budget) |
1131 | { | |
1132 | struct send_queue *sq = container_of(napi, struct send_queue, napi); | |
1133 | struct virtnet_info *vi = sq->vq->vdev->priv; | |
1134 | struct netdev_queue *txq = netdev_get_tx_queue(vi->dev, vq2txq(sq->vq)); | |
1135 | ||
1136 | __netif_tx_lock(txq, raw_smp_processor_id()); | |
1137 | free_old_xmit_skbs(sq); | |
1138 | __netif_tx_unlock(txq); | |
1139 | ||
1140 | virtqueue_napi_complete(napi, sq->vq, 0); | |
1141 | ||
1142 | if (sq->vq->num_free >= 2 + MAX_SKB_FRAGS) | |
1143 | netif_tx_wake_queue(txq); | |
1144 | ||
1145 | return 0; | |
1146 | } | |
1147 | ||
e9d7417b | 1148 | static int xmit_skb(struct send_queue *sq, struct sk_buff *skb) |
296f96fc | 1149 | { |
012873d0 | 1150 | struct virtio_net_hdr_mrg_rxbuf *hdr; |
296f96fc | 1151 | const unsigned char *dest = ((struct ethhdr *)skb->data)->h_dest; |
e9d7417b | 1152 | struct virtnet_info *vi = sq->vq->vdev->priv; |
7bedc7dc | 1153 | unsigned num_sg; |
012873d0 | 1154 | unsigned hdr_len = vi->hdr_len; |
e7428e95 | 1155 | bool can_push; |
296f96fc | 1156 | |
e174961c | 1157 | pr_debug("%s: xmit %p %pM\n", vi->dev->name, skb, dest); |
e7428e95 MT |
1158 | |
1159 | can_push = vi->any_header_sg && | |
1160 | !((unsigned long)skb->data & (__alignof__(*hdr) - 1)) && | |
1161 | !skb_header_cloned(skb) && skb_headroom(skb) >= hdr_len; | |
1162 | /* Even if we can, don't push here yet as this would skew | |
1163 | * csum_start offset below. */ | |
1164 | if (can_push) | |
012873d0 | 1165 | hdr = (struct virtio_net_hdr_mrg_rxbuf *)(skb->data - hdr_len); |
e7428e95 MT |
1166 | else |
1167 | hdr = skb_vnet_hdr(skb); | |
296f96fc | 1168 | |
e858fae2 | 1169 | if (virtio_net_hdr_from_skb(skb, &hdr->hdr, |
6391a448 | 1170 | virtio_is_little_endian(vi->vdev), false)) |
e858fae2 | 1171 | BUG(); |
296f96fc | 1172 | |
3f2c31d9 | 1173 | if (vi->mergeable_rx_bufs) |
012873d0 | 1174 | hdr->num_buffers = 0; |
3f2c31d9 | 1175 | |
547c890c | 1176 | sg_init_table(sq->sg, skb_shinfo(skb)->nr_frags + (can_push ? 1 : 2)); |
e7428e95 MT |
1177 | if (can_push) { |
1178 | __skb_push(skb, hdr_len); | |
1179 | num_sg = skb_to_sgvec(skb, sq->sg, 0, skb->len); | |
1180 | /* Pull header back to avoid skew in tx bytes calculations. */ | |
1181 | __skb_pull(skb, hdr_len); | |
1182 | } else { | |
1183 | sg_set_buf(sq->sg, hdr, hdr_len); | |
1184 | num_sg = skb_to_sgvec(skb, sq->sg + 1, 0, skb->len) + 1; | |
1185 | } | |
9dc7b9e4 | 1186 | return virtqueue_add_outbuf(sq->vq, sq->sg, num_sg, skb, GFP_ATOMIC); |
11a3a154 RR |
1187 | } |
1188 | ||
424efe9c | 1189 | static netdev_tx_t start_xmit(struct sk_buff *skb, struct net_device *dev) |
99ffc696 RR |
1190 | { |
1191 | struct virtnet_info *vi = netdev_priv(dev); | |
986a4f4d JW |
1192 | int qnum = skb_get_queue_mapping(skb); |
1193 | struct send_queue *sq = &vi->sq[qnum]; | |
9ed4cb07 | 1194 | int err; |
4b7fd2e6 MT |
1195 | struct netdev_queue *txq = netdev_get_tx_queue(dev, qnum); |
1196 | bool kick = !skb->xmit_more; | |
b92f1e67 | 1197 | bool use_napi = sq->napi.weight; |
2cb9c6ba | 1198 | |
2cb9c6ba | 1199 | /* Free up any pending old buffers before queueing new ones. */ |
e9d7417b | 1200 | free_old_xmit_skbs(sq); |
99ffc696 | 1201 | |
bdb12e0d WB |
1202 | if (use_napi && kick) |
1203 | virtqueue_enable_cb_delayed(sq->vq); | |
1204 | ||
074c3582 JK |
1205 | /* timestamp packet in software */ |
1206 | skb_tx_timestamp(skb); | |
1207 | ||
03f191ba | 1208 | /* Try to transmit */ |
b7dfde95 | 1209 | err = xmit_skb(sq, skb); |
48925e37 | 1210 | |
9ed4cb07 | 1211 | /* This should not happen! */ |
681daee2 | 1212 | if (unlikely(err)) { |
9ed4cb07 RR |
1213 | dev->stats.tx_fifo_errors++; |
1214 | if (net_ratelimit()) | |
1215 | dev_warn(&dev->dev, | |
b7dfde95 | 1216 | "Unexpected TXQ (%d) queue failure: %d\n", qnum, err); |
58eba97d | 1217 | dev->stats.tx_dropped++; |
85e94525 | 1218 | dev_kfree_skb_any(skb); |
58eba97d | 1219 | return NETDEV_TX_OK; |
296f96fc | 1220 | } |
03f191ba | 1221 | |
48925e37 | 1222 | /* Don't wait up for transmitted skbs to be freed. */ |
b92f1e67 WB |
1223 | if (!use_napi) { |
1224 | skb_orphan(skb); | |
1225 | nf_reset(skb); | |
1226 | } | |
48925e37 | 1227 | |
60302ff6 MT |
1228 | /* If running out of space, stop queue to avoid getting packets that we |
1229 | * are then unable to transmit. | |
1230 | * An alternative would be to force queuing layer to requeue the skb by | |
1231 | * returning NETDEV_TX_BUSY. However, NETDEV_TX_BUSY should not be | |
1232 | * returned in a normal path of operation: it means that driver is not | |
1233 | * maintaining the TX queue stop/start state properly, and causes | |
1234 | * the stack to do a non-trivial amount of useless work. | |
1235 | * Since most packets only take 1 or 2 ring slots, stopping the queue | |
1236 | * early means 16 slots are typically wasted. | |
d631b94e | 1237 | */ |
b7dfde95 | 1238 | if (sq->vq->num_free < 2+MAX_SKB_FRAGS) { |
986a4f4d | 1239 | netif_stop_subqueue(dev, qnum); |
b92f1e67 WB |
1240 | if (!use_napi && |
1241 | unlikely(!virtqueue_enable_cb_delayed(sq->vq))) { | |
48925e37 | 1242 | /* More just got used, free them then recheck. */ |
b7dfde95 LT |
1243 | free_old_xmit_skbs(sq); |
1244 | if (sq->vq->num_free >= 2+MAX_SKB_FRAGS) { | |
986a4f4d | 1245 | netif_start_subqueue(dev, qnum); |
e9d7417b | 1246 | virtqueue_disable_cb(sq->vq); |
48925e37 RR |
1247 | } |
1248 | } | |
99ffc696 | 1249 | } |
48925e37 | 1250 | |
4b7fd2e6 | 1251 | if (kick || netif_xmit_stopped(txq)) |
0b725a2c | 1252 | virtqueue_kick(sq->vq); |
296f96fc | 1253 | |
0b725a2c | 1254 | return NETDEV_TX_OK; |
c223a078 DM |
1255 | } |
1256 | ||
40cbfc37 AK |
1257 | /* |
1258 | * Send command via the control virtqueue and check status. Commands | |
1259 | * supported by the hypervisor, as indicated by feature bits, should | |
788a8b6d | 1260 | * never fail unless improperly formatted. |
40cbfc37 AK |
1261 | */ |
1262 | static bool virtnet_send_command(struct virtnet_info *vi, u8 class, u8 cmd, | |
d24bae32 | 1263 | struct scatterlist *out) |
40cbfc37 | 1264 | { |
f7bc9594 | 1265 | struct scatterlist *sgs[4], hdr, stat; |
d24bae32 | 1266 | unsigned out_num = 0, tmp; |
40cbfc37 AK |
1267 | |
1268 | /* Caller should know better */ | |
f7bc9594 | 1269 | BUG_ON(!virtio_has_feature(vi->vdev, VIRTIO_NET_F_CTRL_VQ)); |
40cbfc37 | 1270 | |
2ac46030 MT |
1271 | vi->ctrl_status = ~0; |
1272 | vi->ctrl_hdr.class = class; | |
1273 | vi->ctrl_hdr.cmd = cmd; | |
f7bc9594 | 1274 | /* Add header */ |
2ac46030 | 1275 | sg_init_one(&hdr, &vi->ctrl_hdr, sizeof(vi->ctrl_hdr)); |
f7bc9594 | 1276 | sgs[out_num++] = &hdr; |
40cbfc37 | 1277 | |
f7bc9594 RR |
1278 | if (out) |
1279 | sgs[out_num++] = out; | |
40cbfc37 | 1280 | |
f7bc9594 | 1281 | /* Add return status. */ |
2ac46030 | 1282 | sg_init_one(&stat, &vi->ctrl_status, sizeof(vi->ctrl_status)); |
d24bae32 | 1283 | sgs[out_num] = &stat; |
40cbfc37 | 1284 | |
d24bae32 | 1285 | BUG_ON(out_num + 1 > ARRAY_SIZE(sgs)); |
a7c58146 | 1286 | virtqueue_add_sgs(vi->cvq, sgs, out_num, 1, vi, GFP_ATOMIC); |
40cbfc37 | 1287 | |
67975901 | 1288 | if (unlikely(!virtqueue_kick(vi->cvq))) |
2ac46030 | 1289 | return vi->ctrl_status == VIRTIO_NET_OK; |
40cbfc37 AK |
1290 | |
1291 | /* Spin for a response, the kick causes an ioport write, trapping | |
1292 | * into the hypervisor, so the request should be handled immediately. | |
1293 | */ | |
047b9b94 HG |
1294 | while (!virtqueue_get_buf(vi->cvq, &tmp) && |
1295 | !virtqueue_is_broken(vi->cvq)) | |
40cbfc37 AK |
1296 | cpu_relax(); |
1297 | ||
2ac46030 | 1298 | return vi->ctrl_status == VIRTIO_NET_OK; |
40cbfc37 AK |
1299 | } |
1300 | ||
9c46f6d4 AW |
1301 | static int virtnet_set_mac_address(struct net_device *dev, void *p) |
1302 | { | |
1303 | struct virtnet_info *vi = netdev_priv(dev); | |
1304 | struct virtio_device *vdev = vi->vdev; | |
f2f2c8b4 | 1305 | int ret; |
e37e2ff3 | 1306 | struct sockaddr *addr; |
7e58d5ae | 1307 | struct scatterlist sg; |
9c46f6d4 | 1308 | |
801822d1 | 1309 | addr = kmemdup(p, sizeof(*addr), GFP_KERNEL); |
e37e2ff3 AL |
1310 | if (!addr) |
1311 | return -ENOMEM; | |
