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> |
ba5e4426 SS |
33 | #include <linux/netdevice.h> |
34 | #include <linux/pci.h> | |
d85b758f | 35 | #include <net/route.h> |
754b8a21 | 36 | #include <net/xdp.h> |
ba5e4426 | 37 | #include <net/net_failover.h> |
296f96fc | 38 | |
d34710e3 | 39 | static int napi_weight = NAPI_POLL_WEIGHT; |
6c0cd7c0 DL |
40 | module_param(napi_weight, int, 0444); |
41 | ||
b92f1e67 | 42 | static bool csum = true, gso = true, napi_tx; |
34a48579 RR |
43 | module_param(csum, bool, 0444); |
44 | module_param(gso, bool, 0444); | |
b92f1e67 | 45 | module_param(napi_tx, bool, 0644); |
34a48579 | 46 | |
296f96fc | 47 | /* FIXME: MTU in config. */ |
5061de36 | 48 | #define GOOD_PACKET_LEN (ETH_HLEN + VLAN_HLEN + ETH_DATA_LEN) |
3f2c31d9 | 49 | #define GOOD_COPY_LEN 128 |
296f96fc | 50 | |
f6b10209 JW |
51 | #define VIRTNET_RX_PAD (NET_IP_ALIGN + NET_SKB_PAD) |
52 | ||
2de2f7f4 JF |
53 | /* Amount of XDP headroom to prepend to packets for use by xdp_adjust_head */ |
54 | #define VIRTIO_XDP_HEADROOM 256 | |
55 | ||
5377d758 JB |
56 | /* RX packet size EWMA. The average packet size is used to determine the packet |
57 | * buffer size when refilling RX rings. As the entire RX ring may be refilled | |
58 | * at once, the weight is chosen so that the EWMA will be insensitive to short- | |
59 | * term, transient changes in packet size. | |
ab7db917 | 60 | */ |
eb1e011a | 61 | DECLARE_EWMA(pkt_len, 0, 64) |
ab7db917 | 62 | |
66846048 | 63 | #define VIRTNET_DRIVER_VERSION "1.0.0" |
2a41f71d | 64 | |
7acd4329 CIK |
65 | static const unsigned long guest_offloads[] = { |
66 | VIRTIO_NET_F_GUEST_TSO4, | |
67 | VIRTIO_NET_F_GUEST_TSO6, | |
68 | VIRTIO_NET_F_GUEST_ECN, | |
69 | VIRTIO_NET_F_GUEST_UFO | |
70 | }; | |
3f93522f | 71 | |
d7dfc5cf TM |
72 | struct virtnet_stat_desc { |
73 | char desc[ETH_GSTRING_LEN]; | |
74 | size_t offset; | |
3fa2a1df | 75 | }; |
76 | ||
d7dfc5cf TM |
77 | struct virtnet_sq_stats { |
78 | struct u64_stats_sync syncp; | |
79 | u64 packets; | |
80 | u64 bytes; | |
81 | }; | |
82 | ||
83 | struct virtnet_rq_stats { | |
84 | struct u64_stats_sync syncp; | |
85 | u64 packets; | |
86 | u64 bytes; | |
87 | }; | |
88 | ||
89 | #define VIRTNET_SQ_STAT(m) offsetof(struct virtnet_sq_stats, m) | |
90 | #define VIRTNET_RQ_STAT(m) offsetof(struct virtnet_rq_stats, m) | |
91 | ||
92 | static const struct virtnet_stat_desc virtnet_sq_stats_desc[] = { | |
93 | { "packets", VIRTNET_SQ_STAT(packets) }, | |
94 | { "bytes", VIRTNET_SQ_STAT(bytes) }, | |
95 | }; | |
96 | ||
97 | static const struct virtnet_stat_desc virtnet_rq_stats_desc[] = { | |
98 | { "packets", VIRTNET_RQ_STAT(packets) }, | |
99 | { "bytes", VIRTNET_RQ_STAT(bytes) }, | |
100 | }; | |
101 | ||
102 | #define VIRTNET_SQ_STATS_LEN ARRAY_SIZE(virtnet_sq_stats_desc) | |
103 | #define VIRTNET_RQ_STATS_LEN ARRAY_SIZE(virtnet_rq_stats_desc) | |
104 | ||
e9d7417b JW |
105 | /* Internal representation of a send virtqueue */ |
106 | struct send_queue { | |
107 | /* Virtqueue associated with this send _queue */ | |
108 | struct virtqueue *vq; | |
109 | ||
110 | /* TX: fragments + linear part + virtio header */ | |
111 | struct scatterlist sg[MAX_SKB_FRAGS + 2]; | |
986a4f4d JW |
112 | |
113 | /* Name of the send queue: output.$index */ | |
114 | char name[40]; | |
b92f1e67 | 115 | |
d7dfc5cf TM |
116 | struct virtnet_sq_stats stats; |
117 | ||
b92f1e67 | 118 | struct napi_struct napi; |
e9d7417b JW |
119 | }; |
120 | ||
121 | /* Internal representation of a receive virtqueue */ | |
122 | struct receive_queue { | |
123 | /* Virtqueue associated with this receive_queue */ | |
124 | struct virtqueue *vq; | |
125 | ||
296f96fc RR |
126 | struct napi_struct napi; |
127 | ||
f600b690 JF |
128 | struct bpf_prog __rcu *xdp_prog; |
129 | ||
d7dfc5cf TM |
130 | struct virtnet_rq_stats stats; |
131 | ||
e9d7417b JW |
132 | /* Chain pages by the private ptr. */ |
133 | struct page *pages; | |
134 | ||
ab7db917 | 135 | /* Average packet length for mergeable receive buffers. */ |
5377d758 | 136 | struct ewma_pkt_len mrg_avg_pkt_len; |
ab7db917 | 137 | |
fb51879d MD |
138 | /* Page frag for packet buffer allocation. */ |
139 | struct page_frag alloc_frag; | |
140 | ||
e9d7417b JW |
141 | /* RX: fragments + linear part + virtio header */ |
142 | struct scatterlist sg[MAX_SKB_FRAGS + 2]; | |
986a4f4d | 143 | |
d85b758f MT |
144 | /* Min single buffer size for mergeable buffers case. */ |
145 | unsigned int min_buf_len; | |
146 | ||
986a4f4d JW |
147 | /* Name of this receive queue: input.$index */ |
148 | char name[40]; | |
754b8a21 JDB |
149 | |
150 | struct xdp_rxq_info xdp_rxq; | |
e9d7417b JW |
151 | }; |
152 | ||
12e57169 MT |
153 | /* Control VQ buffers: protected by the rtnl lock */ |
154 | struct control_buf { | |
155 | struct virtio_net_ctrl_hdr hdr; | |
156 | virtio_net_ctrl_ack status; | |
157 | struct virtio_net_ctrl_mq mq; | |
158 | u8 promisc; | |
159 | u8 allmulti; | |
d7fad4c8 | 160 | __virtio16 vid; |
f4ee703a | 161 | __virtio64 offloads; |
12e57169 MT |
162 | }; |
163 | ||
e9d7417b JW |
164 | struct virtnet_info { |
165 | struct virtio_device *vdev; | |
166 | struct virtqueue *cvq; | |
167 | struct net_device *dev; | |
986a4f4d JW |
168 | struct send_queue *sq; |
169 | struct receive_queue *rq; | |
e9d7417b JW |
170 | unsigned int status; |
171 | ||
986a4f4d JW |
172 | /* Max # of queue pairs supported by the device */ |
173 | u16 max_queue_pairs; | |
174 | ||
175 | /* # of queue pairs currently used by the driver */ | |
176 | u16 curr_queue_pairs; | |
177 | ||
672aafd5 JF |
178 | /* # of XDP queue pairs currently used by the driver */ |
179 | u16 xdp_queue_pairs; | |
180 | ||
97402b96 HX |
181 | /* I like... big packets and I cannot lie! */ |
182 | bool big_packets; | |
183 | ||
3f2c31d9 MM |
184 | /* Host will merge rx buffers for big packets (shake it! shake it!) */ |
185 | bool mergeable_rx_bufs; | |
186 | ||
986a4f4d JW |
187 | /* Has control virtqueue */ |
188 | bool has_cvq; | |
189 | ||
e7428e95 MT |
190 | /* Host can handle any s/g split between our header and packet data */ |
191 | bool any_header_sg; | |
192 | ||
012873d0 MT |
193 | /* Packet virtio header size */ |
194 | u8 hdr_len; | |
195 | ||
3161e453 RR |
196 | /* Work struct for refilling if we run low on memory. */ |
197 | struct delayed_work refill; | |
198 | ||
586d17c5 JW |
199 | /* Work struct for config space updates */ |
200 | struct work_struct config_work; | |
201 | ||
986a4f4d JW |
202 | /* Does the affinity hint is set for virtqueues? */ |
203 | bool affinity_hint_set; | |
47be2479 | 204 | |
8017c279 SAS |
205 | /* CPU hotplug instances for online & dead */ |
206 | struct hlist_node node; | |
207 | struct hlist_node node_dead; | |
2ac46030 | 208 | |
12e57169 | 209 | struct control_buf *ctrl; |
16032be5 NA |
210 | |
211 | /* Ethtool settings */ | |
212 | u8 duplex; | |
213 | u32 speed; | |
3f93522f JW |
214 | |
215 | unsigned long guest_offloads; | |
ba5e4426 SS |
216 | |
217 | /* failover when STANDBY feature enabled */ | |
218 | struct failover *failover; | |
296f96fc RR |
219 | }; |
220 | ||
9ab86bbc | 221 | struct padded_vnet_hdr { |
012873d0 | 222 | struct virtio_net_hdr_mrg_rxbuf hdr; |
9ab86bbc | 223 | /* |
012873d0 MT |
224 | * hdr is in a separate sg buffer, and data sg buffer shares same page |
225 | * with this header sg. This padding makes next sg 16 byte aligned | |
226 | * after the header. | |
9ab86bbc | 227 | */ |
012873d0 | 228 | char padding[4]; |
9ab86bbc SM |
229 | }; |
230 | ||
986a4f4d JW |
231 | /* Converting between virtqueue no. and kernel tx/rx queue no. |
232 | * 0:rx0 1:tx0 2:rx1 3:tx1 ... 2N:rxN 2N+1:txN 2N+2:cvq | |
233 | */ | |
234 | static int vq2txq(struct virtqueue *vq) | |
235 | { | |
9d0ca6ed | 236 | return (vq->index - 1) / 2; |
986a4f4d JW |
237 | } |
238 | ||
239 | static int txq2vq(int txq) | |
240 | { | |
241 | return txq * 2 + 1; | |
242 | } | |
243 | ||
244 | static int vq2rxq(struct virtqueue *vq) | |
245 | { | |
9d0ca6ed | 246 | return vq->index / 2; |
986a4f4d JW |
247 | } |
248 | ||
249 | static int rxq2vq(int rxq) | |
250 | { | |
251 | return rxq * 2; | |
252 | } | |
253 | ||
012873d0 | 254 | static inline struct virtio_net_hdr_mrg_rxbuf *skb_vnet_hdr(struct sk_buff *skb) |
296f96fc | 255 | { |
012873d0 | 256 | return (struct virtio_net_hdr_mrg_rxbuf *)skb->cb; |
296f96fc RR |
257 | } |
258 | ||
9ab86bbc SM |
259 | /* |
260 | * private is used to chain pages for big packets, put the whole | |
261 | * most recent used list in the beginning for reuse | |
262 | */ | |
e9d7417b | 263 | static void give_pages(struct receive_queue *rq, struct page *page) |
0a888fd1 | 264 | { |
9ab86bbc | 265 | struct page *end; |
0a888fd1 | 266 | |
e9d7417b | 267 | /* Find end of list, sew whole thing into vi->rq.pages. */ |
9ab86bbc | 268 | for (end = page; end->private; end = (struct page *)end->private); |
e9d7417b JW |
269 | end->private = (unsigned long)rq->pages; |
270 | rq->pages = page; | |
0a888fd1 MM |
271 | } |
272 | ||
e9d7417b | 273 | static struct page *get_a_page(struct receive_queue *rq, gfp_t gfp_mask) |
fb6813f4 | 274 | { |
e9d7417b | 275 | struct page *p = rq->pages; |
fb6813f4 | 276 | |
9ab86bbc | 277 | if (p) { |
e9d7417b | 278 | rq->pages = (struct page *)p->private; |
9ab86bbc SM |
279 | /* clear private here, it is used to chain pages */ |
280 | p->private = 0; | |
281 | } else | |
fb6813f4 RR |
282 | p = alloc_page(gfp_mask); |
283 | return p; | |
284 | } | |
285 | ||
e4e8452a WB |
286 | static void virtqueue_napi_schedule(struct napi_struct *napi, |
287 | struct virtqueue *vq) | |
288 | { | |
289 | if (napi_schedule_prep(napi)) { | |
290 | virtqueue_disable_cb(vq); | |
291 | __napi_schedule(napi); | |
292 | } | |
293 | } | |
294 | ||
295 | static void virtqueue_napi_complete(struct napi_struct *napi, | |
296 | struct virtqueue *vq, int processed) | |
297 | { | |
298 | int opaque; | |
299 | ||
300 | opaque = virtqueue_enable_cb_prepare(vq); | |
fdaa767a TM |
301 | if (napi_complete_done(napi, processed)) { |
302 | if (unlikely(virtqueue_poll(vq, opaque))) | |
303 | virtqueue_napi_schedule(napi, vq); | |
304 | } else { | |
305 | virtqueue_disable_cb(vq); | |
306 | } | |
e4e8452a WB |
307 | } |
308 | ||
e9d7417b | 309 | static void skb_xmit_done(struct virtqueue *vq) |
296f96fc | 310 | { |
e9d7417b | 311 | struct virtnet_info *vi = vq->vdev->priv; |
b92f1e67 | 312 | struct napi_struct *napi = &vi->sq[vq2txq(vq)].napi; |
296f96fc | 313 | |
2cb9c6ba | 314 | /* Suppress further interrupts. */ |
e9d7417b | 315 | virtqueue_disable_cb(vq); |
11a3a154 | 316 | |
b92f1e67 WB |
317 | if (napi->weight) |
318 | virtqueue_napi_schedule(napi, vq); | |
319 | else | |
320 | /* We were probably waiting for more output buffers. */ | |
321 | netif_wake_subqueue(vi->dev, vq2txq(vq)); | |
296f96fc RR |
322 | } |
323 | ||
28b39bc7 JW |
324 | #define MRG_CTX_HEADER_SHIFT 22 |
325 | static void *mergeable_len_to_ctx(unsigned int truesize, | |
326 | unsigned int headroom) | |
327 | { | |
328 | return (void *)(unsigned long)((headroom << MRG_CTX_HEADER_SHIFT) | truesize); | |
329 | } | |
330 | ||
331 | static unsigned int mergeable_ctx_to_headroom(void *mrg_ctx) | |
332 | { | |
333 | return (unsigned long)mrg_ctx >> MRG_CTX_HEADER_SHIFT; | |
334 | } | |
335 | ||
336 | static unsigned int mergeable_ctx_to_truesize(void *mrg_ctx) | |
337 | { | |
338 | return (unsigned long)mrg_ctx & ((1 << MRG_CTX_HEADER_SHIFT) - 1); | |
339 | } | |
340 | ||
3464645a | 341 | /* Called from bottom half context */ |
946fa564 MT |
342 | static struct sk_buff *page_to_skb(struct virtnet_info *vi, |
343 | struct receive_queue *rq, | |
2613af0e MD |
344 | struct page *page, unsigned int offset, |
345 | unsigned int len, unsigned int truesize) | |
9ab86bbc SM |
346 | { |
347 | struct sk_buff *skb; | |
012873d0 | 348 | struct virtio_net_hdr_mrg_rxbuf *hdr; |
2613af0e | 349 | unsigned int copy, hdr_len, hdr_padded_len; |
9ab86bbc | 350 | char *p; |
fb6813f4 | 351 | |
2613af0e | 352 | p = page_address(page) + offset; |
3f2c31d9 | 353 | |
9ab86bbc | 354 | /* copy small packet so we can reuse these pages for small data */ |
c67f5db8 | 355 | skb = napi_alloc_skb(&rq->napi, GOOD_COPY_LEN); |
9ab86bbc SM |
356 | if (unlikely(!skb)) |
357 | return NULL; | |
3f2c31d9 | 358 | |
9ab86bbc | 359 | hdr = skb_vnet_hdr(skb); |
3f2c31d9 | 360 | |
012873d0 MT |
361 | hdr_len = vi->hdr_len; |
362 | if (vi->mergeable_rx_bufs) | |
a4a76503 | 363 | hdr_padded_len = sizeof(*hdr); |
012873d0 | 364 | else |
2613af0e | 365 | hdr_padded_len = sizeof(struct padded_vnet_hdr); |
3f2c31d9 | 366 | |
9ab86bbc | 367 | memcpy(hdr, p, hdr_len); |
3f2c31d9 | 368 | |
9ab86bbc | 369 | len -= hdr_len; |
2613af0e MD |
370 | offset += hdr_padded_len; |
371 | p += hdr_padded_len; | |
3f2c31d9 | 372 | |
9ab86bbc SM |
373 | copy = len; |
374 | if (copy > skb_tailroom(skb)) | |
375 | copy = skb_tailroom(skb); | |
59ae1d12 | 376 | skb_put_data(skb, p, copy); |
3f2c31d9 | 377 | |
9ab86bbc SM |
378 | len -= copy; |
379 | offset += copy; | |
3f2c31d9 | 380 | |
2613af0e MD |
381 | if (vi->mergeable_rx_bufs) { |
382 | if (len) | |
383 | skb_add_rx_frag(skb, 0, page, offset, len, truesize); | |
384 | else | |
385 | put_page(page); | |
386 | return skb; | |
387 | } | |
388 | ||
e878d78b SL |
389 | /* |
390 | * Verify that we can indeed put this data into a skb. | |
391 | * This is here to handle cases when the device erroneously | |
392 | * tries to receive more than is possible. This is usually | |
393 | * the case of a broken device. | |
394 | */ | |
395 | if (unlikely(len > MAX_SKB_FRAGS * PAGE_SIZE)) { | |
be443899 | 396 | net_dbg_ratelimited("%s: too much data\n", skb->dev->name); |
e878d78b SL |
397 | dev_kfree_skb(skb); |
398 | return NULL; | |
399 | } | |
2613af0e | 400 | BUG_ON(offset >= PAGE_SIZE); |
9ab86bbc | 401 | while (len) { |
2613af0e MD |
402 | unsigned int frag_size = min((unsigned)PAGE_SIZE - offset, len); |
403 | skb_add_rx_frag(skb, skb_shinfo(skb)->nr_frags, page, offset, | |
404 | frag_size, truesize); | |
405 | len -= frag_size; | |
9ab86bbc SM |
406 | page = (struct page *)page->private; |
407 | offset = 0; | |
408 | } | |
3f2c31d9 | 409 | |
9ab86bbc | 410 | if (page) |
e9d7417b | 411 | give_pages(rq, page); |
3f2c31d9 | 412 | |
9ab86bbc SM |
413 | return skb; |
414 | } | |
3f2c31d9 | 415 | |
735fc405 JDB |
416 | static int __virtnet_xdp_xmit_one(struct virtnet_info *vi, |
417 | struct send_queue *sq, | |
418 | struct xdp_frame *xdpf) | |
56434a01 | 419 | { |
56434a01 | 420 | struct virtio_net_hdr_mrg_rxbuf *hdr; |
56434a01 JF |
421 | int err; |
422 | ||
cac320c8 JDB |
423 | /* virtqueue want to use data area in-front of packet */ |
424 | if (unlikely(xdpf->metasize > 0)) | |
425 | return -EOPNOTSUPP; | |
56434a01 | 426 | |
cac320c8 JDB |
427 | if (unlikely(xdpf->headroom < vi->hdr_len)) |
428 | return -EOVERFLOW; | |
429 | ||
430 | /* Make room for virtqueue hdr (also change xdpf->headroom?) */ | |
431 | xdpf->data -= vi->hdr_len; | |
f6b10209 | 432 | /* Zero header and leave csum up to XDP layers */ |
cac320c8 | 433 | hdr = xdpf->data; |
f6b10209 | 434 | memset(hdr, 0, vi->hdr_len); |
cac320c8 | 435 | xdpf->len += vi->hdr_len; |
bb91accf | 436 | |
cac320c8 | 437 | sg_init_one(sq->sg, xdpf->data, xdpf->len); |
bb91accf | 438 | |
