]>
Commit | Line | Data |
---|---|---|
48925e37 | 1 | /* A network driver using virtio. |
296f96fc RR |
2 | * |
3 | * Copyright 2007 Rusty Russell <rusty@rustcorp.com.au> IBM Corporation | |
4 | * | |
5 | * This program is free software; you can redistribute it and/or modify | |
6 | * it under the terms of the GNU General Public License as published by | |
7 | * the Free Software Foundation; either version 2 of the License, or | |
8 | * (at your option) any later version. | |
9 | * | |
10 | * This program is distributed in the hope that it will be useful, | |
11 | * but WITHOUT ANY WARRANTY; without even the implied warranty of | |
12 | * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the | |
13 | * GNU General Public License for more details. | |
14 | * | |
15 | * You should have received a copy of the GNU General Public License | |
adf8d3ff | 16 | * along with this program; if not, see <http://www.gnu.org/licenses/>. |
296f96fc RR |
17 | */ |
18 | //#define DEBUG | |
19 | #include <linux/netdevice.h> | |
20 | #include <linux/etherdevice.h> | |
a9ea3fc6 | 21 | #include <linux/ethtool.h> |
296f96fc RR |
22 | #include <linux/module.h> |
23 | #include <linux/virtio.h> | |
24 | #include <linux/virtio_net.h> | |
f600b690 | 25 | #include <linux/bpf.h> |
a67edbf4 | 26 | #include <linux/bpf_trace.h> |
296f96fc | 27 | #include <linux/scatterlist.h> |
e918085a | 28 | #include <linux/if_vlan.h> |
5a0e3ad6 | 29 | #include <linux/slab.h> |
8de4b2f3 | 30 | #include <linux/cpu.h> |
ab7db917 | 31 | #include <linux/average.h> |
296f96fc | 32 | |
d34710e3 | 33 | static int napi_weight = NAPI_POLL_WEIGHT; |
6c0cd7c0 DL |
34 | module_param(napi_weight, int, 0444); |
35 | ||
eb939922 | 36 | static bool csum = true, gso = true; |
34a48579 RR |
37 | module_param(csum, bool, 0444); |
38 | module_param(gso, bool, 0444); | |
39 | ||
296f96fc | 40 | /* FIXME: MTU in config. */ |
5061de36 | 41 | #define GOOD_PACKET_LEN (ETH_HLEN + VLAN_HLEN + ETH_DATA_LEN) |
3f2c31d9 | 42 | #define GOOD_COPY_LEN 128 |
296f96fc | 43 | |
2de2f7f4 JF |
44 | /* Amount of XDP headroom to prepend to packets for use by xdp_adjust_head */ |
45 | #define VIRTIO_XDP_HEADROOM 256 | |
46 | ||
5377d758 JB |
47 | /* RX packet size EWMA. The average packet size is used to determine the packet |
48 | * buffer size when refilling RX rings. As the entire RX ring may be refilled | |
49 | * at once, the weight is chosen so that the EWMA will be insensitive to short- | |
50 | * term, transient changes in packet size. | |
ab7db917 | 51 | */ |
5377d758 | 52 | DECLARE_EWMA(pkt_len, 1, 64) |
ab7db917 | 53 | |
d0fa28f0 MT |
54 | /* With mergeable buffers we align buffer address and use the low bits to |
55 | * encode its true size. Buffer size is up to 1 page so we need to align to | |
56 | * square root of page size to ensure we reserve enough bits to encode the true | |
57 | * size. | |
58 | */ | |
59 | #define MERGEABLE_BUFFER_MIN_ALIGN_SHIFT ((PAGE_SHIFT + 1) / 2) | |
60 | ||
ab7db917 | 61 | /* Minimum alignment for mergeable packet buffers. */ |
d0fa28f0 MT |
62 | #define MERGEABLE_BUFFER_ALIGN max(L1_CACHE_BYTES, \ |
63 | 1 << MERGEABLE_BUFFER_MIN_ALIGN_SHIFT) | |
ab7db917 | 64 | |
66846048 | 65 | #define VIRTNET_DRIVER_VERSION "1.0.0" |
2a41f71d | 66 | |
3fa2a1df | 67 | struct virtnet_stats { |
83a27052 ED |
68 | struct u64_stats_sync tx_syncp; |
69 | struct u64_stats_sync rx_syncp; | |
3fa2a1df | 70 | u64 tx_bytes; |
71 | u64 tx_packets; | |
72 | ||
73 | u64 rx_bytes; | |
74 | u64 rx_packets; | |
75 | }; | |
76 | ||
e9d7417b JW |
77 | /* Internal representation of a send virtqueue */ |
78 | struct send_queue { | |
79 | /* Virtqueue associated with this send _queue */ | |
80 | struct virtqueue *vq; | |
81 | ||
82 | /* TX: fragments + linear part + virtio header */ | |
83 | struct scatterlist sg[MAX_SKB_FRAGS + 2]; | |
986a4f4d JW |
84 | |
85 | /* Name of the send queue: output.$index */ | |
86 | char name[40]; | |
e9d7417b JW |
87 | }; |
88 | ||
89 | /* Internal representation of a receive virtqueue */ | |
90 | struct receive_queue { | |
91 | /* Virtqueue associated with this receive_queue */ | |
92 | struct virtqueue *vq; | |
93 | ||
296f96fc RR |
94 | struct napi_struct napi; |
95 | ||
f600b690 JF |
96 | struct bpf_prog __rcu *xdp_prog; |
97 | ||
e9d7417b JW |
98 | /* Chain pages by the private ptr. */ |
99 | struct page *pages; | |
100 | ||
ab7db917 | 101 | /* Average packet length for mergeable receive buffers. */ |
5377d758 | 102 | struct ewma_pkt_len mrg_avg_pkt_len; |
ab7db917 | 103 | |
fb51879d MD |
104 | /* Page frag for packet buffer allocation. */ |
105 | struct page_frag alloc_frag; | |
106 | ||
e9d7417b JW |
107 | /* RX: fragments + linear part + virtio header */ |
108 | struct scatterlist sg[MAX_SKB_FRAGS + 2]; | |
986a4f4d JW |
109 | |
110 | /* Name of this receive queue: input.$index */ | |
111 | char name[40]; | |
e9d7417b JW |
112 | }; |
113 | ||
114 | struct virtnet_info { | |
115 | struct virtio_device *vdev; | |
116 | struct virtqueue *cvq; | |
117 | struct net_device *dev; | |
986a4f4d JW |
118 | struct send_queue *sq; |
119 | struct receive_queue *rq; | |
e9d7417b JW |
120 | unsigned int status; |
121 | ||
986a4f4d JW |
122 | /* Max # of queue pairs supported by the device */ |
123 | u16 max_queue_pairs; | |
124 | ||
125 | /* # of queue pairs currently used by the driver */ | |
126 | u16 curr_queue_pairs; | |
127 | ||
672aafd5 JF |
128 | /* # of XDP queue pairs currently used by the driver */ |
129 | u16 xdp_queue_pairs; | |
130 | ||
97402b96 HX |
131 | /* I like... big packets and I cannot lie! */ |
132 | bool big_packets; | |
133 | ||
3f2c31d9 MM |
134 | /* Host will merge rx buffers for big packets (shake it! shake it!) */ |
135 | bool mergeable_rx_bufs; | |
136 | ||
986a4f4d JW |
137 | /* Has control virtqueue */ |
138 | bool has_cvq; | |
139 | ||
e7428e95 MT |
140 | /* Host can handle any s/g split between our header and packet data */ |
141 | bool any_header_sg; | |
142 | ||
012873d0 MT |
143 | /* Packet virtio header size */ |
144 | u8 hdr_len; | |
145 | ||
3fa2a1df | 146 | /* Active statistics */ |
147 | struct virtnet_stats __percpu *stats; | |
148 | ||
3161e453 RR |
149 | /* Work struct for refilling if we run low on memory. */ |
150 | struct delayed_work refill; | |
151 | ||
586d17c5 JW |
152 | /* Work struct for config space updates */ |
153 | struct work_struct config_work; | |
154 | ||
986a4f4d JW |
155 | /* Does the affinity hint is set for virtqueues? */ |
156 | bool affinity_hint_set; | |
47be2479 | 157 | |
8017c279 SAS |
158 | /* CPU hotplug instances for online & dead */ |
159 | struct hlist_node node; | |
160 | struct hlist_node node_dead; | |
2ac46030 MT |
161 | |
162 | /* Control VQ buffers: protected by the rtnl lock */ | |
163 | struct virtio_net_ctrl_hdr ctrl_hdr; | |
164 | virtio_net_ctrl_ack ctrl_status; | |
a725ee3e | 165 | struct virtio_net_ctrl_mq ctrl_mq; |
2ac46030 MT |
166 | u8 ctrl_promisc; |
167 | u8 ctrl_allmulti; | |
a725ee3e | 168 | u16 ctrl_vid; |
16032be5 NA |
169 | |
170 | /* Ethtool settings */ | |
171 | u8 duplex; | |
172 | u32 speed; | |
296f96fc RR |
173 | }; |
174 | ||
9ab86bbc | 175 | struct padded_vnet_hdr { |
012873d0 | 176 | struct virtio_net_hdr_mrg_rxbuf hdr; |
9ab86bbc | 177 | /* |
012873d0 MT |
178 | * hdr is in a separate sg buffer, and data sg buffer shares same page |
179 | * with this header sg. This padding makes next sg 16 byte aligned | |
180 | * after the header. | |
9ab86bbc | 181 | */ |
012873d0 | 182 | char padding[4]; |
9ab86bbc SM |
183 | }; |
184 | ||
986a4f4d JW |
185 | /* Converting between virtqueue no. and kernel tx/rx queue no. |
186 | * 0:rx0 1:tx0 2:rx1 3:tx1 ... 2N:rxN 2N+1:txN 2N+2:cvq | |
187 | */ | |
188 | static int vq2txq(struct virtqueue *vq) | |
189 | { | |
9d0ca6ed | 190 | return (vq->index - 1) / 2; |
986a4f4d JW |
191 | } |
192 | ||
193 | static int txq2vq(int txq) | |
194 | { | |
195 | return txq * 2 + 1; | |
196 | } | |
197 | ||
198 | static int vq2rxq(struct virtqueue *vq) | |
199 | { | |
9d0ca6ed | 200 | return vq->index / 2; |
986a4f4d JW |
201 | } |
202 | ||
203 | static int rxq2vq(int rxq) | |
204 | { | |
205 | return rxq * 2; | |
206 | } | |
207 | ||
012873d0 | 208 | static inline struct virtio_net_hdr_mrg_rxbuf *skb_vnet_hdr(struct sk_buff *skb) |
296f96fc | 209 | { |
012873d0 | 210 | return (struct virtio_net_hdr_mrg_rxbuf *)skb->cb; |
296f96fc RR |
211 | } |
212 | ||
9ab86bbc SM |
213 | /* |
214 | * private is used to chain pages for big packets, put the whole | |
215 | * most recent used list in the beginning for reuse | |
216 | */ | |
e9d7417b | 217 | static void give_pages(struct receive_queue *rq, struct page *page) |
0a888fd1 | 218 | { |
9ab86bbc | 219 | struct page *end; |
0a888fd1 | 220 | |
e9d7417b | 221 | /* Find end of list, sew whole thing into vi->rq.pages. */ |
9ab86bbc | 222 | for (end = page; end->private; end = (struct page *)end->private); |
e9d7417b JW |
223 | end->private = (unsigned long)rq->pages; |
224 | rq->pages = page; | |
0a888fd1 MM |
225 | } |
226 | ||
e9d7417b | 227 | static struct page *get_a_page(struct receive_queue *rq, gfp_t gfp_mask) |
fb6813f4 | 228 | { |
e9d7417b | 229 | struct page *p = rq->pages; |
fb6813f4 | 230 | |
9ab86bbc | 231 | if (p) { |
e9d7417b | 232 | rq->pages = (struct page *)p->private; |
9ab86bbc SM |
233 | /* clear private here, it is used to chain pages */ |
234 | p->private = 0; | |
235 | } else | |
fb6813f4 RR |
236 | p = alloc_page(gfp_mask); |
237 | return p; | |
238 | } | |
239 | ||
e9d7417b | 240 | static void skb_xmit_done(struct virtqueue *vq) |
296f96fc | 241 | { |
e9d7417b | 242 | struct virtnet_info *vi = vq->vdev->priv; |
296f96fc | 243 | |
2cb9c6ba | 244 | /* Suppress further interrupts. */ |
e9d7417b | 245 | virtqueue_disable_cb(vq); |
11a3a154 | 246 | |
363f1514 | 247 | /* We were probably waiting for more output buffers. */ |
986a4f4d | 248 | netif_wake_subqueue(vi->dev, vq2txq(vq)); |
296f96fc RR |
249 | } |
250 | ||
ab7db917 MD |
251 | static unsigned int mergeable_ctx_to_buf_truesize(unsigned long mrg_ctx) |
252 | { | |
253 | unsigned int truesize = mrg_ctx & (MERGEABLE_BUFFER_ALIGN - 1); | |
254 | return (truesize + 1) * MERGEABLE_BUFFER_ALIGN; | |
255 | } | |
256 | ||
257 | static void *mergeable_ctx_to_buf_address(unsigned long mrg_ctx) | |
258 | { | |
259 | return (void *)(mrg_ctx & -MERGEABLE_BUFFER_ALIGN); | |
260 | ||
261 | } | |
262 | ||
263 | static unsigned long mergeable_buf_to_ctx(void *buf, unsigned int truesize) | |
264 | { | |
265 | unsigned int size = truesize / MERGEABLE_BUFFER_ALIGN; | |
266 | return (unsigned long)buf | (size - 1); | |
267 | } | |
268 | ||
3464645a | 269 | /* Called from bottom half context */ |
946fa564 MT |
270 | static struct sk_buff *page_to_skb(struct virtnet_info *vi, |
271 | struct receive_queue *rq, | |
2613af0e MD |
272 | struct page *page, unsigned int offset, |
273 | unsigned int len, unsigned int truesize) | |
9ab86bbc SM |
274 | { |
275 | struct sk_buff *skb; | |
012873d0 | 276 | struct virtio_net_hdr_mrg_rxbuf *hdr; |
2613af0e | 277 | unsigned int copy, hdr_len, hdr_padded_len; |
9ab86bbc | 278 | char *p; |
fb6813f4 | 279 | |
2613af0e | 280 | p = page_address(page) + offset; |
3f2c31d9 | 281 | |
9ab86bbc | 282 | /* copy small packet so we can reuse these pages for small data */ |
c67f5db8 | 283 | skb = napi_alloc_skb(&rq->napi, GOOD_COPY_LEN); |
9ab86bbc SM |
284 | if (unlikely(!skb)) |
285 | return NULL; | |
3f2c31d9 | 286 | |
9ab86bbc | 287 | hdr = skb_vnet_hdr(skb); |
3f2c31d9 | 288 | |
012873d0 MT |
289 | hdr_len = vi->hdr_len; |
290 | if (vi->mergeable_rx_bufs) | |
291 | hdr_padded_len = sizeof *hdr; | |
292 | else | |
2613af0e | 293 | hdr_padded_len = sizeof(struct padded_vnet_hdr); |
3f2c31d9 | 294 | |
9ab86bbc | 295 | memcpy(hdr, p, hdr_len); |
3f2c31d9 | 296 | |
9ab86bbc | 297 | len -= hdr_len; |
2613af0e MD |
298 | offset += hdr_padded_len; |
299 | p += hdr_padded_len; | |
3f2c31d9 | 300 | |
9ab86bbc SM |
301 | copy = len; |
302 | if (copy > skb_tailroom(skb)) | |
303 | copy = skb_tailroom(skb); | |
304 | memcpy(skb_put(skb, copy), p, copy); | |
3f2c31d9 | 305 | |
9ab86bbc SM |
306 | len -= copy; |
307 | offset += copy; | |
3f2c31d9 | 308 | |
2613af0e MD |
309 | if (vi->mergeable_rx_bufs) { |
310 | if (len) | |
311 | skb_add_rx_frag(skb, 0, page, offset, len, truesize); | |
312 | else | |
313 | put_page(page); | |
314 | return skb; | |
315 | } | |
316 | ||
e878d78b SL |
317 | /* |
318 | * Verify that we can indeed put this data into a skb. | |
319 | * This is here to handle cases when the device erroneously | |
320 | * tries to receive more than is possible. This is usually | |
321 | * the case of a broken device. | |
322 | */ | |
323 | if (unlikely(len > MAX_SKB_FRAGS * PAGE_SIZE)) { | |
be443899 | 324 | net_dbg_ratelimited("%s: too much data\n", skb->dev->name); |
e878d78b SL |
325 | dev_kfree_skb(skb); |
