Commit | Line | Data |
---|---|---|
1ccea77e | 1 | // SPDX-License-Identifier: GPL-2.0-or-later |
48925e37 | 2 | /* A network driver using virtio. |
296f96fc RR |
3 | * |
4 | * Copyright 2007 Rusty Russell <rusty@rustcorp.com.au> IBM Corporation | |
296f96fc RR |
5 | */ |
6 | //#define DEBUG | |
7 | #include <linux/netdevice.h> | |
8 | #include <linux/etherdevice.h> | |
a9ea3fc6 | 9 | #include <linux/ethtool.h> |
296f96fc RR |
10 | #include <linux/module.h> |
11 | #include <linux/virtio.h> | |
12 | #include <linux/virtio_net.h> | |
f600b690 | 13 | #include <linux/bpf.h> |
a67edbf4 | 14 | #include <linux/bpf_trace.h> |
296f96fc | 15 | #include <linux/scatterlist.h> |
e918085a | 16 | #include <linux/if_vlan.h> |
5a0e3ad6 | 17 | #include <linux/slab.h> |
8de4b2f3 | 18 | #include <linux/cpu.h> |
ab7db917 | 19 | #include <linux/average.h> |
186b3c99 | 20 | #include <linux/filter.h> |
2ca653d6 | 21 | #include <linux/kernel.h> |
d85b758f | 22 | #include <net/route.h> |
754b8a21 | 23 | #include <net/xdp.h> |
ba5e4426 | 24 | #include <net/net_failover.h> |
296f96fc | 25 | |
d34710e3 | 26 | static int napi_weight = NAPI_POLL_WEIGHT; |
6c0cd7c0 DL |
27 | module_param(napi_weight, int, 0444); |
28 | ||
31c03aef | 29 | static bool csum = true, gso = true, napi_tx = true; |
34a48579 RR |
30 | module_param(csum, bool, 0444); |
31 | module_param(gso, bool, 0444); | |
b92f1e67 | 32 | module_param(napi_tx, bool, 0644); |
34a48579 | 33 | |
296f96fc | 34 | /* FIXME: MTU in config. */ |
5061de36 | 35 | #define GOOD_PACKET_LEN (ETH_HLEN + VLAN_HLEN + ETH_DATA_LEN) |
3f2c31d9 | 36 | #define GOOD_COPY_LEN 128 |
296f96fc | 37 | |
f6b10209 JW |
38 | #define VIRTNET_RX_PAD (NET_IP_ALIGN + NET_SKB_PAD) |
39 | ||
2de2f7f4 JF |
40 | /* Amount of XDP headroom to prepend to packets for use by xdp_adjust_head */ |
41 | #define VIRTIO_XDP_HEADROOM 256 | |
42 | ||
2471c75e JDB |
43 | /* Separating two types of XDP xmit */ |
44 | #define VIRTIO_XDP_TX BIT(0) | |
45 | #define VIRTIO_XDP_REDIR BIT(1) | |
46 | ||
5050471d TM |
47 | #define VIRTIO_XDP_FLAG BIT(0) |
48 | ||
5377d758 JB |
49 | /* RX packet size EWMA. The average packet size is used to determine the packet |
50 | * buffer size when refilling RX rings. As the entire RX ring may be refilled | |
51 | * at once, the weight is chosen so that the EWMA will be insensitive to short- | |
52 | * term, transient changes in packet size. | |
ab7db917 | 53 | */ |
eb1e011a | 54 | DECLARE_EWMA(pkt_len, 0, 64) |
ab7db917 | 55 | |
66846048 | 56 | #define VIRTNET_DRIVER_VERSION "1.0.0" |
2a41f71d | 57 | |
7acd4329 CIK |
58 | static const unsigned long guest_offloads[] = { |
59 | VIRTIO_NET_F_GUEST_TSO4, | |
60 | VIRTIO_NET_F_GUEST_TSO6, | |
61 | VIRTIO_NET_F_GUEST_ECN, | |
e59ff2c4 JW |
62 | VIRTIO_NET_F_GUEST_UFO, |
63 | VIRTIO_NET_F_GUEST_CSUM | |
7acd4329 | 64 | }; |
3f93522f | 65 | |
1a03b8a3 TZ |
66 | #define GUEST_OFFLOAD_LRO_MASK ((1ULL << VIRTIO_NET_F_GUEST_TSO4) | \ |
67 | (1ULL << VIRTIO_NET_F_GUEST_TSO6) | \ | |
68 | (1ULL << VIRTIO_NET_F_GUEST_ECN) | \ | |
69 | (1ULL << VIRTIO_NET_F_GUEST_UFO)) | |
70 | ||
d7dfc5cf TM |
71 | struct virtnet_stat_desc { |
72 | char desc[ETH_GSTRING_LEN]; | |
73 | size_t offset; | |
3fa2a1df | 74 | }; |
75 | ||
d7dfc5cf TM |
76 | struct virtnet_sq_stats { |
77 | struct u64_stats_sync syncp; | |
78 | u64 packets; | |
79 | u64 bytes; | |
5b8f3c8d TM |
80 | u64 xdp_tx; |
81 | u64 xdp_tx_drops; | |
461f03dc | 82 | u64 kicks; |
d7dfc5cf TM |
83 | }; |
84 | ||
d46eeeaf JW |
85 | struct virtnet_rq_stats { |
86 | struct u64_stats_sync syncp; | |
d7dfc5cf TM |
87 | u64 packets; |
88 | u64 bytes; | |
2c4a2f7d | 89 | u64 drops; |
5b8f3c8d TM |
90 | u64 xdp_packets; |
91 | u64 xdp_tx; | |
92 | u64 xdp_redirects; | |
93 | u64 xdp_drops; | |
461f03dc | 94 | u64 kicks; |
d7dfc5cf TM |
95 | }; |
96 | ||
97 | #define VIRTNET_SQ_STAT(m) offsetof(struct virtnet_sq_stats, m) | |
d46eeeaf | 98 | #define VIRTNET_RQ_STAT(m) offsetof(struct virtnet_rq_stats, m) |
d7dfc5cf TM |
99 | |
100 | static const struct virtnet_stat_desc virtnet_sq_stats_desc[] = { | |
5b8f3c8d TM |
101 | { "packets", VIRTNET_SQ_STAT(packets) }, |
102 | { "bytes", VIRTNET_SQ_STAT(bytes) }, | |
103 | { "xdp_tx", VIRTNET_SQ_STAT(xdp_tx) }, | |
104 | { "xdp_tx_drops", VIRTNET_SQ_STAT(xdp_tx_drops) }, | |
461f03dc | 105 | { "kicks", VIRTNET_SQ_STAT(kicks) }, |
d7dfc5cf TM |
106 | }; |
107 | ||
108 | static const struct virtnet_stat_desc virtnet_rq_stats_desc[] = { | |
5b8f3c8d TM |
109 | { "packets", VIRTNET_RQ_STAT(packets) }, |
110 | { "bytes", VIRTNET_RQ_STAT(bytes) }, | |
111 | { "drops", VIRTNET_RQ_STAT(drops) }, | |
112 | { "xdp_packets", VIRTNET_RQ_STAT(xdp_packets) }, | |
113 | { "xdp_tx", VIRTNET_RQ_STAT(xdp_tx) }, | |
114 | { "xdp_redirects", VIRTNET_RQ_STAT(xdp_redirects) }, | |
115 | { "xdp_drops", VIRTNET_RQ_STAT(xdp_drops) }, | |
461f03dc | 116 | { "kicks", VIRTNET_RQ_STAT(kicks) }, |
d7dfc5cf TM |
117 | }; |
118 | ||
119 | #define VIRTNET_SQ_STATS_LEN ARRAY_SIZE(virtnet_sq_stats_desc) | |
120 | #define VIRTNET_RQ_STATS_LEN ARRAY_SIZE(virtnet_rq_stats_desc) | |
121 | ||
e9d7417b JW |
122 | /* Internal representation of a send virtqueue */ |
123 | struct send_queue { | |
124 | /* Virtqueue associated with this send _queue */ | |
125 | struct virtqueue *vq; | |
126 | ||
127 | /* TX: fragments + linear part + virtio header */ | |
128 | struct scatterlist sg[MAX_SKB_FRAGS + 2]; | |
986a4f4d JW |
129 | |
130 | /* Name of the send queue: output.$index */ | |
131 | char name[40]; | |
b92f1e67 | 132 | |
d7dfc5cf TM |
133 | struct virtnet_sq_stats stats; |
134 | ||
b92f1e67 | 135 | struct napi_struct napi; |
e9d7417b JW |
136 | }; |
137 | ||
138 | /* Internal representation of a receive virtqueue */ | |
139 | struct receive_queue { | |
140 | /* Virtqueue associated with this receive_queue */ | |
141 | struct virtqueue *vq; | |
142 | ||
296f96fc RR |
143 | struct napi_struct napi; |
144 | ||
f600b690 JF |
145 | struct bpf_prog __rcu *xdp_prog; |
146 | ||
d7dfc5cf TM |
147 | struct virtnet_rq_stats stats; |
148 | ||
e9d7417b JW |
149 | /* Chain pages by the private ptr. */ |
150 | struct page *pages; | |
151 | ||
ab7db917 | 152 | /* Average packet length for mergeable receive buffers. */ |
5377d758 | 153 | struct ewma_pkt_len mrg_avg_pkt_len; |
ab7db917 | 154 | |
fb51879d MD |
155 | /* Page frag for packet buffer allocation. */ |
156 | struct page_frag alloc_frag; | |
157 | ||
e9d7417b JW |
158 | /* RX: fragments + linear part + virtio header */ |
159 | struct scatterlist sg[MAX_SKB_FRAGS + 2]; | |
986a4f4d | 160 | |
d85b758f MT |
161 | /* Min single buffer size for mergeable buffers case. */ |
162 | unsigned int min_buf_len; | |
163 | ||
986a4f4d JW |
164 | /* Name of this receive queue: input.$index */ |
165 | char name[40]; | |
754b8a21 JDB |
166 | |
167 | struct xdp_rxq_info xdp_rxq; | |
e9d7417b JW |
168 | }; |
169 | ||
12e57169 MT |
170 | /* Control VQ buffers: protected by the rtnl lock */ |
171 | struct control_buf { | |
172 | struct virtio_net_ctrl_hdr hdr; | |
173 | virtio_net_ctrl_ack status; | |
174 | struct virtio_net_ctrl_mq mq; | |
175 | u8 promisc; | |
176 | u8 allmulti; | |
d7fad4c8 | 177 | __virtio16 vid; |
f4ee703a | 178 | __virtio64 offloads; |
12e57169 MT |
179 | }; |
180 | ||
e9d7417b JW |
181 | struct virtnet_info { |
182 | struct virtio_device *vdev; | |
183 | struct virtqueue *cvq; | |
184 | struct net_device *dev; | |
986a4f4d JW |
185 | struct send_queue *sq; |
186 | struct receive_queue *rq; | |
e9d7417b JW |
187 | unsigned int status; |
188 | ||
986a4f4d JW |
189 | /* Max # of queue pairs supported by the device */ |
190 | u16 max_queue_pairs; | |
191 | ||
192 | /* # of queue pairs currently used by the driver */ | |
193 | u16 curr_queue_pairs; | |
194 | ||
672aafd5 JF |
195 | /* # of XDP queue pairs currently used by the driver */ |
196 | u16 xdp_queue_pairs; | |
197 | ||
97c2c69e XZ |
198 | /* xdp_queue_pairs may be 0, when xdp is already loaded. So add this. */ |
199 | bool xdp_enabled; | |
200 | ||
97402b96 HX |
201 | /* I like... big packets and I cannot lie! */ |
202 | bool big_packets; | |
203 | ||
3f2c31d9 MM |
204 | /* Host will merge rx buffers for big packets (shake it! shake it!) */ |
205 | bool mergeable_rx_bufs; | |
206 | ||
986a4f4d JW |
207 | /* Has control virtqueue */ |
208 | bool has_cvq; | |
209 | ||
e7428e95 MT |
210 | /* Host can handle any s/g split between our header and packet data */ |
211 | bool any_header_sg; | |
212 | ||
012873d0 MT |
213 | /* Packet virtio header size */ |
214 | u8 hdr_len; | |
215 | ||
3161e453 RR |
216 | /* Work struct for refilling if we run low on memory. */ |
217 | struct delayed_work refill; | |
218 | ||
586d17c5 JW |
219 | /* Work struct for config space updates */ |
220 | struct work_struct config_work; | |
221 | ||
986a4f4d JW |
222 | /* Does the affinity hint is set for virtqueues? */ |
223 | bool affinity_hint_set; | |
47be2479 | 224 | |
8017c279 SAS |
225 | /* CPU hotplug instances for online & dead */ |
226 | struct hlist_node node; | |
227 | struct hlist_node node_dead; | |
2ac46030 | 228 | |
12e57169 | 229 | struct control_buf *ctrl; |
16032be5 NA |
230 | |
231 | /* Ethtool settings */ | |
232 | u8 duplex; | |
233 | u32 speed; | |
3f93522f JW |
234 | |
235 | unsigned long guest_offloads; | |
a02e8964 | 236 | unsigned long guest_offloads_capable; |
ba5e4426 SS |
237 | |
238 | /* failover when STANDBY feature enabled */ | |
239 | struct failover *failover; | |
296f96fc RR |
240 | }; |
241 | ||
9ab86bbc | 242 | struct padded_vnet_hdr { |
012873d0 | 243 | struct virtio_net_hdr_mrg_rxbuf hdr; |
9ab86bbc | 244 | /* |
012873d0 MT |
245 | * hdr is in a separate sg buffer, and data sg buffer shares same page |
246 | * with this header sg. This padding makes next sg 16 byte aligned | |
247 | * after the header. | |
9ab86bbc | 248 | */ |
012873d0 | 249 | char padding[4]; |
9ab86bbc SM |
250 | }; |
251 | ||
5050471d TM |
252 | static bool is_xdp_frame(void *ptr) |
253 | { | |
254 | return (unsigned long)ptr & VIRTIO_XDP_FLAG; | |
255 | } | |
256 | ||
257 | static void *xdp_to_ptr(struct xdp_frame *ptr) | |
258 | { | |
259 | return (void *)((unsigned long)ptr | VIRTIO_XDP_FLAG); | |
260 | } | |
261 | ||
262 | static struct xdp_frame *ptr_to_xdp(void *ptr) | |
263 | { | |
264 | return (struct xdp_frame *)((unsigned long)ptr & ~VIRTIO_XDP_FLAG); | |
265 | } | |
266 | ||
986a4f4d JW |
267 | /* Converting between virtqueue no. and kernel tx/rx queue no. |
268 | * 0:rx0 1:tx0 2:rx1 3:tx1 ... 2N:rxN 2N+1:txN 2N+2:cvq | |
269 | */ | |
270 | static int vq2txq(struct virtqueue *vq) | |
271 | { | |
9d0ca6ed | 272 | return (vq->index - 1) / 2; |
986a4f4d JW |
273 | } |
274 | ||
275 | static int txq2vq(int txq) | |
276 | { | |
277 | return txq * 2 + 1; | |
278 | } | |
279 | ||
280 | static int vq2rxq(struct virtqueue *vq) | |
281 | { | |
9d0ca6ed | 282 | return vq->index / 2; |
986a4f4d JW |
283 | } |
284 | ||
285 | static int rxq2vq(int rxq) | |
286 | { | |
287 | return rxq * 2; | |
288 | } | |
289 | ||
012873d0 | 290 | static inline struct virtio_net_hdr_mrg_rxbuf *skb_vnet_hdr(struct sk_buff *skb) |
296f96fc | 291 | { |
012873d0 | 292 | return (struct virtio_net_hdr_mrg_rxbuf *)skb->cb; |
296f96fc RR |
293 | } |
294 | ||
9ab86bbc SM |
295 | /* |
296 | * private is used to chain pages for big packets, put the whole | |
297 | * most recent used list in the beginning for reuse | |
298 | */ | |
e9d7417b | 299 | static void give_pages(struct receive_queue *rq, struct page *page) |
0a888fd1 | 300 | { |
9ab86bbc | 301 | struct page *end; |
0a888fd1 | 302 | |
e9d7417b | 303 | /* Find end of list, sew whole thing into vi->rq.pages. */ |
9ab86bbc | 304 | for (end = page; end->private; end = (struct page *)end->private); |
e9d7417b JW |
305 | end->private = (unsigned long)rq->pages; |
306 | rq->pages = page; | |
0a888fd1 MM |
307 | } |
308 | ||
e9d7417b | 309 | static struct page *get_a_page(struct receive_queue *rq, gfp_t gfp_mask) |
fb6813f4 | 310 | { |
e9d7417b | 311 | struct page *p = rq->pages; |
fb6813f4 | 312 | |
9ab86bbc | 313 | if (p) { |
e9d7417b | 314 | rq->pages = (struct page *)p->private; |
9ab86bbc SM |
315 | /* clear private here, it is used to chain pages */ |
316 | p->private = 0; | |
317 | } else | |
fb6813f4 RR |
318 | p = alloc_page(gfp_mask); |
319 | return p; | |
320 | } | |
321 | ||
e4e8452a WB |
322 | static void virtqueue_napi_schedule(struct napi_struct *napi, |
323 | struct virtqueue *vq) | |
324 | { | |
325 | if (napi_schedule_prep(napi)) { | |
326 | virtqueue_disable_cb(vq); | |
327 | __napi_schedule(napi); | |
328 | } | |
329 | } | |
330 | ||
331 | static void virtqueue_napi_complete(struct napi_struct *napi, | |
332 | struct virtqueue *vq, int processed) | |
333 | { | |
334 | int opaque; | |
335 | ||
336 | opaque = virtqueue_enable_cb_prepare(vq); | |
fdaa767a TM |
337 | if (napi_complete_done(napi, processed)) { |
338 | if (unlikely(virtqueue_poll(vq, opaque))) | |
339 | virtqueue_napi_schedule(napi, vq); | |
340 | } else { | |
341 | virtqueue_disable_cb(vq); | |
342 | } | |
e4e8452a WB |
343 | } |
344 | ||
e9d7417b | 345 | static void skb_xmit_done(struct virtqueue *vq) |
296f96fc | 346 | { |
e9d7417b | 347 | struct virtnet_info *vi = vq->vdev->priv; |
b92f1e67 | 348 | struct napi_struct *napi = &vi->sq[vq2txq(vq)].napi; |
296f96fc | 349 | |
2cb9c6ba | 350 | /* Suppress further interrupts. */ |
e9d7417b | 351 | virtqueue_disable_cb(vq); |
11a3a154 | 352 | |
b92f1e67 WB |
353 | if (napi->weight) |
354 | virtqueue_napi_schedule(napi, vq); | |
355 | else | |
356 | /* We were probably waiting for more output buffers. */ | |
357 | netif_wake_subqueue(vi->dev, vq2txq(vq)); | |
296f96fc RR |
358 | } |
359 | ||
28b39bc7 JW |
360 | #define MRG_CTX_HEADER_SHIFT 22 |
361 | static void *mergeable_len_to_ctx(unsigned int truesize, | |
362 | unsigned int headroom) | |
363 | { | |
364 | return (void *)(unsigned long)((headroom << MRG_CTX_HEADER_SHIFT) | truesize); | |
365 | } | |
366 | ||
367 | static unsigned int mergeable_ctx_to_headroom(void *mrg_ctx) | |
368 | { | |
369 | return (unsigned long)mrg_ctx >> MRG_CTX_HEADER_SHIFT; | |
370 | } | |
371 | ||
372 | static unsigned int mergeable_ctx_to_truesize(void *mrg_ctx) | |
373 | { | |
374 | return (unsigned long)mrg_ctx & ((1 << MRG_CTX_HEADER_SHIFT) - 1); | |
375 | } | |
376 | ||
3464645a | 377 | /* Called from bottom half context */ |
946fa564 MT |
378 | static struct sk_buff *page_to_skb(struct virtnet_info *vi, |
379 | struct receive_queue *rq, | |
2613af0e | 380 | struct page *page, unsigned int offset, |
436c9453 | 381 | unsigned int len, unsigned int truesize, |
fb32856b | 382 | bool hdr_valid, unsigned int metasize, |
6c66c147 | 383 | bool whole_page) |
9ab86bbc SM |
384 | { |
385 | struct sk_buff *skb; | |
012873d0 | 386 | struct virtio_net_hdr_mrg_rxbuf *hdr; |
2613af0e | 387 | unsigned int copy, hdr_len, hdr_padded_len; |
af39c8f7 | 388 | struct page *page_to_free = NULL; |
fb32856b | 389 | int tailroom, shinfo_size; |
f80bd740 | 390 | char *p, *hdr_p, *buf; |
fb6813f4 | 391 | |
2613af0e | 392 | p = page_address(page) + offset; |
fb32856b | 393 | hdr_p = p; |
3f2c31d9 | 394 | |
012873d0 MT |
395 | hdr_len = vi->hdr_len; |
396 | if (vi->mergeable_rx_bufs) | |
a4a76503 | 397 | hdr_padded_len = sizeof(*hdr); |
012873d0 | 398 | else |
2613af0e | 399 | hdr_padded_len = sizeof(struct padded_vnet_hdr); |
3f2c31d9 | 400 | |
6c66c147 | 401 | /* If whole_page, there is an offset between the beginning of the |
fb32856b XZ |
402 | * data and the allocated space, otherwise the data and the allocated |
403 | * space are aligned. | |
8fb7da9e XZ |
404 | * |
405 | * Buffers with headroom use PAGE_SIZE as alloc size, see | |
406 | * add_recvbuf_mergeable() + get_mergeable_buf_len() | |
fb32856b | 407 | */ |
6c66c147 XZ |
408 | if (whole_page) { |
409 | /* Buffers with whole_page use PAGE_SIZE as alloc size, | |
f80bd740 XZ |
410 | * see add_recvbuf_mergeable() + get_mergeable_buf_len() |
411 | */ | |
fb32856b | 412 | truesize = PAGE_SIZE; |
7bf64460 XZ |
413 | |
414 | /* page maybe head page, so we should get the buf by p, not the | |
415 | * page | |
416 | */ | |
417 | tailroom = truesize - len - offset_in_page(p); | |
418 | buf = (char *)((unsigned long)p & PAGE_MASK); | |
fb32856b XZ |
419 | } else { |
420 | tailroom = truesize - len; | |
f80bd740 | 421 | buf = p; |
fb32856b | 422 | } |
3f2c31d9 | 423 | |
9ab86bbc | 424 | len -= hdr_len; |
2613af0e MD |
425 | offset += hdr_padded_len; |
426 | p += hdr_padded_len; | |
3f2c31d9 | 427 | |
fb32856b XZ |
428 | shinfo_size = SKB_DATA_ALIGN(sizeof(struct skb_shared_info)); |
429 | ||
f80bd740 | 430 | /* copy small packet so we can reuse these pages */ |
f5d7872a | 431 | if (!NET_IP_ALIGN && len > GOOD_COPY_LEN && tailroom >= shinfo_size) { |
f80bd740 | 432 | skb = build_skb(buf, truesize); |
fb32856b XZ |
433 | if (unlikely(!skb)) |
434 | return NULL; | |
435 | ||
f80bd740 | 436 | skb_reserve(skb, p - buf); |
fb32856b XZ |
437 | skb_put(skb, len); |
