Commit | Line | Data |
---|---|---|
1ccea77e | 1 | // SPDX-License-Identifier: GPL-2.0-or-later |
48925e37 | 2 | /* A network driver using virtio. |
296f96fc RR |
3 | * |
4 | * Copyright 2007 Rusty Russell <rusty@rustcorp.com.au> IBM Corporation | |
296f96fc RR |
5 | */ |
6 | //#define DEBUG | |
7 | #include <linux/netdevice.h> | |
8 | #include <linux/etherdevice.h> | |
a9ea3fc6 | 9 | #include <linux/ethtool.h> |
296f96fc RR |
10 | #include <linux/module.h> |
11 | #include <linux/virtio.h> | |
12 | #include <linux/virtio_net.h> | |
f600b690 | 13 | #include <linux/bpf.h> |
a67edbf4 | 14 | #include <linux/bpf_trace.h> |
296f96fc | 15 | #include <linux/scatterlist.h> |
e918085a | 16 | #include <linux/if_vlan.h> |
5a0e3ad6 | 17 | #include <linux/slab.h> |
8de4b2f3 | 18 | #include <linux/cpu.h> |
ab7db917 | 19 | #include <linux/average.h> |
186b3c99 | 20 | #include <linux/filter.h> |
2ca653d6 | 21 | #include <linux/kernel.h> |
d85b758f | 22 | #include <net/route.h> |
754b8a21 | 23 | #include <net/xdp.h> |
ba5e4426 | 24 | #include <net/net_failover.h> |
296f96fc | 25 | |
d34710e3 | 26 | static int napi_weight = NAPI_POLL_WEIGHT; |
6c0cd7c0 DL |
27 | module_param(napi_weight, int, 0444); |
28 | ||
31c03aef | 29 | static bool csum = true, gso = true, napi_tx = true; |
34a48579 RR |
30 | module_param(csum, bool, 0444); |
31 | module_param(gso, bool, 0444); | |
b92f1e67 | 32 | module_param(napi_tx, bool, 0644); |
34a48579 | 33 | |
296f96fc | 34 | /* FIXME: MTU in config. */ |
5061de36 | 35 | #define GOOD_PACKET_LEN (ETH_HLEN + VLAN_HLEN + ETH_DATA_LEN) |
3f2c31d9 | 36 | #define GOOD_COPY_LEN 128 |
296f96fc | 37 | |
f6b10209 JW |
38 | #define VIRTNET_RX_PAD (NET_IP_ALIGN + NET_SKB_PAD) |
39 | ||
2de2f7f4 JF |
40 | /* Amount of XDP headroom to prepend to packets for use by xdp_adjust_head */ |
41 | #define VIRTIO_XDP_HEADROOM 256 | |
42 | ||
2471c75e JDB |
43 | /* Separating two types of XDP xmit */ |
44 | #define VIRTIO_XDP_TX BIT(0) | |
45 | #define VIRTIO_XDP_REDIR BIT(1) | |
46 | ||
5050471d TM |
47 | #define VIRTIO_XDP_FLAG BIT(0) |
48 | ||
5377d758 JB |
49 | /* RX packet size EWMA. The average packet size is used to determine the packet |
50 | * buffer size when refilling RX rings. As the entire RX ring may be refilled | |
51 | * at once, the weight is chosen so that the EWMA will be insensitive to short- | |
52 | * term, transient changes in packet size. | |
ab7db917 | 53 | */ |
eb1e011a | 54 | DECLARE_EWMA(pkt_len, 0, 64) |
ab7db917 | 55 | |
66846048 | 56 | #define VIRTNET_DRIVER_VERSION "1.0.0" |
2a41f71d | 57 | |
7acd4329 CIK |
58 | static const unsigned long guest_offloads[] = { |
59 | VIRTIO_NET_F_GUEST_TSO4, | |
60 | VIRTIO_NET_F_GUEST_TSO6, | |
61 | VIRTIO_NET_F_GUEST_ECN, | |
e59ff2c4 | 62 | VIRTIO_NET_F_GUEST_UFO, |
418044e1 AM |
63 | VIRTIO_NET_F_GUEST_CSUM, |
64 | VIRTIO_NET_F_GUEST_USO4, | |
be50da3e JP |
65 | VIRTIO_NET_F_GUEST_USO6, |
66 | VIRTIO_NET_F_GUEST_HDRLEN | |
7acd4329 | 67 | }; |
3f93522f | 68 | |
dbcf24d1 | 69 | #define GUEST_OFFLOAD_GRO_HW_MASK ((1ULL << VIRTIO_NET_F_GUEST_TSO4) | \ |
1a03b8a3 TZ |
70 | (1ULL << VIRTIO_NET_F_GUEST_TSO6) | \ |
71 | (1ULL << VIRTIO_NET_F_GUEST_ECN) | \ | |
418044e1 AM |
72 | (1ULL << VIRTIO_NET_F_GUEST_UFO) | \ |
73 | (1ULL << VIRTIO_NET_F_GUEST_USO4) | \ | |
74 | (1ULL << VIRTIO_NET_F_GUEST_USO6)) | |
1a03b8a3 | 75 | |
d7dfc5cf TM |
76 | struct virtnet_stat_desc { |
77 | char desc[ETH_GSTRING_LEN]; | |
78 | size_t offset; | |
3fa2a1df | 79 | }; |
80 | ||
d7dfc5cf TM |
81 | struct virtnet_sq_stats { |
82 | struct u64_stats_sync syncp; | |
83 | u64 packets; | |