326 | return NULL; | |
327 | } | |
2613af0e | 328 | BUG_ON(offset >= PAGE_SIZE); |
9ab86bbc | 329 | while (len) { |
2613af0e MD |
330 | unsigned int frag_size = min((unsigned)PAGE_SIZE - offset, len); |
331 | skb_add_rx_frag(skb, skb_shinfo(skb)->nr_frags, page, offset, | |
332 | frag_size, truesize); | |
333 | len -= frag_size; | |
9ab86bbc SM |
334 | page = (struct page *)page->private; |
335 | offset = 0; | |
336 | } | |
3f2c31d9 | 337 | |
9ab86bbc | 338 | if (page) |
e9d7417b | 339 | give_pages(rq, page); |
3f2c31d9 | 340 | |
9ab86bbc SM |
341 | return skb; |
342 | } | |
3f2c31d9 | 343 | |
a67edbf4 | 344 | static bool virtnet_xdp_xmit(struct virtnet_info *vi, |
56434a01 | 345 | struct receive_queue *rq, |
bb91accf JW |
346 | struct xdp_buff *xdp, |
347 | void *data) | |
56434a01 | 348 | { |
56434a01 JF |
349 | struct virtio_net_hdr_mrg_rxbuf *hdr; |
350 | unsigned int num_sg, len; | |
722d8283 JF |
351 | struct send_queue *sq; |
352 | unsigned int qp; | |
56434a01 JF |
353 | void *xdp_sent; |
354 | int err; | |
355 | ||
722d8283 JF |
356 | qp = vi->curr_queue_pairs - vi->xdp_queue_pairs + smp_processor_id(); |
357 | sq = &vi->sq[qp]; | |
358 | ||
56434a01 JF |
359 | /* Free up any pending old buffers before queueing new ones. */ |
360 | while ((xdp_sent = virtqueue_get_buf(sq->vq, &len)) != NULL) { | |
bb91accf JW |
361 | if (vi->mergeable_rx_bufs) { |
362 | struct page *sent_page = virt_to_head_page(xdp_sent); | |
363 | ||
364 | put_page(sent_page); | |
365 | } else { /* small buffer */ | |
366 | struct sk_buff *skb = xdp_sent; | |
56434a01 | 367 | |
bb91accf JW |
368 | kfree_skb(skb); |
369 | } | |
370 | } | |
56434a01 | 371 | |
bb91accf | 372 | if (vi->mergeable_rx_bufs) { |
2de2f7f4 | 373 | xdp->data -= sizeof(struct virtio_net_hdr_mrg_rxbuf); |
bb91accf JW |
374 | /* Zero header and leave csum up to XDP layers */ |
375 | hdr = xdp->data; | |
376 | memset(hdr, 0, vi->hdr_len); | |
377 | ||
378 | num_sg = 1; | |
379 | sg_init_one(sq->sg, xdp->data, xdp->data_end - xdp->data); | |
380 | } else { /* small buffer */ | |
381 | struct sk_buff *skb = data; | |
382 | ||
383 | /* Zero header and leave csum up to XDP layers */ | |
384 | hdr = skb_vnet_hdr(skb); | |
385 | memset(hdr, 0, vi->hdr_len); | |
386 | ||
387 | num_sg = 2; | |
388 | sg_init_table(sq->sg, 2); | |
389 | sg_set_buf(sq->sg, hdr, vi->hdr_len); | |
2de2f7f4 JF |
390 | skb_to_sgvec(skb, sq->sg + 1, |
391 | xdp->data - xdp->data_hard_start, | |
392 | xdp->data_end - xdp->data); | |
bb91accf | 393 | } |
56434a01 | 394 | err = virtqueue_add_outbuf(sq->vq, sq->sg, num_sg, |
bb91accf | 395 | data, GFP_ATOMIC); |
56434a01 | 396 | if (unlikely(err)) { |
bb91accf JW |
397 | if (vi->mergeable_rx_bufs) { |
398 | struct page *page = virt_to_head_page(xdp->data); | |
399 | ||
400 | put_page(page); | |
401 | } else /* small buffer */ | |
402 | kfree_skb(data); | |
a67edbf4 DB |
403 | /* On error abort to avoid unnecessary kick */ |
404 | return false; | |
56434a01 JF |
405 | } |
406 | ||
407 | virtqueue_kick(sq->vq); | |
a67edbf4 | 408 | return true; |
56434a01 JF |
409 | } |
410 | ||
bb91accf JW |
411 | static struct sk_buff *receive_small(struct net_device *dev, |
412 | struct virtnet_info *vi, | |
413 | struct receive_queue *rq, | |
414 | void *buf, unsigned int len) | |
f121159d MT |
415 | { |
416 | struct sk_buff * skb = buf; | |
bb91accf | 417 | struct bpf_prog *xdp_prog; |
f121159d | 418 | |
012873d0 | 419 | len -= vi->hdr_len; |
f121159d | 420 | |
bb91accf JW |
421 | rcu_read_lock(); |
422 | xdp_prog = rcu_dereference(rq->xdp_prog); | |
423 | if (xdp_prog) { | |
424 | struct virtio_net_hdr_mrg_rxbuf *hdr = buf; | |
0354e4d1 | 425 | struct xdp_buff xdp; |
bb91accf JW |
426 | u32 act; |
427 | ||
428 | if (unlikely(hdr->hdr.gso_type || hdr->hdr.flags)) | |
429 | goto err_xdp; | |
0354e4d1 | 430 | |
2de2f7f4 JF |
431 | xdp.data_hard_start = skb->data; |
432 | xdp.data = skb->data + VIRTIO_XDP_HEADROOM; | |
0354e4d1 JF |
433 | xdp.data_end = xdp.data + len; |
434 | act = bpf_prog_run_xdp(xdp_prog, &xdp); | |
435 | ||
bb91accf JW |
436 | switch (act) { |
437 | case XDP_PASS: | |
2de2f7f4 JF |
438 | /* Recalculate length in case bpf program changed it */ |
439 | __skb_pull(skb, xdp.data - xdp.data_hard_start); | |
440 | len = xdp.data_end - xdp.data; | |
bb91accf JW |
441 | break; |
442 | case XDP_TX: | |
722d8283 | 443 | if (unlikely(!virtnet_xdp_xmit(vi, rq, &xdp, skb))) |
0354e4d1 | 444 | trace_xdp_exception(vi->dev, xdp_prog, act); |
bb91accf JW |
445 | rcu_read_unlock(); |
446 | goto xdp_xmit; | |
bb91accf | 447 | default: |
0354e4d1 JF |
448 | bpf_warn_invalid_xdp_action(act); |
449 | case XDP_ABORTED: | |
450 | trace_xdp_exception(vi->dev, xdp_prog, act); | |
451 | case XDP_DROP: | |
bb91accf JW |
452 | goto err_xdp; |
453 | } | |
454 | } | |
455 | rcu_read_unlock(); | |
456 | ||
2de2f7f4 | 457 | skb_trim(skb, len); |
f121159d | 458 | return skb; |
bb91accf JW |
459 | |
460 | err_xdp: | |
461 | rcu_read_unlock(); | |
462 | dev->stats.rx_dropped++; | |
463 | kfree_skb(skb); | |
464 | xdp_xmit: | |
465 | return NULL; | |
f121159d MT |
466 | } |
467 | ||
468 | static struct sk_buff *receive_big(struct net_device *dev, | |
946fa564 | 469 | struct virtnet_info *vi, |
f121159d MT |
470 | struct receive_queue *rq, |
471 | void *buf, | |
472 | unsigned int len) | |
473 | { | |
474 | struct page *page = buf; | |
c47a43d3 | 475 | struct sk_buff *skb = page_to_skb(vi, rq, page, 0, len, PAGE_SIZE); |
f600b690 | 476 | |
f121159d MT |
477 | if (unlikely(!skb)) |
478 | goto err; | |
479 | ||
480 | return skb; | |
481 | ||
482 | err: | |
483 | dev->stats.rx_dropped++; | |
484 | give_pages(rq, page); | |
485 | return NULL; | |
486 | } | |
487 | ||
72979a6c JF |
488 | /* The conditions to enable XDP should preclude the underlying device from |
489 | * sending packets across multiple buffers (num_buf > 1). However per spec | |
490 | * it does not appear to be illegal to do so but rather just against convention. | |
491 | * So in order to avoid making a system unresponsive the packets are pushed | |
492 | * into a page and the XDP program is run. This will be extremely slow and we | |
493 | * push a warning to the user to fix this as soon as possible. Fixing this may | |
494 | * require resolving the underlying hardware to determine why multiple buffers | |
495 | * are being received or simply loading the XDP program in the ingress stack | |
496 | * after the skb is built because there is no advantage to running it here | |
497 | * anymore. | |
498 | */ | |
499 | static struct page *xdp_linearize_page(struct receive_queue *rq, | |
56a86f84 | 500 | u16 *num_buf, |
72979a6c JF |
501 | struct page *p, |
502 | int offset, | |
503 | unsigned int *len) | |
504 | { | |
505 | struct page *page = alloc_page(GFP_ATOMIC); | |
2de2f7f4 | 506 | unsigned int page_off = VIRTIO_XDP_HEADROOM; |
72979a6c JF |
507 | |
508 | if (!page) | |
509 | return NULL; | |
510 | ||
511 | memcpy(page_address(page) + page_off, page_address(p) + offset, *len); | |
512 | page_off += *len; | |
513 | ||
56a86f84 | 514 | while (--*num_buf) { |
72979a6c JF |
515 | unsigned int buflen; |
516 | unsigned long ctx; | |
517 | void *buf; | |
518 | int off; | |
519 | ||
520 | ctx = (unsigned long)virtqueue_get_buf(rq->vq, &buflen); | |
521 | if (unlikely(!ctx)) | |
522 | goto err_buf; | |
523 | ||
56a86f84 JW |
524 | buf = mergeable_ctx_to_buf_address(ctx); |
525 | p = virt_to_head_page(buf); | |
526 | off = buf - page_address(p); | |
527 | ||
72979a6c JF |
528 | /* guard against a misconfigured or uncooperative backend that |
529 | * is sending packet larger than the MTU. | |
530 | */ | |
56a86f84 JW |
531 | if ((page_off + buflen) > PAGE_SIZE) { |
532 | put_page(p); | |
72979a6c | 533 | goto err_buf; |
56a86f84 | 534 | } |
72979a6c JF |
535 | |
536 | memcpy(page_address(page) + page_off, | |
537 | page_address(p) + off, buflen); | |
538 | page_off += buflen; | |
56a86f84 | 539 | put_page(p); |
72979a6c JF |
540 | } |
541 | ||
2de2f7f4 JF |
542 | /* Headroom does not contribute to packet length */ |
543 | *len = page_off - VIRTIO_XDP_HEADROOM; | |
72979a6c JF |
544 | return page; |
545 | err_buf: | |
546 | __free_pages(page, 0); | |
547 | return NULL; | |
548 | } | |
549 | ||
8fc3b9e9 | 550 | static struct sk_buff *receive_mergeable(struct net_device *dev, |
fdd819b2 | 551 | struct virtnet_info *vi, |
8fc3b9e9 | 552 | struct receive_queue *rq, |
ab7db917 | 553 | unsigned long ctx, |
8fc3b9e9 | 554 | unsigned int len) |
9ab86bbc | 555 | { |
ab7db917 | 556 | void *buf = mergeable_ctx_to_buf_address(ctx); |
012873d0 MT |
557 | struct virtio_net_hdr_mrg_rxbuf *hdr = buf; |
558 | u16 num_buf = virtio16_to_cpu(vi->vdev, hdr->num_buffers); | |
8fc3b9e9 MT |
559 | struct page *page = virt_to_head_page(buf); |
560 | int offset = buf - page_address(page); | |
f600b690 JF |
561 | struct sk_buff *head_skb, *curr_skb; |
562 | struct bpf_prog *xdp_prog; | |
563 | unsigned int truesize; | |
564 | ||
56434a01 JF |
565 | head_skb = NULL; |
566 | ||
f600b690 JF |
567 | rcu_read_lock(); |
568 | xdp_prog = rcu_dereference(rq->xdp_prog); | |
569 | if (xdp_prog) { | |
72979a6c | 570 | struct page *xdp_page; |
0354e4d1 | 571 | struct xdp_buff xdp; |
0354e4d1 | 572 | void *data; |
f600b690 JF |
573 | u32 act; |
574 | ||
73b62bd0 | 575 | /* This happens when rx buffer size is underestimated */ |
f600b690 | 576 | if (unlikely(num_buf > 1)) { |
72979a6c | 577 | /* linearize data for XDP */ |
56a86f84 | 578 | xdp_page = xdp_linearize_page(rq, &num_buf, |
72979a6c JF |
579 | page, offset, &len); |
580 | if (!xdp_page) | |
581 | goto err_xdp; | |
2de2f7f4 | 582 | offset = VIRTIO_XDP_HEADROOM; |
72979a6c JF |
583 | } else { |
584 | xdp_page = page; | |
f600b690 JF |
585 | } |
586 | ||
587 | /* Transient failure which in theory could occur if | |
588 | * in-flight packets from before XDP was enabled reach | |
589 | * the receive path after XDP is loaded. In practice I | |
590 | * was not able to create this condition. | |
591 | */ | |
b00f70b0 | 592 | if (unlikely(hdr->hdr.gso_type)) |
f600b690 JF |
593 | goto err_xdp; |
594 | ||
2de2f7f4 JF |
595 | /* Allow consuming headroom but reserve enough space to push |
596 | * the descriptor on if we get an XDP_TX return code. | |
597 | */ | |
0354e4d1 | 598 | data = page_address(xdp_page) + offset; |
2de2f7f4 | 599 | xdp.data_hard_start = data - VIRTIO_XDP_HEADROOM + vi->hdr_len; |
0354e4d1 JF |
600 | xdp.data = data + vi->hdr_len; |
601 | xdp.data_end = xdp.data + (len - vi->hdr_len); | |
602 | act = bpf_prog_run_xdp(xdp_prog, &xdp); | |
603 | ||
56434a01 JF |
604 | switch (act) { |
605 | case XDP_PASS: | |
2de2f7f4 JF |
606 | /* recalculate offset to account for any header |
607 | * adjustments. Note other cases do not build an | |
608 | * skb and avoid using offset | |
609 | */ | |
610 | offset = xdp.data - | |
611 | page_address(xdp_page) - vi->hdr_len; | |
612 | ||
1830f893 JW |
613 | /* We can only create skb based on xdp_page. */ |
614 | if (unlikely(xdp_page != page)) { | |
615 | rcu_read_unlock(); | |
616 | put_page(page); | |
617 | head_skb = page_to_skb(vi, rq, xdp_page, | |
2de2f7f4 | 618 | offset, len, PAGE_SIZE); |
5c33474d | 619 | ewma_pkt_len_add(&rq->mrg_avg_pkt_len, len); |
1830f893 JW |
620 | return head_skb; |
621 | } | |
56434a01 JF |
622 | break; |
623 | case XDP_TX: | |
722d8283 | 624 | if (unlikely(!virtnet_xdp_xmit(vi, rq, &xdp, data))) |
0354e4d1 | 625 | trace_xdp_exception(vi->dev, xdp_prog, act); |
5c33474d | 626 | ewma_pkt_len_add(&rq->mrg_avg_pkt_len, len); |
72979a6c JF |
627 | if (unlikely(xdp_page != page)) |
628 | goto err_xdp; | |
56434a01 JF |
629 | rcu_read_unlock(); |
630 | goto xdp_xmit; | |
56434a01 | 631 | default: |
0354e4d1 JF |
632 | bpf_warn_invalid_xdp_action(act); |
633 | case XDP_ABORTED: | |
634 | trace_xdp_exception(vi->dev, xdp_prog, act); | |
635 | case XDP_DROP: | |
72979a6c JF |
636 | if (unlikely(xdp_page != page)) |
637 | __free_pages(xdp_page, 0); | |
5c33474d | 638 | ewma_pkt_len_add(&rq->mrg_avg_pkt_len, len); |
f600b690 | 639 | goto err_xdp; |
56434a01 | 640 | } |
f600b690 JF |
641 | } |
642 | rcu_read_unlock(); | |
ab7db917 | 643 | |
f600b690 JF |
644 | truesize = max(len, mergeable_ctx_to_buf_truesize(ctx)); |
645 | head_skb = page_to_skb(vi, rq, page, offset, len, truesize); | |
646 | curr_skb = head_skb; | |
9ab86bbc | 647 | |
8fc3b9e9 MT |
648 | if (unlikely(!curr_skb)) |
649 | goto err_skb; | |
9ab86bbc | 650 | while (--num_buf) { |
8fc3b9e9 MT |
651 | int num_skb_frags; |
652 | ||
ab7db917 MD |
653 | ctx = (unsigned long)virtqueue_get_buf(rq->vq, &len); |
654 | if (unlikely(!ctx)) { | |
8fc3b9e9 | 655 | pr_debug("%s: rx error: %d buffers out of %d missing\n", |
fdd819b2 | 656 | dev->name, num_buf, |
012873d0 MT |
657 | virtio16_to_cpu(vi->vdev, |
658 | hdr->num_buffers)); | |
8fc3b9e9 MT |
659 | dev->stats.rx_length_errors++; |