438 | goto ok; | |
439 | } | |
440 | ||
441 | /* copy small packet so we can reuse these pages for small data */ | |
442 | skb = napi_alloc_skb(&rq->napi, GOOD_COPY_LEN); | |
443 | if (unlikely(!skb)) | |
444 | return NULL; | |
445 | ||
0f6925b3 ED |
446 | /* Copy all frame if it fits skb->head, otherwise |
447 | * we let virtio_net_hdr_to_skb() and GRO pull headers as needed. | |
448 | */ | |
449 | if (len <= skb_tailroom(skb)) | |
450 | copy = len; | |
451 | else | |
452 | copy = ETH_HLEN + metasize; | |
59ae1d12 | 453 | skb_put_data(skb, p, copy); |
3f2c31d9 | 454 | |
9ab86bbc SM |
455 | len -= copy; |
456 | offset += copy; | |
3f2c31d9 | 457 | |
2613af0e MD |
458 | if (vi->mergeable_rx_bufs) { |
459 | if (len) | |
460 | skb_add_rx_frag(skb, 0, page, offset, len, truesize); | |
461 | else | |
af39c8f7 | 462 | page_to_free = page; |
fb32856b | 463 | goto ok; |
2613af0e MD |
464 | } |
465 | ||
e878d78b SL |
466 | /* |
467 | * Verify that we can indeed put this data into a skb. | |
468 | * This is here to handle cases when the device erroneously | |
469 | * tries to receive more than is possible. This is usually | |
470 | * the case of a broken device. | |
471 | */ | |
472 | if (unlikely(len > MAX_SKB_FRAGS * PAGE_SIZE)) { | |
be443899 | 473 | net_dbg_ratelimited("%s: too much data\n", skb->dev->name); |
e878d78b SL |
474 | dev_kfree_skb(skb); |
475 | return NULL; | |
476 | } | |
2613af0e | 477 | BUG_ON(offset >= PAGE_SIZE); |
9ab86bbc | 478 | while (len) { |
2613af0e MD |
479 | unsigned int frag_size = min((unsigned)PAGE_SIZE - offset, len); |
480 | skb_add_rx_frag(skb, skb_shinfo(skb)->nr_frags, page, offset, | |
481 | frag_size, truesize); | |
482 | len -= frag_size; | |
9ab86bbc SM |
483 | page = (struct page *)page->private; |
484 | offset = 0; | |
485 | } | |
3f2c31d9 | 486 | |
9ab86bbc | 487 | if (page) |
e9d7417b | 488 | give_pages(rq, page); |
3f2c31d9 | 489 | |
fb32856b XZ |
490 | ok: |
491 | /* hdr_valid means no XDP, so we can copy the vnet header */ | |
492 | if (hdr_valid) { | |
493 | hdr = skb_vnet_hdr(skb); | |
494 | memcpy(hdr, hdr_p, hdr_len); | |
495 | } | |
af39c8f7 ED |
496 | if (page_to_free) |
497 | put_page(page_to_free); | |
fb32856b XZ |
498 | |
499 | if (metasize) { | |
500 | __skb_pull(skb, metasize); | |
501 | skb_metadata_set(skb, metasize); | |
502 | } | |
503 | ||
9ab86bbc SM |
504 | return skb; |
505 | } | |
3f2c31d9 | 506 | |
735fc405 JDB |
507 | static int __virtnet_xdp_xmit_one(struct virtnet_info *vi, |
508 | struct send_queue *sq, | |
509 | struct xdp_frame *xdpf) | |
56434a01 | 510 | { |
56434a01 | 511 | struct virtio_net_hdr_mrg_rxbuf *hdr; |
56434a01 JF |
512 | int err; |
513 | ||
cac320c8 JDB |
514 | if (unlikely(xdpf->headroom < vi->hdr_len)) |
515 | return -EOVERFLOW; | |
516 | ||
517 | /* Make room for virtqueue hdr (also change xdpf->headroom?) */ | |
518 | xdpf->data -= vi->hdr_len; | |
f6b10209 | 519 | /* Zero header and leave csum up to XDP layers */ |
cac320c8 | 520 | hdr = xdpf->data; |
f6b10209 | 521 | memset(hdr, 0, vi->hdr_len); |
cac320c8 | 522 | xdpf->len += vi->hdr_len; |
bb91accf | 523 | |
cac320c8 | 524 | sg_init_one(sq->sg, xdpf->data, xdpf->len); |
bb91accf | 525 | |
5050471d TM |
526 | err = virtqueue_add_outbuf(sq->vq, sq->sg, 1, xdp_to_ptr(xdpf), |
527 | GFP_ATOMIC); | |
11b7d897 | 528 | if (unlikely(err)) |
cac320c8 | 529 | return -ENOSPC; /* Caller handle free/refcnt */ |
56434a01 | 530 | |
cac320c8 | 531 | return 0; |
56434a01 JF |
532 | } |
533 | ||
97c2c69e XZ |
534 | /* when vi->curr_queue_pairs > nr_cpu_ids, the txq/sq is only used for xdp tx on |
535 | * the current cpu, so it does not need to be locked. | |
536 | * | |
537 | * Here we use marco instead of inline functions because we have to deal with | |
538 | * three issues at the same time: 1. the choice of sq. 2. judge and execute the | |
539 | * lock/unlock of txq 3. make sparse happy. It is difficult for two inline | |
540 | * functions to perfectly solve these three problems at the same time. | |
541 | */ | |
542 | #define virtnet_xdp_get_sq(vi) ({ \ | |
543 | struct netdev_queue *txq; \ | |
544 | typeof(vi) v = (vi); \ | |
545 | unsigned int qp; \ | |
546 | \ | |
547 | if (v->curr_queue_pairs > nr_cpu_ids) { \ | |
548 | qp = v->curr_queue_pairs - v->xdp_queue_pairs; \ | |
549 | qp += smp_processor_id(); \ | |
550 | txq = netdev_get_tx_queue(v->dev, qp); \ | |
551 | __netif_tx_acquire(txq); \ | |
552 | } else { \ | |
553 | qp = smp_processor_id() % v->curr_queue_pairs; \ | |
554 | txq = netdev_get_tx_queue(v->dev, qp); \ | |
555 | __netif_tx_lock(txq, raw_smp_processor_id()); \ | |
556 | } \ | |
557 | v->sq + qp; \ | |
558 | }) | |
559 | ||
560 | #define virtnet_xdp_put_sq(vi, q) { \ | |
561 | struct netdev_queue *txq; \ | |
562 | typeof(vi) v = (vi); \ | |
563 | \ | |
564 | txq = netdev_get_tx_queue(v->dev, (q) - v->sq); \ | |
565 | if (v->curr_queue_pairs > nr_cpu_ids) \ | |
566 | __netif_tx_release(txq); \ | |
567 | else \ | |
568 | __netif_tx_unlock(txq); \ | |
2a43565c TM |
569 | } |
570 | ||
735fc405 | 571 | static int virtnet_xdp_xmit(struct net_device *dev, |
42b33468 | 572 | int n, struct xdp_frame **frames, u32 flags) |
186b3c99 JW |
573 | { |
574 | struct virtnet_info *vi = netdev_priv(dev); | |
8dcc5b0a JDB |
575 | struct receive_queue *rq = vi->rq; |
576 | struct bpf_prog *xdp_prog; | |
735fc405 JDB |
577 | struct send_queue *sq; |
578 | unsigned int len; | |
546f2897 TM |
579 | int packets = 0; |
580 | int bytes = 0; | |
fdc13979 | 581 | int nxmit = 0; |
461f03dc | 582 | int kicks = 0; |
5050471d | 583 | void *ptr; |
fdc13979 | 584 | int ret; |
735fc405 JDB |
585 | int i; |
586 | ||
8dcc5b0a JDB |
587 | /* Only allow ndo_xdp_xmit if XDP is loaded on dev, as this |
588 | * indicate XDP resources have been successfully allocated. | |
589 | */ | |
9719c6b9 | 590 | xdp_prog = rcu_access_pointer(rq->xdp_prog); |
1667c08a TM |
591 | if (!xdp_prog) |
592 | return -ENXIO; | |
593 | ||
97c2c69e | 594 | sq = virtnet_xdp_get_sq(vi); |
1667c08a TM |
595 | |
596 | if (unlikely(flags & ~XDP_XMIT_FLAGS_MASK)) { | |
597 | ret = -EINVAL; | |
5b8f3c8d TM |
598 | goto out; |
599 | } | |
8dcc5b0a | 600 | |
735fc405 | 601 | /* Free up any pending old buffers before queueing new ones. */ |
5050471d | 602 | while ((ptr = virtqueue_get_buf(sq->vq, &len)) != NULL) { |
546f2897 TM |
603 | if (likely(is_xdp_frame(ptr))) { |
604 | struct xdp_frame *frame = ptr_to_xdp(ptr); | |
605 | ||
606 | bytes += frame->len; | |
607 | xdp_return_frame(frame); | |
608 | } else { | |
609 | struct sk_buff *skb = ptr; | |
610 | ||
611 | bytes += skb->len; | |
612 | napi_consume_skb(skb, false); | |
613 | } | |
614 | packets++; | |
5050471d | 615 | } |
735fc405 JDB |
616 | |
617 | for (i = 0; i < n; i++) { | |
618 | struct xdp_frame *xdpf = frames[i]; | |
619 | ||
fdc13979 LB |
620 | if (__virtnet_xdp_xmit_one(vi, sq, xdpf)) |
621 | break; | |
622 | nxmit++; | |
735fc405 | 623 | } |
fdc13979 | 624 | ret = nxmit; |
5d274cb4 | 625 | |
461f03dc TM |
626 | if (flags & XDP_XMIT_FLUSH) { |
627 | if (virtqueue_kick_prepare(sq->vq) && virtqueue_notify(sq->vq)) | |
628 | kicks = 1; | |
629 | } | |
5b8f3c8d TM |
630 | out: |
631 | u64_stats_update_begin(&sq->stats.syncp); | |
546f2897 TM |
632 | sq->stats.bytes += bytes; |
633 | sq->stats.packets += packets; | |
5b8f3c8d | 634 | sq->stats.xdp_tx += n; |
fdc13979 | 635 | sq->stats.xdp_tx_drops += n - nxmit; |
461f03dc | 636 | sq->stats.kicks += kicks; |
5b8f3c8d | 637 | u64_stats_update_end(&sq->stats.syncp); |
5d274cb4 | 638 | |
97c2c69e | 639 | virtnet_xdp_put_sq(vi, sq); |
5b8f3c8d | 640 | return ret; |
186b3c99 JW |
641 | } |
642 | ||
f6b10209 JW |
643 | static unsigned int virtnet_get_headroom(struct virtnet_info *vi) |
644 | { | |
97c2c69e | 645 | return vi->xdp_enabled ? VIRTIO_XDP_HEADROOM : 0; |
f6b10209 JW |
646 | } |
647 | ||
4941d472 JW |
648 | /* We copy the packet for XDP in the following cases: |
649 | * | |
650 | * 1) Packet is scattered across multiple rx buffers. | |
651 | * 2) Headroom space is insufficient. | |
652 | * | |
653 | * This is inefficient but it's a temporary condition that | |
654 | * we hit right after XDP is enabled and until queue is refilled | |
655 | * with large buffers with sufficient headroom - so it should affect | |
656 | * at most queue size packets. | |
657 | * Afterwards, the conditions to enable | |
658 | * XDP should preclude the underlying device from sending packets | |
659 | * across multiple buffers (num_buf > 1), and we make sure buffers | |
660 | * have enough headroom. | |
661 | */ | |
662 | static struct page *xdp_linearize_page(struct receive_queue *rq, | |
663 | u16 *num_buf, | |
664 | struct page *p, | |
665 | int offset, | |
666 | int page_off, | |
667 | unsigned int *len) | |
668 | { | |
669 | struct page *page = alloc_page(GFP_ATOMIC); | |
670 | ||
671 | if (!page) | |
672 | return NULL; | |
673 | ||
674 | memcpy(page_address(page) + page_off, page_address(p) + offset, *len); | |
675 | page_off += *len; | |
676 | ||
677 | while (--*num_buf) { | |
3cc81a9a | 678 | int tailroom = SKB_DATA_ALIGN(sizeof(struct skb_shared_info)); |
4941d472 JW |
679 | unsigned int buflen; |
680 | void *buf; | |
681 | int off; | |
682 | ||
683 | buf = virtqueue_get_buf(rq->vq, &buflen); | |
684 | if (unlikely(!buf)) | |
685 | goto err_buf; | |
686 | ||
687 | p = virt_to_head_page(buf); | |
688 | off = buf - page_address(p); | |
689 | ||
690 | /* guard against a misconfigured or uncooperative backend that | |
691 | * is sending packet larger than the MTU. | |
692 | */ | |
3cc81a9a | 693 | if ((page_off + buflen + tailroom) > PAGE_SIZE) { |
4941d472 JW |
694 | put_page(p); |
695 | goto err_buf; | |
696 | } | |
697 | ||
698 | memcpy(page_address(page) + page_off, | |
699 | page_address(p) + off, buflen); | |
700 | page_off += buflen; | |
701 | put_page(p); | |
702 | } | |
703 | ||
704 | /* Headroom does not contribute to packet length */ | |
705 | *len = page_off - VIRTIO_XDP_HEADROOM; | |
706 | return page; | |
707 | err_buf: | |
708 | __free_pages(page, 0); | |
709 | return NULL; | |
710 | } | |
711 | ||
bb91accf JW |
712 | static struct sk_buff *receive_small(struct net_device *dev, |
713 | struct virtnet_info *vi, | |
714 | struct receive_queue *rq, | |
192f68cf | 715 | void *buf, void *ctx, |
186b3c99 | 716 | unsigned int len, |
7d9d60fd | 717 | unsigned int *xdp_xmit, |
d46eeeaf | 718 | struct virtnet_rq_stats *stats) |
f121159d | 719 | { |
f6b10209 | 720 | struct sk_buff *skb; |
bb91accf | 721 | struct bpf_prog *xdp_prog; |
4941d472 | 722 | unsigned int xdp_headroom = (unsigned long)ctx; |
f6b10209 JW |
723 | unsigned int header_offset = VIRTNET_RX_PAD + xdp_headroom; |
724 | unsigned int headroom = vi->hdr_len + header_offset; | |
725 | unsigned int buflen = SKB_DATA_ALIGN(GOOD_PACKET_LEN + headroom) + | |
726 | SKB_DATA_ALIGN(sizeof(struct skb_shared_info)); | |
4941d472 | 727 | struct page *page = virt_to_head_page(buf); |
11b7d897 | 728 | unsigned int delta = 0; |
4941d472 | 729 | struct page *xdp_page; |
11b7d897 | 730 | int err; |
503d539a | 731 | unsigned int metasize = 0; |
11b7d897 | 732 | |
012873d0 | 733 | len -= vi->hdr_len; |
d46eeeaf | 734 | stats->bytes += len; |
f121159d | 735 | |
ad993a95 XY |
736 | if (unlikely(len > GOOD_PACKET_LEN)) { |
737 | pr_debug("%s: rx error: len %u exceeds max size %d\n", | |
738 | dev->name, len, GOOD_PACKET_LEN); | |
739 | dev->stats.rx_length_errors++; | |
740 | goto err_len; | |
741 | } | |
bb91accf JW |
742 | rcu_read_lock(); |
743 | xdp_prog = rcu_dereference(rq->xdp_prog); | |
744 | if (xdp_prog) { | |
f6b10209 | 745 | struct virtio_net_hdr_mrg_rxbuf *hdr = buf + header_offset; |
44fa2dbd | 746 | struct xdp_frame *xdpf; |
0354e4d1 | 747 | struct xdp_buff xdp; |
f6b10209 | 748 | void *orig_data; |
bb91accf JW |
749 | u32 act; |
750 | ||
95dbe9e7 | 751 | if (unlikely(hdr->hdr.gso_type)) |
bb91accf | 752 | goto err_xdp; |
0354e4d1 | 753 | |
4941d472 JW |
754 | if (unlikely(xdp_headroom < virtnet_get_headroom(vi))) { |
755 | int offset = buf - page_address(page) + header_offset; | |
756 | unsigned int tlen = len + vi->hdr_len; | |
757 | u16 num_buf = 1; | |
758 | ||
759 | xdp_headroom = virtnet_get_headroom(vi); | |
760 | header_offset = VIRTNET_RX_PAD + xdp_headroom; | |
761 | headroom = vi->hdr_len + header_offset; | |
762 | buflen = SKB_DATA_ALIGN(GOOD_PACKET_LEN + headroom) + | |
763 | SKB_DATA_ALIGN(sizeof(struct skb_shared_info)); | |
764 | xdp_page = xdp_linearize_page(rq, &num_buf, page, | |
765 | offset, header_offset, | |
766 | &tlen); | |
767 | if (!xdp_page) | |
768 | goto err_xdp; | |
769 | ||
770 | buf = page_address(xdp_page); | |
771 | put_page(page); | |
772 | page = xdp_page; | |
773 | } | |
774 | ||
43b5169d | 775 | xdp_init_buff(&xdp, buflen, &rq->xdp_rxq); |
be9df4af LB |
776 | xdp_prepare_buff(&xdp, buf + VIRTNET_RX_PAD + vi->hdr_len, |
777 | xdp_headroom, len, true); | |
f6b10209 | 778 | orig_data = xdp.data; |
0354e4d1 | 779 | act = bpf_prog_run_xdp(xdp_prog, &xdp); |
d46eeeaf | 780 | stats->xdp_packets++; |
0354e4d1 | 781 | |
bb91accf JW |
782 | switch (act) { |
783 | case XDP_PASS: | |
2de2f7f4 | 784 | /* Recalculate length in case bpf program changed it */ |
f6b10209 | 785 | delta = orig_data - xdp.data; |
6870de43 | 786 | len = xdp.data_end - xdp.data; |
503d539a | 787 | metasize = xdp.data - xdp.data_meta; |
bb91accf JW |
788 | break; |
789 | case XDP_TX: | |
d46eeeaf | 790 | stats->xdp_tx++; |
1b698fa5 | 791 | xdpf = xdp_convert_buff_to_frame(&xdp); |
44fa2dbd JDB |
792 | if (unlikely(!xdpf)) |
793 | goto err_xdp; | |
ca9e83b4 | 794 | err = virtnet_xdp_xmit(dev, 1, &xdpf, 0); |
fdc13979 LB |
795 | if (unlikely(!err)) { |
796 | xdp_return_frame_rx_napi(xdpf); | |
797 | } else if (unlikely(err < 0)) { | |
0354e4d1 | 798 | trace_xdp_exception(vi->dev, xdp_prog, act); |
11b7d897 JDB |
799 | goto err_xdp; |
800 | } | |
2471c75e | 801 | *xdp_xmit |= VIRTIO_XDP_TX; |
186b3c99 JW |
802 | rcu_read_unlock(); |
803 | goto xdp_xmit; | |
804 | case XDP_REDIRECT: | |
d46eeeaf | 805 | stats->xdp_redirects++; |
186b3c99 | 806 | err = xdp_do_redirect(dev, &xdp, xdp_prog); |
11b7d897 JDB |
807 | if (err) |
808 | goto err_xdp; | |
2471c75e | 809 | *xdp_xmit |= VIRTIO_XDP_REDIR; |
bb91accf JW |
810 | rcu_read_unlock(); |
811 | goto xdp_xmit; | |
bb91accf | 812 | default: |
0354e4d1 | 813 | bpf_warn_invalid_xdp_action(act); |
df561f66 | 814 | fallthrough; |
0354e4d1 JF |
815 | case XDP_ABORTED: |
816 | trace_xdp_exception(vi->dev, xdp_prog, act); | |
95efabf0 | 817 | goto err_xdp; |
0354e4d1 | 818 | case XDP_DROP: |
bb91accf JW |
819 | goto err_xdp; |
820 | } | |
821 | } | |
822 | rcu_read_unlock(); | |
823 | ||
f6b10209 JW |
824 | skb = build_skb(buf, buflen); |
825 | if (!skb) { | |
4941d472 | 826 | put_page(page); |
f6b10209 JW |
827 | goto err; |
828 | } | |
829 | skb_reserve(skb, headroom - delta); | |
6870de43 | 830 | skb_put(skb, len); |
f1d4884d | 831 | if (!xdp_prog) { |
f6b10209 JW |
832 | buf += header_offset; |
833 | memcpy(skb_vnet_hdr(skb), buf, vi->hdr_len); | |
f1d4884d | 834 | } /* keep zeroed vnet hdr since XDP is loaded */ |
f6b10209 | 835 | |
503d539a YK |
836 | if (metasize) |
837 | skb_metadata_set(skb, metasize); | |
838 | ||
f6b10209 | 839 | err: |
f121159d | 840 | return skb; |
bb91accf JW |
841 | |
842 | err_xdp: | |
843 | rcu_read_unlock(); | |
d46eeeaf | 844 | stats->xdp_drops++; |
ad993a95 | 845 | err_len: |
d46eeeaf | 846 | stats->drops++; |
4941d472 | 847 | put_page(page); |
bb91accf JW |
848 | xdp_xmit: |
849 | return NULL; | |
f121159d MT |
850 | } |
851 | ||
852 | static struct sk_buff *receive_big(struct net_device *dev, | |
946fa564 | 853 | struct virtnet_info *vi, |
f121159d MT |
854 | struct receive_queue *rq, |
855 | void *buf, | |
7d9d60fd | 856 | unsigned int len, |
d46eeeaf | 857 | struct virtnet_rq_stats *stats) |
f121159d MT |
858 | { |
859 | struct page *page = buf; | |
503d539a | 860 | struct sk_buff *skb = |
fb32856b | 861 | page_to_skb(vi, rq, page, 0, len, PAGE_SIZE, true, 0, 0); |
f600b690 | 862 | |
d46eeeaf | 863 | stats->bytes += len - vi->hdr_len; |
f121159d MT |
864 | if (unlikely(!skb)) |
865 | goto err; | |
866 | ||
867 | return skb; | |
868 | ||
869 | err: | |
d46eeeaf | 870 | stats->drops++; |
f121159d MT |
871 | give_pages(rq, page); |
872 | return NULL; | |
873 | } | |
874 | ||
8fc3b9e9 | 875 | static struct sk_buff *receive_mergeable(struct net_device *dev, |
fdd819b2 | 876 | struct virtnet_info *vi, |
8fc3b9e9 | 877 | struct receive_queue *rq, |
680557cf MT |
878 | void *buf, |
879 | void *ctx, | |
186b3c99 | 880 | unsigned int len, |
7d9d60fd | 881 | unsigned int *xdp_xmit, |
d46eeeaf | 882 | struct virtnet_rq_stats *stats) |
9ab86bbc | 883 | { |
012873d0 MT |
884 | struct virtio_net_hdr_mrg_rxbuf *hdr = buf; |
885 | u16 num_buf = virtio16_to_cpu(vi->vdev, hdr->num_buffers); | |
8fc3b9e9 MT |
886 | struct page *page = virt_to_head_page(buf); |
887 | int offset = buf - page_address(page); | |
f600b690 JF |
888 | struct sk_buff *head_skb, *curr_skb; |