84 | u64 bytes; | |
5b8f3c8d TM |
85 | u64 xdp_tx; |
86 | u64 xdp_tx_drops; | |
461f03dc | 87 | u64 kicks; |
a520794b | 88 | u64 tx_timeouts; |
d7dfc5cf TM |
89 | }; |
90 | ||
d46eeeaf JW |
91 | struct virtnet_rq_stats { |
92 | struct u64_stats_sync syncp; | |
d7dfc5cf TM |
93 | u64 packets; |
94 | u64 bytes; | |
2c4a2f7d | 95 | u64 drops; |
5b8f3c8d TM |
96 | u64 xdp_packets; |
97 | u64 xdp_tx; | |
98 | u64 xdp_redirects; | |
99 | u64 xdp_drops; | |
461f03dc | 100 | u64 kicks; |
d7dfc5cf TM |
101 | }; |
102 | ||
103 | #define VIRTNET_SQ_STAT(m) offsetof(struct virtnet_sq_stats, m) | |
d46eeeaf | 104 | #define VIRTNET_RQ_STAT(m) offsetof(struct virtnet_rq_stats, m) |
d7dfc5cf TM |
105 | |
106 | static const struct virtnet_stat_desc virtnet_sq_stats_desc[] = { | |
5b8f3c8d TM |
107 | { "packets", VIRTNET_SQ_STAT(packets) }, |
108 | { "bytes", VIRTNET_SQ_STAT(bytes) }, | |
109 | { "xdp_tx", VIRTNET_SQ_STAT(xdp_tx) }, | |
110 | { "xdp_tx_drops", VIRTNET_SQ_STAT(xdp_tx_drops) }, | |
461f03dc | 111 | { "kicks", VIRTNET_SQ_STAT(kicks) }, |
a520794b | 112 | { "tx_timeouts", VIRTNET_SQ_STAT(tx_timeouts) }, |
d7dfc5cf TM |
113 | }; |
114 | ||
115 | static const struct virtnet_stat_desc virtnet_rq_stats_desc[] = { | |
5b8f3c8d TM |
116 | { "packets", VIRTNET_RQ_STAT(packets) }, |
117 | { "bytes", VIRTNET_RQ_STAT(bytes) }, | |
118 | { "drops", VIRTNET_RQ_STAT(drops) }, | |
119 | { "xdp_packets", VIRTNET_RQ_STAT(xdp_packets) }, | |
120 | { "xdp_tx", VIRTNET_RQ_STAT(xdp_tx) }, | |
121 | { "xdp_redirects", VIRTNET_RQ_STAT(xdp_redirects) }, | |
122 | { "xdp_drops", VIRTNET_RQ_STAT(xdp_drops) }, | |
461f03dc | 123 | { "kicks", VIRTNET_RQ_STAT(kicks) }, |
d7dfc5cf TM |
124 | }; |
125 | ||
126 | #define VIRTNET_SQ_STATS_LEN ARRAY_SIZE(virtnet_sq_stats_desc) | |
127 | #define VIRTNET_RQ_STATS_LEN ARRAY_SIZE(virtnet_rq_stats_desc) | |
128 | ||
e9d7417b JW |
129 | /* Internal representation of a send virtqueue */ |
130 | struct send_queue { | |
131 | /* Virtqueue associated with this send _queue */ | |
132 | struct virtqueue *vq; | |
133 | ||
134 | /* TX: fragments + linear part + virtio header */ | |
135 | struct scatterlist sg[MAX_SKB_FRAGS + 2]; | |
986a4f4d JW |
136 | |
137 | /* Name of the send queue: output.$index */ | |
d0671115 | 138 | char name[16]; |
b92f1e67 | 139 | |
d7dfc5cf TM |
140 | struct virtnet_sq_stats stats; |
141 | ||
b92f1e67 | 142 | struct napi_struct napi; |
ebcce492 XZ |
143 | |
144 | /* Record whether sq is in reset state. */ | |
145 | bool reset; | |
e9d7417b JW |
146 | }; |
147 | ||
148 | /* Internal representation of a receive virtqueue */ | |
149 | struct receive_queue { | |
150 | /* Virtqueue associated with this receive_queue */ | |
151 | struct virtqueue *vq; | |
152 | ||
296f96fc RR |
153 | struct napi_struct napi; |
154 | ||
f600b690 JF |
155 | struct bpf_prog __rcu *xdp_prog; |
156 | ||
d7dfc5cf TM |
157 | struct virtnet_rq_stats stats; |
158 | ||
e9d7417b JW |
159 | /* Chain pages by the private ptr. */ |
160 | struct page *pages; | |
161 | ||
ab7db917 | 162 | /* Average packet length for mergeable receive buffers. */ |
5377d758 | 163 | struct ewma_pkt_len mrg_avg_pkt_len; |
ab7db917 | 164 | |
fb51879d MD |
165 | /* Page frag for packet buffer allocation. */ |
166 | struct page_frag alloc_frag; | |
167 | ||
e9d7417b JW |
168 | /* RX: fragments + linear part + virtio header */ |
169 | struct scatterlist sg[MAX_SKB_FRAGS + 2]; | |
986a4f4d | 170 | |
d85b758f MT |
171 | /* Min single buffer size for mergeable buffers case. */ |
172 | unsigned int min_buf_len; | |
173 | ||
