]>
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 | |
d7dfc5cf TM |
66 | struct virtnet_stat_desc { |
67 | char desc[ETH_GSTRING_LEN]; | |
68 | size_t offset; | |
3fa2a1df | 69 | }; |
70 | ||
d7dfc5cf TM |
71 | struct virtnet_sq_stats { |
72 | struct u64_stats_sync syncp; | |
73 | u64 packets; | |
74 | u64 bytes; | |
5b8f3c8d TM |
75 | u64 xdp_tx; |
76 | u64 xdp_tx_drops; | |
461f03dc | 77 | u64 kicks; |
d7dfc5cf TM |
78 | }; |
79 | ||
d46eeeaf JW |
80 | struct virtnet_rq_stats { |
81 | struct u64_stats_sync syncp; | |
d7dfc5cf TM |
82 | u64 packets; |
83 | u64 bytes; | |
2c4a2f7d | 84 | u64 drops; |
5b8f3c8d TM |
85 | u64 xdp_packets; |
86 | u64 xdp_tx; | |
87 | u64 xdp_redirects; | |
88 | u64 xdp_drops; | |
461f03dc | 89 | u64 kicks; |
d7dfc5cf TM |
90 | }; |
91 | ||
92 | #define VIRTNET_SQ_STAT(m) offsetof(struct virtnet_sq_stats, m) | |
d46eeeaf | 93 | #define VIRTNET_RQ_STAT(m) offsetof(struct virtnet_rq_stats, m) |
d7dfc5cf TM |
94 | |
95 | static const struct virtnet_stat_desc virtnet_sq_stats_desc[] = { | |
5b8f3c8d TM |
96 | { "packets", VIRTNET_SQ_STAT(packets) }, |
97 | { "bytes", VIRTNET_SQ_STAT(bytes) }, | |
98 | { "xdp_tx", VIRTNET_SQ_STAT(xdp_tx) }, | |
99 | { "xdp_tx_drops", VIRTNET_SQ_STAT(xdp_tx_drops) }, | |
461f03dc | 100 | { "kicks", VIRTNET_SQ_STAT(kicks) }, |
d7dfc5cf TM |
101 | }; |
102 | ||
103 | static const struct virtnet_stat_desc virtnet_rq_stats_desc[] = { | |
5b8f3c8d TM |
104 | { "packets", VIRTNET_RQ_STAT(packets) }, |
105 | { "bytes", VIRTNET_RQ_STAT(bytes) }, | |
106 | { "drops", VIRTNET_RQ_STAT(drops) }, | |
107 | { "xdp_packets", VIRTNET_RQ_STAT(xdp_packets) }, | |
108 | { "xdp_tx", VIRTNET_RQ_STAT(xdp_tx) }, | |
109 | { "xdp_redirects", VIRTNET_RQ_STAT(xdp_redirects) }, | |
110 | { "xdp_drops", VIRTNET_RQ_STAT(xdp_drops) }, | |
461f03dc | 111 | { "kicks", VIRTNET_RQ_STAT(kicks) }, |
d7dfc5cf TM |
112 | }; |
113 | ||
114 | #define VIRTNET_SQ_STATS_LEN ARRAY_SIZE(virtnet_sq_stats_desc) | |
115 | #define VIRTNET_RQ_STATS_LEN ARRAY_SIZE(virtnet_rq_stats_desc) | |
116 | ||
e9d7417b JW |
117 | /* Internal representation of a send virtqueue */ |
118 | struct send_queue { | |
119 | /* Virtqueue associated with this send _queue */ | |
120 | struct virtqueue *vq; | |
121 | ||
122 | /* TX: fragments + linear part + virtio header */ | |
123 | struct scatterlist sg[MAX_SKB_FRAGS + 2]; | |
986a4f4d JW |
124 | |
125 | /* Name of the send queue: output.$index */ | |
126 | char name[40]; | |
b92f1e67 | 127 | |
d7dfc5cf TM |
128 | struct virtnet_sq_stats stats; |
129 | ||
b92f1e67 | 130 | struct napi_struct napi; |
e9d7417b JW |
131 | }; |
132 | ||
133 | /* Internal representation of a receive virtqueue */ | |
134 | struct receive_queue { | |
135 | /* Virtqueue associated with this receive_queue */ | |
136 | struct virtqueue *vq; | |
137 | ||
296f96fc RR |
138 | struct napi_struct napi; |
139 | ||
f600b690 JF |
140 | struct bpf_prog __rcu *xdp_prog; |
141 | ||
d7dfc5cf TM |
142 | struct virtnet_rq_stats stats; |
143 | ||
e9d7417b JW |
144 | /* Chain pages by the private ptr. */ |
145 | struct page *pages; | |
146 | ||
ab7db917 | 147 | /* Average packet length for mergeable receive buffers. */ |
5377d758 | 148 | struct ewma_pkt_len mrg_avg_pkt_len; |
ab7db917 | 149 | |
fb51879d MD |
150 | /* Page frag for packet buffer allocation. */ |
151 | struct page_frag alloc_frag; | |
152 | ||
e9d7417b JW |
153 | /* RX: fragments + linear part + virtio header */ |
154 | struct scatterlist sg[MAX_SKB_FRAGS + 2]; | |
986a4f4d | 155 | |
d85b758f MT |
156 | /* Min single buffer size for mergeable buffers case. */ |
157 | unsigned int min_buf_len; | |
158 | ||
986a4f4d JW |
159 | /* Name of this receive queue: input.$index */ |
160 | char name[40]; | |
754b8a21 JDB |
161 | |
162 | struct xdp_rxq_info xdp_rxq; | |
e9d7417b JW |
163 | }; |
164 | ||
12e57169 MT |
165 | /* Control VQ buffers: protected by the rtnl lock */ |
166 | struct control_buf { | |
167 | struct virtio_net_ctrl_hdr hdr; | |
168 | virtio_net_ctrl_ack status; | |
169 | struct virtio_net_ctrl_mq mq; | |
170 | u8 promisc; | |
171 | u8 allmulti; | |
d7fad4c8 | 172 | __virtio16 vid; |
f4ee703a | 173 | __virtio64 offloads; |
12e57169 MT |
174 | }; |
175 | ||
e9d7417b JW |
176 | struct virtnet_info { |
177 | struct virtio_device *vdev; | |
178 | struct virtqueue *cvq; | |
179 | struct net_device *dev; | |
986a4f4d JW |
180 | struct send_queue *sq; |
181 | struct receive_queue *rq; | |
e9d7417b JW |
182 | unsigned int status; |
183 | ||
986a4f4d JW |
184 | /* Max # of queue pairs supported by the device */ |
185 | u16 max_queue_pairs; | |
186 | ||
187 | /* # of queue pairs currently used by the driver */ | |
188 | u16 curr_queue_pairs; | |
189 | ||
672aafd5 JF |
190 | /* # of XDP queue pairs currently used by the driver */ |
191 | u16 xdp_queue_pairs; | |
192 | ||
97402b96 HX |
193 | /* I like... big packets and I cannot lie! */ |
194 | bool big_packets; | |
195 | ||
3f2c31d9 MM |
196 | /* Host will merge rx buffers for big packets (shake it! shake it!) */ |
197 | bool mergeable_rx_bufs; | |
198 | ||
986a4f4d JW |
199 | /* Has control virtqueue */ |
200 | bool has_cvq; | |
201 | ||
e7428e95 MT |
202 | /* Host can handle any s/g split between our header and packet data */ |
203 | bool any_header_sg; | |
204 | ||
012873d0 MT |
205 | /* Packet virtio header size */ |
206 | u8 hdr_len; | |
207 | ||
3161e453 RR |
208 | /* Work struct for refilling if we run low on memory. */ |
209 | struct delayed_work refill; | |
210 | ||
586d17c5 JW |
211 | /* Work struct for config space updates */ |
212 | struct work_struct config_work; | |
213 | ||
986a4f4d JW |
214 | /* Does the affinity hint is set for virtqueues? */ |
215 | bool affinity_hint_set; | |
47be2479 | 216 | |
8017c279 SAS |
217 | /* CPU hotplug instances for online & dead */ |
218 | struct hlist_node node; | |
219 | struct hlist_node node_dead; | |
2ac46030 | 220 | |
12e57169 | 221 | struct control_buf *ctrl; |
16032be5 NA |
222 | |
223 | /* Ethtool settings */ | |
224 | u8 duplex; | |
225 | u32 speed; | |
3f93522f JW |
226 | |
227 | unsigned long guest_offloads; | |
a02e8964 | 228 | unsigned long guest_offloads_capable; |
ba5e4426 SS |
229 | |
230 | /* failover when STANDBY feature enabled */ | |
231 | struct failover *failover; | |
296f96fc RR |
232 | }; |
233 | ||
9ab86bbc | 234 | struct padded_vnet_hdr { |
012873d0 | 235 | struct virtio_net_hdr_mrg_rxbuf hdr; |
9ab86bbc | 236 | /* |
012873d0 MT |
237 | * hdr is in a separate sg buffer, and data sg buffer shares same page |
238 | * with this header sg. This padding makes next sg 16 byte aligned | |
239 | * after the header. | |
9ab86bbc | 240 | */ |
012873d0 | 241 | char padding[4]; |
9ab86bbc SM |
242 | }; |
243 | ||
5050471d TM |
244 | static bool is_xdp_frame(void *ptr) |
245 | { | |
246 | return (unsigned long)ptr & VIRTIO_XDP_FLAG; | |
247 | } | |
248 | ||
249 | static void *xdp_to_ptr(struct xdp_frame *ptr) | |
250 | { | |
251 | return (void *)((unsigned long)ptr | VIRTIO_XDP_FLAG); | |
252 | } | |
253 | ||
254 | static struct xdp_frame *ptr_to_xdp(void *ptr) | |
255 | { | |
256 | return (struct xdp_frame *)((unsigned long)ptr & ~VIRTIO_XDP_FLAG); | |
257 | } | |
258 | ||
986a4f4d JW |
259 | /* Converting between virtqueue no. and kernel tx/rx queue no. |
260 | * 0:rx0 1:tx0 2:rx1 3:tx1 ... 2N:rxN 2N+1:txN 2N+2:cvq | |
261 | */ | |
262 | static int vq2txq(struct virtqueue *vq) | |
263 | { | |
9d0ca6ed | 264 | return (vq->index - 1) / 2; |
986a4f4d JW |
265 | } |
266 | ||
267 | static int txq2vq(int txq) | |
268 | { | |
269 | return txq * 2 + 1; | |
270 | } | |
271 | ||
272 | static int vq2rxq(struct virtqueue *vq) | |
273 | { | |
9d0ca6ed | 274 | return vq->index / 2; |
986a4f4d JW |
275 | } |
276 | ||
277 | static int rxq2vq(int rxq) | |
278 | { | |
279 | return rxq * 2; | |
280 | } | |
281 | ||
012873d0 | 282 | static inline struct virtio_net_hdr_mrg_rxbuf *skb_vnet_hdr(struct sk_buff *skb) |
296f96fc | 283 | { |
012873d0 | 284 | return (struct virtio_net_hdr_mrg_rxbuf *)skb->cb; |
296f96fc RR |
285 | } |
286 | ||
9ab86bbc SM |
287 | /* |
288 | * private is used to chain pages for big packets, put the whole | |
289 | * most recent used list in the beginning for reuse | |
290 | */ | |
e9d7417b | 291 | static void give_pages(struct receive_queue *rq, struct page *page) |
0a888fd1 | 292 | { |
9ab86bbc | 293 | struct page *end; |
0a888fd1 | 294 | |
e9d7417b | 295 | /* Find end of list, sew whole thing into vi->rq.pages. */ |
9ab86bbc | 296 | for (end = page; end->private; end = (struct page *)end->private); |
e9d7417b JW |
297 | end->private = (unsigned long)rq->pages; |
298 | rq->pages = page; | |
0a888fd1 MM |
299 | } |
300 | ||
e9d7417b | 301 | static struct page *get_a_page(struct receive_queue *rq, gfp_t gfp_mask) |
fb6813f4 | 302 | { |
e9d7417b | 303 | struct page *p = rq->pages; |
fb6813f4 | 304 | |
9ab86bbc | 305 | if (p) { |
e9d7417b | 306 | rq->pages = (struct page *)p->private; |
9ab86bbc SM |
307 | /* clear private here, it is used to chain pages */ |
308 | p->private = 0; | |
309 | } else | |
fb6813f4 RR |
310 | p = alloc_page(gfp_mask); |
311 | return p; | |
312 | } | |
313 | ||
e4e8452a WB |
314 | static void virtqueue_napi_schedule(struct napi_struct *napi, |
315 | struct virtqueue *vq) | |
316 | { | |
317 | if (napi_schedule_prep(napi)) { | |
318 | virtqueue_disable_cb(vq); | |
319 | __napi_schedule(napi); | |
320 | } | |
321 | } | |
322 | ||
323 | static void virtqueue_napi_complete(struct napi_struct *napi, | |
324 | struct virtqueue *vq, int processed) | |
325 | { | |
326 | int opaque; | |
327 | ||
328 | opaque = virtqueue_enable_cb_prepare(vq); | |
fdaa767a TM |
329 | if (napi_complete_done(napi, processed)) { |
330 | if (unlikely(virtqueue_poll(vq, opaque))) | |
331 | virtqueue_napi_schedule(napi, vq); | |
332 | } else { | |
333 | virtqueue_disable_cb(vq); | |
334 | } | |
e4e8452a WB |
335 | } |
336 | ||
e9d7417b | 337 | static void skb_xmit_done(struct virtqueue *vq) |
296f96fc | 338 | { |
e9d7417b | 339 | struct virtnet_info *vi = vq->vdev->priv; |
b92f1e67 | 340 | struct napi_struct *napi = &vi->sq[vq2txq(vq)].napi; |
296f96fc | 341 | |
2cb9c6ba | 342 | /* Suppress further interrupts. */ |
e9d7417b | 343 | virtqueue_disable_cb(vq); |
11a3a154 | 344 | |
b92f1e67 WB |
345 | if (napi->weight) |
346 | virtqueue_napi_schedule(napi, vq); | |
347 | else | |
348 | /* We were probably waiting for more output buffers. */ | |
349 | netif_wake_subqueue(vi->dev, vq2txq(vq)); | |
296f96fc RR |
350 | } |
351 | ||
28b39bc7 JW |
352 | #define MRG_CTX_HEADER_SHIFT 22 |
353 | static void *mergeable_len_to_ctx(unsigned int truesize, | |
354 | unsigned int headroom) | |
355 | { | |
356 | return (void *)(unsigned long)((headroom << MRG_CTX_HEADER_SHIFT) | truesize); | |
357 | } | |
358 | ||
359 | static unsigned int mergeable_ctx_to_headroom(void *mrg_ctx) | |
360 | { | |
361 | return (unsigned long)mrg_ctx >> MRG_CTX_HEADER_SHIFT; | |
362 | } | |
363 | ||
364 | static unsigned int mergeable_ctx_to_truesize(void *mrg_ctx) | |
365 | { | |
366 | return (unsigned long)mrg_ctx & ((1 << MRG_CTX_HEADER_SHIFT) - 1); | |
367 | } | |
368 | ||
3464645a | 369 | /* Called from bottom half context */ |
946fa564 MT |
370 | static struct sk_buff *page_to_skb(struct virtnet_info *vi, |
371 | struct receive_queue *rq, | |
2613af0e | 372 | struct page *page, unsigned int offset, |
436c9453 | 373 | unsigned int len, unsigned int truesize, |
503d539a | 374 | bool hdr_valid, unsigned int metasize) |
9ab86bbc SM |
375 | { |
376 | struct sk_buff *skb; | |
012873d0 | 377 | struct virtio_net_hdr_mrg_rxbuf *hdr; |
2613af0e | 378 | unsigned int copy, hdr_len, hdr_padded_len; |
9ab86bbc | 379 | char *p; |
fb6813f4 | 380 | |
2613af0e | 381 | p = page_address(page) + offset; |
3f2c31d9 | 382 | |
9ab86bbc | 383 | /* copy small packet so we can reuse these pages for small data */ |
c67f5db8 | 384 | skb = napi_alloc_skb(&rq->napi, GOOD_COPY_LEN); |
9ab86bbc SM |
385 | if (unlikely(!skb)) |
386 | return NULL; | |
3f2c31d9 | 387 | |
9ab86bbc | 388 | hdr = skb_vnet_hdr(skb); |
3f2c31d9 | 389 | |
012873d0 MT |
390 | hdr_len = vi->hdr_len; |
391 | if (vi->mergeable_rx_bufs) | |
a4a76503 | 392 | hdr_padded_len = sizeof(*hdr); |
012873d0 | 393 | else |
2613af0e | 394 | hdr_padded_len = sizeof(struct padded_vnet_hdr); |
3f2c31d9 | 395 | |
503d539a | 396 | /* hdr_valid means no XDP, so we can copy the vnet header */ |
436c9453 JW |
397 | if (hdr_valid) |
398 | memcpy(hdr, p, hdr_len); | |
3f2c31d9 | 399 | |
9ab86bbc | 400 | len -= hdr_len; |
2613af0e MD |
401 | offset += hdr_padded_len; |
402 | p += hdr_padded_len; | |
3f2c31d9 | 403 | |
9ab86bbc SM |
404 | copy = len; |
405 | if (copy > skb_tailroom(skb)) | |
406 | copy = skb_tailroom(skb); | |
59ae1d12 | 407 | skb_put_data(skb, p, copy); |
3f2c31d9 | 408 | |
503d539a YK |
409 | if (metasize) { |
410 | __skb_pull(skb, metasize); | |
411 | skb_metadata_set(skb, metasize); | |
412 | } | |
413 | ||
9ab86bbc SM |
414 | len -= copy; |
415 | offset += copy; | |
3f2c31d9 | 416 | |
2613af0e MD |
417 | if (vi->mergeable_rx_bufs) { |
418 | if (len) | |
419 | skb_add_rx_frag(skb, 0, page, offset, len, truesize); | |
420 | else | |
421 | put_page(page); | |
422 | return skb; | |
423 | } | |
424 | ||
e878d78b SL |
425 | /* |
426 | * Verify that we can indeed put this data into a skb. | |
427 | * This is here to handle cases when the device erroneously | |
428 | * tries to receive more than is possible. This is usually | |
429 | * the case of a broken device. | |
430 | */ | |
431 | if (unlikely(len > MAX_SKB_FRAGS * PAGE_SIZE)) { | |
be443899 | 432 | net_dbg_ratelimited("%s: too much data\n", skb->dev->name); |
e878d78b SL |
433 | dev_kfree_skb(skb); |
434 | return NULL; | |
435 | } | |
2613af0e | 436 | BUG_ON(offset >= PAGE_SIZE); |
9ab86bbc | 437 | while (len) { |
2613af0e MD |
438 | unsigned int frag_size = min((unsigned)PAGE_SIZE - offset, len); |
439 | skb_add_rx_frag(skb, skb_shinfo(skb)->nr_frags, page, offset, | |
440 | frag_size, truesize); | |
441 | len -= frag_size; | |
9ab86bbc SM |
442 | page = (struct page *)page->private; |
443 | offset = 0; | |
444 | } | |
3f2c31d9 | 445 | |
9ab86bbc | 446 | if (page) |
e9d7417b | 447 | give_pages(rq, page); |
3f2c31d9 | 448 | |
9ab86bbc SM |
449 | return skb; |
450 | } | |
3f2c31d9 | 451 | |
735fc405 JDB |
452 | static int __virtnet_xdp_xmit_one(struct virtnet_info *vi, |
453 | struct send_queue *sq, | |
454 | struct xdp_frame *xdpf) | |
56434a01 | 455 | { |
56434a01 | 456 | struct virtio_net_hdr_mrg_rxbuf *hdr; |
56434a01 JF |
457 | int err; |
458 | ||
cac320c8 JDB |
459 | if (unlikely(xdpf->headroom < vi->hdr_len)) |
460 | return -EOVERFLOW; | |
461 | ||
462 | /* Make room for virtqueue hdr (also change xdpf->headroom?) */ | |
463 | xdpf->data -= vi->hdr_len; | |
f6b10209 | 464 | /* Zero header and leave csum up to XDP layers */ |
cac320c8 | 465 | hdr = xdpf->data; |
f6b10209 | 466 | memset(hdr, 0, vi->hdr_len); |
cac320c8 | 467 | xdpf->len += vi->hdr_len; |
bb91accf | 468 | |
cac320c8 | 469 | sg_init_one(sq->sg, xdpf->data, xdpf->len); |
bb91accf | 470 | |
5050471d TM |
471 | err = virtqueue_add_outbuf(sq->vq, sq->sg, 1, xdp_to_ptr(xdpf), |
472 | GFP_ATOMIC); | |
11b7d897 | 473 | if (unlikely(err)) |
cac320c8 | 474 | return -ENOSPC; /* Caller handle free/refcnt */ |
56434a01 | 475 | |
cac320c8 | 476 | return 0; |
56434a01 JF |
477 | } |
478 | ||
2a43565c TM |
479 | static struct send_queue *virtnet_xdp_sq(struct virtnet_info *vi) |
480 | { | |
481 | unsigned int qp; | |
482 | ||
483 | qp = vi->curr_queue_pairs - vi->xdp_queue_pairs + smp_processor_id(); | |
484 | return &vi->sq[qp]; | |
485 | } | |
486 | ||
735fc405 | 487 | static int virtnet_xdp_xmit(struct net_device *dev, |
42b33468 | 488 | int n, struct xdp_frame **frames, u32 flags) |
186b3c99 JW |
489 | { |
490 | struct virtnet_info *vi = netdev_priv(dev); | |
8dcc5b0a JDB |
491 | struct receive_queue *rq = vi->rq; |
492 | struct bpf_prog *xdp_prog; | |
735fc405 JDB |
493 | struct send_queue *sq; |
494 | unsigned int len; | |
546f2897 TM |
495 | int packets = 0; |
496 | int bytes = 0; | |
