]>
Commit | Line | Data |
---|---|---|
48925e37 | 1 | /* A network driver using virtio. |
296f96fc RR |
2 | * |
3 | * Copyright 2007 Rusty Russell <rusty@rustcorp.com.au> IBM Corporation | |
4 | * | |
5 | * This program is free software; you can redistribute it and/or modify | |
6 | * it under the terms of the GNU General Public License as published by | |
7 | * the Free Software Foundation; either version 2 of the License, or | |
8 | * (at your option) any later version. | |
9 | * | |
10 | * This program is distributed in the hope that it will be useful, | |
11 | * but WITHOUT ANY WARRANTY; without even the implied warranty of | |
12 | * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the | |
13 | * GNU General Public License for more details. | |
14 | * | |
15 | * You should have received a copy of the GNU General Public License | |
adf8d3ff | 16 | * along with this program; if not, see <http://www.gnu.org/licenses/>. |
296f96fc RR |
17 | */ |
18 | //#define DEBUG | |
19 | #include <linux/netdevice.h> | |
20 | #include <linux/etherdevice.h> | |
a9ea3fc6 | 21 | #include <linux/ethtool.h> |
296f96fc RR |
22 | #include <linux/module.h> |
23 | #include <linux/virtio.h> | |
24 | #include <linux/virtio_net.h> | |
f600b690 | 25 | #include <linux/bpf.h> |
a67edbf4 | 26 | #include <linux/bpf_trace.h> |
296f96fc | 27 | #include <linux/scatterlist.h> |
e918085a | 28 | #include <linux/if_vlan.h> |
5a0e3ad6 | 29 | #include <linux/slab.h> |
8de4b2f3 | 30 | #include <linux/cpu.h> |
ab7db917 | 31 | #include <linux/average.h> |
186b3c99 | 32 | #include <linux/filter.h> |
d85b758f | 33 | #include <net/route.h> |
296f96fc | 34 | |
d34710e3 | 35 | static int napi_weight = NAPI_POLL_WEIGHT; |
6c0cd7c0 DL |
36 | module_param(napi_weight, int, 0444); |
37 | ||
b92f1e67 | 38 | static bool csum = true, gso = true, napi_tx; |
34a48579 RR |
39 | module_param(csum, bool, 0444); |
40 | module_param(gso, bool, 0444); | |
b92f1e67 | 41 | module_param(napi_tx, bool, 0644); |
34a48579 | 42 | |
296f96fc | 43 | /* FIXME: MTU in config. */ |
5061de36 | 44 | #define GOOD_PACKET_LEN (ETH_HLEN + VLAN_HLEN + ETH_DATA_LEN) |
3f2c31d9 | 45 | #define GOOD_COPY_LEN 128 |
296f96fc | 46 | |
f6b10209 JW |
47 | #define VIRTNET_RX_PAD (NET_IP_ALIGN + NET_SKB_PAD) |
48 | ||
2de2f7f4 JF |
49 | /* Amount of XDP headroom to prepend to packets for use by xdp_adjust_head */ |
50 | #define VIRTIO_XDP_HEADROOM 256 | |
51 | ||
5377d758 JB |
52 | /* RX packet size EWMA. The average packet size is used to determine the packet |
53 | * buffer size when refilling RX rings. As the entire RX ring may be refilled | |
54 | * at once, the weight is chosen so that the EWMA will be insensitive to short- | |
55 | * term, transient changes in packet size. | |
ab7db917 | 56 | */ |
eb1e011a | 57 | DECLARE_EWMA(pkt_len, 0, 64) |
ab7db917 | 58 | |
66846048 | 59 | #define VIRTNET_DRIVER_VERSION "1.0.0" |
2a41f71d | 60 | |
7acd4329 CIK |
61 | static const unsigned long guest_offloads[] = { |
62 | VIRTIO_NET_F_GUEST_TSO4, | |
63 | VIRTIO_NET_F_GUEST_TSO6, | |
64 | VIRTIO_NET_F_GUEST_ECN, | |
465688ce JW |
65 | VIRTIO_NET_F_GUEST_UFO, |
66 | VIRTIO_NET_F_GUEST_CSUM | |
7acd4329 | 67 | }; |
3f93522f | 68 | |
3fa2a1df | 69 | struct virtnet_stats { |
83a27052 ED |
70 | struct u64_stats_sync tx_syncp; |
71 | struct u64_stats_sync rx_syncp; | |
3fa2a1df | 72 | u64 tx_bytes; |
73 | u64 tx_packets; | |
74 | ||
75 | u64 rx_bytes; | |
76 | u64 rx_packets; | |
77 | }; | |
78 | ||
e9d7417b JW |
79 | /* Internal representation of a send virtqueue */ |
80 | struct send_queue { | |
81 | /* Virtqueue associated with this send _queue */ | |
82 | struct virtqueue *vq; | |
83 | ||
84 | /* TX: fragments + linear part + virtio header */ | |
85 | struct scatterlist sg[MAX_SKB_FRAGS + 2]; | |
986a4f4d JW |
86 | |
87 | /* Name of the send queue: output.$index */ | |
88 | char name[40]; | |
b92f1e67 WB |
89 | |
90 | struct napi_struct napi; | |
e9d7417b JW |
91 | }; |
92 | ||
93 | /* Internal representation of a receive virtqueue */ | |
94 | struct receive_queue { | |
95 | /* Virtqueue associated with this receive_queue */ | |
96 | struct virtqueue *vq; | |
97 | ||
296f96fc RR |
98 | struct napi_struct napi; |
99 | ||
f600b690 JF |
100 | struct bpf_prog __rcu *xdp_prog; |
101 | ||
e9d7417b JW |
102 | /* Chain pages by the private ptr. */ |
103 | struct page *pages; | |
104 | ||
ab7db917 | 105 | /* Average packet length for mergeable receive buffers. */ |
5377d758 | 106 | struct ewma_pkt_len mrg_avg_pkt_len; |
ab7db917 | 107 | |
fb51879d MD |
108 | /* Page frag for packet buffer allocation. */ |
109 | struct page_frag alloc_frag; | |
110 | ||
e9d7417b JW |
111 | /* RX: fragments + linear part + virtio header */ |
112 | struct scatterlist sg[MAX_SKB_FRAGS + 2]; | |
986a4f4d | 113 | |
d85b758f MT |
114 | /* Min single buffer size for mergeable buffers case. */ |
115 | unsigned int min_buf_len; | |
116 | ||
986a4f4d JW |
117 | /* Name of this receive queue: input.$index */ |
118 | char name[40]; | |
e9d7417b JW |
119 | }; |
120 | ||
603343b9 MT |
121 | /* Control VQ buffers: protected by the rtnl lock */ |
122 | struct control_buf { | |
123 | struct virtio_net_ctrl_hdr hdr; | |
124 | virtio_net_ctrl_ack status; | |
125 | struct virtio_net_ctrl_mq mq; | |
126 | u8 promisc; | |
127 | u8 allmulti; | |
c285a7d5 | 128 | __virtio16 vid; |
603343b9 MT |
129 | u64 offloads; |
130 | }; | |
131 | ||
e9d7417b JW |
132 | struct virtnet_info { |
133 | struct virtio_device *vdev; | |
134 | struct virtqueue *cvq; | |
135 | struct net_device *dev; | |
986a4f4d JW |
136 | struct send_queue *sq; |
137 | struct receive_queue *rq; | |
e9d7417b JW |
138 | unsigned int status; |
139 | ||
986a4f4d JW |
140 | /* Max # of queue pairs supported by the device */ |
141 | u16 max_queue_pairs; | |
142 | ||
143 | /* # of queue pairs currently used by the driver */ | |
144 | u16 curr_queue_pairs; | |
145 | ||
672aafd5 JF |
146 | /* # of XDP queue pairs currently used by the driver */ |
147 | u16 xdp_queue_pairs; | |
148 | ||
97402b96 HX |
149 | /* I like... big packets and I cannot lie! */ |
150 | bool big_packets; | |
151 | ||
3f2c31d9 MM |
152 | /* Host will merge rx buffers for big packets (shake it! shake it!) */ |
153 | bool mergeable_rx_bufs; | |
154 | ||
986a4f4d JW |
155 | /* Has control virtqueue */ |
156 | bool has_cvq; | |
157 | ||
e7428e95 MT |
158 | /* Host can handle any s/g split between our header and packet data */ |
159 | bool any_header_sg; | |
160 | ||
012873d0 MT |
161 | /* Packet virtio header size */ |
162 | u8 hdr_len; | |
163 | ||
3fa2a1df | 164 | /* Active statistics */ |
165 | struct virtnet_stats __percpu *stats; | |
166 | ||
3161e453 RR |
167 | /* Work struct for refilling if we run low on memory. */ |
168 | struct delayed_work refill; | |
169 | ||
586d17c5 JW |
170 | /* Work struct for config space updates */ |
171 | struct work_struct config_work; | |
172 | ||
986a4f4d JW |
173 | /* Does the affinity hint is set for virtqueues? */ |
174 | bool affinity_hint_set; | |
47be2479 | 175 | |
8017c279 SAS |
176 | /* CPU hotplug instances for online & dead */ |
177 | struct hlist_node node; | |
178 | struct hlist_node node_dead; | |
2ac46030 | 179 | |
603343b9 | 180 | struct control_buf *ctrl; |
16032be5 NA |
181 | |
182 | /* Ethtool settings */ | |
183 | u8 duplex; | |
184 | u32 speed; | |
3f93522f JW |
185 | |
186 | unsigned long guest_offloads; | |
296f96fc RR |
187 | }; |
188 | ||
9ab86bbc | 189 | struct padded_vnet_hdr { |
012873d0 | 190 | struct virtio_net_hdr_mrg_rxbuf hdr; |
9ab86bbc | 191 | /* |
012873d0 MT |
192 | * hdr is in a separate sg buffer, and data sg buffer shares same page |
193 | * with this header sg. This padding makes next sg 16 byte aligned | |
194 | * after the header. | |
9ab86bbc | 195 | */ |
012873d0 | 196 | char padding[4]; |
9ab86bbc SM |
197 | }; |
198 | ||
986a4f4d JW |
199 | /* Converting between virtqueue no. and kernel tx/rx queue no. |
200 | * 0:rx0 1:tx0 2:rx1 3:tx1 ... 2N:rxN 2N+1:txN 2N+2:cvq | |
201 | */ | |
202 | static int vq2txq(struct virtqueue *vq) | |
203 | { | |
9d0ca6ed | 204 | return (vq->index - 1) / 2; |
986a4f4d JW |
205 | } |
206 | ||
207 | static int txq2vq(int txq) | |
208 | { | |
209 | return txq * 2 + 1; | |
210 | } | |
211 | ||
212 | static int vq2rxq(struct virtqueue *vq) | |
213 | { | |
9d0ca6ed | 214 | return vq->index / 2; |
986a4f4d JW |
215 | } |
216 | ||
217 | static int rxq2vq(int rxq) | |
218 | { | |
219 | return rxq * 2; | |
220 | } | |
221 | ||
012873d0 | 222 | static inline struct virtio_net_hdr_mrg_rxbuf *skb_vnet_hdr(struct sk_buff *skb) |
296f96fc | 223 | { |
012873d0 | 224 | return (struct virtio_net_hdr_mrg_rxbuf *)skb->cb; |
296f96fc RR |
225 | } |
226 | ||
9ab86bbc SM |
227 | /* |
228 | * private is used to chain pages for big packets, put the whole | |
229 | * most recent used list in the beginning for reuse | |
230 | */ | |
e9d7417b | 231 | static void give_pages(struct receive_queue *rq, struct page *page) |
0a888fd1 | 232 | { |
9ab86bbc | 233 | struct page *end; |
0a888fd1 | 234 | |
e9d7417b | 235 | /* Find end of list, sew whole thing into vi->rq.pages. */ |
9ab86bbc | 236 | for (end = page; end->private; end = (struct page *)end->private); |
e9d7417b JW |
237 | end->private = (unsigned long)rq->pages; |
238 | rq->pages = page; | |
0a888fd1 MM |
239 | } |
240 | ||
e9d7417b | 241 | static struct page *get_a_page(struct receive_queue *rq, gfp_t gfp_mask) |
fb6813f4 | 242 | { |
e9d7417b | 243 | struct page *p = rq->pages; |
fb6813f4 | 244 | |
9ab86bbc | 245 | if (p) { |
e9d7417b | 246 | rq->pages = (struct page *)p->private; |
9ab86bbc SM |
247 | /* clear private here, it is used to chain pages */ |
248 | p->private = 0; | |
249 | } else | |
fb6813f4 RR |
250 | p = alloc_page(gfp_mask); |
251 | return p; | |
252 | } | |
253 | ||
e4e8452a WB |
254 | static void virtqueue_napi_schedule(struct napi_struct *napi, |
255 | struct virtqueue *vq) | |
256 | { | |
257 | if (napi_schedule_prep(napi)) { | |
258 | virtqueue_disable_cb(vq); | |
259 | __napi_schedule(napi); | |
260 | } | |
261 | } | |
262 | ||
263 | static void virtqueue_napi_complete(struct napi_struct *napi, | |
264 | struct virtqueue *vq, int processed) | |
265 | { | |
266 | int opaque; | |
267 | ||
268 | opaque = virtqueue_enable_cb_prepare(vq); | |
e3eb8fe2 TM |
269 | if (napi_complete_done(napi, processed)) { |
270 | if (unlikely(virtqueue_poll(vq, opaque))) | |
271 | virtqueue_napi_schedule(napi, vq); | |
272 | } else { | |
273 | virtqueue_disable_cb(vq); | |
274 | } | |
e4e8452a WB |
275 | } |
276 | ||
e9d7417b | 277 | static void skb_xmit_done(struct virtqueue *vq) |
296f96fc | 278 | { |
e9d7417b | 279 | struct virtnet_info *vi = vq->vdev->priv; |
b92f1e67 | 280 | struct napi_struct *napi = &vi->sq[vq2txq(vq)].napi; |
296f96fc | 281 | |
2cb9c6ba | 282 | /* Suppress further interrupts. */ |
e9d7417b | 283 | virtqueue_disable_cb(vq); |
11a3a154 | 284 | |
b92f1e67 WB |
285 | if (napi->weight) |
286 | virtqueue_napi_schedule(napi, vq); | |
287 | else | |
288 | /* We were probably waiting for more output buffers. */ | |
289 | netif_wake_subqueue(vi->dev, vq2txq(vq)); | |
296f96fc RR |
290 | } |
291 | ||
28b39bc7 JW |
292 | #define MRG_CTX_HEADER_SHIFT 22 |
293 | static void *mergeable_len_to_ctx(unsigned int truesize, | |
294 | unsigned int headroom) | |
295 | { | |
296 | return (void *)(unsigned long)((headroom << MRG_CTX_HEADER_SHIFT) | truesize); | |
297 | } | |
298 | ||
299 | static unsigned int mergeable_ctx_to_headroom(void *mrg_ctx) | |
300 | { | |
301 | return (unsigned long)mrg_ctx >> MRG_CTX_HEADER_SHIFT; | |
302 | } | |
303 | ||
304 | static unsigned int mergeable_ctx_to_truesize(void *mrg_ctx) | |
305 | { | |
306 | return (unsigned long)mrg_ctx & ((1 << MRG_CTX_HEADER_SHIFT) - 1); | |
307 | } | |
308 | ||
3464645a | 309 | /* Called from bottom half context */ |
946fa564 MT |
310 | static struct sk_buff *page_to_skb(struct virtnet_info *vi, |
311 | struct receive_queue *rq, | |
2613af0e | 312 | struct page *page, unsigned int offset, |
5e229598 JW |
313 | unsigned int len, unsigned int truesize, |
314 | bool hdr_valid) | |
9ab86bbc SM |
315 | { |
316 | struct sk_buff *skb; | |
012873d0 | 317 | struct virtio_net_hdr_mrg_rxbuf *hdr; |
2613af0e | 318 | unsigned int copy, hdr_len, hdr_padded_len; |
9ab86bbc | 319 | char *p; |
fb6813f4 | 320 | |
2613af0e | 321 | p = page_address(page) + offset; |
3f2c31d9 | 322 | |
9ab86bbc | 323 | /* copy small packet so we can reuse these pages for small data */ |
c67f5db8 | 324 | skb = napi_alloc_skb(&rq->napi, GOOD_COPY_LEN); |
9ab86bbc SM |
325 | if (unlikely(!skb)) |
326 | return NULL; | |
3f2c31d9 | 327 | |
9ab86bbc | 328 | hdr = skb_vnet_hdr(skb); |
3f2c31d9 | 329 | |
012873d0 MT |
330 | hdr_len = vi->hdr_len; |
331 | if (vi->mergeable_rx_bufs) | |
a4a76503 | 332 | hdr_padded_len = sizeof(*hdr); |
012873d0 | 333 | else |
2613af0e | 334 | hdr_padded_len = sizeof(struct padded_vnet_hdr); |
3f2c31d9 | 335 | |
5e229598 JW |
336 | if (hdr_valid) |
337 | memcpy(hdr, p, hdr_len); | |
3f2c31d9 | 338 | |
9ab86bbc | 339 | len -= hdr_len; |
2613af0e MD |
340 | offset += hdr_padded_len; |
341 | p += hdr_padded_len; | |
3f2c31d9 | 342 | |
9ab86bbc SM |
343 | copy = len; |
344 | if (copy > skb_tailroom(skb)) | |
345 | copy = skb_tailroom(skb); | |
59ae1d12 | 346 | skb_put_data(skb, p, copy); |
3f2c31d9 | 347 | |
9ab86bbc SM |
348 | len -= copy; |
349 | offset += copy; | |
3f2c31d9 | 350 | |
2613af0e MD |
351 | if (vi->mergeable_rx_bufs) { |
352 | if (len) | |
353 | skb_add_rx_frag(skb, 0, page, offset, len, truesize); | |
354 | else | |
355 | put_page(page); | |
356 | return skb; | |
357 | } | |
358 | ||
e878d78b SL |
359 | /* |
360 | * Verify that we can indeed put this data into a skb. | |
361 | * This is here to handle cases when the device erroneously | |
362 | * tries to receive more than is possible. This is usually | |
363 | * the case of a broken device. | |
364 | */ | |
365 | if (unlikely(len > MAX_SKB_FRAGS * PAGE_SIZE)) { | |
be443899 | 366 | net_dbg_ratelimited("%s: too much data\n", skb->dev->name); |
e878d78b SL |
367 | dev_kfree_skb(skb); |
368 | return NULL; | |
369 | } | |
2613af0e | 370 | BUG_ON(offset >= PAGE_SIZE); |
9ab86bbc | 371 | while (len) { |
2613af0e MD |
372 | unsigned int frag_size = min((unsigned)PAGE_SIZE - offset, len); |
373 | skb_add_rx_frag(skb, skb_shinfo(skb)->nr_frags, page, offset, | |
374 | frag_size, truesize); | |
375 | len -= frag_size; | |
9ab86bbc SM |
376 | page = (struct page *)page->private; |