e37e2ff3 AL |
1312 | |
1313 | ret = eth_prepare_mac_addr_change(dev, addr); | |
f2f2c8b4 | 1314 | if (ret) |
e37e2ff3 | 1315 | goto out; |
9c46f6d4 | 1316 | |
7e58d5ae AK |
1317 | if (virtio_has_feature(vdev, VIRTIO_NET_F_CTRL_MAC_ADDR)) { |
1318 | sg_init_one(&sg, addr->sa_data, dev->addr_len); | |
1319 | if (!virtnet_send_command(vi, VIRTIO_NET_CTRL_MAC, | |
d24bae32 | 1320 | VIRTIO_NET_CTRL_MAC_ADDR_SET, &sg)) { |
7e58d5ae AK |
1321 | dev_warn(&vdev->dev, |
1322 | "Failed to set mac address by vq command.\n"); | |
e37e2ff3 AL |
1323 | ret = -EINVAL; |
1324 | goto out; | |
7e58d5ae | 1325 | } |
7e93a02f MT |
1326 | } else if (virtio_has_feature(vdev, VIRTIO_NET_F_MAC) && |
1327 | !virtio_has_feature(vdev, VIRTIO_F_VERSION_1)) { | |
855e0c52 RR |
1328 | unsigned int i; |
1329 | ||
1330 | /* Naturally, this has an atomicity problem. */ | |
1331 | for (i = 0; i < dev->addr_len; i++) | |
1332 | virtio_cwrite8(vdev, | |
1333 | offsetof(struct virtio_net_config, mac) + | |
1334 | i, addr->sa_data[i]); | |
7e58d5ae AK |
1335 | } |
1336 | ||
1337 | eth_commit_mac_addr_change(dev, p); | |
e37e2ff3 | 1338 | ret = 0; |
9c46f6d4 | 1339 | |
e37e2ff3 AL |
1340 | out: |
1341 | kfree(addr); | |
1342 | return ret; | |
9c46f6d4 AW |
1343 | } |
1344 | ||
bc1f4470 | 1345 | static void virtnet_stats(struct net_device *dev, |
1346 | struct rtnl_link_stats64 *tot) | |
3fa2a1df | 1347 | { |
1348 | struct virtnet_info *vi = netdev_priv(dev); | |
1349 | int cpu; | |
1350 | unsigned int start; | |
1351 | ||
1352 | for_each_possible_cpu(cpu) { | |
58472a76 | 1353 | struct virtnet_stats *stats = per_cpu_ptr(vi->stats, cpu); |
3fa2a1df | 1354 | u64 tpackets, tbytes, rpackets, rbytes; |
1355 | ||
1356 | do { | |
57a7744e | 1357 | start = u64_stats_fetch_begin_irq(&stats->tx_syncp); |
3fa2a1df | 1358 | tpackets = stats->tx_packets; |
1359 | tbytes = stats->tx_bytes; | |
57a7744e | 1360 | } while (u64_stats_fetch_retry_irq(&stats->tx_syncp, start)); |
83a27052 ED |
1361 | |
1362 | do { | |
57a7744e | 1363 | start = u64_stats_fetch_begin_irq(&stats->rx_syncp); |
3fa2a1df | 1364 | rpackets = stats->rx_packets; |
1365 | rbytes = stats->rx_bytes; | |
57a7744e | 1366 | } while (u64_stats_fetch_retry_irq(&stats->rx_syncp, start)); |
3fa2a1df | 1367 | |
1368 | tot->rx_packets += rpackets; | |
1369 | tot->tx_packets += tpackets; | |
1370 | tot->rx_bytes += rbytes; | |
1371 | tot->tx_bytes += tbytes; | |
1372 | } | |
1373 | ||
1374 | tot->tx_dropped = dev->stats.tx_dropped; | |
021ac8d3 | 1375 | tot->tx_fifo_errors = dev->stats.tx_fifo_errors; |
3fa2a1df | 1376 | tot->rx_dropped = dev->stats.rx_dropped; |
1377 | tot->rx_length_errors = dev->stats.rx_length_errors; | |
1378 | tot->rx_frame_errors = dev->stats.rx_frame_errors; | |
3fa2a1df | 1379 | } |
1380 | ||
da74e89d AS |
1381 | #ifdef CONFIG_NET_POLL_CONTROLLER |
1382 | static void virtnet_netpoll(struct net_device *dev) | |
1383 | { | |
1384 | struct virtnet_info *vi = netdev_priv(dev); | |
986a4f4d | 1385 | int i; |
da74e89d | 1386 | |
986a4f4d JW |
1387 | for (i = 0; i < vi->curr_queue_pairs; i++) |
1388 | napi_schedule(&vi->rq[i].napi); | |
da74e89d AS |
1389 | } |
1390 | #endif | |
1391 | ||
586d17c5 JW |
1392 | static void virtnet_ack_link_announce(struct virtnet_info *vi) |
1393 | { | |
1394 | rtnl_lock(); | |
1395 | if (!virtnet_send_command(vi, VIRTIO_NET_CTRL_ANNOUNCE, | |
d24bae32 | 1396 | VIRTIO_NET_CTRL_ANNOUNCE_ACK, NULL)) |
586d17c5 JW |
1397 | dev_warn(&vi->dev->dev, "Failed to ack link announce.\n"); |
1398 | rtnl_unlock(); | |
1399 | } | |
1400 | ||
47315329 | 1401 | static int _virtnet_set_queues(struct virtnet_info *vi, u16 queue_pairs) |
986a4f4d JW |
1402 | { |
1403 | struct scatterlist sg; | |
986a4f4d JW |
1404 | struct net_device *dev = vi->dev; |
1405 | ||
1406 | if (!vi->has_cvq || !virtio_has_feature(vi->vdev, VIRTIO_NET_F_MQ)) | |
1407 | return 0; | |
1408 | ||
a725ee3e AL |
1409 | vi->ctrl_mq.virtqueue_pairs = cpu_to_virtio16(vi->vdev, queue_pairs); |
1410 | sg_init_one(&sg, &vi->ctrl_mq, sizeof(vi->ctrl_mq)); | |
986a4f4d JW |
1411 | |
1412 | if (!virtnet_send_command(vi, VIRTIO_NET_CTRL_MQ, | |
d24bae32 | 1413 | VIRTIO_NET_CTRL_MQ_VQ_PAIRS_SET, &sg)) { |
986a4f4d JW |
1414 | dev_warn(&dev->dev, "Fail to set num of queue pairs to %d\n", |
1415 | queue_pairs); | |
1416 | return -EINVAL; | |
55257d72 | 1417 | } else { |
986a4f4d | 1418 | vi->curr_queue_pairs = queue_pairs; |
35ed159b JW |
1419 | /* virtnet_open() will refill when device is going to up. */ |
1420 | if (dev->flags & IFF_UP) | |
1421 | schedule_delayed_work(&vi->refill, 0); | |
55257d72 | 1422 | } |
986a4f4d JW |
1423 | |
1424 | return 0; | |
1425 | } | |
1426 | ||
47315329 JF |
1427 | static int virtnet_set_queues(struct virtnet_info *vi, u16 queue_pairs) |
1428 | { | |
1429 | int err; | |
1430 | ||
1431 | rtnl_lock(); | |
1432 | err = _virtnet_set_queues(vi, queue_pairs); | |
1433 | rtnl_unlock(); | |
1434 | return err; | |
1435 | } | |
1436 | ||
296f96fc RR |
1437 | static int virtnet_close(struct net_device *dev) |
1438 | { | |
1439 | struct virtnet_info *vi = netdev_priv(dev); | |
986a4f4d | 1440 | int i; |
296f96fc | 1441 | |
b2baed69 RR |
1442 | /* Make sure refill_work doesn't re-enable napi! */ |
1443 | cancel_delayed_work_sync(&vi->refill); | |
986a4f4d | 1444 | |
b92f1e67 | 1445 | for (i = 0; i < vi->max_queue_pairs; i++) { |
986a4f4d | 1446 | napi_disable(&vi->rq[i].napi); |
78a57b48 | 1447 | virtnet_napi_tx_disable(&vi->sq[i].napi); |
b92f1e67 | 1448 | } |
296f96fc | 1449 | |
296f96fc RR |
1450 | return 0; |
1451 | } | |
1452 | ||
2af7698e AW |
1453 | static void virtnet_set_rx_mode(struct net_device *dev) |
1454 | { | |
1455 | struct virtnet_info *vi = netdev_priv(dev); | |
f565a7c2 | 1456 | struct scatterlist sg[2]; |
f565a7c2 | 1457 | struct virtio_net_ctrl_mac *mac_data; |
ccffad25 | 1458 | struct netdev_hw_addr *ha; |
32e7bfc4 | 1459 | int uc_count; |
4cd24eaf | 1460 | int mc_count; |
f565a7c2 AW |
1461 | void *buf; |
1462 | int i; | |
2af7698e | 1463 | |
788a8b6d | 1464 | /* We can't dynamically set ndo_set_rx_mode, so return gracefully */ |
2af7698e AW |
1465 | if (!virtio_has_feature(vi->vdev, VIRTIO_NET_F_CTRL_RX)) |
1466 | return; | |
1467 | ||
2ac46030 MT |
1468 | vi->ctrl_promisc = ((dev->flags & IFF_PROMISC) != 0); |
1469 | vi->ctrl_allmulti = ((dev->flags & IFF_ALLMULTI) != 0); | |
2af7698e | 1470 | |
2ac46030 | 1471 | sg_init_one(sg, &vi->ctrl_promisc, sizeof(vi->ctrl_promisc)); |
2af7698e AW |
1472 | |
1473 | if (!virtnet_send_command(vi, VIRTIO_NET_CTRL_RX, | |
d24bae32 | 1474 | VIRTIO_NET_CTRL_RX_PROMISC, sg)) |
2af7698e | 1475 | dev_warn(&dev->dev, "Failed to %sable promisc mode.\n", |
2ac46030 | 1476 | vi->ctrl_promisc ? "en" : "dis"); |
2af7698e | 1477 | |
2ac46030 | 1478 | sg_init_one(sg, &vi->ctrl_allmulti, sizeof(vi->ctrl_allmulti)); |
2af7698e AW |
1479 | |
1480 | if (!virtnet_send_command(vi, VIRTIO_NET_CTRL_RX, | |
d24bae32 | 1481 | VIRTIO_NET_CTRL_RX_ALLMULTI, sg)) |
2af7698e | 1482 | dev_warn(&dev->dev, "Failed to %sable allmulti mode.\n", |
2ac46030 | 1483 | vi->ctrl_allmulti ? "en" : "dis"); |
f565a7c2 | 1484 | |
32e7bfc4 | 1485 | uc_count = netdev_uc_count(dev); |
4cd24eaf | 1486 | mc_count = netdev_mc_count(dev); |
f565a7c2 | 1487 | /* MAC filter - use one buffer for both lists */ |
4cd24eaf JP |
1488 | buf = kzalloc(((uc_count + mc_count) * ETH_ALEN) + |
1489 | (2 * sizeof(mac_data->entries)), GFP_ATOMIC); | |
1490 | mac_data = buf; | |
e68ed8f0 | 1491 | if (!buf) |
f565a7c2 | 1492 | return; |
f565a7c2 | 1493 | |
23e258e1 AW |
1494 | sg_init_table(sg, 2); |
1495 | ||
f565a7c2 | 1496 | /* Store the unicast list and count in the front of the buffer */ |
fdd819b2 | 1497 | mac_data->entries = cpu_to_virtio32(vi->vdev, uc_count); |
ccffad25 | 1498 | i = 0; |
32e7bfc4 | 1499 | netdev_for_each_uc_addr(ha, dev) |
ccffad25 | 1500 | memcpy(&mac_data->macs[i++][0], ha->addr, ETH_ALEN); |
f565a7c2 AW |
1501 | |
1502 | sg_set_buf(&sg[0], mac_data, | |
32e7bfc4 | 1503 | sizeof(mac_data->entries) + (uc_count * ETH_ALEN)); |
f565a7c2 AW |
1504 | |