889 | struct bpf_prog *xdp_prog; | |
9ce6146e | 890 | unsigned int truesize = mergeable_ctx_to_truesize(ctx); |
4941d472 | 891 | unsigned int headroom = mergeable_ctx_to_headroom(ctx); |
503d539a | 892 | unsigned int metasize = 0; |
9ce6146e JDB |
893 | unsigned int frame_sz; |
894 | int err; | |
f600b690 | 895 | |
56434a01 | 896 | head_skb = NULL; |
d46eeeaf | 897 | stats->bytes += len - vi->hdr_len; |
56434a01 | 898 | |
ad993a95 XY |
899 | if (unlikely(len > truesize)) { |
900 | pr_debug("%s: rx error: len %u exceeds truesize %lu\n", | |
901 | dev->name, len, (unsigned long)ctx); | |
902 | dev->stats.rx_length_errors++; | |
903 | goto err_skb; | |
904 | } | |
f600b690 JF |
905 | rcu_read_lock(); |
906 | xdp_prog = rcu_dereference(rq->xdp_prog); | |
907 | if (xdp_prog) { | |
44fa2dbd | 908 | struct xdp_frame *xdpf; |
72979a6c | 909 | struct page *xdp_page; |
0354e4d1 | 910 | struct xdp_buff xdp; |
0354e4d1 | 911 | void *data; |
f600b690 JF |
912 | u32 act; |
913 | ||
3d62b2a0 JW |
914 | /* Transient failure which in theory could occur if |
915 | * in-flight packets from before XDP was enabled reach | |
916 | * the receive path after XDP is loaded. | |
917 | */ | |
918 | if (unlikely(hdr->hdr.gso_type)) | |
919 | goto err_xdp; | |
920 | ||
9ce6146e JDB |
921 | /* Buffers with headroom use PAGE_SIZE as alloc size, |
922 | * see add_recvbuf_mergeable() + get_mergeable_buf_len() | |
923 | */ | |
924 | frame_sz = headroom ? PAGE_SIZE : truesize; | |
925 | ||
3cc81a9a JW |
926 | /* This happens when rx buffer size is underestimated |
927 | * or headroom is not enough because of the buffer | |
928 | * was refilled before XDP is set. This should only | |
929 | * happen for the first several packets, so we don't | |
930 | * care much about its performance. | |
931 | */ | |
4941d472 JW |
932 | if (unlikely(num_buf > 1 || |
933 | headroom < virtnet_get_headroom(vi))) { | |
72979a6c | 934 | /* linearize data for XDP */ |
56a86f84 | 935 | xdp_page = xdp_linearize_page(rq, &num_buf, |
4941d472 JW |
936 | page, offset, |
937 | VIRTIO_XDP_HEADROOM, | |
938 | &len); | |
9ce6146e JDB |
939 | frame_sz = PAGE_SIZE; |
940 | ||
72979a6c JF |
941 | if (!xdp_page) |
942 | goto err_xdp; | |
2de2f7f4 | 943 | offset = VIRTIO_XDP_HEADROOM; |
72979a6c JF |
944 | } else { |
945 | xdp_page = page; | |
f600b690 JF |
946 | } |
947 | ||
2de2f7f4 JF |
948 | /* Allow consuming headroom but reserve enough space to push |
949 | * the descriptor on if we get an XDP_TX return code. | |
950 | */ | |
0354e4d1 | 951 | data = page_address(xdp_page) + offset; |
43b5169d | 952 | xdp_init_buff(&xdp, frame_sz - vi->hdr_len, &rq->xdp_rxq); |
be9df4af LB |
953 | xdp_prepare_buff(&xdp, data - VIRTIO_XDP_HEADROOM + vi->hdr_len, |
954 | VIRTIO_XDP_HEADROOM, len - vi->hdr_len, true); | |
754b8a21 | 955 | |
0354e4d1 | 956 | act = bpf_prog_run_xdp(xdp_prog, &xdp); |
d46eeeaf | 957 | stats->xdp_packets++; |
0354e4d1 | 958 | |
56434a01 JF |
959 | switch (act) { |
960 | case XDP_PASS: | |
503d539a YK |
961 | metasize = xdp.data - xdp.data_meta; |
962 | ||
2de2f7f4 | 963 | /* recalculate offset to account for any header |
503d539a YK |
964 | * adjustments and minus the metasize to copy the |
965 | * metadata in page_to_skb(). Note other cases do not | |
966 | * build an skb and avoid using offset | |
2de2f7f4 | 967 | */ |
503d539a YK |
968 | offset = xdp.data - page_address(xdp_page) - |
969 | vi->hdr_len - metasize; | |
2de2f7f4 | 970 | |
503d539a YK |
971 | /* recalculate len if xdp.data, xdp.data_end or |
972 | * xdp.data_meta were adjusted | |
6870de43 | 973 | */ |
503d539a | 974 | len = xdp.data_end - xdp.data + vi->hdr_len + metasize; |
1830f893 JW |
975 | /* We can only create skb based on xdp_page. */ |
976 | if (unlikely(xdp_page != page)) { | |
977 | rcu_read_unlock(); | |
978 | put_page(page); | |
503d539a YK |
979 | head_skb = page_to_skb(vi, rq, xdp_page, offset, |
980 | len, PAGE_SIZE, false, | |
6c66c147 | 981 | metasize, true); |
1830f893 JW |
982 | return head_skb; |
983 | } | |
56434a01 JF |
984 | break; |
985 | case XDP_TX: | |
d46eeeaf | 986 | stats->xdp_tx++; |
1b698fa5 | 987 | xdpf = xdp_convert_buff_to_frame(&xdp); |
44fa2dbd JDB |
988 | if (unlikely(!xdpf)) |
989 | goto err_xdp; | |
ca9e83b4 | 990 | err = virtnet_xdp_xmit(dev, 1, &xdpf, 0); |
fdc13979 LB |
991 | if (unlikely(!err)) { |
992 | xdp_return_frame_rx_napi(xdpf); | |
993 | } else if (unlikely(err < 0)) { | |
0354e4d1 | 994 | trace_xdp_exception(vi->dev, xdp_prog, act); |
11b7d897 JDB |
995 | if (unlikely(xdp_page != page)) |
996 | put_page(xdp_page); | |
997 | goto err_xdp; | |
998 | } | |
2471c75e | 999 | *xdp_xmit |= VIRTIO_XDP_TX; |
72979a6c | 1000 | if (unlikely(xdp_page != page)) |
5d458a13 | 1001 | put_page(page); |
56434a01 JF |
1002 | rcu_read_unlock(); |
1003 | goto xdp_xmit; | |
3cc81a9a | 1004 | case XDP_REDIRECT: |
d46eeeaf | 1005 | stats->xdp_redirects++; |
3cc81a9a JW |
1006 | err = xdp_do_redirect(dev, &xdp, xdp_prog); |
1007 | if (err) { | |
1008 | if (unlikely(xdp_page != page)) | |
1009 | put_page(xdp_page); | |
1010 | goto err_xdp; | |
1011 | } | |
2471c75e | 1012 | *xdp_xmit |= VIRTIO_XDP_REDIR; |
3cc81a9a | 1013 | if (unlikely(xdp_page != page)) |
6890418b | 1014 | put_page(page); |
3cc81a9a JW |
1015 | rcu_read_unlock(); |
1016 | goto xdp_xmit; | |
56434a01 | 1017 | default: |
0354e4d1 | 1018 | bpf_warn_invalid_xdp_action(act); |
df561f66 | 1019 | fallthrough; |
0354e4d1 JF |
1020 | case XDP_ABORTED: |
1021 | trace_xdp_exception(vi->dev, xdp_prog, act); | |
df561f66 | 1022 | fallthrough; |
0354e4d1 | 1023 | case XDP_DROP: |
72979a6c JF |
1024 | if (unlikely(xdp_page != page)) |
1025 | __free_pages(xdp_page, 0); | |
f600b690 | 1026 | goto err_xdp; |
56434a01 | 1027 | } |
f600b690 JF |
1028 | } |
1029 | rcu_read_unlock(); | |
ab7db917 | 1030 | |
503d539a | 1031 | head_skb = page_to_skb(vi, rq, page, offset, len, truesize, !xdp_prog, |
6c66c147 | 1032 | metasize, !!headroom); |
f600b690 | 1033 | curr_skb = head_skb; |
9ab86bbc | 1034 | |
8fc3b9e9 MT |
1035 | if (unlikely(!curr_skb)) |
1036 | goto err_skb; | |
9ab86bbc | 1037 | while (--num_buf) { |
8fc3b9e9 MT |
1038 | int num_skb_frags; |
1039 | ||
680557cf | 1040 | buf = virtqueue_get_buf_ctx(rq->vq, &len, &ctx); |
03e9f8a0 | 1041 | if (unlikely(!buf)) { |
8fc3b9e9 | 1042 | pr_debug("%s: rx error: %d buffers out of %d missing\n", |
fdd819b2 | 1043 | dev->name, num_buf, |
012873d0 MT |
1044 | virtio16_to_cpu(vi->vdev, |
1045 | hdr->num_buffers)); | |
8fc3b9e9 MT |
1046 | dev->stats.rx_length_errors++; |
1047 | goto err_buf; | |
3f2c31d9 | 1048 | } |
8fc3b9e9 | 1049 | |
d46eeeaf | 1050 | stats->bytes += len; |
8fc3b9e9 | 1051 | page = virt_to_head_page(buf); |
28b39bc7 JW |
1052 | |
1053 | truesize = mergeable_ctx_to_truesize(ctx); | |
1054 | if (unlikely(len > truesize)) { | |
56da5fd0 | 1055 | pr_debug("%s: rx error: len %u exceeds truesize %lu\n", |
680557cf MT |
1056 | dev->name, len, (unsigned long)ctx); |
1057 | dev->stats.rx_length_errors++; | |
1058 | goto err_skb; | |
1059 | } | |
8fc3b9e9 MT |
1060 | |
1061 | num_skb_frags = skb_shinfo(curr_skb)->nr_frags; | |
2613af0e MD |
1062 | if (unlikely(num_skb_frags == MAX_SKB_FRAGS)) { |
1063 | struct sk_buff *nskb = alloc_skb(0, GFP_ATOMIC); | |
8fc3b9e9 MT |
1064 | |
1065 | if (unlikely(!nskb)) | |
1066 | goto err_skb; | |
2613af0e MD |
1067 | if (curr_skb == head_skb) |
1068 | skb_shinfo(curr_skb)->frag_list = nskb; | |
1069 | else | |
1070 | curr_skb->next = nskb; | |
1071 | curr_skb = nskb; | |
1072 | head_skb->truesize += nskb->truesize; | |
1073 | num_skb_frags = 0; | |
1074 | } | |
1075 | if (curr_skb != head_skb) { | |
1076 | head_skb->data_len += len; | |
1077 | head_skb->len += len; | |
fb51879d | 1078 | head_skb->truesize += truesize; |
2613af0e | 1079 | } |
8fc3b9e9 | 1080 | offset = buf - page_address(page); |
ba275241 JW |
1081 | if (skb_can_coalesce(curr_skb, num_skb_frags, page, offset)) { |
1082 | put_page(page); | |
1083 | skb_coalesce_rx_frag(curr_skb, num_skb_frags - 1, | |
fb51879d | 1084 | len, truesize); |
ba275241 JW |
1085 | } else { |
1086 | skb_add_rx_frag(curr_skb, num_skb_frags, page, | |
fb51879d | 1087 | offset, len, truesize); |
ba275241 | 1088 | } |
8fc3b9e9 MT |
1089 | } |
1090 | ||
5377d758 | 1091 | ewma_pkt_len_add(&rq->mrg_avg_pkt_len, head_skb->len); |
8fc3b9e9 MT |
1092 | return head_skb; |
1093 | ||
f600b690 JF |
1094 | err_xdp: |
1095 | rcu_read_unlock(); | |
d46eeeaf | 1096 | stats->xdp_drops++; |
8fc3b9e9 MT |
1097 | err_skb: |
1098 | put_page(page); | |
850e088d | 1099 | while (num_buf-- > 1) { |
680557cf MT |
1100 | buf = virtqueue_get_buf(rq->vq, &len); |
1101 | if (unlikely(!buf)) { | |
8fc3b9e9 MT |
1102 | pr_debug("%s: rx error: %d buffers missing\n", |
1103 | dev->name, num_buf); | |
1104 | dev->stats.rx_length_errors++; | |
1105 | break; | |
1106 | } | |
d46eeeaf | 1107 | stats->bytes += len; |
680557cf | 1108 | page = virt_to_head_page(buf); |
8fc3b9e9 | 1109 | put_page(page); |
9ab86bbc | 1110 | } |
8fc3b9e9 | 1111 | err_buf: |
d46eeeaf | 1112 | stats->drops++; |
8fc3b9e9 | 1113 | dev_kfree_skb(head_skb); |
56434a01 | 1114 | xdp_xmit: |
8fc3b9e9 | 1115 | return NULL; |
9ab86bbc SM |
1116 | } |
1117 | ||
7d9d60fd TM |
1118 | static void receive_buf(struct virtnet_info *vi, struct receive_queue *rq, |
1119 | void *buf, unsigned int len, void **ctx, | |
a0929a44 | 1120 | unsigned int *xdp_xmit, |
d46eeeaf | 1121 | struct virtnet_rq_stats *stats) |
9ab86bbc | 1122 | { |
e9d7417b | 1123 | struct net_device *dev = vi->dev; |
9ab86bbc | 1124 | struct sk_buff *skb; |
012873d0 | 1125 | struct virtio_net_hdr_mrg_rxbuf *hdr; |
3f2c31d9 | 1126 | |
bcff3162 | 1127 | if (unlikely(len < vi->hdr_len + ETH_HLEN)) { |
9ab86bbc SM |
1128 | pr_debug("%s: short packet %i\n", dev->name, len); |
1129 | dev->stats.rx_length_errors++; | |
ab7db917 | 1130 | if (vi->mergeable_rx_bufs) { |
680557cf | 1131 | put_page(virt_to_head_page(buf)); |
ab7db917 | 1132 | } else if (vi->big_packets) { |
98bfd23c | 1133 | give_pages(rq, buf); |
ab7db917 | 1134 | } else { |
f6b10209 | 1135 | put_page(virt_to_head_page(buf)); |
ab7db917 | 1136 | } |
7d9d60fd | 1137 | return; |
9ab86bbc | 1138 | } |
3f2c31d9 | 1139 | |
f121159d | 1140 | if (vi->mergeable_rx_bufs) |
7d9d60fd | 1141 | skb = receive_mergeable(dev, vi, rq, buf, ctx, len, xdp_xmit, |
a0929a44 | 1142 | stats); |
f121159d | 1143 | else if (vi->big_packets) |
a0929a44 | 1144 | skb = receive_big(dev, vi, rq, buf, len, stats); |
f121159d | 1145 | else |
a0929a44 | 1146 | skb = receive_small(dev, vi, rq, buf, ctx, len, xdp_xmit, stats); |
f121159d MT |
1147 | |
1148 | if (unlikely(!skb)) | |
7d9d60fd | 1149 | return; |
3f2c31d9 | 1150 | |
9ab86bbc | 1151 | hdr = skb_vnet_hdr(skb); |
3fa2a1df | 1152 | |
e858fae2 | 1153 | if (hdr->hdr.flags & VIRTIO_NET_HDR_F_DATA_VALID) |
10a8d94a | 1154 | skb->ip_summed = CHECKSUM_UNNECESSARY; |
296f96fc | 1155 | |
e858fae2 MR |
1156 | if (virtio_net_hdr_to_skb(skb, &hdr->hdr, |
1157 | virtio_is_little_endian(vi->vdev))) { | |
1158 | net_warn_ratelimited("%s: bad gso: type: %u, size: %u\n", | |
1159 | dev->name, hdr->hdr.gso_type, | |
1160 | hdr->hdr.gso_size); | |
1161 | goto frame_err; | |
296f96fc RR |
1162 | } |
1163 | ||
133bbb18 | 1164 | skb_record_rx_queue(skb, vq2rxq(rq->vq)); |
d1dc06dc MR |
1165 | skb->protocol = eth_type_trans(skb, dev); |
1166 | pr_debug("Receiving skb proto 0x%04x len %i type %i\n", | |
1167 | ntohs(skb->protocol), skb->len, skb->pkt_type); | |
1168 | ||
0fbd050a | 1169 | napi_gro_receive(&rq->napi, skb); |
7d9d60fd | 1170 | return; |
296f96fc RR |
1171 | |
1172 | frame_err: | |
1173 | dev->stats.rx_frame_errors++; | |
296f96fc RR |
1174 | dev_kfree_skb(skb); |
1175 | } | |
1176 | ||
192f68cf JW |
1177 | /* Unlike mergeable buffers, all buffers are allocated to the |
1178 | * same size, except for the headroom. For this reason we do | |
1179 | * not need to use mergeable_len_to_ctx here - it is enough | |
1180 | * to store the headroom as the context ignoring the truesize. | |
1181 | */ | |
946fa564 MT |
1182 | static int add_recvbuf_small(struct virtnet_info *vi, struct receive_queue *rq, |
1183 | gfp_t gfp) | |
296f96fc | 1184 | { |
f6b10209 JW |
1185 | struct page_frag *alloc_frag = &rq->alloc_frag; |
1186 | char *buf; | |
2de2f7f4 | 1187 | unsigned int xdp_headroom = virtnet_get_headroom(vi); |
192f68cf | 1188 | void *ctx = (void *)(unsigned long)xdp_headroom; |
f6b10209 | 1189 | int len = vi->hdr_len + VIRTNET_RX_PAD + GOOD_PACKET_LEN + xdp_headroom; |
9ab86bbc | 1190 | int err; |
3f2c31d9 | 1191 | |
f6b10209 JW |
1192 | len = SKB_DATA_ALIGN(len) + |
1193 | SKB_DATA_ALIGN(sizeof(struct skb_shared_info)); | |
1194 | if (unlikely(!skb_page_frag_refill(len, alloc_frag, gfp))) | |
9ab86bbc | 1195 | return -ENOMEM; |
296f96fc | 1196 | |
f6b10209 JW |
1197 | buf = (char *)page_address(alloc_frag->page) + alloc_frag->offset; |
1198 | get_page(alloc_frag->page); | |
1199 | alloc_frag->offset += len; | |
1200 | sg_init_one(rq->sg, buf + VIRTNET_RX_PAD + xdp_headroom, | |
1201 | vi->hdr_len + GOOD_PACKET_LEN); | |
192f68cf | 1202 | err = virtqueue_add_inbuf_ctx(rq->vq, rq->sg, 1, buf, ctx, gfp); |
9ab86bbc | 1203 | if (err < 0) |
f6b10209 | 1204 | put_page(virt_to_head_page(buf)); |
9ab86bbc SM |
1205 | return err; |
1206 | } | |
97402b96 | 1207 | |
012873d0 MT |
1208 | static int add_recvbuf_big(struct virtnet_info *vi, struct receive_queue *rq, |
1209 | gfp_t gfp) | |
9ab86bbc | 1210 | { |
9ab86bbc SM |
1211 | struct page *first, *list = NULL; |
1212 | char *p; | |
1213 | int i, err, offset; | |
1214 | ||
a5835440 RR |
1215 | sg_init_table(rq->sg, MAX_SKB_FRAGS + 2); |
1216 | ||
e9d7417b | 1217 | /* page in rq->sg[MAX_SKB_FRAGS + 1] is list tail */ |
9ab86bbc | 1218 | for (i = MAX_SKB_FRAGS + 1; i > 1; --i) { |
e9d7417b | 1219 | first = get_a_page(rq, gfp); |
9ab86bbc SM |
1220 | if (!first) { |
1221 | if (list) | |
e9d7417b | 1222 | give_pages(rq, list); |
9ab86bbc | 1223 | return -ENOMEM; |
97402b96 | 1224 | } |
e9d7417b | 1225 | sg_set_buf(&rq->sg[i], page_address(first), PAGE_SIZE); |
97402b96 | 1226 | |
9ab86bbc SM |
1227 | /* chain new page in list head to match sg */ |
1228 | first->private = (unsigned long)list; | |
1229 | list = first; | |
1230 | } | |
296f96fc | 1231 | |
e9d7417b | 1232 | first = get_a_page(rq, gfp); |
9ab86bbc | 1233 | if (!first) { |
e9d7417b | 1234 | give_pages(rq, list); |
9ab86bbc SM |
1235 | return -ENOMEM; |
1236 | } | |
1237 | p = page_address(first); | |
1238 | ||
e9d7417b | 1239 | /* rq->sg[0], rq->sg[1] share the same page */ |
012873d0 MT |
1240 | /* a separated rq->sg[0] for header - required in case !any_header_sg */ |
1241 | sg_set_buf(&rq->sg[0], p, vi->hdr_len); | |
9ab86bbc | 1242 | |
e9d7417b | 1243 | /* rq->sg[1] for data packet, from offset */ |
9ab86bbc | 1244 | offset = sizeof(struct padded_vnet_hdr); |
e9d7417b | 1245 | sg_set_buf(&rq->sg[1], p + offset, PAGE_SIZE - offset); |
9ab86bbc SM |
1246 | |
1247 | /* chain first in list head */ | |
1248 | first->private = (unsigned long)list; | |
9dc7b9e4 RR |
1249 | err = virtqueue_add_inbuf(rq->vq, rq->sg, MAX_SKB_FRAGS + 2, |
1250 | first, gfp); | |
9ab86bbc | 1251 | if (err < 0) |
e9d7417b | 1252 | give_pages(rq, first); |
9ab86bbc SM |
1253 | |
1254 | return err; | |
296f96fc RR |
1255 | } |
1256 | ||
d85b758f | 1257 | static unsigned int get_mergeable_buf_len(struct receive_queue *rq, |
3cc81a9a JW |
1258 | struct ewma_pkt_len *avg_pkt_len, |
1259 | unsigned int room) | |
3f2c31d9 | 1260 | { |
ab7db917 | 1261 | const size_t hdr_len = sizeof(struct virtio_net_hdr_mrg_rxbuf); |
fbf28d78 MD |
1262 | unsigned int len; |
1263 | ||
3cc81a9a JW |
1264 | if (room) |
1265 | return PAGE_SIZE - room; | |
1266 | ||
1267 | len = hdr_len + clamp_t(unsigned int, ewma_pkt_len_read(avg_pkt_len), | |
f0c3192c | 1268 | rq->min_buf_len, PAGE_SIZE - hdr_len); |
3cc81a9a | 1269 | |
e377fcc8 | 1270 | return ALIGN(len, L1_CACHE_BYTES); |
fbf28d78 MD |
1271 | } |
1272 | ||
2de2f7f4 JF |
1273 | static int add_recvbuf_mergeable(struct virtnet_info *vi, |
1274 | struct receive_queue *rq, gfp_t gfp) | |
fbf28d78 | 1275 | { |
fb51879d | 1276 | struct page_frag *alloc_frag = &rq->alloc_frag; |
2de2f7f4 | 1277 | unsigned int headroom = virtnet_get_headroom(vi); |
3cc81a9a JW |
1278 | unsigned int tailroom = headroom ? sizeof(struct skb_shared_info) : 0; |
1279 | unsigned int room = SKB_DATA_ALIGN(headroom + tailroom); | |
fb51879d | 1280 | char *buf; |
680557cf | 1281 | void *ctx; |
3f2c31d9 | 1282 | int err; |
fb51879d | 1283 | unsigned int len, hole; |
3f2c31d9 | 1284 | |
3cc81a9a JW |
1285 | /* Extra tailroom is needed to satisfy XDP's assumption. This |
1286 | * means rx frags coalescing won't work, but consider we've | |
1287 | * disabled GSO for XDP, it won't be a big issue. | |
1288 | */ | |
1289 | len = get_mergeable_buf_len(rq, &rq->mrg_avg_pkt_len, room); | |
1290 | if (unlikely(!skb_page_frag_refill(len + room, alloc_frag, gfp))) | |
9ab86bbc | 1291 | return -ENOMEM; |