377 | offset = 0; | |
378 | } | |
3f2c31d9 | 379 | |
9ab86bbc | 380 | if (page) |
e9d7417b | 381 | give_pages(rq, page); |
3f2c31d9 | 382 | |
9ab86bbc SM |
383 | return skb; |
384 | } | |
3f2c31d9 | 385 | |
186b3c99 JW |
386 | static void virtnet_xdp_flush(struct net_device *dev) |
387 | { | |
388 | struct virtnet_info *vi = netdev_priv(dev); | |
389 | struct send_queue *sq; | |
390 | unsigned int qp; | |
391 | ||
392 | qp = vi->curr_queue_pairs - vi->xdp_queue_pairs + smp_processor_id(); | |
393 | sq = &vi->sq[qp]; | |
394 | ||
395 | virtqueue_kick(sq->vq); | |
396 | } | |
397 | ||
398 | static bool __virtnet_xdp_xmit(struct virtnet_info *vi, | |
399 | struct xdp_buff *xdp) | |
56434a01 | 400 | { |
56434a01 | 401 | struct virtio_net_hdr_mrg_rxbuf *hdr; |
f6b10209 | 402 | unsigned int len; |
722d8283 JF |
403 | struct send_queue *sq; |
404 | unsigned int qp; | |
56434a01 JF |
405 | void *xdp_sent; |
406 | int err; | |
407 | ||
722d8283 JF |
408 | qp = vi->curr_queue_pairs - vi->xdp_queue_pairs + smp_processor_id(); |
409 | sq = &vi->sq[qp]; | |
410 | ||
56434a01 JF |
411 | /* Free up any pending old buffers before queueing new ones. */ |
412 | while ((xdp_sent = virtqueue_get_buf(sq->vq, &len)) != NULL) { | |
f6b10209 | 413 | struct page *sent_page = virt_to_head_page(xdp_sent); |
bb91accf | 414 | |
f6b10209 | 415 | put_page(sent_page); |
bb91accf | 416 | } |
56434a01 | 417 | |
f6b10209 JW |
418 | xdp->data -= vi->hdr_len; |
419 | /* Zero header and leave csum up to XDP layers */ | |
420 | hdr = xdp->data; | |
421 | memset(hdr, 0, vi->hdr_len); | |
bb91accf | 422 | |
f6b10209 | 423 | sg_init_one(sq->sg, xdp->data, xdp->data_end - xdp->data); |
bb91accf | 424 | |
f6b10209 | 425 | err = virtqueue_add_outbuf(sq->vq, sq->sg, 1, xdp->data, GFP_ATOMIC); |
6e49f219 JDB |
426 | if (unlikely(err)) |
427 | return false; /* Caller handle free/refcnt */ | |
56434a01 | 428 | |
a67edbf4 | 429 | return true; |
56434a01 JF |
430 | } |
431 | ||
186b3c99 JW |
432 | static int virtnet_xdp_xmit(struct net_device *dev, struct xdp_buff *xdp) |
433 | { | |
434 | struct virtnet_info *vi = netdev_priv(dev); | |
435 | bool sent = __virtnet_xdp_xmit(vi, xdp); | |
436 | ||
437 | if (!sent) | |
438 | return -ENOSPC; | |
439 | return 0; | |
440 | } | |
441 | ||
f6b10209 JW |
442 | static unsigned int virtnet_get_headroom(struct virtnet_info *vi) |
443 | { | |
444 | return vi->xdp_queue_pairs ? VIRTIO_XDP_HEADROOM : 0; | |
445 | } | |
446 | ||
4941d472 JW |
447 | /* We copy the packet for XDP in the following cases: |
448 | * | |
449 | * 1) Packet is scattered across multiple rx buffers. | |
450 | * 2) Headroom space is insufficient. | |
451 | * | |
452 | * This is inefficient but it's a temporary condition that | |
453 | * we hit right after XDP is enabled and until queue is refilled | |
454 | * with large buffers with sufficient headroom - so it should affect | |
455 | * at most queue size packets. | |
456 | * Afterwards, the conditions to enable | |
457 | * XDP should preclude the underlying device from sending packets | |
458 | * across multiple buffers (num_buf > 1), and we make sure buffers | |
459 | * have enough headroom. | |
460 | */ | |
461 | static struct page *xdp_linearize_page(struct receive_queue *rq, | |
462 | u16 *num_buf, | |
463 | struct page *p, | |
464 | int offset, | |
465 | int page_off, | |
466 | unsigned int *len) | |
467 | { | |
468 | struct page *page = alloc_page(GFP_ATOMIC); | |
469 | ||
470 | if (!page) | |
471 | return NULL; | |
472 | ||
473 | memcpy(page_address(page) + page_off, page_address(p) + offset, *len); | |
474 | page_off += *len; | |
475 | ||
476 | while (--*num_buf) { | |
477 | unsigned int buflen; | |
478 | void *buf; | |
479 | int off; | |
480 | ||
481 | buf = virtqueue_get_buf(rq->vq, &buflen); | |
482 | if (unlikely(!buf)) | |
483 | goto err_buf; | |
484 | ||
485 | p = virt_to_head_page(buf); | |
486 | off = buf - page_address(p); | |
487 | ||
488 | /* guard against a misconfigured or uncooperative backend that | |
489 | * is sending packet larger than the MTU. | |
490 | */ | |
491 | if ((page_off + buflen) > PAGE_SIZE) { | |
492 | put_page(p); | |
493 | goto err_buf; | |
494 | } | |
495 | ||
496 | memcpy(page_address(page) + page_off, | |
497 | page_address(p) + off, buflen); | |
498 | page_off += buflen; | |
499 | put_page(p); | |
500 | } | |
501 | ||
502 | /* Headroom does not contribute to packet length */ | |
503 | *len = page_off - VIRTIO_XDP_HEADROOM; | |
504 | return page; | |
505 | err_buf: | |
506 | __free_pages(page, 0); | |
507 | return NULL; | |
508 | } | |
509 | ||
bb91accf JW |
510 | static struct sk_buff *receive_small(struct net_device *dev, |
511 | struct virtnet_info *vi, | |
512 | struct receive_queue *rq, | |
192f68cf | 513 | void *buf, void *ctx, |
186b3c99 JW |
514 | unsigned int len, |
515 | bool *xdp_xmit) | |
f121159d | 516 | { |
f6b10209 | 517 | struct sk_buff *skb; |
bb91accf | 518 | struct bpf_prog *xdp_prog; |
4941d472 | 519 | unsigned int xdp_headroom = (unsigned long)ctx; |
f6b10209 JW |
520 | unsigned int header_offset = VIRTNET_RX_PAD + xdp_headroom; |
521 | unsigned int headroom = vi->hdr_len + header_offset; | |
522 | unsigned int buflen = SKB_DATA_ALIGN(GOOD_PACKET_LEN + headroom) + | |
523 | SKB_DATA_ALIGN(sizeof(struct skb_shared_info)); | |
4941d472 | 524 | struct page *page = virt_to_head_page(buf); |
6e49f219 | 525 | unsigned int delta = 0; |
4941d472 | 526 | struct page *xdp_page; |
6e49f219 JDB |
527 | bool sent; |
528 | int err; | |
529 | ||
012873d0 | 530 | len -= vi->hdr_len; |
f121159d | 531 | |
bb91accf JW |
532 | rcu_read_lock(); |
533 | xdp_prog = rcu_dereference(rq->xdp_prog); | |
534 | if (xdp_prog) { | |
f6b10209 | 535 | struct virtio_net_hdr_mrg_rxbuf *hdr = buf + header_offset; |
0354e4d1 | 536 | struct xdp_buff xdp; |
f6b10209 | 537 | void *orig_data; |
bb91accf JW |
538 | u32 act; |
539 | ||
36c3dce0 | 540 | if (unlikely(hdr->hdr.gso_type)) |
bb91accf | 541 | goto err_xdp; |
0354e4d1 | 542 | |
4941d472 JW |
543 | if (unlikely(xdp_headroom < virtnet_get_headroom(vi))) { |
544 | int offset = buf - page_address(page) + header_offset; | |
545 | unsigned int tlen = len + vi->hdr_len; | |
546 | u16 num_buf = 1; | |
547 | ||
548 | xdp_headroom = virtnet_get_headroom(vi); | |
549 | header_offset = VIRTNET_RX_PAD + xdp_headroom; | |
550 | headroom = vi->hdr_len + header_offset; | |
551 | buflen = SKB_DATA_ALIGN(GOOD_PACKET_LEN + headroom) + | |
552 | SKB_DATA_ALIGN(sizeof(struct skb_shared_info)); | |
553 | xdp_page = xdp_linearize_page(rq, &num_buf, page, | |
554 | offset, header_offset, | |
555 | &tlen); | |
556 | if (!xdp_page) | |
557 | goto err_xdp; | |
558 | ||
559 | buf = page_address(xdp_page); | |
560 | put_page(page); | |
561 | page = xdp_page; | |
562 | } | |
563 | ||
f6b10209 JW |
564 | xdp.data_hard_start = buf + VIRTNET_RX_PAD + vi->hdr_len; |
565 | xdp.data = xdp.data_hard_start + xdp_headroom; | |
de8f3a83 | 566 | xdp_set_data_meta_invalid(&xdp); |
0354e4d1 | 567 | xdp.data_end = xdp.data + len; |
f6b10209 | 568 | orig_data = xdp.data; |
0354e4d1 JF |
569 | act = bpf_prog_run_xdp(xdp_prog, &xdp); |
570 | ||
bb91accf JW |
571 | switch (act) { |
572 | case XDP_PASS: | |
2de2f7f4 | 573 | /* Recalculate length in case bpf program changed it */ |
f6b10209 | 574 | delta = orig_data - xdp.data; |
bb91accf JW |
575 | break; |
576 | case XDP_TX: | |
6e49f219 JDB |
577 | sent = __virtnet_xdp_xmit(vi, &xdp); |
578 | if (unlikely(!sent)) { | |
0354e4d1 | 579 | trace_xdp_exception(vi->dev, xdp_prog, act); |
6e49f219 JDB |
580 | goto err_xdp; |
581 | } | |
582 | *xdp_xmit = true; | |
186b3c99 JW |
583 | rcu_read_unlock(); |
584 | goto xdp_xmit; | |
585 | case XDP_REDIRECT: | |
586 | err = xdp_do_redirect(dev, &xdp, xdp_prog); | |
6e49f219 JDB |
587 | if (err) |
588 | goto err_xdp; | |
589 | *xdp_xmit = true; | |
bb91accf JW |
590 | rcu_read_unlock(); |
591 | goto xdp_xmit; | |
bb91accf | 592 | default: |
0354e4d1 JF |
593 | bpf_warn_invalid_xdp_action(act); |
594 | case XDP_ABORTED: | |
595 | trace_xdp_exception(vi->dev, xdp_prog, act); | |
596 | case XDP_DROP: | |
bb91accf JW |
597 | goto err_xdp; |
598 | } | |
599 | } | |
600 | rcu_read_unlock(); | |
601 | ||
f6b10209 JW |
602 | skb = build_skb(buf, buflen); |
603 | if (!skb) { | |
4941d472 | 604 | put_page(page); |
f6b10209 JW |
605 | goto err; |
606 | } | |
607 | skb_reserve(skb, headroom - delta); | |
608 | skb_put(skb, len + delta); | |
609 | if (!delta) { | |
610 | buf += header_offset; | |
611 | memcpy(skb_vnet_hdr(skb), buf, vi->hdr_len); | |
612 | } /* keep zeroed vnet hdr since packet was changed by bpf */ | |
613 | ||
614 | err: | |
f121159d | 615 | return skb; |
bb91accf JW |
616 | |
617 | err_xdp: | |
618 | rcu_read_unlock(); | |
619 | dev->stats.rx_dropped++; | |
4941d472 | 620 | put_page(page); |
bb91accf JW |
621 | xdp_xmit: |
622 | return NULL; | |
f121159d MT |
623 | } |
624 | ||
625 | static struct sk_buff *receive_big(struct net_device *dev, | |
946fa564 | 626 | struct virtnet_info *vi, |
f121159d MT |
627 | struct receive_queue *rq, |
628 | void *buf, | |
629 | unsigned int len) | |
630 | { | |
631 | struct page *page = buf; | |
5e229598 JW |
632 | struct sk_buff *skb = page_to_skb(vi, rq, page, 0, len, |
633 | PAGE_SIZE, true); | |
f600b690 | 634 | |
f121159d MT |
635 | if (unlikely(!skb)) |
636 | goto err; | |
637 | ||
638 | return skb; | |
639 | ||
640 | err: | |
641 | dev->stats.rx_dropped++; | |
642 | give_pages(rq, page); | |
643 | return NULL; | |
644 | } | |
645 | ||
8fc3b9e9 | 646 | static struct sk_buff *receive_mergeable(struct net_device *dev, |
fdd819b2 | 647 | struct virtnet_info *vi, |
8fc3b9e9 | 648 | struct receive_queue *rq, |
680557cf MT |
649 | void *buf, |
650 | void *ctx, | |
186b3c99 JW |
651 | unsigned int len, |
652 | bool *xdp_xmit) | |
9ab86bbc | 653 | { |
012873d0 MT |
654 | struct virtio_net_hdr_mrg_rxbuf *hdr = buf; |
655 | u16 num_buf = virtio16_to_cpu(vi->vdev, hdr->num_buffers); | |
8fc3b9e9 MT |
656 | struct page *page = virt_to_head_page(buf); |
657 | int offset = buf - page_address(page); | |
f600b690 JF |
658 | struct sk_buff *head_skb, *curr_skb; |
659 | struct bpf_prog *xdp_prog; | |
660 | unsigned int truesize; | |
4941d472 | 661 | unsigned int headroom = mergeable_ctx_to_headroom(ctx); |
186b3c99 | 662 | int err; |
6e49f219 | 663 | bool sent; |
f600b690 | 664 | |
56434a01 JF |
665 | head_skb = NULL; |
666 | ||
f600b690 JF |
667 | rcu_read_lock(); |
668 | xdp_prog = rcu_dereference(rq->xdp_prog); | |
669 | if (xdp_prog) { | |
72979a6c | 670 | struct page *xdp_page; |
0354e4d1 | 671 | struct xdp_buff xdp; |
0354e4d1 | 672 | void *data; |
f600b690 JF |
673 | u32 act; |
674 | ||
5cf0b8a6 JW |
675 | /* Transient failure which in theory could occur if |
676 | * in-flight packets from before XDP was enabled reach | |
677 | * the receive path after XDP is loaded. | |
678 | */ | |
679 | if (unlikely(hdr->hdr.gso_type)) | |
680 | goto err_xdp; | |
681 | ||
73b62bd0 | 682 | /* This happens when rx buffer size is underestimated */ |
4941d472 JW |
683 | if (unlikely(num_buf > 1 || |
684 | headroom < virtnet_get_headroom(vi))) { | |
72979a6c | 685 | /* linearize data for XDP */ |
56a86f84 | 686 | xdp_page = xdp_linearize_page(rq, &num_buf, |
4941d472 JW |
687 | page, offset, |
688 | VIRTIO_XDP_HEADROOM, | |
689 | &len); | |
72979a6c JF |
690 | if (!xdp_page) |
691 | goto err_xdp; | |
2de2f7f4 | 692 | offset = VIRTIO_XDP_HEADROOM; |
72979a6c JF |
693 | } else { |
694 | xdp_page = page; | |
f600b690 JF |
695 | } |
696 | ||
2de2f7f4 JF |
697 | /* Allow consuming headroom but reserve enough space to push |
698 | * the descriptor on if we get an XDP_TX return code. | |
699 | */ | |
0354e4d1 | 700 | data = page_address(xdp_page) + offset; |
2de2f7f4 | 701 | xdp.data_hard_start = data - VIRTIO_XDP_HEADROOM + vi->hdr_len; |
0354e4d1 | 702 | xdp.data = data + vi->hdr_len; |
de8f3a83 | 703 | xdp_set_data_meta_invalid(&xdp); |
0354e4d1 JF |
704 | xdp.data_end = xdp.data + (len - vi->hdr_len); |
705 | act = bpf_prog_run_xdp(xdp_prog, &xdp); | |
706 | ||
31240345 JW |
707 | if (act != XDP_PASS) |
708 | ewma_pkt_len_add(&rq->mrg_avg_pkt_len, len); | |
709 | ||
56434a01 JF |
710 | switch (act) { |
711 | case XDP_PASS: | |
2de2f7f4 JF |
712 | /* recalculate offset to account for any header |
713 | * adjustments. Note other cases do not build an | |
714 | * skb and avoid using offset | |
715 | */ | |
716 | offset = xdp.data - | |
717 | page_address(xdp_page) - vi->hdr_len; | |
718 | ||
1830f893 JW |
719 | /* We can only create skb based on xdp_page. */ |
720 | if (unlikely(xdp_page != page)) { | |
721 | rcu_read_unlock(); | |
722 | put_page(page); | |
723 | head_skb = page_to_skb(vi, rq, xdp_page, | |
5e229598 JW |
724 | offset, len, |
725 | PAGE_SIZE, false); | |
1830f893 JW |
726 | return head_skb; |
727 | } | |
56434a01 JF |
728 | break; |
729 | case XDP_TX: | |
6e49f219 JDB |
730 | sent = __virtnet_xdp_xmit(vi, &xdp); |
731 | if (unlikely(!sent)) { | |
0354e4d1 | 732 | trace_xdp_exception(vi->dev, xdp_prog, act); |
6e49f219 JDB |
733 | if (unlikely(xdp_page != page)) |
734 | put_page(xdp_page); | |
735 | goto err_xdp; | |
736 | } | |
737 | *xdp_xmit = true; | |
72979a6c | 738 | if (unlikely(xdp_page != page)) |
62b92b48 | 739 | put_page(page); |
56434a01 JF |
740 | rcu_read_unlock(); |
741 | goto xdp_xmit; | |
186b3c99 JW |
742 | case XDP_REDIRECT: |
743 | err = xdp_do_redirect(dev, &xdp, xdp_prog); | |
dd543797 | 744 | if (!err) |
186b3c99 JW |
745 | *xdp_xmit = true; |
746 | rcu_read_unlock(); | |
747 | goto xdp_xmit; | |
56434a01 | 748 | default: |
0354e4d1 JF |
749 | bpf_warn_invalid_xdp_action(act); |
750 | case XDP_ABORTED: | |
751 | trace_xdp_exception(vi->dev, xdp_prog, act); | |
752 | case XDP_DROP: | |
72979a6c JF |
753 | if (unlikely(xdp_page != page)) |
754 | __free_pages(xdp_page, 0); | |
f600b690 | 755 | goto err_xdp; |
56434a01 | 756 | } |
f600b690 JF |
757 | } |
758 | rcu_read_unlock(); | |
ab7db917 | 759 | |
28b39bc7 JW |
760 | truesize = mergeable_ctx_to_truesize(ctx); |
761 | if (unlikely(len > truesize)) { | |
56da5fd0 | 762 | pr_debug("%s: rx error: len %u exceeds truesize %lu\n", |
680557cf MT |
763 | dev->name, len, (unsigned long)ctx); |
764 | dev->stats.rx_length_errors++; | |
765 | goto err_skb; | |
766 | } | |
28b39bc7 | 767 | |
5e229598 | 768 | head_skb = page_to_skb(vi, rq, page, offset, len, truesize, !xdp_prog); |
f600b690 | 769 | curr_skb = head_skb; |
9ab86bbc | 770 | |
8fc3b9e9 MT |
771 | if (unlikely(!curr_skb)) |