735fc405 | 497 | int drops = 0; |
461f03dc | 498 | int kicks = 0; |
5b8f3c8d | 499 | int ret, err; |
5050471d | 500 | void *ptr; |
735fc405 JDB |
501 | int i; |
502 | ||
8dcc5b0a JDB |
503 | /* Only allow ndo_xdp_xmit if XDP is loaded on dev, as this |
504 | * indicate XDP resources have been successfully allocated. | |
505 | */ | |
9719c6b9 | 506 | xdp_prog = rcu_access_pointer(rq->xdp_prog); |
1667c08a TM |
507 | if (!xdp_prog) |
508 | return -ENXIO; | |
509 | ||
510 | sq = virtnet_xdp_sq(vi); | |
511 | ||
512 | if (unlikely(flags & ~XDP_XMIT_FLAGS_MASK)) { | |
513 | ret = -EINVAL; | |
5b8f3c8d TM |
514 | drops = n; |
515 | goto out; | |
516 | } | |
8dcc5b0a | 517 | |
735fc405 | 518 | /* Free up any pending old buffers before queueing new ones. */ |
5050471d | 519 | while ((ptr = virtqueue_get_buf(sq->vq, &len)) != NULL) { |
546f2897 TM |
520 | if (likely(is_xdp_frame(ptr))) { |
521 | struct xdp_frame *frame = ptr_to_xdp(ptr); | |
522 | ||
523 | bytes += frame->len; | |
524 | xdp_return_frame(frame); | |
525 | } else { | |
526 | struct sk_buff *skb = ptr; | |
527 | ||
528 | bytes += skb->len; | |
529 | napi_consume_skb(skb, false); | |
530 | } | |
531 | packets++; | |
5050471d | 532 | } |
735fc405 JDB |
533 | |
534 | for (i = 0; i < n; i++) { | |
535 | struct xdp_frame *xdpf = frames[i]; | |
536 | ||
537 | err = __virtnet_xdp_xmit_one(vi, sq, xdpf); | |
538 | if (err) { | |
539 | xdp_return_frame_rx_napi(xdpf); | |
540 | drops++; | |
541 | } | |
542 | } | |
5b8f3c8d | 543 | ret = n - drops; |
5d274cb4 | 544 | |
461f03dc TM |
545 | if (flags & XDP_XMIT_FLUSH) { |
546 | if (virtqueue_kick_prepare(sq->vq) && virtqueue_notify(sq->vq)) | |
547 | kicks = 1; | |
548 | } | |
5b8f3c8d TM |
549 | out: |
550 | u64_stats_update_begin(&sq->stats.syncp); | |
546f2897 TM |
551 | sq->stats.bytes += bytes; |
552 | sq->stats.packets += packets; | |
5b8f3c8d TM |
553 | sq->stats.xdp_tx += n; |
554 | sq->stats.xdp_tx_drops += drops; | |
461f03dc | 555 | sq->stats.kicks += kicks; |
5b8f3c8d | 556 | u64_stats_update_end(&sq->stats.syncp); |
5d274cb4 | 557 | |
5b8f3c8d | 558 | return ret; |
186b3c99 JW |
559 | } |
560 | ||
f6b10209 JW |
561 | static unsigned int virtnet_get_headroom(struct virtnet_info *vi) |
562 | { | |
563 | return vi->xdp_queue_pairs ? VIRTIO_XDP_HEADROOM : 0; | |
564 | } | |
565 | ||
4941d472 JW |
566 | /* We copy the packet for XDP in the following cases: |
567 | * | |
568 | * 1) Packet is scattered across multiple rx buffers. | |
569 | * 2) Headroom space is insufficient. | |
570 | * | |
571 | * This is inefficient but it's a temporary condition that | |
572 | * we hit right after XDP is enabled and until queue is refilled | |
573 | * with large buffers with sufficient headroom - so it should affect | |
574 | * at most queue size packets. | |
575 | * Afterwards, the conditions to enable | |
576 | * XDP should preclude the underlying device from sending packets | |
577 | * across multiple buffers (num_buf > 1), and we make sure buffers | |
578 | * have enough headroom. | |
579 | */ | |
580 | static struct page *xdp_linearize_page(struct receive_queue *rq, | |
581 | u16 *num_buf, | |
582 | struct page *p, | |
583 | int offset, | |
584 | int page_off, | |
585 | unsigned int *len) | |
586 | { | |
587 | struct page *page = alloc_page(GFP_ATOMIC); | |
588 | ||
589 | if (!page) | |
590 | return NULL; | |
591 | ||
592 | memcpy(page_address(page) + page_off, page_address(p) + offset, *len); | |
593 | page_off += *len; | |
594 | ||
595 | while (--*num_buf) { | |
3cc81a9a | 596 | int tailroom = SKB_DATA_ALIGN(sizeof(struct skb_shared_info)); |
4941d472 JW |
597 | unsigned int buflen; |
598 | void *buf; | |
599 | int off; | |
600 | ||
601 | buf = virtqueue_get_buf(rq->vq, &buflen); | |
602 | if (unlikely(!buf)) | |
603 | goto err_buf; | |
604 | ||
605 | p = virt_to_head_page(buf); | |
606 | off = buf - page_address(p); | |
607 | ||
608 | /* guard against a misconfigured or uncooperative backend that | |
609 | * is sending packet larger than the MTU. | |
610 | */ | |
3cc81a9a | 611 | if ((page_off + buflen + tailroom) > PAGE_SIZE) { |
4941d472 JW |
612 | put_page(p); |
613 | goto err_buf; | |
614 | } | |
615 | ||
616 | memcpy(page_address(page) + page_off, | |
617 | page_address(p) + off, buflen); | |
618 | page_off += buflen; | |
619 | put_page(p); | |
620 | } | |
621 | ||
622 | /* Headroom does not contribute to packet length */ | |
623 | *len = page_off - VIRTIO_XDP_HEADROOM; | |
624 | return page; | |
625 | err_buf: | |
626 | __free_pages(page, 0); | |
627 | return NULL; | |
628 | } | |
629 | ||
bb91accf JW |
630 | static struct sk_buff *receive_small(struct net_device *dev, |
631 | struct virtnet_info *vi, | |
632 | struct receive_queue *rq, | |
192f68cf | 633 | void *buf, void *ctx, |
186b3c99 | 634 | unsigned int len, |
7d9d60fd | 635 | unsigned int *xdp_xmit, |
d46eeeaf | 636 | struct virtnet_rq_stats *stats) |
f121159d | 637 | { |
f6b10209 | 638 | struct sk_buff *skb; |
bb91accf | 639 | struct bpf_prog *xdp_prog; |
4941d472 | 640 | unsigned int xdp_headroom = (unsigned long)ctx; |
f6b10209 JW |
641 | unsigned int header_offset = VIRTNET_RX_PAD + xdp_headroom; |
642 | unsigned int headroom = vi->hdr_len + header_offset; | |
643 | unsigned int buflen = SKB_DATA_ALIGN(GOOD_PACKET_LEN + headroom) + | |
644 | SKB_DATA_ALIGN(sizeof(struct skb_shared_info)); | |
4941d472 | 645 | struct page *page = virt_to_head_page(buf); |
11b7d897 | 646 | unsigned int delta = 0; |
4941d472 | 647 | struct page *xdp_page; |
11b7d897 | 648 | int err; |
503d539a | 649 | unsigned int metasize = 0; |
11b7d897 | 650 | |
012873d0 | 651 | len -= vi->hdr_len; |
d46eeeaf | 652 | stats->bytes += len; |
f121159d | 653 | |
bb91accf JW |
654 | rcu_read_lock(); |
655 | xdp_prog = rcu_dereference(rq->xdp_prog); | |
656 | if (xdp_prog) { | |
f6b10209 | 657 | struct virtio_net_hdr_mrg_rxbuf *hdr = buf + header_offset; |
44fa2dbd | 658 | struct xdp_frame *xdpf; |
0354e4d1 | 659 | struct xdp_buff xdp; |
f6b10209 | 660 | void *orig_data; |
bb91accf JW |
661 | u32 act; |
662 | ||
95dbe9e7 | 663 | if (unlikely(hdr->hdr.gso_type)) |
bb91accf | 664 | goto err_xdp; |
0354e4d1 | 665 | |
4941d472 JW |
666 | if (unlikely(xdp_headroom < virtnet_get_headroom(vi))) { |
667 | int offset = buf - page_address(page) + header_offset; | |
668 | unsigned int tlen = len + vi->hdr_len; | |
669 | u16 num_buf = 1; | |
670 | ||
671 | xdp_headroom = virtnet_get_headroom(vi); | |
672 | header_offset = VIRTNET_RX_PAD + xdp_headroom; | |
673 | headroom = vi->hdr_len + header_offset; | |
674 | buflen = SKB_DATA_ALIGN(GOOD_PACKET_LEN + headroom) + | |
675 | SKB_DATA_ALIGN(sizeof(struct skb_shared_info)); | |
676 | xdp_page = xdp_linearize_page(rq, &num_buf, page, | |
677 | offset, header_offset, | |
678 | &tlen); | |
679 | if (!xdp_page) | |
680 | goto err_xdp; | |
681 | ||
682 | buf = page_address(xdp_page); | |
683 | put_page(page); | |
684 | page = xdp_page; | |
685 | } | |
686 | ||
f6b10209 JW |
687 | xdp.data_hard_start = buf + VIRTNET_RX_PAD + vi->hdr_len; |
688 | xdp.data = xdp.data_hard_start + xdp_headroom; | |
0354e4d1 | 689 | xdp.data_end = xdp.data + len; |
503d539a | 690 | xdp.data_meta = xdp.data; |
754b8a21 | 691 | xdp.rxq = &rq->xdp_rxq; |
9ce6146e | 692 | xdp.frame_sz = buflen; |
f6b10209 | 693 | orig_data = xdp.data; |
0354e4d1 | 694 | act = bpf_prog_run_xdp(xdp_prog, &xdp); |
d46eeeaf | 695 | stats->xdp_packets++; |
0354e4d1 | 696 | |
bb91accf JW |
697 | switch (act) { |
698 | case XDP_PASS: | |
2de2f7f4 | 699 | /* Recalculate length in case bpf program changed it */ |
f6b10209 | 700 | delta = orig_data - xdp.data; |
6870de43 | 701 | len = xdp.data_end - xdp.data; |
503d539a | 702 | metasize = xdp.data - xdp.data_meta; |
bb91accf JW |
703 | break; |
704 | case XDP_TX: | |
d46eeeaf | 705 | stats->xdp_tx++; |
44fa2dbd JDB |
706 | xdpf = convert_to_xdp_frame(&xdp); |
707 | if (unlikely(!xdpf)) | |
708 | goto err_xdp; | |
ca9e83b4 JW |
709 | err = virtnet_xdp_xmit(dev, 1, &xdpf, 0); |
710 | if (unlikely(err < 0)) { | |
0354e4d1 | 711 | trace_xdp_exception(vi->dev, xdp_prog, act); |
11b7d897 JDB |
712 | goto err_xdp; |
713 | } | |
2471c75e | 714 | *xdp_xmit |= VIRTIO_XDP_TX; |
186b3c99 JW |
715 | rcu_read_unlock(); |
716 | goto xdp_xmit; | |
717 | case XDP_REDIRECT: | |
d46eeeaf | 718 | stats->xdp_redirects++; |
186b3c99 | 719 | err = xdp_do_redirect(dev, &xdp, xdp_prog); |
11b7d897 JDB |
720 | if (err) |
721 | goto err_xdp; | |
2471c75e | 722 | *xdp_xmit |= VIRTIO_XDP_REDIR; |
bb91accf JW |
723 | rcu_read_unlock(); |
724 | goto xdp_xmit; | |
bb91accf | 725 | default: |
0354e4d1 | 726 | bpf_warn_invalid_xdp_action(act); |
b633d440 | 727 | /* fall through */ |
0354e4d1 JF |
728 | case XDP_ABORTED: |
729 | trace_xdp_exception(vi->dev, xdp_prog, act); | |
730 | case XDP_DROP: | |
bb91accf JW |
731 | goto err_xdp; |
732 | } | |
733 | } | |
734 | rcu_read_unlock(); | |
735 | ||
f6b10209 JW |
736 | skb = build_skb(buf, buflen); |
737 | if (!skb) { | |
4941d472 | 738 | put_page(page); |
f6b10209 JW |
739 | goto err; |
740 | } | |
741 | skb_reserve(skb, headroom - delta); | |
6870de43 | 742 | skb_put(skb, len); |
f1d4884d | 743 | if (!xdp_prog) { |
f6b10209 JW |
744 | buf += header_offset; |
745 | memcpy(skb_vnet_hdr(skb), buf, vi->hdr_len); | |
f1d4884d | 746 | } /* keep zeroed vnet hdr since XDP is loaded */ |
f6b10209 | 747 | |
503d539a YK |
748 | if (metasize) |
749 | skb_metadata_set(skb, metasize); | |
750 | ||
f6b10209 | 751 | err: |
f121159d | 752 | return skb; |
bb91accf JW |
753 | |
754 | err_xdp: | |
755 | rcu_read_unlock(); | |
d46eeeaf JW |
756 | stats->xdp_drops++; |
757 | stats->drops++; | |
4941d472 | 758 | put_page(page); |
bb91accf JW |
759 | xdp_xmit: |
760 | return NULL; | |
f121159d MT |
761 | } |
762 | ||
763 | static struct sk_buff *receive_big(struct net_device *dev, | |
946fa564 | 764 | struct virtnet_info *vi, |
f121159d MT |
765 | struct receive_queue *rq, |
766 | void *buf, | |
7d9d60fd | 767 | unsigned int len, |
d46eeeaf | 768 | struct virtnet_rq_stats *stats) |
f121159d MT |
769 | { |
770 | struct page *page = buf; | |
503d539a YK |
771 | struct sk_buff *skb = |
772 | page_to_skb(vi, rq, page, 0, len, PAGE_SIZE, true, 0); | |
f600b690 | 773 | |
d46eeeaf | 774 | stats->bytes += len - vi->hdr_len; |
f121159d MT |
775 | if (unlikely(!skb)) |
776 | goto err; | |
777 | ||
778 | return skb; | |
779 | ||
780 | err: | |
d46eeeaf | 781 | stats->drops++; |
f121159d MT |
782 | give_pages(rq, page); |
783 | return NULL; | |
784 | } | |
785 | ||
8fc3b9e9 | 786 | static struct sk_buff *receive_mergeable(struct net_device *dev, |
fdd819b2 | 787 | struct virtnet_info *vi, |
8fc3b9e9 | 788 | struct receive_queue *rq, |
680557cf MT |
789 | void *buf, |
790 | void *ctx, | |
186b3c99 | 791 | unsigned int len, |
7d9d60fd | 792 | unsigned int *xdp_xmit, |
d46eeeaf | 793 | struct virtnet_rq_stats *stats) |
9ab86bbc | 794 | { |
012873d0 MT |
795 | struct virtio_net_hdr_mrg_rxbuf *hdr = buf; |
796 | u16 num_buf = virtio16_to_cpu(vi->vdev, hdr->num_buffers); | |
8fc3b9e9 MT |
797 | struct page *page = virt_to_head_page(buf); |
798 | int offset = buf - page_address(page); | |
f600b690 JF |
799 | struct sk_buff *head_skb, *curr_skb; |
800 | struct bpf_prog *xdp_prog; | |
9ce6146e | 801 | unsigned int truesize = mergeable_ctx_to_truesize(ctx); |
4941d472 | 802 | unsigned int headroom = mergeable_ctx_to_headroom(ctx); |
503d539a | 803 | unsigned int metasize = 0; |
9ce6146e JDB |
804 | unsigned int frame_sz; |
805 | int err; | |
f600b690 | 806 | |
56434a01 | 807 | head_skb = NULL; |
d46eeeaf | 808 | stats->bytes += len - vi->hdr_len; |
56434a01 | 809 | |
f600b690 JF |
810 | rcu_read_lock(); |
811 | xdp_prog = rcu_dereference(rq->xdp_prog); | |
812 | if (xdp_prog) { | |
44fa2dbd | 813 | struct xdp_frame *xdpf; |
72979a6c | 814 | struct page *xdp_page; |
0354e4d1 | 815 | struct xdp_buff xdp; |
0354e4d1 | 816 | void *data; |
f600b690 JF |
817 | u32 act; |
818 | ||
3d62b2a0 JW |
819 | /* Transient failure which in theory could occur if |
820 | * in-flight packets from before XDP was enabled reach | |
821 | * the receive path after XDP is loaded. | |
822 | */ | |
823 | if (unlikely(hdr->hdr.gso_type)) | |
824 | goto err_xdp; | |
825 | ||
9ce6146e JDB |
826 | /* Buffers with headroom use PAGE_SIZE as alloc size, |
827 | * see add_recvbuf_mergeable() + get_mergeable_buf_len() | |
828 | */ | |
829 | frame_sz = headroom ? PAGE_SIZE : truesize; | |
830 | ||
3cc81a9a JW |
831 | /* This happens when rx buffer size is underestimated |
832 | * or headroom is not enough because of the buffer | |
833 | * was refilled before XDP is set. This should only | |
834 | * happen for the first several packets, so we don't | |
835 | * care much about its performance. | |
836 | */ | |
4941d472 JW |
837 | if (unlikely(num_buf > 1 || |
838 | headroom < virtnet_get_headroom(vi))) { | |
72979a6c | 839 | /* linearize data for XDP */ |
56a86f84 | 840 | xdp_page = xdp_linearize_page(rq, &num_buf, |
4941d472 JW |
841 | page, offset, |
842 | VIRTIO_XDP_HEADROOM, | |
843 | &len); | |
9ce6146e JDB |
844 | frame_sz = PAGE_SIZE; |
845 | ||
72979a6c JF |
846 | if (!xdp_page) |
847 | goto err_xdp; | |
2de2f7f4 | 848 | offset = VIRTIO_XDP_HEADROOM; |
72979a6c JF |
849 | } else { |
850 | xdp_page = page; | |
f600b690 JF |
851 | } |
852 | ||
2de2f7f4 JF |
853 | /* Allow consuming headroom but reserve enough space to push |
854 | * the descriptor on if we get an XDP_TX return code. | |
855 | */ | |
0354e4d1 | 856 | data = page_address(xdp_page) + offset; |
2de2f7f4 | 857 | xdp.data_hard_start = data - VIRTIO_XDP_HEADROOM + vi->hdr_len; |
0354e4d1 JF |
858 | xdp.data = data + vi->hdr_len; |
859 | xdp.data_end = xdp.data + (len - vi->hdr_len); | |
503d539a | 860 | xdp.data_meta = xdp.data; |
754b8a21 | 861 | xdp.rxq = &rq->xdp_rxq; |
9ce6146e | 862 | xdp.frame_sz = frame_sz - vi->hdr_len; |
754b8a21 | 863 | |
0354e4d1 | 864 | act = bpf_prog_run_xdp(xdp_prog, &xdp); |
d46eeeaf | 865 | stats->xdp_packets++; |
0354e4d1 | 866 | |
56434a01 JF |
867 | switch (act) { |
868 | case XDP_PASS: | |
503d539a YK |
869 | metasize = xdp.data - xdp.data_meta; |
870 | ||
2de2f7f4 | 871 | /* recalculate offset to account for any header |
503d539a YK |
872 | * adjustments and minus the metasize to copy the |
873 | * metadata in page_to_skb(). Note other cases do not | |
874 | * build an skb and avoid using offset | |
2de2f7f4 | 875 | */ |
503d539a YK |
876 | offset = xdp.data - page_address(xdp_page) - |
877 | vi->hdr_len - metasize; | |
2de2f7f4 | 878 | |
503d539a YK |
879 | /* recalculate len if xdp.data, xdp.data_end or |
880 | * xdp.data_meta were adjusted | |
6870de43 | 881 | */ |
503d539a | 882 | len = xdp.data_end - xdp.data + vi->hdr_len + metasize; |
1830f893 JW |
883 | /* We can only create skb based on xdp_page. */ |
884 | if (unlikely(xdp_page != page)) { | |
885 | rcu_read_unlock(); | |
886 | put_page(page); | |
503d539a YK |
887 | head_skb = page_to_skb(vi, rq, xdp_page, offset, |
888 | len, PAGE_SIZE, false, | |
889 | metasize); | |
1830f893 JW |
890 | return head_skb; |
891 | } | |
56434a01 JF |
892 | break; |
893 | case XDP_TX: | |
d46eeeaf | 894 | stats->xdp_tx++; |
44fa2dbd JDB |
895 | xdpf = convert_to_xdp_frame(&xdp); |
896 | if (unlikely(!xdpf)) | |
897 | goto err_xdp; | |
ca9e83b4 JW |
898 | err = virtnet_xdp_xmit(dev, 1, &xdpf, 0); |
899 | if (unlikely(err < 0)) { | |
0354e4d1 | 900 | trace_xdp_exception(vi->dev, xdp_prog, act); |
11b7d897 JDB |
901 | if (unlikely(xdp_page != page)) |
902 | put_page(xdp_page); | |
903 | goto err_xdp; | |
904 | } | |
2471c75e | 905 | *xdp_xmit |= VIRTIO_XDP_TX; |
72979a6c | 906 | if (unlikely(xdp_page != page)) |
5d458a13 | 907 | put_page(page); |
56434a01 JF |
908 | rcu_read_unlock(); |
909 | goto xdp_xmit; | |
3cc81a9a | 910 | case XDP_REDIRECT: |
d46eeeaf | 911 | stats->xdp_redirects++; |
3cc81a9a JW |
912 | err = xdp_do_redirect(dev, &xdp, xdp_prog); |
913 | if (err) { | |
914 | if (unlikely(xdp_page != page)) | |
915 | put_page(xdp_page); | |
916 | goto err_xdp; | |
917 | } | |
2471c75e | 918 | *xdp_xmit |= VIRTIO_XDP_REDIR; |
3cc81a9a | 919 | if (unlikely(xdp_page != page)) |