ab7db917 | 1292 | |
fb51879d | 1293 | buf = (char *)page_address(alloc_frag->page) + alloc_frag->offset; |
2de2f7f4 | 1294 | buf += headroom; /* advance address leaving hole at front of pkt */ |
fb51879d | 1295 | get_page(alloc_frag->page); |
3cc81a9a | 1296 | alloc_frag->offset += len + room; |
fb51879d | 1297 | hole = alloc_frag->size - alloc_frag->offset; |
3cc81a9a | 1298 | if (hole < len + room) { |
ab7db917 MD |
1299 | /* To avoid internal fragmentation, if there is very likely not |
1300 | * enough space for another buffer, add the remaining space to | |
1daa8790 | 1301 | * the current buffer. |
ab7db917 | 1302 | */ |
fb51879d MD |
1303 | len += hole; |
1304 | alloc_frag->offset += hole; | |
1305 | } | |
3f2c31d9 | 1306 | |
fb51879d | 1307 | sg_init_one(rq->sg, buf, len); |
29fda25a | 1308 | ctx = mergeable_len_to_ctx(len, headroom); |
680557cf | 1309 | err = virtqueue_add_inbuf_ctx(rq->vq, rq->sg, 1, buf, ctx, gfp); |
9ab86bbc | 1310 | if (err < 0) |
2613af0e | 1311 | put_page(virt_to_head_page(buf)); |
3f2c31d9 | 1312 | |
9ab86bbc SM |
1313 | return err; |
1314 | } | |
3f2c31d9 | 1315 | |
b2baed69 RR |
1316 | /* |
1317 | * Returns false if we couldn't fill entirely (OOM). | |
1318 | * | |
1319 | * Normally run in the receive path, but can also be run from ndo_open | |
1320 | * before we're receiving packets, or from refill_work which is | |
1321 | * careful to disable receiving (using napi_disable). | |
1322 | */ | |
946fa564 MT |
1323 | static bool try_fill_recv(struct virtnet_info *vi, struct receive_queue *rq, |
1324 | gfp_t gfp) | |
9ab86bbc SM |
1325 | { |
1326 | int err; | |
1788f495 | 1327 | bool oom; |
3f2c31d9 | 1328 | |
9ab86bbc SM |
1329 | do { |
1330 | if (vi->mergeable_rx_bufs) | |
2de2f7f4 | 1331 | err = add_recvbuf_mergeable(vi, rq, gfp); |
9ab86bbc | 1332 | else if (vi->big_packets) |
012873d0 | 1333 | err = add_recvbuf_big(vi, rq, gfp); |
9ab86bbc | 1334 | else |
946fa564 | 1335 | err = add_recvbuf_small(vi, rq, gfp); |
3f2c31d9 | 1336 | |
1788f495 | 1337 | oom = err == -ENOMEM; |
9ed4cb07 | 1338 | if (err) |
3f2c31d9 | 1339 | break; |
b7dfde95 | 1340 | } while (rq->vq->num_free); |
461f03dc | 1341 | if (virtqueue_kick_prepare(rq->vq) && virtqueue_notify(rq->vq)) { |
01c32598 MT |
1342 | unsigned long flags; |
1343 | ||
1344 | flags = u64_stats_update_begin_irqsave(&rq->stats.syncp); | |
d46eeeaf | 1345 | rq->stats.kicks++; |
01c32598 | 1346 | u64_stats_update_end_irqrestore(&rq->stats.syncp, flags); |
461f03dc TM |
1347 | } |
1348 | ||
3161e453 | 1349 | return !oom; |
3f2c31d9 MM |
1350 | } |
1351 | ||
18445c4d | 1352 | static void skb_recv_done(struct virtqueue *rvq) |
296f96fc RR |
1353 | { |
1354 | struct virtnet_info *vi = rvq->vdev->priv; | |
986a4f4d | 1355 | struct receive_queue *rq = &vi->rq[vq2rxq(rvq)]; |
e9d7417b | 1356 | |
e4e8452a | 1357 | virtqueue_napi_schedule(&rq->napi, rvq); |
296f96fc RR |
1358 | } |
1359 | ||
e4e8452a | 1360 | static void virtnet_napi_enable(struct virtqueue *vq, struct napi_struct *napi) |
3e9d08ec | 1361 | { |
e4e8452a | 1362 | napi_enable(napi); |
3e9d08ec BR |
1363 | |
1364 | /* If all buffers were filled by other side before we napi_enabled, we | |
e4e8452a WB |
1365 | * won't get another interrupt, so process any outstanding packets now. |
1366 | * Call local_bh_enable after to trigger softIRQ processing. | |
1367 | */ | |
1368 | local_bh_disable(); | |
1369 | virtqueue_napi_schedule(napi, vq); | |
1370 | local_bh_enable(); | |
3e9d08ec BR |
1371 | } |
1372 | ||
b92f1e67 WB |
1373 | static void virtnet_napi_tx_enable(struct virtnet_info *vi, |
1374 | struct virtqueue *vq, | |
1375 | struct napi_struct *napi) | |
1376 | { | |
1377 | if (!napi->weight) | |
1378 | return; | |
1379 | ||
1380 | /* Tx napi touches cachelines on the cpu handling tx interrupts. Only | |
1381 | * enable the feature if this is likely affine with the transmit path. | |
1382 | */ | |
1383 | if (!vi->affinity_hint_set) { | |
1384 | napi->weight = 0; | |
1385 | return; | |
1386 | } | |
1387 | ||
1388 | return virtnet_napi_enable(vq, napi); | |
1389 | } | |
1390 | ||
78a57b48 WB |
1391 | static void virtnet_napi_tx_disable(struct napi_struct *napi) |
1392 | { | |
1393 | if (napi->weight) | |
1394 | napi_disable(napi); | |
1395 | } | |
1396 | ||
3161e453 RR |
1397 | static void refill_work(struct work_struct *work) |
1398 | { | |
e9d7417b JW |
1399 | struct virtnet_info *vi = |
1400 | container_of(work, struct virtnet_info, refill.work); | |
3161e453 | 1401 | bool still_empty; |
986a4f4d JW |
1402 | int i; |
1403 | ||
55257d72 | 1404 | for (i = 0; i < vi->curr_queue_pairs; i++) { |
986a4f4d | 1405 | struct receive_queue *rq = &vi->rq[i]; |
3161e453 | 1406 | |
986a4f4d | 1407 | napi_disable(&rq->napi); |
946fa564 | 1408 | still_empty = !try_fill_recv(vi, rq, GFP_KERNEL); |
e4e8452a | 1409 | virtnet_napi_enable(rq->vq, &rq->napi); |
3161e453 | 1410 | |
986a4f4d JW |
1411 | /* In theory, this can happen: if we don't get any buffers in |
1412 | * we will *never* try to fill again. | |
1413 | */ | |
1414 | if (still_empty) | |
1415 | schedule_delayed_work(&vi->refill, HZ/2); | |
1416 | } | |
3161e453 RR |
1417 | } |
1418 | ||
2471c75e JDB |
1419 | static int virtnet_receive(struct receive_queue *rq, int budget, |
1420 | unsigned int *xdp_xmit) | |
296f96fc | 1421 | { |
e9d7417b | 1422 | struct virtnet_info *vi = rq->vq->vdev->priv; |
d46eeeaf | 1423 | struct virtnet_rq_stats stats = {}; |
a0929a44 | 1424 | unsigned int len; |
9ab86bbc | 1425 | void *buf; |
a0929a44 | 1426 | int i; |
296f96fc | 1427 | |
192f68cf | 1428 | if (!vi->big_packets || vi->mergeable_rx_bufs) { |
680557cf MT |
1429 | void *ctx; |
1430 | ||
d46eeeaf | 1431 | while (stats.packets < budget && |
680557cf | 1432 | (buf = virtqueue_get_buf_ctx(rq->vq, &len, &ctx))) { |
a0929a44 | 1433 | receive_buf(vi, rq, buf, len, ctx, xdp_xmit, &stats); |
d46eeeaf | 1434 | stats.packets++; |
680557cf MT |
1435 | } |
1436 | } else { | |
d46eeeaf | 1437 | while (stats.packets < budget && |
680557cf | 1438 | (buf = virtqueue_get_buf(rq->vq, &len)) != NULL) { |
a0929a44 | 1439 | receive_buf(vi, rq, buf, len, NULL, xdp_xmit, &stats); |
d46eeeaf | 1440 | stats.packets++; |
680557cf | 1441 | } |
296f96fc RR |
1442 | } |
1443 | ||
718be6ba | 1444 | if (rq->vq->num_free > min((unsigned int)budget, virtqueue_get_vring_size(rq->vq)) / 2) { |
946fa564 | 1445 | if (!try_fill_recv(vi, rq, GFP_ATOMIC)) |
3b07e9ca | 1446 | schedule_delayed_work(&vi->refill, 0); |
3161e453 | 1447 | } |
296f96fc | 1448 | |
d7dfc5cf | 1449 | u64_stats_update_begin(&rq->stats.syncp); |
a0929a44 TM |
1450 | for (i = 0; i < VIRTNET_RQ_STATS_LEN; i++) { |
1451 | size_t offset = virtnet_rq_stats_desc[i].offset; | |
1452 | u64 *item; | |
1453 | ||
d46eeeaf JW |
1454 | item = (u64 *)((u8 *)&rq->stats + offset); |
1455 | *item += *(u64 *)((u8 *)&stats + offset); | |
a0929a44 | 1456 | } |
d7dfc5cf | 1457 | u64_stats_update_end(&rq->stats.syncp); |
61845d20 | 1458 | |
d46eeeaf | 1459 | return stats.packets; |
2ffa7598 JW |
1460 | } |
1461 | ||
df133f3f | 1462 | static void free_old_xmit_skbs(struct send_queue *sq, bool in_napi) |
ea7735d9 | 1463 | { |
ea7735d9 | 1464 | unsigned int len; |
ea7735d9 WB |
1465 | unsigned int packets = 0; |
1466 | unsigned int bytes = 0; | |
5050471d | 1467 | void *ptr; |
ea7735d9 | 1468 | |
5050471d TM |
1469 | while ((ptr = virtqueue_get_buf(sq->vq, &len)) != NULL) { |
1470 | if (likely(!is_xdp_frame(ptr))) { | |
1471 | struct sk_buff *skb = ptr; | |
ea7735d9 | 1472 | |
5050471d | 1473 | pr_debug("Sent skb %p\n", skb); |
ea7735d9 | 1474 | |
5050471d TM |
1475 | bytes += skb->len; |
1476 | napi_consume_skb(skb, in_napi); | |
1477 | } else { | |
1478 | struct xdp_frame *frame = ptr_to_xdp(ptr); | |
ea7735d9 | 1479 | |
5050471d TM |
1480 | bytes += frame->len; |
1481 | xdp_return_frame(frame); | |
1482 | } | |
1483 | packets++; | |
ea7735d9 WB |
1484 | } |
1485 | ||
1486 | /* Avoid overhead when no packets have been processed | |
1487 | * happens when called speculatively from start_xmit. | |
1488 | */ | |
1489 | if (!packets) | |
1490 | return; | |
1491 | ||
d7dfc5cf TM |
1492 | u64_stats_update_begin(&sq->stats.syncp); |
1493 | sq->stats.bytes += bytes; | |
1494 | sq->stats.packets += packets; | |
1495 | u64_stats_update_end(&sq->stats.syncp); | |
ea7735d9 WB |
1496 | } |
1497 | ||
534da5e8 TM |
1498 | static bool is_xdp_raw_buffer_queue(struct virtnet_info *vi, int q) |
1499 | { | |
1500 | if (q < (vi->curr_queue_pairs - vi->xdp_queue_pairs)) | |
1501 | return false; | |
1502 | else if (q < vi->curr_queue_pairs) | |
1503 | return true; | |
1504 | else | |
1505 | return false; | |
1506 | } | |
1507 | ||
7b0411ef WB |
1508 | static void virtnet_poll_cleantx(struct receive_queue *rq) |
1509 | { | |
1510 | struct virtnet_info *vi = rq->vq->vdev->priv; | |
1511 | unsigned int index = vq2rxq(rq->vq); | |
1512 | struct send_queue *sq = &vi->sq[index]; | |
1513 | struct netdev_queue *txq = netdev_get_tx_queue(vi->dev, index); | |
1514 | ||
534da5e8 | 1515 | if (!sq->napi.weight || is_xdp_raw_buffer_queue(vi, index)) |
7b0411ef WB |
1516 | return; |
1517 | ||
1518 | if (__netif_tx_trylock(txq)) { | |
df133f3f | 1519 | free_old_xmit_skbs(sq, true); |
7b0411ef WB |
1520 | __netif_tx_unlock(txq); |
1521 | } | |
1522 | ||
1523 | if (sq->vq->num_free >= 2 + MAX_SKB_FRAGS) | |
1524 | netif_tx_wake_queue(txq); | |
1525 | } | |
1526 | ||
2ffa7598 JW |
1527 | static int virtnet_poll(struct napi_struct *napi, int budget) |
1528 | { | |
1529 | struct receive_queue *rq = | |
1530 | container_of(napi, struct receive_queue, napi); | |
9267c430 JW |
1531 | struct virtnet_info *vi = rq->vq->vdev->priv; |
1532 | struct send_queue *sq; | |
2a43565c | 1533 | unsigned int received; |
2471c75e | 1534 | unsigned int xdp_xmit = 0; |
2ffa7598 | 1535 | |
7b0411ef WB |
1536 | virtnet_poll_cleantx(rq); |
1537 | ||
186b3c99 | 1538 | received = virtnet_receive(rq, budget, &xdp_xmit); |
2ffa7598 | 1539 | |
8329d98e | 1540 | /* Out of packets? */ |
e4e8452a WB |
1541 | if (received < budget) |
1542 | virtqueue_napi_complete(napi, rq->vq, received); | |
296f96fc | 1543 | |
2471c75e | 1544 | if (xdp_xmit & VIRTIO_XDP_REDIR) |
1d233886 | 1545 | xdp_do_flush(); |
2471c75e JDB |
1546 | |
1547 | if (xdp_xmit & VIRTIO_XDP_TX) { | |
97c2c69e | 1548 | sq = virtnet_xdp_get_sq(vi); |
461f03dc TM |
1549 | if (virtqueue_kick_prepare(sq->vq) && virtqueue_notify(sq->vq)) { |
1550 | u64_stats_update_begin(&sq->stats.syncp); | |
1551 | sq->stats.kicks++; | |
1552 | u64_stats_update_end(&sq->stats.syncp); | |
1553 | } | |
97c2c69e | 1554 | virtnet_xdp_put_sq(vi, sq); |
9267c430 | 1555 | } |
186b3c99 | 1556 | |
296f96fc RR |
1557 | return received; |
1558 | } | |
1559 | ||
986a4f4d JW |
1560 | static int virtnet_open(struct net_device *dev) |
1561 | { | |
1562 | struct virtnet_info *vi = netdev_priv(dev); | |
754b8a21 | 1563 | int i, err; |
986a4f4d | 1564 | |
e4166625 JW |
1565 | for (i = 0; i < vi->max_queue_pairs; i++) { |
1566 | if (i < vi->curr_queue_pairs) | |
1567 | /* Make sure we have some buffers: if oom use wq. */ | |
946fa564 | 1568 | if (!try_fill_recv(vi, &vi->rq[i], GFP_KERNEL)) |
e4166625 | 1569 | schedule_delayed_work(&vi->refill, 0); |
754b8a21 | 1570 | |
b02e5a0e | 1571 | err = xdp_rxq_info_reg(&vi->rq[i].xdp_rxq, dev, i, vi->rq[i].napi.napi_id); |
754b8a21 JDB |
1572 | if (err < 0) |
1573 | return err; | |
1574 | ||
8d5d8852 JDB |
1575 | err = xdp_rxq_info_reg_mem_model(&vi->rq[i].xdp_rxq, |
1576 | MEM_TYPE_PAGE_SHARED, NULL); | |
1577 | if (err < 0) { | |
1578 | xdp_rxq_info_unreg(&vi->rq[i].xdp_rxq); | |
1579 | return err; | |
1580 | } | |
1581 | ||
e4e8452a | 1582 | virtnet_napi_enable(vi->rq[i].vq, &vi->rq[i].napi); |
b92f1e67 | 1583 | virtnet_napi_tx_enable(vi, vi->sq[i].vq, &vi->sq[i].napi); |
986a4f4d JW |
1584 | } |
1585 | ||
1586 | return 0; | |
1587 | } | |
1588 | ||
b92f1e67 WB |
1589 | static int virtnet_poll_tx(struct napi_struct *napi, int budget) |
1590 | { | |
1591 | struct send_queue *sq = container_of(napi, struct send_queue, napi); | |
1592 | struct virtnet_info *vi = sq->vq->vdev->priv; | |
534da5e8 TM |
1593 | unsigned int index = vq2txq(sq->vq); |
1594 | struct netdev_queue *txq; | |
b92f1e67 | 1595 | |
534da5e8 TM |
1596 | if (unlikely(is_xdp_raw_buffer_queue(vi, index))) { |
1597 | /* We don't need to enable cb for XDP */ | |
1598 | napi_complete_done(napi, 0); | |
1599 | return 0; | |
1600 | } | |
1601 | ||
1602 | txq = netdev_get_tx_queue(vi->dev, index); | |
b92f1e67 | 1603 | __netif_tx_lock(txq, raw_smp_processor_id()); |
df133f3f | 1604 | free_old_xmit_skbs(sq, true); |
b92f1e67 WB |
1605 | __netif_tx_unlock(txq); |
1606 | ||
1607 | virtqueue_napi_complete(napi, sq->vq, 0); | |
1608 | ||
1609 | if (sq->vq->num_free >= 2 + MAX_SKB_FRAGS) | |
1610 | netif_tx_wake_queue(txq); | |
1611 | ||
1612 | return 0; | |
1613 | } | |
1614 | ||
e9d7417b | 1615 | static int xmit_skb(struct send_queue *sq, struct sk_buff *skb) |
296f96fc | 1616 | { |
012873d0 | 1617 | struct virtio_net_hdr_mrg_rxbuf *hdr; |
296f96fc | 1618 | const unsigned char *dest = ((struct ethhdr *)skb->data)->h_dest; |
e9d7417b | 1619 | struct virtnet_info *vi = sq->vq->vdev->priv; |
e2fcad58 | 1620 | int num_sg; |
012873d0 | 1621 | unsigned hdr_len = vi->hdr_len; |
e7428e95 | 1622 | bool can_push; |
296f96fc | 1623 | |
e174961c | 1624 | pr_debug("%s: xmit %p %pM\n", vi->dev->name, skb, dest); |
e7428e95 MT |
1625 | |
1626 | can_push = vi->any_header_sg && | |
1627 | !((unsigned long)skb->data & (__alignof__(*hdr) - 1)) && | |
1628 | !skb_header_cloned(skb) && skb_headroom(skb) >= hdr_len; | |
1629 | /* Even if we can, don't push here yet as this would skew | |
1630 | * csum_start offset below. */ | |
1631 | if (can_push) | |
012873d0 | 1632 | hdr = (struct virtio_net_hdr_mrg_rxbuf *)(skb->data - hdr_len); |
e7428e95 MT |
1633 | else |
1634 | hdr = skb_vnet_hdr(skb); | |
296f96fc | 1635 | |
e858fae2 | 1636 | if (virtio_net_hdr_from_skb(skb, &hdr->hdr, |
fd3a8862 WB |
1637 | virtio_is_little_endian(vi->vdev), false, |
1638 | 0)) | |
85eb1389 | 1639 | return -EPROTO; |
296f96fc | 1640 | |
3f2c31d9 | 1641 | if (vi->mergeable_rx_bufs) |
012873d0 | 1642 | hdr->num_buffers = 0; |
3f2c31d9 | 1643 | |
547c890c | 1644 | sg_init_table(sq->sg, skb_shinfo(skb)->nr_frags + (can_push ? 1 : 2)); |
e7428e95 MT |
1645 | if (can_push) { |
1646 | __skb_push(skb, hdr_len); | |
1647 | num_sg = skb_to_sgvec(skb, sq->sg, 0, skb->len); | |
e2fcad58 JD |
1648 | if (unlikely(num_sg < 0)) |
1649 | return num_sg; | |
e7428e95 MT |
1650 | /* Pull header back to avoid skew in tx bytes calculations. */ |
1651 | __skb_pull(skb, hdr_len); | |
1652 | } else { | |
1653 | sg_set_buf(sq->sg, hdr, hdr_len); | |
e2fcad58 JD |
1654 | num_sg = skb_to_sgvec(skb, sq->sg + 1, 0, skb->len); |
1655 | if (unlikely(num_sg < 0)) | |
1656 | return num_sg; | |
1657 | num_sg++; | |
e7428e95 | 1658 | } |
9dc7b9e4 | 1659 | return virtqueue_add_outbuf(sq->vq, sq->sg, num_sg, skb, GFP_ATOMIC); |
11a3a154 RR |
1660 | } |
1661 | ||
424efe9c | 1662 | static netdev_tx_t start_xmit(struct sk_buff *skb, struct net_device *dev) |
99ffc696 RR |
1663 | { |
1664 | struct virtnet_info *vi = netdev_priv(dev); | |
986a4f4d JW |
1665 | int qnum = skb_get_queue_mapping(skb); |
1666 | struct send_queue *sq = &vi->sq[qnum]; | |
9ed4cb07 | 1667 | int err; |
4b7fd2e6 | 1668 | struct netdev_queue *txq = netdev_get_tx_queue(dev, qnum); |
6b16f9ee | 1669 | bool kick = !netdev_xmit_more(); |
b92f1e67 | 1670 | bool use_napi = sq->napi.weight; |
2cb9c6ba | 1671 | |
2cb9c6ba | 1672 | /* Free up any pending old buffers before queueing new ones. */ |
df133f3f | 1673 | free_old_xmit_skbs(sq, false); |
99ffc696 | 1674 | |
bdb12e0d WB |
1675 | if (use_napi && kick) |
1676 | virtqueue_enable_cb_delayed(sq->vq); | |
1677 | ||
074c3582 JK |
1678 | /* timestamp packet in software */ |
1679 | skb_tx_timestamp(skb); | |
1680 | ||
03f191ba | 1681 | /* Try to transmit */ |
b7dfde95 | 1682 | err = xmit_skb(sq, skb); |
48925e37 | 1683 | |
9ed4cb07 | 1684 | /* This should not happen! */ |
681daee2 | 1685 | if (unlikely(err)) { |
9ed4cb07 RR |
1686 | dev->stats.tx_fifo_errors++; |
1687 | if (net_ratelimit()) | |
1688 | dev_warn(&dev->dev, | |
7934b481 YS |
1689 | "Unexpected TXQ (%d) queue failure: %d\n", |
1690 | qnum, err); | |
58eba97d | 1691 | dev->stats.tx_dropped++; |
85e94525 | 1692 | dev_kfree_skb_any(skb); |
58eba97d | 1693 | return NETDEV_TX_OK; |
296f96fc | 1694 | } |
03f191ba | 1695 | |
48925e37 | 1696 | /* Don't wait up for transmitted skbs to be freed. */ |
b92f1e67 WB |
1697 | if (!use_napi) { |
1698 | skb_orphan(skb); | |
895b5c9f | 1699 | nf_reset_ct(skb); |
b92f1e67 | 1700 | } |
48925e37 | 1701 | |
60302ff6 MT |
1702 | /* If running out of space, stop queue to avoid getting packets that we |
1703 | * are then unable to transmit. | |
1704 | * An alternative would be to force queuing layer to requeue the skb by | |
1705 | * returning NETDEV_TX_BUSY. However, NETDEV_TX_BUSY should not be | |