772 | goto err_skb; | |
9ab86bbc | 773 | while (--num_buf) { |
8fc3b9e9 MT |
774 | int num_skb_frags; |
775 | ||
680557cf | 776 | buf = virtqueue_get_buf_ctx(rq->vq, &len, &ctx); |
03e9f8a0 | 777 | if (unlikely(!buf)) { |
8fc3b9e9 | 778 | pr_debug("%s: rx error: %d buffers out of %d missing\n", |
fdd819b2 | 779 | dev->name, num_buf, |
012873d0 MT |
780 | virtio16_to_cpu(vi->vdev, |
781 | hdr->num_buffers)); | |
8fc3b9e9 MT |
782 | dev->stats.rx_length_errors++; |
783 | goto err_buf; | |
3f2c31d9 | 784 | } |
8fc3b9e9 MT |
785 | |
786 | page = virt_to_head_page(buf); | |
28b39bc7 JW |
787 | |
788 | truesize = mergeable_ctx_to_truesize(ctx); | |
789 | if (unlikely(len > truesize)) { | |
56da5fd0 | 790 | pr_debug("%s: rx error: len %u exceeds truesize %lu\n", |
680557cf MT |
791 | dev->name, len, (unsigned long)ctx); |
792 | dev->stats.rx_length_errors++; | |
793 | goto err_skb; | |
794 | } | |
8fc3b9e9 MT |
795 | |
796 | num_skb_frags = skb_shinfo(curr_skb)->nr_frags; | |
2613af0e MD |
797 | if (unlikely(num_skb_frags == MAX_SKB_FRAGS)) { |
798 | struct sk_buff *nskb = alloc_skb(0, GFP_ATOMIC); | |
8fc3b9e9 MT |
799 | |
800 | if (unlikely(!nskb)) | |
801 | goto err_skb; | |
2613af0e MD |
802 | if (curr_skb == head_skb) |
803 | skb_shinfo(curr_skb)->frag_list = nskb; | |
804 | else | |
805 | curr_skb->next = nskb; | |
806 | curr_skb = nskb; | |
807 | head_skb->truesize += nskb->truesize; | |
808 | num_skb_frags = 0; | |
809 | } | |
810 | if (curr_skb != head_skb) { | |
811 | head_skb->data_len += len; | |
812 | head_skb->len += len; | |
fb51879d | 813 | head_skb->truesize += truesize; |
2613af0e | 814 | } |
8fc3b9e9 | 815 | offset = buf - page_address(page); |
ba275241 JW |
816 | if (skb_can_coalesce(curr_skb, num_skb_frags, page, offset)) { |
817 | put_page(page); | |
818 | skb_coalesce_rx_frag(curr_skb, num_skb_frags - 1, | |
fb51879d | 819 | len, truesize); |
ba275241 JW |
820 | } else { |
821 | skb_add_rx_frag(curr_skb, num_skb_frags, page, | |
fb51879d | 822 | offset, len, truesize); |
ba275241 | 823 | } |
8fc3b9e9 MT |
824 | } |
825 | ||
5377d758 | 826 | ewma_pkt_len_add(&rq->mrg_avg_pkt_len, head_skb->len); |
8fc3b9e9 MT |
827 | return head_skb; |
828 | ||
f600b690 JF |
829 | err_xdp: |
830 | rcu_read_unlock(); | |
8fc3b9e9 MT |
831 | err_skb: |
832 | put_page(page); | |
26f1676b | 833 | while (num_buf-- > 1) { |
680557cf MT |
834 | buf = virtqueue_get_buf(rq->vq, &len); |
835 | if (unlikely(!buf)) { | |
8fc3b9e9 MT |
836 | pr_debug("%s: rx error: %d buffers missing\n", |
837 | dev->name, num_buf); | |
838 | dev->stats.rx_length_errors++; | |
839 | break; | |
840 | } | |
680557cf | 841 | page = virt_to_head_page(buf); |
8fc3b9e9 | 842 | put_page(page); |
9ab86bbc | 843 | } |
8fc3b9e9 MT |
844 | err_buf: |
845 | dev->stats.rx_dropped++; | |
846 | dev_kfree_skb(head_skb); | |
56434a01 | 847 | xdp_xmit: |
8fc3b9e9 | 848 | return NULL; |
9ab86bbc SM |
849 | } |
850 | ||
61845d20 | 851 | static int receive_buf(struct virtnet_info *vi, struct receive_queue *rq, |
186b3c99 | 852 | void *buf, unsigned int len, void **ctx, bool *xdp_xmit) |
9ab86bbc | 853 | { |
e9d7417b | 854 | struct net_device *dev = vi->dev; |
9ab86bbc | 855 | struct sk_buff *skb; |
012873d0 | 856 | struct virtio_net_hdr_mrg_rxbuf *hdr; |
61845d20 | 857 | int ret; |
3f2c31d9 | 858 | |
bcff3162 | 859 | if (unlikely(len < vi->hdr_len + ETH_HLEN)) { |
9ab86bbc SM |
860 | pr_debug("%s: short packet %i\n", dev->name, len); |
861 | dev->stats.rx_length_errors++; | |
ab7db917 | 862 | if (vi->mergeable_rx_bufs) { |
680557cf | 863 | put_page(virt_to_head_page(buf)); |
ab7db917 | 864 | } else if (vi->big_packets) { |
98bfd23c | 865 | give_pages(rq, buf); |
ab7db917 | 866 | } else { |
f6b10209 | 867 | put_page(virt_to_head_page(buf)); |
ab7db917 | 868 | } |
61845d20 | 869 | return 0; |
9ab86bbc | 870 | } |
3f2c31d9 | 871 | |
f121159d | 872 | if (vi->mergeable_rx_bufs) |
186b3c99 | 873 | skb = receive_mergeable(dev, vi, rq, buf, ctx, len, xdp_xmit); |
f121159d | 874 | else if (vi->big_packets) |
946fa564 | 875 | skb = receive_big(dev, vi, rq, buf, len); |
f121159d | 876 | else |
186b3c99 | 877 | skb = receive_small(dev, vi, rq, buf, ctx, len, xdp_xmit); |
f121159d MT |
878 | |
879 | if (unlikely(!skb)) | |
61845d20 | 880 | return 0; |
3f2c31d9 | 881 | |
9ab86bbc | 882 | hdr = skb_vnet_hdr(skb); |
3fa2a1df | 883 | |
61845d20 | 884 | ret = skb->len; |
296f96fc | 885 | |
e858fae2 | 886 | if (hdr->hdr.flags & VIRTIO_NET_HDR_F_DATA_VALID) |
10a8d94a | 887 | skb->ip_summed = CHECKSUM_UNNECESSARY; |
296f96fc | 888 | |
e858fae2 MR |
889 | if (virtio_net_hdr_to_skb(skb, &hdr->hdr, |
890 | virtio_is_little_endian(vi->vdev))) { | |
891 | net_warn_ratelimited("%s: bad gso: type: %u, size: %u\n", | |
892 | dev->name, hdr->hdr.gso_type, | |
893 | hdr->hdr.gso_size); | |
894 | goto frame_err; | |
296f96fc RR |
895 | } |
896 | ||
d1dc06dc MR |
897 | skb->protocol = eth_type_trans(skb, dev); |
898 | pr_debug("Receiving skb proto 0x%04x len %i type %i\n", | |
899 | ntohs(skb->protocol), skb->len, skb->pkt_type); | |
900 | ||
0fbd050a | 901 | napi_gro_receive(&rq->napi, skb); |
61845d20 | 902 | return ret; |
296f96fc RR |
903 | |
904 | frame_err: | |
905 | dev->stats.rx_frame_errors++; | |
296f96fc | 906 | dev_kfree_skb(skb); |
61845d20 | 907 | return 0; |
296f96fc RR |
908 | } |
909 | ||
192f68cf JW |
910 | /* Unlike mergeable buffers, all buffers are allocated to the |
911 | * same size, except for the headroom. For this reason we do | |
912 | * not need to use mergeable_len_to_ctx here - it is enough | |
913 | * to store the headroom as the context ignoring the truesize. | |
914 | */ | |
946fa564 MT |
915 | static int add_recvbuf_small(struct virtnet_info *vi, struct receive_queue *rq, |
916 | gfp_t gfp) | |
296f96fc | 917 | { |
f6b10209 JW |
918 | struct page_frag *alloc_frag = &rq->alloc_frag; |
919 | char *buf; | |
2de2f7f4 | 920 | unsigned int xdp_headroom = virtnet_get_headroom(vi); |
192f68cf | 921 | void *ctx = (void *)(unsigned long)xdp_headroom; |
f6b10209 | 922 | int len = vi->hdr_len + VIRTNET_RX_PAD + GOOD_PACKET_LEN + xdp_headroom; |
9ab86bbc | 923 | int err; |
3f2c31d9 | 924 | |
f6b10209 JW |
925 | len = SKB_DATA_ALIGN(len) + |
926 | SKB_DATA_ALIGN(sizeof(struct skb_shared_info)); | |
927 | if (unlikely(!skb_page_frag_refill(len, alloc_frag, gfp))) | |
9ab86bbc | 928 | return -ENOMEM; |
296f96fc | 929 | |
f6b10209 JW |
930 | buf = (char *)page_address(alloc_frag->page) + alloc_frag->offset; |
931 | get_page(alloc_frag->page); | |
932 | alloc_frag->offset += len; | |
933 | sg_init_one(rq->sg, buf + VIRTNET_RX_PAD + xdp_headroom, | |
934 | vi->hdr_len + GOOD_PACKET_LEN); | |
192f68cf | 935 | err = virtqueue_add_inbuf_ctx(rq->vq, rq->sg, 1, buf, ctx, gfp); |
9ab86bbc | 936 | if (err < 0) |
f6b10209 | 937 | put_page(virt_to_head_page(buf)); |
9ab86bbc SM |
938 | return err; |
939 | } | |
97402b96 | 940 | |
012873d0 MT |
941 | static int add_recvbuf_big(struct virtnet_info *vi, struct receive_queue *rq, |
942 | gfp_t gfp) | |
9ab86bbc | 943 | { |
9ab86bbc SM |
944 | struct page *first, *list = NULL; |
945 | char *p; | |
946 | int i, err, offset; | |
947 | ||
a5835440 RR |
948 | sg_init_table(rq->sg, MAX_SKB_FRAGS + 2); |
949 | ||
e9d7417b | 950 | /* page in rq->sg[MAX_SKB_FRAGS + 1] is list tail */ |
9ab86bbc | 951 | for (i = MAX_SKB_FRAGS + 1; i > 1; --i) { |
e9d7417b | 952 | first = get_a_page(rq, gfp); |
9ab86bbc SM |
953 | if (!first) { |
954 | if (list) | |
e9d7417b | 955 | give_pages(rq, list); |
9ab86bbc | 956 | return -ENOMEM; |
97402b96 | 957 | } |
e9d7417b | 958 | sg_set_buf(&rq->sg[i], page_address(first), PAGE_SIZE); |
97402b96 | 959 | |
9ab86bbc SM |
960 | /* chain new page in list head to match sg */ |
961 | first->private = (unsigned long)list; | |
962 | list = first; | |
963 | } | |
296f96fc | 964 | |
e9d7417b | 965 | first = get_a_page(rq, gfp); |
9ab86bbc | 966 | if (!first) { |
e9d7417b | 967 | give_pages(rq, list); |
9ab86bbc SM |
968 | return -ENOMEM; |
969 | } | |
970 | p = page_address(first); | |
971 | ||
e9d7417b | 972 | /* rq->sg[0], rq->sg[1] share the same page */ |
012873d0 MT |
973 | /* a separated rq->sg[0] for header - required in case !any_header_sg */ |
974 | sg_set_buf(&rq->sg[0], p, vi->hdr_len); | |
9ab86bbc | 975 | |
e9d7417b | 976 | /* rq->sg[1] for data packet, from offset */ |
9ab86bbc | 977 | offset = sizeof(struct padded_vnet_hdr); |
e9d7417b | 978 | sg_set_buf(&rq->sg[1], p + offset, PAGE_SIZE - offset); |
9ab86bbc SM |
979 | |
980 | /* chain first in list head */ | |
981 | first->private = (unsigned long)list; | |
9dc7b9e4 RR |
982 | err = virtqueue_add_inbuf(rq->vq, rq->sg, MAX_SKB_FRAGS + 2, |
983 | first, gfp); | |
9ab86bbc | 984 | if (err < 0) |
e9d7417b | 985 | give_pages(rq, first); |
9ab86bbc SM |
986 | |
987 | return err; | |
296f96fc RR |
988 | } |
989 | ||
d85b758f MT |
990 | static unsigned int get_mergeable_buf_len(struct receive_queue *rq, |
991 | struct ewma_pkt_len *avg_pkt_len) | |
3f2c31d9 | 992 | { |
ab7db917 | 993 | const size_t hdr_len = sizeof(struct virtio_net_hdr_mrg_rxbuf); |
fbf28d78 MD |
994 | unsigned int len; |
995 | ||
5377d758 | 996 | len = hdr_len + clamp_t(unsigned int, ewma_pkt_len_read(avg_pkt_len), |
f0c3192c | 997 | rq->min_buf_len, PAGE_SIZE - hdr_len); |
e377fcc8 | 998 | return ALIGN(len, L1_CACHE_BYTES); |
fbf28d78 MD |
999 | } |
1000 | ||
2de2f7f4 JF |
1001 | static int add_recvbuf_mergeable(struct virtnet_info *vi, |
1002 | struct receive_queue *rq, gfp_t gfp) | |
fbf28d78 | 1003 | { |
fb51879d | 1004 | struct page_frag *alloc_frag = &rq->alloc_frag; |
2de2f7f4 | 1005 | unsigned int headroom = virtnet_get_headroom(vi); |
fb51879d | 1006 | char *buf; |
680557cf | 1007 | void *ctx; |
3f2c31d9 | 1008 | int err; |
fb51879d | 1009 | unsigned int len, hole; |
3f2c31d9 | 1010 | |
d85b758f | 1011 | len = get_mergeable_buf_len(rq, &rq->mrg_avg_pkt_len); |
2de2f7f4 | 1012 | if (unlikely(!skb_page_frag_refill(len + headroom, alloc_frag, gfp))) |
9ab86bbc | 1013 | return -ENOMEM; |
ab7db917 | 1014 | |
fb51879d | 1015 | buf = (char *)page_address(alloc_frag->page) + alloc_frag->offset; |
2de2f7f4 | 1016 | buf += headroom; /* advance address leaving hole at front of pkt */ |
fb51879d | 1017 | get_page(alloc_frag->page); |
2de2f7f4 | 1018 | alloc_frag->offset += len + headroom; |
fb51879d | 1019 | hole = alloc_frag->size - alloc_frag->offset; |
2de2f7f4 | 1020 | if (hole < len + headroom) { |
ab7db917 MD |
1021 | /* To avoid internal fragmentation, if there is very likely not |
1022 | * enough space for another buffer, add the remaining space to | |
1daa8790 | 1023 | * the current buffer. |
ab7db917 | 1024 | */ |
fb51879d MD |
1025 | len += hole; |
1026 | alloc_frag->offset += hole; | |
1027 | } | |
3f2c31d9 | 1028 | |
fb51879d | 1029 | sg_init_one(rq->sg, buf, len); |
29fda25a | 1030 | ctx = mergeable_len_to_ctx(len, headroom); |
680557cf | 1031 | err = virtqueue_add_inbuf_ctx(rq->vq, rq->sg, 1, buf, ctx, gfp); |
9ab86bbc | 1032 | if (err < 0) |
2613af0e | 1033 | put_page(virt_to_head_page(buf)); |
3f2c31d9 | 1034 | |
9ab86bbc SM |
1035 | return err; |
1036 | } | |
3f2c31d9 | 1037 | |
b2baed69 RR |
1038 | /* |
1039 | * Returns false if we couldn't fill entirely (OOM). | |
1040 | * | |
1041 | * Normally run in the receive path, but can also be run from ndo_open | |
1042 | * before we're receiving packets, or from refill_work which is | |
1043 | * careful to disable receiving (using napi_disable). | |
1044 | */ | |
946fa564 MT |
1045 | static bool try_fill_recv(struct virtnet_info *vi, struct receive_queue *rq, |
1046 | gfp_t gfp) | |
9ab86bbc SM |
1047 | { |
1048 | int err; | |
1788f495 | 1049 | bool oom; |
3f2c31d9 | 1050 | |
9ab86bbc SM |
1051 | do { |
1052 | if (vi->mergeable_rx_bufs) | |
2de2f7f4 | 1053 | err = add_recvbuf_mergeable(vi, rq, gfp); |
9ab86bbc | 1054 | else if (vi->big_packets) |
012873d0 | 1055 | err = add_recvbuf_big(vi, rq, gfp); |
9ab86bbc | 1056 | else |
946fa564 | 1057 | err = add_recvbuf_small(vi, rq, gfp); |
3f2c31d9 | 1058 | |
1788f495 | 1059 | oom = err == -ENOMEM; |
9ed4cb07 | 1060 | if (err) |
3f2c31d9 | 1061 | break; |
b7dfde95 | 1062 | } while (rq->vq->num_free); |
681daee2 | 1063 | virtqueue_kick(rq->vq); |
3161e453 | 1064 | return !oom; |
3f2c31d9 MM |
1065 | } |
1066 | ||
18445c4d | 1067 | static void skb_recv_done(struct virtqueue *rvq) |
296f96fc RR |
1068 | { |
1069 | struct virtnet_info *vi = rvq->vdev->priv; | |
986a4f4d | 1070 | struct receive_queue *rq = &vi->rq[vq2rxq(rvq)]; |
e9d7417b | 1071 | |
e4e8452a | 1072 | virtqueue_napi_schedule(&rq->napi, rvq); |
296f96fc RR |
1073 | } |
1074 | ||
e4e8452a | 1075 | static void virtnet_napi_enable(struct virtqueue *vq, struct napi_struct *napi) |
3e9d08ec | 1076 | { |
e4e8452a | 1077 | napi_enable(napi); |
3e9d08ec BR |
1078 | |
1079 | /* If all buffers were filled by other side before we napi_enabled, we | |
e4e8452a WB |
1080 | * won't get another interrupt, so process any outstanding packets now. |
1081 | * Call local_bh_enable after to trigger softIRQ processing. | |
1082 | */ | |
1083 | local_bh_disable(); | |
1084 | virtqueue_napi_schedule(napi, vq); | |
1085 | local_bh_enable(); | |
3e9d08ec BR |
1086 | } |
1087 | ||
b92f1e67 WB |
1088 | static void virtnet_napi_tx_enable(struct virtnet_info *vi, |
1089 | struct virtqueue *vq, | |
1090 | struct napi_struct *napi) | |
1091 | { | |
1092 | if (!napi->weight) | |
1093 | return; | |
1094 | ||
1095 | /* Tx napi touches cachelines on the cpu handling tx interrupts. Only | |
1096 | * enable the feature if this is likely affine with the transmit path. | |
1097 | */ | |
1098 | if (!vi->affinity_hint_set) { | |
1099 | napi->weight = 0; | |
1100 | return; | |
1101 | } | |
1102 | ||
1103 | return virtnet_napi_enable(vq, napi); | |
1104 | } | |
1105 | ||
78a57b48 WB |
1106 | static void virtnet_napi_tx_disable(struct napi_struct *napi) |
1107 | { | |
1108 | if (napi->weight) | |
1109 | napi_disable(napi); | |
1110 | } | |
1111 | ||
3161e453 RR |
1112 | static void refill_work(struct work_struct *work) |
1113 | { | |
e9d7417b JW |
1114 | struct virtnet_info *vi = |
1115 | container_of(work, struct virtnet_info, refill.work); | |
3161e453 | 1116 | bool still_empty; |
986a4f4d JW |
1117 | int i; |
1118 | ||
55257d72 | 1119 | for (i = 0; i < vi->curr_queue_pairs; i++) { |
986a4f4d | 1120 | struct receive_queue *rq = &vi->rq[i]; |
3161e453 | 1121 | |
986a4f4d | 1122 | napi_disable(&rq->napi); |
946fa564 | 1123 | still_empty = !try_fill_recv(vi, rq, GFP_KERNEL); |