986a4f4d | 174 | /* Name of this receive queue: input.$index */ |
d0671115 | 175 | char name[16]; |
754b8a21 JDB |
176 | |
177 | struct xdp_rxq_info xdp_rxq; | |
e9d7417b JW |
178 | }; |
179 | ||
c7114b12 AM |
180 | /* This structure can contain rss message with maximum settings for indirection table and keysize |
181 | * Note, that default structure that describes RSS configuration virtio_net_rss_config | |
182 | * contains same info but can't handle table values. | |
183 | * In any case, structure would be passed to virtio hw through sg_buf split by parts | |
184 | * because table sizes may be differ according to the device configuration. | |
185 | */ | |
186 | #define VIRTIO_NET_RSS_MAX_KEY_SIZE 40 | |
187 | #define VIRTIO_NET_RSS_MAX_TABLE_LEN 128 | |
188 | struct virtio_net_ctrl_rss { | |
189 | u32 hash_types; | |
190 | u16 indirection_table_mask; | |
191 | u16 unclassified_queue; | |
192 | u16 indirection_table[VIRTIO_NET_RSS_MAX_TABLE_LEN]; | |
193 | u16 max_tx_vq; | |
194 | u8 hash_key_length; | |
195 | u8 key[VIRTIO_NET_RSS_MAX_KEY_SIZE]; | |
196 | }; | |
197 | ||
12e57169 MT |
198 | /* Control VQ buffers: protected by the rtnl lock */ |
199 | struct control_buf { | |
200 | struct virtio_net_ctrl_hdr hdr; | |
201 | virtio_net_ctrl_ack status; | |
202 | struct virtio_net_ctrl_mq mq; | |
203 | u8 promisc; | |
204 | u8 allmulti; | |
d7fad4c8 | 205 | __virtio16 vid; |
f4ee703a | 206 | __virtio64 offloads; |
c7114b12 | 207 | struct virtio_net_ctrl_rss rss; |
12e57169 MT |
208 | }; |
209 | ||
e9d7417b JW |
210 | struct virtnet_info { |
211 | struct virtio_device *vdev; | |
212 | struct virtqueue *cvq; | |
213 | struct net_device *dev; | |
986a4f4d JW |
214 | struct send_queue *sq; |
215 | struct receive_queue *rq; | |
e9d7417b JW |
216 | unsigned int status; |
217 | ||
986a4f4d JW |
218 | /* Max # of queue pairs supported by the device */ |
219 | u16 max_queue_pairs; | |
220 | ||
221 | /* # of queue pairs currently used by the driver */ | |
222 | u16 curr_queue_pairs; | |
223 | ||
672aafd5 JF |
224 | /* # of XDP queue pairs currently used by the driver */ |
225 | u16 xdp_queue_pairs; | |
226 | ||
97c2c69e XZ |
227 | /* xdp_queue_pairs may be 0, when xdp is already loaded. So add this. */ |
228 | bool xdp_enabled; | |
229 | ||
97402b96 HX |
230 | /* I like... big packets and I cannot lie! */ |
231 | bool big_packets; | |
232 | ||
4959aebb GL |
233 | /* number of sg entries allocated for big packets */ |
234 | unsigned int big_packets_num_skbfrags; | |
235 | ||
3f2c31d9 MM |
236 | /* Host will merge rx buffers for big packets (shake it! shake it!) */ |
237 | bool mergeable_rx_bufs; | |
238 | ||
c7114b12 AM |
239 | /* Host supports rss and/or hash report */ |
240 | bool has_rss; | |
91f41f01 | 241 | bool has_rss_hash_report; |
c7114b12 AM |
242 | u8 rss_key_size; |
243 | u16 rss_indir_table_size; | |
244 | u32 rss_hash_types_supported; | |
c1170820 | 245 | u32 rss_hash_types_saved; |
c7114b12 | 246 | |
986a4f4d JW |
247 | /* Has control virtqueue */ |
248 | bool has_cvq; | |
249 | ||
e7428e95 MT |
250 | /* Host can handle any s/g split between our header and packet data */ |
251 | bool any_header_sg; | |
252 | ||
012873d0 MT |
253 | /* Packet virtio header size */ |
254 | u8 hdr_len; | |
255 | ||
5a159128 | 256 | /* Work struct for delayed refilling if we run low on memory. */ |
3161e453 RR |
257 | struct delayed_work refill; |
258 | ||
5a159128 JW |
259 | /* Is delayed refill enabled? */ |
260 | bool refill_enabled; | |
261 | ||
262 | /* The lock to synchronize the access to refill_enabled */ | |
263 | spinlock_t refill_lock; | |