1505 | /* multicast list and count fill the end */ | |
32e7bfc4 | 1506 | mac_data = (void *)&mac_data->macs[uc_count][0]; |
f565a7c2 | 1507 | |
fdd819b2 | 1508 | mac_data->entries = cpu_to_virtio32(vi->vdev, mc_count); |
567ec874 | 1509 | i = 0; |
22bedad3 JP |
1510 | netdev_for_each_mc_addr(ha, dev) |
1511 | memcpy(&mac_data->macs[i++][0], ha->addr, ETH_ALEN); | |
f565a7c2 AW |
1512 | |
1513 | sg_set_buf(&sg[1], mac_data, | |
4cd24eaf | 1514 | sizeof(mac_data->entries) + (mc_count * ETH_ALEN)); |
f565a7c2 AW |
1515 | |
1516 | if (!virtnet_send_command(vi, VIRTIO_NET_CTRL_MAC, | |
d24bae32 | 1517 | VIRTIO_NET_CTRL_MAC_TABLE_SET, sg)) |
99e872ae | 1518 | dev_warn(&dev->dev, "Failed to set MAC filter table.\n"); |
f565a7c2 AW |
1519 | |
1520 | kfree(buf); | |
2af7698e AW |
1521 | } |
1522 | ||
80d5c368 PM |
1523 | static int virtnet_vlan_rx_add_vid(struct net_device *dev, |
1524 | __be16 proto, u16 vid) | |
0bde9569 AW |
1525 | { |
1526 | struct virtnet_info *vi = netdev_priv(dev); | |
1527 | struct scatterlist sg; | |
1528 | ||
a725ee3e AL |
1529 | vi->ctrl_vid = vid; |
1530 | sg_init_one(&sg, &vi->ctrl_vid, sizeof(vi->ctrl_vid)); | |
0bde9569 AW |
1531 | |
1532 | if (!virtnet_send_command(vi, VIRTIO_NET_CTRL_VLAN, | |
d24bae32 | 1533 | VIRTIO_NET_CTRL_VLAN_ADD, &sg)) |
0bde9569 | 1534 | dev_warn(&dev->dev, "Failed to add VLAN ID %d.\n", vid); |
8e586137 | 1535 | return 0; |
0bde9569 AW |
1536 | } |
1537 | ||
80d5c368 PM |
1538 | static int virtnet_vlan_rx_kill_vid(struct net_device *dev, |
1539 | __be16 proto, u16 vid) | |
0bde9569 AW |
1540 | { |
1541 | struct virtnet_info *vi = netdev_priv(dev); | |
1542 | struct scatterlist sg; | |
1543 | ||
a725ee3e AL |
1544 | vi->ctrl_vid = vid; |
1545 | sg_init_one(&sg, &vi->ctrl_vid, sizeof(vi->ctrl_vid)); | |
0bde9569 AW |
1546 | |
1547 | if (!virtnet_send_command(vi, VIRTIO_NET_CTRL_VLAN, | |
d24bae32 | 1548 | VIRTIO_NET_CTRL_VLAN_DEL, &sg)) |
0bde9569 | 1549 | dev_warn(&dev->dev, "Failed to kill VLAN ID %d.\n", vid); |
8e586137 | 1550 | return 0; |
0bde9569 AW |
1551 | } |
1552 | ||
8898c21c | 1553 | static void virtnet_clean_affinity(struct virtnet_info *vi, long hcpu) |
986a4f4d JW |
1554 | { |
1555 | int i; | |
1556 | ||
8898c21c WG |
1557 | if (vi->affinity_hint_set) { |
1558 | for (i = 0; i < vi->max_queue_pairs; i++) { | |
47be2479 WG |
1559 | virtqueue_set_affinity(vi->rq[i].vq, -1); |
1560 | virtqueue_set_affinity(vi->sq[i].vq, -1); | |
1561 | } | |
1562 | ||
8898c21c WG |
1563 | vi->affinity_hint_set = false; |
1564 | } | |
8898c21c | 1565 | } |
47be2479 | 1566 | |
8898c21c WG |
1567 | static void virtnet_set_affinity(struct virtnet_info *vi) |
1568 | { | |
1569 | int i; | |
1570 | int cpu; | |
986a4f4d JW |
1571 | |
1572 | /* In multiqueue mode, when the number of cpu is equal to the number of | |
1573 | * queue pairs, we let the queue pairs to be private to one cpu by | |
1574 | * setting the affinity hint to eliminate the contention. | |
1575 | */ | |
8898c21c WG |
1576 | if (vi->curr_queue_pairs == 1 || |
1577 | vi->max_queue_pairs != num_online_cpus()) { | |
1578 | virtnet_clean_affinity(vi, -1); | |
1579 | return; | |
986a4f4d JW |
1580 | } |
1581 | ||
8898c21c WG |
1582 | i = 0; |
1583 | for_each_online_cpu(cpu) { | |
986a4f4d JW |
1584 | virtqueue_set_affinity(vi->rq[i].vq, cpu); |
1585 | virtqueue_set_affinity(vi->sq[i].vq, cpu); | |
9bb8ca86 | 1586 | netif_set_xps_queue(vi->dev, cpumask_of(cpu), i); |
8898c21c | 1587 | i++; |
986a4f4d JW |
1588 | } |
1589 | ||
8898c21c | 1590 | vi->affinity_hint_set = true; |
986a4f4d JW |
1591 | } |
1592 | ||
8017c279 | 1593 | static int virtnet_cpu_online(unsigned int cpu, struct hlist_node *node) |
8de4b2f3 | 1594 | { |
8017c279 SAS |
1595 | struct virtnet_info *vi = hlist_entry_safe(node, struct virtnet_info, |
1596 | node); | |
1597 | virtnet_set_affinity(vi); | |
1598 | return 0; | |
1599 | } | |
8de4b2f3 | 1600 | |
8017c279 SAS |
1601 | static int virtnet_cpu_dead(unsigned int cpu, struct hlist_node *node) |
1602 | { | |
1603 | struct virtnet_info *vi = hlist_entry_safe(node, struct virtnet_info, | |
1604 | node_dead); | |
1605 | virtnet_set_affinity(vi); | |
1606 | return 0; | |
1607 | } | |
3ab098df | 1608 | |
8017c279 SAS |
1609 | static int virtnet_cpu_down_prep(unsigned int cpu, struct hlist_node *node) |
1610 | { | |
1611 | struct virtnet_info *vi = hlist_entry_safe(node, struct virtnet_info, | |
1612 | node); | |
1613 | ||
1614 | virtnet_clean_affinity(vi, cpu); | |
1615 | return 0; | |
1616 | } | |
1617 | ||
1618 | static enum cpuhp_state virtionet_online; | |
1619 | ||
1620 | static int virtnet_cpu_notif_add(struct virtnet_info *vi) | |
1621 | { | |
1622 | int ret; | |
1623 | ||
1624 | ret = cpuhp_state_add_instance_nocalls(virtionet_online, &vi->node); | |
1625 | if (ret) | |
1626 | return ret; | |
1627 | ret = cpuhp_state_add_instance_nocalls(CPUHP_VIRT_NET_DEAD, | |
1628 | &vi->node_dead); | |
1629 | if (!ret) | |
1630 | return ret; | |
1631 | cpuhp_state_remove_instance_nocalls(virtionet_online, &vi->node); | |
1632 | return ret; | |
1633 | } | |
1634 | ||
1635 | static void virtnet_cpu_notif_remove(struct virtnet_info *vi) | |
1636 | { | |
1637 | cpuhp_state_remove_instance_nocalls(virtionet_online, &vi->node); | |
1638 | cpuhp_state_remove_instance_nocalls(CPUHP_VIRT_NET_DEAD, | |
1639 | &vi->node_dead); | |
986a4f4d JW |
1640 | } |
1641 | ||
8f9f4668 RJ |
1642 | static void virtnet_get_ringparam(struct net_device *dev, |
1643 | struct ethtool_ringparam *ring) | |
1644 | { | |
1645 | struct virtnet_info *vi = netdev_priv(dev); | |
1646 | ||
986a4f4d JW |
1647 | ring->rx_max_pending = virtqueue_get_vring_size(vi->rq[0].vq); |
1648 | ring->tx_max_pending = virtqueue_get_vring_size(vi->sq[0].vq); | |
8f9f4668 RJ |
1649 | ring->rx_pending = ring->rx_max_pending; |
1650 | ring->tx_pending = ring->tx_max_pending; | |
8f9f4668 RJ |
1651 | } |
1652 | ||
66846048 RJ |
1653 | |
1654 | static void virtnet_get_drvinfo(struct net_device *dev, | |
1655 | struct ethtool_drvinfo *info) | |
1656 | { | |
1657 | struct virtnet_info *vi = netdev_priv(dev); | |
1658 | struct virtio_device *vdev = vi->vdev; | |
1659 | ||
1660 | strlcpy(info->driver, KBUILD_MODNAME, sizeof(info->driver)); | |
1661 | strlcpy(info->version, VIRTNET_DRIVER_VERSION, sizeof(info->version)); | |
1662 | strlcpy(info->bus_info, virtio_bus_name(vdev), sizeof(info->bus_info)); | |
1663 | ||
1664 | } | |
1665 | ||
d73bcd2c JW |
1666 | /* TODO: Eliminate OOO packets during switching */ |
1667 | static int virtnet_set_channels(struct net_device *dev, | |
1668 | struct ethtool_channels *channels) | |
1669 | { | |
1670 | struct virtnet_info *vi = netdev_priv(dev); | |
1671 | u16 queue_pairs = channels->combined_count; | |
1672 | int err; | |
1673 | ||
1674 | /* We don't support separate rx/tx channels. | |
1675 | * We don't allow setting 'other' channels. | |
1676 | */ | |
1677 | if (channels->rx_count || channels->tx_count || channels->other_count) | |
1678 | return -EINVAL; | |
1679 | ||
c18e9cd6 | 1680 | if (queue_pairs > vi->max_queue_pairs || queue_pairs == 0) |
d73bcd2c JW |
1681 | return -EINVAL; |
1682 | ||
f600b690 JF |
1683 | /* For now we don't support modifying channels while XDP is loaded |
1684 | * also when XDP is loaded all RX queues have XDP programs so we only | |
1685 | * need to check a single RX queue. | |
1686 | */ | |
1687 | if (vi->rq[0].xdp_prog) | |
1688 | return -EINVAL; | |
1689 | ||
47be2479 | 1690 | get_online_cpus(); |
47315329 | 1691 | err = _virtnet_set_queues(vi, queue_pairs); |
d73bcd2c JW |
1692 | if (!err) { |
1693 | netif_set_real_num_tx_queues(dev, queue_pairs); | |
1694 | netif_set_real_num_rx_queues(dev, queue_pairs); | |
1695 | ||
8898c21c | 1696 | virtnet_set_affinity(vi); |
d73bcd2c | 1697 | } |
47be2479 | 1698 | put_online_cpus(); |
d73bcd2c JW |
1699 | |
1700 | return err; | |
1701 | } | |
1702 | ||
1703 | static void virtnet_get_channels(struct net_device *dev, | |
1704 | struct ethtool_channels *channels) | |
1705 | { | |
1706 | struct virtnet_info *vi = netdev_priv(dev); | |
1707 | ||