cac320c8 | 439 | err = virtqueue_add_outbuf(sq->vq, sq->sg, 1, xdpf, GFP_ATOMIC); |
11b7d897 | 440 | if (unlikely(err)) |
cac320c8 | 441 | return -ENOSPC; /* Caller handle free/refcnt */ |
56434a01 | 442 | |
cac320c8 | 443 | return 0; |
56434a01 JF |
444 | } |
445 | ||
735fc405 JDB |
446 | static int __virtnet_xdp_tx_xmit(struct virtnet_info *vi, |
447 | struct xdp_frame *xdpf) | |
448 | { | |
449 | struct xdp_frame *xdpf_sent; | |
450 | struct send_queue *sq; | |
451 | unsigned int len; | |
452 | unsigned int qp; | |
453 | ||
454 | qp = vi->curr_queue_pairs - vi->xdp_queue_pairs + smp_processor_id(); | |
455 | sq = &vi->sq[qp]; | |
456 | ||
457 | /* Free up any pending old buffers before queueing new ones. */ | |
458 | while ((xdpf_sent = virtqueue_get_buf(sq->vq, &len)) != NULL) | |
459 | xdp_return_frame(xdpf_sent); | |
460 | ||
461 | return __virtnet_xdp_xmit_one(vi, sq, xdpf); | |
462 | } | |
463 | ||
464 | static int virtnet_xdp_xmit(struct net_device *dev, | |
42b33468 | 465 | int n, struct xdp_frame **frames, u32 flags) |
186b3c99 JW |
466 | { |
467 | struct virtnet_info *vi = netdev_priv(dev); | |
8dcc5b0a | 468 | struct receive_queue *rq = vi->rq; |
735fc405 | 469 | struct xdp_frame *xdpf_sent; |
8dcc5b0a | 470 | struct bpf_prog *xdp_prog; |
735fc405 JDB |
471 | struct send_queue *sq; |
472 | unsigned int len; | |
473 | unsigned int qp; | |
474 | int drops = 0; | |
475 | int err; | |
476 | int i; | |
477 | ||
5d274cb4 | 478 | if (unlikely(flags & ~XDP_XMIT_FLAGS_MASK)) |
42b33468 JDB |
479 | return -EINVAL; |
480 | ||
735fc405 JDB |
481 | qp = vi->curr_queue_pairs - vi->xdp_queue_pairs + smp_processor_id(); |
482 | sq = &vi->sq[qp]; | |
186b3c99 | 483 | |
8dcc5b0a JDB |
484 | /* Only allow ndo_xdp_xmit if XDP is loaded on dev, as this |
485 | * indicate XDP resources have been successfully allocated. | |
486 | */ | |
487 | xdp_prog = rcu_dereference(rq->xdp_prog); | |
488 | if (!xdp_prog) | |
489 | return -ENXIO; | |
490 | ||
735fc405 JDB |
491 | /* Free up any pending old buffers before queueing new ones. */ |
492 | while ((xdpf_sent = virtqueue_get_buf(sq->vq, &len)) != NULL) | |
493 | xdp_return_frame(xdpf_sent); | |
494 | ||
495 | for (i = 0; i < n; i++) { | |
496 | struct xdp_frame *xdpf = frames[i]; | |
497 | ||
498 | err = __virtnet_xdp_xmit_one(vi, sq, xdpf); | |
499 | if (err) { | |
500 | xdp_return_frame_rx_napi(xdpf); | |
501 | drops++; | |
502 | } | |
503 | } | |
5d274cb4 JDB |
504 | |
505 | if (flags & XDP_XMIT_FLUSH) | |
506 | virtqueue_kick(sq->vq); | |
507 | ||
735fc405 | 508 | return n - drops; |
186b3c99 JW |
509 | } |
510 | ||
f6b10209 JW |
511 | static unsigned int virtnet_get_headroom(struct virtnet_info *vi) |
512 | { | |
513 | return vi->xdp_queue_pairs ? VIRTIO_XDP_HEADROOM : 0; | |
514 | } | |
515 | ||
4941d472 JW |
516 | /* We copy the packet for XDP in the following cases: |
517 | * | |
518 | * 1) Packet is scattered across multiple rx buffers. | |
519 | * 2) Headroom space is insufficient. | |
520 | * | |
521 | * This is inefficient but it's a temporary condition that | |
522 | * we hit right after XDP is enabled and until queue is refilled | |
523 | * with large buffers with sufficient headroom - so it should affect | |
524 | * at most queue size packets. | |
525 | * Afterwards, the conditions to enable | |
526 | * XDP should preclude the underlying device from sending packets | |
527 | * across multiple buffers (num_buf > 1), and we make sure buffers | |
528 | * have enough headroom. | |
529 | */ | |
530 | static struct page *xdp_linearize_page(struct receive_queue *rq, | |
531 | u16 *num_buf, | |
532 | struct page *p, | |
533 | int offset, | |
534 | int page_off, | |
535 | unsigned int *len) | |
536 | { | |
537 | struct page *page = alloc_page(GFP_ATOMIC); | |
538 | ||
539 | if (!page) | |
540 | return NULL; | |
541 | ||
542 | memcpy(page_address(page) + page_off, page_address(p) + offset, *len); | |
543 | page_off += *len; | |
544 | ||
545 | while (--*num_buf) { | |
3cc81a9a | 546 | int tailroom = SKB_DATA_ALIGN(sizeof(struct skb_shared_info)); |
4941d472 JW |
547 | unsigned int buflen; |
548 | void *buf; | |
549 | int off; | |
550 | ||
551 | buf = virtqueue_get_buf(rq->vq, &buflen); | |
552 | if (unlikely(!buf)) | |
553 | goto err_buf; | |
554 | ||
555 | p = virt_to_head_page(buf); | |
556 | off = buf - page_address(p); | |
557 | ||
558 | /* guard against a misconfigured or uncooperative backend that | |
559 | * is sending packet larger than the MTU. | |
560 | */ | |
3cc81a9a | 561 | if ((page_off + buflen + tailroom) > PAGE_SIZE) { |
4941d472 JW |
562 | put_page(p); |
563 | goto err_buf; | |
564 | } | |
565 | ||
566 | memcpy(page_address(page) + page_off, | |
567 | page_address(p) + off, buflen); | |
568 | page_off += buflen; | |
569 | put_page(p); | |
570 | } | |
571 | ||
572 | /* Headroom does not contribute to packet length */ | |
573 | *len = page_off - VIRTIO_XDP_HEADROOM; | |
574 | return page; | |
575 | err_buf: | |
576 | __free_pages(page, 0); | |
577 | return NULL; | |
578 | } | |
579 | ||
bb91accf JW |
580 | static struct sk_buff *receive_small(struct net_device *dev, |
581 | struct virtnet_info *vi, | |
582 | struct receive_queue *rq, | |
192f68cf | 583 | void *buf, void *ctx, |
186b3c99 JW |
584 | unsigned int len, |
585 | bool *xdp_xmit) | |
f121159d | 586 | { |
f6b10209 | 587 | struct sk_buff *skb; |
bb91accf | 588 | struct bpf_prog *xdp_prog; |
4941d472 | 589 | unsigned int xdp_headroom = (unsigned long)ctx; |
f6b10209 JW |
590 | unsigned int header_offset = VIRTNET_RX_PAD + xdp_headroom; |
591 | unsigned int headroom = vi->hdr_len + header_offset; | |
592 | unsigned int buflen = SKB_DATA_ALIGN(GOOD_PACKET_LEN + headroom) + | |
593 | SKB_DATA_ALIGN(sizeof(struct skb_shared_info)); | |
4941d472 | 594 | struct page *page = virt_to_head_page(buf); |
11b7d897 | 595 | unsigned int delta = 0; |
4941d472 | 596 | struct page *xdp_page; |
11b7d897 JDB |
597 | int err; |
598 | ||
012873d0 | 599 | len -= vi->hdr_len; |
f121159d | 600 | |
bb91accf JW |
601 | rcu_read_lock(); |
602 | xdp_prog = rcu_dereference(rq->xdp_prog); | |
603 | if (xdp_prog) { | |
f6b10209 | 604 | struct virtio_net_hdr_mrg_rxbuf *hdr = buf + header_offset; |
44fa2dbd | 605 | struct xdp_frame *xdpf; |
0354e4d1 | 606 | struct xdp_buff xdp; |
f6b10209 | 607 | void *orig_data; |
bb91accf JW |
608 | u32 act; |
609 | ||
95dbe9e7 | 610 | if (unlikely(hdr->hdr.gso_type)) |
bb91accf | 611 | goto err_xdp; |
0354e4d1 | 612 | |
4941d472 JW |
613 | if (unlikely(xdp_headroom < virtnet_get_headroom(vi))) { |
614 | int offset = buf - page_address(page) + header_offset; | |
615 | unsigned int tlen = len + vi->hdr_len; | |
616 | u16 num_buf = 1; | |
617 | ||
618 | xdp_headroom = virtnet_get_headroom(vi); | |
619 | header_offset = VIRTNET_RX_PAD + xdp_headroom; | |
620 | headroom = vi->hdr_len + header_offset; | |
621 | buflen = SKB_DATA_ALIGN(GOOD_PACKET_LEN + headroom) + | |
622 | SKB_DATA_ALIGN(sizeof(struct skb_shared_info)); | |
623 | xdp_page = xdp_linearize_page(rq, &num_buf, page, | |
624 | offset, header_offset, | |
625 | &tlen); | |
626 | if (!xdp_page) | |
627 | goto err_xdp; | |
628 | ||
629 | buf = page_address(xdp_page); | |
630 | put_page(page); | |
631 | page = xdp_page; | |
632 | } | |
633 | ||
f6b10209 JW |
634 | xdp.data_hard_start = buf + VIRTNET_RX_PAD + vi->hdr_len; |
635 | xdp.data = xdp.data_hard_start + xdp_headroom; | |
de8f3a83 | 636 | xdp_set_data_meta_invalid(&xdp); |
0354e4d1 | 637 | xdp.data_end = xdp.data + len; |
754b8a21 | 638 | xdp.rxq = &rq->xdp_rxq; |
f6b10209 | 639 | orig_data = xdp.data; |
0354e4d1 JF |
640 | act = bpf_prog_run_xdp(xdp_prog, &xdp); |
641 | ||
bb91accf JW |
642 | switch (act) { |
643 | case XDP_PASS: | |
2de2f7f4 | 644 | /* Recalculate length in case bpf program changed it */ |
f6b10209 | 645 | delta = orig_data - xdp.data; |
6870de43 | 646 | len = xdp.data_end - xdp.data; |
bb91accf JW |
647 | break; |
648 | case XDP_TX: | |
44fa2dbd JDB |
649 | xdpf = convert_to_xdp_frame(&xdp); |
650 | if (unlikely(!xdpf)) | |
651 | goto err_xdp; | |
735fc405 | 652 | err = __virtnet_xdp_tx_xmit(vi, xdpf); |
cac320c8 | 653 | if (unlikely(err)) { |
0354e4d1 | 654 | trace_xdp_exception(vi->dev, xdp_prog, act); |
11b7d897 JDB |
655 | goto err_xdp; |
656 | } | |
657 | *xdp_xmit = true; | |
186b3c99 JW |
658 | rcu_read_unlock(); |
659 | goto xdp_xmit; | |
660 | case XDP_REDIRECT: | |
661 | err = xdp_do_redirect(dev, &xdp, xdp_prog); | |
11b7d897 JDB |
662 | if (err) |
663 | goto err_xdp; | |
664 | *xdp_xmit = true; | |
bb91accf JW |
665 | rcu_read_unlock(); |
666 | goto xdp_xmit; | |
bb91accf | 667 | default: |
0354e4d1 JF |
668 | bpf_warn_invalid_xdp_action(act); |
669 | case XDP_ABORTED: | |
670 | trace_xdp_exception(vi->dev, xdp_prog, act); | |
671 | case XDP_DROP: | |
bb91accf JW |
672 | goto err_xdp; |
673 | } | |
674 | } | |
675 | rcu_read_unlock(); | |
676 | ||
f6b10209 JW |
677 | skb = build_skb(buf, buflen); |
678 | if (!skb) { | |
4941d472 | 679 | put_page(page); |
f6b10209 JW |
680 | goto err; |
681 | } | |
682 | skb_reserve(skb, headroom - delta); | |
6870de43 | 683 | skb_put(skb, len); |
f6b10209 JW |
684 | if (!delta) { |
685 | buf += header_offset; | |
686 | memcpy(skb_vnet_hdr(skb), buf, vi->hdr_len); | |
687 | } /* keep zeroed vnet hdr since packet was changed by bpf */ | |
688 | ||
689 | err: | |
f121159d | 690 | return skb; |
bb91accf JW |
691 | |
692 | err_xdp: | |
693 | rcu_read_unlock(); | |
694 | dev->stats.rx_dropped++; | |
4941d472 | 695 | put_page(page); |
bb91accf JW |
696 | xdp_xmit: |
697 | return NULL; | |
f121159d MT |
698 | } |
699 | ||
700 | static struct sk_buff *receive_big(struct net_device *dev, | |
946fa564 | 701 | struct virtnet_info *vi, |
f121159d MT |
702 | struct receive_queue *rq, |
703 | void *buf, | |
704 | unsigned int len) | |
705 | { | |
706 | struct page *page = buf; | |
c47a43d3 | 707 | struct sk_buff *skb = page_to_skb(vi, rq, page, 0, len, PAGE_SIZE); |
f600b690 | 708 | |
f121159d MT |
709 | if (unlikely(!skb)) |
710 | goto err; | |
711 | ||
712 | return skb; | |
713 | ||
714 | err: | |
715 | dev->stats.rx_dropped++; | |
716 | give_pages(rq, page); | |
717 | return NULL; | |
718 | } | |
719 | ||
8fc3b9e9 | 720 | static struct sk_buff *receive_mergeable(struct net_device *dev, |
fdd819b2 | 721 | struct virtnet_info *vi, |
8fc3b9e9 | 722 | struct receive_queue *rq, |
680557cf MT |
723 | void *buf, |
724 | void *ctx, | |
186b3c99 JW |
725 | unsigned int len, |
726 | bool *xdp_xmit) | |
9ab86bbc | 727 | { |
012873d0 MT |
728 | struct virtio_net_hdr_mrg_rxbuf *hdr = buf; |
729 | u16 num_buf = virtio16_to_cpu(vi->vdev, hdr->num_buffers); | |
8fc3b9e9 MT |
730 | struct page *page = virt_to_head_page(buf); |
731 | int offset = buf - page_address(page); | |
f600b690 JF |
732 | struct sk_buff *head_skb, *curr_skb; |
733 | struct bpf_prog *xdp_prog; | |
734 | unsigned int truesize; | |
4941d472 | 735 | unsigned int headroom = mergeable_ctx_to_headroom(ctx); |
3cc81a9a | 736 | int err; |
f600b690 | 737 | |
56434a01 JF |
738 | head_skb = NULL; |
739 | ||
f600b690 JF |
740 | rcu_read_lock(); |
741 | xdp_prog = rcu_dereference(rq->xdp_prog); | |
742 | if (xdp_prog) { | |
44fa2dbd | 743 | struct xdp_frame *xdpf; |
72979a6c | 744 | struct page *xdp_page; |
0354e4d1 | 745 | struct xdp_buff xdp; |
0354e4d1 | 746 | void *data; |
f600b690 JF |
747 | u32 act; |
748 | ||
3d62b2a0 JW |
749 | /* Transient failure which in theory could occur if |
750 | * in-flight packets from before XDP was enabled reach | |
751 | * the receive path after XDP is loaded. | |
752 | */ | |
753 | if (unlikely(hdr->hdr.gso_type)) | |
754 | goto err_xdp; | |
755 | ||
3cc81a9a JW |
756 | /* This happens when rx buffer size is underestimated |
757 | * or headroom is not enough because of the buffer | |
758 | * was refilled before XDP is set. This should only | |
759 | * happen for the first several packets, so we don't | |
760 | * care much about its performance. | |
761 | */ | |
4941d472 JW |
762 | if (unlikely(num_buf > 1 || |
763 | headroom < virtnet_get_headroom(vi))) { | |
72979a6c | 764 | /* linearize data for XDP */ |
56a86f84 | 765 | xdp_page = xdp_linearize_page(rq, &num_buf, |
4941d472 JW |
766 | page, offset, |
767 | VIRTIO_XDP_HEADROOM, | |
768 | &len); | |
72979a6c JF |
769 | if (!xdp_page) |
770 | goto err_xdp; | |
2de2f7f4 | 771 | offset = VIRTIO_XDP_HEADROOM; |
72979a6c JF |
772 | } else { |
773 | xdp_page = page; | |
f600b690 JF |
774 | } |
775 | ||
2de2f7f4 JF |
776 | /* Allow consuming headroom but reserve enough space to push |
777 | * the descriptor on if we get an XDP_TX return code. | |
778 | */ | |
0354e4d1 | 779 | data = page_address(xdp_page) + offset; |
2de2f7f4 | 780 | xdp.data_hard_start = data - VIRTIO_XDP_HEADROOM + vi->hdr_len; |
0354e4d1 | 781 | xdp.data = data + vi->hdr_len; |
de8f3a83 | 782 | xdp_set_data_meta_invalid(&xdp); |
0354e4d1 | 783 | xdp.data_end = xdp.data + (len - vi->hdr_len); |
754b8a21 JDB |
784 | xdp.rxq = &rq->xdp_rxq; |
785 | ||
0354e4d1 JF |
786 | act = bpf_prog_run_xdp(xdp_prog, &xdp); |
787 | ||
56434a01 JF |
788 | switch (act) { |
789 | case XDP_PASS: | |
2de2f7f4 JF |
790 | /* recalculate offset to account for any header |
791 | * adjustments. Note other cases do not build an | |
792 | * skb and avoid using offset | |
793 | */ | |
794 | offset = xdp.data - | |
795 | page_address(xdp_page) - vi->hdr_len; | |
796 | ||
6870de43 NS |
797 | /* recalculate len if xdp.data or xdp.data_end were |
798 | * adjusted | |
799 | */ | |
aaa64527 | 800 | len = xdp.data_end - xdp.data + vi->hdr_len; |
1830f893 JW |
801 | /* We can only create skb based on xdp_page. */ |
802 | if (unlikely(xdp_page != page)) { | |
803 | rcu_read_unlock(); | |
804 | put_page(page); | |
805 | head_skb = page_to_skb(vi, rq, xdp_page, | |
2de2f7f4 | 806 | offset, len, PAGE_SIZE); |
1830f893 JW |
807 | return head_skb; |
808 | } | |
56434a01 JF |
809 | break; |
810 | case XDP_TX: | |
44fa2dbd JDB |
811 | xdpf = convert_to_xdp_frame(&xdp); |
812 | if (unlikely(!xdpf)) | |
813 | goto err_xdp; | |
735fc405 | 814 | err = __virtnet_xdp_tx_xmit(vi, xdpf); |
cac320c8 | 815 | if (unlikely(err)) { |
0354e4d1 | 816 | trace_xdp_exception(vi->dev, xdp_prog, act); |
11b7d897 JDB |
817 | if (unlikely(xdp_page != page)) |
818 | put_page(xdp_page); | |
819 | goto err_xdp; | |
820 | } | |
821 | *xdp_xmit = true; | |
72979a6c | 822 | if (unlikely(xdp_page != page)) |
5d458a13 | 823 | put_page(page); |
56434a01 JF |
824 | rcu_read_unlock(); |
825 | goto xdp_xmit; | |
3cc81a9a JW |
826 | case XDP_REDIRECT: |
827 | err = xdp_do_redirect(dev, &xdp, xdp_prog); | |
828 | if (err) { | |
829 | if (unlikely(xdp_page != page)) | |
830 | put_page(xdp_page); | |
831 | goto err_xdp; | |
832 | } | |
833 | *xdp_xmit = true; | |
834 | if (unlikely(xdp_page != page)) | |
6890418b | 835 | put_page(page); |
3cc81a9a JW |
836 | rcu_read_unlock(); |
837 | goto xdp_xmit; | |
56434a01 | 838 | default: |
0354e4d1 JF |
839 | bpf_warn_invalid_xdp_action(act); |
840 | case XDP_ABORTED: | |
841 | trace_xdp_exception(vi->dev, xdp_prog, act); | |
842 | case XDP_DROP: | |
72979a6c JF |
843 | if (unlikely(xdp_page != page)) |
844 | __free_pages(xdp_page, 0); | |
f600b690 | 845 | goto err_xdp; |
56434a01 | 846 | } |
f600b690 JF |
847 | } |
848 | rcu_read_unlock(); | |
ab7db917 | 849 | |