264 | ||
586d17c5 JW |
265 | /* Work struct for config space updates */ |
266 | struct work_struct config_work; | |
267 | ||
986a4f4d JW |
268 | /* Does the affinity hint is set for virtqueues? */ |
269 | bool affinity_hint_set; | |
47be2479 | 270 | |
8017c279 SAS |
271 | /* CPU hotplug instances for online & dead */ |
272 | struct hlist_node node; | |
273 | struct hlist_node node_dead; | |
2ac46030 | 274 | |
12e57169 | 275 | struct control_buf *ctrl; |
16032be5 NA |
276 | |
277 | /* Ethtool settings */ | |
278 | u8 duplex; | |
279 | u32 speed; | |
3f93522f | 280 | |
699b045a AK |
281 | /* Interrupt coalescing settings */ |
282 | u32 tx_usecs; | |
283 | u32 rx_usecs; | |
284 | u32 tx_max_packets; | |
285 | u32 rx_max_packets; | |
286 | ||
3f93522f | 287 | unsigned long guest_offloads; |
a02e8964 | 288 | unsigned long guest_offloads_capable; |
ba5e4426 SS |
289 | |
290 | /* failover when STANDBY feature enabled */ | |
291 | struct failover *failover; | |
296f96fc RR |
292 | }; |
293 | ||
9ab86bbc | 294 | struct padded_vnet_hdr { |
c1ddc42d | 295 | struct virtio_net_hdr_v1_hash hdr; |
9ab86bbc | 296 | /* |
012873d0 MT |
297 | * hdr is in a separate sg buffer, and data sg buffer shares same page |
298 | * with this header sg. This padding makes next sg 16 byte aligned | |
299 | * after the header. | |
9ab86bbc | 300 | */ |
c1ddc42d | 301 | char padding[12]; |
9ab86bbc SM |
302 | }; |
303 | ||
6a4763e2 | 304 | static void virtnet_rq_free_unused_buf(struct virtqueue *vq, void *buf); |
ebcce492 | 305 | static void virtnet_sq_free_unused_buf(struct virtqueue *vq, void *buf); |
6a4763e2 | 306 | |
5050471d TM |
307 | static bool is_xdp_frame(void *ptr) |
308 | { | |
309 | return (unsigned long)ptr & VIRTIO_XDP_FLAG; | |
310 | } | |
311 | ||
312 | static void *xdp_to_ptr(struct xdp_frame *ptr) | |
313 | { | |
314 | return (void *)((unsigned long)ptr | VIRTIO_XDP_FLAG); | |
315 | } | |
316 | ||
317 | static struct xdp_frame *ptr_to_xdp(void *ptr) | |
318 | { | |
319 | return (struct xdp_frame *)((unsigned long)ptr & ~VIRTIO_XDP_FLAG); | |
320 | } | |
321 | ||
986a4f4d JW |
322 | /* Converting between virtqueue no. and kernel tx/rx queue no. |
323 | * 0:rx0 1:tx0 2:rx1 3:tx1 ... 2N:rxN 2N+1:txN 2N+2:cvq | |
324 | */ | |
325 | static int vq2txq(struct virtqueue *vq) | |
326 | { | |
9d0ca6ed | 327 | return (vq->index - 1) / 2; |
986a4f4d JW |
328 | } |
329 | ||
330 | static int txq2vq(int txq) | |
331 | { | |
332 | return txq * 2 + 1; | |
333 | } | |
334 | ||
335 | static int vq2rxq(struct virtqueue *vq) | |
336 | { | |
9d0ca6ed | 337 | return vq->index / 2; |
986a4f4d JW |
338 | } |
339 | ||
340 | static int rxq2vq(int rxq) | |
341 | { | |
342 | return rxq * 2; | |
343 | } | |
344 | ||
012873d0 | 345 | static inline struct virtio_net_hdr_mrg_rxbuf *skb_vnet_hdr(struct sk_buff *skb) |
296f96fc | 346 | { |
012873d0 | 347 | return (struct virtio_net_hdr_mrg_rxbuf *)skb->cb; |
296f96fc RR |
348 | } |
349 | ||
9ab86bbc SM |
350 | /* |
351 | * private is used to chain pages for big packets, put the whole | |
352 | * most recent used list in the beginning for reuse | |
353 | */ | |
e9d7417b | 354 | static void give_pages(struct receive_queue *rq, struct page *page) |
0a888fd1 | 355 | { |
9ab86bbc | 356 | struct page *end; |
0a888fd1 | 357 | |
e9d7417b | 358 | /* Find end of list, sew whole thing into vi->rq.pages. */ |
9ab86bbc | 359 | for (end = page; end->private; end = (struct page *)end->private); |
e9d7417b JW |
360 | end->private = (unsigned long)rq->pages; |
361 | rq->pages = page; | |
0a888fd1 MM |
362 | } |
363 | ||