6890418b | 920 | put_page(page); |
3cc81a9a JW |
921 | rcu_read_unlock(); |
922 | goto xdp_xmit; | |
56434a01 | 923 | default: |
0354e4d1 | 924 | bpf_warn_invalid_xdp_action(act); |
b633d440 | 925 | /* fall through */ |
0354e4d1 JF |
926 | case XDP_ABORTED: |
927 | trace_xdp_exception(vi->dev, xdp_prog, act); | |
b633d440 | 928 | /* fall through */ |
0354e4d1 | 929 | case XDP_DROP: |
72979a6c JF |
930 | if (unlikely(xdp_page != page)) |
931 | __free_pages(xdp_page, 0); | |
f600b690 | 932 | goto err_xdp; |
56434a01 | 933 | } |
f600b690 JF |
934 | } |
935 | rcu_read_unlock(); | |
ab7db917 | 936 | |
28b39bc7 | 937 | if (unlikely(len > truesize)) { |
56da5fd0 | 938 | pr_debug("%s: rx error: len %u exceeds truesize %lu\n", |
680557cf MT |
939 | dev->name, len, (unsigned long)ctx); |
940 | dev->stats.rx_length_errors++; | |
941 | goto err_skb; | |
942 | } | |
28b39bc7 | 943 | |
503d539a YK |
944 | head_skb = page_to_skb(vi, rq, page, offset, len, truesize, !xdp_prog, |
945 | metasize); | |
f600b690 | 946 | curr_skb = head_skb; |
9ab86bbc | 947 | |
8fc3b9e9 MT |
948 | if (unlikely(!curr_skb)) |
949 | goto err_skb; | |
9ab86bbc | 950 | while (--num_buf) { |
8fc3b9e9 MT |
951 | int num_skb_frags; |
952 | ||
680557cf | 953 | buf = virtqueue_get_buf_ctx(rq->vq, &len, &ctx); |
03e9f8a0 | 954 | if (unlikely(!buf)) { |
8fc3b9e9 | 955 | pr_debug("%s: rx error: %d buffers out of %d missing\n", |
fdd819b2 | 956 | dev->name, num_buf, |
012873d0 MT |
957 | virtio16_to_cpu(vi->vdev, |
958 | hdr->num_buffers)); | |
8fc3b9e9 MT |
959 | dev->stats.rx_length_errors++; |
960 | goto err_buf; | |
3f2c31d9 | 961 | } |
8fc3b9e9 | 962 | |
d46eeeaf | 963 | stats->bytes += len; |
8fc3b9e9 | 964 | page = virt_to_head_page(buf); |
28b39bc7 JW |
965 | |
966 | truesize = mergeable_ctx_to_truesize(ctx); | |
967 | if (unlikely(len > truesize)) { | |
56da5fd0 | 968 | pr_debug("%s: rx error: len %u exceeds truesize %lu\n", |
680557cf MT |
969 | dev->name, len, (unsigned long)ctx); |
970 | dev->stats.rx_length_errors++; | |
971 | goto err_skb; | |
972 | } | |
8fc3b9e9 MT |
973 | |
974 | num_skb_frags = skb_shinfo(curr_skb)->nr_frags; | |
2613af0e MD |
975 | if (unlikely(num_skb_frags == MAX_SKB_FRAGS)) { |
976 | struct sk_buff *nskb = alloc_skb(0, GFP_ATOMIC); | |
8fc3b9e9 MT |
977 | |
978 | if (unlikely(!nskb)) | |
979 | goto err_skb; | |
2613af0e MD |
980 | if (curr_skb == head_skb) |
981 | skb_shinfo(curr_skb)->frag_list = nskb; | |
982 | else | |
983 | curr_skb->next = nskb; | |
984 | curr_skb = nskb; | |
985 | head_skb->truesize += nskb->truesize; | |
986 | num_skb_frags = 0; | |
987 | } | |
988 | if (curr_skb != head_skb) { | |
989 | head_skb->data_len += len; | |
990 | head_skb->len += len; | |
fb51879d | 991 | head_skb->truesize += truesize; |
2613af0e | 992 | } |
8fc3b9e9 | 993 | offset = buf - page_address(page); |
ba275241 JW |
994 | if (skb_can_coalesce(curr_skb, num_skb_frags, page, offset)) { |
995 | put_page(page); | |
996 | skb_coalesce_rx_frag(curr_skb, num_skb_frags - 1, | |
fb51879d | 997 | len, truesize); |
ba275241 JW |
998 | } else { |
999 | skb_add_rx_frag(curr_skb, num_skb_frags, page, | |
fb51879d | 1000 | offset, len, truesize); |
ba275241 | 1001 | } |
8fc3b9e9 MT |
1002 | } |
1003 | ||
5377d758 | 1004 | ewma_pkt_len_add(&rq->mrg_avg_pkt_len, head_skb->len); |
8fc3b9e9 MT |
1005 | return head_skb; |
1006 | ||
f600b690 JF |
1007 | err_xdp: |
1008 | rcu_read_unlock(); | |
d46eeeaf | 1009 | stats->xdp_drops++; |
8fc3b9e9 MT |
1010 | err_skb: |
1011 | put_page(page); | |
850e088d | 1012 | while (num_buf-- > 1) { |
680557cf MT |
1013 | buf = virtqueue_get_buf(rq->vq, &len); |
1014 | if (unlikely(!buf)) { | |
8fc3b9e9 MT |
1015 | pr_debug("%s: rx error: %d buffers missing\n", |
1016 | dev->name, num_buf); | |
1017 | dev->stats.rx_length_errors++; | |
1018 | break; | |
1019 | } | |
d46eeeaf | 1020 | stats->bytes += len; |
680557cf | 1021 | page = virt_to_head_page(buf); |
8fc3b9e9 | 1022 | put_page(page); |
9ab86bbc | 1023 | } |
8fc3b9e9 | 1024 | err_buf: |
d46eeeaf | 1025 | stats->drops++; |
8fc3b9e9 | 1026 | dev_kfree_skb(head_skb); |
56434a01 | 1027 | xdp_xmit: |
8fc3b9e9 | 1028 | return NULL; |
9ab86bbc SM |
1029 | } |
1030 | ||
7d9d60fd TM |
1031 | static void receive_buf(struct virtnet_info *vi, struct receive_queue *rq, |
1032 | void *buf, unsigned int len, void **ctx, | |
a0929a44 | 1033 | unsigned int *xdp_xmit, |
d46eeeaf | 1034 | struct virtnet_rq_stats *stats) |
9ab86bbc | 1035 | { |
e9d7417b | 1036 | struct net_device *dev = vi->dev; |
9ab86bbc | 1037 | struct sk_buff *skb; |
012873d0 | 1038 | struct virtio_net_hdr_mrg_rxbuf *hdr; |
3f2c31d9 | 1039 | |
bcff3162 | 1040 | if (unlikely(len < vi->hdr_len + ETH_HLEN)) { |
9ab86bbc SM |
1041 | pr_debug("%s: short packet %i\n", dev->name, len); |
1042 | dev->stats.rx_length_errors++; | |
ab7db917 | 1043 | if (vi->mergeable_rx_bufs) { |
680557cf | 1044 | put_page(virt_to_head_page(buf)); |
ab7db917 | 1045 | } else if (vi->big_packets) { |
98bfd23c | 1046 | give_pages(rq, buf); |
ab7db917 | 1047 | } else { |
f6b10209 | 1048 | put_page(virt_to_head_page(buf)); |
ab7db917 | 1049 | } |
7d9d60fd | 1050 | return; |
9ab86bbc | 1051 | } |
3f2c31d9 | 1052 | |
f121159d | 1053 | if (vi->mergeable_rx_bufs) |
7d9d60fd | 1054 | skb = receive_mergeable(dev, vi, rq, buf, ctx, len, xdp_xmit, |
a0929a44 | 1055 | stats); |
f121159d | 1056 | else if (vi->big_packets) |
a0929a44 | 1057 | skb = receive_big(dev, vi, rq, buf, len, stats); |
f121159d | 1058 | else |
a0929a44 | 1059 | skb = receive_small(dev, vi, rq, buf, ctx, len, xdp_xmit, stats); |
f121159d MT |
1060 | |
1061 | if (unlikely(!skb)) | |
7d9d60fd | 1062 | return; |
3f2c31d9 | 1063 | |
9ab86bbc | 1064 | hdr = skb_vnet_hdr(skb); |
3fa2a1df | 1065 | |
e858fae2 | 1066 | if (hdr->hdr.flags & VIRTIO_NET_HDR_F_DATA_VALID) |
10a8d94a | 1067 | skb->ip_summed = CHECKSUM_UNNECESSARY; |
296f96fc | 1068 | |
e858fae2 MR |
1069 | if (virtio_net_hdr_to_skb(skb, &hdr->hdr, |
1070 | virtio_is_little_endian(vi->vdev))) { | |
1071 | net_warn_ratelimited("%s: bad gso: type: %u, size: %u\n", | |
1072 | dev->name, hdr->hdr.gso_type, | |
1073 | hdr->hdr.gso_size); | |
1074 | goto frame_err; | |
296f96fc RR |
1075 | } |
1076 | ||
133bbb18 | 1077 | skb_record_rx_queue(skb, vq2rxq(rq->vq)); |
d1dc06dc MR |
1078 | skb->protocol = eth_type_trans(skb, dev); |
1079 | pr_debug("Receiving skb proto 0x%04x len %i type %i\n", | |
1080 | ntohs(skb->protocol), skb->len, skb->pkt_type); | |
1081 | ||
0fbd050a | 1082 | napi_gro_receive(&rq->napi, skb); |
7d9d60fd | 1083 | return; |
296f96fc RR |
1084 | |
1085 | frame_err: | |
1086 | dev->stats.rx_frame_errors++; | |
296f96fc RR |
1087 | dev_kfree_skb(skb); |
1088 | } | |
1089 | ||
192f68cf JW |
1090 | /* Unlike mergeable buffers, all buffers are allocated to the |
1091 | * same size, except for the headroom. For this reason we do | |
1092 | * not need to use mergeable_len_to_ctx here - it is enough | |
1093 | * to store the headroom as the context ignoring the truesize. | |
1094 | */ | |
946fa564 MT |
1095 | static int add_recvbuf_small(struct virtnet_info *vi, struct receive_queue *rq, |
1096 | gfp_t gfp) | |
296f96fc | 1097 | { |
f6b10209 JW |
1098 | struct page_frag *alloc_frag = &rq->alloc_frag; |
1099 | char *buf; | |
2de2f7f4 | 1100 | unsigned int xdp_headroom = virtnet_get_headroom(vi); |
192f68cf | 1101 | void *ctx = (void *)(unsigned long)xdp_headroom; |
f6b10209 | 1102 | int len = vi->hdr_len + VIRTNET_RX_PAD + GOOD_PACKET_LEN + xdp_headroom; |
9ab86bbc | 1103 | int err; |
3f2c31d9 | 1104 | |
f6b10209 JW |
1105 | len = SKB_DATA_ALIGN(len) + |
1106 | SKB_DATA_ALIGN(sizeof(struct skb_shared_info)); | |
1107 | if (unlikely(!skb_page_frag_refill(len, alloc_frag, gfp))) | |
9ab86bbc | 1108 | return -ENOMEM; |
296f96fc | 1109 | |
f6b10209 JW |
1110 | buf = (char *)page_address(alloc_frag->page) + alloc_frag->offset; |
1111 | get_page(alloc_frag->page); | |
1112 | alloc_frag->offset += len; | |
1113 | sg_init_one(rq->sg, buf + VIRTNET_RX_PAD + xdp_headroom, | |
1114 | vi->hdr_len + GOOD_PACKET_LEN); | |
192f68cf | 1115 | err = virtqueue_add_inbuf_ctx(rq->vq, rq->sg, 1, buf, ctx, gfp); |
9ab86bbc | 1116 | if (err < 0) |
f6b10209 | 1117 | put_page(virt_to_head_page(buf)); |
9ab86bbc SM |
1118 | return err; |
1119 | } | |
97402b96 | 1120 | |
012873d0 MT |
1121 | static int add_recvbuf_big(struct virtnet_info *vi, struct receive_queue *rq, |
1122 | gfp_t gfp) | |
9ab86bbc | 1123 | { |
9ab86bbc SM |
1124 | struct page *first, *list = NULL; |
1125 | char *p; | |
1126 | int i, err, offset; | |
1127 | ||
a5835440 RR |
1128 | sg_init_table(rq->sg, MAX_SKB_FRAGS + 2); |
1129 | ||
e9d7417b | 1130 | /* page in rq->sg[MAX_SKB_FRAGS + 1] is list tail */ |
9ab86bbc | 1131 | for (i = MAX_SKB_FRAGS + 1; i > 1; --i) { |
e9d7417b | 1132 | first = get_a_page(rq, gfp); |
9ab86bbc SM |
1133 | if (!first) { |
1134 | if (list) | |
e9d7417b | 1135 | give_pages(rq, list); |
9ab86bbc | 1136 | return -ENOMEM; |
97402b96 | 1137 | } |
e9d7417b | 1138 | sg_set_buf(&rq->sg[i], page_address(first), PAGE_SIZE); |
97402b96 | 1139 | |
9ab86bbc SM |
1140 | /* chain new page in list head to match sg */ |
1141 | first->private = (unsigned long)list; | |
1142 | list = first; | |
1143 | } | |
296f96fc | 1144 | |
e9d7417b | 1145 | first = get_a_page(rq, gfp); |
9ab86bbc | 1146 | if (!first) { |
e9d7417b | 1147 | give_pages(rq, list); |
9ab86bbc SM |
1148 | return -ENOMEM; |
1149 | } | |
1150 | p = page_address(first); | |
1151 | ||
e9d7417b | 1152 | /* rq->sg[0], rq->sg[1] share the same page */ |
012873d0 MT |
1153 | /* a separated rq->sg[0] for header - required in case !any_header_sg */ |
1154 | sg_set_buf(&rq->sg[0], p, vi->hdr_len); | |
9ab86bbc | 1155 | |
e9d7417b | 1156 | /* rq->sg[1] for data packet, from offset */ |
9ab86bbc | 1157 | offset = sizeof(struct padded_vnet_hdr); |
e9d7417b | 1158 | sg_set_buf(&rq->sg[1], p + offset, PAGE_SIZE - offset); |
9ab86bbc SM |
1159 | |
1160 | /* chain first in list head */ | |
1161 | first->private = (unsigned long)list; | |
9dc7b9e4 RR |
1162 | err = virtqueue_add_inbuf(rq->vq, rq->sg, MAX_SKB_FRAGS + 2, |
1163 | first, gfp); | |
9ab86bbc | 1164 | if (err < 0) |
e9d7417b | 1165 | give_pages(rq, first); |
9ab86bbc SM |
1166 | |
1167 | return err; | |
296f96fc RR |
1168 | } |
1169 | ||
d85b758f | 1170 | static unsigned int get_mergeable_buf_len(struct receive_queue *rq, |
3cc81a9a JW |
1171 | struct ewma_pkt_len *avg_pkt_len, |
1172 | unsigned int room) | |
3f2c31d9 | 1173 | { |
ab7db917 | 1174 | const size_t hdr_len = sizeof(struct virtio_net_hdr_mrg_rxbuf); |
fbf28d78 MD |
1175 | unsigned int len; |
1176 | ||
3cc81a9a JW |
1177 | if (room) |
1178 | return PAGE_SIZE - room; | |
1179 | ||
1180 | len = hdr_len + clamp_t(unsigned int, ewma_pkt_len_read(avg_pkt_len), | |
f0c3192c | 1181 | rq->min_buf_len, PAGE_SIZE - hdr_len); |
3cc81a9a | 1182 | |
e377fcc8 | 1183 | return ALIGN(len, L1_CACHE_BYTES); |
fbf28d78 MD |
1184 | } |
1185 | ||
2de2f7f4 JF |
1186 | static int add_recvbuf_mergeable(struct virtnet_info *vi, |
1187 | struct receive_queue *rq, gfp_t gfp) | |
fbf28d78 | 1188 | { |
fb51879d | 1189 | struct page_frag *alloc_frag = &rq->alloc_frag; |
2de2f7f4 | 1190 | unsigned int headroom = virtnet_get_headroom(vi); |
3cc81a9a JW |
1191 | unsigned int tailroom = headroom ? sizeof(struct skb_shared_info) : 0; |
1192 | unsigned int room = SKB_DATA_ALIGN(headroom + tailroom); | |
fb51879d | 1193 | char *buf; |
680557cf | 1194 | void *ctx; |
3f2c31d9 | 1195 | int err; |
fb51879d | 1196 | unsigned int len, hole; |
3f2c31d9 | 1197 | |
3cc81a9a JW |
1198 | /* Extra tailroom is needed to satisfy XDP's assumption. This |
1199 | * means rx frags coalescing won't work, but consider we've | |
1200 | * disabled GSO for XDP, it won't be a big issue. | |
1201 | */ | |
1202 | len = get_mergeable_buf_len(rq, &rq->mrg_avg_pkt_len, room); | |
1203 | if (unlikely(!skb_page_frag_refill(len + room, alloc_frag, gfp))) | |
9ab86bbc | 1204 | return -ENOMEM; |
ab7db917 | 1205 | |
fb51879d | 1206 | buf = (char *)page_address(alloc_frag->page) + alloc_frag->offset; |
2de2f7f4 | 1207 | buf += headroom; /* advance address leaving hole at front of pkt */ |
fb51879d | 1208 | get_page(alloc_frag->page); |
3cc81a9a | 1209 | alloc_frag->offset += len + room; |
fb51879d | 1210 | hole = alloc_frag->size - alloc_frag->offset; |
3cc81a9a | 1211 | if (hole < len + room) { |
ab7db917 MD |
1212 | /* To avoid internal fragmentation, if there is very likely not |
1213 | * enough space for another buffer, add the remaining space to | |
1daa8790 | 1214 | * the current buffer. |
ab7db917 | 1215 | */ |
fb51879d MD |
1216 | len += hole; |
1217 | alloc_frag->offset += hole; | |
1218 | } | |
3f2c31d9 | 1219 | |
fb51879d | 1220 | sg_init_one(rq->sg, buf, len); |
29fda25a | 1221 | ctx = mergeable_len_to_ctx(len, headroom); |
680557cf | 1222 | err = virtqueue_add_inbuf_ctx(rq->vq, rq->sg, 1, buf, ctx, gfp); |
9ab86bbc | 1223 | if (err < 0) |
2613af0e | 1224 | put_page(virt_to_head_page(buf)); |
3f2c31d9 | 1225 | |
9ab86bbc SM |
1226 | return err; |
1227 | } | |
3f2c31d9 | 1228 | |
b2baed69 RR |
1229 | /* |
1230 | * Returns false if we couldn't fill entirely (OOM). | |
1231 | * | |
1232 | * Normally run in the receive path, but can also be run from ndo_open | |
1233 | * before we're receiving packets, or from refill_work which is | |
1234 | * careful to disable receiving (using napi_disable). | |
1235 | */ | |
946fa564 MT |
1236 | static bool try_fill_recv(struct virtnet_info *vi, struct receive_queue *rq, |
1237 | gfp_t gfp) | |
9ab86bbc SM |
1238 | { |
1239 | int err; | |
1788f495 | 1240 | bool oom; |
3f2c31d9 | 1241 | |
9ab86bbc SM |
1242 | do { |
1243 | if (vi->mergeable_rx_bufs) | |
2de2f7f4 | 1244 | err = add_recvbuf_mergeable(vi, rq, gfp); |
9ab86bbc | 1245 | else if (vi->big_packets) |
012873d0 | 1246 | err = add_recvbuf_big(vi, rq, gfp); |
9ab86bbc | 1247 | else |
946fa564 | 1248 | err = add_recvbuf_small(vi, rq, gfp); |
3f2c31d9 | 1249 | |
1788f495 | 1250 | oom = err == -ENOMEM; |
9ed4cb07 | 1251 | if (err) |
3f2c31d9 | 1252 | break; |
b7dfde95 | 1253 | } while (rq->vq->num_free); |
461f03dc TM |
1254 | if (virtqueue_kick_prepare(rq->vq) && virtqueue_notify(rq->vq)) { |
1255 | u64_stats_update_begin(&rq->stats.syncp); | |
d46eeeaf | 1256 | rq->stats.kicks++; |
461f03dc TM |
1257 | u64_stats_update_end(&rq->stats.syncp); |
1258 | } | |
1259 | ||
3161e453 | 1260 | return !oom; |
3f2c31d9 MM |
1261 | } |
1262 | ||
18445c4d | 1263 | static void skb_recv_done(struct virtqueue *rvq) |
296f96fc RR |
1264 | { |
1265 | struct virtnet_info *vi = rvq->vdev->priv; | |
986a4f4d | 1266 | struct receive_queue *rq = &vi->rq[vq2rxq(rvq)]; |
e9d7417b | 1267 | |
e4e8452a | 1268 | virtqueue_napi_schedule(&rq->napi, rvq); |
296f96fc RR |
1269 | } |
1270 | ||
e4e8452a | 1271 | static void virtnet_napi_enable(struct virtqueue *vq, struct napi_struct *napi) |
3e9d08ec | 1272 | { |
e4e8452a | 1273 | napi_enable(napi); |
3e9d08ec BR |
1274 | |
1275 | /* If all buffers were filled by other side before we napi_enabled, we | |
e4e8452a WB |
1276 | * won't get another interrupt, so process any outstanding packets now. |
1277 | * Call local_bh_enable after to trigger softIRQ processing. | |
1278 | */ | |
1279 | local_bh_disable(); | |
1280 | virtqueue_napi_schedule(napi, vq); | |
1281 | local_bh_enable(); | |
3e9d08ec BR |
1282 | } |
1283 | ||
b92f1e67 WB |
1284 | static void virtnet_napi_tx_enable(struct virtnet_info *vi, |
1285 | struct virtqueue *vq, | |
1286 | struct napi_struct *napi) | |
1287 | { | |
1288 | if (!napi->weight) | |
1289 | return; | |
1290 | ||
1291 | /* Tx napi touches cachelines on the cpu handling tx interrupts. Only | |
1292 | * enable the feature if this is likely affine with the transmit path. | |
1293 | */ | |
1294 | if (!vi->affinity_hint_set) { | |
1295 | napi->weight = 0; | |
1296 | return; | |
1297 | } | |
1298 | ||
1299 | return virtnet_napi_enable(vq, napi); | |
1300 | } | |
1301 | ||
78a57b48 WB |
1302 | static void virtnet_napi_tx_disable(struct napi_struct *napi) |
1303 | { | |
1304 | if (napi->weight) | |