1708 | channels->combined_count = vi->curr_queue_pairs; | |
1709 | channels->max_combined = vi->max_queue_pairs; | |
1710 | channels->max_other = 0; | |
1711 | channels->rx_count = 0; | |
1712 | channels->tx_count = 0; | |
1713 | channels->other_count = 0; | |
1714 | } | |
1715 | ||
16032be5 | 1716 | /* Check if the user is trying to change anything besides speed/duplex */ |
ebb6b4b1 PR |
1717 | static bool |
1718 | virtnet_validate_ethtool_cmd(const struct ethtool_link_ksettings *cmd) | |
16032be5 | 1719 | { |
ebb6b4b1 PR |
1720 | struct ethtool_link_ksettings diff1 = *cmd; |
1721 | struct ethtool_link_ksettings diff2 = {}; | |
16032be5 | 1722 | |
0cf3ace9 NA |
1723 | /* cmd is always set so we need to clear it, validate the port type |
1724 | * and also without autonegotiation we can ignore advertising | |
1725 | */ | |
ebb6b4b1 PR |
1726 | diff1.base.speed = 0; |
1727 | diff2.base.port = PORT_OTHER; | |
1728 | ethtool_link_ksettings_zero_link_mode(&diff1, advertising); | |
1729 | diff1.base.duplex = 0; | |
1730 | diff1.base.cmd = 0; | |
1731 | diff1.base.link_mode_masks_nwords = 0; | |
1732 | ||
1733 | return !memcmp(&diff1.base, &diff2.base, sizeof(diff1.base)) && | |
1734 | bitmap_empty(diff1.link_modes.supported, | |
1735 | __ETHTOOL_LINK_MODE_MASK_NBITS) && | |
1736 | bitmap_empty(diff1.link_modes.advertising, | |
1737 | __ETHTOOL_LINK_MODE_MASK_NBITS) && | |
1738 | bitmap_empty(diff1.link_modes.lp_advertising, | |
1739 | __ETHTOOL_LINK_MODE_MASK_NBITS); | |
16032be5 NA |
1740 | } |
1741 | ||
ebb6b4b1 PR |
1742 | static int virtnet_set_link_ksettings(struct net_device *dev, |
1743 | const struct ethtool_link_ksettings *cmd) | |
16032be5 NA |
1744 | { |
1745 | struct virtnet_info *vi = netdev_priv(dev); | |
1746 | u32 speed; | |
1747 | ||
ebb6b4b1 | 1748 | speed = cmd->base.speed; |
16032be5 NA |
1749 | /* don't allow custom speed and duplex */ |
1750 | if (!ethtool_validate_speed(speed) || | |
ebb6b4b1 | 1751 | !ethtool_validate_duplex(cmd->base.duplex) || |
16032be5 NA |
1752 | !virtnet_validate_ethtool_cmd(cmd)) |
1753 | return -EINVAL; | |
1754 | vi->speed = speed; | |
ebb6b4b1 | 1755 | vi->duplex = cmd->base.duplex; |
16032be5 NA |
1756 | |
1757 | return 0; | |
1758 | } | |
1759 | ||
ebb6b4b1 PR |
1760 | static int virtnet_get_link_ksettings(struct net_device *dev, |
1761 | struct ethtool_link_ksettings *cmd) | |
16032be5 NA |
1762 | { |
1763 | struct virtnet_info *vi = netdev_priv(dev); | |
1764 | ||
ebb6b4b1 PR |
1765 | cmd->base.speed = vi->speed; |
1766 | cmd->base.duplex = vi->duplex; | |
1767 | cmd->base.port = PORT_OTHER; | |
16032be5 NA |
1768 | |
1769 | return 0; | |
1770 | } | |
1771 | ||
1772 | static void virtnet_init_settings(struct net_device *dev) | |
1773 | { | |
1774 | struct virtnet_info *vi = netdev_priv(dev); | |
1775 | ||
1776 | vi->speed = SPEED_UNKNOWN; | |
1777 | vi->duplex = DUPLEX_UNKNOWN; | |
1778 | } | |
1779 | ||
0fc0b732 | 1780 | static const struct ethtool_ops virtnet_ethtool_ops = { |
66846048 | 1781 | .get_drvinfo = virtnet_get_drvinfo, |
9f4d26d0 | 1782 | .get_link = ethtool_op_get_link, |
8f9f4668 | 1783 | .get_ringparam = virtnet_get_ringparam, |
d73bcd2c JW |
1784 | .set_channels = virtnet_set_channels, |
1785 | .get_channels = virtnet_get_channels, | |
074c3582 | 1786 | .get_ts_info = ethtool_op_get_ts_info, |
ebb6b4b1 PR |
1787 | .get_link_ksettings = virtnet_get_link_ksettings, |
1788 | .set_link_ksettings = virtnet_set_link_ksettings, | |
a9ea3fc6 HX |
1789 | }; |
1790 | ||
9fe7bfce JF |
1791 | static void virtnet_freeze_down(struct virtio_device *vdev) |
1792 | { | |
1793 | struct virtnet_info *vi = vdev->priv; | |
1794 | int i; | |
1795 | ||
1796 | /* Make sure no work handler is accessing the device */ | |
1797 | flush_work(&vi->config_work); | |
1798 | ||
1799 | netif_device_detach(vi->dev); | |
713a98d9 | 1800 | netif_tx_disable(vi->dev); |
9fe7bfce JF |
1801 | cancel_delayed_work_sync(&vi->refill); |
1802 | ||
1803 | if (netif_running(vi->dev)) { | |
b92f1e67 | 1804 | for (i = 0; i < vi->max_queue_pairs; i++) { |
9fe7bfce | 1805 | napi_disable(&vi->rq[i].napi); |
78a57b48 | 1806 | virtnet_napi_tx_disable(&vi->sq[i].napi); |
b92f1e67 | 1807 | } |
9fe7bfce JF |
1808 | } |
1809 | } | |
1810 | ||
1811 | static int init_vqs(struct virtnet_info *vi); | |
2de2f7f4 | 1812 | static void _remove_vq_common(struct virtnet_info *vi); |
9fe7bfce JF |
1813 | |
1814 | static int virtnet_restore_up(struct virtio_device *vdev) | |
1815 | { | |
1816 | struct virtnet_info *vi = vdev->priv; | |
1817 | int err, i; | |
1818 | ||
1819 | err = init_vqs(vi); | |
1820 | if (err) | |
1821 | return err; | |
1822 | ||
1823 | virtio_device_ready(vdev); | |
1824 | ||
1825 | if (netif_running(vi->dev)) { | |
1826 | for (i = 0; i < vi->curr_queue_pairs; i++) | |
1827 | if (!try_fill_recv(vi, &vi->rq[i], GFP_KERNEL)) | |
1828 | schedule_delayed_work(&vi->refill, 0); | |
1829 | ||
b92f1e67 | 1830 | for (i = 0; i < vi->max_queue_pairs; i++) { |
e4e8452a | 1831 | virtnet_napi_enable(vi->rq[i].vq, &vi->rq[i].napi); |
b92f1e67 WB |
1832 | virtnet_napi_tx_enable(vi, vi->sq[i].vq, |
1833 | &vi->sq[i].napi); | |
1834 | } | |
9fe7bfce JF |
1835 | } |
1836 | ||
1837 | netif_device_attach(vi->dev); | |
1838 | return err; | |
1839 | } | |
1840 | ||
017b29c3 | 1841 | static int virtnet_reset(struct virtnet_info *vi, int curr_qp, int xdp_qp) |
2de2f7f4 JF |
1842 | { |
1843 | struct virtio_device *dev = vi->vdev; | |
1844 | int ret; | |
1845 | ||
1846 | virtio_config_disable(dev); | |
1847 | dev->failed = dev->config->get_status(dev) & VIRTIO_CONFIG_S_FAILED; | |
1848 | virtnet_freeze_down(dev); | |
1849 | _remove_vq_common(vi); | |
1850 | ||
2de2f7f4 JF |
1851 | virtio_add_status(dev, VIRTIO_CONFIG_S_ACKNOWLEDGE); |
1852 | virtio_add_status(dev, VIRTIO_CONFIG_S_DRIVER); | |
1853 | ||
1854 | ret = virtio_finalize_features(dev); | |
1855 | if (ret) | |
1856 | goto err; | |
1857 | ||
017b29c3 | 1858 | vi->xdp_queue_pairs = xdp_qp; |
2de2f7f4 JF |
1859 | ret = virtnet_restore_up(dev); |
1860 | if (ret) | |
1861 | goto err; | |
017b29c3 | 1862 | ret = _virtnet_set_queues(vi, curr_qp); |
2de2f7f4 JF |
1863 | if (ret) |
1864 | goto err; | |
1865 | ||
1866 | virtio_add_status(dev, VIRTIO_CONFIG_S_DRIVER_OK); | |
1867 | virtio_config_enable(dev); | |
1868 | return 0; | |
1869 | err: | |
1870 | virtio_add_status(dev, VIRTIO_CONFIG_S_FAILED); | |
1871 | return ret; | |
1872 | } | |
1873 | ||
9861ce03 JK |
1874 | static int virtnet_xdp_set(struct net_device *dev, struct bpf_prog *prog, |
1875 | struct netlink_ext_ack *extack) | |
f600b690 JF |
1876 | { |
1877 | unsigned long int max_sz = PAGE_SIZE - sizeof(struct padded_vnet_hdr); | |
1878 | struct virtnet_info *vi = netdev_priv(dev); | |
1879 | struct bpf_prog *old_prog; | |
017b29c3 | 1880 | u16 xdp_qp = 0, curr_qp; |
672aafd5 | 1881 | int i, err; |
f600b690 JF |
1882 | |
1883 | if (virtio_has_feature(vi->vdev, VIRTIO_NET_F_GUEST_TSO4) || | |
92502fe8 JW |
1884 | virtio_has_feature(vi->vdev, VIRTIO_NET_F_GUEST_TSO6) || |
1885 | virtio_has_feature(vi->vdev, VIRTIO_NET_F_GUEST_ECN) || | |
1886 | virtio_has_feature(vi->vdev, VIRTIO_NET_F_GUEST_UFO)) { | |
4d463c4d | 1887 | NL_SET_ERR_MSG_MOD(extack, "Can't set XDP while host is implementing LRO, disable LRO first"); |
f600b690 JF |
1888 | return -EOPNOTSUPP; |
1889 | } | |
1890 | ||
1891 | if (vi->mergeable_rx_bufs && !vi->any_header_sg) { | |
4d463c4d | 1892 | NL_SET_ERR_MSG_MOD(extack, "XDP expects header/data in single page, any_header_sg required"); |
f600b690 JF |
1893 | return -EINVAL; |
1894 | } | |
1895 | ||
1896 | if (dev->mtu > max_sz) { | |
4d463c4d | 1897 | NL_SET_ERR_MSG_MOD(extack, "MTU too large to enable XDP"); |
f600b690 JF |
1898 | netdev_warn(dev, "XDP requires MTU less than %lu\n", max_sz); |
1899 | return -EINVAL; | |
1900 | } | |
1901 | ||
672aafd5 JF |
1902 | curr_qp = vi->curr_queue_pairs - vi->xdp_queue_pairs; |
1903 | if (prog) | |
1904 | xdp_qp = nr_cpu_ids; | |
1905 | ||
1906 | /* XDP requires extra queues for XDP_TX */ | |
1907 | if (curr_qp + xdp_qp > vi->max_queue_pairs) { | |