e9d7417b | 364 | static struct page *get_a_page(struct receive_queue *rq, gfp_t gfp_mask) |
fb6813f4 | 365 | { |
e9d7417b | 366 | struct page *p = rq->pages; |
fb6813f4 | 367 | |
9ab86bbc | 368 | if (p) { |
e9d7417b | 369 | rq->pages = (struct page *)p->private; |
9ab86bbc SM |
370 | /* clear private here, it is used to chain pages */ |
371 | p->private = 0; | |
372 | } else | |
fb6813f4 RR |
373 | p = alloc_page(gfp_mask); |
374 | return p; | |
375 | } | |
376 | ||
5a159128 JW |
377 | static void enable_delayed_refill(struct virtnet_info *vi) |
378 | { | |
379 | spin_lock_bh(&vi->refill_lock); | |
380 | vi->refill_enabled = true; | |
381 | spin_unlock_bh(&vi->refill_lock); | |
382 | } | |
383 | ||
384 | static void disable_delayed_refill(struct virtnet_info *vi) | |
385 | { | |
386 | spin_lock_bh(&vi->refill_lock); | |
387 | vi->refill_enabled = false; | |
388 | spin_unlock_bh(&vi->refill_lock); | |
389 | } | |
390 | ||
e4e8452a WB |
391 | static void virtqueue_napi_schedule(struct napi_struct *napi, |
392 | struct virtqueue *vq) | |
393 | { | |
394 | if (napi_schedule_prep(napi)) { | |
395 | virtqueue_disable_cb(vq); | |
396 | __napi_schedule(napi); | |
397 | } | |
398 | } | |
399 | ||
400 | static void virtqueue_napi_complete(struct napi_struct *napi, | |
401 | struct virtqueue *vq, int processed) | |
402 | { | |
403 | int opaque; | |
404 | ||
405 | opaque = virtqueue_enable_cb_prepare(vq); | |
fdaa767a TM |
406 | if (napi_complete_done(napi, processed)) { |
407 | if (unlikely(virtqueue_poll(vq, opaque))) | |
408 | virtqueue_napi_schedule(napi, vq); | |
409 | } else { | |
410 | virtqueue_disable_cb(vq); | |
411 | } | |
e4e8452a WB |
412 | } |
413 | ||
e9d7417b | 414 | static void skb_xmit_done(struct virtqueue *vq) |
296f96fc | 415 | { |
e9d7417b | 416 | struct virtnet_info *vi = vq->vdev->priv; |
b92f1e67 | 417 | struct napi_struct *napi = &vi->sq[vq2txq(vq)].napi; |
296f96fc | 418 | |
2cb9c6ba | 419 | /* Suppress further interrupts. */ |
e9d7417b | 420 | virtqueue_disable_cb(vq); |
11a3a154 | 421 | |
b92f1e67 WB |
422 | if (napi->weight) |
423 | virtqueue_napi_schedule(napi, vq); | |
424 | else | |
425 | /* We were probably waiting for more output buffers. */ | |
426 | netif_wake_subqueue(vi->dev, vq2txq(vq)); | |
296f96fc RR |
427 | } |
428 | ||
28b39bc7 JW |
429 | #define MRG_CTX_HEADER_SHIFT 22 |
430 | static void *mergeable_len_to_ctx(unsigned int truesize, | |
431 | unsigned int headroom) | |
432 | { | |
433 | return (void *)(unsigned long)((headroom << MRG_CTX_HEADER_SHIFT) | truesize); | |
434 | } | |
435 | ||
436 | static unsigned int mergeable_ctx_to_headroom(void *mrg_ctx) | |
437 | { | |
438 | return (unsigned long)mrg_ctx >> MRG_CTX_HEADER_SHIFT; | |
439 | } | |
440 | ||
441 | static unsigned int mergeable_ctx_to_truesize(void *mrg_ctx) | |
442 | { | |
443 | return (unsigned long)mrg_ctx & ((1 << MRG_CTX_HEADER_SHIFT) - 1); | |
444 | } | |
445 | ||
3464645a | 446 | /* Called from bottom half context */ |
946fa564 MT |
447 | static struct sk_buff *page_to_skb(struct virtnet_info *vi, |
448 | struct receive_queue *rq, | |
2613af0e | 449 | struct page *page, unsigned int offset, |
fa0f1ba7 XZ |
450 | unsigned int len, unsigned int truesize, |
451 | unsigned int headroom) | |
9ab86bbc SM |
452 | { |
453 | struct sk_buff *skb; | |
012873d0 | 454 | struct virtio_net_hdr_mrg_rxbuf *hdr; |
2613af0e | 455 | unsigned int copy, hdr_len, hdr_padded_len; |
af39c8f7 | 456 | struct page *page_to_free = NULL; |
fb32856b | 457 | int tailroom, shinfo_size; |
f80bd740 | 458 | char *p, *hdr_p, *buf; |
fb6813f4 | 459 | |