28b39bc7 JW |
850 | truesize = mergeable_ctx_to_truesize(ctx); |
851 | if (unlikely(len > truesize)) { | |
56da5fd0 | 852 | pr_debug("%s: rx error: len %u exceeds truesize %lu\n", |
680557cf MT |
853 | dev->name, len, (unsigned long)ctx); |
854 | dev->stats.rx_length_errors++; | |
855 | goto err_skb; | |
856 | } | |
28b39bc7 | 857 | |
f600b690 JF |
858 | head_skb = page_to_skb(vi, rq, page, offset, len, truesize); |
859 | curr_skb = head_skb; | |
9ab86bbc | 860 | |
8fc3b9e9 MT |
861 | if (unlikely(!curr_skb)) |
862 | goto err_skb; | |
9ab86bbc | 863 | while (--num_buf) { |
8fc3b9e9 MT |
864 | int num_skb_frags; |
865 | ||
680557cf | 866 | buf = virtqueue_get_buf_ctx(rq->vq, &len, &ctx); |
03e9f8a0 | 867 | if (unlikely(!buf)) { |
8fc3b9e9 | 868 | pr_debug("%s: rx error: %d buffers out of %d missing\n", |
fdd819b2 | 869 | dev->name, num_buf, |
012873d0 MT |
870 | virtio16_to_cpu(vi->vdev, |
871 | hdr->num_buffers)); | |
8fc3b9e9 MT |
872 | dev->stats.rx_length_errors++; |
873 | goto err_buf; | |
3f2c31d9 | 874 | } |
8fc3b9e9 MT |
875 | |
876 | page = virt_to_head_page(buf); | |
28b39bc7 JW |
877 | |
878 | truesize = mergeable_ctx_to_truesize(ctx); | |
879 | if (unlikely(len > truesize)) { | |
56da5fd0 | 880 | pr_debug("%s: rx error: len %u exceeds truesize %lu\n", |
680557cf MT |
881 | dev->name, len, (unsigned long)ctx); |
882 | dev->stats.rx_length_errors++; | |
883 | goto err_skb; | |
884 | } | |
8fc3b9e9 MT |
885 | |
886 | num_skb_frags = skb_shinfo(curr_skb)->nr_frags; | |
2613af0e MD |
887 | if (unlikely(num_skb_frags == MAX_SKB_FRAGS)) { |
888 | struct sk_buff *nskb = alloc_skb(0, GFP_ATOMIC); | |
8fc3b9e9 MT |
889 | |
890 | if (unlikely(!nskb)) | |
891 | goto err_skb; | |
2613af0e MD |
892 | if (curr_skb == head_skb) |
893 | skb_shinfo(curr_skb)->frag_list = nskb; | |
894 | else | |
895 | curr_skb->next = nskb; | |
896 | curr_skb = nskb; | |
897 | head_skb->truesize += nskb->truesize; | |
898 | num_skb_frags = 0; | |
899 | } | |
900 | if (curr_skb != head_skb) { | |
901 | head_skb->data_len += len; | |
902 | head_skb->len += len; | |
fb51879d | 903 | head_skb->truesize += truesize; |
2613af0e | 904 | } |
8fc3b9e9 | 905 | offset = buf - page_address(page); |
ba275241 JW |
906 | if (skb_can_coalesce(curr_skb, num_skb_frags, page, offset)) { |
907 | put_page(page); | |
908 | skb_coalesce_rx_frag(curr_skb, num_skb_frags - 1, | |
fb51879d | 909 | len, truesize); |
ba275241 JW |
910 | } else { |
911 | skb_add_rx_frag(curr_skb, num_skb_frags, page, | |
fb51879d | 912 | offset, len, truesize); |
ba275241 | 913 | } |
8fc3b9e9 MT |
914 | } |
915 | ||
5377d758 | 916 | ewma_pkt_len_add(&rq->mrg_avg_pkt_len, head_skb->len); |
8fc3b9e9 MT |
917 | return head_skb; |
918 | ||
f600b690 JF |
919 | err_xdp: |
920 | rcu_read_unlock(); | |
8fc3b9e9 MT |
921 | err_skb: |
922 | put_page(page); | |
850e088d | 923 | while (num_buf-- > 1) { |
680557cf MT |
924 | buf = virtqueue_get_buf(rq->vq, &len); |
925 | if (unlikely(!buf)) { | |
8fc3b9e9 MT |
926 | pr_debug("%s: rx error: %d buffers missing\n", |
927 | dev->name, num_buf); | |
928 | dev->stats.rx_length_errors++; | |
929 | break; | |
930 | } | |
680557cf | 931 | page = virt_to_head_page(buf); |
8fc3b9e9 | 932 | put_page(page); |
9ab86bbc | 933 | } |
8fc3b9e9 MT |
934 | err_buf: |
935 | dev->stats.rx_dropped++; | |
936 | dev_kfree_skb(head_skb); | |
56434a01 | 937 | xdp_xmit: |
8fc3b9e9 | 938 | return NULL; |
9ab86bbc SM |
939 | } |
940 | ||
61845d20 | 941 | static int receive_buf(struct virtnet_info *vi, struct receive_queue *rq, |
186b3c99 | 942 | void *buf, unsigned int len, void **ctx, bool *xdp_xmit) |
9ab86bbc | 943 | { |
e9d7417b | 944 | struct net_device *dev = vi->dev; |
9ab86bbc | 945 | struct sk_buff *skb; |
012873d0 | 946 | struct virtio_net_hdr_mrg_rxbuf *hdr; |
61845d20 | 947 | int ret; |
3f2c31d9 | 948 | |
bcff3162 | 949 | if (unlikely(len < vi->hdr_len + ETH_HLEN)) { |
9ab86bbc SM |
950 | pr_debug("%s: short packet %i\n", dev->name, len); |
951 | dev->stats.rx_length_errors++; | |
ab7db917 | 952 | if (vi->mergeable_rx_bufs) { |
680557cf | 953 | put_page(virt_to_head_page(buf)); |
ab7db917 | 954 | } else if (vi->big_packets) { |
98bfd23c | 955 | give_pages(rq, buf); |
ab7db917 | 956 | } else { |
f6b10209 | 957 | put_page(virt_to_head_page(buf)); |
ab7db917 | 958 | } |
61845d20 | 959 | return 0; |
9ab86bbc | 960 | } |
3f2c31d9 | 961 | |
f121159d | 962 | if (vi->mergeable_rx_bufs) |
186b3c99 | 963 | skb = receive_mergeable(dev, vi, rq, buf, ctx, len, xdp_xmit); |
f121159d | 964 | else if (vi->big_packets) |
946fa564 | 965 | skb = receive_big(dev, vi, rq, buf, len); |
f121159d | 966 | else |
186b3c99 | 967 | skb = receive_small(dev, vi, rq, buf, ctx, len, xdp_xmit); |
f121159d MT |
968 | |
969 | if (unlikely(!skb)) | |
61845d20 | 970 | return 0; |
3f2c31d9 | 971 | |
9ab86bbc | 972 | hdr = skb_vnet_hdr(skb); |
3fa2a1df | 973 | |
61845d20 | 974 | ret = skb->len; |
296f96fc | 975 | |
e858fae2 | 976 | if (hdr->hdr.flags & VIRTIO_NET_HDR_F_DATA_VALID) |
10a8d94a | 977 | skb->ip_summed = CHECKSUM_UNNECESSARY; |
296f96fc | 978 | |
e858fae2 MR |
979 | if (virtio_net_hdr_to_skb(skb, &hdr->hdr, |
980 | virtio_is_little_endian(vi->vdev))) { | |
981 | net_warn_ratelimited("%s: bad gso: type: %u, size: %u\n", | |
982 | dev->name, hdr->hdr.gso_type, | |
983 | hdr->hdr.gso_size); | |
984 | goto frame_err; | |
296f96fc RR |
985 | } |
986 | ||
d1dc06dc MR |
987 | skb->protocol = eth_type_trans(skb, dev); |
988 | pr_debug("Receiving skb proto 0x%04x len %i type %i\n", | |
989 | ntohs(skb->protocol), skb->len, skb->pkt_type); | |
990 | ||
0fbd050a | 991 | napi_gro_receive(&rq->napi, skb); |
61845d20 | 992 | return ret; |
296f96fc RR |
993 | |
994 | frame_err: | |
995 | dev->stats.rx_frame_errors++; | |
296f96fc | 996 | dev_kfree_skb(skb); |
61845d20 | 997 | return 0; |
296f96fc RR |
998 | } |
999 | ||
192f68cf JW |
1000 | /* Unlike mergeable buffers, all buffers are allocated to the |
1001 | * same size, except for the headroom. For this reason we do | |
1002 | * not need to use mergeable_len_to_ctx here - it is enough | |
1003 | * to store the headroom as the context ignoring the truesize. | |
1004 | */ | |
946fa564 MT |
1005 | static int add_recvbuf_small(struct virtnet_info *vi, struct receive_queue *rq, |
1006 | gfp_t gfp) | |
296f96fc | 1007 | { |
f6b10209 JW |
1008 | struct page_frag *alloc_frag = &rq->alloc_frag; |
1009 | char *buf; | |
2de2f7f4 | 1010 | unsigned int xdp_headroom = virtnet_get_headroom(vi); |
192f68cf | 1011 | void *ctx = (void *)(unsigned long)xdp_headroom; |
f6b10209 | 1012 | int len = vi->hdr_len + VIRTNET_RX_PAD + GOOD_PACKET_LEN + xdp_headroom; |
9ab86bbc | 1013 | int err; |
3f2c31d9 | 1014 | |
f6b10209 JW |
1015 | len = SKB_DATA_ALIGN(len) + |
1016 | SKB_DATA_ALIGN(sizeof(struct skb_shared_info)); | |
1017 | if (unlikely(!skb_page_frag_refill(len, alloc_frag, gfp))) | |
9ab86bbc | 1018 | return -ENOMEM; |
296f96fc | 1019 | |
f6b10209 JW |
1020 | buf = (char *)page_address(alloc_frag->page) + alloc_frag->offset; |
1021 | get_page(alloc_frag->page); | |
1022 | alloc_frag->offset += len; | |
1023 | sg_init_one(rq->sg, buf + VIRTNET_RX_PAD + xdp_headroom, | |
1024 | vi->hdr_len + GOOD_PACKET_LEN); | |
192f68cf | 1025 | err = virtqueue_add_inbuf_ctx(rq->vq, rq->sg, 1, buf, ctx, gfp); |
9ab86bbc | 1026 | if (err < 0) |
f6b10209 | 1027 | put_page(virt_to_head_page(buf)); |
9ab86bbc SM |
1028 | return err; |
1029 | } | |
97402b96 | 1030 | |
012873d0 MT |
1031 | static int add_recvbuf_big(struct virtnet_info *vi, struct receive_queue *rq, |
1032 | gfp_t gfp) | |
9ab86bbc | 1033 | { |
9ab86bbc SM |
1034 | struct page *first, *list = NULL; |
1035 | char *p; | |
1036 | int i, err, offset; | |
1037 | ||
a5835440 RR |
1038 | sg_init_table(rq->sg, MAX_SKB_FRAGS + 2); |
1039 | ||
e9d7417b | 1040 | /* page in rq->sg[MAX_SKB_FRAGS + 1] is list tail */ |
9ab86bbc | 1041 | for (i = MAX_SKB_FRAGS + 1; i > 1; --i) { |
e9d7417b | 1042 | first = get_a_page(rq, gfp); |
9ab86bbc SM |
1043 | if (!first) { |
1044 | if (list) | |
e9d7417b | 1045 | give_pages(rq, list); |
9ab86bbc | 1046 | return -ENOMEM; |
97402b96 | 1047 | } |
e9d7417b | 1048 | sg_set_buf(&rq->sg[i], page_address(first), PAGE_SIZE); |
97402b96 | 1049 | |
9ab86bbc SM |
1050 | /* chain new page in list head to match sg */ |
1051 | first->private = (unsigned long)list; | |
1052 | list = first; | |
1053 | } | |
296f96fc | 1054 | |
e9d7417b | 1055 | first = get_a_page(rq, gfp); |
9ab86bbc | 1056 | if (!first) { |
e9d7417b | 1057 | give_pages(rq, list); |
9ab86bbc SM |
1058 | return -ENOMEM; |
1059 | } | |
1060 | p = page_address(first); | |
1061 | ||
e9d7417b | 1062 | /* rq->sg[0], rq->sg[1] share the same page */ |
012873d0 MT |
1063 | /* a separated rq->sg[0] for header - required in case !any_header_sg */ |
1064 | sg_set_buf(&rq->sg[0], p, vi->hdr_len); | |
9ab86bbc | 1065 | |
e9d7417b | 1066 | /* rq->sg[1] for data packet, from offset */ |
9ab86bbc | 1067 | offset = sizeof(struct padded_vnet_hdr); |
e9d7417b | 1068 | sg_set_buf(&rq->sg[1], p + offset, PAGE_SIZE - offset); |
9ab86bbc SM |
1069 | |
1070 | /* chain first in list head */ | |
1071 | first->private = (unsigned long)list; | |
9dc7b9e4 RR |
1072 | err = virtqueue_add_inbuf(rq->vq, rq->sg, MAX_SKB_FRAGS + 2, |
1073 | first, gfp); | |
9ab86bbc | 1074 | if (err < 0) |
e9d7417b | 1075 | give_pages(rq, first); |
9ab86bbc SM |
1076 | |
1077 | return err; | |
296f96fc RR |
1078 | } |
1079 | ||
d85b758f | 1080 | static unsigned int get_mergeable_buf_len(struct receive_queue *rq, |
3cc81a9a JW |
1081 | struct ewma_pkt_len *avg_pkt_len, |
1082 | unsigned int room) | |
3f2c31d9 | 1083 | { |
ab7db917 | 1084 | const size_t hdr_len = sizeof(struct virtio_net_hdr_mrg_rxbuf); |
fbf28d78 MD |
1085 | unsigned int len; |
1086 | ||
3cc81a9a JW |
1087 | if (room) |
1088 | return PAGE_SIZE - room; | |
1089 | ||
1090 | len = hdr_len + clamp_t(unsigned int, ewma_pkt_len_read(avg_pkt_len), | |
f0c3192c | 1091 | rq->min_buf_len, PAGE_SIZE - hdr_len); |
3cc81a9a | 1092 | |
e377fcc8 | 1093 | return ALIGN(len, L1_CACHE_BYTES); |
fbf28d78 MD |
1094 | } |
1095 | ||
2de2f7f4 JF |
1096 | static int add_recvbuf_mergeable(struct virtnet_info *vi, |
1097 | struct receive_queue *rq, gfp_t gfp) | |
fbf28d78 | 1098 | { |
fb51879d | 1099 | struct page_frag *alloc_frag = &rq->alloc_frag; |
2de2f7f4 | 1100 | unsigned int headroom = virtnet_get_headroom(vi); |
3cc81a9a JW |
1101 | unsigned int tailroom = headroom ? sizeof(struct skb_shared_info) : 0; |
1102 | unsigned int room = SKB_DATA_ALIGN(headroom + tailroom); | |
fb51879d | 1103 | char *buf; |
680557cf | 1104 | void *ctx; |
3f2c31d9 | 1105 | int err; |
fb51879d | 1106 | unsigned int len, hole; |
3f2c31d9 | 1107 | |
3cc81a9a JW |
1108 | /* Extra tailroom is needed to satisfy XDP's assumption. This |
1109 | * means rx frags coalescing won't work, but consider we've | |
1110 | * disabled GSO for XDP, it won't be a big issue. | |
1111 | */ | |
1112 | len = get_mergeable_buf_len(rq, &rq->mrg_avg_pkt_len, room); | |
1113 | if (unlikely(!skb_page_frag_refill(len + room, alloc_frag, gfp))) | |
9ab86bbc | 1114 | return -ENOMEM; |
ab7db917 | 1115 | |
fb51879d | 1116 | buf = (char *)page_address(alloc_frag->page) + alloc_frag->offset; |
2de2f7f4 | 1117 | buf += headroom; /* advance address leaving hole at front of pkt */ |
fb51879d | 1118 | get_page(alloc_frag->page); |
3cc81a9a | 1119 | alloc_frag->offset += len + room; |
fb51879d | 1120 | hole = alloc_frag->size - alloc_frag->offset; |
3cc81a9a | 1121 | if (hole < len + room) { |
ab7db917 MD |
1122 | /* To avoid internal fragmentation, if there is very likely not |
1123 | * enough space for another buffer, add the remaining space to | |
1daa8790 | 1124 | * the current buffer. |
ab7db917 | 1125 | */ |
fb51879d MD |
1126 | len += hole; |
1127 | alloc_frag->offset += hole; | |
1128 | } | |
3f2c31d9 | 1129 | |
fb51879d | 1130 | sg_init_one(rq->sg, buf, len); |
29fda25a | 1131 | ctx = mergeable_len_to_ctx(len, headroom); |
680557cf | 1132 | err = virtqueue_add_inbuf_ctx(rq->vq, rq->sg, 1, buf, ctx, gfp); |
9ab86bbc | 1133 | if (err < 0) |
2613af0e | 1134 | put_page(virt_to_head_page(buf)); |
3f2c31d9 | 1135 | |
9ab86bbc SM |
1136 | return err; |
1137 | } | |
3f2c31d9 | 1138 | |
b2baed69 RR |
1139 | /* |
1140 | * Returns false if we couldn't fill entirely (OOM). | |
1141 | * | |
1142 | * Normally run in the receive path, but can also be run from ndo_open | |
1143 | * before we're receiving packets, or from refill_work which is | |
1144 | * careful to disable receiving (using napi_disable). | |
1145 | */ | |
946fa564 MT |
1146 | static bool try_fill_recv(struct virtnet_info *vi, struct receive_queue *rq, |
1147 | gfp_t gfp) | |
9ab86bbc SM |
1148 | { |
1149 | int err; | |
1788f495 | 1150 | bool oom; |
3f2c31d9 | 1151 | |
9ab86bbc SM |
1152 | do { |
1153 | if (vi->mergeable_rx_bufs) | |
2de2f7f4 | 1154 | err = add_recvbuf_mergeable(vi, rq, gfp); |
9ab86bbc | 1155 | else if (vi->big_packets) |
012873d0 | 1156 | err = add_recvbuf_big(vi, rq, gfp); |
9ab86bbc | 1157 | else |
946fa564 | 1158 | err = add_recvbuf_small(vi, rq, gfp); |
3f2c31d9 | 1159 | |
1788f495 | 1160 | oom = err == -ENOMEM; |
9ed4cb07 | 1161 | if (err) |
3f2c31d9 | 1162 | break; |
b7dfde95 | 1163 | } while (rq->vq->num_free); |
681daee2 | 1164 | virtqueue_kick(rq->vq); |
3161e453 | 1165 | return !oom; |
3f2c31d9 MM |
1166 | } |
1167 | ||
18445c4d | 1168 | static void skb_recv_done(struct virtqueue *rvq) |
296f96fc RR |
1169 | { |
1170 | struct virtnet_info *vi = rvq->vdev->priv; | |
986a4f4d | 1171 | struct receive_queue *rq = &vi->rq[vq2rxq(rvq)]; |
e9d7417b | 1172 | |
e4e8452a | 1173 | virtqueue_napi_schedule(&rq->napi, rvq); |
296f96fc RR |
1174 | } |
1175 | ||
e4e8452a | 1176 | static void virtnet_napi_enable(struct virtqueue *vq, struct napi_struct *napi) |
3e9d08ec | 1177 | { |
e4e8452a | 1178 | napi_enable(napi); |
3e9d08ec BR |
1179 | |
1180 | /* If all buffers were filled by other side before we napi_enabled, we | |
e4e8452a WB |
1181 | * won't get another interrupt, so process any outstanding packets now. |
1182 | * Call local_bh_enable after to trigger softIRQ processing. | |
1183 | */ | |
1184 | local_bh_disable(); | |
1185 | virtqueue_napi_schedule(napi, vq); | |
1186 | local_bh_enable(); | |
3e9d08ec BR |
1187 | } |
1188 | ||
b92f1e67 WB |
1189 | static void virtnet_napi_tx_enable(struct virtnet_info *vi, |
1190 | struct virtqueue *vq, | |
1191 | struct napi_struct *napi) | |
1192 | { | |
1193 | if (!napi->weight) | |
1194 | return; | |
1195 | ||
1196 | /* Tx napi touches cachelines on the cpu handling tx interrupts. Only | |
1197 | * enable the feature if this is likely affine with the transmit path. | |
1198 | */ | |
1199 | if (!vi->affinity_hint_set) { | |
1200 | napi->weight = 0; | |
1201 | return; | |
1202 | } | |
1203 | ||
1204 | return virtnet_napi_enable(vq, napi); | |
1205 | } | |
1206 | ||
78a57b48 WB |
1207 | static void virtnet_napi_tx_disable(struct napi_struct *napi) |
1208 | { | |
1209 | if (napi->weight) | |
1210 | napi_disable(napi); | |
1211 | } | |
1212 | ||
3161e453 RR |
1213 | static void refill_work(struct work_struct *work) |
1214 | { | |
e9d7417b JW |
1215 | struct virtnet_info *vi = |
1216 | container_of(work, struct virtnet_info, refill.work); | |
3161e453 | 1217 | bool still_empty; |