660 | goto err_buf; | |
3f2c31d9 | 661 | } |
8fc3b9e9 | 662 | |
ab7db917 | 663 | buf = mergeable_ctx_to_buf_address(ctx); |
8fc3b9e9 | 664 | page = virt_to_head_page(buf); |
8fc3b9e9 MT |
665 | |
666 | num_skb_frags = skb_shinfo(curr_skb)->nr_frags; | |
2613af0e MD |
667 | if (unlikely(num_skb_frags == MAX_SKB_FRAGS)) { |
668 | struct sk_buff *nskb = alloc_skb(0, GFP_ATOMIC); | |
8fc3b9e9 MT |
669 | |
670 | if (unlikely(!nskb)) | |
671 | goto err_skb; | |
2613af0e MD |
672 | if (curr_skb == head_skb) |
673 | skb_shinfo(curr_skb)->frag_list = nskb; | |
674 | else | |
675 | curr_skb->next = nskb; | |
676 | curr_skb = nskb; | |
677 | head_skb->truesize += nskb->truesize; | |
678 | num_skb_frags = 0; | |
679 | } | |
ab7db917 | 680 | truesize = max(len, mergeable_ctx_to_buf_truesize(ctx)); |
2613af0e MD |
681 | if (curr_skb != head_skb) { |
682 | head_skb->data_len += len; | |
683 | head_skb->len += len; | |
fb51879d | 684 | head_skb->truesize += truesize; |
2613af0e | 685 | } |
8fc3b9e9 | 686 | offset = buf - page_address(page); |
ba275241 JW |
687 | if (skb_can_coalesce(curr_skb, num_skb_frags, page, offset)) { |
688 | put_page(page); | |
689 | skb_coalesce_rx_frag(curr_skb, num_skb_frags - 1, | |
fb51879d | 690 | len, truesize); |
ba275241 JW |
691 | } else { |
692 | skb_add_rx_frag(curr_skb, num_skb_frags, page, | |
fb51879d | 693 | offset, len, truesize); |
ba275241 | 694 | } |
8fc3b9e9 MT |
695 | } |
696 | ||
5377d758 | 697 | ewma_pkt_len_add(&rq->mrg_avg_pkt_len, head_skb->len); |
8fc3b9e9 MT |
698 | return head_skb; |
699 | ||
f600b690 JF |
700 | err_xdp: |
701 | rcu_read_unlock(); | |
8fc3b9e9 MT |
702 | err_skb: |
703 | put_page(page); | |
704 | while (--num_buf) { | |
ab7db917 MD |
705 | ctx = (unsigned long)virtqueue_get_buf(rq->vq, &len); |
706 | if (unlikely(!ctx)) { | |
8fc3b9e9 MT |
707 | pr_debug("%s: rx error: %d buffers missing\n", |
708 | dev->name, num_buf); | |
709 | dev->stats.rx_length_errors++; | |
710 | break; | |
711 | } | |
ab7db917 | 712 | page = virt_to_head_page(mergeable_ctx_to_buf_address(ctx)); |
8fc3b9e9 | 713 | put_page(page); |
9ab86bbc | 714 | } |
8fc3b9e9 MT |
715 | err_buf: |
716 | dev->stats.rx_dropped++; | |
717 | dev_kfree_skb(head_skb); | |
56434a01 | 718 | xdp_xmit: |
8fc3b9e9 | 719 | return NULL; |
9ab86bbc SM |
720 | } |
721 | ||
946fa564 MT |
722 | static void receive_buf(struct virtnet_info *vi, struct receive_queue *rq, |
723 | void *buf, unsigned int len) | |
9ab86bbc | 724 | { |
e9d7417b | 725 | struct net_device *dev = vi->dev; |
58472a76 | 726 | struct virtnet_stats *stats = this_cpu_ptr(vi->stats); |
9ab86bbc | 727 | struct sk_buff *skb; |
012873d0 | 728 | struct virtio_net_hdr_mrg_rxbuf *hdr; |
3f2c31d9 | 729 | |
bcff3162 | 730 | if (unlikely(len < vi->hdr_len + ETH_HLEN)) { |
9ab86bbc SM |
731 | pr_debug("%s: short packet %i\n", dev->name, len); |
732 | dev->stats.rx_length_errors++; | |
ab7db917 MD |
733 | if (vi->mergeable_rx_bufs) { |
734 | unsigned long ctx = (unsigned long)buf; | |
735 | void *base = mergeable_ctx_to_buf_address(ctx); | |
736 | put_page(virt_to_head_page(base)); | |
737 | } else if (vi->big_packets) { | |
98bfd23c | 738 | give_pages(rq, buf); |
ab7db917 | 739 | } else { |
9ab86bbc | 740 | dev_kfree_skb(buf); |
ab7db917 | 741 | } |
9ab86bbc SM |
742 | return; |
743 | } | |
3f2c31d9 | 744 | |
f121159d | 745 | if (vi->mergeable_rx_bufs) |
fdd819b2 | 746 | skb = receive_mergeable(dev, vi, rq, (unsigned long)buf, len); |
f121159d | 747 | else if (vi->big_packets) |
946fa564 | 748 | skb = receive_big(dev, vi, rq, buf, len); |
f121159d | 749 | else |
bb91accf | 750 | skb = receive_small(dev, vi, rq, buf, len); |
f121159d MT |
751 | |
752 | if (unlikely(!skb)) | |
753 | return; | |
3f2c31d9 | 754 | |
9ab86bbc | 755 | hdr = skb_vnet_hdr(skb); |
3fa2a1df | 756 | |
83a27052 | 757 | u64_stats_update_begin(&stats->rx_syncp); |
3fa2a1df | 758 | stats->rx_bytes += skb->len; |
759 | stats->rx_packets++; | |
83a27052 | 760 | u64_stats_update_end(&stats->rx_syncp); |
296f96fc | 761 | |
e858fae2 | 762 | if (hdr->hdr.flags & VIRTIO_NET_HDR_F_DATA_VALID) |
10a8d94a | 763 | skb->ip_summed = CHECKSUM_UNNECESSARY; |
296f96fc | 764 | |
e858fae2 MR |
765 | if (virtio_net_hdr_to_skb(skb, &hdr->hdr, |
766 | virtio_is_little_endian(vi->vdev))) { | |
767 | net_warn_ratelimited("%s: bad gso: type: %u, size: %u\n", | |
768 | dev->name, hdr->hdr.gso_type, | |
769 | hdr->hdr.gso_size); | |
770 | goto frame_err; | |
296f96fc RR |
771 | } |
772 | ||
d1dc06dc MR |
773 | skb->protocol = eth_type_trans(skb, dev); |
774 | pr_debug("Receiving skb proto 0x%04x len %i type %i\n", | |
775 | ntohs(skb->protocol), skb->len, skb->pkt_type); | |
776 | ||
0fbd050a | 777 | napi_gro_receive(&rq->napi, skb); |
296f96fc RR |
778 | return; |
779 | ||
780 | frame_err: | |
781 | dev->stats.rx_frame_errors++; | |
296f96fc RR |
782 | dev_kfree_skb(skb); |
783 | } | |
784 | ||
2de2f7f4 JF |
785 | static unsigned int virtnet_get_headroom(struct virtnet_info *vi) |
786 | { | |
787 | return vi->xdp_queue_pairs ? VIRTIO_XDP_HEADROOM : 0; | |
788 | } | |
789 | ||
946fa564 MT |
790 | static int add_recvbuf_small(struct virtnet_info *vi, struct receive_queue *rq, |
791 | gfp_t gfp) | |
296f96fc | 792 | { |
2de2f7f4 JF |
793 | int headroom = GOOD_PACKET_LEN + virtnet_get_headroom(vi); |
794 | unsigned int xdp_headroom = virtnet_get_headroom(vi); | |
296f96fc | 795 | struct sk_buff *skb; |
012873d0 | 796 | struct virtio_net_hdr_mrg_rxbuf *hdr; |
9ab86bbc | 797 | int err; |
3f2c31d9 | 798 | |
2de2f7f4 | 799 | skb = __netdev_alloc_skb_ip_align(vi->dev, headroom, gfp); |
9ab86bbc SM |
800 | if (unlikely(!skb)) |
801 | return -ENOMEM; | |
296f96fc | 802 | |
2de2f7f4 | 803 | skb_put(skb, headroom); |
3f2c31d9 | 804 | |
9ab86bbc | 805 | hdr = skb_vnet_hdr(skb); |
547c890c | 806 | sg_init_table(rq->sg, 2); |
012873d0 | 807 | sg_set_buf(rq->sg, hdr, vi->hdr_len); |
2de2f7f4 | 808 | skb_to_sgvec(skb, rq->sg + 1, xdp_headroom, skb->len - xdp_headroom); |
97402b96 | 809 | |
9dc7b9e4 | 810 | err = virtqueue_add_inbuf(rq->vq, rq->sg, 2, skb, gfp); |
9ab86bbc SM |
811 | if (err < 0) |
812 | dev_kfree_skb(skb); | |
97402b96 | 813 | |
9ab86bbc SM |
814 | return err; |
815 | } | |
97402b96 | 816 | |
012873d0 MT |
817 | static int add_recvbuf_big(struct virtnet_info *vi, struct receive_queue *rq, |
818 | gfp_t gfp) | |
9ab86bbc | 819 | { |
9ab86bbc SM |
820 | struct page *first, *list = NULL; |
821 | char *p; | |
822 | int i, err, offset; | |
823 | ||
a5835440 RR |
824 | sg_init_table(rq->sg, MAX_SKB_FRAGS + 2); |
825 | ||
e9d7417b | 826 | /* page in rq->sg[MAX_SKB_FRAGS + 1] is list tail */ |
9ab86bbc | 827 | for (i = MAX_SKB_FRAGS + 1; i > 1; --i) { |
e9d7417b | 828 | first = get_a_page(rq, gfp); |
9ab86bbc SM |
829 | if (!first) { |
830 | if (list) | |
e9d7417b | 831 | give_pages(rq, list); |
9ab86bbc | 832 | return -ENOMEM; |
97402b96 | 833 | } |
e9d7417b | 834 | sg_set_buf(&rq->sg[i], page_address(first), PAGE_SIZE); |
97402b96 | 835 | |
9ab86bbc SM |
836 | /* chain new page in list head to match sg */ |
837 | first->private = (unsigned long)list; | |
838 | list = first; | |
839 | } | |
296f96fc | 840 | |
e9d7417b | 841 | first = get_a_page(rq, gfp); |
9ab86bbc | 842 | if (!first) { |
e9d7417b | 843 | give_pages(rq, list); |
9ab86bbc SM |
844 | return -ENOMEM; |
845 | } | |
846 | p = page_address(first); | |
847 | ||
e9d7417b | 848 | /* rq->sg[0], rq->sg[1] share the same page */ |
012873d0 MT |
849 | /* a separated rq->sg[0] for header - required in case !any_header_sg */ |
850 | sg_set_buf(&rq->sg[0], p, vi->hdr_len); | |
9ab86bbc | 851 | |
e9d7417b | 852 | /* rq->sg[1] for data packet, from offset */ |
9ab86bbc | 853 | offset = sizeof(struct padded_vnet_hdr); |
e9d7417b | 854 | sg_set_buf(&rq->sg[1], p + offset, PAGE_SIZE - offset); |
9ab86bbc SM |
855 | |
856 | /* chain first in list head */ | |
857 | first->private = (unsigned long)list; | |
9dc7b9e4 RR |
858 | err = virtqueue_add_inbuf(rq->vq, rq->sg, MAX_SKB_FRAGS + 2, |
859 | first, gfp); | |
9ab86bbc | 860 | if (err < 0) |
e9d7417b | 861 | give_pages(rq, first); |
9ab86bbc SM |
862 | |
863 | return err; | |
296f96fc RR |
864 | } |
865 | ||
5377d758 | 866 | static unsigned int get_mergeable_buf_len(struct ewma_pkt_len *avg_pkt_len) |
3f2c31d9 | 867 | { |
ab7db917 | 868 | const size_t hdr_len = sizeof(struct virtio_net_hdr_mrg_rxbuf); |
fbf28d78 MD |
869 | unsigned int len; |
870 | ||
5377d758 | 871 | len = hdr_len + clamp_t(unsigned int, ewma_pkt_len_read(avg_pkt_len), |
fbf28d78 MD |
872 | GOOD_PACKET_LEN, PAGE_SIZE - hdr_len); |
873 | return ALIGN(len, MERGEABLE_BUFFER_ALIGN); | |
874 | } | |
875 | ||
2de2f7f4 JF |
876 | static int add_recvbuf_mergeable(struct virtnet_info *vi, |
877 | struct receive_queue *rq, gfp_t gfp) | |
fbf28d78 | 878 | { |
fb51879d | 879 | struct page_frag *alloc_frag = &rq->alloc_frag; |
2de2f7f4 | 880 | unsigned int headroom = virtnet_get_headroom(vi); |
fb51879d | 881 | char *buf; |
ab7db917 | 882 | unsigned long ctx; |
3f2c31d9 | 883 | int err; |
fb51879d | 884 | unsigned int len, hole; |
3f2c31d9 | 885 | |
fbf28d78 | 886 | len = get_mergeable_buf_len(&rq->mrg_avg_pkt_len); |
2de2f7f4 | 887 | if (unlikely(!skb_page_frag_refill(len + headroom, alloc_frag, gfp))) |
9ab86bbc | 888 | return -ENOMEM; |
ab7db917 | 889 | |
fb51879d | 890 | buf = (char *)page_address(alloc_frag->page) + alloc_frag->offset; |
2de2f7f4 | 891 | buf += headroom; /* advance address leaving hole at front of pkt */ |
ab7db917 | 892 | ctx = mergeable_buf_to_ctx(buf, len); |
fb51879d | 893 | get_page(alloc_frag->page); |
2de2f7f4 | 894 | alloc_frag->offset += len + headroom; |
fb51879d | 895 | hole = alloc_frag->size - alloc_frag->offset; |
2de2f7f4 | 896 | if (hole < len + headroom) { |
ab7db917 MD |
897 | /* To avoid internal fragmentation, if there is very likely not |
898 | * enough space for another buffer, add the remaining space to | |
899 | * the current buffer. This extra space is not included in | |
900 | * the truesize stored in ctx. | |
901 | */ | |
fb51879d MD |
902 | len += hole; |
903 | alloc_frag->offset += hole; | |
904 | } | |
3f2c31d9 | 905 | |
fb51879d | 906 | sg_init_one(rq->sg, buf, len); |
ab7db917 | 907 | err = virtqueue_add_inbuf(rq->vq, rq->sg, 1, (void *)ctx, gfp); |
9ab86bbc | 908 | if (err < 0) |
2613af0e | 909 | put_page(virt_to_head_page(buf)); |
3f2c31d9 | 910 | |
9ab86bbc SM |
911 | return err; |
912 | } | |
3f2c31d9 | 913 | |
b2baed69 RR |
914 | /* |
915 | * Returns false if we couldn't fill entirely (OOM). | |
916 | * | |
917 | * Normally run in the receive path, but can also be run from ndo_open | |
918 | * before we're receiving packets, or from refill_work which is | |
919 | * careful to disable receiving (using napi_disable). | |
920 | */ | |
946fa564 MT |
921 | static bool try_fill_recv(struct virtnet_info *vi, struct receive_queue *rq, |
922 | gfp_t gfp) | |
9ab86bbc SM |
923 | { |
924 | int err; | |
1788f495 | 925 | bool oom; |
3f2c31d9 | 926 | |
fb51879d | 927 | gfp |= __GFP_COLD; |
9ab86bbc SM |
928 | do { |
929 | if (vi->mergeable_rx_bufs) | |
2de2f7f4 | 930 | err = add_recvbuf_mergeable(vi, rq, gfp); |
9ab86bbc | 931 | else if (vi->big_packets) |
012873d0 | 932 | err = add_recvbuf_big(vi, rq, gfp); |
9ab86bbc | 933 | else |
946fa564 | 934 | err = add_recvbuf_small(vi, rq, gfp); |
3f2c31d9 | 935 | |
1788f495 | 936 | oom = err == -ENOMEM; |
9ed4cb07 | 937 | if (err) |
3f2c31d9 | 938 | break; |
b7dfde95 | 939 | } while (rq->vq->num_free); |
681daee2 | 940 | virtqueue_kick(rq->vq); |
3161e453 | 941 | return !oom; |
3f2c31d9 MM |
942 | } |
943 | ||
18445c4d | 944 | static void skb_recv_done(struct virtqueue *rvq) |
296f96fc RR |
945 | { |
946 | struct virtnet_info *vi = rvq->vdev->priv; | |
986a4f4d | 947 | struct receive_queue *rq = &vi->rq[vq2rxq(rvq)]; |
e9d7417b | 948 | |
18445c4d | 949 | /* Schedule NAPI, Suppress further interrupts if successful. */ |
e9d7417b | 950 | if (napi_schedule_prep(&rq->napi)) { |
1915a712 | 951 | virtqueue_disable_cb(rvq); |
e9d7417b | 952 | __napi_schedule(&rq->napi); |
18445c4d | 953 | } |
296f96fc RR |
954 | } |
955 | ||
e9d7417b | 956 | static void virtnet_napi_enable(struct receive_queue *rq) |
3e9d08ec | 957 | { |
e9d7417b | 958 | napi_enable(&rq->napi); |
3e9d08ec BR |
959 | |
960 | /* If all buffers were filled by other side before we napi_enabled, we | |
961 | * won't get another interrupt, so process any outstanding packets | |
962 | * now. virtnet_poll wants re-enable the queue, so we disable here. | |
963 | * We synchronize against interrupts via NAPI_STATE_SCHED */ | |
e9d7417b JW |
964 | if (napi_schedule_prep(&rq->napi)) { |
965 | virtqueue_disable_cb(rq->vq); | |
ec13ee80 | 966 | local_bh_disable(); |
e9d7417b | 967 | __napi_schedule(&rq->napi); |
ec13ee80 | 968 | local_bh_enable(); |
3e9d08ec BR |
969 | } |
970 | } | |
971 | ||
3161e453 RR |