1305 | napi_disable(napi); | |
1306 | } | |
1307 | ||
3161e453 RR |
1308 | static void refill_work(struct work_struct *work) |
1309 | { | |
e9d7417b JW |
1310 | struct virtnet_info *vi = |
1311 | container_of(work, struct virtnet_info, refill.work); | |
3161e453 | 1312 | bool still_empty; |
986a4f4d JW |
1313 | int i; |
1314 | ||
55257d72 | 1315 | for (i = 0; i < vi->curr_queue_pairs; i++) { |
986a4f4d | 1316 | struct receive_queue *rq = &vi->rq[i]; |
3161e453 | 1317 | |
986a4f4d | 1318 | napi_disable(&rq->napi); |
946fa564 | 1319 | still_empty = !try_fill_recv(vi, rq, GFP_KERNEL); |
e4e8452a | 1320 | virtnet_napi_enable(rq->vq, &rq->napi); |
3161e453 | 1321 | |
986a4f4d JW |
1322 | /* In theory, this can happen: if we don't get any buffers in |
1323 | * we will *never* try to fill again. | |
1324 | */ | |
1325 | if (still_empty) | |
1326 | schedule_delayed_work(&vi->refill, HZ/2); | |
1327 | } | |
3161e453 RR |
1328 | } |
1329 | ||
2471c75e JDB |
1330 | static int virtnet_receive(struct receive_queue *rq, int budget, |
1331 | unsigned int *xdp_xmit) | |
296f96fc | 1332 | { |
e9d7417b | 1333 | struct virtnet_info *vi = rq->vq->vdev->priv; |
d46eeeaf | 1334 | struct virtnet_rq_stats stats = {}; |
a0929a44 | 1335 | unsigned int len; |
9ab86bbc | 1336 | void *buf; |
a0929a44 | 1337 | int i; |
296f96fc | 1338 | |
192f68cf | 1339 | if (!vi->big_packets || vi->mergeable_rx_bufs) { |
680557cf MT |
1340 | void *ctx; |
1341 | ||
d46eeeaf | 1342 | while (stats.packets < budget && |
680557cf | 1343 | (buf = virtqueue_get_buf_ctx(rq->vq, &len, &ctx))) { |
a0929a44 | 1344 | receive_buf(vi, rq, buf, len, ctx, xdp_xmit, &stats); |
d46eeeaf | 1345 | stats.packets++; |
680557cf MT |
1346 | } |
1347 | } else { | |
d46eeeaf | 1348 | while (stats.packets < budget && |
680557cf | 1349 | (buf = virtqueue_get_buf(rq->vq, &len)) != NULL) { |
a0929a44 | 1350 | receive_buf(vi, rq, buf, len, NULL, xdp_xmit, &stats); |
d46eeeaf | 1351 | stats.packets++; |
680557cf | 1352 | } |
296f96fc RR |
1353 | } |
1354 | ||
718be6ba | 1355 | if (rq->vq->num_free > min((unsigned int)budget, virtqueue_get_vring_size(rq->vq)) / 2) { |
946fa564 | 1356 | if (!try_fill_recv(vi, rq, GFP_ATOMIC)) |
3b07e9ca | 1357 | schedule_delayed_work(&vi->refill, 0); |
3161e453 | 1358 | } |
296f96fc | 1359 | |
d7dfc5cf | 1360 | u64_stats_update_begin(&rq->stats.syncp); |
a0929a44 TM |
1361 | for (i = 0; i < VIRTNET_RQ_STATS_LEN; i++) { |
1362 | size_t offset = virtnet_rq_stats_desc[i].offset; | |
1363 | u64 *item; | |
1364 | ||
d46eeeaf JW |
1365 | item = (u64 *)((u8 *)&rq->stats + offset); |
1366 | *item += *(u64 *)((u8 *)&stats + offset); | |
a0929a44 | 1367 | } |
d7dfc5cf | 1368 | u64_stats_update_end(&rq->stats.syncp); |
61845d20 | 1369 | |
d46eeeaf | 1370 | return stats.packets; |
2ffa7598 JW |
1371 | } |
1372 | ||
df133f3f | 1373 | static void free_old_xmit_skbs(struct send_queue *sq, bool in_napi) |
ea7735d9 | 1374 | { |
ea7735d9 | 1375 | unsigned int len; |
ea7735d9 WB |
1376 | unsigned int packets = 0; |
1377 | unsigned int bytes = 0; | |
5050471d | 1378 | void *ptr; |
ea7735d9 | 1379 | |
5050471d TM |
1380 | while ((ptr = virtqueue_get_buf(sq->vq, &len)) != NULL) { |
1381 | if (likely(!is_xdp_frame(ptr))) { | |
1382 | struct sk_buff *skb = ptr; | |
ea7735d9 | 1383 | |
5050471d | 1384 | pr_debug("Sent skb %p\n", skb); |
ea7735d9 | 1385 | |
5050471d TM |
1386 | bytes += skb->len; |
1387 | napi_consume_skb(skb, in_napi); | |
1388 | } else { | |
1389 | struct xdp_frame *frame = ptr_to_xdp(ptr); | |
ea7735d9 | 1390 | |
5050471d TM |
1391 | bytes += frame->len; |
1392 | xdp_return_frame(frame); | |
1393 | } | |
1394 | packets++; | |
ea7735d9 WB |
1395 | } |
1396 | ||
1397 | /* Avoid overhead when no packets have been processed | |
1398 | * happens when called speculatively from start_xmit. | |
1399 | */ | |
1400 | if (!packets) | |
1401 | return; | |
1402 | ||
d7dfc5cf TM |
1403 | u64_stats_update_begin(&sq->stats.syncp); |
1404 | sq->stats.bytes += bytes; | |
1405 | sq->stats.packets += packets; | |
1406 | u64_stats_update_end(&sq->stats.syncp); | |
ea7735d9 WB |
1407 | } |
1408 | ||
534da5e8 TM |
1409 | static bool is_xdp_raw_buffer_queue(struct virtnet_info *vi, int q) |
1410 | { | |
1411 | if (q < (vi->curr_queue_pairs - vi->xdp_queue_pairs)) | |
1412 | return false; | |
1413 | else if (q < vi->curr_queue_pairs) | |
1414 | return true; | |
1415 | else | |
1416 | return false; | |
1417 | } | |
1418 | ||
7b0411ef WB |
1419 | static void virtnet_poll_cleantx(struct receive_queue *rq) |
1420 | { | |
1421 | struct virtnet_info *vi = rq->vq->vdev->priv; | |
1422 | unsigned int index = vq2rxq(rq->vq); | |
1423 | struct send_queue *sq = &vi->sq[index]; | |
1424 | struct netdev_queue *txq = netdev_get_tx_queue(vi->dev, index); | |
1425 | ||
534da5e8 | 1426 | if (!sq->napi.weight || is_xdp_raw_buffer_queue(vi, index)) |
7b0411ef WB |
1427 | return; |
1428 | ||
1429 | if (__netif_tx_trylock(txq)) { | |
df133f3f | 1430 | free_old_xmit_skbs(sq, true); |
7b0411ef WB |
1431 | __netif_tx_unlock(txq); |
1432 | } | |
1433 | ||
1434 | if (sq->vq->num_free >= 2 + MAX_SKB_FRAGS) | |
1435 | netif_tx_wake_queue(txq); | |
1436 | } | |
1437 | ||
2ffa7598 JW |
1438 | static int virtnet_poll(struct napi_struct *napi, int budget) |
1439 | { | |
1440 | struct receive_queue *rq = | |
1441 | container_of(napi, struct receive_queue, napi); | |
9267c430 JW |
1442 | struct virtnet_info *vi = rq->vq->vdev->priv; |
1443 | struct send_queue *sq; | |
2a43565c | 1444 | unsigned int received; |
2471c75e | 1445 | unsigned int xdp_xmit = 0; |
2ffa7598 | 1446 | |
7b0411ef WB |
1447 | virtnet_poll_cleantx(rq); |
1448 | ||
186b3c99 | 1449 | received = virtnet_receive(rq, budget, &xdp_xmit); |
2ffa7598 | 1450 | |
8329d98e | 1451 | /* Out of packets? */ |
e4e8452a WB |
1452 | if (received < budget) |
1453 | virtqueue_napi_complete(napi, rq->vq, received); | |
296f96fc | 1454 | |
2471c75e | 1455 | if (xdp_xmit & VIRTIO_XDP_REDIR) |
1d233886 | 1456 | xdp_do_flush(); |
2471c75e JDB |
1457 | |
1458 | if (xdp_xmit & VIRTIO_XDP_TX) { | |
2a43565c | 1459 | sq = virtnet_xdp_sq(vi); |
461f03dc TM |
1460 | if (virtqueue_kick_prepare(sq->vq) && virtqueue_notify(sq->vq)) { |
1461 | u64_stats_update_begin(&sq->stats.syncp); | |
1462 | sq->stats.kicks++; | |
1463 | u64_stats_update_end(&sq->stats.syncp); | |
1464 | } | |
9267c430 | 1465 | } |
186b3c99 | 1466 | |
296f96fc RR |
1467 | return received; |
1468 | } | |
1469 | ||
986a4f4d JW |
1470 | static int virtnet_open(struct net_device *dev) |
1471 | { | |
1472 | struct virtnet_info *vi = netdev_priv(dev); | |
754b8a21 | 1473 | int i, err; |
986a4f4d | 1474 | |
e4166625 JW |
1475 | for (i = 0; i < vi->max_queue_pairs; i++) { |
1476 | if (i < vi->curr_queue_pairs) | |
1477 | /* Make sure we have some buffers: if oom use wq. */ | |
946fa564 | 1478 | if (!try_fill_recv(vi, &vi->rq[i], GFP_KERNEL)) |
e4166625 | 1479 | schedule_delayed_work(&vi->refill, 0); |
754b8a21 JDB |
1480 | |
1481 | err = xdp_rxq_info_reg(&vi->rq[i].xdp_rxq, dev, i); | |
1482 | if (err < 0) | |
1483 | return err; | |
1484 | ||
8d5d8852 JDB |
1485 | err = xdp_rxq_info_reg_mem_model(&vi->rq[i].xdp_rxq, |
1486 | MEM_TYPE_PAGE_SHARED, NULL); | |
1487 | if (err < 0) { | |
1488 | xdp_rxq_info_unreg(&vi->rq[i].xdp_rxq); | |
1489 | return err; | |
1490 | } | |
1491 | ||
e4e8452a | 1492 | virtnet_napi_enable(vi->rq[i].vq, &vi->rq[i].napi); |
b92f1e67 | 1493 | virtnet_napi_tx_enable(vi, vi->sq[i].vq, &vi->sq[i].napi); |
986a4f4d JW |
1494 | } |
1495 | ||
1496 | return 0; | |
1497 | } | |
1498 | ||
b92f1e67 WB |
1499 | static int virtnet_poll_tx(struct napi_struct *napi, int budget) |
1500 | { | |
1501 | struct send_queue *sq = container_of(napi, struct send_queue, napi); | |
1502 | struct virtnet_info *vi = sq->vq->vdev->priv; | |
534da5e8 TM |
1503 | unsigned int index = vq2txq(sq->vq); |
1504 | struct netdev_queue *txq; | |
b92f1e67 | 1505 | |
534da5e8 TM |
1506 | if (unlikely(is_xdp_raw_buffer_queue(vi, index))) { |
1507 | /* We don't need to enable cb for XDP */ | |
1508 | napi_complete_done(napi, 0); | |
1509 | return 0; | |
1510 | } | |
1511 | ||
1512 | txq = netdev_get_tx_queue(vi->dev, index); | |
b92f1e67 | 1513 | __netif_tx_lock(txq, raw_smp_processor_id()); |
df133f3f | 1514 | free_old_xmit_skbs(sq, true); |
b92f1e67 WB |
1515 | __netif_tx_unlock(txq); |
1516 | ||
1517 | virtqueue_napi_complete(napi, sq->vq, 0); | |
1518 | ||
1519 | if (sq->vq->num_free >= 2 + MAX_SKB_FRAGS) | |
1520 | netif_tx_wake_queue(txq); | |
1521 | ||
1522 | return 0; | |
1523 | } | |
1524 | ||
e9d7417b | 1525 | static int xmit_skb(struct send_queue *sq, struct sk_buff *skb) |
296f96fc | 1526 | { |
012873d0 | 1527 | struct virtio_net_hdr_mrg_rxbuf *hdr; |
296f96fc | 1528 | const unsigned char *dest = ((struct ethhdr *)skb->data)->h_dest; |
e9d7417b | 1529 | struct virtnet_info *vi = sq->vq->vdev->priv; |
e2fcad58 | 1530 | int num_sg; |
012873d0 | 1531 | unsigned hdr_len = vi->hdr_len; |
e7428e95 | 1532 | bool can_push; |
296f96fc | 1533 | |
e174961c | 1534 | pr_debug("%s: xmit %p %pM\n", vi->dev->name, skb, dest); |
e7428e95 MT |
1535 | |
1536 | can_push = vi->any_header_sg && | |
1537 | !((unsigned long)skb->data & (__alignof__(*hdr) - 1)) && | |
1538 | !skb_header_cloned(skb) && skb_headroom(skb) >= hdr_len; | |
1539 | /* Even if we can, don't push here yet as this would skew | |
1540 | * csum_start offset below. */ | |
1541 | if (can_push) | |
012873d0 | 1542 | hdr = (struct virtio_net_hdr_mrg_rxbuf *)(skb->data - hdr_len); |
e7428e95 MT |
1543 | else |
1544 | hdr = skb_vnet_hdr(skb); | |
296f96fc | 1545 | |
e858fae2 | 1546 | if (virtio_net_hdr_from_skb(skb, &hdr->hdr, |
fd3a8862 WB |
1547 | virtio_is_little_endian(vi->vdev), false, |
1548 | 0)) | |
e858fae2 | 1549 | BUG(); |
296f96fc | 1550 | |
3f2c31d9 | 1551 | if (vi->mergeable_rx_bufs) |
012873d0 | 1552 | hdr->num_buffers = 0; |
3f2c31d9 | 1553 | |
547c890c | 1554 | sg_init_table(sq->sg, skb_shinfo(skb)->nr_frags + (can_push ? 1 : 2)); |
e7428e95 MT |
1555 | if (can_push) { |
1556 | __skb_push(skb, hdr_len); | |
1557 | num_sg = skb_to_sgvec(skb, sq->sg, 0, skb->len); | |
e2fcad58 JD |
1558 | if (unlikely(num_sg < 0)) |
1559 | return num_sg; | |
e7428e95 MT |
1560 | /* Pull header back to avoid skew in tx bytes calculations. */ |
1561 | __skb_pull(skb, hdr_len); | |
1562 | } else { | |
1563 | sg_set_buf(sq->sg, hdr, hdr_len); | |
e2fcad58 JD |
1564 | num_sg = skb_to_sgvec(skb, sq->sg + 1, 0, skb->len); |
1565 | if (unlikely(num_sg < 0)) | |
1566 | return num_sg; | |
1567 | num_sg++; | |
e7428e95 | 1568 | } |
9dc7b9e4 | 1569 | return virtqueue_add_outbuf(sq->vq, sq->sg, num_sg, skb, GFP_ATOMIC); |
11a3a154 RR |
1570 | } |
1571 | ||
424efe9c | 1572 | static netdev_tx_t start_xmit(struct sk_buff *skb, struct net_device *dev) |
99ffc696 RR |
1573 | { |
1574 | struct virtnet_info *vi = netdev_priv(dev); | |
986a4f4d JW |
1575 | int qnum = skb_get_queue_mapping(skb); |
1576 | struct send_queue *sq = &vi->sq[qnum]; | |
9ed4cb07 | 1577 | int err; |
4b7fd2e6 | 1578 | struct netdev_queue *txq = netdev_get_tx_queue(dev, qnum); |
6b16f9ee | 1579 | bool kick = !netdev_xmit_more(); |
b92f1e67 | 1580 | bool use_napi = sq->napi.weight; |
2cb9c6ba | 1581 | |
2cb9c6ba | 1582 | /* Free up any pending old buffers before queueing new ones. */ |
df133f3f | 1583 | free_old_xmit_skbs(sq, false); |
99ffc696 | 1584 | |
bdb12e0d WB |
1585 | if (use_napi && kick) |
1586 | virtqueue_enable_cb_delayed(sq->vq); | |
1587 | ||
074c3582 JK |
1588 | /* timestamp packet in software */ |
1589 | skb_tx_timestamp(skb); | |
1590 | ||
03f191ba | 1591 | /* Try to transmit */ |
b7dfde95 | 1592 | err = xmit_skb(sq, skb); |
48925e37 | 1593 | |
9ed4cb07 | 1594 | /* This should not happen! */ |
681daee2 | 1595 | if (unlikely(err)) { |
9ed4cb07 RR |
1596 | dev->stats.tx_fifo_errors++; |
1597 | if (net_ratelimit()) | |
1598 | dev_warn(&dev->dev, | |
7934b481 YS |
1599 | "Unexpected TXQ (%d) queue failure: %d\n", |
1600 | qnum, err); | |
58eba97d | 1601 | dev->stats.tx_dropped++; |
85e94525 | 1602 | dev_kfree_skb_any(skb); |
58eba97d | 1603 | return NETDEV_TX_OK; |
296f96fc | 1604 | } |
03f191ba | 1605 | |
48925e37 | 1606 | /* Don't wait up for transmitted skbs to be freed. */ |
b92f1e67 WB |
1607 | if (!use_napi) { |
1608 | skb_orphan(skb); | |
895b5c9f | 1609 | nf_reset_ct(skb); |
b92f1e67 | 1610 | } |
48925e37 | 1611 | |
60302ff6 MT |
1612 | /* If running out of space, stop queue to avoid getting packets that we |
1613 | * are then unable to transmit. | |
1614 | * An alternative would be to force queuing layer to requeue the skb by | |
1615 | * returning NETDEV_TX_BUSY. However, NETDEV_TX_BUSY should not be | |
1616 | * returned in a normal path of operation: it means that driver is not | |
1617 | * maintaining the TX queue stop/start state properly, and causes | |
1618 | * the stack to do a non-trivial amount of useless work. | |
1619 | * Since most packets only take 1 or 2 ring slots, stopping the queue | |
1620 | * early means 16 slots are typically wasted. | |
d631b94e | 1621 | */ |
b7dfde95 | 1622 | if (sq->vq->num_free < 2+MAX_SKB_FRAGS) { |
986a4f4d | 1623 | netif_stop_subqueue(dev, qnum); |
b92f1e67 WB |
1624 | if (!use_napi && |
1625 | unlikely(!virtqueue_enable_cb_delayed(sq->vq))) { | |
48925e37 | 1626 | /* More just got used, free them then recheck. */ |
df133f3f | 1627 | free_old_xmit_skbs(sq, false); |
b7dfde95 | 1628 | if (sq->vq->num_free >= 2+MAX_SKB_FRAGS) { |
986a4f4d | 1629 | netif_start_subqueue(dev, qnum); |
e9d7417b | 1630 | virtqueue_disable_cb(sq->vq); |
48925e37 RR |
1631 | } |
1632 | } | |
99ffc696 | 1633 | } |
48925e37 | 1634 | |
461f03dc TM |
1635 | if (kick || netif_xmit_stopped(txq)) { |
1636 | if (virtqueue_kick_prepare(sq->vq) && virtqueue_notify(sq->vq)) { | |
1637 | u64_stats_update_begin(&sq->stats.syncp); | |
1638 | sq->stats.kicks++; | |
1639 | u64_stats_update_end(&sq->stats.syncp); | |
1640 | } | |
1641 | } | |
296f96fc | 1642 | |
0b725a2c | 1643 | return NETDEV_TX_OK; |
c223a078 DM |
1644 | } |
1645 | ||
40cbfc37 AK |
1646 | /* |
1647 | * Send command via the control virtqueue and check status. Commands | |
1648 | * supported by the hypervisor, as indicated by feature bits, should | |
788a8b6d | 1649 | * never fail unless improperly formatted. |
40cbfc37 AK |
1650 | */ |
1651 | static bool virtnet_send_command(struct virtnet_info *vi, u8 class, u8 cmd, | |
d24bae32 | 1652 | struct scatterlist *out) |
40cbfc37 | 1653 | { |
f7bc9594 | 1654 | struct scatterlist *sgs[4], hdr, stat; |
d24bae32 | 1655 | unsigned out_num = 0, tmp; |
40cbfc37 AK |
1656 | |
1657 | /* Caller should know better */ | |
f7bc9594 | 1658 | BUG_ON(!virtio_has_feature(vi->vdev, VIRTIO_NET_F_CTRL_VQ)); |
40cbfc37 | 1659 | |
12e57169 MT |
1660 | vi->ctrl->status = ~0; |
1661 | vi->ctrl->hdr.class = class; | |
1662 | vi->ctrl->hdr.cmd = cmd; | |
f7bc9594 | 1663 | /* Add header */ |
12e57169 | 1664 | sg_init_one(&hdr, &vi->ctrl->hdr, sizeof(vi->ctrl->hdr)); |
f7bc9594 | 1665 | sgs[out_num++] = &hdr; |
40cbfc37 | 1666 | |
f7bc9594 RR |
1667 | if (out) |
1668 | sgs[out_num++] = out; | |
40cbfc37 | 1669 | |
f7bc9594 | 1670 | /* Add return status. */ |
12e57169 | 1671 | sg_init_one(&stat, &vi->ctrl->status, sizeof(vi->ctrl->status)); |
d24bae32 | 1672 | sgs[out_num] = &stat; |
40cbfc37 | 1673 | |
d24bae32 | 1674 | BUG_ON(out_num + 1 > ARRAY_SIZE(sgs)); |
a7c58146 | 1675 | virtqueue_add_sgs(vi->cvq, sgs, out_num, 1, vi, GFP_ATOMIC); |
40cbfc37 | 1676 | |
67975901 | 1677 | if (unlikely(!virtqueue_kick(vi->cvq))) |
12e57169 | 1678 | return vi->ctrl->status == VIRTIO_NET_OK; |
40cbfc37 AK |
1679 | |
1680 | /* Spin for a response, the kick causes an ioport write, trapping | |
1681 | * into the hypervisor, so the request should be handled immediately. | |
1682 | */ | |
047b9b94 HG |
1683 | while (!virtqueue_get_buf(vi->cvq, &tmp) && |
1684 | !virtqueue_is_broken(vi->cvq)) | |
40cbfc37 AK |
1685 | cpu_relax(); |
1686 | ||
12e57169 | 1687 | return vi->ctrl->status == VIRTIO_NET_OK; |
40cbfc37 AK |
1688 | } |
1689 | ||
9c46f6d4 AW |
1690 | static int virtnet_set_mac_address(struct net_device *dev, void *p) |