986a4f4d JW |
1218 | int i; |
1219 | ||
55257d72 | 1220 | for (i = 0; i < vi->curr_queue_pairs; i++) { |
986a4f4d | 1221 | struct receive_queue *rq = &vi->rq[i]; |
3161e453 | 1222 | |
986a4f4d | 1223 | napi_disable(&rq->napi); |
946fa564 | 1224 | still_empty = !try_fill_recv(vi, rq, GFP_KERNEL); |
e4e8452a | 1225 | virtnet_napi_enable(rq->vq, &rq->napi); |
3161e453 | 1226 | |
986a4f4d JW |
1227 | /* In theory, this can happen: if we don't get any buffers in |
1228 | * we will *never* try to fill again. | |
1229 | */ | |
1230 | if (still_empty) | |
1231 | schedule_delayed_work(&vi->refill, HZ/2); | |
1232 | } | |
3161e453 RR |
1233 | } |
1234 | ||
186b3c99 | 1235 | static int virtnet_receive(struct receive_queue *rq, int budget, bool *xdp_xmit) |
296f96fc | 1236 | { |
e9d7417b | 1237 | struct virtnet_info *vi = rq->vq->vdev->priv; |
61845d20 | 1238 | unsigned int len, received = 0, bytes = 0; |
9ab86bbc | 1239 | void *buf; |
296f96fc | 1240 | |
192f68cf | 1241 | if (!vi->big_packets || vi->mergeable_rx_bufs) { |
680557cf MT |
1242 | void *ctx; |
1243 | ||
1244 | while (received < budget && | |
1245 | (buf = virtqueue_get_buf_ctx(rq->vq, &len, &ctx))) { | |
186b3c99 | 1246 | bytes += receive_buf(vi, rq, buf, len, ctx, xdp_xmit); |
680557cf MT |
1247 | received++; |
1248 | } | |
1249 | } else { | |
1250 | while (received < budget && | |
1251 | (buf = virtqueue_get_buf(rq->vq, &len)) != NULL) { | |
186b3c99 | 1252 | bytes += receive_buf(vi, rq, buf, len, NULL, xdp_xmit); |
680557cf MT |
1253 | received++; |
1254 | } | |
296f96fc RR |
1255 | } |
1256 | ||
be121f46 | 1257 | if (rq->vq->num_free > virtqueue_get_vring_size(rq->vq) / 2) { |
946fa564 | 1258 | if (!try_fill_recv(vi, rq, GFP_ATOMIC)) |
3b07e9ca | 1259 | schedule_delayed_work(&vi->refill, 0); |
3161e453 | 1260 | } |
296f96fc | 1261 | |
d7dfc5cf TM |
1262 | u64_stats_update_begin(&rq->stats.syncp); |
1263 | rq->stats.bytes += bytes; | |
1264 | rq->stats.packets += received; | |
1265 | u64_stats_update_end(&rq->stats.syncp); | |
61845d20 | 1266 | |
2ffa7598 JW |
1267 | return received; |
1268 | } | |
1269 | ||
ea7735d9 WB |
1270 | static void free_old_xmit_skbs(struct send_queue *sq) |
1271 | { | |
1272 | struct sk_buff *skb; | |
1273 | unsigned int len; | |
ea7735d9 WB |
1274 | unsigned int packets = 0; |
1275 | unsigned int bytes = 0; | |
1276 | ||
1277 | while ((skb = virtqueue_get_buf(sq->vq, &len)) != NULL) { | |
1278 | pr_debug("Sent skb %p\n", skb); | |
1279 | ||
1280 | bytes += skb->len; | |
1281 | packets++; | |
1282 | ||
dadc0736 | 1283 | dev_consume_skb_any(skb); |
ea7735d9 WB |
1284 | } |
1285 | ||
1286 | /* Avoid overhead when no packets have been processed | |
1287 | * happens when called speculatively from start_xmit. | |
1288 | */ | |
1289 | if (!packets) | |
1290 | return; | |
1291 | ||
d7dfc5cf TM |
1292 | u64_stats_update_begin(&sq->stats.syncp); |
1293 | sq->stats.bytes += bytes; | |
1294 | sq->stats.packets += packets; | |
1295 | u64_stats_update_end(&sq->stats.syncp); | |
ea7735d9 WB |
1296 | } |
1297 | ||
7b0411ef WB |
1298 | static void virtnet_poll_cleantx(struct receive_queue *rq) |
1299 | { | |
1300 | struct virtnet_info *vi = rq->vq->vdev->priv; | |
1301 | unsigned int index = vq2rxq(rq->vq); | |
1302 | struct send_queue *sq = &vi->sq[index]; | |
1303 | struct netdev_queue *txq = netdev_get_tx_queue(vi->dev, index); | |
1304 | ||
1305 | if (!sq->napi.weight) | |
1306 | return; | |
1307 | ||
1308 | if (__netif_tx_trylock(txq)) { | |
1309 | free_old_xmit_skbs(sq); | |
1310 | __netif_tx_unlock(txq); | |
1311 | } | |
1312 | ||
1313 | if (sq->vq->num_free >= 2 + MAX_SKB_FRAGS) | |
1314 | netif_tx_wake_queue(txq); | |
1315 | } | |
1316 | ||
2ffa7598 JW |
1317 | static int virtnet_poll(struct napi_struct *napi, int budget) |
1318 | { | |
1319 | struct receive_queue *rq = | |
1320 | container_of(napi, struct receive_queue, napi); | |
9267c430 JW |
1321 | struct virtnet_info *vi = rq->vq->vdev->priv; |
1322 | struct send_queue *sq; | |
1323 | unsigned int received, qp; | |
186b3c99 | 1324 | bool xdp_xmit = false; |
2ffa7598 | 1325 | |
7b0411ef WB |
1326 | virtnet_poll_cleantx(rq); |
1327 | ||
186b3c99 | 1328 | received = virtnet_receive(rq, budget, &xdp_xmit); |
2ffa7598 | 1329 | |
8329d98e | 1330 | /* Out of packets? */ |
e4e8452a WB |
1331 | if (received < budget) |
1332 | virtqueue_napi_complete(napi, rq->vq, received); | |
296f96fc | 1333 | |
9267c430 JW |
1334 | if (xdp_xmit) { |
1335 | qp = vi->curr_queue_pairs - vi->xdp_queue_pairs + | |
1336 | smp_processor_id(); | |
1337 | sq = &vi->sq[qp]; | |
1338 | virtqueue_kick(sq->vq); | |
186b3c99 | 1339 | xdp_do_flush_map(); |
9267c430 | 1340 | } |
186b3c99 | 1341 | |
296f96fc RR |
1342 | return received; |
1343 | } | |
1344 | ||
986a4f4d JW |
1345 | static int virtnet_open(struct net_device *dev) |
1346 | { | |
1347 | struct virtnet_info *vi = netdev_priv(dev); | |
754b8a21 | 1348 | int i, err; |
986a4f4d | 1349 | |
e4166625 JW |
1350 | for (i = 0; i < vi->max_queue_pairs; i++) { |
1351 | if (i < vi->curr_queue_pairs) | |
1352 | /* Make sure we have some buffers: if oom use wq. */ | |
946fa564 | 1353 | if (!try_fill_recv(vi, &vi->rq[i], GFP_KERNEL)) |
e4166625 | 1354 | schedule_delayed_work(&vi->refill, 0); |
754b8a21 JDB |
1355 | |
1356 | err = xdp_rxq_info_reg(&vi->rq[i].xdp_rxq, dev, i); | |
1357 | if (err < 0) | |
1358 | return err; | |
1359 | ||
8d5d8852 JDB |
1360 | err = xdp_rxq_info_reg_mem_model(&vi->rq[i].xdp_rxq, |
1361 | MEM_TYPE_PAGE_SHARED, NULL); | |
1362 | if (err < 0) { | |
1363 | xdp_rxq_info_unreg(&vi->rq[i].xdp_rxq); | |
1364 | return err; | |
1365 | } | |
1366 | ||
e4e8452a | 1367 | virtnet_napi_enable(vi->rq[i].vq, &vi->rq[i].napi); |
b92f1e67 | 1368 | virtnet_napi_tx_enable(vi, vi->sq[i].vq, &vi->sq[i].napi); |
986a4f4d JW |
1369 | } |
1370 | ||
1371 | return 0; | |
1372 | } | |
1373 | ||
b92f1e67 WB |
1374 | static int virtnet_poll_tx(struct napi_struct *napi, int budget) |
1375 | { | |
1376 | struct send_queue *sq = container_of(napi, struct send_queue, napi); | |
1377 | struct virtnet_info *vi = sq->vq->vdev->priv; | |
1378 | struct netdev_queue *txq = netdev_get_tx_queue(vi->dev, vq2txq(sq->vq)); | |
1379 | ||
1380 | __netif_tx_lock(txq, raw_smp_processor_id()); | |
1381 | free_old_xmit_skbs(sq); | |
1382 | __netif_tx_unlock(txq); | |
1383 | ||
1384 | virtqueue_napi_complete(napi, sq->vq, 0); | |
1385 | ||
1386 | if (sq->vq->num_free >= 2 + MAX_SKB_FRAGS) | |
1387 | netif_tx_wake_queue(txq); | |
1388 | ||
1389 | return 0; | |
1390 | } | |
1391 | ||
e9d7417b | 1392 | static int xmit_skb(struct send_queue *sq, struct sk_buff *skb) |
296f96fc | 1393 | { |
012873d0 | 1394 | struct virtio_net_hdr_mrg_rxbuf *hdr; |
296f96fc | 1395 | const unsigned char *dest = ((struct ethhdr *)skb->data)->h_dest; |
e9d7417b | 1396 | struct virtnet_info *vi = sq->vq->vdev->priv; |
e2fcad58 | 1397 | int num_sg; |
012873d0 | 1398 | unsigned hdr_len = vi->hdr_len; |
e7428e95 | 1399 | bool can_push; |
296f96fc | 1400 | |
e174961c | 1401 | pr_debug("%s: xmit %p %pM\n", vi->dev->name, skb, dest); |
e7428e95 MT |
1402 | |
1403 | can_push = vi->any_header_sg && | |
1404 | !((unsigned long)skb->data & (__alignof__(*hdr) - 1)) && | |
1405 | !skb_header_cloned(skb) && skb_headroom(skb) >= hdr_len; | |
1406 | /* Even if we can, don't push here yet as this would skew | |
1407 | * csum_start offset below. */ | |
1408 | if (can_push) | |
012873d0 | 1409 | hdr = (struct virtio_net_hdr_mrg_rxbuf *)(skb->data - hdr_len); |
e7428e95 MT |
1410 | else |
1411 | hdr = skb_vnet_hdr(skb); | |
296f96fc | 1412 | |
e858fae2 | 1413 | if (virtio_net_hdr_from_skb(skb, &hdr->hdr, |
fd3a8862 WB |
1414 | virtio_is_little_endian(vi->vdev), false, |
1415 | 0)) | |
e858fae2 | 1416 | BUG(); |
296f96fc | 1417 | |
3f2c31d9 | 1418 | if (vi->mergeable_rx_bufs) |
012873d0 | 1419 | hdr->num_buffers = 0; |
3f2c31d9 | 1420 | |
547c890c | 1421 | sg_init_table(sq->sg, skb_shinfo(skb)->nr_frags + (can_push ? 1 : 2)); |
e7428e95 MT |
1422 | if (can_push) { |
1423 | __skb_push(skb, hdr_len); | |
1424 | num_sg = skb_to_sgvec(skb, sq->sg, 0, skb->len); | |
e2fcad58 JD |
1425 | if (unlikely(num_sg < 0)) |
1426 | return num_sg; | |
e7428e95 MT |
1427 | /* Pull header back to avoid skew in tx bytes calculations. */ |
1428 | __skb_pull(skb, hdr_len); | |
1429 | } else { | |
1430 | sg_set_buf(sq->sg, hdr, hdr_len); | |
e2fcad58 JD |
1431 | num_sg = skb_to_sgvec(skb, sq->sg + 1, 0, skb->len); |
1432 | if (unlikely(num_sg < 0)) | |
1433 | return num_sg; | |
1434 | num_sg++; | |
e7428e95 | 1435 | } |
9dc7b9e4 | 1436 | return virtqueue_add_outbuf(sq->vq, sq->sg, num_sg, skb, GFP_ATOMIC); |
11a3a154 RR |
1437 | } |
1438 | ||
424efe9c | 1439 | static netdev_tx_t start_xmit(struct sk_buff *skb, struct net_device *dev) |
99ffc696 RR |
1440 | { |
1441 | struct virtnet_info *vi = netdev_priv(dev); | |
986a4f4d JW |
1442 | int qnum = skb_get_queue_mapping(skb); |
1443 | struct send_queue *sq = &vi->sq[qnum]; | |
9ed4cb07 | 1444 | int err; |
4b7fd2e6 MT |
1445 | struct netdev_queue *txq = netdev_get_tx_queue(dev, qnum); |
1446 | bool kick = !skb->xmit_more; | |
b92f1e67 | 1447 | bool use_napi = sq->napi.weight; |
2cb9c6ba | 1448 | |
2cb9c6ba | 1449 | /* Free up any pending old buffers before queueing new ones. */ |
e9d7417b | 1450 | free_old_xmit_skbs(sq); |
99ffc696 | 1451 | |
bdb12e0d WB |
1452 | if (use_napi && kick) |
1453 | virtqueue_enable_cb_delayed(sq->vq); | |
1454 | ||
074c3582 JK |
1455 | /* timestamp packet in software */ |
1456 | skb_tx_timestamp(skb); | |
1457 | ||
03f191ba | 1458 | /* Try to transmit */ |
b7dfde95 | 1459 | err = xmit_skb(sq, skb); |
48925e37 | 1460 | |
9ed4cb07 | 1461 | /* This should not happen! */ |
681daee2 | 1462 | if (unlikely(err)) { |
9ed4cb07 RR |
1463 | dev->stats.tx_fifo_errors++; |
1464 | if (net_ratelimit()) | |
1465 | dev_warn(&dev->dev, | |
b7dfde95 | 1466 | "Unexpected TXQ (%d) queue failure: %d\n", qnum, err); |
58eba97d | 1467 | dev->stats.tx_dropped++; |
85e94525 | 1468 | dev_kfree_skb_any(skb); |
58eba97d | 1469 | return NETDEV_TX_OK; |
296f96fc | 1470 | } |
03f191ba | 1471 | |
48925e37 | 1472 | /* Don't wait up for transmitted skbs to be freed. */ |
b92f1e67 WB |
1473 | if (!use_napi) { |
1474 | skb_orphan(skb); | |
1475 | nf_reset(skb); | |
1476 | } | |
48925e37 | 1477 | |
60302ff6 MT |
1478 | /* If running out of space, stop queue to avoid getting packets that we |
1479 | * are then unable to transmit. | |
1480 | * An alternative would be to force queuing layer to requeue the skb by | |
1481 | * returning NETDEV_TX_BUSY. However, NETDEV_TX_BUSY should not be | |
1482 | * returned in a normal path of operation: it means that driver is not | |
1483 | * maintaining the TX queue stop/start state properly, and causes | |
1484 | * the stack to do a non-trivial amount of useless work. | |
1485 | * Since most packets only take 1 or 2 ring slots, stopping the queue | |
1486 | * early means 16 slots are typically wasted. | |
d631b94e | 1487 | */ |
b7dfde95 | 1488 | if (sq->vq->num_free < 2+MAX_SKB_FRAGS) { |
986a4f4d | 1489 | netif_stop_subqueue(dev, qnum); |
b92f1e67 WB |
1490 | if (!use_napi && |
1491 | unlikely(!virtqueue_enable_cb_delayed(sq->vq))) { | |
48925e37 | 1492 | /* More just got used, free them then recheck. */ |
b7dfde95 LT |
1493 | free_old_xmit_skbs(sq); |
1494 | if (sq->vq->num_free >= 2+MAX_SKB_FRAGS) { | |
986a4f4d | 1495 | netif_start_subqueue(dev, qnum); |
e9d7417b | 1496 | virtqueue_disable_cb(sq->vq); |
48925e37 RR |
1497 | } |
1498 | } | |
99ffc696 | 1499 | } |
48925e37 | 1500 | |
4b7fd2e6 | 1501 | if (kick || netif_xmit_stopped(txq)) |
0b725a2c | 1502 | virtqueue_kick(sq->vq); |
296f96fc | 1503 | |
0b725a2c | 1504 | return NETDEV_TX_OK; |
c223a078 DM |
1505 | } |
1506 | ||
40cbfc37 AK |
1507 | /* |
1508 | * Send command via the control virtqueue and check status. Commands | |
1509 | * supported by the hypervisor, as indicated by feature bits, should | |
788a8b6d | 1510 | * never fail unless improperly formatted. |
40cbfc37 AK |
1511 | */ |
1512 | static bool virtnet_send_command(struct virtnet_info *vi, u8 class, u8 cmd, | |
d24bae32 | 1513 | struct scatterlist *out) |
40cbfc37 | 1514 | { |
f7bc9594 | 1515 | struct scatterlist *sgs[4], hdr, stat; |
d24bae32 | 1516 | unsigned out_num = 0, tmp; |
40cbfc37 AK |
1517 | |
1518 | /* Caller should know better */ | |
f7bc9594 | 1519 | BUG_ON(!virtio_has_feature(vi->vdev, VIRTIO_NET_F_CTRL_VQ)); |
40cbfc37 | 1520 | |
12e57169 MT |
1521 | vi->ctrl->status = ~0; |
1522 | vi->ctrl->hdr.class = class; | |
1523 | vi->ctrl->hdr.cmd = cmd; | |
f7bc9594 | 1524 | /* Add header */ |
12e57169 | 1525 | sg_init_one(&hdr, &vi->ctrl->hdr, sizeof(vi->ctrl->hdr)); |
f7bc9594 | 1526 | sgs[out_num++] = &hdr; |
40cbfc37 | 1527 | |
f7bc9594 RR |
1528 | if (out) |
1529 | sgs[out_num++] = out; | |
40cbfc37 | 1530 | |
f7bc9594 | 1531 | /* Add return status. */ |
12e57169 | 1532 | sg_init_one(&stat, &vi->ctrl->status, sizeof(vi->ctrl->status)); |
d24bae32 | 1533 | sgs[out_num] = &stat; |
40cbfc37 | 1534 | |
d24bae32 | 1535 | BUG_ON(out_num + 1 > ARRAY_SIZE(sgs)); |
a7c58146 | 1536 | virtqueue_add_sgs(vi->cvq, sgs, out_num, 1, vi, GFP_ATOMIC); |
40cbfc37 | 1537 | |
67975901 | 1538 | if (unlikely(!virtqueue_kick(vi->cvq))) |
12e57169 | 1539 | return vi->ctrl->status == VIRTIO_NET_OK; |
40cbfc37 AK |
1540 | |
1541 | /* Spin for a response, the kick causes an ioport write, trapping | |
1542 | * into the hypervisor, so the request should be handled immediately. | |
1543 | */ | |
047b9b94 HG |
1544 | while (!virtqueue_get_buf(vi->cvq, &tmp) && |
1545 | !virtqueue_is_broken(vi->cvq)) | |
40cbfc37 AK |
1546 | cpu_relax(); |
1547 | ||
12e57169 | 1548 | return vi->ctrl->status == VIRTIO_NET_OK; |
40cbfc37 AK |
1549 | } |
1550 | ||
9c46f6d4 AW |
1551 | static int virtnet_set_mac_address(struct net_device *dev, void *p) |
1552 | { | |
1553 | struct virtnet_info *vi = netdev_priv(dev); | |
1554 | struct virtio_device *vdev = vi->vdev; | |
f2f2c8b4 | 1555 | int ret; |
e37e2ff3 | 1556 | struct sockaddr *addr; |
7e58d5ae | 1557 | struct scatterlist sg; |
9c46f6d4 | 1558 | |
ba5e4426 SS |
1559 | if (virtio_has_feature(vi->vdev, VIRTIO_NET_F_STANDBY)) |
1560 | return -EOPNOTSUPP; | |
1561 | ||
801822d1 | 1562 | addr = kmemdup(p, sizeof(*addr), GFP_KERNEL); |
e37e2ff3 AL |
1563 | if (!addr) |
1564 | return -ENOMEM; | |
e37e2ff3 AL |
1565 | |
1566 | ret = eth_prepare_mac_addr_change(dev, addr); | |
f2f2c8b4 | 1567 | if (ret) |
e37e2ff3 | 1568 | goto out; |
9c46f6d4 | 1569 | |
7e58d5ae AK |
1570 | if (virtio_has_feature(vdev, VIRTIO_NET_F_CTRL_MAC_ADDR)) { |
1571 | sg_init_one(&sg, addr->sa_data, dev->addr_len); | |
1572 | if (!virtnet_send_command(vi, VIRTIO_NET_CTRL_MAC, | |
d24bae32 | 1573 | VIRTIO_NET_CTRL_MAC_ADDR_SET, &sg)) { |
7e58d5ae AK |
1574 | dev_warn(&vdev->dev, |
1575 | "Failed to set mac address by vq command.\n"); | |
e37e2ff3 AL |
1576 | ret = -EINVAL; |
1577 | goto out; | |
7e58d5ae | 1578 | } |
7e93a02f MT |
1579 | } else if (virtio_has_feature(vdev, VIRTIO_NET_F_MAC) && |
1580 | !virtio_has_feature(vdev, VIRTIO_F_VERSION_1)) { | |
855e0c52 RR |
1581 | unsigned int i; |
1582 | ||
1583 | /* Naturally, this has an atomicity problem. */ | |
1584 | for (i = 0; i < dev->addr_len; i++) | |
1585 | virtio_cwrite8(vdev, | |