972 | static void refill_work(struct work_struct *work) |
973 | { | |
e9d7417b JW |
974 | struct virtnet_info *vi = |
975 | container_of(work, struct virtnet_info, refill.work); | |
3161e453 | 976 | bool still_empty; |
986a4f4d JW |
977 | int i; |
978 | ||
55257d72 | 979 | for (i = 0; i < vi->curr_queue_pairs; i++) { |
986a4f4d | 980 | struct receive_queue *rq = &vi->rq[i]; |
3161e453 | 981 | |
986a4f4d | 982 | napi_disable(&rq->napi); |
946fa564 | 983 | still_empty = !try_fill_recv(vi, rq, GFP_KERNEL); |
986a4f4d | 984 | virtnet_napi_enable(rq); |
3161e453 | 985 | |
986a4f4d JW |
986 | /* In theory, this can happen: if we don't get any buffers in |
987 | * we will *never* try to fill again. | |
988 | */ | |
989 | if (still_empty) | |
990 | schedule_delayed_work(&vi->refill, HZ/2); | |
991 | } | |
3161e453 RR |
992 | } |
993 | ||
2ffa7598 | 994 | static int virtnet_receive(struct receive_queue *rq, int budget) |
296f96fc | 995 | { |
e9d7417b | 996 | struct virtnet_info *vi = rq->vq->vdev->priv; |
2ffa7598 | 997 | unsigned int len, received = 0; |
9ab86bbc | 998 | void *buf; |
296f96fc | 999 | |
296f96fc | 1000 | while (received < budget && |
e9d7417b | 1001 | (buf = virtqueue_get_buf(rq->vq, &len)) != NULL) { |
946fa564 | 1002 | receive_buf(vi, rq, buf, len); |
296f96fc RR |
1003 | received++; |
1004 | } | |
1005 | ||
be121f46 | 1006 | if (rq->vq->num_free > virtqueue_get_vring_size(rq->vq) / 2) { |
946fa564 | 1007 | if (!try_fill_recv(vi, rq, GFP_ATOMIC)) |
3b07e9ca | 1008 | schedule_delayed_work(&vi->refill, 0); |
3161e453 | 1009 | } |
296f96fc | 1010 | |
2ffa7598 JW |
1011 | return received; |
1012 | } | |
1013 | ||
1014 | static int virtnet_poll(struct napi_struct *napi, int budget) | |
1015 | { | |
1016 | struct receive_queue *rq = | |
1017 | container_of(napi, struct receive_queue, napi); | |
faadb05f | 1018 | unsigned int r, received; |
2ffa7598 | 1019 | |
faadb05f | 1020 | received = virtnet_receive(rq, budget); |
2ffa7598 | 1021 | |
8329d98e RR |
1022 | /* Out of packets? */ |
1023 | if (received < budget) { | |
cbdadbbf | 1024 | r = virtqueue_enable_cb_prepare(rq->vq); |
4d6308aa ED |
1025 | if (napi_complete_done(napi, received)) { |
1026 | if (unlikely(virtqueue_poll(rq->vq, r)) && | |
1027 | napi_schedule_prep(napi)) { | |
1028 | virtqueue_disable_cb(rq->vq); | |
1029 | __napi_schedule(napi); | |
1030 | } | |
4265f161 | 1031 | } |
296f96fc RR |
1032 | } |
1033 | ||
1034 | return received; | |
1035 | } | |
1036 | ||
986a4f4d JW |
1037 | static int virtnet_open(struct net_device *dev) |
1038 | { | |
1039 | struct virtnet_info *vi = netdev_priv(dev); | |
1040 | int i; | |
1041 | ||
e4166625 JW |
1042 | for (i = 0; i < vi->max_queue_pairs; i++) { |
1043 | if (i < vi->curr_queue_pairs) | |
1044 | /* Make sure we have some buffers: if oom use wq. */ | |
946fa564 | 1045 | if (!try_fill_recv(vi, &vi->rq[i], GFP_KERNEL)) |
e4166625 | 1046 | schedule_delayed_work(&vi->refill, 0); |
986a4f4d JW |
1047 | virtnet_napi_enable(&vi->rq[i]); |
1048 | } | |
1049 | ||
1050 | return 0; | |
1051 | } | |
1052 | ||
b7dfde95 | 1053 | static void free_old_xmit_skbs(struct send_queue *sq) |
296f96fc RR |
1054 | { |
1055 | struct sk_buff *skb; | |
6ee57bcc | 1056 | unsigned int len; |
e9d7417b | 1057 | struct virtnet_info *vi = sq->vq->vdev->priv; |
58472a76 | 1058 | struct virtnet_stats *stats = this_cpu_ptr(vi->stats); |
296f96fc | 1059 | |
e9d7417b | 1060 | while ((skb = virtqueue_get_buf(sq->vq, &len)) != NULL) { |
296f96fc | 1061 | pr_debug("Sent skb %p\n", skb); |
3fa2a1df | 1062 | |
83a27052 | 1063 | u64_stats_update_begin(&stats->tx_syncp); |
3fa2a1df | 1064 | stats->tx_bytes += skb->len; |
1065 | stats->tx_packets++; | |
83a27052 | 1066 | u64_stats_update_end(&stats->tx_syncp); |
3fa2a1df | 1067 | |
ed79bab8 | 1068 | dev_kfree_skb_any(skb); |
296f96fc RR |
1069 | } |
1070 | } | |
1071 | ||
e9d7417b | 1072 | static int xmit_skb(struct send_queue *sq, struct sk_buff *skb) |
296f96fc | 1073 | { |
012873d0 | 1074 | struct virtio_net_hdr_mrg_rxbuf *hdr; |
296f96fc | 1075 | const unsigned char *dest = ((struct ethhdr *)skb->data)->h_dest; |
e9d7417b | 1076 | struct virtnet_info *vi = sq->vq->vdev->priv; |
7bedc7dc | 1077 | unsigned num_sg; |
012873d0 | 1078 | unsigned hdr_len = vi->hdr_len; |
e7428e95 | 1079 | bool can_push; |
296f96fc | 1080 | |
e174961c | 1081 | pr_debug("%s: xmit %p %pM\n", vi->dev->name, skb, dest); |
e7428e95 MT |
1082 | |
1083 | can_push = vi->any_header_sg && | |
1084 | !((unsigned long)skb->data & (__alignof__(*hdr) - 1)) && | |
1085 | !skb_header_cloned(skb) && skb_headroom(skb) >= hdr_len; | |
1086 | /* Even if we can, don't push here yet as this would skew | |
1087 | * csum_start offset below. */ | |
1088 | if (can_push) | |
012873d0 | 1089 | hdr = (struct virtio_net_hdr_mrg_rxbuf *)(skb->data - hdr_len); |
e7428e95 MT |
1090 | else |
1091 | hdr = skb_vnet_hdr(skb); | |
296f96fc | 1092 | |
e858fae2 | 1093 | if (virtio_net_hdr_from_skb(skb, &hdr->hdr, |
6391a448 | 1094 | virtio_is_little_endian(vi->vdev), false)) |
e858fae2 | 1095 | BUG(); |
296f96fc | 1096 | |
3f2c31d9 | 1097 | if (vi->mergeable_rx_bufs) |
012873d0 | 1098 | hdr->num_buffers = 0; |
3f2c31d9 | 1099 | |
547c890c | 1100 | sg_init_table(sq->sg, skb_shinfo(skb)->nr_frags + (can_push ? 1 : 2)); |
e7428e95 MT |
1101 | if (can_push) { |
1102 | __skb_push(skb, hdr_len); | |
1103 | num_sg = skb_to_sgvec(skb, sq->sg, 0, skb->len); | |
1104 | /* Pull header back to avoid skew in tx bytes calculations. */ | |
1105 | __skb_pull(skb, hdr_len); | |
1106 | } else { | |
1107 | sg_set_buf(sq->sg, hdr, hdr_len); | |
1108 | num_sg = skb_to_sgvec(skb, sq->sg + 1, 0, skb->len) + 1; | |
1109 | } | |
9dc7b9e4 | 1110 | return virtqueue_add_outbuf(sq->vq, sq->sg, num_sg, skb, GFP_ATOMIC); |
11a3a154 RR |
1111 | } |
1112 | ||
424efe9c | 1113 | static netdev_tx_t start_xmit(struct sk_buff *skb, struct net_device *dev) |
99ffc696 RR |
1114 | { |
1115 | struct virtnet_info *vi = netdev_priv(dev); | |
986a4f4d JW |
1116 | int qnum = skb_get_queue_mapping(skb); |
1117 | struct send_queue *sq = &vi->sq[qnum]; | |
9ed4cb07 | 1118 | int err; |
4b7fd2e6 MT |
1119 | struct netdev_queue *txq = netdev_get_tx_queue(dev, qnum); |
1120 | bool kick = !skb->xmit_more; | |
2cb9c6ba | 1121 | |
2cb9c6ba | 1122 | /* Free up any pending old buffers before queueing new ones. */ |
e9d7417b | 1123 | free_old_xmit_skbs(sq); |
99ffc696 | 1124 | |
074c3582 JK |
1125 | /* timestamp packet in software */ |
1126 | skb_tx_timestamp(skb); | |
1127 | ||
03f191ba | 1128 | /* Try to transmit */ |
b7dfde95 | 1129 | err = xmit_skb(sq, skb); |
48925e37 | 1130 | |
9ed4cb07 | 1131 | /* This should not happen! */ |
681daee2 | 1132 | if (unlikely(err)) { |
9ed4cb07 RR |
1133 | dev->stats.tx_fifo_errors++; |
1134 | if (net_ratelimit()) | |
1135 | dev_warn(&dev->dev, | |
b7dfde95 | 1136 | "Unexpected TXQ (%d) queue failure: %d\n", qnum, err); |
58eba97d | 1137 | dev->stats.tx_dropped++; |
85e94525 | 1138 | dev_kfree_skb_any(skb); |
58eba97d | 1139 | return NETDEV_TX_OK; |
296f96fc | 1140 | } |
03f191ba | 1141 | |
48925e37 RR |
1142 | /* Don't wait up for transmitted skbs to be freed. */ |
1143 | skb_orphan(skb); | |
1144 | nf_reset(skb); | |
1145 | ||
60302ff6 MT |
1146 | /* If running out of space, stop queue to avoid getting packets that we |
1147 | * are then unable to transmit. | |
1148 | * An alternative would be to force queuing layer to requeue the skb by | |
1149 | * returning NETDEV_TX_BUSY. However, NETDEV_TX_BUSY should not be | |
1150 | * returned in a normal path of operation: it means that driver is not | |
1151 | * maintaining the TX queue stop/start state properly, and causes | |
1152 | * the stack to do a non-trivial amount of useless work. | |
1153 | * Since most packets only take 1 or 2 ring slots, stopping the queue | |
1154 | * early means 16 slots are typically wasted. | |
d631b94e | 1155 | */ |
b7dfde95 | 1156 | if (sq->vq->num_free < 2+MAX_SKB_FRAGS) { |
986a4f4d | 1157 | netif_stop_subqueue(dev, qnum); |
e9d7417b | 1158 | if (unlikely(!virtqueue_enable_cb_delayed(sq->vq))) { |
48925e37 | 1159 | /* More just got used, free them then recheck. */ |
b7dfde95 LT |
1160 | free_old_xmit_skbs(sq); |
1161 | if (sq->vq->num_free >= 2+MAX_SKB_FRAGS) { | |
986a4f4d | 1162 | netif_start_subqueue(dev, qnum); |
e9d7417b | 1163 | virtqueue_disable_cb(sq->vq); |
48925e37 RR |
1164 | } |
1165 | } | |
99ffc696 | 1166 | } |
48925e37 | 1167 | |
4b7fd2e6 | 1168 | if (kick || netif_xmit_stopped(txq)) |
0b725a2c | 1169 | virtqueue_kick(sq->vq); |
296f96fc | 1170 | |
0b725a2c | 1171 | return NETDEV_TX_OK; |
c223a078 DM |
1172 | } |
1173 | ||
40cbfc37 AK |
1174 | /* |
1175 | * Send command via the control virtqueue and check status. Commands | |
1176 | * supported by the hypervisor, as indicated by feature bits, should | |
788a8b6d | 1177 | * never fail unless improperly formatted. |
40cbfc37 AK |
1178 | */ |
1179 | static bool virtnet_send_command(struct virtnet_info *vi, u8 class, u8 cmd, | |
d24bae32 | 1180 | struct scatterlist *out) |
40cbfc37 | 1181 | { |
f7bc9594 | 1182 | struct scatterlist *sgs[4], hdr, stat; |
d24bae32 | 1183 | unsigned out_num = 0, tmp; |
40cbfc37 AK |
1184 | |
1185 | /* Caller should know better */ | |
f7bc9594 | 1186 | BUG_ON(!virtio_has_feature(vi->vdev, VIRTIO_NET_F_CTRL_VQ)); |
40cbfc37 | 1187 | |
2ac46030 MT |
1188 | vi->ctrl_status = ~0; |
1189 | vi->ctrl_hdr.class = class; | |
1190 | vi->ctrl_hdr.cmd = cmd; | |
f7bc9594 | 1191 | /* Add header */ |
2ac46030 | 1192 | sg_init_one(&hdr, &vi->ctrl_hdr, sizeof(vi->ctrl_hdr)); |
f7bc9594 | 1193 | sgs[out_num++] = &hdr; |
40cbfc37 | 1194 | |
f7bc9594 RR |
1195 | if (out) |
1196 | sgs[out_num++] = out; | |
40cbfc37 | 1197 | |
f7bc9594 | 1198 | /* Add return status. */ |
2ac46030 | 1199 | sg_init_one(&stat, &vi->ctrl_status, sizeof(vi->ctrl_status)); |
d24bae32 | 1200 | sgs[out_num] = &stat; |
40cbfc37 | 1201 | |
d24bae32 | 1202 | BUG_ON(out_num + 1 > ARRAY_SIZE(sgs)); |
a7c58146 | 1203 | virtqueue_add_sgs(vi->cvq, sgs, out_num, 1, vi, GFP_ATOMIC); |
40cbfc37 | 1204 | |
67975901 | 1205 | if (unlikely(!virtqueue_kick(vi->cvq))) |
2ac46030 | 1206 | return vi->ctrl_status == VIRTIO_NET_OK; |
40cbfc37 AK |
1207 | |
1208 | /* Spin for a response, the kick causes an ioport write, trapping | |
1209 | * into the hypervisor, so the request should be handled immediately. | |
1210 | */ | |
047b9b94 HG |
1211 | while (!virtqueue_get_buf(vi->cvq, &tmp) && |
1212 | !virtqueue_is_broken(vi->cvq)) | |
40cbfc37 AK |
1213 | cpu_relax(); |
1214 | ||
2ac46030 | 1215 | return vi->ctrl_status == VIRTIO_NET_OK; |
40cbfc37 AK |
1216 | } |
1217 | ||
9c46f6d4 AW |
1218 | static int virtnet_set_mac_address(struct net_device *dev, void *p) |
1219 | { | |
1220 | struct virtnet_info *vi = netdev_priv(dev); | |
1221 | struct virtio_device *vdev = vi->vdev; | |
f2f2c8b4 | 1222 | int ret; |
e37e2ff3 | 1223 | struct sockaddr *addr; |
7e58d5ae | 1224 | struct scatterlist sg; |
9c46f6d4 | 1225 | |
801822d1 | 1226 | addr = kmemdup(p, sizeof(*addr), GFP_KERNEL); |
e37e2ff3 AL |
1227 | if (!addr) |
1228 | return -ENOMEM; | |
e37e2ff3 AL |
1229 | |
1230 | ret = eth_prepare_mac_addr_change(dev, addr); | |
f2f2c8b4 | 1231 | if (ret) |
e37e2ff3 | 1232 | goto out; |
9c46f6d4 | 1233 | |
7e58d5ae AK |
1234 | if (virtio_has_feature(vdev, VIRTIO_NET_F_CTRL_MAC_ADDR)) { |
1235 | sg_init_one(&sg, addr->sa_data, dev->addr_len); | |
1236 | if (!virtnet_send_command(vi, VIRTIO_NET_CTRL_MAC, | |
d24bae32 | 1237 | VIRTIO_NET_CTRL_MAC_ADDR_SET, &sg)) { |
7e58d5ae AK |
1238 | dev_warn(&vdev->dev, |
1239 | "Failed to set mac address by vq command.\n"); | |
e37e2ff3 AL |
1240 | ret = -EINVAL; |
1241 | goto out; | |
7e58d5ae | 1242 | } |
7e93a02f MT |
1243 | } else if (virtio_has_feature(vdev, VIRTIO_NET_F_MAC) && |
1244 | !virtio_has_feature(vdev, VIRTIO_F_VERSION_1)) { | |
855e0c52 RR |
1245 | unsigned int i; |
1246 | ||
1247 | /* Naturally, this has an atomicity problem. */ | |
1248 | for (i = 0; i < dev->addr_len; i++) | |
1249 | virtio_cwrite8(vdev, | |
1250 | offsetof(struct virtio_net_config, mac) + | |
1251 | i, addr->sa_data[i]); | |
7e58d5ae AK |
1252 | } |
1253 | ||
1254 | eth_commit_mac_addr_change(dev, p); | |
e37e2ff3 | 1255 | ret = 0; |
9c46f6d4 | 1256 | |
e37e2ff3 AL |
1257 | out: |
1258 | kfree(addr); | |
1259 | return ret; | |
9c46f6d4 AW |
1260 | } |
1261 | ||
bc1f4470 | 1262 | static void virtnet_stats(struct net_device *dev, |
1263 | struct rtnl_link_stats64 *tot) | |
3fa2a1df | 1264 | { |
1265 | struct virtnet_info *vi = netdev_priv(dev); | |
1266 | int cpu; | |
1267 | unsigned int start; | |
1268 | ||
1269 | for_each_possible_cpu(cpu) { | |
58472a76 | 1270 | struct virtnet_stats *stats = per_cpu_ptr(vi->stats, cpu); |
3fa2a1df | 1271 | u64 tpackets, tbytes, rpackets, rbytes; |
1272 | ||
1273 | do { | |
57a7744e | 1274 | start = u64_stats_fetch_begin_irq(&stats->tx_syncp); |