1691 | { | |
1692 | struct virtnet_info *vi = netdev_priv(dev); | |
1693 | struct virtio_device *vdev = vi->vdev; | |
f2f2c8b4 | 1694 | int ret; |
e37e2ff3 | 1695 | struct sockaddr *addr; |
7e58d5ae | 1696 | struct scatterlist sg; |
9c46f6d4 | 1697 | |
ba5e4426 SS |
1698 | if (virtio_has_feature(vi->vdev, VIRTIO_NET_F_STANDBY)) |
1699 | return -EOPNOTSUPP; | |
1700 | ||
801822d1 | 1701 | addr = kmemdup(p, sizeof(*addr), GFP_KERNEL); |
e37e2ff3 AL |
1702 | if (!addr) |
1703 | return -ENOMEM; | |
e37e2ff3 AL |
1704 | |
1705 | ret = eth_prepare_mac_addr_change(dev, addr); | |
f2f2c8b4 | 1706 | if (ret) |
e37e2ff3 | 1707 | goto out; |
9c46f6d4 | 1708 | |
7e58d5ae AK |
1709 | if (virtio_has_feature(vdev, VIRTIO_NET_F_CTRL_MAC_ADDR)) { |
1710 | sg_init_one(&sg, addr->sa_data, dev->addr_len); | |
1711 | if (!virtnet_send_command(vi, VIRTIO_NET_CTRL_MAC, | |
d24bae32 | 1712 | VIRTIO_NET_CTRL_MAC_ADDR_SET, &sg)) { |
7e58d5ae AK |
1713 | dev_warn(&vdev->dev, |
1714 | "Failed to set mac address by vq command.\n"); | |
e37e2ff3 AL |
1715 | ret = -EINVAL; |
1716 | goto out; | |
7e58d5ae | 1717 | } |
7e93a02f MT |
1718 | } else if (virtio_has_feature(vdev, VIRTIO_NET_F_MAC) && |
1719 | !virtio_has_feature(vdev, VIRTIO_F_VERSION_1)) { | |
855e0c52 RR |
1720 | unsigned int i; |
1721 | ||
1722 | /* Naturally, this has an atomicity problem. */ | |
1723 | for (i = 0; i < dev->addr_len; i++) | |
1724 | virtio_cwrite8(vdev, | |
1725 | offsetof(struct virtio_net_config, mac) + | |
1726 | i, addr->sa_data[i]); | |
7e58d5ae AK |
1727 | } |
1728 | ||
1729 | eth_commit_mac_addr_change(dev, p); | |
e37e2ff3 | 1730 | ret = 0; |
9c46f6d4 | 1731 | |
e37e2ff3 AL |
1732 | out: |
1733 | kfree(addr); | |
1734 | return ret; | |
9c46f6d4 AW |
1735 | } |
1736 | ||
bc1f4470 | 1737 | static void virtnet_stats(struct net_device *dev, |
1738 | struct rtnl_link_stats64 *tot) | |
3fa2a1df | 1739 | { |
1740 | struct virtnet_info *vi = netdev_priv(dev); | |
3fa2a1df | 1741 | unsigned int start; |
d7dfc5cf | 1742 | int i; |
3fa2a1df | 1743 | |
d7dfc5cf | 1744 | for (i = 0; i < vi->max_queue_pairs; i++) { |
2c4a2f7d | 1745 | u64 tpackets, tbytes, rpackets, rbytes, rdrops; |
d7dfc5cf TM |
1746 | struct receive_queue *rq = &vi->rq[i]; |
1747 | struct send_queue *sq = &vi->sq[i]; | |
3fa2a1df | 1748 | |
1749 | do { | |
d7dfc5cf TM |
1750 | start = u64_stats_fetch_begin_irq(&sq->stats.syncp); |
1751 | tpackets = sq->stats.packets; | |
1752 | tbytes = sq->stats.bytes; | |
1753 | } while (u64_stats_fetch_retry_irq(&sq->stats.syncp, start)); | |
83a27052 ED |
1754 | |
1755 | do { | |
d7dfc5cf | 1756 | start = u64_stats_fetch_begin_irq(&rq->stats.syncp); |
d46eeeaf JW |
1757 | rpackets = rq->stats.packets; |
1758 | rbytes = rq->stats.bytes; | |
1759 | rdrops = rq->stats.drops; | |
d7dfc5cf | 1760 | } while (u64_stats_fetch_retry_irq(&rq->stats.syncp, start)); |
3fa2a1df | 1761 | |
1762 | tot->rx_packets += rpackets; | |
1763 | tot->tx_packets += tpackets; | |
1764 | tot->rx_bytes += rbytes; | |
1765 | tot->tx_bytes += tbytes; | |
2c4a2f7d | 1766 | tot->rx_dropped += rdrops; |
3fa2a1df | 1767 | } |
1768 | ||
1769 | tot->tx_dropped = dev->stats.tx_dropped; | |
021ac8d3 | 1770 | tot->tx_fifo_errors = dev->stats.tx_fifo_errors; |
3fa2a1df | 1771 | tot->rx_length_errors = dev->stats.rx_length_errors; |
1772 | tot->rx_frame_errors = dev->stats.rx_frame_errors; | |
3fa2a1df | 1773 | } |
1774 | ||
586d17c5 JW |
1775 | static void virtnet_ack_link_announce(struct virtnet_info *vi) |
1776 | { | |
1777 | rtnl_lock(); | |
1778 | if (!virtnet_send_command(vi, VIRTIO_NET_CTRL_ANNOUNCE, | |
d24bae32 | 1779 | VIRTIO_NET_CTRL_ANNOUNCE_ACK, NULL)) |
586d17c5 JW |
1780 | dev_warn(&vi->dev->dev, "Failed to ack link announce.\n"); |
1781 | rtnl_unlock(); | |
1782 | } | |
1783 | ||
47315329 | 1784 | static int _virtnet_set_queues(struct virtnet_info *vi, u16 queue_pairs) |
986a4f4d JW |
1785 | { |
1786 | struct scatterlist sg; | |
986a4f4d JW |
1787 | struct net_device *dev = vi->dev; |
1788 | ||
1789 | if (!vi->has_cvq || !virtio_has_feature(vi->vdev, VIRTIO_NET_F_MQ)) | |
1790 | return 0; | |
1791 | ||
12e57169 MT |
1792 | vi->ctrl->mq.virtqueue_pairs = cpu_to_virtio16(vi->vdev, queue_pairs); |
1793 | sg_init_one(&sg, &vi->ctrl->mq, sizeof(vi->ctrl->mq)); | |
986a4f4d JW |
1794 | |
1795 | if (!virtnet_send_command(vi, VIRTIO_NET_CTRL_MQ, | |
d24bae32 | 1796 | VIRTIO_NET_CTRL_MQ_VQ_PAIRS_SET, &sg)) { |
986a4f4d JW |
1797 | dev_warn(&dev->dev, "Fail to set num of queue pairs to %d\n", |
1798 | queue_pairs); | |
1799 | return -EINVAL; | |
55257d72 | 1800 | } else { |
986a4f4d | 1801 | vi->curr_queue_pairs = queue_pairs; |
35ed159b JW |
1802 | /* virtnet_open() will refill when device is going to up. */ |
1803 | if (dev->flags & IFF_UP) | |
1804 | schedule_delayed_work(&vi->refill, 0); | |
55257d72 | 1805 | } |
986a4f4d JW |
1806 | |
1807 | return 0; | |
1808 | } | |
1809 | ||
47315329 JF |
1810 | static int virtnet_set_queues(struct virtnet_info *vi, u16 queue_pairs) |
1811 | { | |
1812 | int err; | |
1813 | ||
1814 | rtnl_lock(); | |
1815 | err = _virtnet_set_queues(vi, queue_pairs); | |
1816 | rtnl_unlock(); | |
1817 | return err; | |
1818 | } | |
1819 | ||
296f96fc RR |
1820 | static int virtnet_close(struct net_device *dev) |
1821 | { | |
1822 | struct virtnet_info *vi = netdev_priv(dev); | |
986a4f4d | 1823 | int i; |
296f96fc | 1824 | |
b2baed69 RR |
1825 | /* Make sure refill_work doesn't re-enable napi! */ |
1826 | cancel_delayed_work_sync(&vi->refill); | |
986a4f4d | 1827 | |
b92f1e67 | 1828 | for (i = 0; i < vi->max_queue_pairs; i++) { |
754b8a21 | 1829 | xdp_rxq_info_unreg(&vi->rq[i].xdp_rxq); |
986a4f4d | 1830 | napi_disable(&vi->rq[i].napi); |
78a57b48 | 1831 | virtnet_napi_tx_disable(&vi->sq[i].napi); |
b92f1e67 | 1832 | } |
296f96fc | 1833 | |
296f96fc RR |
1834 | return 0; |
1835 | } | |
1836 | ||
2af7698e AW |
1837 | static void virtnet_set_rx_mode(struct net_device *dev) |
1838 | { | |
1839 | struct virtnet_info *vi = netdev_priv(dev); | |
f565a7c2 | 1840 | struct scatterlist sg[2]; |
f565a7c2 | 1841 | struct virtio_net_ctrl_mac *mac_data; |
ccffad25 | 1842 | struct netdev_hw_addr *ha; |
32e7bfc4 | 1843 | int uc_count; |
4cd24eaf | 1844 | int mc_count; |
f565a7c2 AW |
1845 | void *buf; |
1846 | int i; | |
2af7698e | 1847 | |
788a8b6d | 1848 | /* We can't dynamically set ndo_set_rx_mode, so return gracefully */ |
2af7698e AW |
1849 | if (!virtio_has_feature(vi->vdev, VIRTIO_NET_F_CTRL_RX)) |
1850 | return; | |
1851 | ||
12e57169 MT |
1852 | vi->ctrl->promisc = ((dev->flags & IFF_PROMISC) != 0); |
1853 | vi->ctrl->allmulti = ((dev->flags & IFF_ALLMULTI) != 0); | |
2af7698e | 1854 | |
12e57169 | 1855 | sg_init_one(sg, &vi->ctrl->promisc, sizeof(vi->ctrl->promisc)); |
2af7698e AW |
1856 | |
1857 | if (!virtnet_send_command(vi, VIRTIO_NET_CTRL_RX, | |
d24bae32 | 1858 | VIRTIO_NET_CTRL_RX_PROMISC, sg)) |
2af7698e | 1859 | dev_warn(&dev->dev, "Failed to %sable promisc mode.\n", |
12e57169 | 1860 | vi->ctrl->promisc ? "en" : "dis"); |
2af7698e | 1861 | |
12e57169 | 1862 | sg_init_one(sg, &vi->ctrl->allmulti, sizeof(vi->ctrl->allmulti)); |
2af7698e AW |
1863 | |
1864 | if (!virtnet_send_command(vi, VIRTIO_NET_CTRL_RX, | |
d24bae32 | 1865 | VIRTIO_NET_CTRL_RX_ALLMULTI, sg)) |
2af7698e | 1866 | dev_warn(&dev->dev, "Failed to %sable allmulti mode.\n", |
12e57169 | 1867 | vi->ctrl->allmulti ? "en" : "dis"); |
f565a7c2 | 1868 | |
32e7bfc4 | 1869 | uc_count = netdev_uc_count(dev); |
4cd24eaf | 1870 | mc_count = netdev_mc_count(dev); |
f565a7c2 | 1871 | /* MAC filter - use one buffer for both lists */ |
4cd24eaf JP |
1872 | buf = kzalloc(((uc_count + mc_count) * ETH_ALEN) + |
1873 | (2 * sizeof(mac_data->entries)), GFP_ATOMIC); | |
1874 | mac_data = buf; | |
e68ed8f0 | 1875 | if (!buf) |
f565a7c2 | 1876 | return; |
f565a7c2 | 1877 | |
23e258e1 AW |
1878 | sg_init_table(sg, 2); |
1879 | ||
f565a7c2 | 1880 | /* Store the unicast list and count in the front of the buffer */ |
fdd819b2 | 1881 | mac_data->entries = cpu_to_virtio32(vi->vdev, uc_count); |
ccffad25 | 1882 | i = 0; |
32e7bfc4 | 1883 | netdev_for_each_uc_addr(ha, dev) |
ccffad25 | 1884 | memcpy(&mac_data->macs[i++][0], ha->addr, ETH_ALEN); |
f565a7c2 AW |
1885 | |
1886 | sg_set_buf(&sg[0], mac_data, | |
32e7bfc4 | 1887 | sizeof(mac_data->entries) + (uc_count * ETH_ALEN)); |
f565a7c2 AW |
1888 | |
1889 | /* multicast list and count fill the end */ | |
32e7bfc4 | 1890 | mac_data = (void *)&mac_data->macs[uc_count][0]; |
f565a7c2 | 1891 | |
fdd819b2 | 1892 | mac_data->entries = cpu_to_virtio32(vi->vdev, mc_count); |
567ec874 | 1893 | i = 0; |
22bedad3 JP |
1894 | netdev_for_each_mc_addr(ha, dev) |
1895 | memcpy(&mac_data->macs[i++][0], ha->addr, ETH_ALEN); | |
f565a7c2 AW |
1896 | |
1897 | sg_set_buf(&sg[1], mac_data, | |
4cd24eaf | 1898 | sizeof(mac_data->entries) + (mc_count * ETH_ALEN)); |
f565a7c2 AW |
1899 | |
1900 | if (!virtnet_send_command(vi, VIRTIO_NET_CTRL_MAC, | |
d24bae32 | 1901 | VIRTIO_NET_CTRL_MAC_TABLE_SET, sg)) |
99e872ae | 1902 | dev_warn(&dev->dev, "Failed to set MAC filter table.\n"); |
f565a7c2 AW |
1903 | |
1904 | kfree(buf); | |
2af7698e AW |
1905 | } |
1906 | ||
80d5c368 PM |
1907 | static int virtnet_vlan_rx_add_vid(struct net_device *dev, |
1908 | __be16 proto, u16 vid) | |
0bde9569 AW |
1909 | { |
1910 | struct virtnet_info *vi = netdev_priv(dev); | |
1911 | struct scatterlist sg; | |
1912 | ||
d7fad4c8 | 1913 | vi->ctrl->vid = cpu_to_virtio16(vi->vdev, vid); |
12e57169 | 1914 | sg_init_one(&sg, &vi->ctrl->vid, sizeof(vi->ctrl->vid)); |
0bde9569 AW |
1915 | |
1916 | if (!virtnet_send_command(vi, VIRTIO_NET_CTRL_VLAN, | |
d24bae32 | 1917 | VIRTIO_NET_CTRL_VLAN_ADD, &sg)) |
0bde9569 | 1918 | dev_warn(&dev->dev, "Failed to add VLAN ID %d.\n", vid); |
8e586137 | 1919 | return 0; |
0bde9569 AW |
1920 | } |
1921 | ||
80d5c368 PM |
1922 | static int virtnet_vlan_rx_kill_vid(struct net_device *dev, |
1923 | __be16 proto, u16 vid) | |
0bde9569 AW |
1924 | { |
1925 | struct virtnet_info *vi = netdev_priv(dev); | |
1926 | struct scatterlist sg; | |
1927 | ||
d7fad4c8 | 1928 | vi->ctrl->vid = cpu_to_virtio16(vi->vdev, vid); |
12e57169 | 1929 | sg_init_one(&sg, &vi->ctrl->vid, sizeof(vi->ctrl->vid)); |
0bde9569 AW |
1930 | |
1931 | if (!virtnet_send_command(vi, VIRTIO_NET_CTRL_VLAN, | |
d24bae32 | 1932 | VIRTIO_NET_CTRL_VLAN_DEL, &sg)) |
0bde9569 | 1933 | dev_warn(&dev->dev, "Failed to kill VLAN ID %d.\n", vid); |
8e586137 | 1934 | return 0; |
0bde9569 AW |
1935 | } |
1936 | ||
310974fa | 1937 | static void virtnet_clean_affinity(struct virtnet_info *vi) |
986a4f4d JW |
1938 | { |
1939 | int i; | |
1940 | ||
8898c21c WG |
1941 | if (vi->affinity_hint_set) { |
1942 | for (i = 0; i < vi->max_queue_pairs; i++) { | |
19e226e8 CR |
1943 | virtqueue_set_affinity(vi->rq[i].vq, NULL); |
1944 | virtqueue_set_affinity(vi->sq[i].vq, NULL); | |
47be2479 WG |
1945 | } |
1946 | ||
8898c21c WG |
1947 | vi->affinity_hint_set = false; |
1948 | } | |
8898c21c | 1949 | } |
47be2479 | 1950 | |
8898c21c WG |
1951 | static void virtnet_set_affinity(struct virtnet_info *vi) |
1952 | { | |
2ca653d6 CR |
1953 | cpumask_var_t mask; |
1954 | int stragglers; | |
1955 | int group_size; | |
1956 | int i, j, cpu; | |
1957 | int num_cpu; | |
1958 | int stride; | |
1959 | ||
1960 | if (!zalloc_cpumask_var(&mask, GFP_KERNEL)) { | |
310974fa | 1961 | virtnet_clean_affinity(vi); |
8898c21c | 1962 | return; |
986a4f4d JW |
1963 | } |
1964 | ||
2ca653d6 CR |
1965 | num_cpu = num_online_cpus(); |
1966 | stride = max_t(int, num_cpu / vi->curr_queue_pairs, 1); | |
1967 | stragglers = num_cpu >= vi->curr_queue_pairs ? | |
1968 | num_cpu % vi->curr_queue_pairs : | |
1969 | 0; | |
1970 | cpu = cpumask_next(-1, cpu_online_mask); | |
4d99f660 | 1971 | |
2ca653d6 CR |
1972 | for (i = 0; i < vi->curr_queue_pairs; i++) { |
1973 | group_size = stride + (i < stragglers ? 1 : 0); | |
1974 | ||
1975 | for (j = 0; j < group_size; j++) { | |
1976 | cpumask_set_cpu(cpu, mask); | |
1977 | cpu = cpumask_next_wrap(cpu, cpu_online_mask, | |
1978 | nr_cpu_ids, false); | |
1979 | } | |
1980 | virtqueue_set_affinity(vi->rq[i].vq, mask); | |
1981 | virtqueue_set_affinity(vi->sq[i].vq, mask); | |
1982 | __netif_set_xps_queue(vi->dev, cpumask_bits(mask), i, false); | |
1983 | cpumask_clear(mask); | |
986a4f4d JW |
1984 | } |
1985 | ||
8898c21c | 1986 | vi->affinity_hint_set = true; |
2ca653d6 | 1987 | free_cpumask_var(mask); |
986a4f4d JW |
1988 | } |
1989 | ||
8017c279 | 1990 | static int virtnet_cpu_online(unsigned int cpu, struct hlist_node *node) |
8de4b2f3 | 1991 | { |
8017c279 SAS |
1992 | struct virtnet_info *vi = hlist_entry_safe(node, struct virtnet_info, |
1993 | node); | |
1994 | virtnet_set_affinity(vi); | |
1995 | return 0; | |
1996 | } | |
8de4b2f3 | 1997 | |
8017c279 SAS |
1998 | static int virtnet_cpu_dead(unsigned int cpu, struct hlist_node *node) |
1999 | { | |
2000 | struct virtnet_info *vi = hlist_entry_safe(node, struct virtnet_info, | |
2001 | node_dead); | |
2002 | virtnet_set_affinity(vi); | |
2003 | return 0; | |
2004 | } | |
3ab098df | 2005 | |
8017c279 SAS |
2006 | static int virtnet_cpu_down_prep(unsigned int cpu, struct hlist_node *node) |
2007 | { | |
2008 | struct virtnet_info *vi = hlist_entry_safe(node, struct virtnet_info, | |
2009 | node); | |
2010 | ||
310974fa | 2011 | virtnet_clean_affinity(vi); |
8017c279 SAS |
2012 | return 0; |
2013 | } | |
2014 | ||
2015 | static enum cpuhp_state virtionet_online; | |
2016 | ||
2017 | static int virtnet_cpu_notif_add(struct virtnet_info *vi) | |
2018 | { | |
2019 | int ret; | |
2020 | ||
2021 | ret = cpuhp_state_add_instance_nocalls(virtionet_online, &vi->node); | |
2022 | if (ret) | |
2023 | return ret; | |
2024 | ret = cpuhp_state_add_instance_nocalls(CPUHP_VIRT_NET_DEAD, | |
2025 | &vi->node_dead); | |
2026 | if (!ret) | |
2027 | return ret; | |
2028 | cpuhp_state_remove_instance_nocalls(virtionet_online, &vi->node); | |
2029 | return ret; | |
2030 | } | |
2031 | ||
2032 | static void virtnet_cpu_notif_remove(struct virtnet_info *vi) | |
2033 | { | |
2034 | cpuhp_state_remove_instance_nocalls(virtionet_online, &vi->node); | |
2035 | cpuhp_state_remove_instance_nocalls(CPUHP_VIRT_NET_DEAD, | |
2036 | &vi->node_dead); | |
986a4f4d JW |
2037 | } |
2038 | ||
8f9f4668 RJ |
2039 | static void virtnet_get_ringparam(struct net_device *dev, |
2040 | struct ethtool_ringparam *ring) | |
2041 | { | |
2042 | struct virtnet_info *vi = netdev_priv(dev); | |
2043 | ||
986a4f4d JW |
2044 | ring->rx_max_pending = virtqueue_get_vring_size(vi->rq[0].vq); |
2045 | ring->tx_max_pending = virtqueue_get_vring_size(vi->sq[0].vq); | |
8f9f4668 RJ |
2046 | ring->rx_pending = ring->rx_max_pending; |
2047 | ring->tx_pending = ring->tx_max_pending; | |
8f9f4668 RJ |
2048 | } |
2049 | ||
66846048 RJ |
2050 | |
2051 | static void virtnet_get_drvinfo(struct net_device *dev, | |
2052 | struct ethtool_drvinfo *info) | |
2053 | { | |
2054 | struct virtnet_info *vi = netdev_priv(dev); | |
2055 | struct virtio_device *vdev = vi->vdev; | |
2056 | ||
2057 | strlcpy(info->driver, KBUILD_MODNAME, sizeof(info->driver)); | |
2058 | strlcpy(info->version, VIRTNET_DRIVER_VERSION, sizeof(info->version)); | |
2059 | strlcpy(info->bus_info, virtio_bus_name(vdev), sizeof(info->bus_info)); | |
2060 | ||
2061 | } | |
2062 | ||
d73bcd2c JW |
2063 | /* TODO: Eliminate OOO packets during switching */ |
2064 | static int virtnet_set_channels(struct net_device *dev, | |
2065 | struct ethtool_channels *channels) | |
2066 | { | |
2067 | struct virtnet_info *vi = netdev_priv(dev); | |
2068 | u16 queue_pairs = channels->combined_count; | |
2069 | int err; | |
2070 | ||
2071 | /* We don't support separate rx/tx channels. | |
2072 | * We don't allow setting 'other' channels. | |
2073 | */ | |
2074 | if (channels->rx_count || channels->tx_count || channels->other_count) | |
2075 | return -EINVAL; | |
2076 | ||
c18e9cd6 | 2077 | if (queue_pairs > vi->max_queue_pairs || queue_pairs == 0) |
d73bcd2c JW |
2078 | return -EINVAL; |
2079 | ||
f600b690 JF |
2080 | /* For now we don't support modifying channels while XDP is loaded |