4d463c4d | 1908 | NL_SET_ERR_MSG_MOD(extack, "Too few free TX rings available"); |
672aafd5 JF |
1909 | netdev_warn(dev, "request %i queues but max is %i\n", |
1910 | curr_qp + xdp_qp, vi->max_queue_pairs); | |
1911 | return -ENOMEM; | |
1912 | } | |
1913 | ||
2de2f7f4 JF |
1914 | if (prog) { |
1915 | prog = bpf_prog_add(prog, vi->max_queue_pairs - 1); | |
1916 | if (IS_ERR(prog)) | |
1917 | return PTR_ERR(prog); | |
1918 | } | |
1919 | ||
2de2f7f4 JF |
1920 | /* Changing the headroom in buffers is a disruptive operation because |
1921 | * existing buffers must be flushed and reallocated. This will happen | |
1922 | * when a xdp program is initially added or xdp is disabled by removing | |
1923 | * the xdp program resulting in number of XDP queues changing. | |
1924 | */ | |
1925 | if (vi->xdp_queue_pairs != xdp_qp) { | |
017b29c3 JW |
1926 | err = virtnet_reset(vi, curr_qp + xdp_qp, xdp_qp); |
1927 | if (err) { | |
1928 | dev_warn(&dev->dev, "XDP reset failure.\n"); | |
2de2f7f4 | 1929 | goto virtio_reset_err; |
017b29c3 | 1930 | } |
f600b690 JF |
1931 | } |
1932 | ||
672aafd5 JF |
1933 | netif_set_real_num_rx_queues(dev, curr_qp + xdp_qp); |
1934 | ||
f600b690 JF |
1935 | for (i = 0; i < vi->max_queue_pairs; i++) { |
1936 | old_prog = rtnl_dereference(vi->rq[i].xdp_prog); | |
1937 | rcu_assign_pointer(vi->rq[i].xdp_prog, prog); | |
1938 | if (old_prog) | |
1939 | bpf_prog_put(old_prog); | |
1940 | } | |
1941 | ||
1942 | return 0; | |
2de2f7f4 JF |
1943 | |
1944 | virtio_reset_err: | |
1945 | /* On reset error do our best to unwind XDP changes inflight and return | |
1946 | * error up to user space for resolution. The underlying reset hung on | |
1947 | * us so not much we can do here. | |
1948 | */ | |
2de2f7f4 JF |
1949 | if (prog) |
1950 | bpf_prog_sub(prog, vi->max_queue_pairs - 1); | |
1951 | return err; | |
f600b690 JF |
1952 | } |
1953 | ||
1954 | static bool virtnet_xdp_query(struct net_device *dev) | |
1955 | { | |
1956 | struct virtnet_info *vi = netdev_priv(dev); | |
1957 | int i; | |
1958 | ||
1959 | for (i = 0; i < vi->max_queue_pairs; i++) { | |
1960 | if (vi->rq[i].xdp_prog) | |
1961 | return true; | |
1962 | } | |
1963 | return false; | |
1964 | } | |
1965 | ||
1966 | static int virtnet_xdp(struct net_device *dev, struct netdev_xdp *xdp) | |
1967 | { | |
1968 | switch (xdp->command) { | |
1969 | case XDP_SETUP_PROG: | |
9861ce03 | 1970 | return virtnet_xdp_set(dev, xdp->prog, xdp->extack); |
f600b690 JF |
1971 | case XDP_QUERY_PROG: |
1972 | xdp->prog_attached = virtnet_xdp_query(dev); | |
1973 | return 0; | |
1974 | default: | |
1975 | return -EINVAL; | |
1976 | } | |
1977 | } | |
1978 | ||
76288b4e SH |
1979 | static const struct net_device_ops virtnet_netdev = { |
1980 | .ndo_open = virtnet_open, | |
1981 | .ndo_stop = virtnet_close, | |
1982 | .ndo_start_xmit = start_xmit, | |
1983 | .ndo_validate_addr = eth_validate_addr, | |
9c46f6d4 | 1984 | .ndo_set_mac_address = virtnet_set_mac_address, |
2af7698e | 1985 | .ndo_set_rx_mode = virtnet_set_rx_mode, |
3fa2a1df | 1986 | .ndo_get_stats64 = virtnet_stats, |
1824a989 AW |
1987 | .ndo_vlan_rx_add_vid = virtnet_vlan_rx_add_vid, |
1988 | .ndo_vlan_rx_kill_vid = virtnet_vlan_rx_kill_vid, | |
76288b4e SH |
1989 | #ifdef CONFIG_NET_POLL_CONTROLLER |
1990 | .ndo_poll_controller = virtnet_netpoll, | |
91815639 | 1991 | #endif |
f600b690 | 1992 | .ndo_xdp = virtnet_xdp, |
2836b4f2 | 1993 | .ndo_features_check = passthru_features_check, |
76288b4e SH |
1994 | }; |
1995 | ||
586d17c5 | 1996 | static void virtnet_config_changed_work(struct work_struct *work) |
9f4d26d0 | 1997 | { |
586d17c5 JW |
1998 | struct virtnet_info *vi = |
1999 | container_of(work, struct virtnet_info, config_work); | |
9f4d26d0 MM |
2000 | u16 v; |
2001 | ||
855e0c52 RR |
2002 | if (virtio_cread_feature(vi->vdev, VIRTIO_NET_F_STATUS, |
2003 | struct virtio_net_config, status, &v) < 0) | |
507613bf | 2004 | return; |
586d17c5 JW |
2005 | |
2006 | if (v & VIRTIO_NET_S_ANNOUNCE) { | |
ee89bab1 | 2007 | netdev_notify_peers(vi->dev); |
586d17c5 JW |
2008 | virtnet_ack_link_announce(vi); |
2009 | } | |
9f4d26d0 MM |
2010 | |
2011 | /* Ignore unknown (future) status bits */ | |
2012 | v &= VIRTIO_NET_S_LINK_UP; | |
2013 | ||
2014 | if (vi->status == v) | |
507613bf | 2015 | return; |
9f4d26d0 MM |
2016 | |
2017 | vi->status = v; | |
2018 | ||
2019 | if (vi->status & VIRTIO_NET_S_LINK_UP) { | |
2020 | netif_carrier_on(vi->dev); | |
986a4f4d | 2021 | netif_tx_wake_all_queues(vi->dev); |
9f4d26d0 MM |
2022 | } else { |
2023 | netif_carrier_off(vi->dev); | |
986a4f4d | 2024 | netif_tx_stop_all_queues(vi->dev); |
9f4d26d0 MM |
2025 | } |
2026 | } | |
2027 | ||
2028 | static void virtnet_config_changed(struct virtio_device *vdev) | |
2029 | { | |
2030 | struct virtnet_info *vi = vdev->priv; | |
2031 | ||
3b07e9ca | 2032 | schedule_work(&vi->config_work); |
9f4d26d0 MM |
2033 | } |
2034 | ||
986a4f4d JW |
2035 | static void virtnet_free_queues(struct virtnet_info *vi) |
2036 | { | |
d4fb84ee AV |
2037 | int i; |
2038 | ||
ab3971b1 JW |
2039 | for (i = 0; i < vi->max_queue_pairs; i++) { |
2040 | napi_hash_del(&vi->rq[i].napi); | |
d4fb84ee | 2041 | netif_napi_del(&vi->rq[i].napi); |
b92f1e67 | 2042 | netif_napi_del(&vi->sq[i].napi); |
ab3971b1 | 2043 | } |
d4fb84ee | 2044 | |
963abe5c ED |
2045 | /* We called napi_hash_del() before netif_napi_del(), |
2046 | * we need to respect an RCU grace period before freeing vi->rq | |
2047 | */ | |
2048 | synchronize_net(); | |
2049 | ||
986a4f4d JW |
2050 | kfree(vi->rq); |
2051 | kfree(vi->sq); | |
2052 | } | |
2053 | ||
47315329 | 2054 | static void _free_receive_bufs(struct virtnet_info *vi) |
986a4f4d | 2055 | { |
f600b690 | 2056 | struct bpf_prog *old_prog; |
986a4f4d JW |
2057 | int i; |
2058 | ||
2059 | for (i = 0; i < vi->max_queue_pairs; i++) { | |
2060 | while (vi->rq[i].pages) | |
2061 | __free_pages(get_a_page(&vi->rq[i], GFP_KERNEL), 0); | |
f600b690 JF |
2062 | |
2063 | old_prog = rtnl_dereference(vi->rq[i].xdp_prog); | |
2064 | RCU_INIT_POINTER(vi->rq[i].xdp_prog, NULL); | |
2065 | if (old_prog) | |
2066 | bpf_prog_put(old_prog); | |
986a4f4d | 2067 | } |
47315329 JF |
2068 | } |
2069 | ||
2070 | static void free_receive_bufs(struct virtnet_info *vi) | |
2071 | { | |
2072 | rtnl_lock(); | |
2073 | _free_receive_bufs(vi); | |
f600b690 | 2074 | rtnl_unlock(); |
986a4f4d JW |
2075 | } |
2076 | ||
fb51879d MD |
2077 | static void free_receive_page_frags(struct virtnet_info *vi) |
2078 | { | |
2079 | int i; | |
2080 | for (i = 0; i < vi->max_queue_pairs; i++) | |
2081 | if (vi->rq[i].alloc_frag.page) | |
2082 | put_page(vi->rq[i].alloc_frag.page); | |
2083 | } | |
2084 | ||
b68df015 | 2085 | static bool is_xdp_raw_buffer_queue(struct virtnet_info *vi, int q) |
56434a01 JF |
2086 | { |
2087 | if (q < (vi->curr_queue_pairs - vi->xdp_queue_pairs)) | |
2088 | return false; | |
2089 | else if (q < vi->curr_queue_pairs) | |
2090 | return true; | |
2091 | else | |
2092 | return false; | |
2093 | } | |
2094 | ||
986a4f4d JW |
2095 | static void free_unused_bufs(struct virtnet_info *vi) |
2096 | { | |
2097 | void *buf; | |
2098 | int i; | |
2099 | ||
2100 | for (i = 0; i < vi->max_queue_pairs; i++) { | |
2101 | struct virtqueue *vq = vi->sq[i].vq; | |
56434a01 | 2102 | while ((buf = virtqueue_detach_unused_buf(vq)) != NULL) { |
b68df015 | 2103 | if (!is_xdp_raw_buffer_queue(vi, i)) |
56434a01 JF |
2104 | dev_kfree_skb(buf); |
2105 | else | |
2106 | put_page(virt_to_head_page(buf)); | |
2107 | } | |
986a4f4d JW |
2108 | } |
2109 | ||
2110 | for (i = 0; i < vi->max_queue_pairs; i++) { | |
2111 | struct virtqueue *vq = vi->rq[i].vq; | |
2112 | ||
2113 | while ((buf = virtqueue_detach_unused_buf(vq)) != NULL) { | |
ab7db917 | 2114 | if (vi->mergeable_rx_bufs) { |
680557cf | 2115 | put_page(virt_to_head_page(buf)); |
ab7db917 | 2116 | } else if (vi->big_packets) { |
fa9fac17 | 2117 | give_pages(&vi->rq[i], buf); |
ab7db917 | 2118 | } else { |
f6b10209 | 2119 | put_page(virt_to_head_page(buf)); |
ab7db917 | 2120 | } |
986a4f4d | 2121 | } |
986a4f4d JW |
2122 | } |
2123 | } | |
2124 | ||