e4e8452a | 1124 | virtnet_napi_enable(rq->vq, &rq->napi); |
3161e453 | 1125 | |
986a4f4d JW |
1126 | /* In theory, this can happen: if we don't get any buffers in |
1127 | * we will *never* try to fill again. | |
1128 | */ | |
1129 | if (still_empty) | |
1130 | schedule_delayed_work(&vi->refill, HZ/2); | |
1131 | } | |
3161e453 RR |
1132 | } |
1133 | ||
186b3c99 | 1134 | static int virtnet_receive(struct receive_queue *rq, int budget, bool *xdp_xmit) |
296f96fc | 1135 | { |
e9d7417b | 1136 | struct virtnet_info *vi = rq->vq->vdev->priv; |
61845d20 | 1137 | unsigned int len, received = 0, bytes = 0; |
9ab86bbc | 1138 | void *buf; |
61845d20 | 1139 | struct virtnet_stats *stats = this_cpu_ptr(vi->stats); |
296f96fc | 1140 | |
192f68cf | 1141 | if (!vi->big_packets || vi->mergeable_rx_bufs) { |
680557cf MT |
1142 | void *ctx; |
1143 | ||
1144 | while (received < budget && | |
1145 | (buf = virtqueue_get_buf_ctx(rq->vq, &len, &ctx))) { | |
186b3c99 | 1146 | bytes += receive_buf(vi, rq, buf, len, ctx, xdp_xmit); |
680557cf MT |
1147 | received++; |
1148 | } | |
1149 | } else { | |
1150 | while (received < budget && | |
1151 | (buf = virtqueue_get_buf(rq->vq, &len)) != NULL) { | |
186b3c99 | 1152 | bytes += receive_buf(vi, rq, buf, len, NULL, xdp_xmit); |
680557cf MT |
1153 | received++; |
1154 | } | |
296f96fc RR |
1155 | } |
1156 | ||
be121f46 | 1157 | if (rq->vq->num_free > virtqueue_get_vring_size(rq->vq) / 2) { |
946fa564 | 1158 | if (!try_fill_recv(vi, rq, GFP_ATOMIC)) |
3b07e9ca | 1159 | schedule_delayed_work(&vi->refill, 0); |
3161e453 | 1160 | } |
296f96fc | 1161 | |
61845d20 JW |
1162 | u64_stats_update_begin(&stats->rx_syncp); |
1163 | stats->rx_bytes += bytes; | |
1164 | stats->rx_packets += received; | |
1165 | u64_stats_update_end(&stats->rx_syncp); | |
1166 | ||
2ffa7598 JW |
1167 | return received; |
1168 | } | |
1169 | ||
ea7735d9 WB |
1170 | static void free_old_xmit_skbs(struct send_queue *sq) |
1171 | { | |
1172 | struct sk_buff *skb; | |
1173 | unsigned int len; | |
1174 | struct virtnet_info *vi = sq->vq->vdev->priv; | |
1175 | struct virtnet_stats *stats = this_cpu_ptr(vi->stats); | |
1176 | unsigned int packets = 0; | |
1177 | unsigned int bytes = 0; | |
1178 | ||
1179 | while ((skb = virtqueue_get_buf(sq->vq, &len)) != NULL) { | |
1180 | pr_debug("Sent skb %p\n", skb); | |
1181 | ||
1182 | bytes += skb->len; | |
1183 | packets++; | |
1184 | ||
dadc0736 | 1185 | dev_consume_skb_any(skb); |
ea7735d9 WB |
1186 | } |
1187 | ||
1188 | /* Avoid overhead when no packets have been processed | |
1189 | * happens when called speculatively from start_xmit. | |
1190 | */ | |
1191 | if (!packets) | |
1192 | return; | |
1193 | ||
1194 | u64_stats_update_begin(&stats->tx_syncp); | |
1195 | stats->tx_bytes += bytes; | |
1196 | stats->tx_packets += packets; | |
1197 | u64_stats_update_end(&stats->tx_syncp); | |
1198 | } | |
1199 | ||
d8c6e7d4 TM |
1200 | static bool is_xdp_raw_buffer_queue(struct virtnet_info *vi, int q) |
1201 | { | |
1202 | if (q < (vi->curr_queue_pairs - vi->xdp_queue_pairs)) | |
1203 | return false; | |
1204 | else if (q < vi->curr_queue_pairs) | |
1205 | return true; | |
1206 | else | |
1207 | return false; | |
1208 | } | |
1209 | ||
7b0411ef WB |
1210 | static void virtnet_poll_cleantx(struct receive_queue *rq) |
1211 | { | |
1212 | struct virtnet_info *vi = rq->vq->vdev->priv; | |
1213 | unsigned int index = vq2rxq(rq->vq); | |
1214 | struct send_queue *sq = &vi->sq[index]; | |
1215 | struct netdev_queue *txq = netdev_get_tx_queue(vi->dev, index); | |
1216 | ||
d8c6e7d4 | 1217 | if (!sq->napi.weight || is_xdp_raw_buffer_queue(vi, index)) |
7b0411ef WB |
1218 | return; |
1219 | ||
1220 | if (__netif_tx_trylock(txq)) { | |
1221 | free_old_xmit_skbs(sq); | |
1222 | __netif_tx_unlock(txq); | |
1223 | } | |
1224 | ||
1225 | if (sq->vq->num_free >= 2 + MAX_SKB_FRAGS) | |
1226 | netif_tx_wake_queue(txq); | |
1227 | } | |
1228 | ||
2ffa7598 JW |
1229 | static int virtnet_poll(struct napi_struct *napi, int budget) |
1230 | { | |
1231 | struct receive_queue *rq = | |
1232 | container_of(napi, struct receive_queue, napi); | |
3c93c06c JW |
1233 | struct virtnet_info *vi = rq->vq->vdev->priv; |
1234 | struct send_queue *sq; | |
1235 | unsigned int received, qp; | |
186b3c99 | 1236 | bool xdp_xmit = false; |
2ffa7598 | 1237 | |
7b0411ef WB |
1238 | virtnet_poll_cleantx(rq); |
1239 | ||
186b3c99 | 1240 | received = virtnet_receive(rq, budget, &xdp_xmit); |
2ffa7598 | 1241 | |
8329d98e | 1242 | /* Out of packets? */ |
e4e8452a WB |
1243 | if (received < budget) |
1244 | virtqueue_napi_complete(napi, rq->vq, received); | |
296f96fc | 1245 | |
3c93c06c JW |
1246 | if (xdp_xmit) { |
1247 | qp = vi->curr_queue_pairs - vi->xdp_queue_pairs + | |
1248 | smp_processor_id(); | |
1249 | sq = &vi->sq[qp]; | |
1250 | virtqueue_kick(sq->vq); | |
186b3c99 | 1251 | xdp_do_flush_map(); |
3c93c06c | 1252 | } |
186b3c99 | 1253 | |
296f96fc RR |
1254 | return received; |
1255 | } | |
1256 | ||
986a4f4d JW |
1257 | static int virtnet_open(struct net_device *dev) |
1258 | { | |
1259 | struct virtnet_info *vi = netdev_priv(dev); | |
1260 | int i; | |
1261 | ||
e4166625 JW |
1262 | for (i = 0; i < vi->max_queue_pairs; i++) { |
1263 | if (i < vi->curr_queue_pairs) | |
1264 | /* Make sure we have some buffers: if oom use wq. */ | |
946fa564 | 1265 | if (!try_fill_recv(vi, &vi->rq[i], GFP_KERNEL)) |
e4166625 | 1266 | schedule_delayed_work(&vi->refill, 0); |
e4e8452a | 1267 | virtnet_napi_enable(vi->rq[i].vq, &vi->rq[i].napi); |
b92f1e67 | 1268 | virtnet_napi_tx_enable(vi, vi->sq[i].vq, &vi->sq[i].napi); |
986a4f4d JW |
1269 | } |
1270 | ||
1271 | return 0; | |
1272 | } | |
1273 | ||
b92f1e67 WB |
1274 | static int virtnet_poll_tx(struct napi_struct *napi, int budget) |
1275 | { | |
1276 | struct send_queue *sq = container_of(napi, struct send_queue, napi); | |
1277 | struct virtnet_info *vi = sq->vq->vdev->priv; | |
d8c6e7d4 TM |
1278 | unsigned int index = vq2txq(sq->vq); |
1279 | struct netdev_queue *txq; | |
b92f1e67 | 1280 | |
d8c6e7d4 TM |
1281 | if (unlikely(is_xdp_raw_buffer_queue(vi, index))) { |
1282 | /* We don't need to enable cb for XDP */ | |
1283 | napi_complete_done(napi, 0); | |
1284 | return 0; | |
1285 | } | |
1286 | ||
1287 | txq = netdev_get_tx_queue(vi->dev, index); | |
b92f1e67 WB |
1288 | __netif_tx_lock(txq, raw_smp_processor_id()); |
1289 | free_old_xmit_skbs(sq); | |
1290 | __netif_tx_unlock(txq); | |
1291 | ||
1292 | virtqueue_napi_complete(napi, sq->vq, 0); | |
1293 | ||
1294 | if (sq->vq->num_free >= 2 + MAX_SKB_FRAGS) | |
1295 | netif_tx_wake_queue(txq); | |
1296 | ||
1297 | return 0; | |
1298 | } | |
1299 | ||
e9d7417b | 1300 | static int xmit_skb(struct send_queue *sq, struct sk_buff *skb) |
296f96fc | 1301 | { |
012873d0 | 1302 | struct virtio_net_hdr_mrg_rxbuf *hdr; |
296f96fc | 1303 | const unsigned char *dest = ((struct ethhdr *)skb->data)->h_dest; |
e9d7417b | 1304 | struct virtnet_info *vi = sq->vq->vdev->priv; |
e2fcad58 | 1305 | int num_sg; |
012873d0 | 1306 | unsigned hdr_len = vi->hdr_len; |
e7428e95 | 1307 | bool can_push; |
296f96fc | 1308 | |
e174961c | 1309 | pr_debug("%s: xmit %p %pM\n", vi->dev->name, skb, dest); |
e7428e95 MT |
1310 | |
1311 | can_push = vi->any_header_sg && | |
1312 | !((unsigned long)skb->data & (__alignof__(*hdr) - 1)) && | |
1313 | !skb_header_cloned(skb) && skb_headroom(skb) >= hdr_len; | |
1314 | /* Even if we can, don't push here yet as this would skew | |
1315 | * csum_start offset below. */ | |
1316 | if (can_push) | |
012873d0 | 1317 | hdr = (struct virtio_net_hdr_mrg_rxbuf *)(skb->data - hdr_len); |
e7428e95 MT |
1318 | else |
1319 | hdr = skb_vnet_hdr(skb); | |
296f96fc | 1320 | |
e858fae2 | 1321 | if (virtio_net_hdr_from_skb(skb, &hdr->hdr, |
34d736aa WB |
1322 | virtio_is_little_endian(vi->vdev), false, |
1323 | 0)) | |
e858fae2 | 1324 | BUG(); |
296f96fc | 1325 | |
3f2c31d9 | 1326 | if (vi->mergeable_rx_bufs) |
012873d0 | 1327 | hdr->num_buffers = 0; |
3f2c31d9 | 1328 | |
547c890c | 1329 | sg_init_table(sq->sg, skb_shinfo(skb)->nr_frags + (can_push ? 1 : 2)); |
e7428e95 MT |
1330 | if (can_push) { |
1331 | __skb_push(skb, hdr_len); | |
1332 | num_sg = skb_to_sgvec(skb, sq->sg, 0, skb->len); | |
e2fcad58 JD |
1333 | if (unlikely(num_sg < 0)) |
1334 | return num_sg; | |
e7428e95 MT |
1335 | /* Pull header back to avoid skew in tx bytes calculations. */ |
1336 | __skb_pull(skb, hdr_len); | |
1337 | } else { | |
1338 | sg_set_buf(sq->sg, hdr, hdr_len); | |
e2fcad58 JD |
1339 | num_sg = skb_to_sgvec(skb, sq->sg + 1, 0, skb->len); |
1340 | if (unlikely(num_sg < 0)) | |
1341 | return num_sg; | |
1342 | num_sg++; | |
e7428e95 | 1343 | } |
9dc7b9e4 | 1344 | return virtqueue_add_outbuf(sq->vq, sq->sg, num_sg, skb, GFP_ATOMIC); |
11a3a154 RR |
1345 | } |
1346 | ||
424efe9c | 1347 | static netdev_tx_t start_xmit(struct sk_buff *skb, struct net_device *dev) |
99ffc696 RR |
1348 | { |
1349 | struct virtnet_info *vi = netdev_priv(dev); | |
986a4f4d JW |
1350 | int qnum = skb_get_queue_mapping(skb); |
1351 | struct send_queue *sq = &vi->sq[qnum]; | |
9ed4cb07 | 1352 | int err; |
4b7fd2e6 MT |
1353 | struct netdev_queue *txq = netdev_get_tx_queue(dev, qnum); |
1354 | bool kick = !skb->xmit_more; | |
b92f1e67 | 1355 | bool use_napi = sq->napi.weight; |
2cb9c6ba | 1356 | |
2cb9c6ba | 1357 | /* Free up any pending old buffers before queueing new ones. */ |
e9d7417b | 1358 | free_old_xmit_skbs(sq); |
99ffc696 | 1359 | |
bdb12e0d WB |
1360 | if (use_napi && kick) |
1361 | virtqueue_enable_cb_delayed(sq->vq); | |
1362 | ||
074c3582 JK |
1363 | /* timestamp packet in software */ |
1364 | skb_tx_timestamp(skb); | |
1365 | ||
03f191ba | 1366 | /* Try to transmit */ |
b7dfde95 | 1367 | err = xmit_skb(sq, skb); |
48925e37 | 1368 | |
9ed4cb07 | 1369 | /* This should not happen! */ |
681daee2 | 1370 | if (unlikely(err)) { |
9ed4cb07 RR |
1371 | dev->stats.tx_fifo_errors++; |
1372 | if (net_ratelimit()) | |
1373 | dev_warn(&dev->dev, | |
b7dfde95 | 1374 | "Unexpected TXQ (%d) queue failure: %d\n", qnum, err); |
58eba97d | 1375 | dev->stats.tx_dropped++; |
85e94525 | 1376 | dev_kfree_skb_any(skb); |
58eba97d | 1377 | return NETDEV_TX_OK; |
296f96fc | 1378 | } |
03f191ba | 1379 | |
48925e37 | 1380 | /* Don't wait up for transmitted skbs to be freed. */ |
b92f1e67 WB |
1381 | if (!use_napi) { |
1382 | skb_orphan(skb); | |
1383 | nf_reset(skb); | |
1384 | } | |
48925e37 | 1385 | |
60302ff6 MT |
1386 | /* If running out of space, stop queue to avoid getting packets that we |
1387 | * are then unable to transmit. | |
1388 | * An alternative would be to force queuing layer to requeue the skb by | |
1389 | * returning NETDEV_TX_BUSY. However, NETDEV_TX_BUSY should not be | |
1390 | * returned in a normal path of operation: it means that driver is not | |
1391 | * maintaining the TX queue stop/start state properly, and causes | |
1392 | * the stack to do a non-trivial amount of useless work. | |
1393 | * Since most packets only take 1 or 2 ring slots, stopping the queue | |
1394 | * early means 16 slots are typically wasted. | |
d631b94e | 1395 | */ |
b7dfde95 | 1396 | if (sq->vq->num_free < 2+MAX_SKB_FRAGS) { |
986a4f4d | 1397 | netif_stop_subqueue(dev, qnum); |
b92f1e67 WB |
1398 | if (!use_napi && |
1399 | unlikely(!virtqueue_enable_cb_delayed(sq->vq))) { | |
48925e37 | 1400 | /* More just got used, free them then recheck. */ |
b7dfde95 LT |
1401 | free_old_xmit_skbs(sq); |
1402 | if (sq->vq->num_free >= 2+MAX_SKB_FRAGS) { | |
986a4f4d | 1403 | netif_start_subqueue(dev, qnum); |
e9d7417b | 1404 | virtqueue_disable_cb(sq->vq); |
48925e37 RR |
1405 | } |
1406 | } | |
99ffc696 | 1407 | } |
48925e37 | 1408 | |
4b7fd2e6 | 1409 | if (kick || netif_xmit_stopped(txq)) |
0b725a2c | 1410 | virtqueue_kick(sq->vq); |
296f96fc | 1411 | |
0b725a2c | 1412 | return NETDEV_TX_OK; |
c223a078 DM |
1413 | } |
1414 | ||
40cbfc37 AK |
1415 | /* |
1416 | * Send command via the control virtqueue and check status. Commands | |
1417 | * supported by the hypervisor, as indicated by feature bits, should | |
788a8b6d | 1418 | * never fail unless improperly formatted. |
40cbfc37 AK |
1419 | */ |
1420 | static bool virtnet_send_command(struct virtnet_info *vi, u8 class, u8 cmd, | |
d24bae32 | 1421 | struct scatterlist *out) |
40cbfc37 | 1422 | { |
f7bc9594 | 1423 | struct scatterlist *sgs[4], hdr, stat; |
d24bae32 | 1424 | unsigned out_num = 0, tmp; |
40cbfc37 AK |
1425 | |
1426 | /* Caller should know better */ | |
f7bc9594 | 1427 | BUG_ON(!virtio_has_feature(vi->vdev, VIRTIO_NET_F_CTRL_VQ)); |
40cbfc37 | 1428 | |
603343b9 MT |
1429 | vi->ctrl->status = ~0; |
1430 | vi->ctrl->hdr.class = class; | |
1431 | vi->ctrl->hdr.cmd = cmd; | |
f7bc9594 | 1432 | /* Add header */ |
603343b9 | 1433 | sg_init_one(&hdr, &vi->ctrl->hdr, sizeof(vi->ctrl->hdr)); |
f7bc9594 | 1434 | sgs[out_num++] = &hdr; |
40cbfc37 | 1435 | |
f7bc9594 RR |
1436 | if (out) |
1437 | sgs[out_num++] = out; | |
40cbfc37 | 1438 | |
f7bc9594 | 1439 | /* Add return status. */ |
603343b9 | 1440 | sg_init_one(&stat, &vi->ctrl->status, sizeof(vi->ctrl->status)); |
d24bae32 | 1441 | sgs[out_num] = &stat; |
40cbfc37 | 1442 | |
d24bae32 | 1443 | BUG_ON(out_num + 1 > ARRAY_SIZE(sgs)); |
a7c58146 | 1444 | virtqueue_add_sgs(vi->cvq, sgs, out_num, 1, vi, GFP_ATOMIC); |
40cbfc37 | 1445 | |
67975901 | 1446 | if (unlikely(!virtqueue_kick(vi->cvq))) |
603343b9 | 1447 | return vi->ctrl->status == VIRTIO_NET_OK; |
40cbfc37 AK |
1448 | |
1449 | /* Spin for a response, the kick causes an ioport write, trapping | |
1450 | * into the hypervisor, so the request should be handled immediately. | |
1451 | */ | |
047b9b94 HG |
1452 | while (!virtqueue_get_buf(vi->cvq, &tmp) && |
1453 | !virtqueue_is_broken(vi->cvq)) | |
40cbfc37 AK |
1454 | cpu_relax(); |
1455 | ||
603343b9 | 1456 | return vi->ctrl->status == VIRTIO_NET_OK; |
40cbfc37 AK |
1457 | } |
1458 | ||
9c46f6d4 AW |
1459 | static int virtnet_set_mac_address(struct net_device *dev, void *p) |
1460 | { | |
1461 | struct virtnet_info *vi = netdev_priv(dev); | |
1462 | struct virtio_device *vdev = vi->vdev; | |
f2f2c8b4 | 1463 | int ret; |
e37e2ff3 | 1464 | struct sockaddr *addr; |
7e58d5ae | 1465 | struct scatterlist sg; |
9c46f6d4 | 1466 | |
801822d1 | 1467 | addr = kmemdup(p, sizeof(*addr), GFP_KERNEL); |
e37e2ff3 AL |
1468 | if (!addr) |
1469 | return -ENOMEM; | |
e37e2ff3 AL |
1470 | |
1471 | ret = eth_prepare_mac_addr_change(dev, addr); | |
f2f2c8b4 | 1472 | if (ret) |
e37e2ff3 | 1473 | goto out; |