1586 | offsetof(struct virtio_net_config, mac) + | |
1587 | i, addr->sa_data[i]); | |
7e58d5ae AK |
1588 | } |
1589 | ||
1590 | eth_commit_mac_addr_change(dev, p); | |
e37e2ff3 | 1591 | ret = 0; |
9c46f6d4 | 1592 | |
e37e2ff3 AL |
1593 | out: |
1594 | kfree(addr); | |
1595 | return ret; | |
9c46f6d4 AW |
1596 | } |
1597 | ||
bc1f4470 | 1598 | static void virtnet_stats(struct net_device *dev, |
1599 | struct rtnl_link_stats64 *tot) | |
3fa2a1df | 1600 | { |
1601 | struct virtnet_info *vi = netdev_priv(dev); | |
3fa2a1df | 1602 | unsigned int start; |
d7dfc5cf | 1603 | int i; |
3fa2a1df | 1604 | |
d7dfc5cf | 1605 | for (i = 0; i < vi->max_queue_pairs; i++) { |
3fa2a1df | 1606 | u64 tpackets, tbytes, rpackets, rbytes; |
d7dfc5cf TM |
1607 | struct receive_queue *rq = &vi->rq[i]; |
1608 | struct send_queue *sq = &vi->sq[i]; | |
3fa2a1df | 1609 | |
1610 | do { | |
d7dfc5cf TM |
1611 | start = u64_stats_fetch_begin_irq(&sq->stats.syncp); |
1612 | tpackets = sq->stats.packets; | |
1613 | tbytes = sq->stats.bytes; | |
1614 | } while (u64_stats_fetch_retry_irq(&sq->stats.syncp, start)); | |
83a27052 ED |
1615 | |
1616 | do { | |
d7dfc5cf TM |
1617 | start = u64_stats_fetch_begin_irq(&rq->stats.syncp); |
1618 | rpackets = rq->stats.packets; | |
1619 | rbytes = rq->stats.bytes; | |
1620 | } while (u64_stats_fetch_retry_irq(&rq->stats.syncp, start)); | |
3fa2a1df | 1621 | |
1622 | tot->rx_packets += rpackets; | |
1623 | tot->tx_packets += tpackets; | |
1624 | tot->rx_bytes += rbytes; | |
1625 | tot->tx_bytes += tbytes; | |
1626 | } | |
1627 | ||
1628 | tot->tx_dropped = dev->stats.tx_dropped; | |
021ac8d3 | 1629 | tot->tx_fifo_errors = dev->stats.tx_fifo_errors; |
3fa2a1df | 1630 | tot->rx_dropped = dev->stats.rx_dropped; |
1631 | tot->rx_length_errors = dev->stats.rx_length_errors; | |
1632 | tot->rx_frame_errors = dev->stats.rx_frame_errors; | |
3fa2a1df | 1633 | } |
1634 | ||
da74e89d AS |
1635 | #ifdef CONFIG_NET_POLL_CONTROLLER |
1636 | static void virtnet_netpoll(struct net_device *dev) | |
1637 | { | |
1638 | struct virtnet_info *vi = netdev_priv(dev); | |
986a4f4d | 1639 | int i; |
da74e89d | 1640 | |
986a4f4d JW |
1641 | for (i = 0; i < vi->curr_queue_pairs; i++) |
1642 | napi_schedule(&vi->rq[i].napi); | |
da74e89d AS |
1643 | } |
1644 | #endif | |
1645 | ||
586d17c5 JW |
1646 | static void virtnet_ack_link_announce(struct virtnet_info *vi) |
1647 | { | |
1648 | rtnl_lock(); | |
1649 | if (!virtnet_send_command(vi, VIRTIO_NET_CTRL_ANNOUNCE, | |
d24bae32 | 1650 | VIRTIO_NET_CTRL_ANNOUNCE_ACK, NULL)) |
586d17c5 JW |
1651 | dev_warn(&vi->dev->dev, "Failed to ack link announce.\n"); |
1652 | rtnl_unlock(); | |
1653 | } | |
1654 | ||
47315329 | 1655 | static int _virtnet_set_queues(struct virtnet_info *vi, u16 queue_pairs) |
986a4f4d JW |
1656 | { |
1657 | struct scatterlist sg; | |
986a4f4d JW |
1658 | struct net_device *dev = vi->dev; |
1659 | ||
1660 | if (!vi->has_cvq || !virtio_has_feature(vi->vdev, VIRTIO_NET_F_MQ)) | |
1661 | return 0; | |
1662 | ||
12e57169 MT |
1663 | vi->ctrl->mq.virtqueue_pairs = cpu_to_virtio16(vi->vdev, queue_pairs); |
1664 | sg_init_one(&sg, &vi->ctrl->mq, sizeof(vi->ctrl->mq)); | |
986a4f4d JW |
1665 | |
1666 | if (!virtnet_send_command(vi, VIRTIO_NET_CTRL_MQ, | |
d24bae32 | 1667 | VIRTIO_NET_CTRL_MQ_VQ_PAIRS_SET, &sg)) { |
986a4f4d JW |
1668 | dev_warn(&dev->dev, "Fail to set num of queue pairs to %d\n", |
1669 | queue_pairs); | |
1670 | return -EINVAL; | |
55257d72 | 1671 | } else { |
986a4f4d | 1672 | vi->curr_queue_pairs = queue_pairs; |
35ed159b JW |
1673 | /* virtnet_open() will refill when device is going to up. */ |
1674 | if (dev->flags & IFF_UP) | |
1675 | schedule_delayed_work(&vi->refill, 0); | |
55257d72 | 1676 | } |
986a4f4d JW |
1677 | |
1678 | return 0; | |
1679 | } | |
1680 | ||
47315329 JF |
1681 | static int virtnet_set_queues(struct virtnet_info *vi, u16 queue_pairs) |
1682 | { | |
1683 | int err; | |
1684 | ||
1685 | rtnl_lock(); | |
1686 | err = _virtnet_set_queues(vi, queue_pairs); | |
1687 | rtnl_unlock(); | |
1688 | return err; | |
1689 | } | |
1690 | ||
296f96fc RR |
1691 | static int virtnet_close(struct net_device *dev) |
1692 | { | |
1693 | struct virtnet_info *vi = netdev_priv(dev); | |
986a4f4d | 1694 | int i; |
296f96fc | 1695 | |
b2baed69 RR |
1696 | /* Make sure refill_work doesn't re-enable napi! */ |
1697 | cancel_delayed_work_sync(&vi->refill); | |
986a4f4d | 1698 | |
b92f1e67 | 1699 | for (i = 0; i < vi->max_queue_pairs; i++) { |
754b8a21 | 1700 | xdp_rxq_info_unreg(&vi->rq[i].xdp_rxq); |
986a4f4d | 1701 | napi_disable(&vi->rq[i].napi); |
78a57b48 | 1702 | virtnet_napi_tx_disable(&vi->sq[i].napi); |
b92f1e67 | 1703 | } |
296f96fc | 1704 | |
296f96fc RR |
1705 | return 0; |
1706 | } | |
1707 | ||
2af7698e AW |
1708 | static void virtnet_set_rx_mode(struct net_device *dev) |
1709 | { | |
1710 | struct virtnet_info *vi = netdev_priv(dev); | |
f565a7c2 | 1711 | struct scatterlist sg[2]; |
f565a7c2 | 1712 | struct virtio_net_ctrl_mac *mac_data; |
ccffad25 | 1713 | struct netdev_hw_addr *ha; |
32e7bfc4 | 1714 | int uc_count; |
4cd24eaf | 1715 | int mc_count; |
f565a7c2 AW |
1716 | void *buf; |
1717 | int i; | |
2af7698e | 1718 | |
788a8b6d | 1719 | /* We can't dynamically set ndo_set_rx_mode, so return gracefully */ |
2af7698e AW |
1720 | if (!virtio_has_feature(vi->vdev, VIRTIO_NET_F_CTRL_RX)) |
1721 | return; | |
1722 | ||
12e57169 MT |
1723 | vi->ctrl->promisc = ((dev->flags & IFF_PROMISC) != 0); |
1724 | vi->ctrl->allmulti = ((dev->flags & IFF_ALLMULTI) != 0); | |
2af7698e | 1725 | |
12e57169 | 1726 | sg_init_one(sg, &vi->ctrl->promisc, sizeof(vi->ctrl->promisc)); |
2af7698e AW |
1727 | |
1728 | if (!virtnet_send_command(vi, VIRTIO_NET_CTRL_RX, | |
d24bae32 | 1729 | VIRTIO_NET_CTRL_RX_PROMISC, sg)) |
2af7698e | 1730 | dev_warn(&dev->dev, "Failed to %sable promisc mode.\n", |
12e57169 | 1731 | vi->ctrl->promisc ? "en" : "dis"); |
2af7698e | 1732 | |
12e57169 | 1733 | sg_init_one(sg, &vi->ctrl->allmulti, sizeof(vi->ctrl->allmulti)); |
2af7698e AW |
1734 | |
1735 | if (!virtnet_send_command(vi, VIRTIO_NET_CTRL_RX, | |
d24bae32 | 1736 | VIRTIO_NET_CTRL_RX_ALLMULTI, sg)) |
2af7698e | 1737 | dev_warn(&dev->dev, "Failed to %sable allmulti mode.\n", |
12e57169 | 1738 | vi->ctrl->allmulti ? "en" : "dis"); |
f565a7c2 | 1739 | |
32e7bfc4 | 1740 | uc_count = netdev_uc_count(dev); |
4cd24eaf | 1741 | mc_count = netdev_mc_count(dev); |
f565a7c2 | 1742 | /* MAC filter - use one buffer for both lists */ |
4cd24eaf JP |
1743 | buf = kzalloc(((uc_count + mc_count) * ETH_ALEN) + |
1744 | (2 * sizeof(mac_data->entries)), GFP_ATOMIC); | |
1745 | mac_data = buf; | |
e68ed8f0 | 1746 | if (!buf) |
f565a7c2 | 1747 | return; |
f565a7c2 | 1748 | |
23e258e1 AW |
1749 | sg_init_table(sg, 2); |
1750 | ||
f565a7c2 | 1751 | /* Store the unicast list and count in the front of the buffer */ |
fdd819b2 | 1752 | mac_data->entries = cpu_to_virtio32(vi->vdev, uc_count); |
ccffad25 | 1753 | i = 0; |
32e7bfc4 | 1754 | netdev_for_each_uc_addr(ha, dev) |
ccffad25 | 1755 | memcpy(&mac_data->macs[i++][0], ha->addr, ETH_ALEN); |
f565a7c2 AW |
1756 | |
1757 | sg_set_buf(&sg[0], mac_data, | |
32e7bfc4 | 1758 | sizeof(mac_data->entries) + (uc_count * ETH_ALEN)); |
f565a7c2 AW |
1759 | |
1760 | /* multicast list and count fill the end */ | |
32e7bfc4 | 1761 | mac_data = (void *)&mac_data->macs[uc_count][0]; |
f565a7c2 | 1762 | |
fdd819b2 | 1763 | mac_data->entries = cpu_to_virtio32(vi->vdev, mc_count); |
567ec874 | 1764 | i = 0; |
22bedad3 JP |
1765 | netdev_for_each_mc_addr(ha, dev) |
1766 | memcpy(&mac_data->macs[i++][0], ha->addr, ETH_ALEN); | |
f565a7c2 AW |
1767 | |
1768 | sg_set_buf(&sg[1], mac_data, | |
4cd24eaf | 1769 | sizeof(mac_data->entries) + (mc_count * ETH_ALEN)); |
f565a7c2 AW |
1770 | |
1771 | if (!virtnet_send_command(vi, VIRTIO_NET_CTRL_MAC, | |
d24bae32 | 1772 | VIRTIO_NET_CTRL_MAC_TABLE_SET, sg)) |
99e872ae | 1773 | dev_warn(&dev->dev, "Failed to set MAC filter table.\n"); |
f565a7c2 AW |
1774 | |
1775 | kfree(buf); | |
2af7698e AW |
1776 | } |
1777 | ||
80d5c368 PM |
1778 | static int virtnet_vlan_rx_add_vid(struct net_device *dev, |
1779 | __be16 proto, u16 vid) | |
0bde9569 AW |
1780 | { |
1781 | struct virtnet_info *vi = netdev_priv(dev); | |
1782 | struct scatterlist sg; | |
1783 | ||
d7fad4c8 | 1784 | vi->ctrl->vid = cpu_to_virtio16(vi->vdev, vid); |
12e57169 | 1785 | sg_init_one(&sg, &vi->ctrl->vid, sizeof(vi->ctrl->vid)); |
0bde9569 AW |
1786 | |
1787 | if (!virtnet_send_command(vi, VIRTIO_NET_CTRL_VLAN, | |
d24bae32 | 1788 | VIRTIO_NET_CTRL_VLAN_ADD, &sg)) |
0bde9569 | 1789 | dev_warn(&dev->dev, "Failed to add VLAN ID %d.\n", vid); |
8e586137 | 1790 | return 0; |
0bde9569 AW |
1791 | } |
1792 | ||
80d5c368 PM |
1793 | static int virtnet_vlan_rx_kill_vid(struct net_device *dev, |
1794 | __be16 proto, u16 vid) | |
0bde9569 AW |
1795 | { |
1796 | struct virtnet_info *vi = netdev_priv(dev); | |
1797 | struct scatterlist sg; | |
1798 | ||
d7fad4c8 | 1799 | vi->ctrl->vid = cpu_to_virtio16(vi->vdev, vid); |
12e57169 | 1800 | sg_init_one(&sg, &vi->ctrl->vid, sizeof(vi->ctrl->vid)); |
0bde9569 AW |
1801 | |
1802 | if (!virtnet_send_command(vi, VIRTIO_NET_CTRL_VLAN, | |
d24bae32 | 1803 | VIRTIO_NET_CTRL_VLAN_DEL, &sg)) |
0bde9569 | 1804 | dev_warn(&dev->dev, "Failed to kill VLAN ID %d.\n", vid); |
8e586137 | 1805 | return 0; |
0bde9569 AW |
1806 | } |
1807 | ||
8898c21c | 1808 | static void virtnet_clean_affinity(struct virtnet_info *vi, long hcpu) |
986a4f4d JW |
1809 | { |
1810 | int i; | |
1811 | ||
8898c21c WG |
1812 | if (vi->affinity_hint_set) { |
1813 | for (i = 0; i < vi->max_queue_pairs; i++) { | |
47be2479 WG |
1814 | virtqueue_set_affinity(vi->rq[i].vq, -1); |
1815 | virtqueue_set_affinity(vi->sq[i].vq, -1); | |
1816 | } | |
1817 | ||
8898c21c WG |
1818 | vi->affinity_hint_set = false; |
1819 | } | |
8898c21c | 1820 | } |
47be2479 | 1821 | |
8898c21c WG |
1822 | static void virtnet_set_affinity(struct virtnet_info *vi) |
1823 | { | |
1824 | int i; | |
1825 | int cpu; | |
986a4f4d JW |
1826 | |
1827 | /* In multiqueue mode, when the number of cpu is equal to the number of | |
1828 | * queue pairs, we let the queue pairs to be private to one cpu by | |
1829 | * setting the affinity hint to eliminate the contention. | |
1830 | */ | |
8898c21c WG |
1831 | if (vi->curr_queue_pairs == 1 || |
1832 | vi->max_queue_pairs != num_online_cpus()) { | |
1833 | virtnet_clean_affinity(vi, -1); | |
1834 | return; | |
986a4f4d JW |
1835 | } |
1836 | ||
8898c21c WG |
1837 | i = 0; |
1838 | for_each_online_cpu(cpu) { | |
986a4f4d JW |
1839 | virtqueue_set_affinity(vi->rq[i].vq, cpu); |
1840 | virtqueue_set_affinity(vi->sq[i].vq, cpu); | |
9bb8ca86 | 1841 | netif_set_xps_queue(vi->dev, cpumask_of(cpu), i); |
8898c21c | 1842 | i++; |
986a4f4d JW |
1843 | } |
1844 | ||
8898c21c | 1845 | vi->affinity_hint_set = true; |
986a4f4d JW |
1846 | } |
1847 | ||
8017c279 | 1848 | static int virtnet_cpu_online(unsigned int cpu, struct hlist_node *node) |
8de4b2f3 | 1849 | { |
8017c279 SAS |
1850 | struct virtnet_info *vi = hlist_entry_safe(node, struct virtnet_info, |
1851 | node); | |
1852 | virtnet_set_affinity(vi); | |
1853 | return 0; | |
1854 | } | |
8de4b2f3 | 1855 | |
8017c279 SAS |
1856 | static int virtnet_cpu_dead(unsigned int cpu, struct hlist_node *node) |
1857 | { | |
1858 | struct virtnet_info *vi = hlist_entry_safe(node, struct virtnet_info, | |
1859 | node_dead); | |
1860 | virtnet_set_affinity(vi); | |
1861 | return 0; | |
1862 | } | |
3ab098df | 1863 | |
8017c279 SAS |
1864 | static int virtnet_cpu_down_prep(unsigned int cpu, struct hlist_node *node) |
1865 | { | |
1866 | struct virtnet_info *vi = hlist_entry_safe(node, struct virtnet_info, | |
1867 | node); | |
1868 | ||
1869 | virtnet_clean_affinity(vi, cpu); | |
1870 | return 0; | |
1871 | } | |
1872 | ||
1873 | static enum cpuhp_state virtionet_online; | |
1874 | ||
1875 | static int virtnet_cpu_notif_add(struct virtnet_info *vi) | |
1876 | { | |
1877 | int ret; | |
1878 | ||
1879 | ret = cpuhp_state_add_instance_nocalls(virtionet_online, &vi->node); | |
1880 | if (ret) | |
1881 | return ret; | |
1882 | ret = cpuhp_state_add_instance_nocalls(CPUHP_VIRT_NET_DEAD, | |
1883 | &vi->node_dead); | |
1884 | if (!ret) | |
1885 | return ret; | |
1886 | cpuhp_state_remove_instance_nocalls(virtionet_online, &vi->node); | |
1887 | return ret; | |
1888 | } | |
1889 | ||
1890 | static void virtnet_cpu_notif_remove(struct virtnet_info *vi) | |
1891 | { | |
1892 | cpuhp_state_remove_instance_nocalls(virtionet_online, &vi->node); | |
1893 | cpuhp_state_remove_instance_nocalls(CPUHP_VIRT_NET_DEAD, | |
1894 | &vi->node_dead); | |
986a4f4d JW |
1895 | } |
1896 | ||
8f9f4668 RJ |
1897 | static void virtnet_get_ringparam(struct net_device *dev, |
1898 | struct ethtool_ringparam *ring) | |
1899 | { | |
1900 | struct virtnet_info *vi = netdev_priv(dev); | |
1901 | ||
986a4f4d JW |
1902 | ring->rx_max_pending = virtqueue_get_vring_size(vi->rq[0].vq); |
1903 | ring->tx_max_pending = virtqueue_get_vring_size(vi->sq[0].vq); | |
8f9f4668 RJ |
1904 | ring->rx_pending = ring->rx_max_pending; |
1905 | ring->tx_pending = ring->tx_max_pending; | |
8f9f4668 RJ |
1906 | } |
1907 | ||
66846048 RJ |
1908 | |
1909 | static void virtnet_get_drvinfo(struct net_device *dev, | |
1910 | struct ethtool_drvinfo *info) | |
1911 | { | |
1912 | struct virtnet_info *vi = netdev_priv(dev); | |
1913 | struct virtio_device *vdev = vi->vdev; | |
1914 | ||
1915 | strlcpy(info->driver, KBUILD_MODNAME, sizeof(info->driver)); | |
1916 | strlcpy(info->version, VIRTNET_DRIVER_VERSION, sizeof(info->version)); | |
1917 | strlcpy(info->bus_info, virtio_bus_name(vdev), sizeof(info->bus_info)); | |
1918 | ||
1919 | } | |
1920 | ||
d73bcd2c JW |
1921 | /* TODO: Eliminate OOO packets during switching */ |
1922 | static int virtnet_set_channels(struct net_device *dev, | |
1923 | struct ethtool_channels *channels) | |
1924 | { | |
1925 | struct virtnet_info *vi = netdev_priv(dev); | |
1926 | u16 queue_pairs = channels->combined_count; | |
1927 | int err; | |
1928 | ||
1929 | /* We don't support separate rx/tx channels. | |
1930 | * We don't allow setting 'other' channels. | |
1931 | */ | |
1932 | if (channels->rx_count || channels->tx_count || channels->other_count) | |
1933 | return -EINVAL; | |
1934 | ||
c18e9cd6 | 1935 | if (queue_pairs > vi->max_queue_pairs || queue_pairs == 0) |
d73bcd2c JW |
1936 | return -EINVAL; |
1937 | ||
f600b690 JF |
1938 | /* For now we don't support modifying channels while XDP is loaded |
1939 | * also when XDP is loaded all RX queues have XDP programs so we only | |
1940 | * need to check a single RX queue. | |
1941 | */ | |
1942 | if (vi->rq[0].xdp_prog) | |
1943 | return -EINVAL; | |
1944 | ||
47be2479 | 1945 | get_online_cpus(); |
47315329 | 1946 | err = _virtnet_set_queues(vi, queue_pairs); |
d73bcd2c JW |
1947 | if (!err) { |
1948 | netif_set_real_num_tx_queues(dev, queue_pairs); | |
1949 | netif_set_real_num_rx_queues(dev, queue_pairs); | |
1950 | ||
8898c21c | 1951 | virtnet_set_affinity(vi); |
d73bcd2c | 1952 | } |
47be2479 | 1953 | put_online_cpus(); |
d73bcd2c JW |
1954 | |
1955 | return err; | |
1956 | } | |
1957 | ||
d7dfc5cf TM |
1958 | static void virtnet_get_strings(struct net_device *dev, u32 stringset, u8 *data) |