2081 | * also when XDP is loaded all RX queues have XDP programs so we only | |
2082 | * need to check a single RX queue. | |
2083 | */ | |
2084 | if (vi->rq[0].xdp_prog) | |
2085 | return -EINVAL; | |
2086 | ||
47be2479 | 2087 | get_online_cpus(); |
47315329 | 2088 | err = _virtnet_set_queues(vi, queue_pairs); |
d73bcd2c JW |
2089 | if (!err) { |
2090 | netif_set_real_num_tx_queues(dev, queue_pairs); | |
2091 | netif_set_real_num_rx_queues(dev, queue_pairs); | |
2092 | ||
8898c21c | 2093 | virtnet_set_affinity(vi); |
d73bcd2c | 2094 | } |
47be2479 | 2095 | put_online_cpus(); |
d73bcd2c JW |
2096 | |
2097 | return err; | |
2098 | } | |
2099 | ||
d7dfc5cf TM |
2100 | static void virtnet_get_strings(struct net_device *dev, u32 stringset, u8 *data) |
2101 | { | |
2102 | struct virtnet_info *vi = netdev_priv(dev); | |
2103 | char *p = (char *)data; | |
2104 | unsigned int i, j; | |
2105 | ||
2106 | switch (stringset) { | |
2107 | case ETH_SS_STATS: | |
2108 | for (i = 0; i < vi->curr_queue_pairs; i++) { | |
2109 | for (j = 0; j < VIRTNET_RQ_STATS_LEN; j++) { | |
2110 | snprintf(p, ETH_GSTRING_LEN, "rx_queue_%u_%s", | |
2111 | i, virtnet_rq_stats_desc[j].desc); | |
2112 | p += ETH_GSTRING_LEN; | |
2113 | } | |
2114 | } | |
2115 | ||
2116 | for (i = 0; i < vi->curr_queue_pairs; i++) { | |
2117 | for (j = 0; j < VIRTNET_SQ_STATS_LEN; j++) { | |
2118 | snprintf(p, ETH_GSTRING_LEN, "tx_queue_%u_%s", | |
2119 | i, virtnet_sq_stats_desc[j].desc); | |
2120 | p += ETH_GSTRING_LEN; | |
2121 | } | |
2122 | } | |
2123 | break; | |
2124 | } | |
2125 | } | |
2126 | ||
2127 | static int virtnet_get_sset_count(struct net_device *dev, int sset) | |
2128 | { | |
2129 | struct virtnet_info *vi = netdev_priv(dev); | |
2130 | ||
2131 | switch (sset) { | |
2132 | case ETH_SS_STATS: | |
2133 | return vi->curr_queue_pairs * (VIRTNET_RQ_STATS_LEN + | |
2134 | VIRTNET_SQ_STATS_LEN); | |
2135 | default: | |
2136 | return -EOPNOTSUPP; | |
2137 | } | |
2138 | } | |
2139 | ||
2140 | static void virtnet_get_ethtool_stats(struct net_device *dev, | |
2141 | struct ethtool_stats *stats, u64 *data) | |
2142 | { | |
2143 | struct virtnet_info *vi = netdev_priv(dev); | |
2144 | unsigned int idx = 0, start, i, j; | |
2145 | const u8 *stats_base; | |
2146 | size_t offset; | |
2147 | ||
2148 | for (i = 0; i < vi->curr_queue_pairs; i++) { | |
2149 | struct receive_queue *rq = &vi->rq[i]; | |
2150 | ||
d46eeeaf | 2151 | stats_base = (u8 *)&rq->stats; |
d7dfc5cf TM |
2152 | do { |
2153 | start = u64_stats_fetch_begin_irq(&rq->stats.syncp); | |
2154 | for (j = 0; j < VIRTNET_RQ_STATS_LEN; j++) { | |
2155 | offset = virtnet_rq_stats_desc[j].offset; | |
2156 | data[idx + j] = *(u64 *)(stats_base + offset); | |
2157 | } | |
2158 | } while (u64_stats_fetch_retry_irq(&rq->stats.syncp, start)); | |
2159 | idx += VIRTNET_RQ_STATS_LEN; | |
2160 | } | |
2161 | ||
2162 | for (i = 0; i < vi->curr_queue_pairs; i++) { | |
2163 | struct send_queue *sq = &vi->sq[i]; | |
2164 | ||
2165 | stats_base = (u8 *)&sq->stats; | |
2166 | do { | |
2167 | start = u64_stats_fetch_begin_irq(&sq->stats.syncp); | |
2168 | for (j = 0; j < VIRTNET_SQ_STATS_LEN; j++) { | |
2169 | offset = virtnet_sq_stats_desc[j].offset; | |
2170 | data[idx + j] = *(u64 *)(stats_base + offset); | |
2171 | } | |
2172 | } while (u64_stats_fetch_retry_irq(&sq->stats.syncp, start)); | |
2173 | idx += VIRTNET_SQ_STATS_LEN; | |
2174 | } | |
2175 | } | |
2176 | ||
d73bcd2c JW |
2177 | static void virtnet_get_channels(struct net_device *dev, |
2178 | struct ethtool_channels *channels) | |
2179 | { | |
2180 | struct virtnet_info *vi = netdev_priv(dev); | |
2181 | ||
2182 | channels->combined_count = vi->curr_queue_pairs; | |
2183 | channels->max_combined = vi->max_queue_pairs; | |
2184 | channels->max_other = 0; | |
2185 | channels->rx_count = 0; | |
2186 | channels->tx_count = 0; | |
2187 | channels->other_count = 0; | |
2188 | } | |
2189 | ||
ebb6b4b1 PR |
2190 | static int virtnet_set_link_ksettings(struct net_device *dev, |
2191 | const struct ethtool_link_ksettings *cmd) | |
16032be5 NA |
2192 | { |
2193 | struct virtnet_info *vi = netdev_priv(dev); | |
16032be5 | 2194 | |
9aedc6e2 CF |
2195 | return ethtool_virtdev_set_link_ksettings(dev, cmd, |
2196 | &vi->speed, &vi->duplex); | |
16032be5 NA |
2197 | } |
2198 | ||
ebb6b4b1 PR |
2199 | static int virtnet_get_link_ksettings(struct net_device *dev, |
2200 | struct ethtool_link_ksettings *cmd) | |
16032be5 NA |
2201 | { |
2202 | struct virtnet_info *vi = netdev_priv(dev); | |
2203 | ||
ebb6b4b1 PR |
2204 | cmd->base.speed = vi->speed; |
2205 | cmd->base.duplex = vi->duplex; | |
2206 | cmd->base.port = PORT_OTHER; | |
16032be5 NA |
2207 | |
2208 | return 0; | |
2209 | } | |
2210 | ||
0c465be1 JW |
2211 | static int virtnet_set_coalesce(struct net_device *dev, |
2212 | struct ethtool_coalesce *ec) | |
2213 | { | |
0c465be1 JW |
2214 | struct virtnet_info *vi = netdev_priv(dev); |
2215 | int i, napi_weight; | |
2216 | ||
a51e5206 JK |
2217 | if (ec->tx_max_coalesced_frames > 1 || |
2218 | ec->rx_max_coalesced_frames != 1) | |
0c465be1 JW |
2219 | return -EINVAL; |
2220 | ||
0c465be1 | 2221 | napi_weight = ec->tx_max_coalesced_frames ? NAPI_POLL_WEIGHT : 0; |
0c465be1 JW |
2222 | if (napi_weight ^ vi->sq[0].napi.weight) { |
2223 | if (dev->flags & IFF_UP) | |
2224 | return -EBUSY; | |
2225 | for (i = 0; i < vi->max_queue_pairs; i++) | |
2226 | vi->sq[i].napi.weight = napi_weight; | |
2227 | } | |
2228 | ||
2229 | return 0; | |
2230 | } | |
2231 | ||
2232 | static int virtnet_get_coalesce(struct net_device *dev, | |
2233 | struct ethtool_coalesce *ec) | |
2234 | { | |
2235 | struct ethtool_coalesce ec_default = { | |
2236 | .cmd = ETHTOOL_GCOALESCE, | |
2237 | .rx_max_coalesced_frames = 1, | |
2238 | }; | |
2239 | struct virtnet_info *vi = netdev_priv(dev); | |
2240 | ||
2241 | memcpy(ec, &ec_default, sizeof(ec_default)); | |
2242 | ||
2243 | if (vi->sq[0].napi.weight) | |
2244 | ec->tx_max_coalesced_frames = 1; | |
2245 | ||
2246 | return 0; | |
2247 | } | |
2248 | ||
16032be5 NA |
2249 | static void virtnet_init_settings(struct net_device *dev) |
2250 | { | |
2251 | struct virtnet_info *vi = netdev_priv(dev); | |
2252 | ||
2253 | vi->speed = SPEED_UNKNOWN; | |
2254 | vi->duplex = DUPLEX_UNKNOWN; | |
2255 | } | |
2256 | ||
faa9b39f JB |
2257 | static void virtnet_update_settings(struct virtnet_info *vi) |
2258 | { | |
2259 | u32 speed; | |
2260 | u8 duplex; | |
2261 | ||
2262 | if (!virtio_has_feature(vi->vdev, VIRTIO_NET_F_SPEED_DUPLEX)) | |
2263 | return; | |
2264 | ||
2265 | speed = virtio_cread32(vi->vdev, offsetof(struct virtio_net_config, | |
2266 | speed)); | |
2267 | if (ethtool_validate_speed(speed)) | |
2268 | vi->speed = speed; | |
2269 | duplex = virtio_cread8(vi->vdev, offsetof(struct virtio_net_config, | |
2270 | duplex)); | |
2271 | if (ethtool_validate_duplex(duplex)) | |
2272 | vi->duplex = duplex; | |
2273 | } | |
2274 | ||
0fc0b732 | 2275 | static const struct ethtool_ops virtnet_ethtool_ops = { |
a51e5206 | 2276 | .supported_coalesce_params = ETHTOOL_COALESCE_MAX_FRAMES, |
66846048 | 2277 | .get_drvinfo = virtnet_get_drvinfo, |
9f4d26d0 | 2278 | .get_link = ethtool_op_get_link, |
8f9f4668 | 2279 | .get_ringparam = virtnet_get_ringparam, |
d7dfc5cf TM |
2280 | .get_strings = virtnet_get_strings, |
2281 | .get_sset_count = virtnet_get_sset_count, | |
2282 | .get_ethtool_stats = virtnet_get_ethtool_stats, | |
d73bcd2c JW |
2283 | .set_channels = virtnet_set_channels, |
2284 | .get_channels = virtnet_get_channels, | |
074c3582 | 2285 | .get_ts_info = ethtool_op_get_ts_info, |
ebb6b4b1 PR |
2286 | .get_link_ksettings = virtnet_get_link_ksettings, |
2287 | .set_link_ksettings = virtnet_set_link_ksettings, | |
0c465be1 JW |
2288 | .set_coalesce = virtnet_set_coalesce, |
2289 | .get_coalesce = virtnet_get_coalesce, | |
a9ea3fc6 HX |
2290 | }; |
2291 | ||
9fe7bfce JF |
2292 | static void virtnet_freeze_down(struct virtio_device *vdev) |
2293 | { | |
2294 | struct virtnet_info *vi = vdev->priv; | |
2295 | int i; | |
2296 | ||
2297 | /* Make sure no work handler is accessing the device */ | |
2298 | flush_work(&vi->config_work); | |
2299 | ||
05c998b7 | 2300 | netif_tx_lock_bh(vi->dev); |
9fe7bfce | 2301 | netif_device_detach(vi->dev); |
05c998b7 | 2302 | netif_tx_unlock_bh(vi->dev); |
9fe7bfce JF |
2303 | cancel_delayed_work_sync(&vi->refill); |
2304 | ||
2305 | if (netif_running(vi->dev)) { | |
b92f1e67 | 2306 | for (i = 0; i < vi->max_queue_pairs; i++) { |
9fe7bfce | 2307 | napi_disable(&vi->rq[i].napi); |
78a57b48 | 2308 | virtnet_napi_tx_disable(&vi->sq[i].napi); |
b92f1e67 | 2309 | } |
9fe7bfce JF |
2310 | } |
2311 | } | |
2312 | ||
2313 | static int init_vqs(struct virtnet_info *vi); | |
2314 | ||
2315 | static int virtnet_restore_up(struct virtio_device *vdev) | |
2316 | { | |
2317 | struct virtnet_info *vi = vdev->priv; | |
2318 | int err, i; | |
2319 | ||
2320 | err = init_vqs(vi); | |
2321 | if (err) | |
2322 | return err; | |
2323 | ||
2324 | virtio_device_ready(vdev); | |
2325 | ||
2326 | if (netif_running(vi->dev)) { | |
2327 | for (i = 0; i < vi->curr_queue_pairs; i++) | |
2328 | if (!try_fill_recv(vi, &vi->rq[i], GFP_KERNEL)) | |
2329 | schedule_delayed_work(&vi->refill, 0); | |
2330 | ||
b92f1e67 | 2331 | for (i = 0; i < vi->max_queue_pairs; i++) { |
e4e8452a | 2332 | virtnet_napi_enable(vi->rq[i].vq, &vi->rq[i].napi); |
b92f1e67 WB |
2333 | virtnet_napi_tx_enable(vi, vi->sq[i].vq, |
2334 | &vi->sq[i].napi); | |
2335 | } | |
9fe7bfce JF |
2336 | } |
2337 | ||
05c998b7 | 2338 | netif_tx_lock_bh(vi->dev); |
9fe7bfce | 2339 | netif_device_attach(vi->dev); |
05c998b7 | 2340 | netif_tx_unlock_bh(vi->dev); |
9fe7bfce JF |
2341 | return err; |
2342 | } | |
2343 | ||
3f93522f JW |
2344 | static int virtnet_set_guest_offloads(struct virtnet_info *vi, u64 offloads) |
2345 | { | |
2346 | struct scatterlist sg; | |
12e57169 | 2347 | vi->ctrl->offloads = cpu_to_virtio64(vi->vdev, offloads); |
3f93522f | 2348 | |
12e57169 | 2349 | sg_init_one(&sg, &vi->ctrl->offloads, sizeof(vi->ctrl->offloads)); |
3f93522f JW |
2350 | |
2351 | if (!virtnet_send_command(vi, VIRTIO_NET_CTRL_GUEST_OFFLOADS, | |
2352 | VIRTIO_NET_CTRL_GUEST_OFFLOADS_SET, &sg)) { | |
7934b481 | 2353 | dev_warn(&vi->dev->dev, "Fail to set guest offload.\n"); |
3f93522f JW |
2354 | return -EINVAL; |
2355 | } | |
2356 | ||
2357 | return 0; | |
2358 | } | |
2359 | ||
2360 | static int virtnet_clear_guest_offloads(struct virtnet_info *vi) | |
2361 | { | |
2362 | u64 offloads = 0; | |
2363 | ||
2364 | if (!vi->guest_offloads) | |
2365 | return 0; | |
2366 | ||
3f93522f JW |
2367 | return virtnet_set_guest_offloads(vi, offloads); |
2368 | } | |
2369 | ||
2370 | static int virtnet_restore_guest_offloads(struct virtnet_info *vi) | |
2371 | { | |
2372 | u64 offloads = vi->guest_offloads; | |
2373 | ||
2374 | if (!vi->guest_offloads) | |
2375 | return 0; | |
3f93522f JW |
2376 | |
2377 | return virtnet_set_guest_offloads(vi, offloads); | |
2378 | } | |
2379 | ||
9861ce03 JK |
2380 | static int virtnet_xdp_set(struct net_device *dev, struct bpf_prog *prog, |
2381 | struct netlink_ext_ack *extack) | |
f600b690 JF |
2382 | { |
2383 | unsigned long int max_sz = PAGE_SIZE - sizeof(struct padded_vnet_hdr); | |
2384 | struct virtnet_info *vi = netdev_priv(dev); | |
2385 | struct bpf_prog *old_prog; | |
017b29c3 | 2386 | u16 xdp_qp = 0, curr_qp; |
672aafd5 | 2387 | int i, err; |
f600b690 | 2388 | |
3f93522f JW |
2389 | if (!virtio_has_feature(vi->vdev, VIRTIO_NET_F_CTRL_GUEST_OFFLOADS) |
2390 | && (virtio_has_feature(vi->vdev, VIRTIO_NET_F_GUEST_TSO4) || | |
2391 | virtio_has_feature(vi->vdev, VIRTIO_NET_F_GUEST_TSO6) || | |
2392 | virtio_has_feature(vi->vdev, VIRTIO_NET_F_GUEST_ECN) || | |
18ba58e1 JW |
2393 | virtio_has_feature(vi->vdev, VIRTIO_NET_F_GUEST_UFO) || |
2394 | virtio_has_feature(vi->vdev, VIRTIO_NET_F_GUEST_CSUM))) { | |
2395 | NL_SET_ERR_MSG_MOD(extack, "Can't set XDP while host is implementing LRO/CSUM, disable LRO/CSUM first"); | |
f600b690 JF |
2396 | return -EOPNOTSUPP; |
2397 | } | |
2398 | ||
2399 | if (vi->mergeable_rx_bufs && !vi->any_header_sg) { | |
4d463c4d | 2400 | NL_SET_ERR_MSG_MOD(extack, "XDP expects header/data in single page, any_header_sg required"); |
f600b690 JF |
2401 | return -EINVAL; |
2402 | } | |
2403 | ||
2404 | if (dev->mtu > max_sz) { | |
4d463c4d | 2405 | NL_SET_ERR_MSG_MOD(extack, "MTU too large to enable XDP"); |
f600b690 JF |
2406 | netdev_warn(dev, "XDP requires MTU less than %lu\n", max_sz); |
2407 | return -EINVAL; | |
2408 | } | |
2409 | ||
672aafd5 JF |
2410 | curr_qp = vi->curr_queue_pairs - vi->xdp_queue_pairs; |
2411 | if (prog) | |
2412 | xdp_qp = nr_cpu_ids; | |
2413 | ||
2414 | /* XDP requires extra queues for XDP_TX */ | |
2415 | if (curr_qp + xdp_qp > vi->max_queue_pairs) { | |
4d463c4d | 2416 | NL_SET_ERR_MSG_MOD(extack, "Too few free TX rings available"); |
672aafd5 JF |
2417 | netdev_warn(dev, "request %i queues but max is %i\n", |
2418 | curr_qp + xdp_qp, vi->max_queue_pairs); | |
2419 | return -ENOMEM; | |
2420 | } | |
2421 | ||
03aa6d34 TM |
2422 | old_prog = rtnl_dereference(vi->rq[0].xdp_prog); |
2423 | if (!prog && !old_prog) | |
2424 | return 0; | |
2425 | ||
85192dbf AN |
2426 | if (prog) |
2427 | bpf_prog_add(prog, vi->max_queue_pairs - 1); | |
2de2f7f4 | 2428 | |
4941d472 | 2429 | /* Make sure NAPI is not using any XDP TX queues for RX. */ |
534da5e8 TM |
2430 | if (netif_running(dev)) { |
2431 | for (i = 0; i < vi->max_queue_pairs; i++) { | |
4e09ff53 | 2432 | napi_disable(&vi->rq[i].napi); |
534da5e8 TM |
2433 | virtnet_napi_tx_disable(&vi->sq[i].napi); |
2434 | } | |
2435 | } | |
f600b690 | 2436 | |
03aa6d34 TM |
2437 | if (!prog) { |
2438 | for (i = 0; i < vi->max_queue_pairs; i++) { | |
2439 | rcu_assign_pointer(vi->rq[i].xdp_prog, prog); | |
2440 | if (i == 0) | |
2441 | virtnet_restore_guest_offloads(vi); | |
2442 | } | |
2443 | synchronize_net(); | |
2444 | } | |
f600b690 | 2445 | |
4941d472 JW |
2446 | err = _virtnet_set_queues(vi, curr_qp + xdp_qp); |
2447 | if (err) | |
2448 | goto err; | |
188313c1 | 2449 | netif_set_real_num_rx_queues(dev, curr_qp + xdp_qp); |
4941d472 | 2450 | vi->xdp_queue_pairs = xdp_qp; |
672aafd5 | 2451 | |
03aa6d34 TM |
2452 | if (prog) { |
2453 | for (i = 0; i < vi->max_queue_pairs; i++) { | |
2454 | rcu_assign_pointer(vi->rq[i].xdp_prog, prog); | |
2455 | if (i == 0 && !old_prog) | |
3f93522f | 2456 | virtnet_clear_guest_offloads(vi); |
3f93522f | 2457 | } |
03aa6d34 TM |
2458 | } |
2459 | ||
2460 | for (i = 0; i < vi->max_queue_pairs; i++) { | |
f600b690 JF |
2461 | if (old_prog) |
2462 | bpf_prog_put(old_prog); | |
534da5e8 | 2463 | if (netif_running(dev)) { |
4e09ff53 | 2464 | virtnet_napi_enable(vi->rq[i].vq, &vi->rq[i].napi); |
534da5e8 TM |
2465 | virtnet_napi_tx_enable(vi, vi->sq[i].vq, |
2466 | &vi->sq[i].napi); | |
2467 | } | |
f600b690 JF |
2468 | } |
2469 | ||
2470 | return 0; | |
2de2f7f4 | 2471 | |
4941d472 | 2472 | err: |
03aa6d34 TM |
2473 | if (!prog) { |
2474 | virtnet_clear_guest_offloads(vi); | |
2475 | for (i = 0; i < vi->max_queue_pairs; i++) | |
2476 | rcu_assign_pointer(vi->rq[i].xdp_prog, old_prog); | |
2477 | } | |
2478 | ||
8be4d9a4 | 2479 | if (netif_running(dev)) { |
534da5e8 | 2480 | for (i = 0; i < vi->max_queue_pairs; i++) { |
8be4d9a4 | 2481 | virtnet_napi_enable(vi->rq[i].vq, &vi->rq[i].napi); |
534da5e8 TM |
2482 | virtnet_napi_tx_enable(vi, vi->sq[i].vq, |
2483 | &vi->sq[i].napi); | |
2484 | } | |
8be4d9a4 | 2485 | } |
2de2f7f4 JF |
2486 | if (prog) |
2487 | bpf_prog_sub(prog, vi->max_queue_pairs - 1); | |
2488 | return err; | |
f600b690 JF |
2489 | } |
2490 | ||
5b0e6629 | 2491 | static u32 virtnet_xdp_query(struct net_device *dev) |
f600b690 JF |
2492 | { |
2493 | struct virtnet_info *vi = netdev_priv(dev); | |
5b0e6629 | 2494 | const struct bpf_prog *xdp_prog; |
f600b690 JF |
2495 | int i; |
2496 | ||
2497 | for (i = 0; i < vi->max_queue_pairs; i++) { | |
5b0e6629 MKL |
2498 | xdp_prog = rtnl_dereference(vi->rq[i].xdp_prog); |
2499 | if (xdp_prog) | |
2500 | return xdp_prog->aux->id; | |
f600b690 | 2501 | } |