1706 | * returned in a normal path of operation: it means that driver is not | |
1707 | * maintaining the TX queue stop/start state properly, and causes | |
1708 | * the stack to do a non-trivial amount of useless work. | |
1709 | * Since most packets only take 1 or 2 ring slots, stopping the queue | |
1710 | * early means 16 slots are typically wasted. | |
d631b94e | 1711 | */ |
b7dfde95 | 1712 | if (sq->vq->num_free < 2+MAX_SKB_FRAGS) { |
986a4f4d | 1713 | netif_stop_subqueue(dev, qnum); |
b92f1e67 WB |
1714 | if (!use_napi && |
1715 | unlikely(!virtqueue_enable_cb_delayed(sq->vq))) { | |
48925e37 | 1716 | /* More just got used, free them then recheck. */ |
df133f3f | 1717 | free_old_xmit_skbs(sq, false); |
b7dfde95 | 1718 | if (sq->vq->num_free >= 2+MAX_SKB_FRAGS) { |
986a4f4d | 1719 | netif_start_subqueue(dev, qnum); |
e9d7417b | 1720 | virtqueue_disable_cb(sq->vq); |
48925e37 RR |
1721 | } |
1722 | } | |
99ffc696 | 1723 | } |
48925e37 | 1724 | |
461f03dc TM |
1725 | if (kick || netif_xmit_stopped(txq)) { |
1726 | if (virtqueue_kick_prepare(sq->vq) && virtqueue_notify(sq->vq)) { | |
1727 | u64_stats_update_begin(&sq->stats.syncp); | |
1728 | sq->stats.kicks++; | |
1729 | u64_stats_update_end(&sq->stats.syncp); | |
1730 | } | |
1731 | } | |
296f96fc | 1732 | |
0b725a2c | 1733 | return NETDEV_TX_OK; |
c223a078 DM |
1734 | } |
1735 | ||
40cbfc37 AK |
1736 | /* |
1737 | * Send command via the control virtqueue and check status. Commands | |
1738 | * supported by the hypervisor, as indicated by feature bits, should | |
788a8b6d | 1739 | * never fail unless improperly formatted. |
40cbfc37 AK |
1740 | */ |
1741 | static bool virtnet_send_command(struct virtnet_info *vi, u8 class, u8 cmd, | |
d24bae32 | 1742 | struct scatterlist *out) |
40cbfc37 | 1743 | { |
f7bc9594 | 1744 | struct scatterlist *sgs[4], hdr, stat; |
d24bae32 | 1745 | unsigned out_num = 0, tmp; |
40cbfc37 AK |
1746 | |
1747 | /* Caller should know better */ | |
f7bc9594 | 1748 | BUG_ON(!virtio_has_feature(vi->vdev, VIRTIO_NET_F_CTRL_VQ)); |
40cbfc37 | 1749 | |
12e57169 MT |
1750 | vi->ctrl->status = ~0; |
1751 | vi->ctrl->hdr.class = class; | |
1752 | vi->ctrl->hdr.cmd = cmd; | |
f7bc9594 | 1753 | /* Add header */ |
12e57169 | 1754 | sg_init_one(&hdr, &vi->ctrl->hdr, sizeof(vi->ctrl->hdr)); |
f7bc9594 | 1755 | sgs[out_num++] = &hdr; |
40cbfc37 | 1756 | |
f7bc9594 RR |
1757 | if (out) |
1758 | sgs[out_num++] = out; | |
40cbfc37 | 1759 | |
f7bc9594 | 1760 | /* Add return status. */ |
12e57169 | 1761 | sg_init_one(&stat, &vi->ctrl->status, sizeof(vi->ctrl->status)); |
d24bae32 | 1762 | sgs[out_num] = &stat; |
40cbfc37 | 1763 | |
d24bae32 | 1764 | BUG_ON(out_num + 1 > ARRAY_SIZE(sgs)); |
a7c58146 | 1765 | virtqueue_add_sgs(vi->cvq, sgs, out_num, 1, vi, GFP_ATOMIC); |
40cbfc37 | 1766 | |
67975901 | 1767 | if (unlikely(!virtqueue_kick(vi->cvq))) |
12e57169 | 1768 | return vi->ctrl->status == VIRTIO_NET_OK; |
40cbfc37 AK |
1769 | |
1770 | /* Spin for a response, the kick causes an ioport write, trapping | |
1771 | * into the hypervisor, so the request should be handled immediately. | |
1772 | */ | |
047b9b94 HG |
1773 | while (!virtqueue_get_buf(vi->cvq, &tmp) && |
1774 | !virtqueue_is_broken(vi->cvq)) | |
40cbfc37 AK |
1775 | cpu_relax(); |
1776 | ||
12e57169 | 1777 | return vi->ctrl->status == VIRTIO_NET_OK; |
40cbfc37 AK |
1778 | } |
1779 | ||
9c46f6d4 AW |
1780 | static int virtnet_set_mac_address(struct net_device *dev, void *p) |
1781 | { | |
1782 | struct virtnet_info *vi = netdev_priv(dev); | |
1783 | struct virtio_device *vdev = vi->vdev; | |
f2f2c8b4 | 1784 | int ret; |
e37e2ff3 | 1785 | struct sockaddr *addr; |
7e58d5ae | 1786 | struct scatterlist sg; |
9c46f6d4 | 1787 | |
ba5e4426 SS |
1788 | if (virtio_has_feature(vi->vdev, VIRTIO_NET_F_STANDBY)) |
1789 | return -EOPNOTSUPP; | |
1790 | ||
801822d1 | 1791 | addr = kmemdup(p, sizeof(*addr), GFP_KERNEL); |
e37e2ff3 AL |
1792 | if (!addr) |
1793 | return -ENOMEM; | |
e37e2ff3 AL |
1794 | |
1795 | ret = eth_prepare_mac_addr_change(dev, addr); | |
f2f2c8b4 | 1796 | if (ret) |
e37e2ff3 | 1797 | goto out; |
9c46f6d4 | 1798 | |
7e58d5ae AK |
1799 | if (virtio_has_feature(vdev, VIRTIO_NET_F_CTRL_MAC_ADDR)) { |
1800 | sg_init_one(&sg, addr->sa_data, dev->addr_len); | |
1801 | if (!virtnet_send_command(vi, VIRTIO_NET_CTRL_MAC, | |
d24bae32 | 1802 | VIRTIO_NET_CTRL_MAC_ADDR_SET, &sg)) { |
7e58d5ae AK |
1803 | dev_warn(&vdev->dev, |
1804 | "Failed to set mac address by vq command.\n"); | |
e37e2ff3 AL |
1805 | ret = -EINVAL; |
1806 | goto out; | |
7e58d5ae | 1807 | } |
7e93a02f MT |
1808 | } else if (virtio_has_feature(vdev, VIRTIO_NET_F_MAC) && |
1809 | !virtio_has_feature(vdev, VIRTIO_F_VERSION_1)) { | |
855e0c52 RR |
1810 | unsigned int i; |
1811 | ||
1812 | /* Naturally, this has an atomicity problem. */ | |
1813 | for (i = 0; i < dev->addr_len; i++) | |
1814 | virtio_cwrite8(vdev, | |
1815 | offsetof(struct virtio_net_config, mac) + | |
1816 | i, addr->sa_data[i]); | |
7e58d5ae AK |
1817 | } |
1818 | ||
1819 | eth_commit_mac_addr_change(dev, p); | |
e37e2ff3 | 1820 | ret = 0; |
9c46f6d4 | 1821 | |
e37e2ff3 AL |
1822 | out: |
1823 | kfree(addr); | |
1824 | return ret; | |
9c46f6d4 AW |
1825 | } |
1826 | ||
bc1f4470 | 1827 | static void virtnet_stats(struct net_device *dev, |
1828 | struct rtnl_link_stats64 *tot) | |
3fa2a1df | 1829 | { |
1830 | struct virtnet_info *vi = netdev_priv(dev); | |
3fa2a1df | 1831 | unsigned int start; |
d7dfc5cf | 1832 | int i; |
3fa2a1df | 1833 | |
d7dfc5cf | 1834 | for (i = 0; i < vi->max_queue_pairs; i++) { |
2c4a2f7d | 1835 | u64 tpackets, tbytes, rpackets, rbytes, rdrops; |
d7dfc5cf TM |
1836 | struct receive_queue *rq = &vi->rq[i]; |
1837 | struct send_queue *sq = &vi->sq[i]; | |
3fa2a1df | 1838 | |
1839 | do { | |
d7dfc5cf TM |
1840 | start = u64_stats_fetch_begin_irq(&sq->stats.syncp); |
1841 | tpackets = sq->stats.packets; | |
1842 | tbytes = sq->stats.bytes; | |
1843 | } while (u64_stats_fetch_retry_irq(&sq->stats.syncp, start)); | |
83a27052 ED |
1844 | |
1845 | do { | |
d7dfc5cf | 1846 | start = u64_stats_fetch_begin_irq(&rq->stats.syncp); |
d46eeeaf JW |
1847 | rpackets = rq->stats.packets; |
1848 | rbytes = rq->stats.bytes; | |
1849 | rdrops = rq->stats.drops; | |
d7dfc5cf | 1850 | } while (u64_stats_fetch_retry_irq(&rq->stats.syncp, start)); |
3fa2a1df | 1851 | |
1852 | tot->rx_packets += rpackets; | |
1853 | tot->tx_packets += tpackets; | |
1854 | tot->rx_bytes += rbytes; | |
1855 | tot->tx_bytes += tbytes; | |
2c4a2f7d | 1856 | tot->rx_dropped += rdrops; |
3fa2a1df | 1857 | } |
1858 | ||
1859 | tot->tx_dropped = dev->stats.tx_dropped; | |
021ac8d3 | 1860 | tot->tx_fifo_errors = dev->stats.tx_fifo_errors; |
3fa2a1df | 1861 | tot->rx_length_errors = dev->stats.rx_length_errors; |
1862 | tot->rx_frame_errors = dev->stats.rx_frame_errors; | |
3fa2a1df | 1863 | } |
1864 | ||
586d17c5 JW |
1865 | static void virtnet_ack_link_announce(struct virtnet_info *vi) |
1866 | { | |
1867 | rtnl_lock(); | |
1868 | if (!virtnet_send_command(vi, VIRTIO_NET_CTRL_ANNOUNCE, | |
d24bae32 | 1869 | VIRTIO_NET_CTRL_ANNOUNCE_ACK, NULL)) |
586d17c5 JW |
1870 | dev_warn(&vi->dev->dev, "Failed to ack link announce.\n"); |
1871 | rtnl_unlock(); | |
1872 | } | |
1873 | ||
47315329 | 1874 | static int _virtnet_set_queues(struct virtnet_info *vi, u16 queue_pairs) |
986a4f4d JW |
1875 | { |
1876 | struct scatterlist sg; | |
986a4f4d JW |
1877 | struct net_device *dev = vi->dev; |
1878 | ||
1879 | if (!vi->has_cvq || !virtio_has_feature(vi->vdev, VIRTIO_NET_F_MQ)) | |
1880 | return 0; | |
1881 | ||
12e57169 MT |
1882 | vi->ctrl->mq.virtqueue_pairs = cpu_to_virtio16(vi->vdev, queue_pairs); |
1883 | sg_init_one(&sg, &vi->ctrl->mq, sizeof(vi->ctrl->mq)); | |
986a4f4d JW |
1884 | |
1885 | if (!virtnet_send_command(vi, VIRTIO_NET_CTRL_MQ, | |
d24bae32 | 1886 | VIRTIO_NET_CTRL_MQ_VQ_PAIRS_SET, &sg)) { |
986a4f4d JW |
1887 | dev_warn(&dev->dev, "Fail to set num of queue pairs to %d\n", |
1888 | queue_pairs); | |
1889 | return -EINVAL; | |
55257d72 | 1890 | } else { |
986a4f4d | 1891 | vi->curr_queue_pairs = queue_pairs; |
35ed159b JW |
1892 | /* virtnet_open() will refill when device is going to up. */ |
1893 | if (dev->flags & IFF_UP) | |
1894 | schedule_delayed_work(&vi->refill, 0); | |
55257d72 | 1895 | } |
986a4f4d JW |
1896 | |
1897 | return 0; | |
1898 | } | |
1899 | ||
47315329 JF |
1900 | static int virtnet_set_queues(struct virtnet_info *vi, u16 queue_pairs) |
1901 | { | |
1902 | int err; | |
1903 | ||
1904 | rtnl_lock(); | |
1905 | err = _virtnet_set_queues(vi, queue_pairs); | |
1906 | rtnl_unlock(); | |
1907 | return err; | |
1908 | } | |
1909 | ||
296f96fc RR |
1910 | static int virtnet_close(struct net_device *dev) |
1911 | { | |
1912 | struct virtnet_info *vi = netdev_priv(dev); | |
986a4f4d | 1913 | int i; |
296f96fc | 1914 | |
b2baed69 RR |
1915 | /* Make sure refill_work doesn't re-enable napi! */ |
1916 | cancel_delayed_work_sync(&vi->refill); | |
986a4f4d | 1917 | |
b92f1e67 | 1918 | for (i = 0; i < vi->max_queue_pairs; i++) { |
754b8a21 | 1919 | xdp_rxq_info_unreg(&vi->rq[i].xdp_rxq); |
986a4f4d | 1920 | napi_disable(&vi->rq[i].napi); |
78a57b48 | 1921 | virtnet_napi_tx_disable(&vi->sq[i].napi); |
b92f1e67 | 1922 | } |
296f96fc | 1923 | |
296f96fc RR |
1924 | return 0; |
1925 | } | |
1926 | ||
2af7698e AW |
1927 | static void virtnet_set_rx_mode(struct net_device *dev) |
1928 | { | |
1929 | struct virtnet_info *vi = netdev_priv(dev); | |
f565a7c2 | 1930 | struct scatterlist sg[2]; |
f565a7c2 | 1931 | struct virtio_net_ctrl_mac *mac_data; |
ccffad25 | 1932 | struct netdev_hw_addr *ha; |
32e7bfc4 | 1933 | int uc_count; |
4cd24eaf | 1934 | int mc_count; |
f565a7c2 AW |
1935 | void *buf; |
1936 | int i; | |
2af7698e | 1937 | |
788a8b6d | 1938 | /* We can't dynamically set ndo_set_rx_mode, so return gracefully */ |
2af7698e AW |
1939 | if (!virtio_has_feature(vi->vdev, VIRTIO_NET_F_CTRL_RX)) |
1940 | return; | |
1941 | ||
12e57169 MT |
1942 | vi->ctrl->promisc = ((dev->flags & IFF_PROMISC) != 0); |
1943 | vi->ctrl->allmulti = ((dev->flags & IFF_ALLMULTI) != 0); | |
2af7698e | 1944 | |
12e57169 | 1945 | sg_init_one(sg, &vi->ctrl->promisc, sizeof(vi->ctrl->promisc)); |
2af7698e AW |
1946 | |
1947 | if (!virtnet_send_command(vi, VIRTIO_NET_CTRL_RX, | |
d24bae32 | 1948 | VIRTIO_NET_CTRL_RX_PROMISC, sg)) |
2af7698e | 1949 | dev_warn(&dev->dev, "Failed to %sable promisc mode.\n", |
12e57169 | 1950 | vi->ctrl->promisc ? "en" : "dis"); |
2af7698e | 1951 | |
12e57169 | 1952 | sg_init_one(sg, &vi->ctrl->allmulti, sizeof(vi->ctrl->allmulti)); |
2af7698e AW |
1953 | |
1954 | if (!virtnet_send_command(vi, VIRTIO_NET_CTRL_RX, | |
d24bae32 | 1955 | VIRTIO_NET_CTRL_RX_ALLMULTI, sg)) |
2af7698e | 1956 | dev_warn(&dev->dev, "Failed to %sable allmulti mode.\n", |
12e57169 | 1957 | vi->ctrl->allmulti ? "en" : "dis"); |
f565a7c2 | 1958 | |
32e7bfc4 | 1959 | uc_count = netdev_uc_count(dev); |
4cd24eaf | 1960 | mc_count = netdev_mc_count(dev); |
f565a7c2 | 1961 | /* MAC filter - use one buffer for both lists */ |
4cd24eaf JP |
1962 | buf = kzalloc(((uc_count + mc_count) * ETH_ALEN) + |
1963 | (2 * sizeof(mac_data->entries)), GFP_ATOMIC); | |
1964 | mac_data = buf; | |
e68ed8f0 | 1965 | if (!buf) |
f565a7c2 | 1966 | return; |
f565a7c2 | 1967 | |
23e258e1 AW |
1968 | sg_init_table(sg, 2); |
1969 | ||
f565a7c2 | 1970 | /* Store the unicast list and count in the front of the buffer */ |
fdd819b2 | 1971 | mac_data->entries = cpu_to_virtio32(vi->vdev, uc_count); |
ccffad25 | 1972 | i = 0; |
32e7bfc4 | 1973 | netdev_for_each_uc_addr(ha, dev) |
ccffad25 | 1974 | memcpy(&mac_data->macs[i++][0], ha->addr, ETH_ALEN); |
f565a7c2 AW |
1975 | |
1976 | sg_set_buf(&sg[0], mac_data, | |
32e7bfc4 | 1977 | sizeof(mac_data->entries) + (uc_count * ETH_ALEN)); |
f565a7c2 AW |
1978 | |
1979 | /* multicast list and count fill the end */ | |
32e7bfc4 | 1980 | mac_data = (void *)&mac_data->macs[uc_count][0]; |
f565a7c2 | 1981 | |
fdd819b2 | 1982 | mac_data->entries = cpu_to_virtio32(vi->vdev, mc_count); |
567ec874 | 1983 | i = 0; |
22bedad3 JP |
1984 | netdev_for_each_mc_addr(ha, dev) |
1985 | memcpy(&mac_data->macs[i++][0], ha->addr, ETH_ALEN); | |
f565a7c2 AW |
1986 | |
1987 | sg_set_buf(&sg[1], mac_data, | |
4cd24eaf | 1988 | sizeof(mac_data->entries) + (mc_count * ETH_ALEN)); |
f565a7c2 AW |
1989 | |
1990 | if (!virtnet_send_command(vi, VIRTIO_NET_CTRL_MAC, | |
d24bae32 | 1991 | VIRTIO_NET_CTRL_MAC_TABLE_SET, sg)) |
99e872ae | 1992 | dev_warn(&dev->dev, "Failed to set MAC filter table.\n"); |
f565a7c2 AW |
1993 | |
1994 | kfree(buf); | |
2af7698e AW |
1995 | } |
1996 | ||
80d5c368 PM |
1997 | static int virtnet_vlan_rx_add_vid(struct net_device *dev, |
1998 | __be16 proto, u16 vid) | |
0bde9569 AW |
1999 | { |
2000 | struct virtnet_info *vi = netdev_priv(dev); | |
2001 | struct scatterlist sg; | |
2002 | ||
d7fad4c8 | 2003 | vi->ctrl->vid = cpu_to_virtio16(vi->vdev, vid); |
12e57169 | 2004 | sg_init_one(&sg, &vi->ctrl->vid, sizeof(vi->ctrl->vid)); |
0bde9569 AW |
2005 | |
2006 | if (!virtnet_send_command(vi, VIRTIO_NET_CTRL_VLAN, | |
d24bae32 | 2007 | VIRTIO_NET_CTRL_VLAN_ADD, &sg)) |
0bde9569 | 2008 | dev_warn(&dev->dev, "Failed to add VLAN ID %d.\n", vid); |
8e586137 | 2009 | return 0; |
0bde9569 AW |
2010 | } |
2011 | ||
80d5c368 PM |
2012 | static int virtnet_vlan_rx_kill_vid(struct net_device *dev, |
2013 | __be16 proto, u16 vid) | |
0bde9569 AW |
2014 | { |
2015 | struct virtnet_info *vi = netdev_priv(dev); | |
2016 | struct scatterlist sg; | |
2017 | ||
d7fad4c8 | 2018 | vi->ctrl->vid = cpu_to_virtio16(vi->vdev, vid); |
12e57169 | 2019 | sg_init_one(&sg, &vi->ctrl->vid, sizeof(vi->ctrl->vid)); |
0bde9569 AW |
2020 | |
2021 | if (!virtnet_send_command(vi, VIRTIO_NET_CTRL_VLAN, | |
d24bae32 | 2022 | VIRTIO_NET_CTRL_VLAN_DEL, &sg)) |
0bde9569 | 2023 | dev_warn(&dev->dev, "Failed to kill VLAN ID %d.\n", vid); |
8e586137 | 2024 | return 0; |
0bde9569 AW |
2025 | } |
2026 | ||
310974fa | 2027 | static void virtnet_clean_affinity(struct virtnet_info *vi) |
986a4f4d JW |
2028 | { |
2029 | int i; | |
2030 | ||
8898c21c WG |
2031 | if (vi->affinity_hint_set) { |
2032 | for (i = 0; i < vi->max_queue_pairs; i++) { | |
19e226e8 CR |
2033 | virtqueue_set_affinity(vi->rq[i].vq, NULL); |
2034 | virtqueue_set_affinity(vi->sq[i].vq, NULL); | |
47be2479 WG |
2035 | } |
2036 | ||
8898c21c WG |
2037 | vi->affinity_hint_set = false; |
2038 | } | |
8898c21c | 2039 | } |
47be2479 | 2040 | |
8898c21c WG |
2041 | static void virtnet_set_affinity(struct virtnet_info *vi) |
2042 | { | |
2ca653d6 CR |
2043 | cpumask_var_t mask; |
2044 | int stragglers; | |
2045 | int group_size; | |
2046 | int i, j, cpu; | |
2047 | int num_cpu; | |
2048 | int stride; | |
2049 | ||
2050 | if (!zalloc_cpumask_var(&mask, GFP_KERNEL)) { | |
310974fa | 2051 | virtnet_clean_affinity(vi); |
8898c21c | 2052 | return; |
986a4f4d JW |
2053 | } |
2054 | ||
2ca653d6 CR |
2055 | num_cpu = num_online_cpus(); |
2056 | stride = max_t(int, num_cpu / vi->curr_queue_pairs, 1); | |
2057 | stragglers = num_cpu >= vi->curr_queue_pairs ? | |
2058 | num_cpu % vi->curr_queue_pairs : | |
2059 | 0; | |
2060 | cpu = cpumask_next(-1, cpu_online_mask); | |
4d99f660 | 2061 | |
2ca653d6 CR |
2062 | for (i = 0; i < vi->curr_queue_pairs; i++) { |
2063 | group_size = stride + (i < stragglers ? 1 : 0); | |
2064 | ||
2065 | for (j = 0; j < group_size; j++) { | |
2066 | cpumask_set_cpu(cpu, mask); | |
2067 | cpu = cpumask_next_wrap(cpu, cpu_online_mask, | |
2068 | nr_cpu_ids, false); | |
2069 | } | |
2070 | virtqueue_set_affinity(vi->rq[i].vq, mask); | |
2071 | virtqueue_set_affinity(vi->sq[i].vq, mask); | |
044ab86d | 2072 | __netif_set_xps_queue(vi->dev, cpumask_bits(mask), i, XPS_CPUS); |
2ca653d6 | 2073 | cpumask_clear(mask); |
986a4f4d JW |
2074 | } |
2075 | ||
8898c21c | 2076 | vi->affinity_hint_set = true; |
2ca653d6 | 2077 | free_cpumask_var(mask); |
986a4f4d JW |
2078 | } |
2079 | ||
8017c279 | 2080 | static int virtnet_cpu_online(unsigned int cpu, struct hlist_node *node) |
8de4b2f3 | 2081 | { |
8017c279 SAS |
2082 | struct virtnet_info *vi = hlist_entry_safe(node, struct virtnet_info, |
2083 | node); | |
2084 | virtnet_set_affinity(vi); | |
2085 | return 0; | |
2086 | } | |
8de4b2f3 | 2087 | |
8017c279 SAS |
2088 | static int virtnet_cpu_dead(unsigned int cpu, struct hlist_node *node) |
2089 | { | |
2090 | struct virtnet_info *vi = hlist_entry_safe(node, struct virtnet_info, | |
2091 | node_dead); | |
2092 | virtnet_set_affinity(vi); | |
2093 | return 0; | |
2094 | } | |
3ab098df | 2095 | |
8017c279 SAS |
2096 | static int virtnet_cpu_down_prep(unsigned int cpu, struct hlist_node *node) |
2097 | { | |
2098 | struct virtnet_info *vi = hlist_entry_safe(node, struct virtnet_info, | |
2099 | node); | |
2100 | ||
310974fa | 2101 | virtnet_clean_affinity(vi); |
8017c279 SAS |
2102 | return 0; |
2103 | } | |
2104 | ||
2105 | static enum cpuhp_state virtionet_online; | |
2106 | ||