3fa2a1df | 1275 | tpackets = stats->tx_packets; |
1276 | tbytes = stats->tx_bytes; | |
57a7744e | 1277 | } while (u64_stats_fetch_retry_irq(&stats->tx_syncp, start)); |
83a27052 ED |
1278 | |
1279 | do { | |
57a7744e | 1280 | start = u64_stats_fetch_begin_irq(&stats->rx_syncp); |
3fa2a1df | 1281 | rpackets = stats->rx_packets; |
1282 | rbytes = stats->rx_bytes; | |
57a7744e | 1283 | } while (u64_stats_fetch_retry_irq(&stats->rx_syncp, start)); |
3fa2a1df | 1284 | |
1285 | tot->rx_packets += rpackets; | |
1286 | tot->tx_packets += tpackets; | |
1287 | tot->rx_bytes += rbytes; | |
1288 | tot->tx_bytes += tbytes; | |
1289 | } | |
1290 | ||
1291 | tot->tx_dropped = dev->stats.tx_dropped; | |
021ac8d3 | 1292 | tot->tx_fifo_errors = dev->stats.tx_fifo_errors; |
3fa2a1df | 1293 | tot->rx_dropped = dev->stats.rx_dropped; |
1294 | tot->rx_length_errors = dev->stats.rx_length_errors; | |
1295 | tot->rx_frame_errors = dev->stats.rx_frame_errors; | |
3fa2a1df | 1296 | } |
1297 | ||
da74e89d AS |
1298 | #ifdef CONFIG_NET_POLL_CONTROLLER |
1299 | static void virtnet_netpoll(struct net_device *dev) | |
1300 | { | |
1301 | struct virtnet_info *vi = netdev_priv(dev); | |
986a4f4d | 1302 | int i; |
da74e89d | 1303 | |
986a4f4d JW |
1304 | for (i = 0; i < vi->curr_queue_pairs; i++) |
1305 | napi_schedule(&vi->rq[i].napi); | |
da74e89d AS |
1306 | } |
1307 | #endif | |
1308 | ||
586d17c5 JW |
1309 | static void virtnet_ack_link_announce(struct virtnet_info *vi) |
1310 | { | |
1311 | rtnl_lock(); | |
1312 | if (!virtnet_send_command(vi, VIRTIO_NET_CTRL_ANNOUNCE, | |
d24bae32 | 1313 | VIRTIO_NET_CTRL_ANNOUNCE_ACK, NULL)) |
586d17c5 JW |
1314 | dev_warn(&vi->dev->dev, "Failed to ack link announce.\n"); |
1315 | rtnl_unlock(); | |
1316 | } | |
1317 | ||
47315329 | 1318 | static int _virtnet_set_queues(struct virtnet_info *vi, u16 queue_pairs) |
986a4f4d JW |
1319 | { |
1320 | struct scatterlist sg; | |
986a4f4d JW |
1321 | struct net_device *dev = vi->dev; |
1322 | ||
1323 | if (!vi->has_cvq || !virtio_has_feature(vi->vdev, VIRTIO_NET_F_MQ)) | |
1324 | return 0; | |
1325 | ||
a725ee3e AL |
1326 | vi->ctrl_mq.virtqueue_pairs = cpu_to_virtio16(vi->vdev, queue_pairs); |
1327 | sg_init_one(&sg, &vi->ctrl_mq, sizeof(vi->ctrl_mq)); | |
986a4f4d JW |
1328 | |
1329 | if (!virtnet_send_command(vi, VIRTIO_NET_CTRL_MQ, | |
d24bae32 | 1330 | VIRTIO_NET_CTRL_MQ_VQ_PAIRS_SET, &sg)) { |
986a4f4d JW |
1331 | dev_warn(&dev->dev, "Fail to set num of queue pairs to %d\n", |
1332 | queue_pairs); | |
1333 | return -EINVAL; | |
55257d72 | 1334 | } else { |
986a4f4d | 1335 | vi->curr_queue_pairs = queue_pairs; |
35ed159b JW |
1336 | /* virtnet_open() will refill when device is going to up. */ |
1337 | if (dev->flags & IFF_UP) | |
1338 | schedule_delayed_work(&vi->refill, 0); | |
55257d72 | 1339 | } |
986a4f4d JW |
1340 | |
1341 | return 0; | |
1342 | } | |
1343 | ||
47315329 JF |
1344 | static int virtnet_set_queues(struct virtnet_info *vi, u16 queue_pairs) |
1345 | { | |
1346 | int err; | |
1347 | ||
1348 | rtnl_lock(); | |
1349 | err = _virtnet_set_queues(vi, queue_pairs); | |
1350 | rtnl_unlock(); | |
1351 | return err; | |
1352 | } | |
1353 | ||
296f96fc RR |
1354 | static int virtnet_close(struct net_device *dev) |
1355 | { | |
1356 | struct virtnet_info *vi = netdev_priv(dev); | |
986a4f4d | 1357 | int i; |
296f96fc | 1358 | |
b2baed69 RR |
1359 | /* Make sure refill_work doesn't re-enable napi! */ |
1360 | cancel_delayed_work_sync(&vi->refill); | |
986a4f4d JW |
1361 | |
1362 | for (i = 0; i < vi->max_queue_pairs; i++) | |
1363 | napi_disable(&vi->rq[i].napi); | |
296f96fc | 1364 | |
296f96fc RR |
1365 | return 0; |
1366 | } | |
1367 | ||
2af7698e AW |
1368 | static void virtnet_set_rx_mode(struct net_device *dev) |
1369 | { | |
1370 | struct virtnet_info *vi = netdev_priv(dev); | |
f565a7c2 | 1371 | struct scatterlist sg[2]; |
f565a7c2 | 1372 | struct virtio_net_ctrl_mac *mac_data; |
ccffad25 | 1373 | struct netdev_hw_addr *ha; |
32e7bfc4 | 1374 | int uc_count; |
4cd24eaf | 1375 | int mc_count; |
f565a7c2 AW |
1376 | void *buf; |
1377 | int i; | |
2af7698e | 1378 | |
788a8b6d | 1379 | /* We can't dynamically set ndo_set_rx_mode, so return gracefully */ |
2af7698e AW |
1380 | if (!virtio_has_feature(vi->vdev, VIRTIO_NET_F_CTRL_RX)) |
1381 | return; | |
1382 | ||
2ac46030 MT |
1383 | vi->ctrl_promisc = ((dev->flags & IFF_PROMISC) != 0); |
1384 | vi->ctrl_allmulti = ((dev->flags & IFF_ALLMULTI) != 0); | |
2af7698e | 1385 | |
2ac46030 | 1386 | sg_init_one(sg, &vi->ctrl_promisc, sizeof(vi->ctrl_promisc)); |
2af7698e AW |
1387 | |
1388 | if (!virtnet_send_command(vi, VIRTIO_NET_CTRL_RX, | |
d24bae32 | 1389 | VIRTIO_NET_CTRL_RX_PROMISC, sg)) |
2af7698e | 1390 | dev_warn(&dev->dev, "Failed to %sable promisc mode.\n", |
2ac46030 | 1391 | vi->ctrl_promisc ? "en" : "dis"); |
2af7698e | 1392 | |
2ac46030 | 1393 | sg_init_one(sg, &vi->ctrl_allmulti, sizeof(vi->ctrl_allmulti)); |
2af7698e AW |
1394 | |
1395 | if (!virtnet_send_command(vi, VIRTIO_NET_CTRL_RX, | |
d24bae32 | 1396 | VIRTIO_NET_CTRL_RX_ALLMULTI, sg)) |
2af7698e | 1397 | dev_warn(&dev->dev, "Failed to %sable allmulti mode.\n", |
2ac46030 | 1398 | vi->ctrl_allmulti ? "en" : "dis"); |
f565a7c2 | 1399 | |
32e7bfc4 | 1400 | uc_count = netdev_uc_count(dev); |
4cd24eaf | 1401 | mc_count = netdev_mc_count(dev); |
f565a7c2 | 1402 | /* MAC filter - use one buffer for both lists */ |
4cd24eaf JP |
1403 | buf = kzalloc(((uc_count + mc_count) * ETH_ALEN) + |
1404 | (2 * sizeof(mac_data->entries)), GFP_ATOMIC); | |
1405 | mac_data = buf; | |
e68ed8f0 | 1406 | if (!buf) |
f565a7c2 | 1407 | return; |
f565a7c2 | 1408 | |
23e258e1 AW |
1409 | sg_init_table(sg, 2); |
1410 | ||
f565a7c2 | 1411 | /* Store the unicast list and count in the front of the buffer */ |
fdd819b2 | 1412 | mac_data->entries = cpu_to_virtio32(vi->vdev, uc_count); |
ccffad25 | 1413 | i = 0; |
32e7bfc4 | 1414 | netdev_for_each_uc_addr(ha, dev) |
ccffad25 | 1415 | memcpy(&mac_data->macs[i++][0], ha->addr, ETH_ALEN); |
f565a7c2 AW |
1416 | |
1417 | sg_set_buf(&sg[0], mac_data, | |
32e7bfc4 | 1418 | sizeof(mac_data->entries) + (uc_count * ETH_ALEN)); |
f565a7c2 AW |
1419 | |
1420 | /* multicast list and count fill the end */ | |
32e7bfc4 | 1421 | mac_data = (void *)&mac_data->macs[uc_count][0]; |
f565a7c2 | 1422 | |
fdd819b2 | 1423 | mac_data->entries = cpu_to_virtio32(vi->vdev, mc_count); |
567ec874 | 1424 | i = 0; |
22bedad3 JP |
1425 | netdev_for_each_mc_addr(ha, dev) |
1426 | memcpy(&mac_data->macs[i++][0], ha->addr, ETH_ALEN); | |
f565a7c2 AW |
1427 | |
1428 | sg_set_buf(&sg[1], mac_data, | |
4cd24eaf | 1429 | sizeof(mac_data->entries) + (mc_count * ETH_ALEN)); |
f565a7c2 AW |
1430 | |
1431 | if (!virtnet_send_command(vi, VIRTIO_NET_CTRL_MAC, | |
d24bae32 | 1432 | VIRTIO_NET_CTRL_MAC_TABLE_SET, sg)) |
99e872ae | 1433 | dev_warn(&dev->dev, "Failed to set MAC filter table.\n"); |
f565a7c2 AW |
1434 | |
1435 | kfree(buf); | |
2af7698e AW |
1436 | } |
1437 | ||
80d5c368 PM |
1438 | static int virtnet_vlan_rx_add_vid(struct net_device *dev, |
1439 | __be16 proto, u16 vid) | |
0bde9569 AW |
1440 | { |
1441 | struct virtnet_info *vi = netdev_priv(dev); | |
1442 | struct scatterlist sg; | |
1443 | ||
a725ee3e AL |
1444 | vi->ctrl_vid = vid; |
1445 | sg_init_one(&sg, &vi->ctrl_vid, sizeof(vi->ctrl_vid)); | |
0bde9569 AW |
1446 | |
1447 | if (!virtnet_send_command(vi, VIRTIO_NET_CTRL_VLAN, | |
d24bae32 | 1448 | VIRTIO_NET_CTRL_VLAN_ADD, &sg)) |
0bde9569 | 1449 | dev_warn(&dev->dev, "Failed to add VLAN ID %d.\n", vid); |
8e586137 | 1450 | return 0; |
0bde9569 AW |
1451 | } |
1452 | ||
80d5c368 PM |
1453 | static int virtnet_vlan_rx_kill_vid(struct net_device *dev, |
1454 | __be16 proto, u16 vid) | |
0bde9569 AW |
1455 | { |
1456 | struct virtnet_info *vi = netdev_priv(dev); | |
1457 | struct scatterlist sg; | |
1458 | ||
a725ee3e AL |
1459 | vi->ctrl_vid = vid; |
1460 | sg_init_one(&sg, &vi->ctrl_vid, sizeof(vi->ctrl_vid)); | |
0bde9569 AW |
1461 | |
1462 | if (!virtnet_send_command(vi, VIRTIO_NET_CTRL_VLAN, | |
d24bae32 | 1463 | VIRTIO_NET_CTRL_VLAN_DEL, &sg)) |
0bde9569 | 1464 | dev_warn(&dev->dev, "Failed to kill VLAN ID %d.\n", vid); |
8e586137 | 1465 | return 0; |
0bde9569 AW |
1466 | } |
1467 | ||
8898c21c | 1468 | static void virtnet_clean_affinity(struct virtnet_info *vi, long hcpu) |
986a4f4d JW |
1469 | { |
1470 | int i; | |
1471 | ||
8898c21c WG |
1472 | if (vi->affinity_hint_set) { |
1473 | for (i = 0; i < vi->max_queue_pairs; i++) { | |
47be2479 WG |
1474 | virtqueue_set_affinity(vi->rq[i].vq, -1); |
1475 | virtqueue_set_affinity(vi->sq[i].vq, -1); | |
1476 | } | |
1477 | ||
8898c21c WG |
1478 | vi->affinity_hint_set = false; |
1479 | } | |
8898c21c | 1480 | } |
47be2479 | 1481 | |
8898c21c WG |
1482 | static void virtnet_set_affinity(struct virtnet_info *vi) |
1483 | { | |
1484 | int i; | |
1485 | int cpu; | |
986a4f4d JW |
1486 | |
1487 | /* In multiqueue mode, when the number of cpu is equal to the number of | |
1488 | * queue pairs, we let the queue pairs to be private to one cpu by | |
1489 | * setting the affinity hint to eliminate the contention. | |
1490 | */ | |
8898c21c WG |
1491 | if (vi->curr_queue_pairs == 1 || |
1492 | vi->max_queue_pairs != num_online_cpus()) { | |
1493 | virtnet_clean_affinity(vi, -1); | |
1494 | return; | |
986a4f4d JW |
1495 | } |
1496 | ||
8898c21c WG |
1497 | i = 0; |
1498 | for_each_online_cpu(cpu) { | |
986a4f4d JW |
1499 | virtqueue_set_affinity(vi->rq[i].vq, cpu); |
1500 | virtqueue_set_affinity(vi->sq[i].vq, cpu); | |
9bb8ca86 | 1501 | netif_set_xps_queue(vi->dev, cpumask_of(cpu), i); |
8898c21c | 1502 | i++; |
986a4f4d JW |
1503 | } |
1504 | ||
8898c21c | 1505 | vi->affinity_hint_set = true; |
986a4f4d JW |
1506 | } |
1507 | ||
8017c279 | 1508 | static int virtnet_cpu_online(unsigned int cpu, struct hlist_node *node) |
8de4b2f3 | 1509 | { |
8017c279 SAS |
1510 | struct virtnet_info *vi = hlist_entry_safe(node, struct virtnet_info, |
1511 | node); | |
1512 | virtnet_set_affinity(vi); | |
1513 | return 0; | |
1514 | } | |
8de4b2f3 | 1515 | |
8017c279 SAS |
1516 | static int virtnet_cpu_dead(unsigned int cpu, struct hlist_node *node) |
1517 | { | |
1518 | struct virtnet_info *vi = hlist_entry_safe(node, struct virtnet_info, | |
1519 | node_dead); | |
1520 | virtnet_set_affinity(vi); | |
1521 | return 0; | |
1522 | } | |
3ab098df | 1523 | |
8017c279 SAS |
1524 | static int virtnet_cpu_down_prep(unsigned int cpu, struct hlist_node *node) |
1525 | { | |
1526 | struct virtnet_info *vi = hlist_entry_safe(node, struct virtnet_info, | |
1527 | node); | |
1528 | ||
1529 | virtnet_clean_affinity(vi, cpu); | |
1530 | return 0; | |
1531 | } | |
1532 | ||
1533 | static enum cpuhp_state virtionet_online; | |
1534 | ||
1535 | static int virtnet_cpu_notif_add(struct virtnet_info *vi) | |
1536 | { | |
1537 | int ret; | |
1538 | ||
1539 | ret = cpuhp_state_add_instance_nocalls(virtionet_online, &vi->node); | |
1540 | if (ret) | |
1541 | return ret; | |
1542 | ret = cpuhp_state_add_instance_nocalls(CPUHP_VIRT_NET_DEAD, | |
1543 | &vi->node_dead); | |
1544 | if (!ret) | |
1545 | return ret; | |
1546 | cpuhp_state_remove_instance_nocalls(virtionet_online, &vi->node); | |
1547 | return ret; | |
1548 | } | |
1549 | ||
1550 | static void virtnet_cpu_notif_remove(struct virtnet_info *vi) | |
1551 | { | |
1552 | cpuhp_state_remove_instance_nocalls(virtionet_online, &vi->node); | |
1553 | cpuhp_state_remove_instance_nocalls(CPUHP_VIRT_NET_DEAD, | |
1554 | &vi->node_dead); | |
986a4f4d JW |
1555 | } |
1556 | ||
8f9f4668 RJ |
1557 | static void virtnet_get_ringparam(struct net_device *dev, |
1558 | struct ethtool_ringparam *ring) | |
1559 | { | |
1560 | struct virtnet_info *vi = netdev_priv(dev); | |
1561 | ||
986a4f4d JW |
1562 | ring->rx_max_pending = virtqueue_get_vring_size(vi->rq[0].vq); |
1563 | ring->tx_max_pending = virtqueue_get_vring_size(vi->sq[0].vq); | |
8f9f4668 RJ |
1564 | ring->rx_pending = ring->rx_max_pending; |
1565 | ring->tx_pending = ring->tx_max_pending; | |
8f9f4668 RJ |
1566 | } |
1567 | ||
66846048 RJ |
1568 | |
1569 | static void virtnet_get_drvinfo(struct net_device *dev, | |
1570 | struct ethtool_drvinfo *info) | |
1571 | { | |
1572 | struct virtnet_info *vi = netdev_priv(dev); | |
1573 | struct virtio_device *vdev = vi->vdev; | |
1574 | ||
1575 | strlcpy(info->driver, KBUILD_MODNAME, sizeof(info->driver)); | |
1576 | strlcpy(info->version, VIRTNET_DRIVER_VERSION, sizeof(info->version)); | |
1577 | strlcpy(info->bus_info, virtio_bus_name(vdev), sizeof(info->bus_info)); | |
1578 | ||
1579 | } | |
1580 | ||
d73bcd2c JW |
1581 | /* TODO: Eliminate OOO packets during switching */ |
1582 | static int virtnet_set_channels(struct net_device *dev, | |
1583 | struct ethtool_channels *channels) | |
1584 | { | |
1585 | struct virtnet_info *vi = netdev_priv(dev); | |