1959 | { | |
1960 | struct virtnet_info *vi = netdev_priv(dev); | |
1961 | char *p = (char *)data; | |
1962 | unsigned int i, j; | |
1963 | ||
1964 | switch (stringset) { | |
1965 | case ETH_SS_STATS: | |
1966 | for (i = 0; i < vi->curr_queue_pairs; i++) { | |
1967 | for (j = 0; j < VIRTNET_RQ_STATS_LEN; j++) { | |
1968 | snprintf(p, ETH_GSTRING_LEN, "rx_queue_%u_%s", | |
1969 | i, virtnet_rq_stats_desc[j].desc); | |
1970 | p += ETH_GSTRING_LEN; | |
1971 | } | |
1972 | } | |
1973 | ||
1974 | for (i = 0; i < vi->curr_queue_pairs; i++) { | |
1975 | for (j = 0; j < VIRTNET_SQ_STATS_LEN; j++) { | |
1976 | snprintf(p, ETH_GSTRING_LEN, "tx_queue_%u_%s", | |
1977 | i, virtnet_sq_stats_desc[j].desc); | |
1978 | p += ETH_GSTRING_LEN; | |
1979 | } | |
1980 | } | |
1981 | break; | |
1982 | } | |
1983 | } | |
1984 | ||
1985 | static int virtnet_get_sset_count(struct net_device *dev, int sset) | |
1986 | { | |
1987 | struct virtnet_info *vi = netdev_priv(dev); | |
1988 | ||
1989 | switch (sset) { | |
1990 | case ETH_SS_STATS: | |
1991 | return vi->curr_queue_pairs * (VIRTNET_RQ_STATS_LEN + | |
1992 | VIRTNET_SQ_STATS_LEN); | |
1993 | default: | |
1994 | return -EOPNOTSUPP; | |
1995 | } | |
1996 | } | |
1997 | ||
1998 | static void virtnet_get_ethtool_stats(struct net_device *dev, | |
1999 | struct ethtool_stats *stats, u64 *data) | |
2000 | { | |
2001 | struct virtnet_info *vi = netdev_priv(dev); | |
2002 | unsigned int idx = 0, start, i, j; | |
2003 | const u8 *stats_base; | |
2004 | size_t offset; | |
2005 | ||
2006 | for (i = 0; i < vi->curr_queue_pairs; i++) { | |
2007 | struct receive_queue *rq = &vi->rq[i]; | |
2008 | ||
2009 | stats_base = (u8 *)&rq->stats; | |
2010 | do { | |
2011 | start = u64_stats_fetch_begin_irq(&rq->stats.syncp); | |
2012 | for (j = 0; j < VIRTNET_RQ_STATS_LEN; j++) { | |
2013 | offset = virtnet_rq_stats_desc[j].offset; | |
2014 | data[idx + j] = *(u64 *)(stats_base + offset); | |
2015 | } | |
2016 | } while (u64_stats_fetch_retry_irq(&rq->stats.syncp, start)); | |
2017 | idx += VIRTNET_RQ_STATS_LEN; | |
2018 | } | |
2019 | ||
2020 | for (i = 0; i < vi->curr_queue_pairs; i++) { | |
2021 | struct send_queue *sq = &vi->sq[i]; | |
2022 | ||
2023 | stats_base = (u8 *)&sq->stats; | |
2024 | do { | |
2025 | start = u64_stats_fetch_begin_irq(&sq->stats.syncp); | |
2026 | for (j = 0; j < VIRTNET_SQ_STATS_LEN; j++) { | |
2027 | offset = virtnet_sq_stats_desc[j].offset; | |
2028 | data[idx + j] = *(u64 *)(stats_base + offset); | |
2029 | } | |
2030 | } while (u64_stats_fetch_retry_irq(&sq->stats.syncp, start)); | |
2031 | idx += VIRTNET_SQ_STATS_LEN; | |
2032 | } | |
2033 | } | |
2034 | ||
d73bcd2c JW |
2035 | static void virtnet_get_channels(struct net_device *dev, |
2036 | struct ethtool_channels *channels) | |
2037 | { | |
2038 | struct virtnet_info *vi = netdev_priv(dev); | |
2039 | ||
2040 | channels->combined_count = vi->curr_queue_pairs; | |
2041 | channels->max_combined = vi->max_queue_pairs; | |
2042 | channels->max_other = 0; | |
2043 | channels->rx_count = 0; | |
2044 | channels->tx_count = 0; | |
2045 | channels->other_count = 0; | |
2046 | } | |
2047 | ||
16032be5 | 2048 | /* Check if the user is trying to change anything besides speed/duplex */ |
ebb6b4b1 PR |
2049 | static bool |
2050 | virtnet_validate_ethtool_cmd(const struct ethtool_link_ksettings *cmd) | |
16032be5 | 2051 | { |
ebb6b4b1 PR |
2052 | struct ethtool_link_ksettings diff1 = *cmd; |
2053 | struct ethtool_link_ksettings diff2 = {}; | |
16032be5 | 2054 | |
0cf3ace9 NA |
2055 | /* cmd is always set so we need to clear it, validate the port type |
2056 | * and also without autonegotiation we can ignore advertising | |
2057 | */ | |
ebb6b4b1 PR |
2058 | diff1.base.speed = 0; |
2059 | diff2.base.port = PORT_OTHER; | |
2060 | ethtool_link_ksettings_zero_link_mode(&diff1, advertising); | |
2061 | diff1.base.duplex = 0; | |
2062 | diff1.base.cmd = 0; | |
2063 | diff1.base.link_mode_masks_nwords = 0; | |
2064 | ||
2065 | return !memcmp(&diff1.base, &diff2.base, sizeof(diff1.base)) && | |
2066 | bitmap_empty(diff1.link_modes.supported, | |
2067 | __ETHTOOL_LINK_MODE_MASK_NBITS) && | |
2068 | bitmap_empty(diff1.link_modes.advertising, | |
2069 | __ETHTOOL_LINK_MODE_MASK_NBITS) && | |
2070 | bitmap_empty(diff1.link_modes.lp_advertising, | |
2071 | __ETHTOOL_LINK_MODE_MASK_NBITS); | |
16032be5 NA |
2072 | } |
2073 | ||
ebb6b4b1 PR |
2074 | static int virtnet_set_link_ksettings(struct net_device *dev, |
2075 | const struct ethtool_link_ksettings *cmd) | |
16032be5 NA |
2076 | { |
2077 | struct virtnet_info *vi = netdev_priv(dev); | |
2078 | u32 speed; | |
2079 | ||
ebb6b4b1 | 2080 | speed = cmd->base.speed; |
16032be5 NA |
2081 | /* don't allow custom speed and duplex */ |
2082 | if (!ethtool_validate_speed(speed) || | |
ebb6b4b1 | 2083 | !ethtool_validate_duplex(cmd->base.duplex) || |
16032be5 NA |
2084 | !virtnet_validate_ethtool_cmd(cmd)) |
2085 | return -EINVAL; | |
2086 | vi->speed = speed; | |
ebb6b4b1 | 2087 | vi->duplex = cmd->base.duplex; |
16032be5 NA |
2088 | |
2089 | return 0; | |
2090 | } | |
2091 | ||
ebb6b4b1 PR |
2092 | static int virtnet_get_link_ksettings(struct net_device *dev, |
2093 | struct ethtool_link_ksettings *cmd) | |
16032be5 NA |
2094 | { |
2095 | struct virtnet_info *vi = netdev_priv(dev); | |
2096 | ||
ebb6b4b1 PR |
2097 | cmd->base.speed = vi->speed; |
2098 | cmd->base.duplex = vi->duplex; | |
2099 | cmd->base.port = PORT_OTHER; | |
16032be5 NA |
2100 | |
2101 | return 0; | |
2102 | } | |
2103 | ||
2104 | static void virtnet_init_settings(struct net_device *dev) | |
2105 | { | |
2106 | struct virtnet_info *vi = netdev_priv(dev); | |
2107 | ||
2108 | vi->speed = SPEED_UNKNOWN; | |
2109 | vi->duplex = DUPLEX_UNKNOWN; | |
2110 | } | |
2111 | ||
faa9b39f JB |
2112 | static void virtnet_update_settings(struct virtnet_info *vi) |
2113 | { | |
2114 | u32 speed; | |
2115 | u8 duplex; | |
2116 | ||
2117 | if (!virtio_has_feature(vi->vdev, VIRTIO_NET_F_SPEED_DUPLEX)) | |
2118 | return; | |
2119 | ||
2120 | speed = virtio_cread32(vi->vdev, offsetof(struct virtio_net_config, | |
2121 | speed)); | |
2122 | if (ethtool_validate_speed(speed)) | |
2123 | vi->speed = speed; | |
2124 | duplex = virtio_cread8(vi->vdev, offsetof(struct virtio_net_config, | |
2125 | duplex)); | |
2126 | if (ethtool_validate_duplex(duplex)) | |
2127 | vi->duplex = duplex; | |
2128 | } | |
2129 | ||
0fc0b732 | 2130 | static const struct ethtool_ops virtnet_ethtool_ops = { |
66846048 | 2131 | .get_drvinfo = virtnet_get_drvinfo, |
9f4d26d0 | 2132 | .get_link = ethtool_op_get_link, |
8f9f4668 | 2133 | .get_ringparam = virtnet_get_ringparam, |
d7dfc5cf TM |
2134 | .get_strings = virtnet_get_strings, |
2135 | .get_sset_count = virtnet_get_sset_count, | |
2136 | .get_ethtool_stats = virtnet_get_ethtool_stats, | |
d73bcd2c JW |
2137 | .set_channels = virtnet_set_channels, |
2138 | .get_channels = virtnet_get_channels, | |
074c3582 | 2139 | .get_ts_info = ethtool_op_get_ts_info, |
ebb6b4b1 PR |
2140 | .get_link_ksettings = virtnet_get_link_ksettings, |
2141 | .set_link_ksettings = virtnet_set_link_ksettings, | |
a9ea3fc6 HX |
2142 | }; |
2143 | ||
9fe7bfce JF |
2144 | static void virtnet_freeze_down(struct virtio_device *vdev) |
2145 | { | |
2146 | struct virtnet_info *vi = vdev->priv; | |
2147 | int i; | |
2148 | ||
2149 | /* Make sure no work handler is accessing the device */ | |
2150 | flush_work(&vi->config_work); | |
2151 | ||
2152 | netif_device_detach(vi->dev); | |
713a98d9 | 2153 | netif_tx_disable(vi->dev); |
9fe7bfce JF |
2154 | cancel_delayed_work_sync(&vi->refill); |
2155 | ||
2156 | if (netif_running(vi->dev)) { | |
b92f1e67 | 2157 | for (i = 0; i < vi->max_queue_pairs; i++) { |
9fe7bfce | 2158 | napi_disable(&vi->rq[i].napi); |
78a57b48 | 2159 | virtnet_napi_tx_disable(&vi->sq[i].napi); |
b92f1e67 | 2160 | } |
9fe7bfce JF |
2161 | } |
2162 | } | |
2163 | ||
2164 | static int init_vqs(struct virtnet_info *vi); | |
2165 | ||
2166 | static int virtnet_restore_up(struct virtio_device *vdev) | |
2167 | { | |
2168 | struct virtnet_info *vi = vdev->priv; | |
2169 | int err, i; | |
2170 | ||
2171 | err = init_vqs(vi); | |
2172 | if (err) | |
2173 | return err; | |
2174 | ||
2175 | virtio_device_ready(vdev); | |
2176 | ||
2177 | if (netif_running(vi->dev)) { | |
2178 | for (i = 0; i < vi->curr_queue_pairs; i++) | |
2179 | if (!try_fill_recv(vi, &vi->rq[i], GFP_KERNEL)) | |
2180 | schedule_delayed_work(&vi->refill, 0); | |
2181 | ||
b92f1e67 | 2182 | for (i = 0; i < vi->max_queue_pairs; i++) { |
e4e8452a | 2183 | virtnet_napi_enable(vi->rq[i].vq, &vi->rq[i].napi); |
b92f1e67 WB |
2184 | virtnet_napi_tx_enable(vi, vi->sq[i].vq, |
2185 | &vi->sq[i].napi); | |
2186 | } | |
9fe7bfce JF |
2187 | } |
2188 | ||
2189 | netif_device_attach(vi->dev); | |
2190 | return err; | |
2191 | } | |
2192 | ||
3f93522f JW |
2193 | static int virtnet_set_guest_offloads(struct virtnet_info *vi, u64 offloads) |
2194 | { | |
2195 | struct scatterlist sg; | |
12e57169 | 2196 | vi->ctrl->offloads = cpu_to_virtio64(vi->vdev, offloads); |
3f93522f | 2197 | |
12e57169 | 2198 | sg_init_one(&sg, &vi->ctrl->offloads, sizeof(vi->ctrl->offloads)); |
3f93522f JW |
2199 | |
2200 | if (!virtnet_send_command(vi, VIRTIO_NET_CTRL_GUEST_OFFLOADS, | |
2201 | VIRTIO_NET_CTRL_GUEST_OFFLOADS_SET, &sg)) { | |
2202 | dev_warn(&vi->dev->dev, "Fail to set guest offload. \n"); | |
2203 | return -EINVAL; | |
2204 | } | |
2205 | ||
2206 | return 0; | |
2207 | } | |
2208 | ||
2209 | static int virtnet_clear_guest_offloads(struct virtnet_info *vi) | |
2210 | { | |
2211 | u64 offloads = 0; | |
2212 | ||
2213 | if (!vi->guest_offloads) | |
2214 | return 0; | |
2215 | ||
2216 | if (virtio_has_feature(vi->vdev, VIRTIO_NET_F_GUEST_CSUM)) | |
2217 | offloads = 1ULL << VIRTIO_NET_F_GUEST_CSUM; | |
2218 | ||
2219 | return virtnet_set_guest_offloads(vi, offloads); | |
2220 | } | |
2221 | ||
2222 | static int virtnet_restore_guest_offloads(struct virtnet_info *vi) | |
2223 | { | |
2224 | u64 offloads = vi->guest_offloads; | |
2225 | ||
2226 | if (!vi->guest_offloads) | |
2227 | return 0; | |
2228 | if (virtio_has_feature(vi->vdev, VIRTIO_NET_F_GUEST_CSUM)) | |
2229 | offloads |= 1ULL << VIRTIO_NET_F_GUEST_CSUM; | |
2230 | ||
2231 | return virtnet_set_guest_offloads(vi, offloads); | |
2232 | } | |
2233 | ||
9861ce03 JK |
2234 | static int virtnet_xdp_set(struct net_device *dev, struct bpf_prog *prog, |
2235 | struct netlink_ext_ack *extack) | |
f600b690 JF |
2236 | { |
2237 | unsigned long int max_sz = PAGE_SIZE - sizeof(struct padded_vnet_hdr); | |
2238 | struct virtnet_info *vi = netdev_priv(dev); | |
2239 | struct bpf_prog *old_prog; | |
017b29c3 | 2240 | u16 xdp_qp = 0, curr_qp; |
672aafd5 | 2241 | int i, err; |
f600b690 | 2242 | |
3f93522f JW |
2243 | if (!virtio_has_feature(vi->vdev, VIRTIO_NET_F_CTRL_GUEST_OFFLOADS) |
2244 | && (virtio_has_feature(vi->vdev, VIRTIO_NET_F_GUEST_TSO4) || | |
2245 | virtio_has_feature(vi->vdev, VIRTIO_NET_F_GUEST_TSO6) || | |
2246 | virtio_has_feature(vi->vdev, VIRTIO_NET_F_GUEST_ECN) || | |
2247 | virtio_has_feature(vi->vdev, VIRTIO_NET_F_GUEST_UFO))) { | |
4d463c4d | 2248 | NL_SET_ERR_MSG_MOD(extack, "Can't set XDP while host is implementing LRO, disable LRO first"); |
f600b690 JF |
2249 | return -EOPNOTSUPP; |
2250 | } | |
2251 | ||
2252 | if (vi->mergeable_rx_bufs && !vi->any_header_sg) { | |
4d463c4d | 2253 | NL_SET_ERR_MSG_MOD(extack, "XDP expects header/data in single page, any_header_sg required"); |
f600b690 JF |
2254 | return -EINVAL; |
2255 | } | |
2256 | ||
2257 | if (dev->mtu > max_sz) { | |
4d463c4d | 2258 | NL_SET_ERR_MSG_MOD(extack, "MTU too large to enable XDP"); |
f600b690 JF |
2259 | netdev_warn(dev, "XDP requires MTU less than %lu\n", max_sz); |
2260 | return -EINVAL; | |
2261 | } | |
2262 | ||
672aafd5 JF |
2263 | curr_qp = vi->curr_queue_pairs - vi->xdp_queue_pairs; |
2264 | if (prog) | |
2265 | xdp_qp = nr_cpu_ids; | |
2266 | ||
2267 | /* XDP requires extra queues for XDP_TX */ | |
2268 | if (curr_qp + xdp_qp > vi->max_queue_pairs) { | |
4d463c4d | 2269 | NL_SET_ERR_MSG_MOD(extack, "Too few free TX rings available"); |
672aafd5 JF |
2270 | netdev_warn(dev, "request %i queues but max is %i\n", |
2271 | curr_qp + xdp_qp, vi->max_queue_pairs); | |
2272 | return -ENOMEM; | |
2273 | } | |
2274 | ||
2de2f7f4 JF |
2275 | if (prog) { |
2276 | prog = bpf_prog_add(prog, vi->max_queue_pairs - 1); | |
2277 | if (IS_ERR(prog)) | |
2278 | return PTR_ERR(prog); | |
2279 | } | |
2280 | ||
4941d472 | 2281 | /* Make sure NAPI is not using any XDP TX queues for RX. */ |
4e09ff53 JW |
2282 | if (netif_running(dev)) |
2283 | for (i = 0; i < vi->max_queue_pairs; i++) | |
2284 | napi_disable(&vi->rq[i].napi); | |
f600b690 | 2285 | |
672aafd5 | 2286 | netif_set_real_num_rx_queues(dev, curr_qp + xdp_qp); |
4941d472 JW |
2287 | err = _virtnet_set_queues(vi, curr_qp + xdp_qp); |
2288 | if (err) | |
2289 | goto err; | |
2290 | vi->xdp_queue_pairs = xdp_qp; | |
672aafd5 | 2291 | |
f600b690 JF |
2292 | for (i = 0; i < vi->max_queue_pairs; i++) { |
2293 | old_prog = rtnl_dereference(vi->rq[i].xdp_prog); | |
2294 | rcu_assign_pointer(vi->rq[i].xdp_prog, prog); | |
3f93522f JW |
2295 | if (i == 0) { |
2296 | if (!old_prog) | |
2297 | virtnet_clear_guest_offloads(vi); | |
2298 | if (!prog) | |
2299 | virtnet_restore_guest_offloads(vi); | |
2300 | } | |
f600b690 JF |
2301 | if (old_prog) |
2302 | bpf_prog_put(old_prog); | |
4e09ff53 JW |
2303 | if (netif_running(dev)) |
2304 | virtnet_napi_enable(vi->rq[i].vq, &vi->rq[i].napi); | |
f600b690 JF |
2305 | } |
2306 | ||
2307 | return 0; | |
2de2f7f4 | 2308 | |
4941d472 JW |
2309 | err: |
2310 | for (i = 0; i < vi->max_queue_pairs; i++) | |
2311 | virtnet_napi_enable(vi->rq[i].vq, &vi->rq[i].napi); | |
2de2f7f4 JF |
2312 | if (prog) |
2313 | bpf_prog_sub(prog, vi->max_queue_pairs - 1); | |
2314 | return err; | |
f600b690 JF |
2315 | } |
2316 | ||
5b0e6629 | 2317 | static u32 virtnet_xdp_query(struct net_device *dev) |
f600b690 JF |
2318 | { |
2319 | struct virtnet_info *vi = netdev_priv(dev); | |
5b0e6629 | 2320 | const struct bpf_prog *xdp_prog; |
f600b690 JF |
2321 | int i; |
2322 | ||
2323 | for (i = 0; i < vi->max_queue_pairs; i++) { | |
5b0e6629 MKL |
2324 | xdp_prog = rtnl_dereference(vi->rq[i].xdp_prog); |
2325 | if (xdp_prog) | |
2326 | return xdp_prog->aux->id; | |
f600b690 | 2327 | } |
5b0e6629 | 2328 | return 0; |
f600b690 JF |
2329 | } |
2330 | ||
f4e63525 | 2331 | static int virtnet_xdp(struct net_device *dev, struct netdev_bpf *xdp) |
f600b690 JF |
2332 | { |
2333 | switch (xdp->command) { | |
2334 | case XDP_SETUP_PROG: | |
9861ce03 | 2335 | return virtnet_xdp_set(dev, xdp->prog, xdp->extack); |
f600b690 | 2336 | case XDP_QUERY_PROG: |
5b0e6629 MKL |
2337 | xdp->prog_id = virtnet_xdp_query(dev); |
2338 | xdp->prog_attached = !!xdp->prog_id; | |
f600b690 JF |
2339 | return 0; |
2340 | default: | |
2341 | return -EINVAL; | |
2342 | } | |
2343 | } | |
2344 | ||
ba5e4426 SS |
2345 | static int virtnet_get_phys_port_name(struct net_device *dev, char *buf, |
2346 | size_t len) | |
2347 | { | |
2348 | struct virtnet_info *vi = netdev_priv(dev); | |
2349 | int ret; | |
2350 | ||
2351 | if (!virtio_has_feature(vi->vdev, VIRTIO_NET_F_STANDBY)) | |
2352 | return -EOPNOTSUPP; | |
2353 | ||