5b0e6629 | 2502 | return 0; |
f600b690 JF |
2503 | } |
2504 | ||
f4e63525 | 2505 | static int virtnet_xdp(struct net_device *dev, struct netdev_bpf *xdp) |
f600b690 JF |
2506 | { |
2507 | switch (xdp->command) { | |
2508 | case XDP_SETUP_PROG: | |
9861ce03 | 2509 | return virtnet_xdp_set(dev, xdp->prog, xdp->extack); |
f600b690 | 2510 | case XDP_QUERY_PROG: |
5b0e6629 | 2511 | xdp->prog_id = virtnet_xdp_query(dev); |
f600b690 JF |
2512 | return 0; |
2513 | default: | |
2514 | return -EINVAL; | |
2515 | } | |
2516 | } | |
2517 | ||
ba5e4426 SS |
2518 | static int virtnet_get_phys_port_name(struct net_device *dev, char *buf, |
2519 | size_t len) | |
2520 | { | |
2521 | struct virtnet_info *vi = netdev_priv(dev); | |
2522 | int ret; | |
2523 | ||
2524 | if (!virtio_has_feature(vi->vdev, VIRTIO_NET_F_STANDBY)) | |
2525 | return -EOPNOTSUPP; | |
2526 | ||
2527 | ret = snprintf(buf, len, "sby"); | |
2528 | if (ret >= len) | |
2529 | return -EOPNOTSUPP; | |
2530 | ||
2531 | return 0; | |
2532 | } | |
2533 | ||
a02e8964 WB |
2534 | static int virtnet_set_features(struct net_device *dev, |
2535 | netdev_features_t features) | |
2536 | { | |
2537 | struct virtnet_info *vi = netdev_priv(dev); | |
2538 | u64 offloads; | |
2539 | int err; | |
2540 | ||
2541 | if ((dev->features ^ features) & NETIF_F_LRO) { | |
2542 | if (vi->xdp_queue_pairs) | |
2543 | return -EBUSY; | |
2544 | ||
2545 | if (features & NETIF_F_LRO) | |
2546 | offloads = vi->guest_offloads_capable; | |
2547 | else | |
2548 | offloads = 0; | |
2549 | ||
2550 | err = virtnet_set_guest_offloads(vi, offloads); | |
2551 | if (err) | |
2552 | return err; | |
2553 | vi->guest_offloads = offloads; | |
2554 | } | |
2555 | ||
2556 | return 0; | |
2557 | } | |
2558 | ||
76288b4e SH |
2559 | static const struct net_device_ops virtnet_netdev = { |
2560 | .ndo_open = virtnet_open, | |
2561 | .ndo_stop = virtnet_close, | |
2562 | .ndo_start_xmit = start_xmit, | |
2563 | .ndo_validate_addr = eth_validate_addr, | |
9c46f6d4 | 2564 | .ndo_set_mac_address = virtnet_set_mac_address, |
2af7698e | 2565 | .ndo_set_rx_mode = virtnet_set_rx_mode, |
3fa2a1df | 2566 | .ndo_get_stats64 = virtnet_stats, |
1824a989 AW |
2567 | .ndo_vlan_rx_add_vid = virtnet_vlan_rx_add_vid, |
2568 | .ndo_vlan_rx_kill_vid = virtnet_vlan_rx_kill_vid, | |
f4e63525 | 2569 | .ndo_bpf = virtnet_xdp, |
186b3c99 | 2570 | .ndo_xdp_xmit = virtnet_xdp_xmit, |
2836b4f2 | 2571 | .ndo_features_check = passthru_features_check, |
ba5e4426 | 2572 | .ndo_get_phys_port_name = virtnet_get_phys_port_name, |
a02e8964 | 2573 | .ndo_set_features = virtnet_set_features, |
76288b4e SH |
2574 | }; |
2575 | ||
586d17c5 | 2576 | static void virtnet_config_changed_work(struct work_struct *work) |
9f4d26d0 | 2577 | { |
586d17c5 JW |
2578 | struct virtnet_info *vi = |
2579 | container_of(work, struct virtnet_info, config_work); | |
9f4d26d0 MM |
2580 | u16 v; |
2581 | ||
855e0c52 RR |
2582 | if (virtio_cread_feature(vi->vdev, VIRTIO_NET_F_STATUS, |
2583 | struct virtio_net_config, status, &v) < 0) | |
507613bf | 2584 | return; |
586d17c5 JW |
2585 | |
2586 | if (v & VIRTIO_NET_S_ANNOUNCE) { | |
ee89bab1 | 2587 | netdev_notify_peers(vi->dev); |
586d17c5 JW |
2588 | virtnet_ack_link_announce(vi); |
2589 | } | |
9f4d26d0 MM |
2590 | |
2591 | /* Ignore unknown (future) status bits */ | |
2592 | v &= VIRTIO_NET_S_LINK_UP; | |
2593 | ||
2594 | if (vi->status == v) | |
507613bf | 2595 | return; |
9f4d26d0 MM |
2596 | |
2597 | vi->status = v; | |
2598 | ||
2599 | if (vi->status & VIRTIO_NET_S_LINK_UP) { | |
faa9b39f | 2600 | virtnet_update_settings(vi); |
9f4d26d0 | 2601 | netif_carrier_on(vi->dev); |
986a4f4d | 2602 | netif_tx_wake_all_queues(vi->dev); |
9f4d26d0 MM |
2603 | } else { |
2604 | netif_carrier_off(vi->dev); | |
986a4f4d | 2605 | netif_tx_stop_all_queues(vi->dev); |
9f4d26d0 MM |
2606 | } |
2607 | } | |
2608 | ||
2609 | static void virtnet_config_changed(struct virtio_device *vdev) | |
2610 | { | |
2611 | struct virtnet_info *vi = vdev->priv; | |
2612 | ||
3b07e9ca | 2613 | schedule_work(&vi->config_work); |
9f4d26d0 MM |
2614 | } |
2615 | ||
986a4f4d JW |
2616 | static void virtnet_free_queues(struct virtnet_info *vi) |
2617 | { | |
d4fb84ee AV |
2618 | int i; |
2619 | ||
ab3971b1 JW |
2620 | for (i = 0; i < vi->max_queue_pairs; i++) { |
2621 | napi_hash_del(&vi->rq[i].napi); | |
d4fb84ee | 2622 | netif_napi_del(&vi->rq[i].napi); |
b92f1e67 | 2623 | netif_napi_del(&vi->sq[i].napi); |
ab3971b1 | 2624 | } |
d4fb84ee | 2625 | |
963abe5c ED |
2626 | /* We called napi_hash_del() before netif_napi_del(), |
2627 | * we need to respect an RCU grace period before freeing vi->rq | |
2628 | */ | |
2629 | synchronize_net(); | |
2630 | ||
986a4f4d JW |
2631 | kfree(vi->rq); |
2632 | kfree(vi->sq); | |
12e57169 | 2633 | kfree(vi->ctrl); |
986a4f4d JW |
2634 | } |
2635 | ||
47315329 | 2636 | static void _free_receive_bufs(struct virtnet_info *vi) |
986a4f4d | 2637 | { |
f600b690 | 2638 | struct bpf_prog *old_prog; |
986a4f4d JW |
2639 | int i; |
2640 | ||
2641 | for (i = 0; i < vi->max_queue_pairs; i++) { | |
2642 | while (vi->rq[i].pages) | |
2643 | __free_pages(get_a_page(&vi->rq[i], GFP_KERNEL), 0); | |
f600b690 JF |
2644 | |
2645 | old_prog = rtnl_dereference(vi->rq[i].xdp_prog); | |
2646 | RCU_INIT_POINTER(vi->rq[i].xdp_prog, NULL); | |
2647 | if (old_prog) | |
2648 | bpf_prog_put(old_prog); | |
986a4f4d | 2649 | } |
47315329 JF |
2650 | } |
2651 | ||
2652 | static void free_receive_bufs(struct virtnet_info *vi) | |
2653 | { | |
2654 | rtnl_lock(); | |
2655 | _free_receive_bufs(vi); | |
f600b690 | 2656 | rtnl_unlock(); |
986a4f4d JW |
2657 | } |
2658 | ||
fb51879d MD |
2659 | static void free_receive_page_frags(struct virtnet_info *vi) |
2660 | { | |
2661 | int i; | |
2662 | for (i = 0; i < vi->max_queue_pairs; i++) | |
2663 | if (vi->rq[i].alloc_frag.page) | |
2664 | put_page(vi->rq[i].alloc_frag.page); | |
2665 | } | |
2666 | ||
986a4f4d JW |
2667 | static void free_unused_bufs(struct virtnet_info *vi) |
2668 | { | |
2669 | void *buf; | |
2670 | int i; | |
2671 | ||
2672 | for (i = 0; i < vi->max_queue_pairs; i++) { | |
2673 | struct virtqueue *vq = vi->sq[i].vq; | |
56434a01 | 2674 | while ((buf = virtqueue_detach_unused_buf(vq)) != NULL) { |
5050471d | 2675 | if (!is_xdp_frame(buf)) |
56434a01 JF |
2676 | dev_kfree_skb(buf); |
2677 | else | |
5050471d | 2678 | xdp_return_frame(ptr_to_xdp(buf)); |
56434a01 | 2679 | } |
986a4f4d JW |
2680 | } |
2681 | ||
2682 | for (i = 0; i < vi->max_queue_pairs; i++) { | |
2683 | struct virtqueue *vq = vi->rq[i].vq; | |
2684 | ||
2685 | while ((buf = virtqueue_detach_unused_buf(vq)) != NULL) { | |
ab7db917 | 2686 | if (vi->mergeable_rx_bufs) { |
680557cf | 2687 | put_page(virt_to_head_page(buf)); |
ab7db917 | 2688 | } else if (vi->big_packets) { |
fa9fac17 | 2689 | give_pages(&vi->rq[i], buf); |
ab7db917 | 2690 | } else { |
f6b10209 | 2691 | put_page(virt_to_head_page(buf)); |
ab7db917 | 2692 | } |
986a4f4d | 2693 | } |
986a4f4d JW |
2694 | } |
2695 | } | |
2696 | ||
e9d7417b JW |
2697 | static void virtnet_del_vqs(struct virtnet_info *vi) |
2698 | { | |
2699 | struct virtio_device *vdev = vi->vdev; | |
2700 | ||
310974fa | 2701 | virtnet_clean_affinity(vi); |
986a4f4d | 2702 | |
e9d7417b | 2703 | vdev->config->del_vqs(vdev); |
986a4f4d JW |
2704 | |
2705 | virtnet_free_queues(vi); | |
e9d7417b JW |
2706 | } |
2707 | ||
d85b758f MT |
2708 | /* How large should a single buffer be so a queue full of these can fit at |
2709 | * least one full packet? | |
2710 | * Logic below assumes the mergeable buffer header is used. | |
2711 | */ | |
2712 | static unsigned int mergeable_min_buf_len(struct virtnet_info *vi, struct virtqueue *vq) | |
2713 | { | |
2714 | const unsigned int hdr_len = sizeof(struct virtio_net_hdr_mrg_rxbuf); | |
2715 | unsigned int rq_size = virtqueue_get_vring_size(vq); | |
2716 | unsigned int packet_len = vi->big_packets ? IP_MAX_MTU : vi->dev->max_mtu; | |
2717 | unsigned int buf_len = hdr_len + ETH_HLEN + VLAN_HLEN + packet_len; | |
2718 | unsigned int min_buf_len = DIV_ROUND_UP(buf_len, rq_size); | |
2719 | ||
f0c3192c MT |
2720 | return max(max(min_buf_len, hdr_len) - hdr_len, |
2721 | (unsigned int)GOOD_PACKET_LEN); | |
d85b758f MT |
2722 | } |
2723 | ||
986a4f4d | 2724 | static int virtnet_find_vqs(struct virtnet_info *vi) |
3f9c10b0 | 2725 | { |
986a4f4d JW |
2726 | vq_callback_t **callbacks; |
2727 | struct virtqueue **vqs; | |
2728 | int ret = -ENOMEM; | |
2729 | int i, total_vqs; | |
2730 | const char **names; | |
d45b897b | 2731 | bool *ctx; |
986a4f4d JW |
2732 | |
2733 | /* We expect 1 RX virtqueue followed by 1 TX virtqueue, followed by | |
2734 | * possible N-1 RX/TX queue pairs used in multiqueue mode, followed by | |
2735 | * possible control vq. | |
2736 | */ | |
2737 | total_vqs = vi->max_queue_pairs * 2 + | |
2738 | virtio_has_feature(vi->vdev, VIRTIO_NET_F_CTRL_VQ); | |
2739 | ||
2740 | /* Allocate space for find_vqs parameters */ | |
6396bb22 | 2741 | vqs = kcalloc(total_vqs, sizeof(*vqs), GFP_KERNEL); |
986a4f4d JW |
2742 | if (!vqs) |
2743 | goto err_vq; | |
6da2ec56 | 2744 | callbacks = kmalloc_array(total_vqs, sizeof(*callbacks), GFP_KERNEL); |
986a4f4d JW |
2745 | if (!callbacks) |
2746 | goto err_callback; | |
6da2ec56 | 2747 | names = kmalloc_array(total_vqs, sizeof(*names), GFP_KERNEL); |
986a4f4d JW |
2748 | if (!names) |
2749 | goto err_names; | |
192f68cf | 2750 | if (!vi->big_packets || vi->mergeable_rx_bufs) { |
6396bb22 | 2751 | ctx = kcalloc(total_vqs, sizeof(*ctx), GFP_KERNEL); |
d45b897b MT |
2752 | if (!ctx) |
2753 | goto err_ctx; | |
2754 | } else { | |
2755 | ctx = NULL; | |
2756 | } | |
986a4f4d JW |
2757 | |
2758 | /* Parameters for control virtqueue, if any */ | |
2759 | if (vi->has_cvq) { | |
2760 | callbacks[total_vqs - 1] = NULL; | |
2761 | names[total_vqs - 1] = "control"; | |
2762 | } | |
3f9c10b0 | 2763 | |
986a4f4d JW |
2764 | /* Allocate/initialize parameters for send/receive virtqueues */ |
2765 | for (i = 0; i < vi->max_queue_pairs; i++) { | |
2766 | callbacks[rxq2vq(i)] = skb_recv_done; | |
2767 | callbacks[txq2vq(i)] = skb_xmit_done; | |
2768 | sprintf(vi->rq[i].name, "input.%d", i); | |
2769 | sprintf(vi->sq[i].name, "output.%d", i); | |
2770 | names[rxq2vq(i)] = vi->rq[i].name; | |
2771 | names[txq2vq(i)] = vi->sq[i].name; | |
d45b897b MT |
2772 | if (ctx) |
2773 | ctx[rxq2vq(i)] = true; | |
986a4f4d | 2774 | } |
3f9c10b0 | 2775 | |
986a4f4d | 2776 | ret = vi->vdev->config->find_vqs(vi->vdev, total_vqs, vqs, callbacks, |
d45b897b | 2777 | names, ctx, NULL); |
986a4f4d JW |
2778 | if (ret) |
2779 | goto err_find; | |
3f9c10b0 | 2780 | |
986a4f4d JW |
2781 | if (vi->has_cvq) { |
2782 | vi->cvq = vqs[total_vqs - 1]; | |
3f9c10b0 | 2783 | if (virtio_has_feature(vi->vdev, VIRTIO_NET_F_CTRL_VLAN)) |
f646968f | 2784 | vi->dev->features |= NETIF_F_HW_VLAN_CTAG_FILTER; |
3f9c10b0 | 2785 | } |
986a4f4d JW |
2786 | |
2787 | for (i = 0; i < vi->max_queue_pairs; i++) { | |
2788 | vi->rq[i].vq = vqs[rxq2vq(i)]; | |
d85b758f | 2789 | vi->rq[i].min_buf_len = mergeable_min_buf_len(vi, vi->rq[i].vq); |
986a4f4d JW |
2790 | vi->sq[i].vq = vqs[txq2vq(i)]; |
2791 | } | |
2792 | ||
2fa3c8a8 | 2793 | /* run here: ret == 0. */ |
986a4f4d | 2794 | |
986a4f4d JW |
2795 | |
2796 | err_find: | |
d45b897b MT |
2797 | kfree(ctx); |
2798 | err_ctx: | |
986a4f4d JW |
2799 | kfree(names); |
2800 | err_names: | |
2801 | kfree(callbacks); | |
2802 | err_callback: | |
2803 | kfree(vqs); | |
2804 | err_vq: | |
2805 | return ret; | |
2806 | } | |
2807 | ||
2808 | static int virtnet_alloc_queues(struct virtnet_info *vi) | |
2809 | { | |
2810 | int i; | |
2811 | ||
12e57169 MT |
2812 | vi->ctrl = kzalloc(sizeof(*vi->ctrl), GFP_KERNEL); |
2813 | if (!vi->ctrl) | |
2814 | goto err_ctrl; | |
6396bb22 | 2815 | vi->sq = kcalloc(vi->max_queue_pairs, sizeof(*vi->sq), GFP_KERNEL); |
986a4f4d JW |
2816 | if (!vi->sq) |
2817 | goto err_sq; | |
6396bb22 | 2818 | vi->rq = kcalloc(vi->max_queue_pairs, sizeof(*vi->rq), GFP_KERNEL); |
008d4278 | 2819 | if (!vi->rq) |
986a4f4d JW |
2820 | goto err_rq; |
2821 | ||
2822 | INIT_DELAYED_WORK(&vi->refill, refill_work); | |
2823 | for (i = 0; i < vi->max_queue_pairs; i++) { | |
2824 | vi->rq[i].pages = NULL; | |
2825 | netif_napi_add(vi->dev, &vi->rq[i].napi, virtnet_poll, | |
2826 | napi_weight); | |
1d11e732 WB |
2827 | netif_tx_napi_add(vi->dev, &vi->sq[i].napi, virtnet_poll_tx, |
2828 | napi_tx ? napi_weight : 0); | |
986a4f4d JW |
2829 | |
2830 | sg_init_table(vi->rq[i].sg, ARRAY_SIZE(vi->rq[i].sg)); | |
5377d758 | 2831 | ewma_pkt_len_init(&vi->rq[i].mrg_avg_pkt_len); |
986a4f4d | 2832 | sg_init_table(vi->sq[i].sg, ARRAY_SIZE(vi->sq[i].sg)); |
d7dfc5cf TM |
2833 | |
2834 | u64_stats_init(&vi->rq[i].stats.syncp); | |
2835 | u64_stats_init(&vi->sq[i].stats.syncp); | |
986a4f4d JW |
2836 | } |
2837 | ||
2838 | return 0; | |
2839 | ||
2840 | err_rq: | |
2841 | kfree(vi->sq); | |
2842 | err_sq: | |
12e57169 MT |
2843 | kfree(vi->ctrl); |
2844 | err_ctrl: | |
986a4f4d JW |
2845 | return -ENOMEM; |
2846 | } | |
2847 | ||
2848 | static int init_vqs(struct virtnet_info *vi) | |
2849 | { | |
2850 | int ret; | |
2851 | ||
2852 | /* Allocate send & receive queues */ | |
2853 | ret = virtnet_alloc_queues(vi); | |
2854 | if (ret) | |
2855 | goto err; | |
2856 | ||
2857 | ret = virtnet_find_vqs(vi); | |
2858 | if (ret) | |
2859 | goto err_free; | |
2860 | ||
47be2479 | 2861 | get_online_cpus(); |
8898c21c | 2862 | virtnet_set_affinity(vi); |
47be2479 WG |
2863 | put_online_cpus(); |
2864 | ||
986a4f4d JW |
2865 | return 0; |
2866 | ||
2867 | err_free: | |
2868 | virtnet_free_queues(vi); | |
2869 | err: | |
2870 | return ret; | |
3f9c10b0 AS |
2871 | } |
2872 | ||
fbf28d78 MD |
2873 | #ifdef CONFIG_SYSFS |
2874 | static ssize_t mergeable_rx_buffer_size_show(struct netdev_rx_queue *queue, | |
718ad681 | 2875 | char *buf) |
fbf28d78 MD |
2876 | { |
2877 | struct virtnet_info *vi = netdev_priv(queue->dev); | |
2878 | unsigned int queue_index = get_netdev_rx_queue_index(queue); | |
3cc81a9a JW |
2879 | unsigned int headroom = virtnet_get_headroom(vi); |
2880 | unsigned int tailroom = headroom ? sizeof(struct skb_shared_info) : 0; | |
5377d758 | 2881 | struct ewma_pkt_len *avg; |
fbf28d78 MD |
2882 | |
2883 | BUG_ON(queue_index >= vi->max_queue_pairs); | |
2884 | avg = &vi->rq[queue_index].mrg_avg_pkt_len; | |
d85b758f | 2885 | return sprintf(buf, "%u\n", |
3cc81a9a JW |
2886 | get_mergeable_buf_len(&vi->rq[queue_index], avg, |
2887 | SKB_DATA_ALIGN(headroom + tailroom))); | |
fbf28d78 MD |
2888 | } |
2889 | ||
2890 | static struct rx_queue_attribute mergeable_rx_buffer_size_attribute = | |
2891 | __ATTR_RO(mergeable_rx_buffer_size); | |
2892 | ||
2893 | static struct attribute *virtio_net_mrg_rx_attrs[] = { | |
2894 | &mergeable_rx_buffer_size_attribute.attr, | |
2895 | NULL | |
2896 | }; | |
2897 | ||
2898 | static const struct attribute_group virtio_net_mrg_rx_group = { | |
2899 | .name = "virtio_net", | |
2900 | .attrs = virtio_net_mrg_rx_attrs | |
2901 | }; | |
2902 | #endif | |
2903 | ||
892d6eb1 JW |
2904 | static bool virtnet_fail_on_feature(struct virtio_device *vdev, |
2905 | unsigned int fbit, | |
2906 | const char *fname, const char *dname) | |
2907 | { | |
2908 | if (!virtio_has_feature(vdev, fbit)) | |
2909 | return false; | |
2910 | ||
2911 | dev_err(&vdev->dev, "device advertises feature %s but not %s", | |
2912 | fname, dname); | |
2913 | ||
2914 | return true; | |
2915 | } | |
2916 | ||
2917 | #define VIRTNET_FAIL_ON(vdev, fbit, dbit) \ | |
2918 | virtnet_fail_on_feature(vdev, fbit, #fbit, dbit) | |
2919 | ||
2920 | static bool virtnet_validate_features(struct virtio_device *vdev) | |
2921 | { | |
2922 | if (!virtio_has_feature(vdev, VIRTIO_NET_F_CTRL_VQ) && | |
2923 | (VIRTNET_FAIL_ON(vdev, VIRTIO_NET_F_CTRL_RX, | |
2924 | "VIRTIO_NET_F_CTRL_VQ") || | |
2925 | VIRTNET_FAIL_ON(vdev, VIRTIO_NET_F_CTRL_VLAN, | |
2926 | "VIRTIO_NET_F_CTRL_VQ") || | |
2927 | VIRTNET_FAIL_ON(vdev, VIRTIO_NET_F_GUEST_ANNOUNCE, | |
2928 | "VIRTIO_NET_F_CTRL_VQ") || | |
2929 | VIRTNET_FAIL_ON(vdev, VIRTIO_NET_F_MQ, "VIRTIO_NET_F_CTRL_VQ") || | |