e9d7417b JW |
2125 | static void virtnet_del_vqs(struct virtnet_info *vi) |
2126 | { | |
2127 | struct virtio_device *vdev = vi->vdev; | |
2128 | ||
8898c21c | 2129 | virtnet_clean_affinity(vi, -1); |
986a4f4d | 2130 | |
e9d7417b | 2131 | vdev->config->del_vqs(vdev); |
986a4f4d JW |
2132 | |
2133 | virtnet_free_queues(vi); | |
e9d7417b JW |
2134 | } |
2135 | ||
d85b758f MT |
2136 | /* How large should a single buffer be so a queue full of these can fit at |
2137 | * least one full packet? | |
2138 | * Logic below assumes the mergeable buffer header is used. | |
2139 | */ | |
2140 | static unsigned int mergeable_min_buf_len(struct virtnet_info *vi, struct virtqueue *vq) | |
2141 | { | |
2142 | const unsigned int hdr_len = sizeof(struct virtio_net_hdr_mrg_rxbuf); | |
2143 | unsigned int rq_size = virtqueue_get_vring_size(vq); | |
2144 | unsigned int packet_len = vi->big_packets ? IP_MAX_MTU : vi->dev->max_mtu; | |
2145 | unsigned int buf_len = hdr_len + ETH_HLEN + VLAN_HLEN + packet_len; | |
2146 | unsigned int min_buf_len = DIV_ROUND_UP(buf_len, rq_size); | |
2147 | ||
f0c3192c MT |
2148 | return max(max(min_buf_len, hdr_len) - hdr_len, |
2149 | (unsigned int)GOOD_PACKET_LEN); | |
d85b758f MT |
2150 | } |
2151 | ||
986a4f4d | 2152 | static int virtnet_find_vqs(struct virtnet_info *vi) |
3f9c10b0 | 2153 | { |
986a4f4d JW |
2154 | vq_callback_t **callbacks; |
2155 | struct virtqueue **vqs; | |
2156 | int ret = -ENOMEM; | |
2157 | int i, total_vqs; | |
2158 | const char **names; | |
d45b897b | 2159 | bool *ctx; |
986a4f4d JW |
2160 | |
2161 | /* We expect 1 RX virtqueue followed by 1 TX virtqueue, followed by | |
2162 | * possible N-1 RX/TX queue pairs used in multiqueue mode, followed by | |
2163 | * possible control vq. | |
2164 | */ | |
2165 | total_vqs = vi->max_queue_pairs * 2 + | |
2166 | virtio_has_feature(vi->vdev, VIRTIO_NET_F_CTRL_VQ); | |
2167 | ||
2168 | /* Allocate space for find_vqs parameters */ | |
2169 | vqs = kzalloc(total_vqs * sizeof(*vqs), GFP_KERNEL); | |
2170 | if (!vqs) | |
2171 | goto err_vq; | |
2172 | callbacks = kmalloc(total_vqs * sizeof(*callbacks), GFP_KERNEL); | |
2173 | if (!callbacks) | |
2174 | goto err_callback; | |
2175 | names = kmalloc(total_vqs * sizeof(*names), GFP_KERNEL); | |
2176 | if (!names) | |
2177 | goto err_names; | |
d45b897b MT |
2178 | if (vi->mergeable_rx_bufs) { |
2179 | ctx = kzalloc(total_vqs * sizeof(*ctx), GFP_KERNEL); | |
2180 | if (!ctx) | |
2181 | goto err_ctx; | |
2182 | } else { | |
2183 | ctx = NULL; | |
2184 | } | |
986a4f4d JW |
2185 | |
2186 | /* Parameters for control virtqueue, if any */ | |
2187 | if (vi->has_cvq) { | |
2188 | callbacks[total_vqs - 1] = NULL; | |
2189 | names[total_vqs - 1] = "control"; | |
2190 | } | |
3f9c10b0 | 2191 | |
986a4f4d JW |
2192 | /* Allocate/initialize parameters for send/receive virtqueues */ |
2193 | for (i = 0; i < vi->max_queue_pairs; i++) { | |
2194 | callbacks[rxq2vq(i)] = skb_recv_done; | |
2195 | callbacks[txq2vq(i)] = skb_xmit_done; | |
2196 | sprintf(vi->rq[i].name, "input.%d", i); | |
2197 | sprintf(vi->sq[i].name, "output.%d", i); | |
2198 | names[rxq2vq(i)] = vi->rq[i].name; | |
2199 | names[txq2vq(i)] = vi->sq[i].name; | |
d45b897b MT |
2200 | if (ctx) |
2201 | ctx[rxq2vq(i)] = true; | |
986a4f4d | 2202 | } |
3f9c10b0 | 2203 | |
986a4f4d | 2204 | ret = vi->vdev->config->find_vqs(vi->vdev, total_vqs, vqs, callbacks, |
d45b897b | 2205 | names, ctx, NULL); |
986a4f4d JW |
2206 | if (ret) |
2207 | goto err_find; | |
3f9c10b0 | 2208 | |
986a4f4d JW |
2209 | if (vi->has_cvq) { |
2210 | vi->cvq = vqs[total_vqs - 1]; | |
3f9c10b0 | 2211 | if (virtio_has_feature(vi->vdev, VIRTIO_NET_F_CTRL_VLAN)) |
f646968f | 2212 | vi->dev->features |= NETIF_F_HW_VLAN_CTAG_FILTER; |
3f9c10b0 | 2213 | } |
986a4f4d JW |
2214 | |
2215 | for (i = 0; i < vi->max_queue_pairs; i++) { | |
2216 | vi->rq[i].vq = vqs[rxq2vq(i)]; | |
d85b758f | 2217 | vi->rq[i].min_buf_len = mergeable_min_buf_len(vi, vi->rq[i].vq); |
986a4f4d JW |
2218 | vi->sq[i].vq = vqs[txq2vq(i)]; |
2219 | } | |
2220 | ||
2221 | kfree(names); | |
2222 | kfree(callbacks); | |
2223 | kfree(vqs); | |
2224 | ||
3f9c10b0 | 2225 | return 0; |
986a4f4d JW |
2226 | |
2227 | err_find: | |
d45b897b MT |
2228 | kfree(ctx); |
2229 | err_ctx: | |
986a4f4d JW |
2230 | kfree(names); |
2231 | err_names: | |
2232 | kfree(callbacks); | |
2233 | err_callback: | |
2234 | kfree(vqs); | |
2235 | err_vq: | |
2236 | return ret; | |
2237 | } | |
2238 | ||
2239 | static int virtnet_alloc_queues(struct virtnet_info *vi) | |
2240 | { | |
2241 | int i; | |
2242 | ||
2243 | vi->sq = kzalloc(sizeof(*vi->sq) * vi->max_queue_pairs, GFP_KERNEL); | |
2244 | if (!vi->sq) | |
2245 | goto err_sq; | |
2246 | vi->rq = kzalloc(sizeof(*vi->rq) * vi->max_queue_pairs, GFP_KERNEL); | |
008d4278 | 2247 | if (!vi->rq) |
986a4f4d JW |
2248 | goto err_rq; |
2249 | ||
2250 | INIT_DELAYED_WORK(&vi->refill, refill_work); | |
2251 | for (i = 0; i < vi->max_queue_pairs; i++) { | |
2252 | vi->rq[i].pages = NULL; | |
2253 | netif_napi_add(vi->dev, &vi->rq[i].napi, virtnet_poll, | |
2254 | napi_weight); | |
1d11e732 WB |
2255 | netif_tx_napi_add(vi->dev, &vi->sq[i].napi, virtnet_poll_tx, |
2256 | napi_tx ? napi_weight : 0); | |
986a4f4d JW |
2257 | |
2258 | sg_init_table(vi->rq[i].sg, ARRAY_SIZE(vi->rq[i].sg)); | |
5377d758 | 2259 | ewma_pkt_len_init(&vi->rq[i].mrg_avg_pkt_len); |
986a4f4d JW |
2260 | sg_init_table(vi->sq[i].sg, ARRAY_SIZE(vi->sq[i].sg)); |
2261 | } | |
2262 | ||
2263 | return 0; | |
2264 | ||
2265 | err_rq: | |
2266 | kfree(vi->sq); | |
2267 | err_sq: | |
2268 | return -ENOMEM; | |
2269 | } | |
2270 | ||
2271 | static int init_vqs(struct virtnet_info *vi) | |
2272 | { | |
2273 | int ret; | |
2274 | ||
2275 | /* Allocate send & receive queues */ | |
2276 | ret = virtnet_alloc_queues(vi); | |
2277 | if (ret) | |
2278 | goto err; | |
2279 | ||
2280 | ret = virtnet_find_vqs(vi); | |
2281 | if (ret) | |
2282 | goto err_free; | |
2283 | ||
47be2479 | 2284 | get_online_cpus(); |
8898c21c | 2285 | virtnet_set_affinity(vi); |
47be2479 WG |
2286 | put_online_cpus(); |
2287 | ||
986a4f4d JW |
2288 | return 0; |
2289 | ||
2290 | err_free: | |
2291 | virtnet_free_queues(vi); | |
2292 | err: | |
2293 | return ret; | |
3f9c10b0 AS |
2294 | } |
2295 | ||
fbf28d78 MD |
2296 | #ifdef CONFIG_SYSFS |
2297 | static ssize_t mergeable_rx_buffer_size_show(struct netdev_rx_queue *queue, | |
2298 | struct rx_queue_attribute *attribute, char *buf) | |
2299 | { | |
2300 | struct virtnet_info *vi = netdev_priv(queue->dev); | |
2301 | unsigned int queue_index = get_netdev_rx_queue_index(queue); | |
5377d758 | 2302 | struct ewma_pkt_len *avg; |
fbf28d78 MD |
2303 | |
2304 | BUG_ON(queue_index >= vi->max_queue_pairs); | |
2305 | avg = &vi->rq[queue_index].mrg_avg_pkt_len; | |
d85b758f MT |
2306 | return sprintf(buf, "%u\n", |
2307 | get_mergeable_buf_len(&vi->rq[queue_index], avg)); | |
fbf28d78 MD |
2308 | } |
2309 | ||
2310 | static struct rx_queue_attribute mergeable_rx_buffer_size_attribute = | |
2311 | __ATTR_RO(mergeable_rx_buffer_size); | |
2312 | ||
2313 | static struct attribute *virtio_net_mrg_rx_attrs[] = { | |
2314 | &mergeable_rx_buffer_size_attribute.attr, | |
2315 | NULL | |
2316 | }; | |
2317 | ||
2318 | static const struct attribute_group virtio_net_mrg_rx_group = { | |
2319 | .name = "virtio_net", | |
2320 | .attrs = virtio_net_mrg_rx_attrs | |
2321 | }; | |
2322 | #endif | |
2323 | ||
892d6eb1 JW |
2324 | static bool virtnet_fail_on_feature(struct virtio_device *vdev, |
2325 | unsigned int fbit, | |
2326 | const char *fname, const char *dname) | |
2327 | { | |
2328 | if (!virtio_has_feature(vdev, fbit)) | |
2329 | return false; | |
2330 | ||
2331 | dev_err(&vdev->dev, "device advertises feature %s but not %s", | |
2332 | fname, dname); | |
2333 | ||
2334 | return true; | |
2335 | } | |
2336 | ||
2337 | #define VIRTNET_FAIL_ON(vdev, fbit, dbit) \ | |
2338 | virtnet_fail_on_feature(vdev, fbit, #fbit, dbit) | |
2339 | ||
2340 | static bool virtnet_validate_features(struct virtio_device *vdev) | |
2341 | { | |