2354 | ret = snprintf(buf, len, "sby"); | |
2355 | if (ret >= len) | |
2356 | return -EOPNOTSUPP; | |
2357 | ||
2358 | return 0; | |
2359 | } | |
2360 | ||
76288b4e SH |
2361 | static const struct net_device_ops virtnet_netdev = { |
2362 | .ndo_open = virtnet_open, | |
2363 | .ndo_stop = virtnet_close, | |
2364 | .ndo_start_xmit = start_xmit, | |
2365 | .ndo_validate_addr = eth_validate_addr, | |
9c46f6d4 | 2366 | .ndo_set_mac_address = virtnet_set_mac_address, |
2af7698e | 2367 | .ndo_set_rx_mode = virtnet_set_rx_mode, |
3fa2a1df | 2368 | .ndo_get_stats64 = virtnet_stats, |
1824a989 AW |
2369 | .ndo_vlan_rx_add_vid = virtnet_vlan_rx_add_vid, |
2370 | .ndo_vlan_rx_kill_vid = virtnet_vlan_rx_kill_vid, | |
76288b4e SH |
2371 | #ifdef CONFIG_NET_POLL_CONTROLLER |
2372 | .ndo_poll_controller = virtnet_netpoll, | |
91815639 | 2373 | #endif |
f4e63525 | 2374 | .ndo_bpf = virtnet_xdp, |
186b3c99 | 2375 | .ndo_xdp_xmit = virtnet_xdp_xmit, |
2836b4f2 | 2376 | .ndo_features_check = passthru_features_check, |
ba5e4426 | 2377 | .ndo_get_phys_port_name = virtnet_get_phys_port_name, |
76288b4e SH |
2378 | }; |
2379 | ||
586d17c5 | 2380 | static void virtnet_config_changed_work(struct work_struct *work) |
9f4d26d0 | 2381 | { |
586d17c5 JW |
2382 | struct virtnet_info *vi = |
2383 | container_of(work, struct virtnet_info, config_work); | |
9f4d26d0 MM |
2384 | u16 v; |
2385 | ||
855e0c52 RR |
2386 | if (virtio_cread_feature(vi->vdev, VIRTIO_NET_F_STATUS, |
2387 | struct virtio_net_config, status, &v) < 0) | |
507613bf | 2388 | return; |
586d17c5 JW |
2389 | |
2390 | if (v & VIRTIO_NET_S_ANNOUNCE) { | |
ee89bab1 | 2391 | netdev_notify_peers(vi->dev); |
586d17c5 JW |
2392 | virtnet_ack_link_announce(vi); |
2393 | } | |
9f4d26d0 MM |
2394 | |
2395 | /* Ignore unknown (future) status bits */ | |
2396 | v &= VIRTIO_NET_S_LINK_UP; | |
2397 | ||
2398 | if (vi->status == v) | |
507613bf | 2399 | return; |
9f4d26d0 MM |
2400 | |
2401 | vi->status = v; | |
2402 | ||
2403 | if (vi->status & VIRTIO_NET_S_LINK_UP) { | |
faa9b39f | 2404 | virtnet_update_settings(vi); |
9f4d26d0 | 2405 | netif_carrier_on(vi->dev); |
986a4f4d | 2406 | netif_tx_wake_all_queues(vi->dev); |
9f4d26d0 MM |
2407 | } else { |
2408 | netif_carrier_off(vi->dev); | |
986a4f4d | 2409 | netif_tx_stop_all_queues(vi->dev); |
9f4d26d0 MM |
2410 | } |
2411 | } | |
2412 | ||
2413 | static void virtnet_config_changed(struct virtio_device *vdev) | |
2414 | { | |
2415 | struct virtnet_info *vi = vdev->priv; | |
2416 | ||
3b07e9ca | 2417 | schedule_work(&vi->config_work); |
9f4d26d0 MM |
2418 | } |
2419 | ||
986a4f4d JW |
2420 | static void virtnet_free_queues(struct virtnet_info *vi) |
2421 | { | |
d4fb84ee AV |
2422 | int i; |
2423 | ||
ab3971b1 JW |
2424 | for (i = 0; i < vi->max_queue_pairs; i++) { |
2425 | napi_hash_del(&vi->rq[i].napi); | |
d4fb84ee | 2426 | netif_napi_del(&vi->rq[i].napi); |
b92f1e67 | 2427 | netif_napi_del(&vi->sq[i].napi); |
ab3971b1 | 2428 | } |
d4fb84ee | 2429 | |
963abe5c ED |
2430 | /* We called napi_hash_del() before netif_napi_del(), |
2431 | * we need to respect an RCU grace period before freeing vi->rq | |
2432 | */ | |
2433 | synchronize_net(); | |
2434 | ||
986a4f4d JW |
2435 | kfree(vi->rq); |
2436 | kfree(vi->sq); | |
12e57169 | 2437 | kfree(vi->ctrl); |
986a4f4d JW |
2438 | } |
2439 | ||
47315329 | 2440 | static void _free_receive_bufs(struct virtnet_info *vi) |
986a4f4d | 2441 | { |
f600b690 | 2442 | struct bpf_prog *old_prog; |
986a4f4d JW |
2443 | int i; |
2444 | ||
2445 | for (i = 0; i < vi->max_queue_pairs; i++) { | |
2446 | while (vi->rq[i].pages) | |
2447 | __free_pages(get_a_page(&vi->rq[i], GFP_KERNEL), 0); | |
f600b690 JF |
2448 | |
2449 | old_prog = rtnl_dereference(vi->rq[i].xdp_prog); | |
2450 | RCU_INIT_POINTER(vi->rq[i].xdp_prog, NULL); | |
2451 | if (old_prog) | |
2452 | bpf_prog_put(old_prog); | |
986a4f4d | 2453 | } |
47315329 JF |
2454 | } |
2455 | ||
2456 | static void free_receive_bufs(struct virtnet_info *vi) | |
2457 | { | |
2458 | rtnl_lock(); | |
2459 | _free_receive_bufs(vi); | |
f600b690 | 2460 | rtnl_unlock(); |
986a4f4d JW |
2461 | } |
2462 | ||
fb51879d MD |
2463 | static void free_receive_page_frags(struct virtnet_info *vi) |
2464 | { | |
2465 | int i; | |
2466 | for (i = 0; i < vi->max_queue_pairs; i++) | |
2467 | if (vi->rq[i].alloc_frag.page) | |
2468 | put_page(vi->rq[i].alloc_frag.page); | |
2469 | } | |
2470 | ||
b68df015 | 2471 | static bool is_xdp_raw_buffer_queue(struct virtnet_info *vi, int q) |
56434a01 JF |
2472 | { |
2473 | if (q < (vi->curr_queue_pairs - vi->xdp_queue_pairs)) | |
2474 | return false; | |
2475 | else if (q < vi->curr_queue_pairs) | |
2476 | return true; | |
2477 | else | |
2478 | return false; | |
2479 | } | |
2480 | ||
986a4f4d JW |
2481 | static void free_unused_bufs(struct virtnet_info *vi) |
2482 | { | |
2483 | void *buf; | |
2484 | int i; | |
2485 | ||
2486 | for (i = 0; i < vi->max_queue_pairs; i++) { | |
2487 | struct virtqueue *vq = vi->sq[i].vq; | |
56434a01 | 2488 | while ((buf = virtqueue_detach_unused_buf(vq)) != NULL) { |
b68df015 | 2489 | if (!is_xdp_raw_buffer_queue(vi, i)) |
56434a01 JF |
2490 | dev_kfree_skb(buf); |
2491 | else | |
2492 | put_page(virt_to_head_page(buf)); | |
2493 | } | |
986a4f4d JW |
2494 | } |
2495 | ||
2496 | for (i = 0; i < vi->max_queue_pairs; i++) { | |
2497 | struct virtqueue *vq = vi->rq[i].vq; | |
2498 | ||
2499 | while ((buf = virtqueue_detach_unused_buf(vq)) != NULL) { | |
ab7db917 | 2500 | if (vi->mergeable_rx_bufs) { |
680557cf | 2501 | put_page(virt_to_head_page(buf)); |
ab7db917 | 2502 | } else if (vi->big_packets) { |
fa9fac17 | 2503 | give_pages(&vi->rq[i], buf); |
ab7db917 | 2504 | } else { |
f6b10209 | 2505 | put_page(virt_to_head_page(buf)); |
ab7db917 | 2506 | } |
986a4f4d | 2507 | } |
986a4f4d JW |
2508 | } |
2509 | } | |
2510 | ||
e9d7417b JW |
2511 | static void virtnet_del_vqs(struct virtnet_info *vi) |
2512 | { | |
2513 | struct virtio_device *vdev = vi->vdev; | |
2514 | ||
8898c21c | 2515 | virtnet_clean_affinity(vi, -1); |
986a4f4d | 2516 | |
e9d7417b | 2517 | vdev->config->del_vqs(vdev); |
986a4f4d JW |
2518 | |
2519 | virtnet_free_queues(vi); | |
e9d7417b JW |
2520 | } |
2521 | ||
d85b758f MT |
2522 | /* How large should a single buffer be so a queue full of these can fit at |
2523 | * least one full packet? | |
2524 | * Logic below assumes the mergeable buffer header is used. | |
2525 | */ | |
2526 | static unsigned int mergeable_min_buf_len(struct virtnet_info *vi, struct virtqueue *vq) | |
2527 | { | |
2528 | const unsigned int hdr_len = sizeof(struct virtio_net_hdr_mrg_rxbuf); | |
2529 | unsigned int rq_size = virtqueue_get_vring_size(vq); | |
2530 | unsigned int packet_len = vi->big_packets ? IP_MAX_MTU : vi->dev->max_mtu; | |
2531 | unsigned int buf_len = hdr_len + ETH_HLEN + VLAN_HLEN + packet_len; | |
2532 | unsigned int min_buf_len = DIV_ROUND_UP(buf_len, rq_size); | |
2533 | ||
f0c3192c MT |
2534 | return max(max(min_buf_len, hdr_len) - hdr_len, |
2535 | (unsigned int)GOOD_PACKET_LEN); | |
d85b758f MT |
2536 | } |
2537 | ||
986a4f4d | 2538 | static int virtnet_find_vqs(struct virtnet_info *vi) |
3f9c10b0 | 2539 | { |
986a4f4d JW |
2540 | vq_callback_t **callbacks; |
2541 | struct virtqueue **vqs; | |
2542 | int ret = -ENOMEM; | |
2543 | int i, total_vqs; | |
2544 | const char **names; | |
d45b897b | 2545 | bool *ctx; |
986a4f4d JW |
2546 | |
2547 | /* We expect 1 RX virtqueue followed by 1 TX virtqueue, followed by | |
2548 | * possible N-1 RX/TX queue pairs used in multiqueue mode, followed by | |
2549 | * possible control vq. | |
2550 | */ | |
2551 | total_vqs = vi->max_queue_pairs * 2 + | |
2552 | virtio_has_feature(vi->vdev, VIRTIO_NET_F_CTRL_VQ); | |
2553 | ||
2554 | /* Allocate space for find_vqs parameters */ | |
2555 | vqs = kzalloc(total_vqs * sizeof(*vqs), GFP_KERNEL); | |
2556 | if (!vqs) | |
2557 | goto err_vq; | |
6da2ec56 | 2558 | callbacks = kmalloc_array(total_vqs, sizeof(*callbacks), GFP_KERNEL); |
986a4f4d JW |
2559 | if (!callbacks) |
2560 | goto err_callback; | |
6da2ec56 | 2561 | names = kmalloc_array(total_vqs, sizeof(*names), GFP_KERNEL); |
986a4f4d JW |
2562 | if (!names) |
2563 | goto err_names; | |
192f68cf | 2564 | if (!vi->big_packets || vi->mergeable_rx_bufs) { |
d45b897b MT |
2565 | ctx = kzalloc(total_vqs * sizeof(*ctx), GFP_KERNEL); |
2566 | if (!ctx) | |
2567 | goto err_ctx; | |
2568 | } else { | |
2569 | ctx = NULL; | |
2570 | } | |
986a4f4d JW |
2571 | |
2572 | /* Parameters for control virtqueue, if any */ | |
2573 | if (vi->has_cvq) { | |
2574 | callbacks[total_vqs - 1] = NULL; | |
2575 | names[total_vqs - 1] = "control"; | |
2576 | } | |
3f9c10b0 | 2577 | |
986a4f4d JW |
2578 | /* Allocate/initialize parameters for send/receive virtqueues */ |
2579 | for (i = 0; i < vi->max_queue_pairs; i++) { | |
2580 | callbacks[rxq2vq(i)] = skb_recv_done; | |
2581 | callbacks[txq2vq(i)] = skb_xmit_done; | |
2582 | sprintf(vi->rq[i].name, "input.%d", i); | |
2583 | sprintf(vi->sq[i].name, "output.%d", i); | |
2584 | names[rxq2vq(i)] = vi->rq[i].name; | |
2585 | names[txq2vq(i)] = vi->sq[i].name; | |
d45b897b MT |
2586 | if (ctx) |
2587 | ctx[rxq2vq(i)] = true; | |
986a4f4d | 2588 | } |
3f9c10b0 | 2589 | |
986a4f4d | 2590 | ret = vi->vdev->config->find_vqs(vi->vdev, total_vqs, vqs, callbacks, |
d45b897b | 2591 | names, ctx, NULL); |
986a4f4d JW |
2592 | if (ret) |
2593 | goto err_find; | |
3f9c10b0 | 2594 | |
986a4f4d JW |
2595 | if (vi->has_cvq) { |
2596 | vi->cvq = vqs[total_vqs - 1]; | |
3f9c10b0 | 2597 | if (virtio_has_feature(vi->vdev, VIRTIO_NET_F_CTRL_VLAN)) |
f646968f | 2598 | vi->dev->features |= NETIF_F_HW_VLAN_CTAG_FILTER; |
3f9c10b0 | 2599 | } |
986a4f4d JW |
2600 | |
2601 | for (i = 0; i < vi->max_queue_pairs; i++) { | |
2602 | vi->rq[i].vq = vqs[rxq2vq(i)]; | |
d85b758f | 2603 | vi->rq[i].min_buf_len = mergeable_min_buf_len(vi, vi->rq[i].vq); |
986a4f4d JW |
2604 | vi->sq[i].vq = vqs[txq2vq(i)]; |
2605 | } | |
2606 | ||
2fa3c8a8 | 2607 | /* run here: ret == 0. */ |
986a4f4d | 2608 | |
986a4f4d JW |
2609 | |
2610 | err_find: | |
d45b897b MT |
2611 | kfree(ctx); |
2612 | err_ctx: | |
986a4f4d JW |
2613 | kfree(names); |
2614 | err_names: | |
2615 | kfree(callbacks); | |
2616 | err_callback: | |
2617 | kfree(vqs); | |
2618 | err_vq: | |
2619 | return ret; | |
2620 | } | |
2621 | ||
2622 | static int virtnet_alloc_queues(struct virtnet_info *vi) | |
2623 | { | |
2624 | int i; | |
2625 | ||
12e57169 MT |
2626 | vi->ctrl = kzalloc(sizeof(*vi->ctrl), GFP_KERNEL); |
2627 | if (!vi->ctrl) | |
2628 | goto err_ctrl; | |
986a4f4d JW |
2629 | vi->sq = kzalloc(sizeof(*vi->sq) * vi->max_queue_pairs, GFP_KERNEL); |
2630 | if (!vi->sq) | |
2631 | goto err_sq; | |
2632 | vi->rq = kzalloc(sizeof(*vi->rq) * vi->max_queue_pairs, GFP_KERNEL); | |
008d4278 | 2633 | if (!vi->rq) |
986a4f4d JW |
2634 | goto err_rq; |
2635 | ||
2636 | INIT_DELAYED_WORK(&vi->refill, refill_work); | |
2637 | for (i = 0; i < vi->max_queue_pairs; i++) { | |
2638 | vi->rq[i].pages = NULL; | |
2639 | netif_napi_add(vi->dev, &vi->rq[i].napi, virtnet_poll, | |
2640 | napi_weight); | |
1d11e732 WB |
2641 | netif_tx_napi_add(vi->dev, &vi->sq[i].napi, virtnet_poll_tx, |
2642 | napi_tx ? napi_weight : 0); | |
986a4f4d JW |
2643 | |
2644 | sg_init_table(vi->rq[i].sg, ARRAY_SIZE(vi->rq[i].sg)); | |
5377d758 | 2645 | ewma_pkt_len_init(&vi->rq[i].mrg_avg_pkt_len); |
986a4f4d | 2646 | sg_init_table(vi->sq[i].sg, ARRAY_SIZE(vi->sq[i].sg)); |
d7dfc5cf TM |
2647 | |
2648 | u64_stats_init(&vi->rq[i].stats.syncp); | |
2649 | u64_stats_init(&vi->sq[i].stats.syncp); | |
986a4f4d JW |
2650 | } |
2651 | ||
2652 | return 0; | |
2653 | ||
2654 | err_rq: | |
2655 | kfree(vi->sq); | |
2656 | err_sq: | |
12e57169 MT |
2657 | kfree(vi->ctrl); |
2658 | err_ctrl: | |
986a4f4d JW |
2659 | return -ENOMEM; |
2660 | } | |
2661 | ||
2662 | static int init_vqs(struct virtnet_info *vi) | |
2663 | { | |
2664 | int ret; | |
2665 | ||
2666 | /* Allocate send & receive queues */ | |
2667 | ret = virtnet_alloc_queues(vi); | |
2668 | if (ret) | |
2669 | goto err; | |
2670 | ||
2671 | ret = virtnet_find_vqs(vi); | |
2672 | if (ret) | |
2673 | goto err_free; | |
2674 | ||
47be2479 | 2675 | get_online_cpus(); |
8898c21c | 2676 | virtnet_set_affinity(vi); |
47be2479 WG |
2677 | put_online_cpus(); |
2678 | ||
986a4f4d JW |
2679 | return 0; |
2680 | ||
2681 | err_free: | |
2682 | virtnet_free_queues(vi); | |
2683 | err: | |
2684 | return ret; | |
3f9c10b0 AS |
2685 | } |
2686 | ||
fbf28d78 MD |
2687 | #ifdef CONFIG_SYSFS |
2688 | static ssize_t mergeable_rx_buffer_size_show(struct netdev_rx_queue *queue, | |
718ad681 | 2689 | char *buf) |
fbf28d78 MD |
2690 | { |
2691 | struct virtnet_info *vi = netdev_priv(queue->dev); | |
2692 | unsigned int queue_index = get_netdev_rx_queue_index(queue); | |
3cc81a9a JW |
2693 | unsigned int headroom = virtnet_get_headroom(vi); |
2694 | unsigned int tailroom = headroom ? sizeof(struct skb_shared_info) : 0; | |
5377d758 | 2695 | struct ewma_pkt_len *avg; |
fbf28d78 MD |
2696 | |
2697 | BUG_ON(queue_index >= vi->max_queue_pairs); | |
2698 | avg = &vi->rq[queue_index].mrg_avg_pkt_len; | |
d85b758f | 2699 | return sprintf(buf, "%u\n", |
3cc81a9a JW |
2700 | get_mergeable_buf_len(&vi->rq[queue_index], avg, |
2701 | SKB_DATA_ALIGN(headroom + tailroom))); | |
fbf28d78 MD |
2702 | } |
2703 | ||
2704 | static struct rx_queue_attribute mergeable_rx_buffer_size_attribute = | |
2705 | __ATTR_RO(mergeable_rx_buffer_size); | |
2706 | ||
2707 | static struct attribute *virtio_net_mrg_rx_attrs[] = { | |
2708 | &mergeable_rx_buffer_size_attribute.attr, | |
2709 | NULL | |
2710 | }; | |
2711 | ||
2712 | static const struct attribute_group virtio_net_mrg_rx_group = { | |
2713 | .name = "virtio_net", | |
2714 | .attrs = virtio_net_mrg_rx_attrs | |
2715 | }; | |
2716 | #endif | |
2717 | ||
892d6eb1 JW |
2718 | static bool virtnet_fail_on_feature(struct virtio_device *vdev, |
2719 | unsigned int fbit, | |
2720 | const char *fname, const char *dname) | |
2721 | { | |
2722 | if (!virtio_has_feature(vdev, fbit)) | |
2723 | return false; | |
2724 | ||
2725 | dev_err(&vdev->dev, "device advertises feature %s but not %s", | |
2726 | fname, dname); | |
2727 | ||
2728 | return true; | |
2729 | } | |
2730 | ||
2731 | #define VIRTNET_FAIL_ON(vdev, fbit, dbit) \ | |
2732 | virtnet_fail_on_feature(vdev, fbit, #fbit, dbit) | |
2733 | ||
2734 | static bool virtnet_validate_features(struct virtio_device *vdev) | |
2735 | { | |
2736 | if (!virtio_has_feature(vdev, VIRTIO_NET_F_CTRL_VQ) && | |
2737 | (VIRTNET_FAIL_ON(vdev, VIRTIO_NET_F_CTRL_RX, | |
2738 | "VIRTIO_NET_F_CTRL_VQ") || | |
2739 | VIRTNET_FAIL_ON(vdev, VIRTIO_NET_F_CTRL_VLAN, | |
2740 | "VIRTIO_NET_F_CTRL_VQ") || | |
2741 | VIRTNET_FAIL_ON(vdev, VIRTIO_NET_F_GUEST_ANNOUNCE, | |
2742 | "VIRTIO_NET_F_CTRL_VQ") || | |
2743 | VIRTNET_FAIL_ON(vdev, VIRTIO_NET_F_MQ, "VIRTIO_NET_F_CTRL_VQ") || | |
2744 | VIRTNET_FAIL_ON(vdev, VIRTIO_NET_F_CTRL_MAC_ADDR, | |
2745 | "VIRTIO_NET_F_CTRL_VQ"))) { | |
2746 | return false; | |
2747 | } | |
2748 | ||
2749 | return true; | |
2750 | } | |
2751 | ||
d0c2c997 JW |
2752 | #define MIN_MTU ETH_MIN_MTU |
2753 | #define MAX_MTU ETH_MAX_MTU | |
2754 | ||
fe36cbe0 | 2755 | static int virtnet_validate(struct virtio_device *vdev) |
296f96fc | 2756 | { |
6ba42248 MT |