2930 | VIRTNET_FAIL_ON(vdev, VIRTIO_NET_F_CTRL_MAC_ADDR, | |
2931 | "VIRTIO_NET_F_CTRL_VQ"))) { | |
2932 | return false; | |
2933 | } | |
2934 | ||
2935 | return true; | |
2936 | } | |
2937 | ||
d0c2c997 JW |
2938 | #define MIN_MTU ETH_MIN_MTU |
2939 | #define MAX_MTU ETH_MAX_MTU | |
2940 | ||
fe36cbe0 | 2941 | static int virtnet_validate(struct virtio_device *vdev) |
296f96fc | 2942 | { |
6ba42248 MT |
2943 | if (!vdev->config->get) { |
2944 | dev_err(&vdev->dev, "%s failure: config access disabled\n", | |
2945 | __func__); | |
2946 | return -EINVAL; | |
2947 | } | |
2948 | ||
892d6eb1 JW |
2949 | if (!virtnet_validate_features(vdev)) |
2950 | return -EINVAL; | |
2951 | ||
fe36cbe0 MT |
2952 | if (virtio_has_feature(vdev, VIRTIO_NET_F_MTU)) { |
2953 | int mtu = virtio_cread16(vdev, | |
2954 | offsetof(struct virtio_net_config, | |
2955 | mtu)); | |
2956 | if (mtu < MIN_MTU) | |
2957 | __virtio_clear_bit(vdev, VIRTIO_NET_F_MTU); | |
2958 | } | |
2959 | ||
2960 | return 0; | |
2961 | } | |
2962 | ||
2963 | static int virtnet_probe(struct virtio_device *vdev) | |
2964 | { | |
d7dfc5cf | 2965 | int i, err = -ENOMEM; |
fe36cbe0 MT |
2966 | struct net_device *dev; |
2967 | struct virtnet_info *vi; | |
2968 | u16 max_queue_pairs; | |
2969 | int mtu; | |
2970 | ||
986a4f4d | 2971 | /* Find if host supports multiqueue virtio_net device */ |
855e0c52 RR |
2972 | err = virtio_cread_feature(vdev, VIRTIO_NET_F_MQ, |
2973 | struct virtio_net_config, | |
2974 | max_virtqueue_pairs, &max_queue_pairs); | |
986a4f4d JW |
2975 | |
2976 | /* We need at least 2 queue's */ | |
2977 | if (err || max_queue_pairs < VIRTIO_NET_CTRL_MQ_VQ_PAIRS_MIN || | |
2978 | max_queue_pairs > VIRTIO_NET_CTRL_MQ_VQ_PAIRS_MAX || | |
2979 | !virtio_has_feature(vdev, VIRTIO_NET_F_CTRL_VQ)) | |
2980 | max_queue_pairs = 1; | |
296f96fc RR |
2981 | |
2982 | /* Allocate ourselves a network device with room for our info */ | |
986a4f4d | 2983 | dev = alloc_etherdev_mq(sizeof(struct virtnet_info), max_queue_pairs); |
296f96fc RR |
2984 | if (!dev) |
2985 | return -ENOMEM; | |
2986 | ||
2987 | /* Set up network device as normal. */ | |
f2f2c8b4 | 2988 | dev->priv_flags |= IFF_UNICAST_FLT | IFF_LIVE_ADDR_CHANGE; |
76288b4e | 2989 | dev->netdev_ops = &virtnet_netdev; |
296f96fc | 2990 | dev->features = NETIF_F_HIGHDMA; |
3fa2a1df | 2991 | |
7ad24ea4 | 2992 | dev->ethtool_ops = &virtnet_ethtool_ops; |
296f96fc RR |
2993 | SET_NETDEV_DEV(dev, &vdev->dev); |
2994 | ||
2995 | /* Do we support "hardware" checksums? */ | |
98e778c9 | 2996 | if (virtio_has_feature(vdev, VIRTIO_NET_F_CSUM)) { |
296f96fc | 2997 | /* This opens up the world of extra features. */ |
48900cb6 | 2998 | dev->hw_features |= NETIF_F_HW_CSUM | NETIF_F_SG; |
98e778c9 | 2999 | if (csum) |
48900cb6 | 3000 | dev->features |= NETIF_F_HW_CSUM | NETIF_F_SG; |
98e778c9 MM |
3001 | |
3002 | if (virtio_has_feature(vdev, VIRTIO_NET_F_GSO)) { | |
e078de03 | 3003 | dev->hw_features |= NETIF_F_TSO |
34a48579 RR |
3004 | | NETIF_F_TSO_ECN | NETIF_F_TSO6; |
3005 | } | |
5539ae96 | 3006 | /* Individual feature bits: what can host handle? */ |
98e778c9 MM |
3007 | if (virtio_has_feature(vdev, VIRTIO_NET_F_HOST_TSO4)) |
3008 | dev->hw_features |= NETIF_F_TSO; | |
3009 | if (virtio_has_feature(vdev, VIRTIO_NET_F_HOST_TSO6)) | |
3010 | dev->hw_features |= NETIF_F_TSO6; | |
3011 | if (virtio_has_feature(vdev, VIRTIO_NET_F_HOST_ECN)) | |
3012 | dev->hw_features |= NETIF_F_TSO_ECN; | |
98e778c9 | 3013 | |
41f2f127 JW |
3014 | dev->features |= NETIF_F_GSO_ROBUST; |
3015 | ||
98e778c9 | 3016 | if (gso) |
e078de03 | 3017 | dev->features |= dev->hw_features & NETIF_F_ALL_TSO; |
98e778c9 | 3018 | /* (!csum && gso) case will be fixed by register_netdev() */ |
296f96fc | 3019 | } |
4f49129b TH |
3020 | if (virtio_has_feature(vdev, VIRTIO_NET_F_GUEST_CSUM)) |
3021 | dev->features |= NETIF_F_RXCSUM; | |
a02e8964 WB |
3022 | if (virtio_has_feature(vdev, VIRTIO_NET_F_GUEST_TSO4) || |
3023 | virtio_has_feature(vdev, VIRTIO_NET_F_GUEST_TSO6)) | |
3024 | dev->features |= NETIF_F_LRO; | |
3025 | if (virtio_has_feature(vdev, VIRTIO_NET_F_CTRL_GUEST_OFFLOADS)) | |
3026 | dev->hw_features |= NETIF_F_LRO; | |
296f96fc | 3027 | |
4fda8302 JW |
3028 | dev->vlan_features = dev->features; |
3029 | ||
d0c2c997 JW |
3030 | /* MTU range: 68 - 65535 */ |
3031 | dev->min_mtu = MIN_MTU; | |
3032 | dev->max_mtu = MAX_MTU; | |
3033 | ||
296f96fc | 3034 | /* Configuration may specify what MAC to use. Otherwise random. */ |
855e0c52 RR |
3035 | if (virtio_has_feature(vdev, VIRTIO_NET_F_MAC)) |
3036 | virtio_cread_bytes(vdev, | |
3037 | offsetof(struct virtio_net_config, mac), | |
3038 | dev->dev_addr, dev->addr_len); | |
3039 | else | |
f2cedb63 | 3040 | eth_hw_addr_random(dev); |
296f96fc RR |
3041 | |
3042 | /* Set up our device-specific information */ | |
3043 | vi = netdev_priv(dev); | |
296f96fc RR |
3044 | vi->dev = dev; |
3045 | vi->vdev = vdev; | |
d9d5dcc8 | 3046 | vdev->priv = vi; |
827da44c | 3047 | |
586d17c5 | 3048 | INIT_WORK(&vi->config_work, virtnet_config_changed_work); |
296f96fc | 3049 | |
97402b96 | 3050 | /* If we can receive ANY GSO packets, we must allocate large ones. */ |
8e95a202 JP |
3051 | if (virtio_has_feature(vdev, VIRTIO_NET_F_GUEST_TSO4) || |
3052 | virtio_has_feature(vdev, VIRTIO_NET_F_GUEST_TSO6) || | |
e3e3c423 VY |
3053 | virtio_has_feature(vdev, VIRTIO_NET_F_GUEST_ECN) || |
3054 | virtio_has_feature(vdev, VIRTIO_NET_F_GUEST_UFO)) | |
97402b96 HX |
3055 | vi->big_packets = true; |
3056 | ||
3f2c31d9 MM |
3057 | if (virtio_has_feature(vdev, VIRTIO_NET_F_MRG_RXBUF)) |
3058 | vi->mergeable_rx_bufs = true; | |
3059 | ||
d04302b3 MT |
3060 | if (virtio_has_feature(vdev, VIRTIO_NET_F_MRG_RXBUF) || |
3061 | virtio_has_feature(vdev, VIRTIO_F_VERSION_1)) | |
012873d0 MT |
3062 | vi->hdr_len = sizeof(struct virtio_net_hdr_mrg_rxbuf); |
3063 | else | |
3064 | vi->hdr_len = sizeof(struct virtio_net_hdr); | |
3065 | ||
75993300 MT |
3066 | if (virtio_has_feature(vdev, VIRTIO_F_ANY_LAYOUT) || |
3067 | virtio_has_feature(vdev, VIRTIO_F_VERSION_1)) | |
e7428e95 MT |
3068 | vi->any_header_sg = true; |
3069 | ||
986a4f4d JW |
3070 | if (virtio_has_feature(vdev, VIRTIO_NET_F_CTRL_VQ)) |
3071 | vi->has_cvq = true; | |
3072 | ||
14de9d11 AC |
3073 | if (virtio_has_feature(vdev, VIRTIO_NET_F_MTU)) { |
3074 | mtu = virtio_cread16(vdev, | |
3075 | offsetof(struct virtio_net_config, | |
3076 | mtu)); | |
93a205ee | 3077 | if (mtu < dev->min_mtu) { |
fe36cbe0 MT |
3078 | /* Should never trigger: MTU was previously validated |
3079 | * in virtnet_validate. | |
3080 | */ | |
7934b481 YS |
3081 | dev_err(&vdev->dev, |
3082 | "device MTU appears to have changed it is now %d < %d", | |
3083 | mtu, dev->min_mtu); | |
d7dfc5cf | 3084 | goto free; |
93a205ee | 3085 | } |
2e123b44 | 3086 | |
fe36cbe0 MT |
3087 | dev->mtu = mtu; |
3088 | dev->max_mtu = mtu; | |
3089 | ||
2e123b44 MT |
3090 | /* TODO: size buffers correctly in this case. */ |
3091 | if (dev->mtu > ETH_DATA_LEN) | |
3092 | vi->big_packets = true; | |
14de9d11 AC |
3093 | } |
3094 | ||
012873d0 MT |
3095 | if (vi->any_header_sg) |
3096 | dev->needed_headroom = vi->hdr_len; | |
6ebbc1a6 | 3097 | |
44900010 JW |
3098 | /* Enable multiqueue by default */ |
3099 | if (num_online_cpus() >= max_queue_pairs) | |
3100 | vi->curr_queue_pairs = max_queue_pairs; | |
3101 | else | |
3102 | vi->curr_queue_pairs = num_online_cpus(); | |
986a4f4d JW |
3103 | vi->max_queue_pairs = max_queue_pairs; |
3104 | ||
3105 | /* Allocate/initialize the rx/tx queues, and invoke find_vqs */ | |
3f9c10b0 | 3106 | err = init_vqs(vi); |
d2a7ddda | 3107 | if (err) |
d7dfc5cf | 3108 | goto free; |
296f96fc | 3109 | |
fbf28d78 MD |
3110 | #ifdef CONFIG_SYSFS |
3111 | if (vi->mergeable_rx_bufs) | |
3112 | dev->sysfs_rx_queue_group = &virtio_net_mrg_rx_group; | |
3113 | #endif | |
0f13b66b ZYW |
3114 | netif_set_real_num_tx_queues(dev, vi->curr_queue_pairs); |
3115 | netif_set_real_num_rx_queues(dev, vi->curr_queue_pairs); | |
986a4f4d | 3116 | |
16032be5 NA |
3117 | virtnet_init_settings(dev); |
3118 | ||
ba5e4426 SS |
3119 | if (virtio_has_feature(vdev, VIRTIO_NET_F_STANDBY)) { |
3120 | vi->failover = net_failover_create(vi->dev); | |
4b8e6ac4 WY |
3121 | if (IS_ERR(vi->failover)) { |
3122 | err = PTR_ERR(vi->failover); | |
ba5e4426 | 3123 | goto free_vqs; |
4b8e6ac4 | 3124 | } |
ba5e4426 SS |
3125 | } |
3126 | ||
296f96fc RR |
3127 | err = register_netdev(dev); |
3128 | if (err) { | |
3129 | pr_debug("virtio_net: registering device failed\n"); | |
ba5e4426 | 3130 | goto free_failover; |
296f96fc | 3131 | } |
b3369c1f | 3132 | |
4baf1e33 MT |
3133 | virtio_device_ready(vdev); |
3134 | ||
8017c279 | 3135 | err = virtnet_cpu_notif_add(vi); |
8de4b2f3 WG |
3136 | if (err) { |
3137 | pr_debug("virtio_net: registering cpu notifier failed\n"); | |
f00e35e2 | 3138 | goto free_unregister_netdev; |
8de4b2f3 WG |
3139 | } |
3140 | ||
a220871b | 3141 | virtnet_set_queues(vi, vi->curr_queue_pairs); |
44900010 | 3142 | |
167c25e4 JW |
3143 | /* Assume link up if device can't report link status, |
3144 | otherwise get link status from config. */ | |
bda7fab5 | 3145 | netif_carrier_off(dev); |
167c25e4 | 3146 | if (virtio_has_feature(vi->vdev, VIRTIO_NET_F_STATUS)) { |
3b07e9ca | 3147 | schedule_work(&vi->config_work); |
167c25e4 JW |
3148 | } else { |
3149 | vi->status = VIRTIO_NET_S_LINK_UP; | |
faa9b39f | 3150 | virtnet_update_settings(vi); |
167c25e4 JW |
3151 | netif_carrier_on(dev); |
3152 | } | |
9f4d26d0 | 3153 | |
3f93522f JW |
3154 | for (i = 0; i < ARRAY_SIZE(guest_offloads); i++) |
3155 | if (virtio_has_feature(vi->vdev, guest_offloads[i])) | |
3156 | set_bit(guest_offloads[i], &vi->guest_offloads); | |
a02e8964 | 3157 | vi->guest_offloads_capable = vi->guest_offloads; |
3f93522f | 3158 | |
986a4f4d JW |
3159 | pr_debug("virtnet: registered device %s with %d RX and TX vq's\n", |
3160 | dev->name, max_queue_pairs); | |
3161 | ||
296f96fc RR |
3162 | return 0; |
3163 | ||
f00e35e2 | 3164 | free_unregister_netdev: |
02465555 MT |
3165 | vi->vdev->config->reset(vdev); |
3166 | ||
b3369c1f | 3167 | unregister_netdev(dev); |
ba5e4426 SS |
3168 | free_failover: |
3169 | net_failover_destroy(vi->failover); | |
d2a7ddda | 3170 | free_vqs: |
986a4f4d | 3171 | cancel_delayed_work_sync(&vi->refill); |
fb51879d | 3172 | free_receive_page_frags(vi); |
e9d7417b | 3173 | virtnet_del_vqs(vi); |
296f96fc RR |
3174 | free: |
3175 | free_netdev(dev); | |
3176 | return err; | |
3177 | } | |
3178 | ||
04486ed0 | 3179 | static void remove_vq_common(struct virtnet_info *vi) |
296f96fc | 3180 | { |
04486ed0 | 3181 | vi->vdev->config->reset(vi->vdev); |
830a8a97 SM |
3182 | |
3183 | /* Free unused buffers in both send and recv, if any. */ | |
9ab86bbc | 3184 | free_unused_bufs(vi); |
fb6813f4 | 3185 | |
986a4f4d | 3186 | free_receive_bufs(vi); |
d2a7ddda | 3187 | |
fb51879d MD |
3188 | free_receive_page_frags(vi); |
3189 | ||
986a4f4d | 3190 | virtnet_del_vqs(vi); |
04486ed0 AS |
3191 | } |
3192 | ||
8cc085d6 | 3193 | static void virtnet_remove(struct virtio_device *vdev) |
04486ed0 AS |
3194 | { |
3195 | struct virtnet_info *vi = vdev->priv; | |
3196 | ||
8017c279 | 3197 | virtnet_cpu_notif_remove(vi); |
8de4b2f3 | 3198 | |
102a2786 MT |
3199 | /* Make sure no work handler is accessing the device. */ |
3200 | flush_work(&vi->config_work); | |
586d17c5 | 3201 | |
04486ed0 AS |
3202 | unregister_netdev(vi->dev); |
3203 | ||
ba5e4426 SS |
3204 | net_failover_destroy(vi->failover); |
3205 | ||
04486ed0 | 3206 | remove_vq_common(vi); |
fb6813f4 | 3207 | |
74b2553f | 3208 | free_netdev(vi->dev); |
296f96fc RR |
3209 | } |
3210 | ||
67a75194 | 3211 | static __maybe_unused int virtnet_freeze(struct virtio_device *vdev) |
0741bcb5 AS |
3212 | { |
3213 | struct virtnet_info *vi = vdev->priv; | |
3214 | ||
8017c279 | 3215 | virtnet_cpu_notif_remove(vi); |
9fe7bfce | 3216 | virtnet_freeze_down(vdev); |
0741bcb5 AS |
3217 | remove_vq_common(vi); |
3218 | ||
3219 | return 0; | |
3220 | } | |
3221 | ||
67a75194 | 3222 | static __maybe_unused int virtnet_restore(struct virtio_device *vdev) |
0741bcb5 AS |
3223 | { |
3224 | struct virtnet_info *vi = vdev->priv; | |
9fe7bfce | 3225 | int err; |
0741bcb5 | 3226 | |
9fe7bfce | 3227 | err = virtnet_restore_up(vdev); |
0741bcb5 AS |
3228 | if (err) |
3229 | return err; | |
986a4f4d JW |
3230 | virtnet_set_queues(vi, vi->curr_queue_pairs); |
3231 | ||
8017c279 | 3232 | err = virtnet_cpu_notif_add(vi); |
ec9debbd JW |
3233 | if (err) |
3234 | return err; | |
3235 | ||
0741bcb5 AS |
3236 | return 0; |
3237 | } | |
0741bcb5 | 3238 | |
296f96fc RR |
3239 | static struct virtio_device_id id_table[] = { |
3240 | { VIRTIO_ID_NET, VIRTIO_DEV_ANY_ID }, | |
3241 | { 0 }, | |
3242 | }; | |
3243 | ||
f3358507 MT |
3244 | #define VIRTNET_FEATURES \ |
3245 | VIRTIO_NET_F_CSUM, VIRTIO_NET_F_GUEST_CSUM, \ | |
3246 | VIRTIO_NET_F_MAC, \ | |
3247 | VIRTIO_NET_F_HOST_TSO4, VIRTIO_NET_F_HOST_UFO, VIRTIO_NET_F_HOST_TSO6, \ | |
3248 | VIRTIO_NET_F_HOST_ECN, VIRTIO_NET_F_GUEST_TSO4, VIRTIO_NET_F_GUEST_TSO6, \ | |
3249 | VIRTIO_NET_F_GUEST_ECN, VIRTIO_NET_F_GUEST_UFO, \ | |
3250 | VIRTIO_NET_F_MRG_RXBUF, VIRTIO_NET_F_STATUS, VIRTIO_NET_F_CTRL_VQ, \ | |
3251 | VIRTIO_NET_F_CTRL_RX, VIRTIO_NET_F_CTRL_VLAN, \ | |
3252 | VIRTIO_NET_F_GUEST_ANNOUNCE, VIRTIO_NET_F_MQ, \ | |
3253 | VIRTIO_NET_F_CTRL_MAC_ADDR, \ | |
faa9b39f | 3254 | VIRTIO_NET_F_MTU, VIRTIO_NET_F_CTRL_GUEST_OFFLOADS, \ |
9805069d | 3255 | VIRTIO_NET_F_SPEED_DUPLEX, VIRTIO_NET_F_STANDBY |
f3358507 | 3256 | |
c45a6816 | 3257 | static unsigned int features[] = { |
f3358507 MT |
3258 | VIRTNET_FEATURES, |
3259 | }; | |
3260 | ||
3261 | static unsigned int features_legacy[] = { | |
3262 | VIRTNET_FEATURES, | |
3263 | VIRTIO_NET_F_GSO, | |
e7428e95 | 3264 | VIRTIO_F_ANY_LAYOUT, |
c45a6816 RR |
3265 | }; |
3266 | ||
22402529 | 3267 | static struct virtio_driver virtio_net_driver = { |
c45a6816 RR |
3268 | .feature_table = features, |
3269 | .feature_table_size = ARRAY_SIZE(features), | |
f3358507 MT |
3270 | .feature_table_legacy = features_legacy, |
3271 | .feature_table_size_legacy = ARRAY_SIZE(features_legacy), | |
296f96fc RR |
3272 | .driver.name = KBUILD_MODNAME, |
3273 | .driver.owner = THIS_MODULE, | |
3274 | .id_table = id_table, | |
fe36cbe0 | 3275 | .validate = virtnet_validate, |
296f96fc | 3276 | .probe = virtnet_probe, |
8cc085d6 | 3277 | .remove = virtnet_remove, |
9f4d26d0 | 3278 | .config_changed = virtnet_config_changed, |
89107000 | 3279 | #ifdef CONFIG_PM_SLEEP |
0741bcb5 AS |
3280 | .freeze = virtnet_freeze, |
3281 | .restore = virtnet_restore, | |
3282 | #endif | |
296f96fc RR |
3283 | }; |
3284 | ||
8017c279 SAS |
3285 | static __init int virtio_net_driver_init(void) |
3286 | { | |
3287 | int ret; | |
3288 | ||
73c1b41e | 3289 | ret = cpuhp_setup_state_multi(CPUHP_AP_ONLINE_DYN, "virtio/net:online", |
8017c279 SAS |
3290 | virtnet_cpu_online, |
3291 | virtnet_cpu_down_prep); | |
3292 | if (ret < 0) | |
3293 | goto out; | |
3294 | virtionet_online = ret; | |
73c1b41e | 3295 | ret = cpuhp_setup_state_multi(CPUHP_VIRT_NET_DEAD, "virtio/net:dead", |
8017c279 SAS |
3296 | NULL, virtnet_cpu_dead); |
3297 | if (ret) | |
3298 | goto err_dead; | |
3299 | ||
3300 | ret = register_virtio_driver(&virtio_net_driver); | |
3301 | if (ret) | |
3302 | goto err_virtio; | |
3303 | return 0; | |
3304 | err_virtio: | |
3305 | cpuhp_remove_multi_state(CPUHP_VIRT_NET_DEAD); | |
3306 | err_dead: | |
3307 | cpuhp_remove_multi_state(virtionet_online); | |
3308 | out: | |
3309 | return ret; | |
3310 | } | |
3311 | module_init(virtio_net_driver_init); | |
3312 | ||
3313 | static __exit void virtio_net_driver_exit(void) | |
3314 | { | |
cfa0ebc9 | 3315 | unregister_virtio_driver(&virtio_net_driver); |
8017c279 SAS |
3316 | cpuhp_remove_multi_state(CPUHP_VIRT_NET_DEAD); |
3317 | cpuhp_remove_multi_state(virtionet_online); | |
8017c279 SAS |
3318 | } |
3319 | module_exit(virtio_net_driver_exit); | |
296f96fc RR |
3320 | |
3321 | MODULE_DEVICE_TABLE(virtio, id_table); | |
3322 | MODULE_DESCRIPTION("Virtio network driver"); | |
3323 | MODULE_LICENSE("GPL"); |