9c46f6d4 | 1474 | |
7e58d5ae AK |
1475 | if (virtio_has_feature(vdev, VIRTIO_NET_F_CTRL_MAC_ADDR)) { |
1476 | sg_init_one(&sg, addr->sa_data, dev->addr_len); | |
1477 | if (!virtnet_send_command(vi, VIRTIO_NET_CTRL_MAC, | |
d24bae32 | 1478 | VIRTIO_NET_CTRL_MAC_ADDR_SET, &sg)) { |
7e58d5ae AK |
1479 | dev_warn(&vdev->dev, |
1480 | "Failed to set mac address by vq command.\n"); | |
e37e2ff3 AL |
1481 | ret = -EINVAL; |
1482 | goto out; | |
7e58d5ae | 1483 | } |
7e93a02f MT |
1484 | } else if (virtio_has_feature(vdev, VIRTIO_NET_F_MAC) && |
1485 | !virtio_has_feature(vdev, VIRTIO_F_VERSION_1)) { | |
855e0c52 RR |
1486 | unsigned int i; |
1487 | ||
1488 | /* Naturally, this has an atomicity problem. */ | |
1489 | for (i = 0; i < dev->addr_len; i++) | |
1490 | virtio_cwrite8(vdev, | |
1491 | offsetof(struct virtio_net_config, mac) + | |
1492 | i, addr->sa_data[i]); | |
7e58d5ae AK |
1493 | } |
1494 | ||
1495 | eth_commit_mac_addr_change(dev, p); | |
e37e2ff3 | 1496 | ret = 0; |
9c46f6d4 | 1497 | |
e37e2ff3 AL |
1498 | out: |
1499 | kfree(addr); | |
1500 | return ret; | |
9c46f6d4 AW |
1501 | } |
1502 | ||
bc1f4470 | 1503 | static void virtnet_stats(struct net_device *dev, |
1504 | struct rtnl_link_stats64 *tot) | |
3fa2a1df | 1505 | { |
1506 | struct virtnet_info *vi = netdev_priv(dev); | |
1507 | int cpu; | |
1508 | unsigned int start; | |
1509 | ||
1510 | for_each_possible_cpu(cpu) { | |
58472a76 | 1511 | struct virtnet_stats *stats = per_cpu_ptr(vi->stats, cpu); |
3fa2a1df | 1512 | u64 tpackets, tbytes, rpackets, rbytes; |
1513 | ||
1514 | do { | |
57a7744e | 1515 | start = u64_stats_fetch_begin_irq(&stats->tx_syncp); |
3fa2a1df | 1516 | tpackets = stats->tx_packets; |
1517 | tbytes = stats->tx_bytes; | |
57a7744e | 1518 | } while (u64_stats_fetch_retry_irq(&stats->tx_syncp, start)); |
83a27052 ED |
1519 | |
1520 | do { | |
57a7744e | 1521 | start = u64_stats_fetch_begin_irq(&stats->rx_syncp); |
3fa2a1df | 1522 | rpackets = stats->rx_packets; |
1523 | rbytes = stats->rx_bytes; | |
57a7744e | 1524 | } while (u64_stats_fetch_retry_irq(&stats->rx_syncp, start)); |
3fa2a1df | 1525 | |
1526 | tot->rx_packets += rpackets; | |
1527 | tot->tx_packets += tpackets; | |
1528 | tot->rx_bytes += rbytes; | |
1529 | tot->tx_bytes += tbytes; | |
1530 | } | |
1531 | ||
1532 | tot->tx_dropped = dev->stats.tx_dropped; | |
021ac8d3 | 1533 | tot->tx_fifo_errors = dev->stats.tx_fifo_errors; |
3fa2a1df | 1534 | tot->rx_dropped = dev->stats.rx_dropped; |
1535 | tot->rx_length_errors = dev->stats.rx_length_errors; | |
1536 | tot->rx_frame_errors = dev->stats.rx_frame_errors; | |
3fa2a1df | 1537 | } |
1538 | ||
da74e89d AS |
1539 | #ifdef CONFIG_NET_POLL_CONTROLLER |
1540 | static void virtnet_netpoll(struct net_device *dev) | |
1541 | { | |
1542 | struct virtnet_info *vi = netdev_priv(dev); | |
986a4f4d | 1543 | int i; |
da74e89d | 1544 | |
986a4f4d JW |
1545 | for (i = 0; i < vi->curr_queue_pairs; i++) |
1546 | napi_schedule(&vi->rq[i].napi); | |
da74e89d AS |
1547 | } |
1548 | #endif | |
1549 | ||
586d17c5 JW |
1550 | static void virtnet_ack_link_announce(struct virtnet_info *vi) |
1551 | { | |
1552 | rtnl_lock(); | |
1553 | if (!virtnet_send_command(vi, VIRTIO_NET_CTRL_ANNOUNCE, | |
d24bae32 | 1554 | VIRTIO_NET_CTRL_ANNOUNCE_ACK, NULL)) |
586d17c5 JW |
1555 | dev_warn(&vi->dev->dev, "Failed to ack link announce.\n"); |
1556 | rtnl_unlock(); | |
1557 | } | |
1558 | ||
47315329 | 1559 | static int _virtnet_set_queues(struct virtnet_info *vi, u16 queue_pairs) |
986a4f4d JW |
1560 | { |
1561 | struct scatterlist sg; | |
986a4f4d JW |
1562 | struct net_device *dev = vi->dev; |
1563 | ||
1564 | if (!vi->has_cvq || !virtio_has_feature(vi->vdev, VIRTIO_NET_F_MQ)) | |
1565 | return 0; | |
1566 | ||
603343b9 MT |
1567 | vi->ctrl->mq.virtqueue_pairs = cpu_to_virtio16(vi->vdev, queue_pairs); |
1568 | sg_init_one(&sg, &vi->ctrl->mq, sizeof(vi->ctrl->mq)); | |
986a4f4d JW |
1569 | |
1570 | if (!virtnet_send_command(vi, VIRTIO_NET_CTRL_MQ, | |
d24bae32 | 1571 | VIRTIO_NET_CTRL_MQ_VQ_PAIRS_SET, &sg)) { |
986a4f4d JW |
1572 | dev_warn(&dev->dev, "Fail to set num of queue pairs to %d\n", |
1573 | queue_pairs); | |
1574 | return -EINVAL; | |
55257d72 | 1575 | } else { |
986a4f4d | 1576 | vi->curr_queue_pairs = queue_pairs; |
35ed159b JW |
1577 | /* virtnet_open() will refill when device is going to up. */ |
1578 | if (dev->flags & IFF_UP) | |
1579 | schedule_delayed_work(&vi->refill, 0); | |
55257d72 | 1580 | } |
986a4f4d JW |
1581 | |
1582 | return 0; | |
1583 | } | |
1584 | ||
47315329 JF |
1585 | static int virtnet_set_queues(struct virtnet_info *vi, u16 queue_pairs) |
1586 | { | |
1587 | int err; | |
1588 | ||
1589 | rtnl_lock(); | |
1590 | err = _virtnet_set_queues(vi, queue_pairs); | |
1591 | rtnl_unlock(); | |
1592 | return err; | |
1593 | } | |
1594 | ||
296f96fc RR |
1595 | static int virtnet_close(struct net_device *dev) |
1596 | { | |
1597 | struct virtnet_info *vi = netdev_priv(dev); | |
986a4f4d | 1598 | int i; |
296f96fc | 1599 | |
b2baed69 RR |
1600 | /* Make sure refill_work doesn't re-enable napi! */ |
1601 | cancel_delayed_work_sync(&vi->refill); | |
986a4f4d | 1602 | |
b92f1e67 | 1603 | for (i = 0; i < vi->max_queue_pairs; i++) { |
986a4f4d | 1604 | napi_disable(&vi->rq[i].napi); |
78a57b48 | 1605 | virtnet_napi_tx_disable(&vi->sq[i].napi); |
b92f1e67 | 1606 | } |
296f96fc | 1607 | |
296f96fc RR |
1608 | return 0; |
1609 | } | |
1610 | ||
2af7698e AW |
1611 | static void virtnet_set_rx_mode(struct net_device *dev) |
1612 | { | |
1613 | struct virtnet_info *vi = netdev_priv(dev); | |
f565a7c2 | 1614 | struct scatterlist sg[2]; |
f565a7c2 | 1615 | struct virtio_net_ctrl_mac *mac_data; |
ccffad25 | 1616 | struct netdev_hw_addr *ha; |
32e7bfc4 | 1617 | int uc_count; |
4cd24eaf | 1618 | int mc_count; |
f565a7c2 AW |
1619 | void *buf; |
1620 | int i; | |
2af7698e | 1621 | |
788a8b6d | 1622 | /* We can't dynamically set ndo_set_rx_mode, so return gracefully */ |
2af7698e AW |
1623 | if (!virtio_has_feature(vi->vdev, VIRTIO_NET_F_CTRL_RX)) |
1624 | return; | |
1625 | ||
603343b9 MT |
1626 | vi->ctrl->promisc = ((dev->flags & IFF_PROMISC) != 0); |
1627 | vi->ctrl->allmulti = ((dev->flags & IFF_ALLMULTI) != 0); | |
2af7698e | 1628 | |
603343b9 | 1629 | sg_init_one(sg, &vi->ctrl->promisc, sizeof(vi->ctrl->promisc)); |
2af7698e AW |
1630 | |
1631 | if (!virtnet_send_command(vi, VIRTIO_NET_CTRL_RX, | |
d24bae32 | 1632 | VIRTIO_NET_CTRL_RX_PROMISC, sg)) |
2af7698e | 1633 | dev_warn(&dev->dev, "Failed to %sable promisc mode.\n", |
603343b9 | 1634 | vi->ctrl->promisc ? "en" : "dis"); |
2af7698e | 1635 | |
603343b9 | 1636 | sg_init_one(sg, &vi->ctrl->allmulti, sizeof(vi->ctrl->allmulti)); |
2af7698e AW |
1637 | |
1638 | if (!virtnet_send_command(vi, VIRTIO_NET_CTRL_RX, | |
d24bae32 | 1639 | VIRTIO_NET_CTRL_RX_ALLMULTI, sg)) |
2af7698e | 1640 | dev_warn(&dev->dev, "Failed to %sable allmulti mode.\n", |
603343b9 | 1641 | vi->ctrl->allmulti ? "en" : "dis"); |
f565a7c2 | 1642 | |
32e7bfc4 | 1643 | uc_count = netdev_uc_count(dev); |
4cd24eaf | 1644 | mc_count = netdev_mc_count(dev); |
f565a7c2 | 1645 | /* MAC filter - use one buffer for both lists */ |
4cd24eaf JP |
1646 | buf = kzalloc(((uc_count + mc_count) * ETH_ALEN) + |
1647 | (2 * sizeof(mac_data->entries)), GFP_ATOMIC); | |
1648 | mac_data = buf; | |
e68ed8f0 | 1649 | if (!buf) |
f565a7c2 | 1650 | return; |
f565a7c2 | 1651 | |
23e258e1 AW |
1652 | sg_init_table(sg, 2); |
1653 | ||
f565a7c2 | 1654 | /* Store the unicast list and count in the front of the buffer */ |
fdd819b2 | 1655 | mac_data->entries = cpu_to_virtio32(vi->vdev, uc_count); |
ccffad25 | 1656 | i = 0; |
32e7bfc4 | 1657 | netdev_for_each_uc_addr(ha, dev) |
ccffad25 | 1658 | memcpy(&mac_data->macs[i++][0], ha->addr, ETH_ALEN); |
f565a7c2 AW |
1659 | |
1660 | sg_set_buf(&sg[0], mac_data, | |
32e7bfc4 | 1661 | sizeof(mac_data->entries) + (uc_count * ETH_ALEN)); |
f565a7c2 AW |
1662 | |
1663 | /* multicast list and count fill the end */ | |
32e7bfc4 | 1664 | mac_data = (void *)&mac_data->macs[uc_count][0]; |
f565a7c2 | 1665 | |
fdd819b2 | 1666 | mac_data->entries = cpu_to_virtio32(vi->vdev, mc_count); |
567ec874 | 1667 | i = 0; |
22bedad3 JP |
1668 | netdev_for_each_mc_addr(ha, dev) |
1669 | memcpy(&mac_data->macs[i++][0], ha->addr, ETH_ALEN); | |
f565a7c2 AW |
1670 | |
1671 | sg_set_buf(&sg[1], mac_data, | |
4cd24eaf | 1672 | sizeof(mac_data->entries) + (mc_count * ETH_ALEN)); |
f565a7c2 AW |
1673 | |
1674 | if (!virtnet_send_command(vi, VIRTIO_NET_CTRL_MAC, | |
d24bae32 | 1675 | VIRTIO_NET_CTRL_MAC_TABLE_SET, sg)) |
99e872ae | 1676 | dev_warn(&dev->dev, "Failed to set MAC filter table.\n"); |
f565a7c2 AW |
1677 | |
1678 | kfree(buf); | |
2af7698e AW |
1679 | } |
1680 | ||
80d5c368 PM |
1681 | static int virtnet_vlan_rx_add_vid(struct net_device *dev, |
1682 | __be16 proto, u16 vid) | |
0bde9569 AW |
1683 | { |
1684 | struct virtnet_info *vi = netdev_priv(dev); | |
1685 | struct scatterlist sg; | |
1686 | ||
c285a7d5 | 1687 | vi->ctrl->vid = cpu_to_virtio16(vi->vdev, vid); |
603343b9 | 1688 | sg_init_one(&sg, &vi->ctrl->vid, sizeof(vi->ctrl->vid)); |
0bde9569 AW |
1689 | |
1690 | if (!virtnet_send_command(vi, VIRTIO_NET_CTRL_VLAN, | |
d24bae32 | 1691 | VIRTIO_NET_CTRL_VLAN_ADD, &sg)) |
0bde9569 | 1692 | dev_warn(&dev->dev, "Failed to add VLAN ID %d.\n", vid); |
8e586137 | 1693 | return 0; |
0bde9569 AW |
1694 | } |
1695 | ||
80d5c368 PM |
1696 | static int virtnet_vlan_rx_kill_vid(struct net_device *dev, |
1697 | __be16 proto, u16 vid) | |
0bde9569 AW |
1698 | { |
1699 | struct virtnet_info *vi = netdev_priv(dev); | |
1700 | struct scatterlist sg; | |
1701 | ||
c285a7d5 | 1702 | vi->ctrl->vid = cpu_to_virtio16(vi->vdev, vid); |
603343b9 | 1703 | sg_init_one(&sg, &vi->ctrl->vid, sizeof(vi->ctrl->vid)); |
0bde9569 AW |
1704 | |
1705 | if (!virtnet_send_command(vi, VIRTIO_NET_CTRL_VLAN, | |
d24bae32 | 1706 | VIRTIO_NET_CTRL_VLAN_DEL, &sg)) |
0bde9569 | 1707 | dev_warn(&dev->dev, "Failed to kill VLAN ID %d.\n", vid); |
8e586137 | 1708 | return 0; |
0bde9569 AW |
1709 | } |
1710 | ||
8898c21c | 1711 | static void virtnet_clean_affinity(struct virtnet_info *vi, long hcpu) |
986a4f4d JW |
1712 | { |
1713 | int i; | |
1714 | ||
8898c21c WG |
1715 | if (vi->affinity_hint_set) { |
1716 | for (i = 0; i < vi->max_queue_pairs; i++) { | |
47be2479 WG |
1717 | virtqueue_set_affinity(vi->rq[i].vq, -1); |
1718 | virtqueue_set_affinity(vi->sq[i].vq, -1); | |
1719 | } | |
1720 | ||
8898c21c WG |
1721 | vi->affinity_hint_set = false; |
1722 | } | |
8898c21c | 1723 | } |
47be2479 | 1724 | |
8898c21c WG |
1725 | static void virtnet_set_affinity(struct virtnet_info *vi) |
1726 | { | |
1727 | int i; | |
1728 | int cpu; | |
986a4f4d JW |
1729 | |
1730 | /* In multiqueue mode, when the number of cpu is equal to the number of | |
1731 | * queue pairs, we let the queue pairs to be private to one cpu by | |
1732 | * setting the affinity hint to eliminate the contention. | |
1733 | */ | |
8898c21c WG |
1734 | if (vi->curr_queue_pairs == 1 || |
1735 | vi->max_queue_pairs != num_online_cpus()) { | |
1736 | virtnet_clean_affinity(vi, -1); | |
1737 | return; | |
986a4f4d JW |
1738 | } |
1739 | ||
8898c21c WG |
1740 | i = 0; |
1741 | for_each_online_cpu(cpu) { | |
986a4f4d JW |
1742 | virtqueue_set_affinity(vi->rq[i].vq, cpu); |
1743 | virtqueue_set_affinity(vi->sq[i].vq, cpu); | |
9bb8ca86 | 1744 | netif_set_xps_queue(vi->dev, cpumask_of(cpu), i); |
8898c21c | 1745 | i++; |
986a4f4d JW |
1746 | } |
1747 | ||
8898c21c | 1748 | vi->affinity_hint_set = true; |
986a4f4d JW |
1749 | } |
1750 | ||
8017c279 | 1751 | static int virtnet_cpu_online(unsigned int cpu, struct hlist_node *node) |
8de4b2f3 | 1752 | { |
8017c279 SAS |
1753 | struct virtnet_info *vi = hlist_entry_safe(node, struct virtnet_info, |
1754 | node); | |
1755 | virtnet_set_affinity(vi); | |
1756 | return 0; | |
1757 | } | |
8de4b2f3 | 1758 | |
8017c279 SAS |
1759 | static int virtnet_cpu_dead(unsigned int cpu, struct hlist_node *node) |
1760 | { | |
1761 | struct virtnet_info *vi = hlist_entry_safe(node, struct virtnet_info, | |
1762 | node_dead); | |
1763 | virtnet_set_affinity(vi); | |
1764 | return 0; | |
1765 | } | |
3ab098df | 1766 | |
8017c279 SAS |
1767 | static int virtnet_cpu_down_prep(unsigned int cpu, struct hlist_node *node) |
1768 | { | |
1769 | struct virtnet_info *vi = hlist_entry_safe(node, struct virtnet_info, | |
1770 | node); | |
1771 | ||
1772 | virtnet_clean_affinity(vi, cpu); | |
1773 | return 0; | |
1774 | } | |
1775 | ||
1776 | static enum cpuhp_state virtionet_online; | |
1777 | ||
1778 | static int virtnet_cpu_notif_add(struct virtnet_info *vi) | |
1779 | { | |
1780 | int ret; | |
1781 | ||
1782 | ret = cpuhp_state_add_instance_nocalls(virtionet_online, &vi->node); | |
1783 | if (ret) | |
1784 | return ret; | |
1785 | ret = cpuhp_state_add_instance_nocalls(CPUHP_VIRT_NET_DEAD, | |
1786 | &vi->node_dead); | |
1787 | if (!ret) | |
1788 | return ret; | |
1789 | cpuhp_state_remove_instance_nocalls(virtionet_online, &vi->node); | |
1790 | return ret; | |
1791 | } | |
1792 | ||
1793 | static void virtnet_cpu_notif_remove(struct virtnet_info *vi) | |
1794 | { | |
1795 | cpuhp_state_remove_instance_nocalls(virtionet_online, &vi->node); | |
1796 | cpuhp_state_remove_instance_nocalls(CPUHP_VIRT_NET_DEAD, | |
1797 | &vi->node_dead); | |
986a4f4d JW |
1798 | } |
1799 | ||
8f9f4668 RJ |
1800 | static void virtnet_get_ringparam(struct net_device *dev, |
1801 | struct ethtool_ringparam *ring) | |
1802 | { | |
1803 | struct virtnet_info *vi = netdev_priv(dev); | |
1804 | ||
986a4f4d JW |
1805 | ring->rx_max_pending = virtqueue_get_vring_size(vi->rq[0].vq); |
1806 | ring->tx_max_pending = virtqueue_get_vring_size(vi->sq[0].vq); | |
8f9f4668 RJ |
1807 | ring->rx_pending = ring->rx_max_pending; |
1808 | ring->tx_pending = ring->tx_max_pending; | |
8f9f4668 RJ |
1809 | } |
1810 | ||
66846048 RJ |
1811 | |
1812 | static void virtnet_get_drvinfo(struct net_device *dev, | |
1813 | struct ethtool_drvinfo *info) | |
1814 | { | |
1815 | struct virtnet_info *vi = netdev_priv(dev); | |
1816 | struct virtio_device *vdev = vi->vdev; | |
1817 | ||
1818 | strlcpy(info->driver, KBUILD_MODNAME, sizeof(info->driver)); | |
1819 | strlcpy(info->version, VIRTNET_DRIVER_VERSION, sizeof(info->version)); | |
1820 | strlcpy(info->bus_info, virtio_bus_name(vdev), sizeof(info->bus_info)); | |
1821 | ||
1822 | } | |
1823 | ||
d73bcd2c JW |
1824 | /* TODO: Eliminate OOO packets during switching */ |