1586 | u16 queue_pairs = channels->combined_count; | |
1587 | int err; | |
1588 | ||
1589 | /* We don't support separate rx/tx channels. | |
1590 | * We don't allow setting 'other' channels. | |
1591 | */ | |
1592 | if (channels->rx_count || channels->tx_count || channels->other_count) | |
1593 | return -EINVAL; | |
1594 | ||
c18e9cd6 | 1595 | if (queue_pairs > vi->max_queue_pairs || queue_pairs == 0) |
d73bcd2c JW |
1596 | return -EINVAL; |
1597 | ||
f600b690 JF |
1598 | /* For now we don't support modifying channels while XDP is loaded |
1599 | * also when XDP is loaded all RX queues have XDP programs so we only | |
1600 | * need to check a single RX queue. | |
1601 | */ | |
1602 | if (vi->rq[0].xdp_prog) | |
1603 | return -EINVAL; | |
1604 | ||
47be2479 | 1605 | get_online_cpus(); |
47315329 | 1606 | err = _virtnet_set_queues(vi, queue_pairs); |
d73bcd2c JW |
1607 | if (!err) { |
1608 | netif_set_real_num_tx_queues(dev, queue_pairs); | |
1609 | netif_set_real_num_rx_queues(dev, queue_pairs); | |
1610 | ||
8898c21c | 1611 | virtnet_set_affinity(vi); |
d73bcd2c | 1612 | } |
47be2479 | 1613 | put_online_cpus(); |
d73bcd2c JW |
1614 | |
1615 | return err; | |
1616 | } | |
1617 | ||
1618 | static void virtnet_get_channels(struct net_device *dev, | |
1619 | struct ethtool_channels *channels) | |
1620 | { | |
1621 | struct virtnet_info *vi = netdev_priv(dev); | |
1622 | ||
1623 | channels->combined_count = vi->curr_queue_pairs; | |
1624 | channels->max_combined = vi->max_queue_pairs; | |
1625 | channels->max_other = 0; | |
1626 | channels->rx_count = 0; | |
1627 | channels->tx_count = 0; | |
1628 | channels->other_count = 0; | |
1629 | } | |
1630 | ||
16032be5 NA |
1631 | /* Check if the user is trying to change anything besides speed/duplex */ |
1632 | static bool virtnet_validate_ethtool_cmd(const struct ethtool_cmd *cmd) | |
1633 | { | |
1634 | struct ethtool_cmd diff1 = *cmd; | |
1635 | struct ethtool_cmd diff2 = {}; | |
1636 | ||
0cf3ace9 NA |
1637 | /* cmd is always set so we need to clear it, validate the port type |
1638 | * and also without autonegotiation we can ignore advertising | |
1639 | */ | |
16032be5 | 1640 | ethtool_cmd_speed_set(&diff1, 0); |
0cf3ace9 | 1641 | diff2.port = PORT_OTHER; |
16032be5 NA |
1642 | diff1.advertising = 0; |
1643 | diff1.duplex = 0; | |
16032be5 NA |
1644 | diff1.cmd = 0; |
1645 | ||
1646 | return !memcmp(&diff1, &diff2, sizeof(diff1)); | |
1647 | } | |
1648 | ||
1649 | static int virtnet_set_settings(struct net_device *dev, struct ethtool_cmd *cmd) | |
1650 | { | |
1651 | struct virtnet_info *vi = netdev_priv(dev); | |
1652 | u32 speed; | |
1653 | ||
1654 | speed = ethtool_cmd_speed(cmd); | |
1655 | /* don't allow custom speed and duplex */ | |
1656 | if (!ethtool_validate_speed(speed) || | |
1657 | !ethtool_validate_duplex(cmd->duplex) || | |
1658 | !virtnet_validate_ethtool_cmd(cmd)) | |
1659 | return -EINVAL; | |
1660 | vi->speed = speed; | |
1661 | vi->duplex = cmd->duplex; | |
1662 | ||
1663 | return 0; | |
1664 | } | |
1665 | ||
1666 | static int virtnet_get_settings(struct net_device *dev, struct ethtool_cmd *cmd) | |
1667 | { | |
1668 | struct virtnet_info *vi = netdev_priv(dev); | |
1669 | ||
1670 | ethtool_cmd_speed_set(cmd, vi->speed); | |
1671 | cmd->duplex = vi->duplex; | |
1672 | cmd->port = PORT_OTHER; | |
1673 | ||
1674 | return 0; | |
1675 | } | |
1676 | ||
1677 | static void virtnet_init_settings(struct net_device *dev) | |
1678 | { | |
1679 | struct virtnet_info *vi = netdev_priv(dev); | |
1680 | ||
1681 | vi->speed = SPEED_UNKNOWN; | |
1682 | vi->duplex = DUPLEX_UNKNOWN; | |
1683 | } | |
1684 | ||
0fc0b732 | 1685 | static const struct ethtool_ops virtnet_ethtool_ops = { |
66846048 | 1686 | .get_drvinfo = virtnet_get_drvinfo, |
9f4d26d0 | 1687 | .get_link = ethtool_op_get_link, |
8f9f4668 | 1688 | .get_ringparam = virtnet_get_ringparam, |
d73bcd2c JW |
1689 | .set_channels = virtnet_set_channels, |
1690 | .get_channels = virtnet_get_channels, | |
074c3582 | 1691 | .get_ts_info = ethtool_op_get_ts_info, |
16032be5 NA |
1692 | .get_settings = virtnet_get_settings, |
1693 | .set_settings = virtnet_set_settings, | |
a9ea3fc6 HX |
1694 | }; |
1695 | ||
9fe7bfce JF |
1696 | static void virtnet_freeze_down(struct virtio_device *vdev) |
1697 | { | |
1698 | struct virtnet_info *vi = vdev->priv; | |
1699 | int i; | |
1700 | ||
1701 | /* Make sure no work handler is accessing the device */ | |
1702 | flush_work(&vi->config_work); | |
1703 | ||
1704 | netif_device_detach(vi->dev); | |
1705 | cancel_delayed_work_sync(&vi->refill); | |
1706 | ||
1707 | if (netif_running(vi->dev)) { | |
1708 | for (i = 0; i < vi->max_queue_pairs; i++) | |
1709 | napi_disable(&vi->rq[i].napi); | |
1710 | } | |
1711 | } | |
1712 | ||
1713 | static int init_vqs(struct virtnet_info *vi); | |
2de2f7f4 | 1714 | static void _remove_vq_common(struct virtnet_info *vi); |
9fe7bfce JF |
1715 | |
1716 | static int virtnet_restore_up(struct virtio_device *vdev) | |
1717 | { | |
1718 | struct virtnet_info *vi = vdev->priv; | |
1719 | int err, i; | |
1720 | ||
1721 | err = init_vqs(vi); | |
1722 | if (err) | |
1723 | return err; | |
1724 | ||
1725 | virtio_device_ready(vdev); | |
1726 | ||
1727 | if (netif_running(vi->dev)) { | |
1728 | for (i = 0; i < vi->curr_queue_pairs; i++) | |
1729 | if (!try_fill_recv(vi, &vi->rq[i], GFP_KERNEL)) | |
1730 | schedule_delayed_work(&vi->refill, 0); | |
1731 | ||
1732 | for (i = 0; i < vi->max_queue_pairs; i++) | |
1733 | virtnet_napi_enable(&vi->rq[i]); | |
1734 | } | |
1735 | ||
1736 | netif_device_attach(vi->dev); | |
1737 | return err; | |
1738 | } | |
1739 | ||
2de2f7f4 JF |
1740 | static int virtnet_reset(struct virtnet_info *vi) |
1741 | { | |
1742 | struct virtio_device *dev = vi->vdev; | |
1743 | int ret; | |
1744 | ||
1745 | virtio_config_disable(dev); | |
1746 | dev->failed = dev->config->get_status(dev) & VIRTIO_CONFIG_S_FAILED; | |
1747 | virtnet_freeze_down(dev); | |
1748 | _remove_vq_common(vi); | |
1749 | ||
1750 | dev->config->reset(dev); | |
1751 | virtio_add_status(dev, VIRTIO_CONFIG_S_ACKNOWLEDGE); | |
1752 | virtio_add_status(dev, VIRTIO_CONFIG_S_DRIVER); | |
1753 | ||
1754 | ret = virtio_finalize_features(dev); | |
1755 | if (ret) | |
1756 | goto err; | |
1757 | ||
1758 | ret = virtnet_restore_up(dev); | |
1759 | if (ret) | |
1760 | goto err; | |
1761 | ret = _virtnet_set_queues(vi, vi->curr_queue_pairs); | |
1762 | if (ret) | |
1763 | goto err; | |
1764 | ||
1765 | virtio_add_status(dev, VIRTIO_CONFIG_S_DRIVER_OK); | |
1766 | virtio_config_enable(dev); | |
1767 | return 0; | |
1768 | err: | |
1769 | virtio_add_status(dev, VIRTIO_CONFIG_S_FAILED); | |
1770 | return ret; | |
1771 | } | |
1772 | ||
f600b690 JF |
1773 | static int virtnet_xdp_set(struct net_device *dev, struct bpf_prog *prog) |
1774 | { | |
1775 | unsigned long int max_sz = PAGE_SIZE - sizeof(struct padded_vnet_hdr); | |
1776 | struct virtnet_info *vi = netdev_priv(dev); | |
1777 | struct bpf_prog *old_prog; | |
2de2f7f4 | 1778 | u16 oxdp_qp, xdp_qp = 0, curr_qp; |
672aafd5 | 1779 | int i, err; |
f600b690 JF |
1780 | |
1781 | if (virtio_has_feature(vi->vdev, VIRTIO_NET_F_GUEST_TSO4) || | |
92502fe8 JW |
1782 | virtio_has_feature(vi->vdev, VIRTIO_NET_F_GUEST_TSO6) || |
1783 | virtio_has_feature(vi->vdev, VIRTIO_NET_F_GUEST_ECN) || | |
1784 | virtio_has_feature(vi->vdev, VIRTIO_NET_F_GUEST_UFO)) { | |
f600b690 JF |
1785 | netdev_warn(dev, "can't set XDP while host is implementing LRO, disable LRO first\n"); |
1786 | return -EOPNOTSUPP; | |
1787 | } | |
1788 | ||
1789 | if (vi->mergeable_rx_bufs && !vi->any_header_sg) { | |
1790 | netdev_warn(dev, "XDP expects header/data in single page, any_header_sg required\n"); | |
1791 | return -EINVAL; | |
1792 | } | |
1793 | ||
1794 | if (dev->mtu > max_sz) { | |
1795 | netdev_warn(dev, "XDP requires MTU less than %lu\n", max_sz); | |
1796 | return -EINVAL; | |
1797 | } | |
1798 | ||
672aafd5 JF |
1799 | curr_qp = vi->curr_queue_pairs - vi->xdp_queue_pairs; |
1800 | if (prog) | |
1801 | xdp_qp = nr_cpu_ids; | |
1802 | ||
1803 | /* XDP requires extra queues for XDP_TX */ | |
1804 | if (curr_qp + xdp_qp > vi->max_queue_pairs) { | |
1805 | netdev_warn(dev, "request %i queues but max is %i\n", | |
1806 | curr_qp + xdp_qp, vi->max_queue_pairs); | |
1807 | return -ENOMEM; | |
1808 | } | |
1809 | ||
2de2f7f4 JF |
1810 | if (prog) { |
1811 | prog = bpf_prog_add(prog, vi->max_queue_pairs - 1); | |
1812 | if (IS_ERR(prog)) | |
1813 | return PTR_ERR(prog); | |
1814 | } | |
1815 | ||
47315329 | 1816 | err = _virtnet_set_queues(vi, curr_qp + xdp_qp); |
672aafd5 JF |
1817 | if (err) { |
1818 | dev_warn(&dev->dev, "XDP Device queue allocation failure.\n"); | |
2de2f7f4 | 1819 | goto virtio_queue_err; |
672aafd5 JF |
1820 | } |
1821 | ||
2de2f7f4 JF |
1822 | oxdp_qp = vi->xdp_queue_pairs; |
1823 | ||
1824 | /* Changing the headroom in buffers is a disruptive operation because | |
1825 | * existing buffers must be flushed and reallocated. This will happen | |
1826 | * when a xdp program is initially added or xdp is disabled by removing | |
1827 | * the xdp program resulting in number of XDP queues changing. | |
1828 | */ | |
1829 | if (vi->xdp_queue_pairs != xdp_qp) { | |
1830 | vi->xdp_queue_pairs = xdp_qp; | |
1831 | err = virtnet_reset(vi); | |
1832 | if (err) | |
1833 | goto virtio_reset_err; | |
f600b690 JF |
1834 | } |
1835 | ||
672aafd5 JF |
1836 | netif_set_real_num_rx_queues(dev, curr_qp + xdp_qp); |
1837 | ||
f600b690 JF |
1838 | for (i = 0; i < vi->max_queue_pairs; i++) { |
1839 | old_prog = rtnl_dereference(vi->rq[i].xdp_prog); | |
1840 | rcu_assign_pointer(vi->rq[i].xdp_prog, prog); | |
1841 | if (old_prog) | |
1842 | bpf_prog_put(old_prog); | |
1843 | } | |
1844 | ||
1845 | return 0; | |
2de2f7f4 JF |
1846 | |
1847 | virtio_reset_err: | |
1848 | /* On reset error do our best to unwind XDP changes inflight and return | |
1849 | * error up to user space for resolution. The underlying reset hung on | |
1850 | * us so not much we can do here. | |
1851 | */ | |
1852 | dev_warn(&dev->dev, "XDP reset failure and queues unstable\n"); | |
1853 | vi->xdp_queue_pairs = oxdp_qp; | |
1854 | virtio_queue_err: | |
1855 | /* On queue set error we can unwind bpf ref count and user space can | |
1856 | * retry this is most likely an allocation failure. | |
1857 | */ | |
1858 | if (prog) | |
1859 | bpf_prog_sub(prog, vi->max_queue_pairs - 1); | |
1860 | return err; | |
f600b690 JF |
1861 | } |
1862 | ||
1863 | static bool virtnet_xdp_query(struct net_device *dev) | |
1864 | { | |
1865 | struct virtnet_info *vi = netdev_priv(dev); | |
1866 | int i; | |
1867 | ||
1868 | for (i = 0; i < vi->max_queue_pairs; i++) { | |
1869 | if (vi->rq[i].xdp_prog) | |
1870 | return true; | |
1871 | } | |
1872 | return false; | |
1873 | } | |
1874 | ||
1875 | static int virtnet_xdp(struct net_device *dev, struct netdev_xdp *xdp) | |
1876 | { | |
1877 | switch (xdp->command) { | |
1878 | case XDP_SETUP_PROG: | |
1879 | return virtnet_xdp_set(dev, xdp->prog); | |
1880 | case XDP_QUERY_PROG: | |
1881 | xdp->prog_attached = virtnet_xdp_query(dev); | |
1882 | return 0; | |
1883 | default: | |
1884 | return -EINVAL; | |
1885 | } | |
1886 | } | |
1887 | ||
76288b4e SH |
1888 | static const struct net_device_ops virtnet_netdev = { |
1889 | .ndo_open = virtnet_open, | |
1890 | .ndo_stop = virtnet_close, | |
1891 | .ndo_start_xmit = start_xmit, | |
1892 | .ndo_validate_addr = eth_validate_addr, | |
9c46f6d4 | 1893 | .ndo_set_mac_address = virtnet_set_mac_address, |
2af7698e | 1894 | .ndo_set_rx_mode = virtnet_set_rx_mode, |
3fa2a1df | 1895 | .ndo_get_stats64 = virtnet_stats, |
1824a989 AW |
1896 | .ndo_vlan_rx_add_vid = virtnet_vlan_rx_add_vid, |
1897 | .ndo_vlan_rx_kill_vid = virtnet_vlan_rx_kill_vid, | |
76288b4e SH |
1898 | #ifdef CONFIG_NET_POLL_CONTROLLER |
1899 | .ndo_poll_controller = virtnet_netpoll, | |
91815639 | 1900 | #endif |
f600b690 | 1901 | .ndo_xdp = virtnet_xdp, |
76288b4e SH |
1902 | }; |
1903 | ||
586d17c5 | 1904 | static void virtnet_config_changed_work(struct work_struct *work) |
9f4d26d0 | 1905 | { |
586d17c5 JW |
1906 | struct virtnet_info *vi = |
1907 | container_of(work, struct virtnet_info, config_work); | |
9f4d26d0 MM |
1908 | u16 v; |
1909 | ||
855e0c52 RR |
1910 | if (virtio_cread_feature(vi->vdev, VIRTIO_NET_F_STATUS, |
1911 | struct virtio_net_config, status, &v) < 0) | |
507613bf | 1912 | return; |
586d17c5 JW |
1913 | |
1914 | if (v & VIRTIO_NET_S_ANNOUNCE) { | |
ee89bab1 | 1915 | netdev_notify_peers(vi->dev); |
586d17c5 JW |
1916 | virtnet_ack_link_announce(vi); |
1917 | } | |
9f4d26d0 MM |
1918 | |
1919 | /* Ignore unknown (future) status bits */ | |
1920 | v &= VIRTIO_NET_S_LINK_UP; | |
1921 | ||
1922 | if (vi->status == v) | |
507613bf | 1923 | return; |
9f4d26d0 MM |
1924 | |
1925 | vi->status = v; | |
1926 | ||
1927 | if (vi->status & VIRTIO_NET_S_LINK_UP) { | |