2342 | if (!virtio_has_feature(vdev, VIRTIO_NET_F_CTRL_VQ) && | |
2343 | (VIRTNET_FAIL_ON(vdev, VIRTIO_NET_F_CTRL_RX, | |
2344 | "VIRTIO_NET_F_CTRL_VQ") || | |
2345 | VIRTNET_FAIL_ON(vdev, VIRTIO_NET_F_CTRL_VLAN, | |
2346 | "VIRTIO_NET_F_CTRL_VQ") || | |
2347 | VIRTNET_FAIL_ON(vdev, VIRTIO_NET_F_GUEST_ANNOUNCE, | |
2348 | "VIRTIO_NET_F_CTRL_VQ") || | |
2349 | VIRTNET_FAIL_ON(vdev, VIRTIO_NET_F_MQ, "VIRTIO_NET_F_CTRL_VQ") || | |
2350 | VIRTNET_FAIL_ON(vdev, VIRTIO_NET_F_CTRL_MAC_ADDR, | |
2351 | "VIRTIO_NET_F_CTRL_VQ"))) { | |
2352 | return false; | |
2353 | } | |
2354 | ||
2355 | return true; | |
2356 | } | |
2357 | ||
d0c2c997 JW |
2358 | #define MIN_MTU ETH_MIN_MTU |
2359 | #define MAX_MTU ETH_MAX_MTU | |
2360 | ||
fe36cbe0 | 2361 | static int virtnet_validate(struct virtio_device *vdev) |
296f96fc | 2362 | { |
6ba42248 MT |
2363 | if (!vdev->config->get) { |
2364 | dev_err(&vdev->dev, "%s failure: config access disabled\n", | |
2365 | __func__); | |
2366 | return -EINVAL; | |
2367 | } | |
2368 | ||
892d6eb1 JW |
2369 | if (!virtnet_validate_features(vdev)) |
2370 | return -EINVAL; | |
2371 | ||
fe36cbe0 MT |
2372 | if (virtio_has_feature(vdev, VIRTIO_NET_F_MTU)) { |
2373 | int mtu = virtio_cread16(vdev, | |
2374 | offsetof(struct virtio_net_config, | |
2375 | mtu)); | |
2376 | if (mtu < MIN_MTU) | |
2377 | __virtio_clear_bit(vdev, VIRTIO_NET_F_MTU); | |
2378 | } | |
2379 | ||
2380 | return 0; | |
2381 | } | |
2382 | ||
2383 | static int virtnet_probe(struct virtio_device *vdev) | |
2384 | { | |
2385 | int i, err; | |
2386 | struct net_device *dev; | |
2387 | struct virtnet_info *vi; | |
2388 | u16 max_queue_pairs; | |
2389 | int mtu; | |
2390 | ||
986a4f4d | 2391 | /* Find if host supports multiqueue virtio_net device */ |
855e0c52 RR |
2392 | err = virtio_cread_feature(vdev, VIRTIO_NET_F_MQ, |
2393 | struct virtio_net_config, | |
2394 | max_virtqueue_pairs, &max_queue_pairs); | |
986a4f4d JW |
2395 | |
2396 | /* We need at least 2 queue's */ | |
2397 | if (err || max_queue_pairs < VIRTIO_NET_CTRL_MQ_VQ_PAIRS_MIN || | |
2398 | max_queue_pairs > VIRTIO_NET_CTRL_MQ_VQ_PAIRS_MAX || | |
2399 | !virtio_has_feature(vdev, VIRTIO_NET_F_CTRL_VQ)) | |
2400 | max_queue_pairs = 1; | |
296f96fc RR |
2401 | |
2402 | /* Allocate ourselves a network device with room for our info */ | |
986a4f4d | 2403 | dev = alloc_etherdev_mq(sizeof(struct virtnet_info), max_queue_pairs); |
296f96fc RR |
2404 | if (!dev) |
2405 | return -ENOMEM; | |
2406 | ||
2407 | /* Set up network device as normal. */ | |
f2f2c8b4 | 2408 | dev->priv_flags |= IFF_UNICAST_FLT | IFF_LIVE_ADDR_CHANGE; |
76288b4e | 2409 | dev->netdev_ops = &virtnet_netdev; |
296f96fc | 2410 | dev->features = NETIF_F_HIGHDMA; |
3fa2a1df | 2411 | |
7ad24ea4 | 2412 | dev->ethtool_ops = &virtnet_ethtool_ops; |
296f96fc RR |
2413 | SET_NETDEV_DEV(dev, &vdev->dev); |
2414 | ||
2415 | /* Do we support "hardware" checksums? */ | |
98e778c9 | 2416 | if (virtio_has_feature(vdev, VIRTIO_NET_F_CSUM)) { |
296f96fc | 2417 | /* This opens up the world of extra features. */ |
48900cb6 | 2418 | dev->hw_features |= NETIF_F_HW_CSUM | NETIF_F_SG; |
98e778c9 | 2419 | if (csum) |
48900cb6 | 2420 | dev->features |= NETIF_F_HW_CSUM | NETIF_F_SG; |
98e778c9 MM |
2421 | |
2422 | if (virtio_has_feature(vdev, VIRTIO_NET_F_GSO)) { | |
e3e3c423 | 2423 | dev->hw_features |= NETIF_F_TSO | NETIF_F_UFO |
34a48579 RR |
2424 | | NETIF_F_TSO_ECN | NETIF_F_TSO6; |
2425 | } | |
5539ae96 | 2426 | /* Individual feature bits: what can host handle? */ |
98e778c9 MM |
2427 | if (virtio_has_feature(vdev, VIRTIO_NET_F_HOST_TSO4)) |
2428 | dev->hw_features |= NETIF_F_TSO; | |
2429 | if (virtio_has_feature(vdev, VIRTIO_NET_F_HOST_TSO6)) | |
2430 | dev->hw_features |= NETIF_F_TSO6; | |
2431 | if (virtio_has_feature(vdev, VIRTIO_NET_F_HOST_ECN)) | |
2432 | dev->hw_features |= NETIF_F_TSO_ECN; | |
e3e3c423 VY |
2433 | if (virtio_has_feature(vdev, VIRTIO_NET_F_HOST_UFO)) |
2434 | dev->hw_features |= NETIF_F_UFO; | |
98e778c9 | 2435 | |
41f2f127 JW |
2436 | dev->features |= NETIF_F_GSO_ROBUST; |
2437 | ||
98e778c9 | 2438 | if (gso) |
e3e3c423 | 2439 | dev->features |= dev->hw_features & (NETIF_F_ALL_TSO|NETIF_F_UFO); |
98e778c9 | 2440 | /* (!csum && gso) case will be fixed by register_netdev() */ |
296f96fc | 2441 | } |
4f49129b TH |
2442 | if (virtio_has_feature(vdev, VIRTIO_NET_F_GUEST_CSUM)) |
2443 | dev->features |= NETIF_F_RXCSUM; | |
296f96fc | 2444 | |
4fda8302 JW |
2445 | dev->vlan_features = dev->features; |
2446 | ||
d0c2c997 JW |
2447 | /* MTU range: 68 - 65535 */ |
2448 | dev->min_mtu = MIN_MTU; | |
2449 | dev->max_mtu = MAX_MTU; | |
2450 | ||
296f96fc | 2451 | /* Configuration may specify what MAC to use. Otherwise random. */ |
855e0c52 RR |
2452 | if (virtio_has_feature(vdev, VIRTIO_NET_F_MAC)) |
2453 | virtio_cread_bytes(vdev, | |
2454 | offsetof(struct virtio_net_config, mac), | |
2455 | dev->dev_addr, dev->addr_len); | |
2456 | else | |
f2cedb63 | 2457 | eth_hw_addr_random(dev); |
296f96fc RR |
2458 | |
2459 | /* Set up our device-specific information */ | |
2460 | vi = netdev_priv(dev); | |
296f96fc RR |
2461 | vi->dev = dev; |
2462 | vi->vdev = vdev; | |
d9d5dcc8 | 2463 | vdev->priv = vi; |
3fa2a1df | 2464 | vi->stats = alloc_percpu(struct virtnet_stats); |
2465 | err = -ENOMEM; | |
2466 | if (vi->stats == NULL) | |
2467 | goto free; | |
2468 | ||
827da44c JS |
2469 | for_each_possible_cpu(i) { |
2470 | struct virtnet_stats *virtnet_stats; | |
2471 | virtnet_stats = per_cpu_ptr(vi->stats, i); | |
2472 | u64_stats_init(&virtnet_stats->tx_syncp); | |
2473 | u64_stats_init(&virtnet_stats->rx_syncp); | |
2474 | } | |
2475 | ||
586d17c5 | 2476 | INIT_WORK(&vi->config_work, virtnet_config_changed_work); |
296f96fc | 2477 | |
97402b96 | 2478 | /* If we can receive ANY GSO packets, we must allocate large ones. */ |
8e95a202 JP |
2479 | if (virtio_has_feature(vdev, VIRTIO_NET_F_GUEST_TSO4) || |
2480 | virtio_has_feature(vdev, VIRTIO_NET_F_GUEST_TSO6) || | |
e3e3c423 VY |
2481 | virtio_has_feature(vdev, VIRTIO_NET_F_GUEST_ECN) || |
2482 | virtio_has_feature(vdev, VIRTIO_NET_F_GUEST_UFO)) | |
97402b96 HX |
2483 | vi->big_packets = true; |
2484 | ||
3f2c31d9 MM |
2485 | if (virtio_has_feature(vdev, VIRTIO_NET_F_MRG_RXBUF)) |
2486 | vi->mergeable_rx_bufs = true; | |
2487 | ||
d04302b3 MT |
2488 | if (virtio_has_feature(vdev, VIRTIO_NET_F_MRG_RXBUF) || |
2489 | virtio_has_feature(vdev, VIRTIO_F_VERSION_1)) | |
012873d0 MT |
2490 | vi->hdr_len = sizeof(struct virtio_net_hdr_mrg_rxbuf); |
2491 | else | |
2492 | vi->hdr_len = sizeof(struct virtio_net_hdr); | |
2493 | ||
75993300 MT |
2494 | if (virtio_has_feature(vdev, VIRTIO_F_ANY_LAYOUT) || |
2495 | virtio_has_feature(vdev, VIRTIO_F_VERSION_1)) | |
e7428e95 MT |
2496 | vi->any_header_sg = true; |
2497 | ||
986a4f4d JW |
2498 | if (virtio_has_feature(vdev, VIRTIO_NET_F_CTRL_VQ)) |
2499 | vi->has_cvq = true; | |
2500 | ||
14de9d11 AC |
2501 | if (virtio_has_feature(vdev, VIRTIO_NET_F_MTU)) { |
2502 | mtu = virtio_cread16(vdev, | |
2503 | offsetof(struct virtio_net_config, | |
2504 | mtu)); | |
93a205ee | 2505 | if (mtu < dev->min_mtu) { |
fe36cbe0 MT |
2506 | /* Should never trigger: MTU was previously validated |
2507 | * in virtnet_validate. | |
2508 | */ | |
2509 | dev_err(&vdev->dev, "device MTU appears to have changed " | |
2510 | "it is now %d < %d", mtu, dev->min_mtu); | |
2511 | goto free_stats; | |
93a205ee | 2512 | } |
2e123b44 | 2513 | |
fe36cbe0 MT |
2514 | dev->mtu = mtu; |
2515 | dev->max_mtu = mtu; | |
2516 | ||
2e123b44 MT |
2517 | /* TODO: size buffers correctly in this case. */ |
2518 | if (dev->mtu > ETH_DATA_LEN) | |
2519 | vi->big_packets = true; | |
14de9d11 AC |
2520 | } |
2521 | ||
012873d0 MT |
2522 | if (vi->any_header_sg) |
2523 | dev->needed_headroom = vi->hdr_len; | |
6ebbc1a6 | 2524 | |
44900010 JW |
2525 | /* Enable multiqueue by default */ |
2526 | if (num_online_cpus() >= max_queue_pairs) | |
2527 | vi->curr_queue_pairs = max_queue_pairs; | |
2528 | else | |
2529 | vi->curr_queue_pairs = num_online_cpus(); | |
986a4f4d JW |