2107 | static int virtnet_cpu_notif_add(struct virtnet_info *vi) | |
2108 | { | |
2109 | int ret; | |
2110 | ||
2111 | ret = cpuhp_state_add_instance_nocalls(virtionet_online, &vi->node); | |
2112 | if (ret) | |
2113 | return ret; | |
2114 | ret = cpuhp_state_add_instance_nocalls(CPUHP_VIRT_NET_DEAD, | |
2115 | &vi->node_dead); | |
2116 | if (!ret) | |
2117 | return ret; | |
2118 | cpuhp_state_remove_instance_nocalls(virtionet_online, &vi->node); | |
2119 | return ret; | |
2120 | } | |
2121 | ||
2122 | static void virtnet_cpu_notif_remove(struct virtnet_info *vi) | |
2123 | { | |
2124 | cpuhp_state_remove_instance_nocalls(virtionet_online, &vi->node); | |
2125 | cpuhp_state_remove_instance_nocalls(CPUHP_VIRT_NET_DEAD, | |
2126 | &vi->node_dead); | |
986a4f4d JW |
2127 | } |
2128 | ||
8f9f4668 RJ |
2129 | static void virtnet_get_ringparam(struct net_device *dev, |
2130 | struct ethtool_ringparam *ring) | |
2131 | { | |
2132 | struct virtnet_info *vi = netdev_priv(dev); | |
2133 | ||
986a4f4d JW |
2134 | ring->rx_max_pending = virtqueue_get_vring_size(vi->rq[0].vq); |
2135 | ring->tx_max_pending = virtqueue_get_vring_size(vi->sq[0].vq); | |
8f9f4668 RJ |
2136 | ring->rx_pending = ring->rx_max_pending; |
2137 | ring->tx_pending = ring->tx_max_pending; | |
8f9f4668 RJ |
2138 | } |
2139 | ||
66846048 RJ |
2140 | |
2141 | static void virtnet_get_drvinfo(struct net_device *dev, | |
2142 | struct ethtool_drvinfo *info) | |
2143 | { | |
2144 | struct virtnet_info *vi = netdev_priv(dev); | |
2145 | struct virtio_device *vdev = vi->vdev; | |
2146 | ||
2147 | strlcpy(info->driver, KBUILD_MODNAME, sizeof(info->driver)); | |
2148 | strlcpy(info->version, VIRTNET_DRIVER_VERSION, sizeof(info->version)); | |
2149 | strlcpy(info->bus_info, virtio_bus_name(vdev), sizeof(info->bus_info)); | |
2150 | ||
2151 | } | |
2152 | ||
d73bcd2c JW |
2153 | /* TODO: Eliminate OOO packets during switching */ |
2154 | static int virtnet_set_channels(struct net_device *dev, | |
2155 | struct ethtool_channels *channels) | |
2156 | { | |
2157 | struct virtnet_info *vi = netdev_priv(dev); | |
2158 | u16 queue_pairs = channels->combined_count; | |
2159 | int err; | |
2160 | ||
2161 | /* We don't support separate rx/tx channels. | |
2162 | * We don't allow setting 'other' channels. | |
2163 | */ | |
2164 | if (channels->rx_count || channels->tx_count || channels->other_count) | |
2165 | return -EINVAL; | |
2166 | ||
c18e9cd6 | 2167 | if (queue_pairs > vi->max_queue_pairs || queue_pairs == 0) |
d73bcd2c JW |
2168 | return -EINVAL; |
2169 | ||
f600b690 JF |
2170 | /* For now we don't support modifying channels while XDP is loaded |
2171 | * also when XDP is loaded all RX queues have XDP programs so we only | |
2172 | * need to check a single RX queue. | |
2173 | */ | |
2174 | if (vi->rq[0].xdp_prog) | |
2175 | return -EINVAL; | |
2176 | ||
47be2479 | 2177 | get_online_cpus(); |
47315329 | 2178 | err = _virtnet_set_queues(vi, queue_pairs); |
de33212f JD |
2179 | if (err) { |
2180 | put_online_cpus(); | |
2181 | goto err; | |
d73bcd2c | 2182 | } |
de33212f | 2183 | virtnet_set_affinity(vi); |
47be2479 | 2184 | put_online_cpus(); |
d73bcd2c | 2185 | |
de33212f JD |
2186 | netif_set_real_num_tx_queues(dev, queue_pairs); |
2187 | netif_set_real_num_rx_queues(dev, queue_pairs); | |
2188 | err: | |
d73bcd2c JW |
2189 | return err; |
2190 | } | |
2191 | ||
d7dfc5cf TM |
2192 | static void virtnet_get_strings(struct net_device *dev, u32 stringset, u8 *data) |
2193 | { | |
2194 | struct virtnet_info *vi = netdev_priv(dev); | |
d7dfc5cf | 2195 | unsigned int i, j; |
d7a9a01b | 2196 | u8 *p = data; |
d7dfc5cf TM |
2197 | |
2198 | switch (stringset) { | |
2199 | case ETH_SS_STATS: | |
2200 | for (i = 0; i < vi->curr_queue_pairs; i++) { | |
d7a9a01b AD |
2201 | for (j = 0; j < VIRTNET_RQ_STATS_LEN; j++) |
2202 | ethtool_sprintf(&p, "rx_queue_%u_%s", i, | |
2203 | virtnet_rq_stats_desc[j].desc); | |
d7dfc5cf TM |
2204 | } |
2205 | ||
2206 | for (i = 0; i < vi->curr_queue_pairs; i++) { | |
d7a9a01b AD |
2207 | for (j = 0; j < VIRTNET_SQ_STATS_LEN; j++) |
2208 | ethtool_sprintf(&p, "tx_queue_%u_%s", i, | |
2209 | virtnet_sq_stats_desc[j].desc); | |
d7dfc5cf TM |
2210 | } |
2211 | break; | |
2212 | } | |
2213 | } | |
2214 | ||
2215 | static int virtnet_get_sset_count(struct net_device *dev, int sset) | |
2216 | { | |
2217 | struct virtnet_info *vi = netdev_priv(dev); | |
2218 | ||
2219 | switch (sset) { | |
2220 | case ETH_SS_STATS: | |
2221 | return vi->curr_queue_pairs * (VIRTNET_RQ_STATS_LEN + | |
2222 | VIRTNET_SQ_STATS_LEN); | |
2223 | default: | |
2224 | return -EOPNOTSUPP; | |
2225 | } | |
2226 | } | |
2227 | ||
2228 | static void virtnet_get_ethtool_stats(struct net_device *dev, | |
2229 | struct ethtool_stats *stats, u64 *data) | |
2230 | { | |
2231 | struct virtnet_info *vi = netdev_priv(dev); | |
2232 | unsigned int idx = 0, start, i, j; | |
2233 | const u8 *stats_base; | |
2234 | size_t offset; | |
2235 | ||
2236 | for (i = 0; i < vi->curr_queue_pairs; i++) { | |
2237 | struct receive_queue *rq = &vi->rq[i]; | |
2238 | ||
d46eeeaf | 2239 | stats_base = (u8 *)&rq->stats; |
d7dfc5cf TM |
2240 | do { |
2241 | start = u64_stats_fetch_begin_irq(&rq->stats.syncp); | |
2242 | for (j = 0; j < VIRTNET_RQ_STATS_LEN; j++) { | |
2243 | offset = virtnet_rq_stats_desc[j].offset; | |
2244 | data[idx + j] = *(u64 *)(stats_base + offset); | |
2245 | } | |
2246 | } while (u64_stats_fetch_retry_irq(&rq->stats.syncp, start)); | |
2247 | idx += VIRTNET_RQ_STATS_LEN; | |
2248 | } | |
2249 | ||
2250 | for (i = 0; i < vi->curr_queue_pairs; i++) { | |
2251 | struct send_queue *sq = &vi->sq[i]; | |
2252 | ||
2253 | stats_base = (u8 *)&sq->stats; | |
2254 | do { | |
2255 | start = u64_stats_fetch_begin_irq(&sq->stats.syncp); | |
2256 | for (j = 0; j < VIRTNET_SQ_STATS_LEN; j++) { | |
2257 | offset = virtnet_sq_stats_desc[j].offset; | |
2258 | data[idx + j] = *(u64 *)(stats_base + offset); | |
2259 | } | |
2260 | } while (u64_stats_fetch_retry_irq(&sq->stats.syncp, start)); | |
2261 | idx += VIRTNET_SQ_STATS_LEN; | |
2262 | } | |
2263 | } | |
2264 | ||
d73bcd2c JW |
2265 | static void virtnet_get_channels(struct net_device *dev, |
2266 | struct ethtool_channels *channels) | |
2267 | { | |
2268 | struct virtnet_info *vi = netdev_priv(dev); | |
2269 | ||
2270 | channels->combined_count = vi->curr_queue_pairs; | |
2271 | channels->max_combined = vi->max_queue_pairs; | |
2272 | channels->max_other = 0; | |
2273 | channels->rx_count = 0; | |
2274 | channels->tx_count = 0; | |
2275 | channels->other_count = 0; | |
2276 | } | |
2277 | ||
ebb6b4b1 PR |
2278 | static int virtnet_set_link_ksettings(struct net_device *dev, |
2279 | const struct ethtool_link_ksettings *cmd) | |
16032be5 NA |
2280 | { |
2281 | struct virtnet_info *vi = netdev_priv(dev); | |
16032be5 | 2282 | |
9aedc6e2 CF |
2283 | return ethtool_virtdev_set_link_ksettings(dev, cmd, |
2284 | &vi->speed, &vi->duplex); | |
16032be5 NA |
2285 | } |
2286 | ||
ebb6b4b1 PR |
2287 | static int virtnet_get_link_ksettings(struct net_device *dev, |
2288 | struct ethtool_link_ksettings *cmd) | |
16032be5 NA |
2289 | { |
2290 | struct virtnet_info *vi = netdev_priv(dev); | |
2291 | ||
ebb6b4b1 PR |
2292 | cmd->base.speed = vi->speed; |
2293 | cmd->base.duplex = vi->duplex; | |
2294 | cmd->base.port = PORT_OTHER; | |
16032be5 NA |
2295 | |
2296 | return 0; | |
2297 | } | |
2298 | ||
0c465be1 JW |
2299 | static int virtnet_set_coalesce(struct net_device *dev, |
2300 | struct ethtool_coalesce *ec) | |
2301 | { | |
0c465be1 JW |
2302 | struct virtnet_info *vi = netdev_priv(dev); |
2303 | int i, napi_weight; | |
2304 | ||
a51e5206 JK |
2305 | if (ec->tx_max_coalesced_frames > 1 || |
2306 | ec->rx_max_coalesced_frames != 1) | |
0c465be1 JW |
2307 | return -EINVAL; |
2308 | ||
0c465be1 | 2309 | napi_weight = ec->tx_max_coalesced_frames ? NAPI_POLL_WEIGHT : 0; |
0c465be1 JW |
2310 | if (napi_weight ^ vi->sq[0].napi.weight) { |
2311 | if (dev->flags & IFF_UP) | |
2312 | return -EBUSY; | |
2313 | for (i = 0; i < vi->max_queue_pairs; i++) | |
2314 | vi->sq[i].napi.weight = napi_weight; | |
2315 | } | |
2316 | ||
2317 | return 0; | |
2318 | } | |
2319 | ||
2320 | static int virtnet_get_coalesce(struct net_device *dev, | |
2321 | struct ethtool_coalesce *ec) | |
2322 | { | |
2323 | struct ethtool_coalesce ec_default = { | |
2324 | .cmd = ETHTOOL_GCOALESCE, | |
2325 | .rx_max_coalesced_frames = 1, | |
2326 | }; | |
2327 | struct virtnet_info *vi = netdev_priv(dev); | |
2328 | ||
2329 | memcpy(ec, &ec_default, sizeof(ec_default)); | |
2330 | ||
2331 | if (vi->sq[0].napi.weight) | |
2332 | ec->tx_max_coalesced_frames = 1; | |
2333 | ||
2334 | return 0; | |
2335 | } | |
2336 | ||
16032be5 NA |
2337 | static void virtnet_init_settings(struct net_device *dev) |
2338 | { | |
2339 | struct virtnet_info *vi = netdev_priv(dev); | |
2340 | ||
2341 | vi->speed = SPEED_UNKNOWN; | |
2342 | vi->duplex = DUPLEX_UNKNOWN; | |
2343 | } | |
2344 | ||
faa9b39f JB |
2345 | static void virtnet_update_settings(struct virtnet_info *vi) |
2346 | { | |
2347 | u32 speed; | |
2348 | u8 duplex; | |
2349 | ||
2350 | if (!virtio_has_feature(vi->vdev, VIRTIO_NET_F_SPEED_DUPLEX)) | |
2351 | return; | |
2352 | ||
64ffa39d MT |
2353 | virtio_cread_le(vi->vdev, struct virtio_net_config, speed, &speed); |
2354 | ||
faa9b39f JB |
2355 | if (ethtool_validate_speed(speed)) |
2356 | vi->speed = speed; | |
64ffa39d MT |
2357 | |
2358 | virtio_cread_le(vi->vdev, struct virtio_net_config, duplex, &duplex); | |
2359 | ||
faa9b39f JB |
2360 | if (ethtool_validate_duplex(duplex)) |
2361 | vi->duplex = duplex; | |
2362 | } | |
2363 | ||
0fc0b732 | 2364 | static const struct ethtool_ops virtnet_ethtool_ops = { |
a51e5206 | 2365 | .supported_coalesce_params = ETHTOOL_COALESCE_MAX_FRAMES, |
66846048 | 2366 | .get_drvinfo = virtnet_get_drvinfo, |
9f4d26d0 | 2367 | .get_link = ethtool_op_get_link, |
8f9f4668 | 2368 | .get_ringparam = virtnet_get_ringparam, |
d7dfc5cf TM |
2369 | .get_strings = virtnet_get_strings, |
2370 | .get_sset_count = virtnet_get_sset_count, | |
2371 | .get_ethtool_stats = virtnet_get_ethtool_stats, | |
d73bcd2c JW |
2372 | .set_channels = virtnet_set_channels, |
2373 | .get_channels = virtnet_get_channels, | |
074c3582 | 2374 | .get_ts_info = ethtool_op_get_ts_info, |
ebb6b4b1 PR |
2375 | .get_link_ksettings = virtnet_get_link_ksettings, |
2376 | .set_link_ksettings = virtnet_set_link_ksettings, | |
0c465be1 JW |
2377 | .set_coalesce = virtnet_set_coalesce, |
2378 | .get_coalesce = virtnet_get_coalesce, | |
a9ea3fc6 HX |
2379 | }; |
2380 | ||
9fe7bfce JF |
2381 | static void virtnet_freeze_down(struct virtio_device *vdev) |
2382 | { | |
2383 | struct virtnet_info *vi = vdev->priv; | |
2384 | int i; | |
2385 | ||
2386 | /* Make sure no work handler is accessing the device */ | |
2387 | flush_work(&vi->config_work); | |
2388 | ||
05c998b7 | 2389 | netif_tx_lock_bh(vi->dev); |
9fe7bfce | 2390 | netif_device_detach(vi->dev); |
05c998b7 | 2391 | netif_tx_unlock_bh(vi->dev); |
9fe7bfce JF |
2392 | cancel_delayed_work_sync(&vi->refill); |
2393 | ||
2394 | if (netif_running(vi->dev)) { | |
b92f1e67 | 2395 | for (i = 0; i < vi->max_queue_pairs; i++) { |
9fe7bfce | 2396 | napi_disable(&vi->rq[i].napi); |
78a57b48 | 2397 | virtnet_napi_tx_disable(&vi->sq[i].napi); |
b92f1e67 | 2398 | } |
9fe7bfce JF |
2399 | } |
2400 | } | |
2401 | ||
2402 | static int init_vqs(struct virtnet_info *vi); | |
2403 | ||
2404 | static int virtnet_restore_up(struct virtio_device *vdev) | |
2405 | { | |
2406 | struct virtnet_info *vi = vdev->priv; | |
2407 | int err, i; | |
2408 | ||
2409 | err = init_vqs(vi); | |
2410 | if (err) | |
2411 | return err; | |
2412 | ||
2413 | virtio_device_ready(vdev); | |
2414 | ||
2415 | if (netif_running(vi->dev)) { | |
2416 | for (i = 0; i < vi->curr_queue_pairs; i++) | |
2417 | if (!try_fill_recv(vi, &vi->rq[i], GFP_KERNEL)) | |
2418 | schedule_delayed_work(&vi->refill, 0); | |
2419 | ||
b92f1e67 | 2420 | for (i = 0; i < vi->max_queue_pairs; i++) { |
e4e8452a | 2421 | virtnet_napi_enable(vi->rq[i].vq, &vi->rq[i].napi); |
b92f1e67 WB |
2422 | virtnet_napi_tx_enable(vi, vi->sq[i].vq, |
2423 | &vi->sq[i].napi); | |
2424 | } | |
9fe7bfce JF |
2425 | } |
2426 | ||
05c998b7 | 2427 | netif_tx_lock_bh(vi->dev); |
9fe7bfce | 2428 | netif_device_attach(vi->dev); |
05c998b7 | 2429 | netif_tx_unlock_bh(vi->dev); |
9fe7bfce JF |
2430 | return err; |
2431 | } | |
2432 | ||
3f93522f JW |
2433 | static int virtnet_set_guest_offloads(struct virtnet_info *vi, u64 offloads) |
2434 | { | |
2435 | struct scatterlist sg; | |
12e57169 | 2436 | vi->ctrl->offloads = cpu_to_virtio64(vi->vdev, offloads); |
3f93522f | 2437 | |
12e57169 | 2438 | sg_init_one(&sg, &vi->ctrl->offloads, sizeof(vi->ctrl->offloads)); |
3f93522f JW |
2439 | |
2440 | if (!virtnet_send_command(vi, VIRTIO_NET_CTRL_GUEST_OFFLOADS, | |
2441 | VIRTIO_NET_CTRL_GUEST_OFFLOADS_SET, &sg)) { | |
7934b481 | 2442 | dev_warn(&vi->dev->dev, "Fail to set guest offload.\n"); |
3f93522f JW |
2443 | return -EINVAL; |
2444 | } | |
2445 | ||
2446 | return 0; | |
2447 | } | |
2448 | ||
2449 | static int virtnet_clear_guest_offloads(struct virtnet_info *vi) | |
2450 | { | |
2451 | u64 offloads = 0; | |
2452 | ||
2453 | if (!vi->guest_offloads) | |
2454 | return 0; | |
2455 | ||
3f93522f JW |
2456 | return virtnet_set_guest_offloads(vi, offloads); |
2457 | } | |
2458 | ||
2459 | static int virtnet_restore_guest_offloads(struct virtnet_info *vi) | |
2460 | { | |
2461 | u64 offloads = vi->guest_offloads; | |
2462 | ||
2463 | if (!vi->guest_offloads) | |
2464 | return 0; | |
3f93522f JW |
2465 | |
2466 | return virtnet_set_guest_offloads(vi, offloads); | |
2467 | } | |
2468 | ||
9861ce03 JK |
2469 | static int virtnet_xdp_set(struct net_device *dev, struct bpf_prog *prog, |
2470 | struct netlink_ext_ack *extack) | |
f600b690 JF |
2471 | { |
2472 | unsigned long int max_sz = PAGE_SIZE - sizeof(struct padded_vnet_hdr); | |
2473 | struct virtnet_info *vi = netdev_priv(dev); | |
2474 | struct bpf_prog *old_prog; | |
017b29c3 | 2475 | u16 xdp_qp = 0, curr_qp; |
672aafd5 | 2476 | int i, err; |
f600b690 | 2477 | |
3f93522f JW |
2478 | if (!virtio_has_feature(vi->vdev, VIRTIO_NET_F_CTRL_GUEST_OFFLOADS) |
2479 | && (virtio_has_feature(vi->vdev, VIRTIO_NET_F_GUEST_TSO4) || | |
2480 | virtio_has_feature(vi->vdev, VIRTIO_NET_F_GUEST_TSO6) || | |
2481 | virtio_has_feature(vi->vdev, VIRTIO_NET_F_GUEST_ECN) || | |
18ba58e1 JW |
2482 | virtio_has_feature(vi->vdev, VIRTIO_NET_F_GUEST_UFO) || |
2483 | virtio_has_feature(vi->vdev, VIRTIO_NET_F_GUEST_CSUM))) { | |
2484 | NL_SET_ERR_MSG_MOD(extack, "Can't set XDP while host is implementing LRO/CSUM, disable LRO/CSUM first"); | |
f600b690 JF |
2485 | return -EOPNOTSUPP; |
2486 | } | |
2487 | ||
2488 | if (vi->mergeable_rx_bufs && !vi->any_header_sg) { | |
4d463c4d | 2489 | NL_SET_ERR_MSG_MOD(extack, "XDP expects header/data in single page, any_header_sg required"); |
f600b690 JF |
2490 | return -EINVAL; |
2491 | } | |
2492 | ||
2493 | if (dev->mtu > max_sz) { | |
4d463c4d | 2494 | NL_SET_ERR_MSG_MOD(extack, "MTU too large to enable XDP"); |
f600b690 JF |
2495 | netdev_warn(dev, "XDP requires MTU less than %lu\n", max_sz); |
2496 | return -EINVAL; | |
2497 | } | |
2498 | ||
672aafd5 JF |
2499 | curr_qp = vi->curr_queue_pairs - vi->xdp_queue_pairs; |
2500 | if (prog) | |
2501 | xdp_qp = nr_cpu_ids; | |
2502 | ||
2503 | /* XDP requires extra queues for XDP_TX */ | |
2504 | if (curr_qp + xdp_qp > vi->max_queue_pairs) { | |
97c2c69e | 2505 | netdev_warn(dev, "XDP request %i queues but max is %i. XDP_TX and XDP_REDIRECT will operate in a slower locked tx mode.\n", |
672aafd5 | 2506 | curr_qp + xdp_qp, vi->max_queue_pairs); |
97c2c69e | 2507 | xdp_qp = 0; |
672aafd5 JF |
2508 | } |
2509 | ||
03aa6d34 TM |
2510 | old_prog = rtnl_dereference(vi->rq[0].xdp_prog); |
2511 | if (!prog && !old_prog) | |
2512 | return 0; | |
2513 | ||
85192dbf AN |
2514 | if (prog) |
2515 | bpf_prog_add(prog, vi->max_queue_pairs - 1); | |
2de2f7f4 | 2516 | |
4941d472 | 2517 | /* Make sure NAPI is not using any XDP TX queues for RX. */ |
534da5e8 TM |
2518 | if (netif_running(dev)) { |
2519 | for (i = 0; i < vi->max_queue_pairs; i++) { | |
4e09ff53 | 2520 | napi_disable(&vi->rq[i].napi); |
534da5e8 TM |
2521 | virtnet_napi_tx_disable(&vi->sq[i].napi); |
2522 | } | |
2523 | } | |
f600b690 | 2524 | |
03aa6d34 TM |
2525 | if (!prog) { |
2526 | for (i = 0; i < vi->max_queue_pairs; i++) { | |
2527 | rcu_assign_pointer(vi->rq[i].xdp_prog, prog); | |
2528 | if (i == 0) | |
2529 | virtnet_restore_guest_offloads(vi); | |
2530 | } | |
2531 | synchronize_net(); | |
2532 | } | |
f600b690 | 2533 | |
4941d472 JW |
2534 | err = _virtnet_set_queues(vi, curr_qp + xdp_qp); |
2535 | if (err) | |
2536 | goto err; | |
188313c1 | 2537 | netif_set_real_num_rx_queues(dev, curr_qp + xdp_qp); |
4941d472 | 2538 | vi->xdp_queue_pairs = xdp_qp; |
672aafd5 | 2539 | |
03aa6d34 | 2540 | if (prog) { |