1928 | netif_carrier_on(vi->dev); | |
986a4f4d | 1929 | netif_tx_wake_all_queues(vi->dev); |
9f4d26d0 MM |
1930 | } else { |
1931 | netif_carrier_off(vi->dev); | |
986a4f4d | 1932 | netif_tx_stop_all_queues(vi->dev); |
9f4d26d0 MM |
1933 | } |
1934 | } | |
1935 | ||
1936 | static void virtnet_config_changed(struct virtio_device *vdev) | |
1937 | { | |
1938 | struct virtnet_info *vi = vdev->priv; | |
1939 | ||
3b07e9ca | 1940 | schedule_work(&vi->config_work); |
9f4d26d0 MM |
1941 | } |
1942 | ||
986a4f4d JW |
1943 | static void virtnet_free_queues(struct virtnet_info *vi) |
1944 | { | |
d4fb84ee AV |
1945 | int i; |
1946 | ||
ab3971b1 JW |
1947 | for (i = 0; i < vi->max_queue_pairs; i++) { |
1948 | napi_hash_del(&vi->rq[i].napi); | |
d4fb84ee | 1949 | netif_napi_del(&vi->rq[i].napi); |
ab3971b1 | 1950 | } |
d4fb84ee | 1951 | |
963abe5c ED |
1952 | /* We called napi_hash_del() before netif_napi_del(), |
1953 | * we need to respect an RCU grace period before freeing vi->rq | |
1954 | */ | |
1955 | synchronize_net(); | |
1956 | ||
986a4f4d JW |
1957 | kfree(vi->rq); |
1958 | kfree(vi->sq); | |
1959 | } | |
1960 | ||
47315329 | 1961 | static void _free_receive_bufs(struct virtnet_info *vi) |
986a4f4d | 1962 | { |
f600b690 | 1963 | struct bpf_prog *old_prog; |
986a4f4d JW |
1964 | int i; |
1965 | ||
1966 | for (i = 0; i < vi->max_queue_pairs; i++) { | |
1967 | while (vi->rq[i].pages) | |
1968 | __free_pages(get_a_page(&vi->rq[i], GFP_KERNEL), 0); | |
f600b690 JF |
1969 | |
1970 | old_prog = rtnl_dereference(vi->rq[i].xdp_prog); | |
1971 | RCU_INIT_POINTER(vi->rq[i].xdp_prog, NULL); | |
1972 | if (old_prog) | |
1973 | bpf_prog_put(old_prog); | |
986a4f4d | 1974 | } |
47315329 JF |
1975 | } |
1976 | ||
1977 | static void free_receive_bufs(struct virtnet_info *vi) | |
1978 | { | |
1979 | rtnl_lock(); | |
1980 | _free_receive_bufs(vi); | |
f600b690 | 1981 | rtnl_unlock(); |
986a4f4d JW |
1982 | } |
1983 | ||
fb51879d MD |
1984 | static void free_receive_page_frags(struct virtnet_info *vi) |
1985 | { | |
1986 | int i; | |
1987 | for (i = 0; i < vi->max_queue_pairs; i++) | |
1988 | if (vi->rq[i].alloc_frag.page) | |
1989 | put_page(vi->rq[i].alloc_frag.page); | |
1990 | } | |
1991 | ||
b68df015 | 1992 | static bool is_xdp_raw_buffer_queue(struct virtnet_info *vi, int q) |
56434a01 | 1993 | { |
b68df015 JF |
1994 | /* For small receive mode always use kfree_skb variants */ |
1995 | if (!vi->mergeable_rx_bufs) | |
1996 | return false; | |
1997 | ||
56434a01 JF |
1998 | if (q < (vi->curr_queue_pairs - vi->xdp_queue_pairs)) |
1999 | return false; | |
2000 | else if (q < vi->curr_queue_pairs) | |
2001 | return true; | |
2002 | else | |
2003 | return false; | |
2004 | } | |
2005 | ||
986a4f4d JW |
2006 | static void free_unused_bufs(struct virtnet_info *vi) |
2007 | { | |
2008 | void *buf; | |
2009 | int i; | |
2010 | ||
2011 | for (i = 0; i < vi->max_queue_pairs; i++) { | |
2012 | struct virtqueue *vq = vi->sq[i].vq; | |
56434a01 | 2013 | while ((buf = virtqueue_detach_unused_buf(vq)) != NULL) { |
b68df015 | 2014 | if (!is_xdp_raw_buffer_queue(vi, i)) |
56434a01 JF |
2015 | dev_kfree_skb(buf); |
2016 | else | |
2017 | put_page(virt_to_head_page(buf)); | |
2018 | } | |
986a4f4d JW |
2019 | } |
2020 | ||
2021 | for (i = 0; i < vi->max_queue_pairs; i++) { | |
2022 | struct virtqueue *vq = vi->rq[i].vq; | |
2023 | ||
2024 | while ((buf = virtqueue_detach_unused_buf(vq)) != NULL) { | |
ab7db917 MD |
2025 | if (vi->mergeable_rx_bufs) { |
2026 | unsigned long ctx = (unsigned long)buf; | |
2027 | void *base = mergeable_ctx_to_buf_address(ctx); | |
2028 | put_page(virt_to_head_page(base)); | |
2029 | } else if (vi->big_packets) { | |
fa9fac17 | 2030 | give_pages(&vi->rq[i], buf); |
ab7db917 | 2031 | } else { |
986a4f4d | 2032 | dev_kfree_skb(buf); |
ab7db917 | 2033 | } |
986a4f4d | 2034 | } |
986a4f4d JW |
2035 | } |
2036 | } | |
2037 | ||
e9d7417b JW |
2038 | static void virtnet_del_vqs(struct virtnet_info *vi) |
2039 | { | |
2040 | struct virtio_device *vdev = vi->vdev; | |
2041 | ||
8898c21c | 2042 | virtnet_clean_affinity(vi, -1); |
986a4f4d | 2043 | |
e9d7417b | 2044 | vdev->config->del_vqs(vdev); |
986a4f4d JW |
2045 | |
2046 | virtnet_free_queues(vi); | |
e9d7417b JW |
2047 | } |
2048 | ||
986a4f4d | 2049 | static int virtnet_find_vqs(struct virtnet_info *vi) |
3f9c10b0 | 2050 | { |
986a4f4d JW |
2051 | vq_callback_t **callbacks; |
2052 | struct virtqueue **vqs; | |
2053 | int ret = -ENOMEM; | |
2054 | int i, total_vqs; | |
2055 | const char **names; | |
2056 | ||
2057 | /* We expect 1 RX virtqueue followed by 1 TX virtqueue, followed by | |
2058 | * possible N-1 RX/TX queue pairs used in multiqueue mode, followed by | |
2059 | * possible control vq. | |
2060 | */ | |
2061 | total_vqs = vi->max_queue_pairs * 2 + | |
2062 | virtio_has_feature(vi->vdev, VIRTIO_NET_F_CTRL_VQ); | |
2063 | ||
2064 | /* Allocate space for find_vqs parameters */ | |
2065 | vqs = kzalloc(total_vqs * sizeof(*vqs), GFP_KERNEL); | |
2066 | if (!vqs) | |
2067 | goto err_vq; | |
2068 | callbacks = kmalloc(total_vqs * sizeof(*callbacks), GFP_KERNEL); | |
2069 | if (!callbacks) | |
2070 | goto err_callback; | |
2071 | names = kmalloc(total_vqs * sizeof(*names), GFP_KERNEL); | |
2072 | if (!names) | |
2073 | goto err_names; | |
2074 | ||
2075 | /* Parameters for control virtqueue, if any */ | |
2076 | if (vi->has_cvq) { | |
2077 | callbacks[total_vqs - 1] = NULL; | |
2078 | names[total_vqs - 1] = "control"; | |
2079 | } | |
3f9c10b0 | 2080 | |
986a4f4d JW |
2081 | /* Allocate/initialize parameters for send/receive virtqueues */ |
2082 | for (i = 0; i < vi->max_queue_pairs; i++) { | |
2083 | callbacks[rxq2vq(i)] = skb_recv_done; | |
2084 | callbacks[txq2vq(i)] = skb_xmit_done; | |
2085 | sprintf(vi->rq[i].name, "input.%d", i); | |
2086 | sprintf(vi->sq[i].name, "output.%d", i); | |
2087 | names[rxq2vq(i)] = vi->rq[i].name; | |
2088 | names[txq2vq(i)] = vi->sq[i].name; | |
2089 | } | |
3f9c10b0 | 2090 | |
986a4f4d JW |
2091 | ret = vi->vdev->config->find_vqs(vi->vdev, total_vqs, vqs, callbacks, |
2092 | names); | |
2093 | if (ret) | |
2094 | goto err_find; | |
3f9c10b0 | 2095 | |
986a4f4d JW |
2096 | if (vi->has_cvq) { |
2097 | vi->cvq = vqs[total_vqs - 1]; | |
3f9c10b0 | 2098 | if (virtio_has_feature(vi->vdev, VIRTIO_NET_F_CTRL_VLAN)) |
f646968f | 2099 | vi->dev->features |= NETIF_F_HW_VLAN_CTAG_FILTER; |
3f9c10b0 | 2100 | } |
986a4f4d JW |
2101 | |
2102 | for (i = 0; i < vi->max_queue_pairs; i++) { | |
2103 | vi->rq[i].vq = vqs[rxq2vq(i)]; | |
2104 | vi->sq[i].vq = vqs[txq2vq(i)]; | |
2105 | } | |
2106 | ||
2107 | kfree(names); | |
2108 | kfree(callbacks); | |
2109 | kfree(vqs); | |
2110 | ||
3f9c10b0 | 2111 | return 0; |
986a4f4d JW |
2112 | |
2113 | err_find: | |
2114 | kfree(names); | |
2115 | err_names: | |
2116 | kfree(callbacks); | |
2117 | err_callback: | |
2118 | kfree(vqs); | |
2119 | err_vq: | |
2120 | return ret; | |
2121 | } | |
2122 | ||
2123 | static int virtnet_alloc_queues(struct virtnet_info *vi) | |
2124 | { | |
2125 | int i; | |
2126 | ||
2127 | vi->sq = kzalloc(sizeof(*vi->sq) * vi->max_queue_pairs, GFP_KERNEL); | |
2128 | if (!vi->sq) | |
2129 | goto err_sq; | |
2130 | vi->rq = kzalloc(sizeof(*vi->rq) * vi->max_queue_pairs, GFP_KERNEL); | |
008d4278 | 2131 | if (!vi->rq) |
986a4f4d JW |
2132 | goto err_rq; |
2133 | ||
2134 | INIT_DELAYED_WORK(&vi->refill, refill_work); | |
2135 | for (i = 0; i < vi->max_queue_pairs; i++) { | |
2136 | vi->rq[i].pages = NULL; | |
2137 | netif_napi_add(vi->dev, &vi->rq[i].napi, virtnet_poll, | |
2138 | napi_weight); | |
2139 | ||
2140 | sg_init_table(vi->rq[i].sg, ARRAY_SIZE(vi->rq[i].sg)); | |
5377d758 | 2141 | ewma_pkt_len_init(&vi->rq[i].mrg_avg_pkt_len); |
986a4f4d JW |
2142 | sg_init_table(vi->sq[i].sg, ARRAY_SIZE(vi->sq[i].sg)); |
2143 | } | |
2144 | ||
2145 | return 0; | |
2146 | ||
2147 | err_rq: | |
2148 | kfree(vi->sq); | |
2149 | err_sq: | |
2150 | return -ENOMEM; | |
2151 | } | |
2152 | ||
2153 | static int init_vqs(struct virtnet_info *vi) | |
2154 | { | |
2155 | int ret; | |
2156 | ||
2157 | /* Allocate send & receive queues */ | |
2158 | ret = virtnet_alloc_queues(vi); | |
2159 | if (ret) | |
2160 | goto err; | |
2161 | ||
2162 | ret = virtnet_find_vqs(vi); | |
2163 | if (ret) | |
2164 | goto err_free; | |
2165 | ||
47be2479 | 2166 | get_online_cpus(); |
8898c21c | 2167 | virtnet_set_affinity(vi); |
47be2479 WG |
2168 | put_online_cpus(); |
2169 | ||
986a4f4d JW |
2170 | return 0; |
2171 | ||
2172 | err_free: | |
2173 | virtnet_free_queues(vi); | |
2174 | err: | |
2175 | return ret; | |
3f9c10b0 AS |
2176 | } |
2177 | ||
fbf28d78 MD |
2178 | #ifdef CONFIG_SYSFS |
2179 | static ssize_t mergeable_rx_buffer_size_show(struct netdev_rx_queue *queue, | |
2180 | struct rx_queue_attribute *attribute, char *buf) | |
2181 | { | |
2182 | struct virtnet_info *vi = netdev_priv(queue->dev); | |
2183 | unsigned int queue_index = get_netdev_rx_queue_index(queue); | |
5377d758 | 2184 | struct ewma_pkt_len *avg; |
fbf28d78 MD |
2185 | |
2186 | BUG_ON(queue_index >= vi->max_queue_pairs); | |
2187 | avg = &vi->rq[queue_index].mrg_avg_pkt_len; | |
2188 | return sprintf(buf, "%u\n", get_mergeable_buf_len(avg)); | |
2189 | } | |
2190 | ||
2191 | static struct rx_queue_attribute mergeable_rx_buffer_size_attribute = | |
2192 | __ATTR_RO(mergeable_rx_buffer_size); | |
2193 | ||
2194 | static struct attribute *virtio_net_mrg_rx_attrs[] = { | |
2195 | &mergeable_rx_buffer_size_attribute.attr, | |
2196 | NULL | |
2197 | }; | |
2198 | ||
2199 | static const struct attribute_group virtio_net_mrg_rx_group = { | |
2200 | .name = "virtio_net", | |
2201 | .attrs = virtio_net_mrg_rx_attrs | |
2202 | }; | |
2203 | #endif | |
2204 | ||
892d6eb1 JW |
2205 | static bool virtnet_fail_on_feature(struct virtio_device *vdev, |
2206 | unsigned int fbit, | |
2207 | const char *fname, const char *dname) | |
2208 | { | |
2209 | if (!virtio_has_feature(vdev, fbit)) | |
2210 | return false; | |
2211 | ||
2212 | dev_err(&vdev->dev, "device advertises feature %s but not %s", | |
2213 | fname, dname); | |
2214 | ||
2215 | return true; | |
2216 | } | |
2217 | ||
2218 | #define VIRTNET_FAIL_ON(vdev, fbit, dbit) \ | |
2219 | virtnet_fail_on_feature(vdev, fbit, #fbit, dbit) | |
2220 | ||
2221 | static bool virtnet_validate_features(struct virtio_device *vdev) | |
2222 | { | |
2223 | if (!virtio_has_feature(vdev, VIRTIO_NET_F_CTRL_VQ) && | |
2224 | (VIRTNET_FAIL_ON(vdev, VIRTIO_NET_F_CTRL_RX, | |
2225 | "VIRTIO_NET_F_CTRL_VQ") || | |
2226 | VIRTNET_FAIL_ON(vdev, VIRTIO_NET_F_CTRL_VLAN, | |
2227 | "VIRTIO_NET_F_CTRL_VQ") || | |
2228 | VIRTNET_FAIL_ON(vdev, VIRTIO_NET_F_GUEST_ANNOUNCE, | |
2229 | "VIRTIO_NET_F_CTRL_VQ") || | |
2230 | VIRTNET_FAIL_ON(vdev, VIRTIO_NET_F_MQ, "VIRTIO_NET_F_CTRL_VQ") || | |
2231 | VIRTNET_FAIL_ON(vdev, VIRTIO_NET_F_CTRL_MAC_ADDR, | |
2232 | "VIRTIO_NET_F_CTRL_VQ"))) { | |
2233 | return false; | |
2234 | } | |
2235 | ||
2236 | return true; | |
2237 | } | |
2238 | ||
d0c2c997 JW |
2239 | #define MIN_MTU ETH_MIN_MTU |
2240 | #define MAX_MTU ETH_MAX_MTU | |
2241 | ||
296f96fc RR |
2242 | static int virtnet_probe(struct virtio_device *vdev) |
2243 | { | |
986a4f4d | 2244 | int i, err; |
296f96fc RR |
2245 | struct net_device *dev; |
2246 | struct virtnet_info *vi; | |
986a4f4d | 2247 | u16 max_queue_pairs; |
14de9d11 | 2248 | int mtu; |
986a4f4d | 2249 | |
6ba42248 MT |
2250 | if (!vdev->config->get) { |
2251 | dev_err(&vdev->dev, "%s failure: config access disabled\n", | |
2252 | __func__); | |
2253 | return -EINVAL; | |
2254 | } | |
2255 | ||
892d6eb1 JW |
2256 | if (!virtnet_validate_features(vdev)) |
2257 | return -EINVAL; | |
2258 | ||
986a4f4d | 2259 | /* Find if host supports multiqueue virtio_net device */ |
855e0c52 RR |
2260 | err = virtio_cread_feature(vdev, VIRTIO_NET_F_MQ, |
2261 | struct virtio_net_config, | |
2262 | max_virtqueue_pairs, &max_queue_pairs); | |
986a4f4d JW |
2263 | |
2264 | /* We need at least 2 queue's */ | |
2265 | if (err || max_queue_pairs < VIRTIO_NET_CTRL_MQ_VQ_PAIRS_MIN || | |
2266 | max_queue_pairs > VIRTIO_NET_CTRL_MQ_VQ_PAIRS_MAX || | |
2267 | !virtio_has_feature(vdev, VIRTIO_NET_F_CTRL_VQ)) | |
2268 | max_queue_pairs = 1; | |
296f96fc RR |
2269 | |
2270 | /* Allocate ourselves a network device with room for our info */ | |
986a4f4d | 2271 | dev = alloc_etherdev_mq(sizeof(struct virtnet_info), max_queue_pairs); |
296f96fc RR |
2272 | if (!dev) |
2273 | return -ENOMEM; | |
2274 | ||
2275 | /* Set up network device as normal. */ | |
f2f2c8b4 | 2276 | dev->priv_flags |= IFF_UNICAST_FLT | IFF_LIVE_ADDR_CHANGE; |
76288b4e | 2277 | dev->netdev_ops = &virtnet_netdev; |
296f96fc | 2278 | dev->features = NETIF_F_HIGHDMA; |
3fa2a1df | 2279 | |
7ad24ea4 | 2280 | dev->ethtool_ops = &virtnet_ethtool_ops; |
296f96fc RR |
2281 | SET_NETDEV_DEV(dev, &vdev->dev); |
2282 | ||