2757 | if (!vdev->config->get) { |
2758 | dev_err(&vdev->dev, "%s failure: config access disabled\n", | |
2759 | __func__); | |
2760 | return -EINVAL; | |
2761 | } | |
2762 | ||
892d6eb1 JW |
2763 | if (!virtnet_validate_features(vdev)) |
2764 | return -EINVAL; | |
2765 | ||
fe36cbe0 MT |
2766 | if (virtio_has_feature(vdev, VIRTIO_NET_F_MTU)) { |
2767 | int mtu = virtio_cread16(vdev, | |
2768 | offsetof(struct virtio_net_config, | |
2769 | mtu)); | |
2770 | if (mtu < MIN_MTU) | |
2771 | __virtio_clear_bit(vdev, VIRTIO_NET_F_MTU); | |
2772 | } | |
2773 | ||
2774 | return 0; | |
2775 | } | |
2776 | ||
2777 | static int virtnet_probe(struct virtio_device *vdev) | |
2778 | { | |
d7dfc5cf | 2779 | int i, err = -ENOMEM; |
fe36cbe0 MT |
2780 | struct net_device *dev; |
2781 | struct virtnet_info *vi; | |
2782 | u16 max_queue_pairs; | |
2783 | int mtu; | |
2784 | ||
986a4f4d | 2785 | /* Find if host supports multiqueue virtio_net device */ |
855e0c52 RR |
2786 | err = virtio_cread_feature(vdev, VIRTIO_NET_F_MQ, |
2787 | struct virtio_net_config, | |
2788 | max_virtqueue_pairs, &max_queue_pairs); | |
986a4f4d JW |
2789 | |
2790 | /* We need at least 2 queue's */ | |
2791 | if (err || max_queue_pairs < VIRTIO_NET_CTRL_MQ_VQ_PAIRS_MIN || | |
2792 | max_queue_pairs > VIRTIO_NET_CTRL_MQ_VQ_PAIRS_MAX || | |
2793 | !virtio_has_feature(vdev, VIRTIO_NET_F_CTRL_VQ)) | |
2794 | max_queue_pairs = 1; | |
296f96fc RR |
2795 | |
2796 | /* Allocate ourselves a network device with room for our info */ | |
986a4f4d | 2797 | dev = alloc_etherdev_mq(sizeof(struct virtnet_info), max_queue_pairs); |
296f96fc RR |
2798 | if (!dev) |
2799 | return -ENOMEM; | |
2800 | ||
2801 | /* Set up network device as normal. */ | |
f2f2c8b4 | 2802 | dev->priv_flags |= IFF_UNICAST_FLT | IFF_LIVE_ADDR_CHANGE; |
76288b4e | 2803 | dev->netdev_ops = &virtnet_netdev; |
296f96fc | 2804 | dev->features = NETIF_F_HIGHDMA; |
3fa2a1df | 2805 | |
7ad24ea4 | 2806 | dev->ethtool_ops = &virtnet_ethtool_ops; |
296f96fc RR |
2807 | SET_NETDEV_DEV(dev, &vdev->dev); |
2808 | ||
2809 | /* Do we support "hardware" checksums? */ | |
98e778c9 | 2810 | if (virtio_has_feature(vdev, VIRTIO_NET_F_CSUM)) { |
296f96fc | 2811 | /* This opens up the world of extra features. */ |
48900cb6 | 2812 | dev->hw_features |= NETIF_F_HW_CSUM | NETIF_F_SG; |
98e778c9 | 2813 | if (csum) |
48900cb6 | 2814 | dev->features |= NETIF_F_HW_CSUM | NETIF_F_SG; |
98e778c9 MM |
2815 | |
2816 | if (virtio_has_feature(vdev, VIRTIO_NET_F_GSO)) { | |
e078de03 | 2817 | dev->hw_features |= NETIF_F_TSO |
34a48579 RR |
2818 | | NETIF_F_TSO_ECN | NETIF_F_TSO6; |
2819 | } | |
5539ae96 | 2820 | /* Individual feature bits: what can host handle? */ |
98e778c9 MM |
2821 | if (virtio_has_feature(vdev, VIRTIO_NET_F_HOST_TSO4)) |
2822 | dev->hw_features |= NETIF_F_TSO; | |
2823 | if (virtio_has_feature(vdev, VIRTIO_NET_F_HOST_TSO6)) | |
2824 | dev->hw_features |= NETIF_F_TSO6; | |
2825 | if (virtio_has_feature(vdev, VIRTIO_NET_F_HOST_ECN)) | |
2826 | dev->hw_features |= NETIF_F_TSO_ECN; | |
98e778c9 | 2827 | |
41f2f127 JW |
2828 | dev->features |= NETIF_F_GSO_ROBUST; |
2829 | ||
98e778c9 | 2830 | if (gso) |
e078de03 | 2831 | dev->features |= dev->hw_features & NETIF_F_ALL_TSO; |
98e778c9 | 2832 | /* (!csum && gso) case will be fixed by register_netdev() */ |
296f96fc | 2833 | } |
4f49129b TH |
2834 | if (virtio_has_feature(vdev, VIRTIO_NET_F_GUEST_CSUM)) |
2835 | dev->features |= NETIF_F_RXCSUM; | |
296f96fc | 2836 | |
4fda8302 JW |
2837 | dev->vlan_features = dev->features; |
2838 | ||
d0c2c997 JW |
2839 | /* MTU range: 68 - 65535 */ |
2840 | dev->min_mtu = MIN_MTU; | |
2841 | dev->max_mtu = MAX_MTU; | |
2842 | ||
296f96fc | 2843 | /* Configuration may specify what MAC to use. Otherwise random. */ |
855e0c52 RR |
2844 | if (virtio_has_feature(vdev, VIRTIO_NET_F_MAC)) |
2845 | virtio_cread_bytes(vdev, | |
2846 | offsetof(struct virtio_net_config, mac), | |
2847 | dev->dev_addr, dev->addr_len); | |
2848 | else | |
f2cedb63 | 2849 | eth_hw_addr_random(dev); |
296f96fc RR |
2850 | |
2851 | /* Set up our device-specific information */ | |
2852 | vi = netdev_priv(dev); | |
296f96fc RR |
2853 | vi->dev = dev; |
2854 | vi->vdev = vdev; | |
d9d5dcc8 | 2855 | vdev->priv = vi; |
827da44c | 2856 | |
586d17c5 | 2857 | INIT_WORK(&vi->config_work, virtnet_config_changed_work); |
296f96fc | 2858 | |
97402b96 | 2859 | /* If we can receive ANY GSO packets, we must allocate large ones. */ |
8e95a202 JP |
2860 | if (virtio_has_feature(vdev, VIRTIO_NET_F_GUEST_TSO4) || |
2861 | virtio_has_feature(vdev, VIRTIO_NET_F_GUEST_TSO6) || | |
e3e3c423 VY |
2862 | virtio_has_feature(vdev, VIRTIO_NET_F_GUEST_ECN) || |
2863 | virtio_has_feature(vdev, VIRTIO_NET_F_GUEST_UFO)) | |
97402b96 HX |
2864 | vi->big_packets = true; |
2865 | ||
3f2c31d9 MM |
2866 | if (virtio_has_feature(vdev, VIRTIO_NET_F_MRG_RXBUF)) |
2867 | vi->mergeable_rx_bufs = true; | |
2868 | ||
d04302b3 MT |
2869 | if (virtio_has_feature(vdev, VIRTIO_NET_F_MRG_RXBUF) || |
2870 | virtio_has_feature(vdev, VIRTIO_F_VERSION_1)) | |
012873d0 MT |
2871 | vi->hdr_len = sizeof(struct virtio_net_hdr_mrg_rxbuf); |
2872 | else | |
2873 | vi->hdr_len = sizeof(struct virtio_net_hdr); | |
2874 | ||
75993300 MT |
2875 | if (virtio_has_feature(vdev, VIRTIO_F_ANY_LAYOUT) || |
2876 | virtio_has_feature(vdev, VIRTIO_F_VERSION_1)) | |
e7428e95 MT |
2877 | vi->any_header_sg = true; |
2878 | ||
986a4f4d JW |
2879 | if (virtio_has_feature(vdev, VIRTIO_NET_F_CTRL_VQ)) |
2880 | vi->has_cvq = true; | |
2881 | ||
14de9d11 AC |
2882 | if (virtio_has_feature(vdev, VIRTIO_NET_F_MTU)) { |
2883 | mtu = virtio_cread16(vdev, | |
2884 | offsetof(struct virtio_net_config, | |
2885 | mtu)); | |
93a205ee | 2886 | if (mtu < dev->min_mtu) { |
fe36cbe0 MT |
2887 | /* Should never trigger: MTU was previously validated |
2888 | * in virtnet_validate. | |
2889 | */ | |
2890 | dev_err(&vdev->dev, "device MTU appears to have changed " | |
2891 | "it is now %d < %d", mtu, dev->min_mtu); | |
d7dfc5cf | 2892 | goto free; |
93a205ee | 2893 | } |
2e123b44 | 2894 | |
fe36cbe0 MT |
2895 | dev->mtu = mtu; |
2896 | dev->max_mtu = mtu; | |
2897 | ||
2e123b44 MT |
2898 | /* TODO: size buffers correctly in this case. */ |
2899 | if (dev->mtu > ETH_DATA_LEN) | |
2900 | vi->big_packets = true; | |
14de9d11 AC |
2901 | } |
2902 | ||
012873d0 MT |
2903 | if (vi->any_header_sg) |
2904 | dev->needed_headroom = vi->hdr_len; | |
6ebbc1a6 | 2905 | |
44900010 JW |
2906 | /* Enable multiqueue by default */ |
2907 | if (num_online_cpus() >= max_queue_pairs) | |
2908 | vi->curr_queue_pairs = max_queue_pairs; | |
2909 | else | |
2910 | vi->curr_queue_pairs = num_online_cpus(); | |
986a4f4d JW |
2911 | vi->max_queue_pairs = max_queue_pairs; |
2912 | ||
2913 | /* Allocate/initialize the rx/tx queues, and invoke find_vqs */ | |
3f9c10b0 | 2914 | err = init_vqs(vi); |
d2a7ddda | 2915 | if (err) |
d7dfc5cf | 2916 | goto free; |
296f96fc | 2917 | |
fbf28d78 MD |
2918 | #ifdef CONFIG_SYSFS |
2919 | if (vi->mergeable_rx_bufs) | |
2920 | dev->sysfs_rx_queue_group = &virtio_net_mrg_rx_group; | |
2921 | #endif | |
0f13b66b ZYW |
2922 | netif_set_real_num_tx_queues(dev, vi->curr_queue_pairs); |
2923 | netif_set_real_num_rx_queues(dev, vi->curr_queue_pairs); | |
986a4f4d | 2924 | |
16032be5 NA |
2925 | virtnet_init_settings(dev); |
2926 | ||
ba5e4426 SS |
2927 | if (virtio_has_feature(vdev, VIRTIO_NET_F_STANDBY)) { |
2928 | vi->failover = net_failover_create(vi->dev); | |
4b8e6ac4 WY |
2929 | if (IS_ERR(vi->failover)) { |
2930 | err = PTR_ERR(vi->failover); | |
ba5e4426 | 2931 | goto free_vqs; |
4b8e6ac4 | 2932 | } |
ba5e4426 SS |
2933 | } |
2934 | ||
296f96fc RR |
2935 | err = register_netdev(dev); |
2936 | if (err) { | |
2937 | pr_debug("virtio_net: registering device failed\n"); | |
ba5e4426 | 2938 | goto free_failover; |
296f96fc | 2939 | } |
b3369c1f | 2940 | |
4baf1e33 MT |
2941 | virtio_device_ready(vdev); |
2942 | ||
8017c279 | 2943 | err = virtnet_cpu_notif_add(vi); |
8de4b2f3 WG |
2944 | if (err) { |
2945 | pr_debug("virtio_net: registering cpu notifier failed\n"); | |
f00e35e2 | 2946 | goto free_unregister_netdev; |
8de4b2f3 WG |
2947 | } |
2948 | ||
a220871b | 2949 | virtnet_set_queues(vi, vi->curr_queue_pairs); |
44900010 | 2950 | |
167c25e4 JW |
2951 | /* Assume link up if device can't report link status, |
2952 | otherwise get link status from config. */ | |
bda7fab5 | 2953 | netif_carrier_off(dev); |
167c25e4 | 2954 | if (virtio_has_feature(vi->vdev, VIRTIO_NET_F_STATUS)) { |
3b07e9ca | 2955 | schedule_work(&vi->config_work); |
167c25e4 JW |
2956 | } else { |
2957 | vi->status = VIRTIO_NET_S_LINK_UP; | |
faa9b39f | 2958 | virtnet_update_settings(vi); |
167c25e4 JW |
2959 | netif_carrier_on(dev); |
2960 | } | |
9f4d26d0 | 2961 | |
3f93522f JW |
2962 | for (i = 0; i < ARRAY_SIZE(guest_offloads); i++) |
2963 | if (virtio_has_feature(vi->vdev, guest_offloads[i])) | |
2964 | set_bit(guest_offloads[i], &vi->guest_offloads); | |
2965 | ||
986a4f4d JW |
2966 | pr_debug("virtnet: registered device %s with %d RX and TX vq's\n", |
2967 | dev->name, max_queue_pairs); | |
2968 | ||
296f96fc RR |
2969 | return 0; |
2970 | ||
f00e35e2 | 2971 | free_unregister_netdev: |
02465555 MT |
2972 | vi->vdev->config->reset(vdev); |
2973 | ||
b3369c1f | 2974 | unregister_netdev(dev); |
ba5e4426 SS |
2975 | free_failover: |
2976 | net_failover_destroy(vi->failover); | |
d2a7ddda | 2977 | free_vqs: |
986a4f4d | 2978 | cancel_delayed_work_sync(&vi->refill); |
fb51879d | 2979 | free_receive_page_frags(vi); |
e9d7417b | 2980 | virtnet_del_vqs(vi); |
296f96fc RR |
2981 | free: |
2982 | free_netdev(dev); | |
2983 | return err; | |
2984 | } | |
2985 | ||
04486ed0 | 2986 | static void remove_vq_common(struct virtnet_info *vi) |
296f96fc | 2987 | { |
04486ed0 | 2988 | vi->vdev->config->reset(vi->vdev); |
830a8a97 SM |
2989 | |
2990 | /* Free unused buffers in both send and recv, if any. */ | |
9ab86bbc | 2991 | free_unused_bufs(vi); |
fb6813f4 | 2992 | |
986a4f4d | 2993 | free_receive_bufs(vi); |
d2a7ddda | 2994 | |
fb51879d MD |
2995 | free_receive_page_frags(vi); |
2996 | ||
986a4f4d | 2997 | virtnet_del_vqs(vi); |
04486ed0 AS |
2998 | } |
2999 | ||
8cc085d6 | 3000 | static void virtnet_remove(struct virtio_device *vdev) |
04486ed0 AS |
3001 | { |
3002 | struct virtnet_info *vi = vdev->priv; | |
3003 | ||
8017c279 | 3004 | virtnet_cpu_notif_remove(vi); |
8de4b2f3 | 3005 | |
102a2786 MT |
3006 | /* Make sure no work handler is accessing the device. */ |
3007 | flush_work(&vi->config_work); | |
586d17c5 | 3008 | |
04486ed0 AS |
3009 | unregister_netdev(vi->dev); |
3010 | ||
ba5e4426 SS |
3011 | net_failover_destroy(vi->failover); |
3012 | ||
04486ed0 | 3013 | remove_vq_common(vi); |
fb6813f4 | 3014 | |
74b2553f | 3015 | free_netdev(vi->dev); |
296f96fc RR |
3016 | } |
3017 | ||
67a75194 | 3018 | static __maybe_unused int virtnet_freeze(struct virtio_device *vdev) |
0741bcb5 AS |
3019 | { |
3020 | struct virtnet_info *vi = vdev->priv; | |
3021 | ||
8017c279 | 3022 | virtnet_cpu_notif_remove(vi); |
9fe7bfce | 3023 | virtnet_freeze_down(vdev); |
0741bcb5 AS |
3024 | remove_vq_common(vi); |
3025 | ||
3026 | return 0; | |
3027 | } | |
3028 | ||
67a75194 | 3029 | static __maybe_unused int virtnet_restore(struct virtio_device *vdev) |
0741bcb5 AS |
3030 | { |
3031 | struct virtnet_info *vi = vdev->priv; | |
9fe7bfce | 3032 | int err; |
0741bcb5 | 3033 | |
9fe7bfce | 3034 | err = virtnet_restore_up(vdev); |
0741bcb5 AS |
3035 | if (err) |
3036 | return err; | |
986a4f4d JW |
3037 | virtnet_set_queues(vi, vi->curr_queue_pairs); |
3038 | ||
8017c279 | 3039 | err = virtnet_cpu_notif_add(vi); |
ec9debbd JW |
3040 | if (err) |
3041 | return err; | |
3042 | ||
0741bcb5 AS |
3043 | return 0; |
3044 | } | |
0741bcb5 | 3045 | |
296f96fc RR |
3046 | static struct virtio_device_id id_table[] = { |
3047 | { VIRTIO_ID_NET, VIRTIO_DEV_ANY_ID }, | |
3048 | { 0 }, | |
3049 | }; | |
3050 | ||
f3358507 MT |
3051 | #define VIRTNET_FEATURES \ |
3052 | VIRTIO_NET_F_CSUM, VIRTIO_NET_F_GUEST_CSUM, \ | |
3053 | VIRTIO_NET_F_MAC, \ | |
3054 | VIRTIO_NET_F_HOST_TSO4, VIRTIO_NET_F_HOST_UFO, VIRTIO_NET_F_HOST_TSO6, \ | |
3055 | VIRTIO_NET_F_HOST_ECN, VIRTIO_NET_F_GUEST_TSO4, VIRTIO_NET_F_GUEST_TSO6, \ | |
3056 | VIRTIO_NET_F_GUEST_ECN, VIRTIO_NET_F_GUEST_UFO, \ | |
3057 | VIRTIO_NET_F_MRG_RXBUF, VIRTIO_NET_F_STATUS, VIRTIO_NET_F_CTRL_VQ, \ | |
3058 | VIRTIO_NET_F_CTRL_RX, VIRTIO_NET_F_CTRL_VLAN, \ | |
3059 | VIRTIO_NET_F_GUEST_ANNOUNCE, VIRTIO_NET_F_MQ, \ | |
3060 | VIRTIO_NET_F_CTRL_MAC_ADDR, \ | |
faa9b39f | 3061 | VIRTIO_NET_F_MTU, VIRTIO_NET_F_CTRL_GUEST_OFFLOADS, \ |
9805069d | 3062 | VIRTIO_NET_F_SPEED_DUPLEX, VIRTIO_NET_F_STANDBY |
f3358507 | 3063 | |
c45a6816 | 3064 | static unsigned int features[] = { |
f3358507 MT |
3065 | VIRTNET_FEATURES, |
3066 | }; | |
3067 | ||
3068 | static unsigned int features_legacy[] = { | |
3069 | VIRTNET_FEATURES, | |
3070 | VIRTIO_NET_F_GSO, | |
e7428e95 | 3071 | VIRTIO_F_ANY_LAYOUT, |
c45a6816 RR |
3072 | }; |
3073 | ||
22402529 | 3074 | static struct virtio_driver virtio_net_driver = { |
c45a6816 RR |
3075 | .feature_table = features, |
3076 | .feature_table_size = ARRAY_SIZE(features), | |
f3358507 MT |
3077 | .feature_table_legacy = features_legacy, |
3078 | .feature_table_size_legacy = ARRAY_SIZE(features_legacy), | |
296f96fc RR |
3079 | .driver.name = KBUILD_MODNAME, |
3080 | .driver.owner = THIS_MODULE, | |
3081 | .id_table = id_table, | |
fe36cbe0 | 3082 | .validate = virtnet_validate, |
296f96fc | 3083 | .probe = virtnet_probe, |
8cc085d6 | 3084 | .remove = virtnet_remove, |
9f4d26d0 | 3085 | .config_changed = virtnet_config_changed, |
89107000 | 3086 | #ifdef CONFIG_PM_SLEEP |
0741bcb5 AS |
3087 | .freeze = virtnet_freeze, |
3088 | .restore = virtnet_restore, | |
3089 | #endif | |
296f96fc RR |
3090 | }; |
3091 | ||
8017c279 SAS |
3092 | static __init int virtio_net_driver_init(void) |
3093 | { | |
3094 | int ret; | |
3095 | ||
73c1b41e | 3096 | ret = cpuhp_setup_state_multi(CPUHP_AP_ONLINE_DYN, "virtio/net:online", |
8017c279 SAS |
3097 | virtnet_cpu_online, |
3098 | virtnet_cpu_down_prep); | |
3099 | if (ret < 0) | |
3100 | goto out; | |
3101 | virtionet_online = ret; | |
73c1b41e | 3102 | ret = cpuhp_setup_state_multi(CPUHP_VIRT_NET_DEAD, "virtio/net:dead", |
8017c279 SAS |
3103 | NULL, virtnet_cpu_dead); |
3104 | if (ret) | |
3105 | goto err_dead; | |
3106 | ||
3107 | ret = register_virtio_driver(&virtio_net_driver); | |
3108 | if (ret) | |
3109 | goto err_virtio; | |
3110 | return 0; | |
3111 | err_virtio: | |
3112 | cpuhp_remove_multi_state(CPUHP_VIRT_NET_DEAD); | |
3113 | err_dead: | |
3114 | cpuhp_remove_multi_state(virtionet_online); | |
3115 | out: | |
3116 | return ret; | |
3117 | } | |
3118 | module_init(virtio_net_driver_init); | |
3119 | ||
3120 | static __exit void virtio_net_driver_exit(void) | |
3121 | { | |
cfa0ebc9 | 3122 | unregister_virtio_driver(&virtio_net_driver); |
8017c279 SAS |
3123 | cpuhp_remove_multi_state(CPUHP_VIRT_NET_DEAD); |
3124 | cpuhp_remove_multi_state(virtionet_online); | |
8017c279 SAS |
3125 | } |
3126 | module_exit(virtio_net_driver_exit); | |
296f96fc RR |
3127 | |
3128 | MODULE_DEVICE_TABLE(virtio, id_table); | |
3129 | MODULE_DESCRIPTION("Virtio network driver"); | |
3130 | MODULE_LICENSE("GPL"); |