97c2c69e | 2541 | vi->xdp_enabled = true; |
03aa6d34 TM |
2542 | for (i = 0; i < vi->max_queue_pairs; i++) { |
2543 | rcu_assign_pointer(vi->rq[i].xdp_prog, prog); | |
2544 | if (i == 0 && !old_prog) | |
3f93522f | 2545 | virtnet_clear_guest_offloads(vi); |
3f93522f | 2546 | } |
97c2c69e XZ |
2547 | } else { |
2548 | vi->xdp_enabled = false; | |
03aa6d34 TM |
2549 | } |
2550 | ||
2551 | for (i = 0; i < vi->max_queue_pairs; i++) { | |
f600b690 JF |
2552 | if (old_prog) |
2553 | bpf_prog_put(old_prog); | |
534da5e8 | 2554 | if (netif_running(dev)) { |
4e09ff53 | 2555 | virtnet_napi_enable(vi->rq[i].vq, &vi->rq[i].napi); |
534da5e8 TM |
2556 | virtnet_napi_tx_enable(vi, vi->sq[i].vq, |
2557 | &vi->sq[i].napi); | |
2558 | } | |
f600b690 JF |
2559 | } |
2560 | ||
2561 | return 0; | |
2de2f7f4 | 2562 | |
4941d472 | 2563 | err: |
03aa6d34 TM |
2564 | if (!prog) { |
2565 | virtnet_clear_guest_offloads(vi); | |
2566 | for (i = 0; i < vi->max_queue_pairs; i++) | |
2567 | rcu_assign_pointer(vi->rq[i].xdp_prog, old_prog); | |
2568 | } | |
2569 | ||
8be4d9a4 | 2570 | if (netif_running(dev)) { |
534da5e8 | 2571 | for (i = 0; i < vi->max_queue_pairs; i++) { |
8be4d9a4 | 2572 | virtnet_napi_enable(vi->rq[i].vq, &vi->rq[i].napi); |
534da5e8 TM |
2573 | virtnet_napi_tx_enable(vi, vi->sq[i].vq, |
2574 | &vi->sq[i].napi); | |
2575 | } | |
8be4d9a4 | 2576 | } |
2de2f7f4 JF |
2577 | if (prog) |
2578 | bpf_prog_sub(prog, vi->max_queue_pairs - 1); | |
2579 | return err; | |
f600b690 JF |
2580 | } |
2581 | ||
f4e63525 | 2582 | static int virtnet_xdp(struct net_device *dev, struct netdev_bpf *xdp) |
f600b690 JF |
2583 | { |
2584 | switch (xdp->command) { | |
2585 | case XDP_SETUP_PROG: | |
9861ce03 | 2586 | return virtnet_xdp_set(dev, xdp->prog, xdp->extack); |
f600b690 JF |
2587 | default: |
2588 | return -EINVAL; | |
2589 | } | |
2590 | } | |
2591 | ||
ba5e4426 SS |
2592 | static int virtnet_get_phys_port_name(struct net_device *dev, char *buf, |
2593 | size_t len) | |
2594 | { | |
2595 | struct virtnet_info *vi = netdev_priv(dev); | |
2596 | int ret; | |
2597 | ||
2598 | if (!virtio_has_feature(vi->vdev, VIRTIO_NET_F_STANDBY)) | |
2599 | return -EOPNOTSUPP; | |
2600 | ||
2601 | ret = snprintf(buf, len, "sby"); | |
2602 | if (ret >= len) | |
2603 | return -EOPNOTSUPP; | |
2604 | ||
2605 | return 0; | |
2606 | } | |
2607 | ||
a02e8964 WB |
2608 | static int virtnet_set_features(struct net_device *dev, |
2609 | netdev_features_t features) | |
2610 | { | |
2611 | struct virtnet_info *vi = netdev_priv(dev); | |
cf8691cb | 2612 | u64 offloads; |
a02e8964 WB |
2613 | int err; |
2614 | ||
3618ad2a | 2615 | if ((dev->features ^ features) & NETIF_F_LRO) { |
97c2c69e | 2616 | if (vi->xdp_enabled) |
cf8691cb MT |
2617 | return -EBUSY; |
2618 | ||
a02e8964 | 2619 | if (features & NETIF_F_LRO) |
cf8691cb | 2620 | offloads = vi->guest_offloads_capable; |
a02e8964 | 2621 | else |
cf8691cb MT |
2622 | offloads = vi->guest_offloads_capable & |
2623 | ~GUEST_OFFLOAD_LRO_MASK; | |
a02e8964 | 2624 | |
cf8691cb MT |
2625 | err = virtnet_set_guest_offloads(vi, offloads); |
2626 | if (err) | |
2627 | return err; | |
2628 | vi->guest_offloads = offloads; | |
a02e8964 WB |
2629 | } |
2630 | ||
2631 | return 0; | |
2632 | } | |
2633 | ||
76288b4e SH |
2634 | static const struct net_device_ops virtnet_netdev = { |
2635 | .ndo_open = virtnet_open, | |
2636 | .ndo_stop = virtnet_close, | |
2637 | .ndo_start_xmit = start_xmit, | |
2638 | .ndo_validate_addr = eth_validate_addr, | |
9c46f6d4 | 2639 | .ndo_set_mac_address = virtnet_set_mac_address, |
2af7698e | 2640 | .ndo_set_rx_mode = virtnet_set_rx_mode, |
3fa2a1df | 2641 | .ndo_get_stats64 = virtnet_stats, |
1824a989 AW |
2642 | .ndo_vlan_rx_add_vid = virtnet_vlan_rx_add_vid, |
2643 | .ndo_vlan_rx_kill_vid = virtnet_vlan_rx_kill_vid, | |
f4e63525 | 2644 | .ndo_bpf = virtnet_xdp, |
186b3c99 | 2645 | .ndo_xdp_xmit = virtnet_xdp_xmit, |
2836b4f2 | 2646 | .ndo_features_check = passthru_features_check, |
ba5e4426 | 2647 | .ndo_get_phys_port_name = virtnet_get_phys_port_name, |
a02e8964 | 2648 | .ndo_set_features = virtnet_set_features, |
76288b4e SH |
2649 | }; |
2650 | ||
586d17c5 | 2651 | static void virtnet_config_changed_work(struct work_struct *work) |
9f4d26d0 | 2652 | { |
586d17c5 JW |
2653 | struct virtnet_info *vi = |
2654 | container_of(work, struct virtnet_info, config_work); | |
9f4d26d0 MM |
2655 | u16 v; |
2656 | ||
855e0c52 RR |
2657 | if (virtio_cread_feature(vi->vdev, VIRTIO_NET_F_STATUS, |
2658 | struct virtio_net_config, status, &v) < 0) | |
507613bf | 2659 | return; |
586d17c5 JW |
2660 | |
2661 | if (v & VIRTIO_NET_S_ANNOUNCE) { | |
ee89bab1 | 2662 | netdev_notify_peers(vi->dev); |
586d17c5 JW |
2663 | virtnet_ack_link_announce(vi); |
2664 | } | |
9f4d26d0 MM |
2665 | |
2666 | /* Ignore unknown (future) status bits */ | |
2667 | v &= VIRTIO_NET_S_LINK_UP; | |
2668 | ||
2669 | if (vi->status == v) | |
507613bf | 2670 | return; |
9f4d26d0 MM |
2671 | |
2672 | vi->status = v; | |
2673 | ||
2674 | if (vi->status & VIRTIO_NET_S_LINK_UP) { | |
faa9b39f | 2675 | virtnet_update_settings(vi); |
9f4d26d0 | 2676 | netif_carrier_on(vi->dev); |
986a4f4d | 2677 | netif_tx_wake_all_queues(vi->dev); |
9f4d26d0 MM |
2678 | } else { |
2679 | netif_carrier_off(vi->dev); | |
986a4f4d | 2680 | netif_tx_stop_all_queues(vi->dev); |
9f4d26d0 MM |
2681 | } |
2682 | } | |
2683 | ||
2684 | static void virtnet_config_changed(struct virtio_device *vdev) | |
2685 | { | |
2686 | struct virtnet_info *vi = vdev->priv; | |
2687 | ||
3b07e9ca | 2688 | schedule_work(&vi->config_work); |
9f4d26d0 MM |
2689 | } |
2690 | ||
986a4f4d JW |
2691 | static void virtnet_free_queues(struct virtnet_info *vi) |
2692 | { | |
d4fb84ee AV |
2693 | int i; |
2694 | ||
ab3971b1 | 2695 | for (i = 0; i < vi->max_queue_pairs; i++) { |
5198d545 JK |
2696 | __netif_napi_del(&vi->rq[i].napi); |
2697 | __netif_napi_del(&vi->sq[i].napi); | |
ab3971b1 | 2698 | } |
d4fb84ee | 2699 | |
5198d545 | 2700 | /* We called __netif_napi_del(), |
963abe5c ED |
2701 | * we need to respect an RCU grace period before freeing vi->rq |
2702 | */ | |
2703 | synchronize_net(); | |
2704 | ||
986a4f4d JW |
2705 | kfree(vi->rq); |
2706 | kfree(vi->sq); | |
12e57169 | 2707 | kfree(vi->ctrl); |
986a4f4d JW |
2708 | } |
2709 | ||
47315329 | 2710 | static void _free_receive_bufs(struct virtnet_info *vi) |
986a4f4d | 2711 | { |
f600b690 | 2712 | struct bpf_prog *old_prog; |
986a4f4d JW |
2713 | int i; |
2714 | ||
2715 | for (i = 0; i < vi->max_queue_pairs; i++) { | |
2716 | while (vi->rq[i].pages) | |
2717 | __free_pages(get_a_page(&vi->rq[i], GFP_KERNEL), 0); | |
f600b690 JF |
2718 | |
2719 | old_prog = rtnl_dereference(vi->rq[i].xdp_prog); | |
2720 | RCU_INIT_POINTER(vi->rq[i].xdp_prog, NULL); | |
2721 | if (old_prog) | |
2722 | bpf_prog_put(old_prog); | |
986a4f4d | 2723 | } |
47315329 JF |
2724 | } |
2725 | ||
2726 | static void free_receive_bufs(struct virtnet_info *vi) | |
2727 | { | |
2728 | rtnl_lock(); | |
2729 | _free_receive_bufs(vi); | |
f600b690 | 2730 | rtnl_unlock(); |
986a4f4d JW |
2731 | } |
2732 | ||
fb51879d MD |
2733 | static void free_receive_page_frags(struct virtnet_info *vi) |
2734 | { | |
2735 | int i; | |
2736 | for (i = 0; i < vi->max_queue_pairs; i++) | |
2737 | if (vi->rq[i].alloc_frag.page) | |
2738 | put_page(vi->rq[i].alloc_frag.page); | |
2739 | } | |
2740 | ||
986a4f4d JW |
2741 | static void free_unused_bufs(struct virtnet_info *vi) |
2742 | { | |
2743 | void *buf; | |
2744 | int i; | |
2745 | ||
2746 | for (i = 0; i < vi->max_queue_pairs; i++) { | |
2747 | struct virtqueue *vq = vi->sq[i].vq; | |
56434a01 | 2748 | while ((buf = virtqueue_detach_unused_buf(vq)) != NULL) { |
5050471d | 2749 | if (!is_xdp_frame(buf)) |
56434a01 JF |
2750 | dev_kfree_skb(buf); |
2751 | else | |
5050471d | 2752 | xdp_return_frame(ptr_to_xdp(buf)); |
56434a01 | 2753 | } |
986a4f4d JW |
2754 | } |
2755 | ||
2756 | for (i = 0; i < vi->max_queue_pairs; i++) { | |
2757 | struct virtqueue *vq = vi->rq[i].vq; | |
2758 | ||
2759 | while ((buf = virtqueue_detach_unused_buf(vq)) != NULL) { | |
ab7db917 | 2760 | if (vi->mergeable_rx_bufs) { |
680557cf | 2761 | put_page(virt_to_head_page(buf)); |
ab7db917 | 2762 | } else if (vi->big_packets) { |
fa9fac17 | 2763 | give_pages(&vi->rq[i], buf); |
ab7db917 | 2764 | } else { |
f6b10209 | 2765 | put_page(virt_to_head_page(buf)); |
ab7db917 | 2766 | } |
986a4f4d | 2767 | } |
986a4f4d JW |
2768 | } |
2769 | } | |
2770 | ||
e9d7417b JW |
2771 | static void virtnet_del_vqs(struct virtnet_info *vi) |
2772 | { | |
2773 | struct virtio_device *vdev = vi->vdev; | |
2774 | ||
310974fa | 2775 | virtnet_clean_affinity(vi); |
986a4f4d | 2776 | |
e9d7417b | 2777 | vdev->config->del_vqs(vdev); |
986a4f4d JW |
2778 | |
2779 | virtnet_free_queues(vi); | |
e9d7417b JW |
2780 | } |
2781 | ||
d85b758f MT |
2782 | /* How large should a single buffer be so a queue full of these can fit at |
2783 | * least one full packet? | |
2784 | * Logic below assumes the mergeable buffer header is used. | |
2785 | */ | |
2786 | static unsigned int mergeable_min_buf_len(struct virtnet_info *vi, struct virtqueue *vq) | |
2787 | { | |
2788 | const unsigned int hdr_len = sizeof(struct virtio_net_hdr_mrg_rxbuf); | |
2789 | unsigned int rq_size = virtqueue_get_vring_size(vq); | |
2790 | unsigned int packet_len = vi->big_packets ? IP_MAX_MTU : vi->dev->max_mtu; | |
2791 | unsigned int buf_len = hdr_len + ETH_HLEN + VLAN_HLEN + packet_len; | |
2792 | unsigned int min_buf_len = DIV_ROUND_UP(buf_len, rq_size); | |
2793 | ||
f0c3192c MT |
2794 | return max(max(min_buf_len, hdr_len) - hdr_len, |
2795 | (unsigned int)GOOD_PACKET_LEN); | |
d85b758f MT |
2796 | } |
2797 | ||
986a4f4d | 2798 | static int virtnet_find_vqs(struct virtnet_info *vi) |
3f9c10b0 | 2799 | { |
986a4f4d JW |
2800 | vq_callback_t **callbacks; |
2801 | struct virtqueue **vqs; | |
2802 | int ret = -ENOMEM; | |
2803 | int i, total_vqs; | |
2804 | const char **names; | |
d45b897b | 2805 | bool *ctx; |
986a4f4d JW |
2806 | |
2807 | /* We expect 1 RX virtqueue followed by 1 TX virtqueue, followed by | |
2808 | * possible N-1 RX/TX queue pairs used in multiqueue mode, followed by | |
2809 | * possible control vq. | |
2810 | */ | |
2811 | total_vqs = vi->max_queue_pairs * 2 + | |
2812 | virtio_has_feature(vi->vdev, VIRTIO_NET_F_CTRL_VQ); | |
2813 | ||
2814 | /* Allocate space for find_vqs parameters */ | |
6396bb22 | 2815 | vqs = kcalloc(total_vqs, sizeof(*vqs), GFP_KERNEL); |
986a4f4d JW |
2816 | if (!vqs) |
2817 | goto err_vq; | |
6da2ec56 | 2818 | callbacks = kmalloc_array(total_vqs, sizeof(*callbacks), GFP_KERNEL); |
986a4f4d JW |
2819 | if (!callbacks) |
2820 | goto err_callback; | |
6da2ec56 | 2821 | names = kmalloc_array(total_vqs, sizeof(*names), GFP_KERNEL); |
986a4f4d JW |
2822 | if (!names) |
2823 | goto err_names; | |
192f68cf | 2824 | if (!vi->big_packets || vi->mergeable_rx_bufs) { |
6396bb22 | 2825 | ctx = kcalloc(total_vqs, sizeof(*ctx), GFP_KERNEL); |
d45b897b MT |
2826 | if (!ctx) |
2827 | goto err_ctx; | |
2828 | } else { | |
2829 | ctx = NULL; | |
2830 | } | |
986a4f4d JW |
2831 | |
2832 | /* Parameters for control virtqueue, if any */ | |
2833 | if (vi->has_cvq) { | |
2834 | callbacks[total_vqs - 1] = NULL; | |
2835 | names[total_vqs - 1] = "control"; | |
2836 | } | |
3f9c10b0 | 2837 | |
986a4f4d JW |
2838 | /* Allocate/initialize parameters for send/receive virtqueues */ |
2839 | for (i = 0; i < vi->max_queue_pairs; i++) { | |
2840 | callbacks[rxq2vq(i)] = skb_recv_done; | |
2841 | callbacks[txq2vq(i)] = skb_xmit_done; | |
2842 | sprintf(vi->rq[i].name, "input.%d", i); | |
2843 | sprintf(vi->sq[i].name, "output.%d", i); | |
2844 | names[rxq2vq(i)] = vi->rq[i].name; | |
2845 | names[txq2vq(i)] = vi->sq[i].name; | |
d45b897b MT |
2846 | if (ctx) |
2847 | ctx[rxq2vq(i)] = true; | |
986a4f4d | 2848 | } |
3f9c10b0 | 2849 | |
a2f7dc00 XT |
2850 | ret = virtio_find_vqs_ctx(vi->vdev, total_vqs, vqs, callbacks, |
2851 | names, ctx, NULL); | |
986a4f4d JW |
2852 | if (ret) |
2853 | goto err_find; | |
3f9c10b0 | 2854 | |
986a4f4d JW |
2855 | if (vi->has_cvq) { |
2856 | vi->cvq = vqs[total_vqs - 1]; | |
3f9c10b0 | 2857 | if (virtio_has_feature(vi->vdev, VIRTIO_NET_F_CTRL_VLAN)) |
f646968f | 2858 | vi->dev->features |= NETIF_F_HW_VLAN_CTAG_FILTER; |
3f9c10b0 | 2859 | } |
986a4f4d JW |
2860 | |
2861 | for (i = 0; i < vi->max_queue_pairs; i++) { | |
2862 | vi->rq[i].vq = vqs[rxq2vq(i)]; | |
d85b758f | 2863 | vi->rq[i].min_buf_len = mergeable_min_buf_len(vi, vi->rq[i].vq); |
986a4f4d JW |
2864 | vi->sq[i].vq = vqs[txq2vq(i)]; |
2865 | } | |
2866 | ||
2fa3c8a8 | 2867 | /* run here: ret == 0. */ |
986a4f4d | 2868 | |
986a4f4d JW |
2869 | |
2870 | err_find: | |
d45b897b MT |
2871 | kfree(ctx); |
2872 | err_ctx: | |
986a4f4d JW |
2873 | kfree(names); |
2874 | err_names: | |
2875 | kfree(callbacks); | |
2876 | err_callback: | |
2877 | kfree(vqs); | |
2878 | err_vq: | |
2879 | return ret; | |
2880 | } | |
2881 | ||
2882 | static int virtnet_alloc_queues(struct virtnet_info *vi) | |
2883 | { | |
2884 | int i; | |
2885 | ||
122b84a1 MG |
2886 | if (vi->has_cvq) { |
2887 | vi->ctrl = kzalloc(sizeof(*vi->ctrl), GFP_KERNEL); | |
2888 | if (!vi->ctrl) | |
2889 | goto err_ctrl; | |
2890 | } else { | |
2891 | vi->ctrl = NULL; | |
2892 | } | |
6396bb22 | 2893 | vi->sq = kcalloc(vi->max_queue_pairs, sizeof(*vi->sq), GFP_KERNEL); |
986a4f4d JW |
2894 | if (!vi->sq) |
2895 | goto err_sq; | |
6396bb22 | 2896 | vi->rq = kcalloc(vi->max_queue_pairs, sizeof(*vi->rq), GFP_KERNEL); |
008d4278 | 2897 | if (!vi->rq) |
986a4f4d JW |
2898 | goto err_rq; |
2899 | ||
2900 | INIT_DELAYED_WORK(&vi->refill, refill_work); | |
2901 | for (i = 0; i < vi->max_queue_pairs; i++) { | |
2902 | vi->rq[i].pages = NULL; | |
2903 | netif_napi_add(vi->dev, &vi->rq[i].napi, virtnet_poll, | |
2904 | napi_weight); | |
1d11e732 WB |
2905 | netif_tx_napi_add(vi->dev, &vi->sq[i].napi, virtnet_poll_tx, |
2906 | napi_tx ? napi_weight : 0); | |
986a4f4d JW |
2907 | |
2908 | sg_init_table(vi->rq[i].sg, ARRAY_SIZE(vi->rq[i].sg)); | |
5377d758 | 2909 | ewma_pkt_len_init(&vi->rq[i].mrg_avg_pkt_len); |
986a4f4d | 2910 | sg_init_table(vi->sq[i].sg, ARRAY_SIZE(vi->sq[i].sg)); |
d7dfc5cf TM |
2911 | |
2912 | u64_stats_init(&vi->rq[i].stats.syncp); | |
2913 | u64_stats_init(&vi->sq[i].stats.syncp); | |
986a4f4d JW |
2914 | } |
2915 | ||
2916 | return 0; | |
2917 | ||
2918 | err_rq: | |
2919 | kfree(vi->sq); | |
2920 | err_sq: | |
12e57169 MT |
2921 | kfree(vi->ctrl); |
2922 | err_ctrl: | |
986a4f4d JW |
2923 | return -ENOMEM; |
2924 | } | |
2925 | ||
2926 | static int init_vqs(struct virtnet_info *vi) | |
2927 | { | |
2928 | int ret; | |
2929 | ||
2930 | /* Allocate send & receive queues */ | |
2931 | ret = virtnet_alloc_queues(vi); | |
2932 | if (ret) | |
2933 | goto err; | |
2934 | ||
2935 | ret = virtnet_find_vqs(vi); | |
2936 | if (ret) | |
2937 | goto err_free; | |
2938 | ||
47be2479 | 2939 | get_online_cpus(); |
8898c21c | 2940 | virtnet_set_affinity(vi); |
47be2479 WG |
2941 | put_online_cpus(); |
2942 | ||
986a4f4d JW |
2943 | return 0; |
2944 | ||
2945 | err_free: | |
2946 | virtnet_free_queues(vi); | |
2947 | err: | |
2948 | return ret; | |
3f9c10b0 AS |
2949 | } |
2950 | ||
fbf28d78 MD |
2951 | #ifdef CONFIG_SYSFS |
2952 | static ssize_t mergeable_rx_buffer_size_show(struct netdev_rx_queue *queue, | |
718ad681 | 2953 | char *buf) |
fbf28d78 MD |
2954 | { |
2955 | struct virtnet_info *vi = netdev_priv(queue->dev); | |
2956 | unsigned int queue_index = get_netdev_rx_queue_index(queue); | |
3cc81a9a JW |
2957 | unsigned int headroom = virtnet_get_headroom(vi); |
2958 | unsigned int tailroom = headroom ? sizeof(struct skb_shared_info) : 0; | |
5377d758 | 2959 | struct ewma_pkt_len *avg; |
fbf28d78 MD |
2960 | |
2961 | BUG_ON(queue_index >= vi->max_queue_pairs); | |
2962 | avg = &vi->rq[queue_index].mrg_avg_pkt_len; | |
d85b758f | 2963 | return sprintf(buf, "%u\n", |
3cc81a9a JW |
2964 | get_mergeable_buf_len(&vi->rq[queue_index], avg, |
2965 | SKB_DATA_ALIGN(headroom + tailroom))); | |
fbf28d78 MD |
2966 | } |
2967 | ||
2968 | static struct rx_queue_attribute mergeable_rx_buffer_size_attribute = | |
2969 | __ATTR_RO(mergeable_rx_buffer_size); | |
2970 | ||
2971 | static struct attribute *virtio_net_mrg_rx_attrs[] = { | |
2972 | &mergeable_rx_buffer_size_attribute.attr, | |
2973 | NULL | |
2974 | }; | |
2975 | ||
2976 | static const struct attribute_group virtio_net_mrg_rx_group = { | |
2977 | .name = "virtio_net", | |
2978 | .attrs = virtio_net_mrg_rx_attrs | |
2979 | }; | |
2980 | #endif | |
2981 | ||
892d6eb1 JW |
2982 | static bool virtnet_fail_on_feature(struct virtio_device *vdev, |
2983 | unsigned int fbit, | |
2984 | const char *fname, const char *dname) | |
2985 | { | |
2986 | if (!virtio_has_feature(vdev, fbit)) | |
2987 | return false; | |
2988 | ||
2989 | dev_err(&vdev->dev, "device advertises feature %s but not %s", | |
2990 | fname, dname); | |
2991 | ||
2992 | return true; | |