2283 | /* Do we support "hardware" checksums? */ | |
98e778c9 | 2284 | if (virtio_has_feature(vdev, VIRTIO_NET_F_CSUM)) { |
296f96fc | 2285 | /* This opens up the world of extra features. */ |
48900cb6 | 2286 | dev->hw_features |= NETIF_F_HW_CSUM | NETIF_F_SG; |
98e778c9 | 2287 | if (csum) |
48900cb6 | 2288 | dev->features |= NETIF_F_HW_CSUM | NETIF_F_SG; |
98e778c9 MM |
2289 | |
2290 | if (virtio_has_feature(vdev, VIRTIO_NET_F_GSO)) { | |
e3e3c423 | 2291 | dev->hw_features |= NETIF_F_TSO | NETIF_F_UFO |
34a48579 RR |
2292 | | NETIF_F_TSO_ECN | NETIF_F_TSO6; |
2293 | } | |
5539ae96 | 2294 | /* Individual feature bits: what can host handle? */ |
98e778c9 MM |
2295 | if (virtio_has_feature(vdev, VIRTIO_NET_F_HOST_TSO4)) |
2296 | dev->hw_features |= NETIF_F_TSO; | |
2297 | if (virtio_has_feature(vdev, VIRTIO_NET_F_HOST_TSO6)) | |
2298 | dev->hw_features |= NETIF_F_TSO6; | |
2299 | if (virtio_has_feature(vdev, VIRTIO_NET_F_HOST_ECN)) | |
2300 | dev->hw_features |= NETIF_F_TSO_ECN; | |
e3e3c423 VY |
2301 | if (virtio_has_feature(vdev, VIRTIO_NET_F_HOST_UFO)) |
2302 | dev->hw_features |= NETIF_F_UFO; | |
98e778c9 | 2303 | |
41f2f127 JW |
2304 | dev->features |= NETIF_F_GSO_ROBUST; |
2305 | ||
98e778c9 | 2306 | if (gso) |
e3e3c423 | 2307 | dev->features |= dev->hw_features & (NETIF_F_ALL_TSO|NETIF_F_UFO); |
98e778c9 | 2308 | /* (!csum && gso) case will be fixed by register_netdev() */ |
296f96fc | 2309 | } |
4f49129b TH |
2310 | if (virtio_has_feature(vdev, VIRTIO_NET_F_GUEST_CSUM)) |
2311 | dev->features |= NETIF_F_RXCSUM; | |
296f96fc | 2312 | |
4fda8302 JW |
2313 | dev->vlan_features = dev->features; |
2314 | ||
d0c2c997 JW |
2315 | /* MTU range: 68 - 65535 */ |
2316 | dev->min_mtu = MIN_MTU; | |
2317 | dev->max_mtu = MAX_MTU; | |
2318 | ||
296f96fc | 2319 | /* Configuration may specify what MAC to use. Otherwise random. */ |
855e0c52 RR |
2320 | if (virtio_has_feature(vdev, VIRTIO_NET_F_MAC)) |
2321 | virtio_cread_bytes(vdev, | |
2322 | offsetof(struct virtio_net_config, mac), | |
2323 | dev->dev_addr, dev->addr_len); | |
2324 | else | |
f2cedb63 | 2325 | eth_hw_addr_random(dev); |
296f96fc RR |
2326 | |
2327 | /* Set up our device-specific information */ | |
2328 | vi = netdev_priv(dev); | |
296f96fc RR |
2329 | vi->dev = dev; |
2330 | vi->vdev = vdev; | |
d9d5dcc8 | 2331 | vdev->priv = vi; |
3fa2a1df | 2332 | vi->stats = alloc_percpu(struct virtnet_stats); |
2333 | err = -ENOMEM; | |
2334 | if (vi->stats == NULL) | |
2335 | goto free; | |
2336 | ||
827da44c JS |
2337 | for_each_possible_cpu(i) { |
2338 | struct virtnet_stats *virtnet_stats; | |
2339 | virtnet_stats = per_cpu_ptr(vi->stats, i); | |
2340 | u64_stats_init(&virtnet_stats->tx_syncp); | |
2341 | u64_stats_init(&virtnet_stats->rx_syncp); | |
2342 | } | |
2343 | ||
586d17c5 | 2344 | INIT_WORK(&vi->config_work, virtnet_config_changed_work); |
296f96fc | 2345 | |
97402b96 | 2346 | /* If we can receive ANY GSO packets, we must allocate large ones. */ |
8e95a202 JP |
2347 | if (virtio_has_feature(vdev, VIRTIO_NET_F_GUEST_TSO4) || |
2348 | virtio_has_feature(vdev, VIRTIO_NET_F_GUEST_TSO6) || | |
e3e3c423 VY |
2349 | virtio_has_feature(vdev, VIRTIO_NET_F_GUEST_ECN) || |
2350 | virtio_has_feature(vdev, VIRTIO_NET_F_GUEST_UFO)) | |
97402b96 HX |
2351 | vi->big_packets = true; |
2352 | ||
3f2c31d9 MM |
2353 | if (virtio_has_feature(vdev, VIRTIO_NET_F_MRG_RXBUF)) |
2354 | vi->mergeable_rx_bufs = true; | |
2355 | ||
d04302b3 MT |
2356 | if (virtio_has_feature(vdev, VIRTIO_NET_F_MRG_RXBUF) || |
2357 | virtio_has_feature(vdev, VIRTIO_F_VERSION_1)) | |
012873d0 MT |
2358 | vi->hdr_len = sizeof(struct virtio_net_hdr_mrg_rxbuf); |
2359 | else | |
2360 | vi->hdr_len = sizeof(struct virtio_net_hdr); | |
2361 | ||
75993300 MT |
2362 | if (virtio_has_feature(vdev, VIRTIO_F_ANY_LAYOUT) || |
2363 | virtio_has_feature(vdev, VIRTIO_F_VERSION_1)) | |
e7428e95 MT |
2364 | vi->any_header_sg = true; |
2365 | ||
986a4f4d JW |
2366 | if (virtio_has_feature(vdev, VIRTIO_NET_F_CTRL_VQ)) |
2367 | vi->has_cvq = true; | |
2368 | ||
14de9d11 AC |
2369 | if (virtio_has_feature(vdev, VIRTIO_NET_F_MTU)) { |
2370 | mtu = virtio_cread16(vdev, | |
2371 | offsetof(struct virtio_net_config, | |
2372 | mtu)); | |
93a205ee | 2373 | if (mtu < dev->min_mtu) { |
14de9d11 | 2374 | __virtio_clear_bit(vdev, VIRTIO_NET_F_MTU); |
93a205ee | 2375 | } else { |
d0c2c997 | 2376 | dev->mtu = mtu; |
93a205ee AC |
2377 | dev->max_mtu = mtu; |
2378 | } | |
14de9d11 AC |
2379 | } |
2380 | ||
012873d0 MT |
2381 | if (vi->any_header_sg) |
2382 | dev->needed_headroom = vi->hdr_len; | |
6ebbc1a6 | 2383 | |
44900010 JW |
2384 | /* Enable multiqueue by default */ |
2385 | if (num_online_cpus() >= max_queue_pairs) | |
2386 | vi->curr_queue_pairs = max_queue_pairs; | |
2387 | else | |
2388 | vi->curr_queue_pairs = num_online_cpus(); | |
986a4f4d JW |
2389 | vi->max_queue_pairs = max_queue_pairs; |
2390 | ||
2391 | /* Allocate/initialize the rx/tx queues, and invoke find_vqs */ | |
3f9c10b0 | 2392 | err = init_vqs(vi); |
d2a7ddda | 2393 | if (err) |
9bb8ca86 | 2394 | goto free_stats; |
296f96fc | 2395 | |
fbf28d78 MD |
2396 | #ifdef CONFIG_SYSFS |
2397 | if (vi->mergeable_rx_bufs) | |
2398 | dev->sysfs_rx_queue_group = &virtio_net_mrg_rx_group; | |
2399 | #endif | |
0f13b66b ZYW |
2400 | netif_set_real_num_tx_queues(dev, vi->curr_queue_pairs); |
2401 | netif_set_real_num_rx_queues(dev, vi->curr_queue_pairs); | |
986a4f4d | 2402 | |
16032be5 NA |
2403 | virtnet_init_settings(dev); |
2404 | ||
296f96fc RR |
2405 | err = register_netdev(dev); |
2406 | if (err) { | |
2407 | pr_debug("virtio_net: registering device failed\n"); | |
d2a7ddda | 2408 | goto free_vqs; |
296f96fc | 2409 | } |
b3369c1f | 2410 | |
4baf1e33 MT |
2411 | virtio_device_ready(vdev); |
2412 | ||
8017c279 | 2413 | err = virtnet_cpu_notif_add(vi); |
8de4b2f3 WG |
2414 | if (err) { |
2415 | pr_debug("virtio_net: registering cpu notifier failed\n"); | |
f00e35e2 | 2416 | goto free_unregister_netdev; |
8de4b2f3 WG |
2417 | } |
2418 | ||
a220871b | 2419 | virtnet_set_queues(vi, vi->curr_queue_pairs); |
44900010 | 2420 | |
167c25e4 JW |
2421 | /* Assume link up if device can't report link status, |
2422 | otherwise get link status from config. */ | |
2423 | if (virtio_has_feature(vi->vdev, VIRTIO_NET_F_STATUS)) { | |
2424 | netif_carrier_off(dev); | |
3b07e9ca | 2425 | schedule_work(&vi->config_work); |
167c25e4 JW |
2426 | } else { |
2427 | vi->status = VIRTIO_NET_S_LINK_UP; | |
2428 | netif_carrier_on(dev); | |
2429 | } | |
9f4d26d0 | 2430 | |
986a4f4d JW |
2431 | pr_debug("virtnet: registered device %s with %d RX and TX vq's\n", |
2432 | dev->name, max_queue_pairs); | |
2433 | ||
296f96fc RR |
2434 | return 0; |
2435 | ||
f00e35e2 | 2436 | free_unregister_netdev: |
02465555 MT |
2437 | vi->vdev->config->reset(vdev); |
2438 | ||
b3369c1f | 2439 | unregister_netdev(dev); |
d2a7ddda | 2440 | free_vqs: |
986a4f4d | 2441 | cancel_delayed_work_sync(&vi->refill); |
fb51879d | 2442 | free_receive_page_frags(vi); |
e9d7417b | 2443 | virtnet_del_vqs(vi); |
3fa2a1df | 2444 | free_stats: |
2445 | free_percpu(vi->stats); | |
296f96fc RR |
2446 | free: |
2447 | free_netdev(dev); | |
2448 | return err; | |
2449 | } | |
2450 | ||
2de2f7f4 JF |
2451 | static void _remove_vq_common(struct virtnet_info *vi) |
2452 | { | |
2453 | vi->vdev->config->reset(vi->vdev); | |
2454 | free_unused_bufs(vi); | |
2455 | _free_receive_bufs(vi); | |
2456 | free_receive_page_frags(vi); | |
2457 | virtnet_del_vqs(vi); | |
2458 | } | |
2459 | ||
04486ed0 | 2460 | static void remove_vq_common(struct virtnet_info *vi) |
296f96fc | 2461 | { |
04486ed0 | 2462 | vi->vdev->config->reset(vi->vdev); |
830a8a97 SM |
2463 | |
2464 | /* Free unused buffers in both send and recv, if any. */ | |
9ab86bbc | 2465 | free_unused_bufs(vi); |
fb6813f4 | 2466 | |
986a4f4d | 2467 | free_receive_bufs(vi); |
d2a7ddda | 2468 | |
fb51879d MD |
2469 | free_receive_page_frags(vi); |
2470 | ||
986a4f4d | 2471 | virtnet_del_vqs(vi); |
04486ed0 AS |
2472 | } |
2473 | ||
8cc085d6 | 2474 | static void virtnet_remove(struct virtio_device *vdev) |
04486ed0 AS |
2475 | { |
2476 | struct virtnet_info *vi = vdev->priv; | |
2477 | ||
8017c279 | 2478 | virtnet_cpu_notif_remove(vi); |
8de4b2f3 | 2479 | |
102a2786 MT |
2480 | /* Make sure no work handler is accessing the device. */ |
2481 | flush_work(&vi->config_work); | |
586d17c5 | 2482 | |
04486ed0 AS |
2483 | unregister_netdev(vi->dev); |
2484 | ||
2485 | remove_vq_common(vi); | |
fb6813f4 | 2486 | |
2e66f55b | 2487 | free_percpu(vi->stats); |
74b2553f | 2488 | free_netdev(vi->dev); |
296f96fc RR |
2489 | } |
2490 | ||
89107000 | 2491 | #ifdef CONFIG_PM_SLEEP |
0741bcb5 AS |
2492 | static int virtnet_freeze(struct virtio_device *vdev) |
2493 | { | |
2494 | struct virtnet_info *vi = vdev->priv; | |
2495 | ||
8017c279 | 2496 | virtnet_cpu_notif_remove(vi); |
9fe7bfce | 2497 | virtnet_freeze_down(vdev); |
0741bcb5 AS |
2498 | remove_vq_common(vi); |
2499 | ||
2500 | return 0; | |
2501 | } | |
2502 | ||
2503 | static int virtnet_restore(struct virtio_device *vdev) | |
2504 | { | |
2505 | struct virtnet_info *vi = vdev->priv; | |
9fe7bfce | 2506 | int err; |
0741bcb5 | 2507 | |
9fe7bfce | 2508 | err = virtnet_restore_up(vdev); |
0741bcb5 AS |
2509 | if (err) |
2510 | return err; | |
986a4f4d JW |
2511 | virtnet_set_queues(vi, vi->curr_queue_pairs); |
2512 | ||
8017c279 | 2513 | err = virtnet_cpu_notif_add(vi); |
ec9debbd JW |
2514 | if (err) |
2515 | return err; | |
2516 | ||
0741bcb5 AS |
2517 | return 0; |
2518 | } | |
2519 | #endif | |
2520 | ||
296f96fc RR |
2521 | static struct virtio_device_id id_table[] = { |
2522 | { VIRTIO_ID_NET, VIRTIO_DEV_ANY_ID }, | |
2523 | { 0 }, | |
2524 | }; | |
2525 | ||
f3358507 MT |
2526 | #define VIRTNET_FEATURES \ |
2527 | VIRTIO_NET_F_CSUM, VIRTIO_NET_F_GUEST_CSUM, \ | |
2528 | VIRTIO_NET_F_MAC, \ | |
2529 | VIRTIO_NET_F_HOST_TSO4, VIRTIO_NET_F_HOST_UFO, VIRTIO_NET_F_HOST_TSO6, \ | |
2530 | VIRTIO_NET_F_HOST_ECN, VIRTIO_NET_F_GUEST_TSO4, VIRTIO_NET_F_GUEST_TSO6, \ | |
2531 | VIRTIO_NET_F_GUEST_ECN, VIRTIO_NET_F_GUEST_UFO, \ | |
2532 | VIRTIO_NET_F_MRG_RXBUF, VIRTIO_NET_F_STATUS, VIRTIO_NET_F_CTRL_VQ, \ | |
2533 | VIRTIO_NET_F_CTRL_RX, VIRTIO_NET_F_CTRL_VLAN, \ | |
2534 | VIRTIO_NET_F_GUEST_ANNOUNCE, VIRTIO_NET_F_MQ, \ | |
2535 | VIRTIO_NET_F_CTRL_MAC_ADDR, \ | |
2536 | VIRTIO_NET_F_MTU | |
2537 | ||
c45a6816 | 2538 | static unsigned int features[] = { |
f3358507 MT |
2539 | VIRTNET_FEATURES, |
2540 | }; | |
2541 | ||
2542 | static unsigned int features_legacy[] = { | |
2543 | VIRTNET_FEATURES, | |
2544 | VIRTIO_NET_F_GSO, | |
e7428e95 | 2545 | VIRTIO_F_ANY_LAYOUT, |
c45a6816 RR |
2546 | }; |
2547 | ||
22402529 | 2548 | static struct virtio_driver virtio_net_driver = { |
c45a6816 RR |
2549 | .feature_table = features, |
2550 | .feature_table_size = ARRAY_SIZE(features), | |
f3358507 MT |
2551 | .feature_table_legacy = features_legacy, |
2552 | .feature_table_size_legacy = ARRAY_SIZE(features_legacy), | |
296f96fc RR |
2553 | .driver.name = KBUILD_MODNAME, |
2554 | .driver.owner = THIS_MODULE, | |
2555 | .id_table = id_table, | |
2556 | .probe = virtnet_probe, | |
8cc085d6 | 2557 | .remove = virtnet_remove, |
9f4d26d0 | 2558 | .config_changed = virtnet_config_changed, |
89107000 | 2559 | #ifdef CONFIG_PM_SLEEP |
0741bcb5 AS |
2560 | .freeze = virtnet_freeze, |
2561 | .restore = virtnet_restore, | |
2562 | #endif | |
296f96fc RR |
2563 | }; |
2564 | ||
8017c279 SAS |
2565 | static __init int virtio_net_driver_init(void) |
2566 | { | |
2567 | int ret; | |
2568 | ||
73c1b41e | 2569 | ret = cpuhp_setup_state_multi(CPUHP_AP_ONLINE_DYN, "virtio/net:online", |
8017c279 SAS |
2570 | virtnet_cpu_online, |
2571 | virtnet_cpu_down_prep); | |
2572 | if (ret < 0) | |
2573 | goto out; | |
2574 | virtionet_online = ret; | |
73c1b41e | 2575 | ret = cpuhp_setup_state_multi(CPUHP_VIRT_NET_DEAD, "virtio/net:dead", |
8017c279 SAS |
2576 | NULL, virtnet_cpu_dead); |
2577 | if (ret) | |
2578 | goto err_dead; | |
2579 | ||
2580 | ret = register_virtio_driver(&virtio_net_driver); | |
2581 | if (ret) | |
2582 | goto err_virtio; | |
2583 | return 0; | |
2584 | err_virtio: | |
2585 | cpuhp_remove_multi_state(CPUHP_VIRT_NET_DEAD); | |
2586 | err_dead: | |
2587 | cpuhp_remove_multi_state(virtionet_online); | |
2588 | out: | |
2589 | return ret; | |
2590 | } | |
2591 | module_init(virtio_net_driver_init); | |
2592 | ||
2593 | static __exit void virtio_net_driver_exit(void) | |
2594 | { | |
2595 | cpuhp_remove_multi_state(CPUHP_VIRT_NET_DEAD); | |
2596 | cpuhp_remove_multi_state(virtionet_online); | |
2597 | unregister_virtio_driver(&virtio_net_driver); | |
2598 | } | |
2599 | module_exit(virtio_net_driver_exit); | |
296f96fc RR |
2600 | |
2601 | MODULE_DEVICE_TABLE(virtio, id_table); | |
2602 | MODULE_DESCRIPTION("Virtio network driver"); | |
2603 | MODULE_LICENSE("GPL"); |