2613af0e | 460 | p = page_address(page) + offset; |
fb32856b | 461 | hdr_p = p; |
3f2c31d9 | 462 | |
012873d0 MT |
463 | hdr_len = vi->hdr_len; |
464 | if (vi->mergeable_rx_bufs) | |
c1ddc42d | 465 | hdr_padded_len = hdr_len; |
012873d0 | 466 | else |
2613af0e | 467 | hdr_padded_len = sizeof(struct padded_vnet_hdr); |
3f2c31d9 | 468 | |
fa0f1ba7 | 469 | buf = p - headroom; |
9ab86bbc | 470 | len -= hdr_len; |
2613af0e MD |
471 | offset += hdr_padded_len; |
472 | p += hdr_padded_len; | |
fa0f1ba7 | 473 | tailroom = truesize - headroom - hdr_padded_len - len; |
3f2c31d9 | 474 | |
fb32856b XZ |
475 | shinfo_size = SKB_DATA_ALIGN(sizeof(struct skb_shared_info)); |
476 | ||
f80bd740 | 477 | /* copy small packet so we can reuse these pages */ |
f5d7872a | 478 | if (!NET_IP_ALIGN && len > GOOD_COPY_LEN && tailroom >= shinfo_size) { |
f80bd740 | 479 | skb = build_skb(buf, truesize); |
fb32856b XZ |
480 | if (unlikely(!skb)) |
481 | return NULL; | |
482 | ||
f80bd740 | 483 | skb_reserve(skb, p - buf); |
fb32856b | 484 | skb_put(skb, len); |
afd92d82 JW |
485 | |
486 | page = (struct page *)page->private; | |
487 | if (page) | |
488 | give_pages(rq, page); | |
fb32856b XZ |
489 | goto ok; |
490 | } | |
491 | ||
492 | /* copy small packet so we can reuse these pages for small data */ | |
493 | skb = napi_alloc_skb(&rq->napi, GOOD_COPY_LEN); | |
494 | if (unlikely(!skb)) | |
495 | return NULL; | |
496 | ||
0f6925b3 ED |
497 | /* Copy all frame if it fits skb->head, otherwise |
498 | * we let virtio_net_hdr_to_skb() and GRO pull headers as needed. | |
499 | */ | |
500 | if (len <= skb_tailroom(skb)) | |
501 | copy = len; | |
502 | else | |
18117a84 | 503 | copy = ETH_HLEN; |
59ae1d12 | 504 | skb_put_data(skb, p, copy); |
3f2c31d9 | 505 | |
9ab86bbc SM |
506 | len -= copy; |
507 | offset += copy; | |
3f2c31d9 | 508 | |
2613af0e MD |
509 | if (vi->mergeable_rx_bufs) { |
510 | if (len) | |
511 | skb_add_rx_frag(skb, 0, page, offset, len, truesize); | |
512 | else | |
af39c8f7 | 513 | page_to_free = page; |
fb32856b | 514 | goto ok; |
2613af0e MD |
515 | } |
516 | ||
e878d78b SL |
517 | /* |
518 | * Verify that we can indeed put this data into a skb. | |
519 | * This is here to handle cases when the device erroneously | |
520 | * tries to receive more than is possible. This is usually | |
521 | * the case of a broken device. | |
522 | */ | |
523 | if (unlikely(len > MAX_SKB_FRAGS * PAGE_SIZE)) { | |
be443899 | 524 | net_dbg_ratelimited("%s: too much data\n", skb->dev->name); |
e878d78b SL |
525 | dev_kfree_skb(skb); |
526 | return NULL; | |
527 | } | |
2613af0e | 528 | BUG_ON(offset >= PAGE_SIZE); |
9ab86bbc | 529 | while (len) { |
2613af0e MD |
530 | unsigned int frag_size = min((unsigned)PAGE_SIZE - offset, len); |
531 | skb_add_rx_frag(skb, skb_shinfo(skb)->nr_frags, page, offset, | |
532 | frag_size, truesize); | |
533 | len -= frag_size; | |
9ab86bbc SM |
534 | page = (struct page *)page->private; |
535 | offset = 0; | |
536 | } | |
3f2c31d9 | 537 | |
9ab86bbc | 538 | if (page) |
e9d7417b | 539 | give_pages(rq, page); |
3f2c31d9 | 540 | |
fb32856b | 541 | ok: |
18117a84 HQ |
542 | hdr = skb_vnet_hdr(skb); |
543 | memcpy(hdr, hdr_p, hdr_len); | |
af39c8f7 ED |
544 | if (page_to_free) |
545 | put_page(page_to_free); | |
fb32856b | 546 | |
9ab86bbc SM |
547 | return skb; |
548 | } | |
3f2c31d9 | 549 | |
25074a44 XZ |
550 | static void free_old_xmit_skbs(struct send_queue *sq, bool in_napi) |
551 | { | |
552 | unsigned int len; | |
553 | unsigned int packets = 0; | |
554 | unsigned int bytes = 0; | |
555 | void *ptr; | |
556 | ||