1825 | static int virtnet_set_channels(struct net_device *dev, | |
1826 | struct ethtool_channels *channels) | |
1827 | { | |
1828 | struct virtnet_info *vi = netdev_priv(dev); | |
1829 | u16 queue_pairs = channels->combined_count; | |
1830 | int err; | |
1831 | ||
1832 | /* We don't support separate rx/tx channels. | |
1833 | * We don't allow setting 'other' channels. | |
1834 | */ | |
1835 | if (channels->rx_count || channels->tx_count || channels->other_count) | |
1836 | return -EINVAL; | |
1837 | ||
c18e9cd6 | 1838 | if (queue_pairs > vi->max_queue_pairs || queue_pairs == 0) |
d73bcd2c JW |
1839 | return -EINVAL; |
1840 | ||
f600b690 JF |
1841 | /* For now we don't support modifying channels while XDP is loaded |
1842 | * also when XDP is loaded all RX queues have XDP programs so we only | |
1843 | * need to check a single RX queue. | |
1844 | */ | |
1845 | if (vi->rq[0].xdp_prog) | |
1846 | return -EINVAL; | |
1847 | ||
47be2479 | 1848 | get_online_cpus(); |
47315329 | 1849 | err = _virtnet_set_queues(vi, queue_pairs); |
d73bcd2c JW |
1850 | if (!err) { |
1851 | netif_set_real_num_tx_queues(dev, queue_pairs); | |
1852 | netif_set_real_num_rx_queues(dev, queue_pairs); | |
1853 | ||
8898c21c | 1854 | virtnet_set_affinity(vi); |
d73bcd2c | 1855 | } |
47be2479 | 1856 | put_online_cpus(); |
d73bcd2c JW |
1857 | |
1858 | return err; | |
1859 | } | |
1860 | ||
1861 | static void virtnet_get_channels(struct net_device *dev, | |
1862 | struct ethtool_channels *channels) | |
1863 | { | |
1864 | struct virtnet_info *vi = netdev_priv(dev); | |
1865 | ||
1866 | channels->combined_count = vi->curr_queue_pairs; | |
1867 | channels->max_combined = vi->max_queue_pairs; | |
1868 | channels->max_other = 0; | |
1869 | channels->rx_count = 0; | |
1870 | channels->tx_count = 0; | |
1871 | channels->other_count = 0; | |
1872 | } | |
1873 | ||
16032be5 | 1874 | /* Check if the user is trying to change anything besides speed/duplex */ |
ebb6b4b1 PR |
1875 | static bool |
1876 | virtnet_validate_ethtool_cmd(const struct ethtool_link_ksettings *cmd) | |
16032be5 | 1877 | { |
ebb6b4b1 PR |
1878 | struct ethtool_link_ksettings diff1 = *cmd; |
1879 | struct ethtool_link_ksettings diff2 = {}; | |
16032be5 | 1880 | |
0cf3ace9 NA |
1881 | /* cmd is always set so we need to clear it, validate the port type |
1882 | * and also without autonegotiation we can ignore advertising | |
1883 | */ | |
ebb6b4b1 PR |
1884 | diff1.base.speed = 0; |
1885 | diff2.base.port = PORT_OTHER; | |
1886 | ethtool_link_ksettings_zero_link_mode(&diff1, advertising); | |
1887 | diff1.base.duplex = 0; | |
1888 | diff1.base.cmd = 0; | |
1889 | diff1.base.link_mode_masks_nwords = 0; | |
1890 | ||
1891 | return !memcmp(&diff1.base, &diff2.base, sizeof(diff1.base)) && | |
1892 | bitmap_empty(diff1.link_modes.supported, | |
1893 | __ETHTOOL_LINK_MODE_MASK_NBITS) && | |
1894 | bitmap_empty(diff1.link_modes.advertising, | |
1895 | __ETHTOOL_LINK_MODE_MASK_NBITS) && | |
1896 | bitmap_empty(diff1.link_modes.lp_advertising, | |
1897 | __ETHTOOL_LINK_MODE_MASK_NBITS); | |
16032be5 NA |
1898 | } |
1899 | ||
ebb6b4b1 PR |
1900 | static int virtnet_set_link_ksettings(struct net_device *dev, |
1901 | const struct ethtool_link_ksettings *cmd) | |
16032be5 NA |
1902 | { |
1903 | struct virtnet_info *vi = netdev_priv(dev); | |
1904 | u32 speed; | |
1905 | ||
ebb6b4b1 | 1906 | speed = cmd->base.speed; |
16032be5 NA |
1907 | /* don't allow custom speed and duplex */ |
1908 | if (!ethtool_validate_speed(speed) || | |
ebb6b4b1 | 1909 | !ethtool_validate_duplex(cmd->base.duplex) || |
16032be5 NA |
1910 | !virtnet_validate_ethtool_cmd(cmd)) |
1911 | return -EINVAL; | |
1912 | vi->speed = speed; | |
ebb6b4b1 | 1913 | vi->duplex = cmd->base.duplex; |
16032be5 NA |
1914 | |
1915 | return 0; | |
1916 | } | |
1917 | ||
ebb6b4b1 PR |
1918 | static int virtnet_get_link_ksettings(struct net_device *dev, |
1919 | struct ethtool_link_ksettings *cmd) | |
16032be5 NA |
1920 | { |
1921 | struct virtnet_info *vi = netdev_priv(dev); | |
1922 | ||
ebb6b4b1 PR |
1923 | cmd->base.speed = vi->speed; |
1924 | cmd->base.duplex = vi->duplex; | |
1925 | cmd->base.port = PORT_OTHER; | |
16032be5 NA |
1926 | |
1927 | return 0; | |
1928 | } | |
1929 | ||
1930 | static void virtnet_init_settings(struct net_device *dev) | |
1931 | { | |
1932 | struct virtnet_info *vi = netdev_priv(dev); | |
1933 | ||
1934 | vi->speed = SPEED_UNKNOWN; | |
1935 | vi->duplex = DUPLEX_UNKNOWN; | |
1936 | } | |
1937 | ||
0fc0b732 | 1938 | static const struct ethtool_ops virtnet_ethtool_ops = { |
66846048 | 1939 | .get_drvinfo = virtnet_get_drvinfo, |
9f4d26d0 | 1940 | .get_link = ethtool_op_get_link, |
8f9f4668 | 1941 | .get_ringparam = virtnet_get_ringparam, |
d73bcd2c JW |
1942 | .set_channels = virtnet_set_channels, |
1943 | .get_channels = virtnet_get_channels, | |
074c3582 | 1944 | .get_ts_info = ethtool_op_get_ts_info, |
ebb6b4b1 PR |
1945 | .get_link_ksettings = virtnet_get_link_ksettings, |
1946 | .set_link_ksettings = virtnet_set_link_ksettings, | |
a9ea3fc6 HX |
1947 | }; |
1948 | ||
9fe7bfce JF |
1949 | static void virtnet_freeze_down(struct virtio_device *vdev) |
1950 | { | |
1951 | struct virtnet_info *vi = vdev->priv; | |
1952 | int i; | |
1953 | ||
1954 | /* Make sure no work handler is accessing the device */ | |
1955 | flush_work(&vi->config_work); | |
1956 | ||
d20ed693 | 1957 | netif_tx_lock_bh(vi->dev); |
9fe7bfce | 1958 | netif_device_detach(vi->dev); |
d20ed693 | 1959 | netif_tx_unlock_bh(vi->dev); |
9fe7bfce JF |
1960 | cancel_delayed_work_sync(&vi->refill); |
1961 | ||
1962 | if (netif_running(vi->dev)) { | |
b92f1e67 | 1963 | for (i = 0; i < vi->max_queue_pairs; i++) { |
9fe7bfce | 1964 | napi_disable(&vi->rq[i].napi); |
78a57b48 | 1965 | virtnet_napi_tx_disable(&vi->sq[i].napi); |
b92f1e67 | 1966 | } |
9fe7bfce JF |
1967 | } |
1968 | } | |
1969 | ||
1970 | static int init_vqs(struct virtnet_info *vi); | |
1971 | ||
1972 | static int virtnet_restore_up(struct virtio_device *vdev) | |
1973 | { | |
1974 | struct virtnet_info *vi = vdev->priv; | |
1975 | int err, i; | |
1976 | ||
1977 | err = init_vqs(vi); | |
1978 | if (err) | |
1979 | return err; | |
1980 | ||
1981 | virtio_device_ready(vdev); | |
1982 | ||
1983 | if (netif_running(vi->dev)) { | |
1984 | for (i = 0; i < vi->curr_queue_pairs; i++) | |
1985 | if (!try_fill_recv(vi, &vi->rq[i], GFP_KERNEL)) | |
1986 | schedule_delayed_work(&vi->refill, 0); | |
1987 | ||
b92f1e67 | 1988 | for (i = 0; i < vi->max_queue_pairs; i++) { |
e4e8452a | 1989 | virtnet_napi_enable(vi->rq[i].vq, &vi->rq[i].napi); |
b92f1e67 WB |
1990 | virtnet_napi_tx_enable(vi, vi->sq[i].vq, |
1991 | &vi->sq[i].napi); | |
1992 | } | |
9fe7bfce JF |
1993 | } |
1994 | ||
d20ed693 | 1995 | netif_tx_lock_bh(vi->dev); |
9fe7bfce | 1996 | netif_device_attach(vi->dev); |
d20ed693 | 1997 | netif_tx_unlock_bh(vi->dev); |
9fe7bfce JF |
1998 | return err; |
1999 | } | |
2000 | ||
3f93522f JW |
2001 | static int virtnet_set_guest_offloads(struct virtnet_info *vi, u64 offloads) |
2002 | { | |
2003 | struct scatterlist sg; | |
603343b9 | 2004 | vi->ctrl->offloads = cpu_to_virtio64(vi->vdev, offloads); |
3f93522f | 2005 | |
603343b9 | 2006 | sg_init_one(&sg, &vi->ctrl->offloads, sizeof(vi->ctrl->offloads)); |
3f93522f JW |
2007 | |
2008 | if (!virtnet_send_command(vi, VIRTIO_NET_CTRL_GUEST_OFFLOADS, | |
2009 | VIRTIO_NET_CTRL_GUEST_OFFLOADS_SET, &sg)) { | |
2010 | dev_warn(&vi->dev->dev, "Fail to set guest offload. \n"); | |
2011 | return -EINVAL; | |
2012 | } | |
2013 | ||
2014 | return 0; | |
2015 | } | |
2016 | ||
2017 | static int virtnet_clear_guest_offloads(struct virtnet_info *vi) | |
2018 | { | |
2019 | u64 offloads = 0; | |
2020 | ||
2021 | if (!vi->guest_offloads) | |
2022 | return 0; | |
2023 | ||
3f93522f JW |
2024 | return virtnet_set_guest_offloads(vi, offloads); |
2025 | } | |
2026 | ||
2027 | static int virtnet_restore_guest_offloads(struct virtnet_info *vi) | |
2028 | { | |
2029 | u64 offloads = vi->guest_offloads; | |
2030 | ||
2031 | if (!vi->guest_offloads) | |
2032 | return 0; | |
3f93522f JW |
2033 | |
2034 | return virtnet_set_guest_offloads(vi, offloads); | |
2035 | } | |
2036 | ||
9861ce03 JK |
2037 | static int virtnet_xdp_set(struct net_device *dev, struct bpf_prog *prog, |
2038 | struct netlink_ext_ack *extack) | |
f600b690 JF |
2039 | { |
2040 | unsigned long int max_sz = PAGE_SIZE - sizeof(struct padded_vnet_hdr); | |
2041 | struct virtnet_info *vi = netdev_priv(dev); | |
2042 | struct bpf_prog *old_prog; | |
017b29c3 | 2043 | u16 xdp_qp = 0, curr_qp; |
672aafd5 | 2044 | int i, err; |
f600b690 | 2045 | |
3f93522f JW |
2046 | if (!virtio_has_feature(vi->vdev, VIRTIO_NET_F_CTRL_GUEST_OFFLOADS) |
2047 | && (virtio_has_feature(vi->vdev, VIRTIO_NET_F_GUEST_TSO4) || | |
2048 | virtio_has_feature(vi->vdev, VIRTIO_NET_F_GUEST_TSO6) || | |
2049 | virtio_has_feature(vi->vdev, VIRTIO_NET_F_GUEST_ECN) || | |
5728cf06 JW |
2050 | virtio_has_feature(vi->vdev, VIRTIO_NET_F_GUEST_UFO) || |
2051 | virtio_has_feature(vi->vdev, VIRTIO_NET_F_GUEST_CSUM))) { | |
2052 | NL_SET_ERR_MSG_MOD(extack, "Can't set XDP while host is implementing LRO/CSUM, disable LRO/CSUM first"); | |
f600b690 JF |
2053 | return -EOPNOTSUPP; |
2054 | } | |
2055 | ||
2056 | if (vi->mergeable_rx_bufs && !vi->any_header_sg) { | |
4d463c4d | 2057 | NL_SET_ERR_MSG_MOD(extack, "XDP expects header/data in single page, any_header_sg required"); |
f600b690 JF |
2058 | return -EINVAL; |
2059 | } | |
2060 | ||
2061 | if (dev->mtu > max_sz) { | |
4d463c4d | 2062 | NL_SET_ERR_MSG_MOD(extack, "MTU too large to enable XDP"); |
f600b690 JF |
2063 | netdev_warn(dev, "XDP requires MTU less than %lu\n", max_sz); |
2064 | return -EINVAL; | |
2065 | } | |
2066 | ||
672aafd5 JF |
2067 | curr_qp = vi->curr_queue_pairs - vi->xdp_queue_pairs; |
2068 | if (prog) | |
2069 | xdp_qp = nr_cpu_ids; | |
2070 | ||
2071 | /* XDP requires extra queues for XDP_TX */ | |
2072 | if (curr_qp + xdp_qp > vi->max_queue_pairs) { | |
4d463c4d | 2073 | NL_SET_ERR_MSG_MOD(extack, "Too few free TX rings available"); |
672aafd5 JF |
2074 | netdev_warn(dev, "request %i queues but max is %i\n", |
2075 | curr_qp + xdp_qp, vi->max_queue_pairs); | |
2076 | return -ENOMEM; | |
2077 | } | |
2078 | ||
9b60b8b6 TM |
2079 | old_prog = rtnl_dereference(vi->rq[0].xdp_prog); |
2080 | if (!prog && !old_prog) | |
2081 | return 0; | |
2082 | ||
2de2f7f4 JF |
2083 | if (prog) { |
2084 | prog = bpf_prog_add(prog, vi->max_queue_pairs - 1); | |
2085 | if (IS_ERR(prog)) | |
2086 | return PTR_ERR(prog); | |
2087 | } | |
2088 | ||
4941d472 | 2089 | /* Make sure NAPI is not using any XDP TX queues for RX. */ |
d8c6e7d4 TM |
2090 | if (netif_running(dev)) { |
2091 | for (i = 0; i < vi->max_queue_pairs; i++) { | |
e527a479 | 2092 | napi_disable(&vi->rq[i].napi); |
d8c6e7d4 TM |
2093 | virtnet_napi_tx_disable(&vi->sq[i].napi); |
2094 | } | |
2095 | } | |
f600b690 | 2096 | |
9b60b8b6 TM |
2097 | if (!prog) { |
2098 | for (i = 0; i < vi->max_queue_pairs; i++) { | |
2099 | rcu_assign_pointer(vi->rq[i].xdp_prog, prog); | |
2100 | if (i == 0) | |
2101 | virtnet_restore_guest_offloads(vi); | |
2102 | } | |
2103 | synchronize_net(); | |
2104 | } | |
2105 | ||
4941d472 JW |
2106 | err = _virtnet_set_queues(vi, curr_qp + xdp_qp); |
2107 | if (err) | |
2108 | goto err; | |
4244c299 | 2109 | netif_set_real_num_rx_queues(dev, curr_qp + xdp_qp); |
4941d472 | 2110 | vi->xdp_queue_pairs = xdp_qp; |
672aafd5 | 2111 | |
9b60b8b6 TM |
2112 | if (prog) { |
2113 | for (i = 0; i < vi->max_queue_pairs; i++) { | |
2114 | rcu_assign_pointer(vi->rq[i].xdp_prog, prog); | |
2115 | if (i == 0 && !old_prog) | |
3f93522f | 2116 | virtnet_clear_guest_offloads(vi); |
3f93522f | 2117 | } |
9b60b8b6 TM |
2118 | } |
2119 | ||
2120 | for (i = 0; i < vi->max_queue_pairs; i++) { | |
f600b690 JF |
2121 | if (old_prog) |
2122 | bpf_prog_put(old_prog); | |
d8c6e7d4 | 2123 | if (netif_running(dev)) { |
e527a479 | 2124 | virtnet_napi_enable(vi->rq[i].vq, &vi->rq[i].napi); |
d8c6e7d4 TM |
2125 | virtnet_napi_tx_enable(vi, vi->sq[i].vq, |
2126 | &vi->sq[i].napi); | |
2127 | } | |
f600b690 JF |
2128 | } |
2129 | ||
2130 | return 0; | |
2de2f7f4 | 2131 | |
4941d472 | 2132 | err: |
9b60b8b6 TM |
2133 | if (!prog) { |
2134 | virtnet_clear_guest_offloads(vi); | |
2135 | for (i = 0; i < vi->max_queue_pairs; i++) | |
2136 | rcu_assign_pointer(vi->rq[i].xdp_prog, old_prog); | |
2137 | } | |
2138 | ||
83977d2c | 2139 | if (netif_running(dev)) { |
d8c6e7d4 | 2140 | for (i = 0; i < vi->max_queue_pairs; i++) { |
83977d2c | 2141 | virtnet_napi_enable(vi->rq[i].vq, &vi->rq[i].napi); |
d8c6e7d4 TM |
2142 | virtnet_napi_tx_enable(vi, vi->sq[i].vq, |
2143 | &vi->sq[i].napi); | |
2144 | } | |
83977d2c | 2145 | } |
2de2f7f4 JF |
2146 | if (prog) |
2147 | bpf_prog_sub(prog, vi->max_queue_pairs - 1); | |
2148 | return err; | |
f600b690 JF |
2149 | } |
2150 | ||
5b0e6629 | 2151 | static u32 virtnet_xdp_query(struct net_device *dev) |
f600b690 JF |
2152 | { |
2153 | struct virtnet_info *vi = netdev_priv(dev); | |
5b0e6629 | 2154 | const struct bpf_prog *xdp_prog; |
f600b690 JF |
2155 | int i; |
2156 | ||
2157 | for (i = 0; i < vi->max_queue_pairs; i++) { | |
5b0e6629 MKL |
2158 | xdp_prog = rtnl_dereference(vi->rq[i].xdp_prog); |
2159 | if (xdp_prog) | |
2160 | return xdp_prog->aux->id; | |
f600b690 | 2161 | } |
5b0e6629 | 2162 | return 0; |
f600b690 JF |
2163 | } |
2164 | ||
f4e63525 | 2165 | static int virtnet_xdp(struct net_device *dev, struct netdev_bpf *xdp) |
f600b690 JF |
2166 | { |
2167 | switch (xdp->command) { | |
2168 | case XDP_SETUP_PROG: | |
9861ce03 | 2169 | return virtnet_xdp_set(dev, xdp->prog, xdp->extack); |
f600b690 | 2170 | case XDP_QUERY_PROG: |
5b0e6629 MKL |
2171 | xdp->prog_id = virtnet_xdp_query(dev); |
2172 | xdp->prog_attached = !!xdp->prog_id; | |
f600b690 JF |
2173 | return 0; |
2174 | default: | |
2175 | return -EINVAL; | |
2176 | } | |
2177 | } | |
2178 | ||
76288b4e SH |
2179 | static const struct net_device_ops virtnet_netdev = { |
2180 | .ndo_open = virtnet_open, | |
2181 | .ndo_stop = virtnet_close, | |
2182 | .ndo_start_xmit = start_xmit, | |
2183 | .ndo_validate_addr = eth_validate_addr, | |
9c46f6d4 | 2184 | .ndo_set_mac_address = virtnet_set_mac_address, |
2af7698e | 2185 | .ndo_set_rx_mode = virtnet_set_rx_mode, |
3fa2a1df | 2186 | .ndo_get_stats64 = virtnet_stats, |
1824a989 AW |
2187 | .ndo_vlan_rx_add_vid = virtnet_vlan_rx_add_vid, |
2188 | .ndo_vlan_rx_kill_vid = virtnet_vlan_rx_kill_vid, | |
76288b4e SH |
2189 | #ifdef CONFIG_NET_POLL_CONTROLLER |