2530 | vi->max_queue_pairs = max_queue_pairs; |
2531 | ||
2532 | /* Allocate/initialize the rx/tx queues, and invoke find_vqs */ | |
3f9c10b0 | 2533 | err = init_vqs(vi); |
d2a7ddda | 2534 | if (err) |
9bb8ca86 | 2535 | goto free_stats; |
296f96fc | 2536 | |
fbf28d78 MD |
2537 | #ifdef CONFIG_SYSFS |
2538 | if (vi->mergeable_rx_bufs) | |
2539 | dev->sysfs_rx_queue_group = &virtio_net_mrg_rx_group; | |
2540 | #endif | |
0f13b66b ZYW |
2541 | netif_set_real_num_tx_queues(dev, vi->curr_queue_pairs); |
2542 | netif_set_real_num_rx_queues(dev, vi->curr_queue_pairs); | |
986a4f4d | 2543 | |
16032be5 NA |
2544 | virtnet_init_settings(dev); |
2545 | ||
296f96fc RR |
2546 | err = register_netdev(dev); |
2547 | if (err) { | |
2548 | pr_debug("virtio_net: registering device failed\n"); | |
d2a7ddda | 2549 | goto free_vqs; |
296f96fc | 2550 | } |
b3369c1f | 2551 | |
4baf1e33 MT |
2552 | virtio_device_ready(vdev); |
2553 | ||
8017c279 | 2554 | err = virtnet_cpu_notif_add(vi); |
8de4b2f3 WG |
2555 | if (err) { |
2556 | pr_debug("virtio_net: registering cpu notifier failed\n"); | |
f00e35e2 | 2557 | goto free_unregister_netdev; |
8de4b2f3 WG |
2558 | } |
2559 | ||
a220871b | 2560 | virtnet_set_queues(vi, vi->curr_queue_pairs); |
44900010 | 2561 | |
167c25e4 JW |
2562 | /* Assume link up if device can't report link status, |
2563 | otherwise get link status from config. */ | |
2564 | if (virtio_has_feature(vi->vdev, VIRTIO_NET_F_STATUS)) { | |
2565 | netif_carrier_off(dev); | |
3b07e9ca | 2566 | schedule_work(&vi->config_work); |
167c25e4 JW |
2567 | } else { |
2568 | vi->status = VIRTIO_NET_S_LINK_UP; | |
2569 | netif_carrier_on(dev); | |
2570 | } | |
9f4d26d0 | 2571 | |
986a4f4d JW |
2572 | pr_debug("virtnet: registered device %s with %d RX and TX vq's\n", |
2573 | dev->name, max_queue_pairs); | |
2574 | ||
296f96fc RR |
2575 | return 0; |
2576 | ||
f00e35e2 | 2577 | free_unregister_netdev: |
02465555 MT |
2578 | vi->vdev->config->reset(vdev); |
2579 | ||
b3369c1f | 2580 | unregister_netdev(dev); |
d2a7ddda | 2581 | free_vqs: |
986a4f4d | 2582 | cancel_delayed_work_sync(&vi->refill); |
fb51879d | 2583 | free_receive_page_frags(vi); |
e9d7417b | 2584 | virtnet_del_vqs(vi); |
3fa2a1df | 2585 | free_stats: |
2586 | free_percpu(vi->stats); | |
296f96fc RR |
2587 | free: |
2588 | free_netdev(dev); | |
2589 | return err; | |
2590 | } | |
2591 | ||
2de2f7f4 JF |
2592 | static void _remove_vq_common(struct virtnet_info *vi) |
2593 | { | |
2594 | vi->vdev->config->reset(vi->vdev); | |
2595 | free_unused_bufs(vi); | |
2596 | _free_receive_bufs(vi); | |
2597 | free_receive_page_frags(vi); | |
2598 | virtnet_del_vqs(vi); | |
2599 | } | |
2600 | ||
04486ed0 | 2601 | static void remove_vq_common(struct virtnet_info *vi) |
296f96fc | 2602 | { |
04486ed0 | 2603 | vi->vdev->config->reset(vi->vdev); |
830a8a97 SM |
2604 | |
2605 | /* Free unused buffers in both send and recv, if any. */ | |
9ab86bbc | 2606 | free_unused_bufs(vi); |
fb6813f4 | 2607 | |
986a4f4d | 2608 | free_receive_bufs(vi); |
d2a7ddda | 2609 | |
fb51879d MD |
2610 | free_receive_page_frags(vi); |
2611 | ||
986a4f4d | 2612 | virtnet_del_vqs(vi); |
04486ed0 AS |
2613 | } |
2614 | ||
8cc085d6 | 2615 | static void virtnet_remove(struct virtio_device *vdev) |
04486ed0 AS |
2616 | { |
2617 | struct virtnet_info *vi = vdev->priv; | |
2618 | ||
8017c279 | 2619 | virtnet_cpu_notif_remove(vi); |
8de4b2f3 | 2620 | |
102a2786 MT |
2621 | /* Make sure no work handler is accessing the device. */ |
2622 | flush_work(&vi->config_work); | |
586d17c5 | 2623 | |
04486ed0 AS |
2624 | unregister_netdev(vi->dev); |
2625 | ||
2626 | remove_vq_common(vi); | |
fb6813f4 | 2627 | |
2e66f55b | 2628 | free_percpu(vi->stats); |
74b2553f | 2629 | free_netdev(vi->dev); |
296f96fc RR |
2630 | } |
2631 | ||
89107000 | 2632 | #ifdef CONFIG_PM_SLEEP |
0741bcb5 AS |
2633 | static int virtnet_freeze(struct virtio_device *vdev) |
2634 | { | |
2635 | struct virtnet_info *vi = vdev->priv; | |
2636 | ||
8017c279 | 2637 | virtnet_cpu_notif_remove(vi); |
9fe7bfce | 2638 | virtnet_freeze_down(vdev); |
0741bcb5 AS |
2639 | remove_vq_common(vi); |
2640 | ||
2641 | return 0; | |
2642 | } | |
2643 | ||
2644 | static int virtnet_restore(struct virtio_device *vdev) | |
2645 | { | |
2646 | struct virtnet_info *vi = vdev->priv; | |
9fe7bfce | 2647 | int err; |
0741bcb5 | 2648 | |
9fe7bfce | 2649 | err = virtnet_restore_up(vdev); |
0741bcb5 AS |
2650 | if (err) |
2651 | return err; | |
986a4f4d JW |
2652 | virtnet_set_queues(vi, vi->curr_queue_pairs); |
2653 | ||
8017c279 | 2654 | err = virtnet_cpu_notif_add(vi); |
ec9debbd JW |
2655 | if (err) |
2656 | return err; | |
2657 | ||
0741bcb5 AS |
2658 | return 0; |
2659 | } | |
2660 | #endif | |
2661 | ||
296f96fc RR |
2662 | static struct virtio_device_id id_table[] = { |
2663 | { VIRTIO_ID_NET, VIRTIO_DEV_ANY_ID }, | |
2664 | { 0 }, | |
2665 | }; | |
2666 | ||
f3358507 MT |
2667 | #define VIRTNET_FEATURES \ |
2668 | VIRTIO_NET_F_CSUM, VIRTIO_NET_F_GUEST_CSUM, \ | |
2669 | VIRTIO_NET_F_MAC, \ | |
2670 | VIRTIO_NET_F_HOST_TSO4, VIRTIO_NET_F_HOST_UFO, VIRTIO_NET_F_HOST_TSO6, \ | |
2671 | VIRTIO_NET_F_HOST_ECN, VIRTIO_NET_F_GUEST_TSO4, VIRTIO_NET_F_GUEST_TSO6, \ | |
2672 | VIRTIO_NET_F_GUEST_ECN, VIRTIO_NET_F_GUEST_UFO, \ | |
2673 | VIRTIO_NET_F_MRG_RXBUF, VIRTIO_NET_F_STATUS, VIRTIO_NET_F_CTRL_VQ, \ | |
2674 | VIRTIO_NET_F_CTRL_RX, VIRTIO_NET_F_CTRL_VLAN, \ | |
2675 | VIRTIO_NET_F_GUEST_ANNOUNCE, VIRTIO_NET_F_MQ, \ | |
2676 | VIRTIO_NET_F_CTRL_MAC_ADDR, \ | |
2677 | VIRTIO_NET_F_MTU | |
2678 | ||
c45a6816 | 2679 | static unsigned int features[] = { |
f3358507 MT |
2680 | VIRTNET_FEATURES, |
2681 | }; | |
2682 | ||
2683 | static unsigned int features_legacy[] = { | |
2684 | VIRTNET_FEATURES, | |
2685 | VIRTIO_NET_F_GSO, | |
e7428e95 | 2686 | VIRTIO_F_ANY_LAYOUT, |
c45a6816 RR |
2687 | }; |
2688 | ||
22402529 | 2689 | static struct virtio_driver virtio_net_driver = { |
c45a6816 RR |
2690 | .feature_table = features, |
2691 | .feature_table_size = ARRAY_SIZE(features), | |
f3358507 MT |
2692 | .feature_table_legacy = features_legacy, |
2693 | .feature_table_size_legacy = ARRAY_SIZE(features_legacy), | |
296f96fc RR |
2694 | .driver.name = KBUILD_MODNAME, |
2695 | .driver.owner = THIS_MODULE, | |
2696 | .id_table = id_table, | |
fe36cbe0 | 2697 | .validate = virtnet_validate, |
296f96fc | 2698 | .probe = virtnet_probe, |
8cc085d6 | 2699 | .remove = virtnet_remove, |
9f4d26d0 | 2700 | .config_changed = virtnet_config_changed, |
89107000 | 2701 | #ifdef CONFIG_PM_SLEEP |
0741bcb5 AS |
2702 | .freeze = virtnet_freeze, |
2703 | .restore = virtnet_restore, | |
2704 | #endif | |
296f96fc RR |
2705 | }; |
2706 | ||
8017c279 SAS |
2707 | static __init int virtio_net_driver_init(void) |
2708 | { | |
2709 | int ret; | |
2710 | ||
73c1b41e | 2711 | ret = cpuhp_setup_state_multi(CPUHP_AP_ONLINE_DYN, "virtio/net:online", |
8017c279 SAS |
2712 | virtnet_cpu_online, |
2713 | virtnet_cpu_down_prep); | |
2714 | if (ret < 0) | |
2715 | goto out; | |
2716 | virtionet_online = ret; | |
73c1b41e | 2717 | ret = cpuhp_setup_state_multi(CPUHP_VIRT_NET_DEAD, "virtio/net:dead", |
8017c279 SAS |
2718 | NULL, virtnet_cpu_dead); |
2719 | if (ret) | |
2720 | goto err_dead; | |
2721 | ||
2722 | ret = register_virtio_driver(&virtio_net_driver); | |
2723 | if (ret) | |
2724 | goto err_virtio; | |
2725 | return 0; | |
2726 | err_virtio: | |
2727 | cpuhp_remove_multi_state(CPUHP_VIRT_NET_DEAD); | |
2728 | err_dead: | |
2729 | cpuhp_remove_multi_state(virtionet_online); | |
2730 | out: | |
2731 | return ret; | |
2732 | } | |
2733 | module_init(virtio_net_driver_init); | |
2734 | ||
2735 | static __exit void virtio_net_driver_exit(void) | |
2736 | { | |
2737 | cpuhp_remove_multi_state(CPUHP_VIRT_NET_DEAD); | |
2738 | cpuhp_remove_multi_state(virtionet_online); | |
2739 | unregister_virtio_driver(&virtio_net_driver); | |
2740 | } | |
2741 | module_exit(virtio_net_driver_exit); | |
296f96fc RR |
2742 | |
2743 | MODULE_DEVICE_TABLE(virtio, id_table); | |
2744 | MODULE_DESCRIPTION("Virtio network driver"); | |
2745 | MODULE_LICENSE("GPL"); |