557 | while ((ptr = virtqueue_get_buf(sq->vq, &len)) != NULL) { | |
558 | if (likely(!is_xdp_frame(ptr))) { | |
559 | struct sk_buff *skb = ptr; | |
560 | ||
561 | pr_debug("Sent skb %p\n", skb); | |
562 | ||
563 | bytes += skb->len; | |
564 | napi_consume_skb(skb, in_napi); | |
565 | } else { | |
566 | struct xdp_frame *frame = ptr_to_xdp(ptr); | |
567 | ||
568 | bytes += xdp_get_frame_len(frame); | |
569 | xdp_return_frame(frame); | |
570 | } | |
571 | packets++; | |
572 | } | |
573 | ||
574 | /* Avoid overhead when no packets have been processed | |
575 | * happens when called speculatively from start_xmit. | |
576 | */ | |
577 | if (!packets) | |
578 | return; | |
579 | ||
580 | u64_stats_update_begin(&sq->stats.syncp); | |
581 | sq->stats.bytes += bytes; | |
582 | sq->stats.packets += packets; | |
583 | u64_stats_update_end(&sq->stats.syncp); | |
584 | } | |
585 | ||
586 | static bool is_xdp_raw_buffer_queue(struct virtnet_info *vi, int q) | |
587 | { | |
588 | if (q < (vi->curr_queue_pairs - vi->xdp_queue_pairs)) | |
589 | return false; | |
590 | else if (q < vi->curr_queue_pairs) | |
591 | return true; | |
592 | else | |
593 | return false; | |
594 | } | |
595 | ||
b8ef4809 XZ |
596 | static void check_sq_full_and_disable(struct virtnet_info *vi, |
597 | struct net_device *dev, | |
598 | struct send_queue *sq) | |
599 | { | |
600 | bool use_napi = sq->napi.weight; | |
601 | int qnum; | |
602 | ||
603 | qnum = sq - vi->sq; | |
604 | ||
605 | /* If running out of space, stop queue to avoid getting packets that we | |
606 | * are then unable to transmit. | |
607 | * An alternative would be to force queuing layer to requeue the skb by | |
608 | * returning NETDEV_TX_BUSY. However, NETDEV_TX_BUSY should not be | |
609 | * returned in a normal path of operation: it means that driver is not | |
610 | * maintaining the TX queue stop/start state properly, and causes | |
611 | * the stack to do a non-trivial amount of useless work. | |
612 | * Since most packets only take 1 or 2 ring slots, stopping the queue | |
613 | * early means 16 slots are typically wasted. | |
614 | */ | |
615 | if (sq->vq->num_free < 2+MAX_SKB_FRAGS) { | |
616 | netif_stop_subqueue(dev, qnum); | |
617 | if (use_napi) { | |
618 | if (unlikely(!virtqueue_enable_cb_delayed(sq->vq))) | |
619 | virtqueue_napi_schedule(&sq->napi, sq->vq); | |
620 | } else if (unlikely(!virtqueue_enable_cb_delayed(sq->vq))) { | |
621 | /* More just got used, free them then recheck. */ | |
622 | free_old_xmit_skbs(sq, false); | |
623 | if (sq->vq->num_free >= 2+MAX_SKB_FRAGS) { | |
624 | netif_start_subqueue(dev, qnum); | |
625 | virtqueue_disable_cb(sq->vq); | |
626 | } | |
627 | } | |
628 | } | |
629 | } | |
630 | ||
735fc405 JDB |
631 | static int __virtnet_xdp_xmit_one(struct virtnet_info *vi, |
632 | struct send_queue *sq, | |
633 | struct xdp_frame *xdpf) | |
56434a01 | 634 | { |
56434a01 | 635 | struct virtio_net_hdr_mrg_rxbuf *hdr; |
97717e8d HQ |
636 | struct skb_shared_info *shinfo; |
637 | u8 nr_frags = 0; | |
638 | int err, i; | |
56434a01 | 639 | |
cac320c8 JDB |
640 | if (unlikely(xdpf->headroom < vi->hdr_len)) |
641 | return -EOVERFLOW; | |
642 | ||
97717e8d HQ |
643 | if (unlikely(xdp_frame_has_frags(xdpf))) { |
644 | shinfo = xdp_get_shared_info_from_frame(xdpf); | |
645 | nr_frags = shinfo->nr_frags; | |
646 | } | |
647 | ||
648 | /* In wrapping function virtnet_xdp_xmit(), we need to free | |
649 | * up the pending old buffers, where we need to calculate the | |
650 | * position of skb_shared_info in xdp_get_frame_len() and | |