2190 | .ndo_poll_controller = virtnet_netpoll, | |
91815639 | 2191 | #endif |
f4e63525 | 2192 | .ndo_bpf = virtnet_xdp, |
186b3c99 JW |
2193 | .ndo_xdp_xmit = virtnet_xdp_xmit, |
2194 | .ndo_xdp_flush = virtnet_xdp_flush, | |
2836b4f2 | 2195 | .ndo_features_check = passthru_features_check, |
76288b4e SH |
2196 | }; |
2197 | ||
586d17c5 | 2198 | static void virtnet_config_changed_work(struct work_struct *work) |
9f4d26d0 | 2199 | { |
586d17c5 JW |
2200 | struct virtnet_info *vi = |
2201 | container_of(work, struct virtnet_info, config_work); | |
9f4d26d0 MM |
2202 | u16 v; |
2203 | ||
855e0c52 RR |
2204 | if (virtio_cread_feature(vi->vdev, VIRTIO_NET_F_STATUS, |
2205 | struct virtio_net_config, status, &v) < 0) | |
507613bf | 2206 | return; |
586d17c5 JW |
2207 | |
2208 | if (v & VIRTIO_NET_S_ANNOUNCE) { | |
ee89bab1 | 2209 | netdev_notify_peers(vi->dev); |
586d17c5 JW |
2210 | virtnet_ack_link_announce(vi); |
2211 | } | |
9f4d26d0 MM |
2212 | |
2213 | /* Ignore unknown (future) status bits */ | |
2214 | v &= VIRTIO_NET_S_LINK_UP; | |
2215 | ||
2216 | if (vi->status == v) | |
507613bf | 2217 | return; |
9f4d26d0 MM |
2218 | |
2219 | vi->status = v; | |
2220 | ||
2221 | if (vi->status & VIRTIO_NET_S_LINK_UP) { | |
2222 | netif_carrier_on(vi->dev); | |
986a4f4d | 2223 | netif_tx_wake_all_queues(vi->dev); |
9f4d26d0 MM |
2224 | } else { |
2225 | netif_carrier_off(vi->dev); | |
986a4f4d | 2226 | netif_tx_stop_all_queues(vi->dev); |
9f4d26d0 MM |
2227 | } |
2228 | } | |
2229 | ||
2230 | static void virtnet_config_changed(struct virtio_device *vdev) | |
2231 | { | |
2232 | struct virtnet_info *vi = vdev->priv; | |
2233 | ||
3b07e9ca | 2234 | schedule_work(&vi->config_work); |
9f4d26d0 MM |
2235 | } |
2236 | ||
986a4f4d JW |
2237 | static void virtnet_free_queues(struct virtnet_info *vi) |
2238 | { | |
d4fb84ee AV |
2239 | int i; |
2240 | ||
ab3971b1 JW |
2241 | for (i = 0; i < vi->max_queue_pairs; i++) { |
2242 | napi_hash_del(&vi->rq[i].napi); | |
d4fb84ee | 2243 | netif_napi_del(&vi->rq[i].napi); |
b92f1e67 | 2244 | netif_napi_del(&vi->sq[i].napi); |
ab3971b1 | 2245 | } |
d4fb84ee | 2246 | |
963abe5c ED |
2247 | /* We called napi_hash_del() before netif_napi_del(), |
2248 | * we need to respect an RCU grace period before freeing vi->rq | |
2249 | */ | |
2250 | synchronize_net(); | |
2251 | ||
986a4f4d JW |
2252 | kfree(vi->rq); |
2253 | kfree(vi->sq); | |
603343b9 | 2254 | kfree(vi->ctrl); |
986a4f4d JW |
2255 | } |
2256 | ||
47315329 | 2257 | static void _free_receive_bufs(struct virtnet_info *vi) |
986a4f4d | 2258 | { |
f600b690 | 2259 | struct bpf_prog *old_prog; |
986a4f4d JW |
2260 | int i; |
2261 | ||
2262 | for (i = 0; i < vi->max_queue_pairs; i++) { | |
2263 | while (vi->rq[i].pages) | |
2264 | __free_pages(get_a_page(&vi->rq[i], GFP_KERNEL), 0); | |
f600b690 JF |
2265 | |
2266 | old_prog = rtnl_dereference(vi->rq[i].xdp_prog); | |
2267 | RCU_INIT_POINTER(vi->rq[i].xdp_prog, NULL); | |
2268 | if (old_prog) | |
2269 | bpf_prog_put(old_prog); | |
986a4f4d | 2270 | } |
47315329 JF |
2271 | } |
2272 | ||
2273 | static void free_receive_bufs(struct virtnet_info *vi) | |
2274 | { | |
2275 | rtnl_lock(); | |
2276 | _free_receive_bufs(vi); | |
f600b690 | 2277 | rtnl_unlock(); |
986a4f4d JW |
2278 | } |
2279 | ||
fb51879d MD |
2280 | static void free_receive_page_frags(struct virtnet_info *vi) |
2281 | { | |
2282 | int i; | |
2283 | for (i = 0; i < vi->max_queue_pairs; i++) | |
2284 | if (vi->rq[i].alloc_frag.page) | |
2285 | put_page(vi->rq[i].alloc_frag.page); | |
2286 | } | |
2287 | ||
986a4f4d JW |
2288 | static void free_unused_bufs(struct virtnet_info *vi) |
2289 | { | |
2290 | void *buf; | |
2291 | int i; | |
2292 | ||
2293 | for (i = 0; i < vi->max_queue_pairs; i++) { | |
2294 | struct virtqueue *vq = vi->sq[i].vq; | |
56434a01 | 2295 | while ((buf = virtqueue_detach_unused_buf(vq)) != NULL) { |
b68df015 | 2296 | if (!is_xdp_raw_buffer_queue(vi, i)) |
56434a01 JF |
2297 | dev_kfree_skb(buf); |
2298 | else | |
2299 | put_page(virt_to_head_page(buf)); | |
2300 | } | |
986a4f4d JW |
2301 | } |
2302 | ||
2303 | for (i = 0; i < vi->max_queue_pairs; i++) { | |
2304 | struct virtqueue *vq = vi->rq[i].vq; | |
2305 | ||
2306 | while ((buf = virtqueue_detach_unused_buf(vq)) != NULL) { | |
ab7db917 | 2307 | if (vi->mergeable_rx_bufs) { |
680557cf | 2308 | put_page(virt_to_head_page(buf)); |
ab7db917 | 2309 | } else if (vi->big_packets) { |
fa9fac17 | 2310 | give_pages(&vi->rq[i], buf); |
ab7db917 | 2311 | } else { |
f6b10209 | 2312 | put_page(virt_to_head_page(buf)); |
ab7db917 | 2313 | } |
986a4f4d | 2314 | } |
986a4f4d JW |
2315 | } |
2316 | } | |
2317 | ||
e9d7417b JW |
2318 | static void virtnet_del_vqs(struct virtnet_info *vi) |
2319 | { | |
2320 | struct virtio_device *vdev = vi->vdev; | |
2321 | ||
8898c21c | 2322 | virtnet_clean_affinity(vi, -1); |
986a4f4d | 2323 | |
e9d7417b | 2324 | vdev->config->del_vqs(vdev); |
986a4f4d JW |
2325 | |
2326 | virtnet_free_queues(vi); | |
e9d7417b JW |
2327 | } |
2328 | ||
d85b758f MT |
2329 | /* How large should a single buffer be so a queue full of these can fit at |
2330 | * least one full packet? | |
2331 | * Logic below assumes the mergeable buffer header is used. | |
2332 | */ | |
2333 | static unsigned int mergeable_min_buf_len(struct virtnet_info *vi, struct virtqueue *vq) | |
2334 | { | |
2335 | const unsigned int hdr_len = sizeof(struct virtio_net_hdr_mrg_rxbuf); | |
2336 | unsigned int rq_size = virtqueue_get_vring_size(vq); | |
2337 | unsigned int packet_len = vi->big_packets ? IP_MAX_MTU : vi->dev->max_mtu; | |
2338 | unsigned int buf_len = hdr_len + ETH_HLEN + VLAN_HLEN + packet_len; | |
2339 | unsigned int min_buf_len = DIV_ROUND_UP(buf_len, rq_size); | |
2340 | ||
f0c3192c MT |
2341 | return max(max(min_buf_len, hdr_len) - hdr_len, |
2342 | (unsigned int)GOOD_PACKET_LEN); | |
d85b758f MT |
2343 | } |
2344 | ||
986a4f4d | 2345 | static int virtnet_find_vqs(struct virtnet_info *vi) |
3f9c10b0 | 2346 | { |
986a4f4d JW |
2347 | vq_callback_t **callbacks; |
2348 | struct virtqueue **vqs; | |
2349 | int ret = -ENOMEM; | |
2350 | int i, total_vqs; | |
2351 | const char **names; | |
d45b897b | 2352 | bool *ctx; |
986a4f4d JW |
2353 | |
2354 | /* We expect 1 RX virtqueue followed by 1 TX virtqueue, followed by | |
2355 | * possible N-1 RX/TX queue pairs used in multiqueue mode, followed by | |
2356 | * possible control vq. | |
2357 | */ | |
2358 | total_vqs = vi->max_queue_pairs * 2 + | |
2359 | virtio_has_feature(vi->vdev, VIRTIO_NET_F_CTRL_VQ); | |
2360 | ||
2361 | /* Allocate space for find_vqs parameters */ | |
2362 | vqs = kzalloc(total_vqs * sizeof(*vqs), GFP_KERNEL); | |
2363 | if (!vqs) | |
2364 | goto err_vq; | |
2365 | callbacks = kmalloc(total_vqs * sizeof(*callbacks), GFP_KERNEL); | |
2366 | if (!callbacks) | |
2367 | goto err_callback; | |
2368 | names = kmalloc(total_vqs * sizeof(*names), GFP_KERNEL); | |
2369 | if (!names) | |
2370 | goto err_names; | |
192f68cf | 2371 | if (!vi->big_packets || vi->mergeable_rx_bufs) { |
d45b897b MT |
2372 | ctx = kzalloc(total_vqs * sizeof(*ctx), GFP_KERNEL); |
2373 | if (!ctx) | |
2374 | goto err_ctx; | |
2375 | } else { | |
2376 | ctx = NULL; | |
2377 | } | |
986a4f4d JW |
2378 | |
2379 | /* Parameters for control virtqueue, if any */ | |
2380 | if (vi->has_cvq) { | |
2381 | callbacks[total_vqs - 1] = NULL; | |
2382 | names[total_vqs - 1] = "control"; | |
2383 | } | |
3f9c10b0 | 2384 | |
986a4f4d JW |
2385 | /* Allocate/initialize parameters for send/receive virtqueues */ |
2386 | for (i = 0; i < vi->max_queue_pairs; i++) { | |
2387 | callbacks[rxq2vq(i)] = skb_recv_done; | |
2388 | callbacks[txq2vq(i)] = skb_xmit_done; | |
2389 | sprintf(vi->rq[i].name, "input.%d", i); | |
2390 | sprintf(vi->sq[i].name, "output.%d", i); | |
2391 | names[rxq2vq(i)] = vi->rq[i].name; | |
2392 | names[txq2vq(i)] = vi->sq[i].name; | |
d45b897b MT |
2393 | if (ctx) |
2394 | ctx[rxq2vq(i)] = true; | |
986a4f4d | 2395 | } |
3f9c10b0 | 2396 | |
986a4f4d | 2397 | ret = vi->vdev->config->find_vqs(vi->vdev, total_vqs, vqs, callbacks, |
d45b897b | 2398 | names, ctx, NULL); |
986a4f4d JW |
2399 | if (ret) |
2400 | goto err_find; | |
3f9c10b0 | 2401 | |
986a4f4d JW |
2402 | if (vi->has_cvq) { |
2403 | vi->cvq = vqs[total_vqs - 1]; | |
3f9c10b0 | 2404 | if (virtio_has_feature(vi->vdev, VIRTIO_NET_F_CTRL_VLAN)) |
f646968f | 2405 | vi->dev->features |= NETIF_F_HW_VLAN_CTAG_FILTER; |
3f9c10b0 | 2406 | } |
986a4f4d JW |
2407 | |
2408 | for (i = 0; i < vi->max_queue_pairs; i++) { | |
2409 | vi->rq[i].vq = vqs[rxq2vq(i)]; | |
d85b758f | 2410 | vi->rq[i].min_buf_len = mergeable_min_buf_len(vi, vi->rq[i].vq); |
986a4f4d JW |
2411 | vi->sq[i].vq = vqs[txq2vq(i)]; |
2412 | } | |
2413 | ||
2414 | kfree(names); | |
2415 | kfree(callbacks); | |
2416 | kfree(vqs); | |
55281621 | 2417 | kfree(ctx); |
986a4f4d | 2418 | |
3f9c10b0 | 2419 | return 0; |
986a4f4d JW |
2420 | |
2421 | err_find: | |
d45b897b MT |
2422 | kfree(ctx); |
2423 | err_ctx: | |
986a4f4d JW |
2424 | kfree(names); |
2425 | err_names: | |
2426 | kfree(callbacks); | |
2427 | err_callback: | |
2428 | kfree(vqs); | |
2429 | err_vq: | |
2430 | return ret; | |
2431 | } | |
2432 | ||
2433 | static int virtnet_alloc_queues(struct virtnet_info *vi) | |
2434 | { | |
2435 | int i; | |
2436 | ||
603343b9 MT |
2437 | vi->ctrl = kzalloc(sizeof(*vi->ctrl), GFP_KERNEL); |
2438 | if (!vi->ctrl) | |
2439 | goto err_ctrl; | |
986a4f4d JW |
2440 | vi->sq = kzalloc(sizeof(*vi->sq) * vi->max_queue_pairs, GFP_KERNEL); |
2441 | if (!vi->sq) | |
2442 | goto err_sq; | |
2443 | vi->rq = kzalloc(sizeof(*vi->rq) * vi->max_queue_pairs, GFP_KERNEL); | |
008d4278 | 2444 | if (!vi->rq) |
986a4f4d JW |
2445 | goto err_rq; |
2446 | ||
2447 | INIT_DELAYED_WORK(&vi->refill, refill_work); | |
2448 | for (i = 0; i < vi->max_queue_pairs; i++) { | |
2449 | vi->rq[i].pages = NULL; | |
2450 | netif_napi_add(vi->dev, &vi->rq[i].napi, virtnet_poll, | |
2451 | napi_weight); | |
1d11e732 WB |
2452 | netif_tx_napi_add(vi->dev, &vi->sq[i].napi, virtnet_poll_tx, |
2453 | napi_tx ? napi_weight : 0); | |
986a4f4d JW |
2454 | |
2455 | sg_init_table(vi->rq[i].sg, ARRAY_SIZE(vi->rq[i].sg)); | |
5377d758 | 2456 | ewma_pkt_len_init(&vi->rq[i].mrg_avg_pkt_len); |
986a4f4d JW |
2457 | sg_init_table(vi->sq[i].sg, ARRAY_SIZE(vi->sq[i].sg)); |
2458 | } | |
2459 | ||
2460 | return 0; | |
2461 | ||
2462 | err_rq: | |
2463 | kfree(vi->sq); | |
2464 | err_sq: | |
603343b9 MT |
2465 | kfree(vi->ctrl); |
2466 | err_ctrl: | |
986a4f4d JW |
2467 | return -ENOMEM; |
2468 | } | |
2469 | ||
2470 | static int init_vqs(struct virtnet_info *vi) | |
2471 | { | |
2472 | int ret; | |
2473 | ||
2474 | /* Allocate send & receive queues */ | |
2475 | ret = virtnet_alloc_queues(vi); | |
2476 | if (ret) | |
2477 | goto err; | |
2478 | ||
2479 | ret = virtnet_find_vqs(vi); | |
2480 | if (ret) | |
2481 | goto err_free; | |
2482 | ||
47be2479 | 2483 | get_online_cpus(); |
8898c21c | 2484 | virtnet_set_affinity(vi); |
47be2479 WG |
2485 | put_online_cpus(); |
2486 | ||
986a4f4d JW |
2487 | return 0; |
2488 | ||
2489 | err_free: | |
2490 | virtnet_free_queues(vi); | |
2491 | err: | |
2492 | return ret; | |
3f9c10b0 AS |
2493 | } |
2494 | ||
fbf28d78 MD |
2495 | #ifdef CONFIG_SYSFS |
2496 | static ssize_t mergeable_rx_buffer_size_show(struct netdev_rx_queue *queue, | |
718ad681 | 2497 | char *buf) |
fbf28d78 MD |
2498 | { |
2499 | struct virtnet_info *vi = netdev_priv(queue->dev); | |
2500 | unsigned int queue_index = get_netdev_rx_queue_index(queue); | |
5377d758 | 2501 | struct ewma_pkt_len *avg; |
fbf28d78 MD |
2502 | |
2503 | BUG_ON(queue_index >= vi->max_queue_pairs); | |
2504 | avg = &vi->rq[queue_index].mrg_avg_pkt_len; | |
d85b758f MT |
2505 | return sprintf(buf, "%u\n", |
2506 | get_mergeable_buf_len(&vi->rq[queue_index], avg)); | |
fbf28d78 MD |
2507 | } |
2508 | ||
2509 | static struct rx_queue_attribute mergeable_rx_buffer_size_attribute = | |
2510 | __ATTR_RO(mergeable_rx_buffer_size); | |
2511 | ||
2512 | static struct attribute *virtio_net_mrg_rx_attrs[] = { | |
2513 | &mergeable_rx_buffer_size_attribute.attr, | |
2514 | NULL | |
2515 | }; | |
2516 | ||
2517 | static const struct attribute_group virtio_net_mrg_rx_group = { | |
2518 | .name = "virtio_net", | |
2519 | .attrs = virtio_net_mrg_rx_attrs | |
2520 | }; | |
2521 | #endif | |
2522 | ||
892d6eb1 JW |
2523 | static bool virtnet_fail_on_feature(struct virtio_device *vdev, |
2524 | unsigned int fbit, | |
2525 | const char *fname, const char *dname) | |
2526 | { | |
2527 | if (!virtio_has_feature(vdev, fbit)) | |
2528 | return false; | |
2529 | ||
2530 | dev_err(&vdev->dev, "device advertises feature %s but not %s", | |
2531 | fname, dname); | |
2532 | ||
2533 | return true; | |
2534 | } | |
2535 | ||
2536 | #define VIRTNET_FAIL_ON(vdev, fbit, dbit) \ | |
2537 | virtnet_fail_on_feature(vdev, fbit, #fbit, dbit) | |
2538 | ||
2539 | static bool virtnet_validate_features(struct virtio_device *vdev) | |
2540 | { | |
2541 | if (!virtio_has_feature(vdev, VIRTIO_NET_F_CTRL_VQ) && | |
2542 | (VIRTNET_FAIL_ON(vdev, VIRTIO_NET_F_CTRL_RX, | |
2543 | "VIRTIO_NET_F_CTRL_VQ") || | |
2544 | VIRTNET_FAIL_ON(vdev, VIRTIO_NET_F_CTRL_VLAN, | |
2545 | "VIRTIO_NET_F_CTRL_VQ") || | |
2546 | VIRTNET_FAIL_ON(vdev, VIRTIO_NET_F_GUEST_ANNOUNCE, | |
2547 | "VIRTIO_NET_F_CTRL_VQ") || | |
2548 | VIRTNET_FAIL_ON(vdev, VIRTIO_NET_F_MQ, "VIRTIO_NET_F_CTRL_VQ") || | |
2549 | VIRTNET_FAIL_ON(vdev, VIRTIO_NET_F_CTRL_MAC_ADDR, | |
2550 | "VIRTIO_NET_F_CTRL_VQ"))) { | |
2551 | return false; | |
2552 | } | |
2553 | ||
2554 | return true; | |
2555 | } | |
2556 | ||
d0c2c997 JW |
2557 | #define MIN_MTU ETH_MIN_MTU |
2558 | #define MAX_MTU ETH_MAX_MTU | |
2559 | ||
fe36cbe0 | 2560 | static int virtnet_validate(struct virtio_device *vdev) |
296f96fc | 2561 | { |
6ba42248 MT |
2562 | if (!vdev->config->get) { |
2563 | dev_err(&vdev->dev, "%s failure: config access disabled\n", | |
2564 | __func__); | |
2565 | return -EINVAL; | |
2566 | } | |
2567 | ||
892d6eb1 JW |
2568 | if (!virtnet_validate_features(vdev)) |
2569 | return -EINVAL; | |
2570 | ||
fe36cbe0 MT |
2571 | if (virtio_has_feature(vdev, VIRTIO_NET_F_MTU)) { |
2572 | int mtu = virtio_cread16(vdev, | |
2573 | offsetof(struct virtio_net_config, | |
2574 | mtu)); | |
2575 | if (mtu < MIN_MTU) | |
2576 | __virtio_clear_bit(vdev, VIRTIO_NET_F_MTU); | |
2577 | } | |
2578 | ||
2579 | return 0; | |
2580 | } | |
2581 | ||
2582 | static int virtnet_probe(struct virtio_device *vdev) | |
2583 | { | |