2993 | } | |
2994 | ||
2995 | #define VIRTNET_FAIL_ON(vdev, fbit, dbit) \ | |
2996 | virtnet_fail_on_feature(vdev, fbit, #fbit, dbit) | |
2997 | ||
2998 | static bool virtnet_validate_features(struct virtio_device *vdev) | |
2999 | { | |
3000 | if (!virtio_has_feature(vdev, VIRTIO_NET_F_CTRL_VQ) && | |
3001 | (VIRTNET_FAIL_ON(vdev, VIRTIO_NET_F_CTRL_RX, | |
3002 | "VIRTIO_NET_F_CTRL_VQ") || | |
3003 | VIRTNET_FAIL_ON(vdev, VIRTIO_NET_F_CTRL_VLAN, | |
3004 | "VIRTIO_NET_F_CTRL_VQ") || | |
3005 | VIRTNET_FAIL_ON(vdev, VIRTIO_NET_F_GUEST_ANNOUNCE, | |
3006 | "VIRTIO_NET_F_CTRL_VQ") || | |
3007 | VIRTNET_FAIL_ON(vdev, VIRTIO_NET_F_MQ, "VIRTIO_NET_F_CTRL_VQ") || | |
3008 | VIRTNET_FAIL_ON(vdev, VIRTIO_NET_F_CTRL_MAC_ADDR, | |
3009 | "VIRTIO_NET_F_CTRL_VQ"))) { | |
3010 | return false; | |
3011 | } | |
3012 | ||
3013 | return true; | |
3014 | } | |
3015 | ||
d0c2c997 JW |
3016 | #define MIN_MTU ETH_MIN_MTU |
3017 | #define MAX_MTU ETH_MAX_MTU | |
3018 | ||
fe36cbe0 | 3019 | static int virtnet_validate(struct virtio_device *vdev) |
296f96fc | 3020 | { |
6ba42248 MT |
3021 | if (!vdev->config->get) { |
3022 | dev_err(&vdev->dev, "%s failure: config access disabled\n", | |
3023 | __func__); | |
3024 | return -EINVAL; | |
3025 | } | |
3026 | ||
892d6eb1 JW |
3027 | if (!virtnet_validate_features(vdev)) |
3028 | return -EINVAL; | |
3029 | ||
fe36cbe0 MT |
3030 | if (virtio_has_feature(vdev, VIRTIO_NET_F_MTU)) { |
3031 | int mtu = virtio_cread16(vdev, | |
3032 | offsetof(struct virtio_net_config, | |
3033 | mtu)); | |
3034 | if (mtu < MIN_MTU) | |
3035 | __virtio_clear_bit(vdev, VIRTIO_NET_F_MTU); | |
3036 | } | |
3037 | ||
3038 | return 0; | |
3039 | } | |
3040 | ||
3041 | static int virtnet_probe(struct virtio_device *vdev) | |
3042 | { | |
d7dfc5cf | 3043 | int i, err = -ENOMEM; |
fe36cbe0 MT |
3044 | struct net_device *dev; |
3045 | struct virtnet_info *vi; | |
3046 | u16 max_queue_pairs; | |
3047 | int mtu; | |
3048 | ||
986a4f4d | 3049 | /* Find if host supports multiqueue virtio_net device */ |
855e0c52 RR |
3050 | err = virtio_cread_feature(vdev, VIRTIO_NET_F_MQ, |
3051 | struct virtio_net_config, | |
3052 | max_virtqueue_pairs, &max_queue_pairs); | |
986a4f4d JW |
3053 | |
3054 | /* We need at least 2 queue's */ | |
3055 | if (err || max_queue_pairs < VIRTIO_NET_CTRL_MQ_VQ_PAIRS_MIN || | |
3056 | max_queue_pairs > VIRTIO_NET_CTRL_MQ_VQ_PAIRS_MAX || | |
3057 | !virtio_has_feature(vdev, VIRTIO_NET_F_CTRL_VQ)) | |
3058 | max_queue_pairs = 1; | |
296f96fc RR |
3059 | |
3060 | /* Allocate ourselves a network device with room for our info */ | |
986a4f4d | 3061 | dev = alloc_etherdev_mq(sizeof(struct virtnet_info), max_queue_pairs); |
296f96fc RR |
3062 | if (!dev) |
3063 | return -ENOMEM; | |
3064 | ||
3065 | /* Set up network device as normal. */ | |
ab5bd583 XZ |
3066 | dev->priv_flags |= IFF_UNICAST_FLT | IFF_LIVE_ADDR_CHANGE | |
3067 | IFF_TX_SKB_NO_LINEAR; | |
76288b4e | 3068 | dev->netdev_ops = &virtnet_netdev; |
296f96fc | 3069 | dev->features = NETIF_F_HIGHDMA; |
3fa2a1df | 3070 | |
7ad24ea4 | 3071 | dev->ethtool_ops = &virtnet_ethtool_ops; |
296f96fc RR |
3072 | SET_NETDEV_DEV(dev, &vdev->dev); |
3073 | ||
3074 | /* Do we support "hardware" checksums? */ | |
98e778c9 | 3075 | if (virtio_has_feature(vdev, VIRTIO_NET_F_CSUM)) { |
296f96fc | 3076 | /* This opens up the world of extra features. */ |
48900cb6 | 3077 | dev->hw_features |= NETIF_F_HW_CSUM | NETIF_F_SG; |
98e778c9 | 3078 | if (csum) |
48900cb6 | 3079 | dev->features |= NETIF_F_HW_CSUM | NETIF_F_SG; |
98e778c9 MM |
3080 | |
3081 | if (virtio_has_feature(vdev, VIRTIO_NET_F_GSO)) { | |
e078de03 | 3082 | dev->hw_features |= NETIF_F_TSO |
34a48579 RR |
3083 | | NETIF_F_TSO_ECN | NETIF_F_TSO6; |
3084 | } | |
5539ae96 | 3085 | /* Individual feature bits: what can host handle? */ |
98e778c9 MM |
3086 | if (virtio_has_feature(vdev, VIRTIO_NET_F_HOST_TSO4)) |
3087 | dev->hw_features |= NETIF_F_TSO; | |
3088 | if (virtio_has_feature(vdev, VIRTIO_NET_F_HOST_TSO6)) | |
3089 | dev->hw_features |= NETIF_F_TSO6; | |
3090 | if (virtio_has_feature(vdev, VIRTIO_NET_F_HOST_ECN)) | |
3091 | dev->hw_features |= NETIF_F_TSO_ECN; | |
98e778c9 | 3092 | |
41f2f127 JW |
3093 | dev->features |= NETIF_F_GSO_ROBUST; |
3094 | ||
98e778c9 | 3095 | if (gso) |
e078de03 | 3096 | dev->features |= dev->hw_features & NETIF_F_ALL_TSO; |
98e778c9 | 3097 | /* (!csum && gso) case will be fixed by register_netdev() */ |
296f96fc | 3098 | } |
4f49129b TH |
3099 | if (virtio_has_feature(vdev, VIRTIO_NET_F_GUEST_CSUM)) |
3100 | dev->features |= NETIF_F_RXCSUM; | |
a02e8964 WB |
3101 | if (virtio_has_feature(vdev, VIRTIO_NET_F_GUEST_TSO4) || |
3102 | virtio_has_feature(vdev, VIRTIO_NET_F_GUEST_TSO6)) | |
3103 | dev->features |= NETIF_F_LRO; | |
cf8691cb | 3104 | if (virtio_has_feature(vdev, VIRTIO_NET_F_CTRL_GUEST_OFFLOADS)) |
a02e8964 | 3105 | dev->hw_features |= NETIF_F_LRO; |
296f96fc | 3106 | |
4fda8302 JW |
3107 | dev->vlan_features = dev->features; |
3108 | ||
d0c2c997 JW |
3109 | /* MTU range: 68 - 65535 */ |
3110 | dev->min_mtu = MIN_MTU; | |
3111 | dev->max_mtu = MAX_MTU; | |
3112 | ||
296f96fc | 3113 | /* Configuration may specify what MAC to use. Otherwise random. */ |
855e0c52 RR |
3114 | if (virtio_has_feature(vdev, VIRTIO_NET_F_MAC)) |
3115 | virtio_cread_bytes(vdev, | |
3116 | offsetof(struct virtio_net_config, mac), | |
3117 | dev->dev_addr, dev->addr_len); | |
3118 | else | |
f2cedb63 | 3119 | eth_hw_addr_random(dev); |
296f96fc RR |
3120 | |
3121 | /* Set up our device-specific information */ | |
3122 | vi = netdev_priv(dev); | |
296f96fc RR |
3123 | vi->dev = dev; |
3124 | vi->vdev = vdev; | |
d9d5dcc8 | 3125 | vdev->priv = vi; |
827da44c | 3126 | |
586d17c5 | 3127 | INIT_WORK(&vi->config_work, virtnet_config_changed_work); |
296f96fc | 3128 | |
97402b96 | 3129 | /* If we can receive ANY GSO packets, we must allocate large ones. */ |
8e95a202 JP |
3130 | if (virtio_has_feature(vdev, VIRTIO_NET_F_GUEST_TSO4) || |
3131 | virtio_has_feature(vdev, VIRTIO_NET_F_GUEST_TSO6) || | |
e3e3c423 VY |
3132 | virtio_has_feature(vdev, VIRTIO_NET_F_GUEST_ECN) || |
3133 | virtio_has_feature(vdev, VIRTIO_NET_F_GUEST_UFO)) | |
97402b96 HX |
3134 | vi->big_packets = true; |
3135 | ||
3f2c31d9 MM |
3136 | if (virtio_has_feature(vdev, VIRTIO_NET_F_MRG_RXBUF)) |
3137 | vi->mergeable_rx_bufs = true; | |
3138 | ||
d04302b3 MT |
3139 | if (virtio_has_feature(vdev, VIRTIO_NET_F_MRG_RXBUF) || |
3140 | virtio_has_feature(vdev, VIRTIO_F_VERSION_1)) | |
012873d0 MT |
3141 | vi->hdr_len = sizeof(struct virtio_net_hdr_mrg_rxbuf); |
3142 | else | |
3143 | vi->hdr_len = sizeof(struct virtio_net_hdr); | |
3144 | ||
75993300 MT |
3145 | if (virtio_has_feature(vdev, VIRTIO_F_ANY_LAYOUT) || |
3146 | virtio_has_feature(vdev, VIRTIO_F_VERSION_1)) | |
e7428e95 MT |
3147 | vi->any_header_sg = true; |
3148 | ||
986a4f4d JW |
3149 | if (virtio_has_feature(vdev, VIRTIO_NET_F_CTRL_VQ)) |
3150 | vi->has_cvq = true; | |
3151 | ||
14de9d11 AC |
3152 | if (virtio_has_feature(vdev, VIRTIO_NET_F_MTU)) { |
3153 | mtu = virtio_cread16(vdev, | |
3154 | offsetof(struct virtio_net_config, | |
3155 | mtu)); | |
93a205ee | 3156 | if (mtu < dev->min_mtu) { |
fe36cbe0 MT |
3157 | /* Should never trigger: MTU was previously validated |
3158 | * in virtnet_validate. | |
3159 | */ | |
7934b481 YS |
3160 | dev_err(&vdev->dev, |
3161 | "device MTU appears to have changed it is now %d < %d", | |
3162 | mtu, dev->min_mtu); | |
411ea23a | 3163 | err = -EINVAL; |
d7dfc5cf | 3164 | goto free; |
93a205ee | 3165 | } |
2e123b44 | 3166 | |
fe36cbe0 MT |
3167 | dev->mtu = mtu; |
3168 | dev->max_mtu = mtu; | |
3169 | ||
2e123b44 MT |
3170 | /* TODO: size buffers correctly in this case. */ |
3171 | if (dev->mtu > ETH_DATA_LEN) | |
3172 | vi->big_packets = true; | |
14de9d11 AC |
3173 | } |
3174 | ||
012873d0 MT |
3175 | if (vi->any_header_sg) |
3176 | dev->needed_headroom = vi->hdr_len; | |
6ebbc1a6 | 3177 | |
44900010 JW |
3178 | /* Enable multiqueue by default */ |
3179 | if (num_online_cpus() >= max_queue_pairs) | |
3180 | vi->curr_queue_pairs = max_queue_pairs; | |
3181 | else | |
3182 | vi->curr_queue_pairs = num_online_cpus(); | |
986a4f4d JW |
3183 | vi->max_queue_pairs = max_queue_pairs; |
3184 | ||
3185 | /* Allocate/initialize the rx/tx queues, and invoke find_vqs */ | |
3f9c10b0 | 3186 | err = init_vqs(vi); |
d2a7ddda | 3187 | if (err) |
d7dfc5cf | 3188 | goto free; |
296f96fc | 3189 | |
fbf28d78 MD |
3190 | #ifdef CONFIG_SYSFS |
3191 | if (vi->mergeable_rx_bufs) | |
3192 | dev->sysfs_rx_queue_group = &virtio_net_mrg_rx_group; | |
3193 | #endif | |
0f13b66b ZYW |
3194 | netif_set_real_num_tx_queues(dev, vi->curr_queue_pairs); |
3195 | netif_set_real_num_rx_queues(dev, vi->curr_queue_pairs); | |
986a4f4d | 3196 | |
16032be5 NA |
3197 | virtnet_init_settings(dev); |
3198 | ||
ba5e4426 SS |
3199 | if (virtio_has_feature(vdev, VIRTIO_NET_F_STANDBY)) { |
3200 | vi->failover = net_failover_create(vi->dev); | |
4b8e6ac4 WY |
3201 | if (IS_ERR(vi->failover)) { |
3202 | err = PTR_ERR(vi->failover); | |
ba5e4426 | 3203 | goto free_vqs; |
4b8e6ac4 | 3204 | } |
ba5e4426 SS |
3205 | } |
3206 | ||
296f96fc RR |
3207 | err = register_netdev(dev); |
3208 | if (err) { | |
3209 | pr_debug("virtio_net: registering device failed\n"); | |
ba5e4426 | 3210 | goto free_failover; |
296f96fc | 3211 | } |
b3369c1f | 3212 | |
4baf1e33 MT |
3213 | virtio_device_ready(vdev); |
3214 | ||
8017c279 | 3215 | err = virtnet_cpu_notif_add(vi); |
8de4b2f3 WG |
3216 | if (err) { |
3217 | pr_debug("virtio_net: registering cpu notifier failed\n"); | |
f00e35e2 | 3218 | goto free_unregister_netdev; |
8de4b2f3 WG |
3219 | } |
3220 | ||
a220871b | 3221 | virtnet_set_queues(vi, vi->curr_queue_pairs); |
44900010 | 3222 | |
167c25e4 JW |
3223 | /* Assume link up if device can't report link status, |
3224 | otherwise get link status from config. */ | |
bda7fab5 | 3225 | netif_carrier_off(dev); |
167c25e4 | 3226 | if (virtio_has_feature(vi->vdev, VIRTIO_NET_F_STATUS)) { |
3b07e9ca | 3227 | schedule_work(&vi->config_work); |
167c25e4 JW |
3228 | } else { |
3229 | vi->status = VIRTIO_NET_S_LINK_UP; | |
faa9b39f | 3230 | virtnet_update_settings(vi); |
167c25e4 JW |
3231 | netif_carrier_on(dev); |
3232 | } | |
9f4d26d0 | 3233 | |
3f93522f JW |
3234 | for (i = 0; i < ARRAY_SIZE(guest_offloads); i++) |
3235 | if (virtio_has_feature(vi->vdev, guest_offloads[i])) | |
3236 | set_bit(guest_offloads[i], &vi->guest_offloads); | |
a02e8964 | 3237 | vi->guest_offloads_capable = vi->guest_offloads; |
3f93522f | 3238 | |
986a4f4d JW |
3239 | pr_debug("virtnet: registered device %s with %d RX and TX vq's\n", |
3240 | dev->name, max_queue_pairs); | |
3241 | ||
296f96fc RR |
3242 | return 0; |
3243 | ||
f00e35e2 | 3244 | free_unregister_netdev: |
02465555 MT |
3245 | vi->vdev->config->reset(vdev); |
3246 | ||
b3369c1f | 3247 | unregister_netdev(dev); |
ba5e4426 SS |
3248 | free_failover: |
3249 | net_failover_destroy(vi->failover); | |
d2a7ddda | 3250 | free_vqs: |
986a4f4d | 3251 | cancel_delayed_work_sync(&vi->refill); |
fb51879d | 3252 | free_receive_page_frags(vi); |
e9d7417b | 3253 | virtnet_del_vqs(vi); |
296f96fc RR |
3254 | free: |
3255 | free_netdev(dev); | |
3256 | return err; | |
3257 | } | |
3258 | ||
04486ed0 | 3259 | static void remove_vq_common(struct virtnet_info *vi) |
296f96fc | 3260 | { |
04486ed0 | 3261 | vi->vdev->config->reset(vi->vdev); |
830a8a97 SM |
3262 | |
3263 | /* Free unused buffers in both send and recv, if any. */ | |
9ab86bbc | 3264 | free_unused_bufs(vi); |
fb6813f4 | 3265 | |
986a4f4d | 3266 | free_receive_bufs(vi); |
d2a7ddda | 3267 | |
fb51879d MD |
3268 | free_receive_page_frags(vi); |
3269 | ||
986a4f4d | 3270 | virtnet_del_vqs(vi); |
04486ed0 AS |
3271 | } |
3272 | ||
8cc085d6 | 3273 | static void virtnet_remove(struct virtio_device *vdev) |
04486ed0 AS |
3274 | { |
3275 | struct virtnet_info *vi = vdev->priv; | |
3276 | ||
8017c279 | 3277 | virtnet_cpu_notif_remove(vi); |
8de4b2f3 | 3278 | |
102a2786 MT |
3279 | /* Make sure no work handler is accessing the device. */ |
3280 | flush_work(&vi->config_work); | |
586d17c5 | 3281 | |
04486ed0 AS |
3282 | unregister_netdev(vi->dev); |
3283 | ||
ba5e4426 SS |
3284 | net_failover_destroy(vi->failover); |
3285 | ||
04486ed0 | 3286 | remove_vq_common(vi); |
fb6813f4 | 3287 | |
74b2553f | 3288 | free_netdev(vi->dev); |
296f96fc RR |
3289 | } |
3290 | ||
67a75194 | 3291 | static __maybe_unused int virtnet_freeze(struct virtio_device *vdev) |
0741bcb5 AS |
3292 | { |
3293 | struct virtnet_info *vi = vdev->priv; | |
3294 | ||
8017c279 | 3295 | virtnet_cpu_notif_remove(vi); |
9fe7bfce | 3296 | virtnet_freeze_down(vdev); |
0741bcb5 AS |
3297 | remove_vq_common(vi); |
3298 | ||
3299 | return 0; | |
3300 | } | |
3301 | ||
67a75194 | 3302 | static __maybe_unused int virtnet_restore(struct virtio_device *vdev) |
0741bcb5 AS |
3303 | { |
3304 | struct virtnet_info *vi = vdev->priv; | |
9fe7bfce | 3305 | int err; |
0741bcb5 | 3306 | |
9fe7bfce | 3307 | err = virtnet_restore_up(vdev); |
0741bcb5 AS |
3308 | if (err) |
3309 | return err; | |
986a4f4d JW |
3310 | virtnet_set_queues(vi, vi->curr_queue_pairs); |
3311 | ||
8017c279 | 3312 | err = virtnet_cpu_notif_add(vi); |
3f2869ca XY |
3313 | if (err) { |
3314 | virtnet_freeze_down(vdev); | |
3315 | remove_vq_common(vi); | |
ec9debbd | 3316 | return err; |
3f2869ca | 3317 | } |
ec9debbd | 3318 | |
0741bcb5 AS |
3319 | return 0; |
3320 | } | |
0741bcb5 | 3321 | |
296f96fc RR |
3322 | static struct virtio_device_id id_table[] = { |
3323 | { VIRTIO_ID_NET, VIRTIO_DEV_ANY_ID }, | |
3324 | { 0 }, | |
3325 | }; | |
3326 | ||
f3358507 MT |
3327 | #define VIRTNET_FEATURES \ |
3328 | VIRTIO_NET_F_CSUM, VIRTIO_NET_F_GUEST_CSUM, \ | |
3329 | VIRTIO_NET_F_MAC, \ | |
3330 | VIRTIO_NET_F_HOST_TSO4, VIRTIO_NET_F_HOST_UFO, VIRTIO_NET_F_HOST_TSO6, \ | |
3331 | VIRTIO_NET_F_HOST_ECN, VIRTIO_NET_F_GUEST_TSO4, VIRTIO_NET_F_GUEST_TSO6, \ | |
3332 | VIRTIO_NET_F_GUEST_ECN, VIRTIO_NET_F_GUEST_UFO, \ | |
3333 | VIRTIO_NET_F_MRG_RXBUF, VIRTIO_NET_F_STATUS, VIRTIO_NET_F_CTRL_VQ, \ | |
3334 | VIRTIO_NET_F_CTRL_RX, VIRTIO_NET_F_CTRL_VLAN, \ | |
3335 | VIRTIO_NET_F_GUEST_ANNOUNCE, VIRTIO_NET_F_MQ, \ | |
3336 | VIRTIO_NET_F_CTRL_MAC_ADDR, \ | |
faa9b39f | 3337 | VIRTIO_NET_F_MTU, VIRTIO_NET_F_CTRL_GUEST_OFFLOADS, \ |
9805069d | 3338 | VIRTIO_NET_F_SPEED_DUPLEX, VIRTIO_NET_F_STANDBY |
f3358507 | 3339 | |
c45a6816 | 3340 | static unsigned int features[] = { |
f3358507 MT |
3341 | VIRTNET_FEATURES, |
3342 | }; | |
3343 | ||
3344 | static unsigned int features_legacy[] = { | |
3345 | VIRTNET_FEATURES, | |
3346 | VIRTIO_NET_F_GSO, | |
e7428e95 | 3347 | VIRTIO_F_ANY_LAYOUT, |
c45a6816 RR |
3348 | }; |
3349 | ||
22402529 | 3350 | static struct virtio_driver virtio_net_driver = { |
c45a6816 RR |
3351 | .feature_table = features, |
3352 | .feature_table_size = ARRAY_SIZE(features), | |
f3358507 MT |
3353 | .feature_table_legacy = features_legacy, |
3354 | .feature_table_size_legacy = ARRAY_SIZE(features_legacy), | |
296f96fc RR |
3355 | .driver.name = KBUILD_MODNAME, |
3356 | .driver.owner = THIS_MODULE, | |
3357 | .id_table = id_table, | |
fe36cbe0 | 3358 | .validate = virtnet_validate, |
296f96fc | 3359 | .probe = virtnet_probe, |
8cc085d6 | 3360 | .remove = virtnet_remove, |
9f4d26d0 | 3361 | .config_changed = virtnet_config_changed, |
89107000 | 3362 | #ifdef CONFIG_PM_SLEEP |
0741bcb5 AS |
3363 | .freeze = virtnet_freeze, |
3364 | .restore = virtnet_restore, | |
3365 | #endif | |
296f96fc RR |
3366 | }; |
3367 | ||
8017c279 SAS |
3368 | static __init int virtio_net_driver_init(void) |
3369 | { | |
3370 | int ret; | |
3371 | ||
73c1b41e | 3372 | ret = cpuhp_setup_state_multi(CPUHP_AP_ONLINE_DYN, "virtio/net:online", |
8017c279 SAS |
3373 | virtnet_cpu_online, |
3374 | virtnet_cpu_down_prep); | |
3375 | if (ret < 0) | |
3376 | goto out; | |
3377 | virtionet_online = ret; | |
73c1b41e | 3378 | ret = cpuhp_setup_state_multi(CPUHP_VIRT_NET_DEAD, "virtio/net:dead", |
8017c279 SAS |
3379 | NULL, virtnet_cpu_dead); |
3380 | if (ret) | |
3381 | goto err_dead; | |
3382 | ||
3383 | ret = register_virtio_driver(&virtio_net_driver); | |
3384 | if (ret) | |
3385 | goto err_virtio; | |
3386 | return 0; | |
3387 | err_virtio: | |
3388 | cpuhp_remove_multi_state(CPUHP_VIRT_NET_DEAD); | |
3389 | err_dead: | |
3390 | cpuhp_remove_multi_state(virtionet_online); | |
3391 | out: | |
3392 | return ret; | |
3393 | } | |
3394 | module_init(virtio_net_driver_init); | |
3395 | ||
3396 | static __exit void virtio_net_driver_exit(void) | |
3397 | { | |
cfa0ebc9 | 3398 | unregister_virtio_driver(&virtio_net_driver); |
8017c279 SAS |
3399 | cpuhp_remove_multi_state(CPUHP_VIRT_NET_DEAD); |
3400 | cpuhp_remove_multi_state(virtionet_online); | |
8017c279 SAS |
3401 | } |
3402 | module_exit(virtio_net_driver_exit); | |
296f96fc RR |
3403 | |
3404 | MODULE_DEVICE_TABLE(virtio, id_table); | |
3405 | MODULE_DESCRIPTION("Virtio network driver"); | |
3406 | MODULE_LICENSE("GPL"); |