2584 | int i, err; | |
2585 | struct net_device *dev; | |
2586 | struct virtnet_info *vi; | |
2587 | u16 max_queue_pairs; | |
2588 | int mtu; | |
2589 | ||
986a4f4d | 2590 | /* Find if host supports multiqueue virtio_net device */ |
855e0c52 RR |
2591 | err = virtio_cread_feature(vdev, VIRTIO_NET_F_MQ, |
2592 | struct virtio_net_config, | |
2593 | max_virtqueue_pairs, &max_queue_pairs); | |
986a4f4d JW |
2594 | |
2595 | /* We need at least 2 queue's */ | |
2596 | if (err || max_queue_pairs < VIRTIO_NET_CTRL_MQ_VQ_PAIRS_MIN || | |
2597 | max_queue_pairs > VIRTIO_NET_CTRL_MQ_VQ_PAIRS_MAX || | |
2598 | !virtio_has_feature(vdev, VIRTIO_NET_F_CTRL_VQ)) | |
2599 | max_queue_pairs = 1; | |
296f96fc RR |
2600 | |
2601 | /* Allocate ourselves a network device with room for our info */ | |
986a4f4d | 2602 | dev = alloc_etherdev_mq(sizeof(struct virtnet_info), max_queue_pairs); |
296f96fc RR |
2603 | if (!dev) |
2604 | return -ENOMEM; | |
2605 | ||
2606 | /* Set up network device as normal. */ | |
f2f2c8b4 | 2607 | dev->priv_flags |= IFF_UNICAST_FLT | IFF_LIVE_ADDR_CHANGE; |
76288b4e | 2608 | dev->netdev_ops = &virtnet_netdev; |
296f96fc | 2609 | dev->features = NETIF_F_HIGHDMA; |
3fa2a1df | 2610 | |
7ad24ea4 | 2611 | dev->ethtool_ops = &virtnet_ethtool_ops; |
296f96fc RR |
2612 | SET_NETDEV_DEV(dev, &vdev->dev); |
2613 | ||
2614 | /* Do we support "hardware" checksums? */ | |
98e778c9 | 2615 | if (virtio_has_feature(vdev, VIRTIO_NET_F_CSUM)) { |
296f96fc | 2616 | /* This opens up the world of extra features. */ |
48900cb6 | 2617 | dev->hw_features |= NETIF_F_HW_CSUM | NETIF_F_SG; |
98e778c9 | 2618 | if (csum) |
48900cb6 | 2619 | dev->features |= NETIF_F_HW_CSUM | NETIF_F_SG; |
98e778c9 MM |
2620 | |
2621 | if (virtio_has_feature(vdev, VIRTIO_NET_F_GSO)) { | |
e078de03 | 2622 | dev->hw_features |= NETIF_F_TSO |
34a48579 RR |
2623 | | NETIF_F_TSO_ECN | NETIF_F_TSO6; |
2624 | } | |
5539ae96 | 2625 | /* Individual feature bits: what can host handle? */ |
98e778c9 MM |
2626 | if (virtio_has_feature(vdev, VIRTIO_NET_F_HOST_TSO4)) |
2627 | dev->hw_features |= NETIF_F_TSO; | |
2628 | if (virtio_has_feature(vdev, VIRTIO_NET_F_HOST_TSO6)) | |
2629 | dev->hw_features |= NETIF_F_TSO6; | |
2630 | if (virtio_has_feature(vdev, VIRTIO_NET_F_HOST_ECN)) | |
2631 | dev->hw_features |= NETIF_F_TSO_ECN; | |
98e778c9 | 2632 | |
41f2f127 JW |
2633 | dev->features |= NETIF_F_GSO_ROBUST; |
2634 | ||
98e778c9 | 2635 | if (gso) |
e078de03 | 2636 | dev->features |= dev->hw_features & NETIF_F_ALL_TSO; |
98e778c9 | 2637 | /* (!csum && gso) case will be fixed by register_netdev() */ |
296f96fc | 2638 | } |
4f49129b TH |
2639 | if (virtio_has_feature(vdev, VIRTIO_NET_F_GUEST_CSUM)) |
2640 | dev->features |= NETIF_F_RXCSUM; | |
296f96fc | 2641 | |
4fda8302 JW |
2642 | dev->vlan_features = dev->features; |
2643 | ||
d0c2c997 JW |
2644 | /* MTU range: 68 - 65535 */ |
2645 | dev->min_mtu = MIN_MTU; | |
2646 | dev->max_mtu = MAX_MTU; | |
2647 | ||
296f96fc | 2648 | /* Configuration may specify what MAC to use. Otherwise random. */ |
855e0c52 RR |
2649 | if (virtio_has_feature(vdev, VIRTIO_NET_F_MAC)) |
2650 | virtio_cread_bytes(vdev, | |
2651 | offsetof(struct virtio_net_config, mac), | |
2652 | dev->dev_addr, dev->addr_len); | |
2653 | else | |
f2cedb63 | 2654 | eth_hw_addr_random(dev); |
296f96fc RR |
2655 | |
2656 | /* Set up our device-specific information */ | |
2657 | vi = netdev_priv(dev); | |
296f96fc RR |
2658 | vi->dev = dev; |
2659 | vi->vdev = vdev; | |
d9d5dcc8 | 2660 | vdev->priv = vi; |
3fa2a1df | 2661 | vi->stats = alloc_percpu(struct virtnet_stats); |
2662 | err = -ENOMEM; | |
2663 | if (vi->stats == NULL) | |
2664 | goto free; | |
2665 | ||
827da44c JS |
2666 | for_each_possible_cpu(i) { |
2667 | struct virtnet_stats *virtnet_stats; | |
2668 | virtnet_stats = per_cpu_ptr(vi->stats, i); | |
2669 | u64_stats_init(&virtnet_stats->tx_syncp); | |
2670 | u64_stats_init(&virtnet_stats->rx_syncp); | |
2671 | } | |
2672 | ||
586d17c5 | 2673 | INIT_WORK(&vi->config_work, virtnet_config_changed_work); |
296f96fc | 2674 | |
97402b96 | 2675 | /* If we can receive ANY GSO packets, we must allocate large ones. */ |
8e95a202 JP |
2676 | if (virtio_has_feature(vdev, VIRTIO_NET_F_GUEST_TSO4) || |
2677 | virtio_has_feature(vdev, VIRTIO_NET_F_GUEST_TSO6) || | |
e3e3c423 VY |
2678 | virtio_has_feature(vdev, VIRTIO_NET_F_GUEST_ECN) || |
2679 | virtio_has_feature(vdev, VIRTIO_NET_F_GUEST_UFO)) | |
97402b96 HX |
2680 | vi->big_packets = true; |
2681 | ||
3f2c31d9 MM |
2682 | if (virtio_has_feature(vdev, VIRTIO_NET_F_MRG_RXBUF)) |
2683 | vi->mergeable_rx_bufs = true; | |
2684 | ||
d04302b3 MT |
2685 | if (virtio_has_feature(vdev, VIRTIO_NET_F_MRG_RXBUF) || |
2686 | virtio_has_feature(vdev, VIRTIO_F_VERSION_1)) | |
012873d0 MT |
2687 | vi->hdr_len = sizeof(struct virtio_net_hdr_mrg_rxbuf); |
2688 | else | |
2689 | vi->hdr_len = sizeof(struct virtio_net_hdr); | |
2690 | ||
75993300 MT |
2691 | if (virtio_has_feature(vdev, VIRTIO_F_ANY_LAYOUT) || |
2692 | virtio_has_feature(vdev, VIRTIO_F_VERSION_1)) | |
e7428e95 MT |
2693 | vi->any_header_sg = true; |
2694 | ||
986a4f4d JW |
2695 | if (virtio_has_feature(vdev, VIRTIO_NET_F_CTRL_VQ)) |
2696 | vi->has_cvq = true; | |
2697 | ||
14de9d11 AC |
2698 | if (virtio_has_feature(vdev, VIRTIO_NET_F_MTU)) { |
2699 | mtu = virtio_cread16(vdev, | |
2700 | offsetof(struct virtio_net_config, | |
2701 | mtu)); | |
93a205ee | 2702 | if (mtu < dev->min_mtu) { |
fe36cbe0 MT |
2703 | /* Should never trigger: MTU was previously validated |
2704 | * in virtnet_validate. | |
2705 | */ | |
2706 | dev_err(&vdev->dev, "device MTU appears to have changed " | |
2707 | "it is now %d < %d", mtu, dev->min_mtu); | |
2708 | goto free_stats; | |
93a205ee | 2709 | } |
2e123b44 | 2710 | |
fe36cbe0 MT |
2711 | dev->mtu = mtu; |
2712 | dev->max_mtu = mtu; | |
2713 | ||
2e123b44 MT |
2714 | /* TODO: size buffers correctly in this case. */ |
2715 | if (dev->mtu > ETH_DATA_LEN) | |
2716 | vi->big_packets = true; | |
14de9d11 AC |
2717 | } |
2718 | ||
012873d0 MT |
2719 | if (vi->any_header_sg) |
2720 | dev->needed_headroom = vi->hdr_len; | |
6ebbc1a6 | 2721 | |
44900010 JW |
2722 | /* Enable multiqueue by default */ |
2723 | if (num_online_cpus() >= max_queue_pairs) | |
2724 | vi->curr_queue_pairs = max_queue_pairs; | |
2725 | else | |
2726 | vi->curr_queue_pairs = num_online_cpus(); | |
986a4f4d JW |
2727 | vi->max_queue_pairs = max_queue_pairs; |
2728 | ||
2729 | /* Allocate/initialize the rx/tx queues, and invoke find_vqs */ | |
3f9c10b0 | 2730 | err = init_vqs(vi); |
d2a7ddda | 2731 | if (err) |
9bb8ca86 | 2732 | goto free_stats; |
296f96fc | 2733 | |
fbf28d78 MD |
2734 | #ifdef CONFIG_SYSFS |
2735 | if (vi->mergeable_rx_bufs) | |
2736 | dev->sysfs_rx_queue_group = &virtio_net_mrg_rx_group; | |
2737 | #endif | |
0f13b66b ZYW |
2738 | netif_set_real_num_tx_queues(dev, vi->curr_queue_pairs); |
2739 | netif_set_real_num_rx_queues(dev, vi->curr_queue_pairs); | |
986a4f4d | 2740 | |
16032be5 NA |
2741 | virtnet_init_settings(dev); |
2742 | ||
296f96fc RR |
2743 | err = register_netdev(dev); |
2744 | if (err) { | |
2745 | pr_debug("virtio_net: registering device failed\n"); | |
d2a7ddda | 2746 | goto free_vqs; |
296f96fc | 2747 | } |
b3369c1f | 2748 | |
4baf1e33 MT |
2749 | virtio_device_ready(vdev); |
2750 | ||
8017c279 | 2751 | err = virtnet_cpu_notif_add(vi); |
8de4b2f3 WG |
2752 | if (err) { |
2753 | pr_debug("virtio_net: registering cpu notifier failed\n"); | |
f00e35e2 | 2754 | goto free_unregister_netdev; |
8de4b2f3 WG |
2755 | } |
2756 | ||
a220871b | 2757 | virtnet_set_queues(vi, vi->curr_queue_pairs); |
44900010 | 2758 | |
167c25e4 JW |
2759 | /* Assume link up if device can't report link status, |
2760 | otherwise get link status from config. */ | |
5bf6a429 | 2761 | netif_carrier_off(dev); |
167c25e4 | 2762 | if (virtio_has_feature(vi->vdev, VIRTIO_NET_F_STATUS)) { |
3b07e9ca | 2763 | schedule_work(&vi->config_work); |
167c25e4 JW |
2764 | } else { |
2765 | vi->status = VIRTIO_NET_S_LINK_UP; | |
2766 | netif_carrier_on(dev); | |
2767 | } | |
9f4d26d0 | 2768 | |
3f93522f JW |
2769 | for (i = 0; i < ARRAY_SIZE(guest_offloads); i++) |
2770 | if (virtio_has_feature(vi->vdev, guest_offloads[i])) | |
2771 | set_bit(guest_offloads[i], &vi->guest_offloads); | |
2772 | ||
986a4f4d JW |
2773 | pr_debug("virtnet: registered device %s with %d RX and TX vq's\n", |
2774 | dev->name, max_queue_pairs); | |
2775 | ||
296f96fc RR |
2776 | return 0; |
2777 | ||
f00e35e2 | 2778 | free_unregister_netdev: |
02465555 MT |
2779 | vi->vdev->config->reset(vdev); |
2780 | ||
b3369c1f | 2781 | unregister_netdev(dev); |
d2a7ddda | 2782 | free_vqs: |
986a4f4d | 2783 | cancel_delayed_work_sync(&vi->refill); |
fb51879d | 2784 | free_receive_page_frags(vi); |
e9d7417b | 2785 | virtnet_del_vqs(vi); |
3fa2a1df | 2786 | free_stats: |
2787 | free_percpu(vi->stats); | |
296f96fc RR |
2788 | free: |
2789 | free_netdev(dev); | |
2790 | return err; | |
2791 | } | |
2792 | ||
04486ed0 | 2793 | static void remove_vq_common(struct virtnet_info *vi) |
296f96fc | 2794 | { |
04486ed0 | 2795 | vi->vdev->config->reset(vi->vdev); |
830a8a97 SM |
2796 | |
2797 | /* Free unused buffers in both send and recv, if any. */ | |
9ab86bbc | 2798 | free_unused_bufs(vi); |
fb6813f4 | 2799 | |
986a4f4d | 2800 | free_receive_bufs(vi); |
d2a7ddda | 2801 | |
fb51879d MD |
2802 | free_receive_page_frags(vi); |
2803 | ||
986a4f4d | 2804 | virtnet_del_vqs(vi); |
04486ed0 AS |
2805 | } |
2806 | ||
8cc085d6 | 2807 | static void virtnet_remove(struct virtio_device *vdev) |
04486ed0 AS |
2808 | { |
2809 | struct virtnet_info *vi = vdev->priv; | |
2810 | ||
8017c279 | 2811 | virtnet_cpu_notif_remove(vi); |
8de4b2f3 | 2812 | |
102a2786 MT |
2813 | /* Make sure no work handler is accessing the device. */ |
2814 | flush_work(&vi->config_work); | |
586d17c5 | 2815 | |
04486ed0 AS |
2816 | unregister_netdev(vi->dev); |
2817 | ||
2818 | remove_vq_common(vi); | |
fb6813f4 | 2819 | |
2e66f55b | 2820 | free_percpu(vi->stats); |
74b2553f | 2821 | free_netdev(vi->dev); |
296f96fc RR |
2822 | } |
2823 | ||
67a75194 | 2824 | static __maybe_unused int virtnet_freeze(struct virtio_device *vdev) |
0741bcb5 AS |
2825 | { |
2826 | struct virtnet_info *vi = vdev->priv; | |
2827 | ||
8017c279 | 2828 | virtnet_cpu_notif_remove(vi); |
9fe7bfce | 2829 | virtnet_freeze_down(vdev); |
0741bcb5 AS |
2830 | remove_vq_common(vi); |
2831 | ||
2832 | return 0; | |
2833 | } | |
2834 | ||
67a75194 | 2835 | static __maybe_unused int virtnet_restore(struct virtio_device *vdev) |
0741bcb5 AS |
2836 | { |
2837 | struct virtnet_info *vi = vdev->priv; | |
9fe7bfce | 2838 | int err; |
0741bcb5 | 2839 | |
9fe7bfce | 2840 | err = virtnet_restore_up(vdev); |
0741bcb5 AS |
2841 | if (err) |
2842 | return err; | |
986a4f4d JW |
2843 | virtnet_set_queues(vi, vi->curr_queue_pairs); |
2844 | ||
8017c279 | 2845 | err = virtnet_cpu_notif_add(vi); |
ec9debbd JW |
2846 | if (err) |
2847 | return err; | |
2848 | ||
0741bcb5 AS |
2849 | return 0; |
2850 | } | |
0741bcb5 | 2851 | |
296f96fc RR |
2852 | static struct virtio_device_id id_table[] = { |
2853 | { VIRTIO_ID_NET, VIRTIO_DEV_ANY_ID }, | |
2854 | { 0 }, | |
2855 | }; | |
2856 | ||
f3358507 MT |
2857 | #define VIRTNET_FEATURES \ |
2858 | VIRTIO_NET_F_CSUM, VIRTIO_NET_F_GUEST_CSUM, \ | |
2859 | VIRTIO_NET_F_MAC, \ | |
2860 | VIRTIO_NET_F_HOST_TSO4, VIRTIO_NET_F_HOST_UFO, VIRTIO_NET_F_HOST_TSO6, \ | |
2861 | VIRTIO_NET_F_HOST_ECN, VIRTIO_NET_F_GUEST_TSO4, VIRTIO_NET_F_GUEST_TSO6, \ | |
2862 | VIRTIO_NET_F_GUEST_ECN, VIRTIO_NET_F_GUEST_UFO, \ | |
2863 | VIRTIO_NET_F_MRG_RXBUF, VIRTIO_NET_F_STATUS, VIRTIO_NET_F_CTRL_VQ, \ | |
2864 | VIRTIO_NET_F_CTRL_RX, VIRTIO_NET_F_CTRL_VLAN, \ | |
2865 | VIRTIO_NET_F_GUEST_ANNOUNCE, VIRTIO_NET_F_MQ, \ | |
2866 | VIRTIO_NET_F_CTRL_MAC_ADDR, \ | |
3f93522f | 2867 | VIRTIO_NET_F_MTU, VIRTIO_NET_F_CTRL_GUEST_OFFLOADS |
f3358507 | 2868 | |
c45a6816 | 2869 | static unsigned int features[] = { |
f3358507 MT |
2870 | VIRTNET_FEATURES, |
2871 | }; | |
2872 | ||
2873 | static unsigned int features_legacy[] = { | |
2874 | VIRTNET_FEATURES, | |
2875 | VIRTIO_NET_F_GSO, | |
e7428e95 | 2876 | VIRTIO_F_ANY_LAYOUT, |
c45a6816 RR |
2877 | }; |
2878 | ||
22402529 | 2879 | static struct virtio_driver virtio_net_driver = { |
c45a6816 RR |
2880 | .feature_table = features, |
2881 | .feature_table_size = ARRAY_SIZE(features), | |
f3358507 MT |
2882 | .feature_table_legacy = features_legacy, |
2883 | .feature_table_size_legacy = ARRAY_SIZE(features_legacy), | |
296f96fc RR |
2884 | .driver.name = KBUILD_MODNAME, |
2885 | .driver.owner = THIS_MODULE, | |
2886 | .id_table = id_table, | |
fe36cbe0 | 2887 | .validate = virtnet_validate, |
296f96fc | 2888 | .probe = virtnet_probe, |
8cc085d6 | 2889 | .remove = virtnet_remove, |
9f4d26d0 | 2890 | .config_changed = virtnet_config_changed, |
89107000 | 2891 | #ifdef CONFIG_PM_SLEEP |
0741bcb5 AS |
2892 | .freeze = virtnet_freeze, |
2893 | .restore = virtnet_restore, | |
2894 | #endif | |
296f96fc RR |
2895 | }; |
2896 | ||
8017c279 SAS |
2897 | static __init int virtio_net_driver_init(void) |
2898 | { | |
2899 | int ret; | |
2900 | ||
73c1b41e | 2901 | ret = cpuhp_setup_state_multi(CPUHP_AP_ONLINE_DYN, "virtio/net:online", |
8017c279 SAS |
2902 | virtnet_cpu_online, |
2903 | virtnet_cpu_down_prep); | |
2904 | if (ret < 0) | |
2905 | goto out; | |
2906 | virtionet_online = ret; | |
73c1b41e | 2907 | ret = cpuhp_setup_state_multi(CPUHP_VIRT_NET_DEAD, "virtio/net:dead", |
8017c279 SAS |
2908 | NULL, virtnet_cpu_dead); |
2909 | if (ret) | |
2910 | goto err_dead; | |
2911 | ||
2912 | ret = register_virtio_driver(&virtio_net_driver); | |
2913 | if (ret) | |
2914 | goto err_virtio; | |
2915 | return 0; | |
2916 | err_virtio: | |
2917 | cpuhp_remove_multi_state(CPUHP_VIRT_NET_DEAD); | |
2918 | err_dead: | |
2919 | cpuhp_remove_multi_state(virtionet_online); | |
2920 | out: | |
2921 | return ret; | |
2922 | } | |
2923 | module_init(virtio_net_driver_init); | |
2924 | ||
2925 | static __exit void virtio_net_driver_exit(void) | |
2926 | { | |
cfa0ebc9 | 2927 | unregister_virtio_driver(&virtio_net_driver); |
8017c279 SAS |
2928 | cpuhp_remove_multi_state(CPUHP_VIRT_NET_DEAD); |
2929 | cpuhp_remove_multi_state(virtionet_online); | |
8017c279 SAS |
2930 | } |
2931 | module_exit(virtio_net_driver_exit); | |
296f96fc RR |
2932 | |
2933 | MODULE_DEVICE_TABLE(virtio, id_table); | |
2934 | MODULE_DESCRIPTION("Virtio network driver"); | |
2935 | MODULE_LICENSE("GPL"); |