651 | * xdp_return_frame(), which will involve to xdpf->data and | |
652 | * xdpf->headroom. Therefore, we need to update the value of | |
653 | * headroom synchronously here. | |
654 | */ | |
655 | xdpf->headroom -= vi->hdr_len; | |
cac320c8 | 656 | xdpf->data -= vi->hdr_len; |
f6b10209 | 657 | /* Zero header and leave csum up to XDP layers */ |
cac320c8 | 658 | hdr = xdpf->data; |
f6b10209 | 659 | memset(hdr, 0, vi->hdr_len); |
cac320c8 | 660 | xdpf->len += vi->hdr_len; |
bb91accf | 661 | |
97717e8d HQ |
662 | sg_init_table(sq->sg, nr_frags + 1); |
663 | sg_set_buf(sq->sg, xdpf->data, xdpf->len); | |
664 | for (i = 0; i < nr_frags; i++) { | |
665 | skb_frag_t *frag = &shinfo->frags[i]; | |
666 | ||
667 | sg_set_page(&sq->sg[i + 1], skb_frag_page(frag), | |
668 | skb_frag_size(frag), skb_frag_off(frag)); | |
669 | } | |
bb91accf | 670 | |
97717e8d HQ |
671 | err = virtqueue_add_outbuf(sq->vq, sq->sg, nr_frags + 1, |
672 | xdp_to_ptr(xdpf), GFP_ATOMIC); | |
11b7d897 | 673 | if (unlikely(err)) |
cac320c8 | 674 | return -ENOSPC; /* Caller handle free/refcnt */ |
56434a01 | 675 | |
cac320c8 | 676 | return 0; |
56434a01 JF |
677 | } |
678 | ||
97c2c69e XZ |
679 | /* when vi->curr_queue_pairs > nr_cpu_ids, the txq/sq is only used for xdp tx on |
680 | * the current cpu, so it does not need to be locked. | |
681 | * | |
682 | * Here we use marco instead of inline functions because we have to deal with | |
683 | * three issues at the same time: 1. the choice of sq. 2. judge and execute the | |
684 | * lock/unlock of txq 3. make sparse happy. It is difficult for two inline | |
685 | * functions to perfectly solve these three problems at the same time. | |
686 | */ | |
687 | #define virtnet_xdp_get_sq(vi) ({ \ | |
3dcc1edc | 688 | int cpu = smp_processor_id(); \ |
97c2c69e XZ |
689 | struct netdev_queue *txq; \ |
690 | typeof(vi) v = (vi); \ | |
691 | unsigned int qp; \ | |
692 | \ | |
693 | if (v->curr_queue_pairs > nr_cpu_ids) { \ | |
694 | qp = v->curr_queue_pairs - v->xdp_queue_pairs; \ | |
3dcc1edc | 695 | qp += cpu; \ |
97c2c69e XZ |
696 | txq = netdev_get_tx_queue(v->dev, qp); \ |
697 | __netif_tx_acquire(txq); \ | |
698 | } else { \ | |
3dcc1edc | 699 | qp = cpu % v->curr_queue_pairs; \ |
97c2c69e | 700 | txq = netdev_get_tx_queue(v->dev, qp); \ |
3dcc1edc | 701 | __netif_tx_lock(txq, cpu); \ |
97c2c69e XZ |
702 | } \ |
703 | v->sq + qp; \ | |
704 | }) | |
705 | ||
706 | #define virtnet_xdp_put_sq(vi, q) { \ | |
707 | struct netdev_queue *txq; \ | |
708 | typeof(vi) v = (vi); \ | |
709 | \ | |
710 | txq = netdev_get_tx_queue(v->dev, (q) - v->sq); \ | |
711 | if (v->curr_queue_pairs > nr_cpu_ids) \ | |
712 | __netif_tx_release(txq); \ | |
713 | else \ | |
714 | __netif_tx_unlock(txq); \ | |
2a43565c TM |
715 | } |
716 | ||
735fc405 | 717 | static int virtnet_xdp_xmit(struct net_device *dev, |
42b33468 | 718 | int n, struct xdp_frame **frames, u32 flags) |
186b3c99 JW |
719 | { |
720 | struct virtnet_info *vi = netdev_priv(dev); | |
8dcc5b0a JDB |
721 | struct receive_queue *rq = vi->rq; |
722 | struct bpf_prog *xdp_prog; | |
735fc405 JDB |
723 | struct send_queue *sq; |
724 | unsigned int len; | |
546f2897 TM |
725 | int packets = 0; |
726 | int bytes = 0; | |
fdc13979 | 727 | int nxmit = 0; |
461f03dc | 728 | int kicks = 0; |
5050471d | 729 | void *ptr; |
fdc13979 | 730 | int ret; |
735fc405 JDB |
731 | int i; |
732 | ||
8dcc5b0a JDB |
733 | /* Only allow ndo_xdp_xmit if XDP is loaded on dev, as this |
734 | * indicate XDP resources have been successfully allocated. | |
735 | */ | |
9719c6b9 | 736 | xdp_prog = rcu_access_pointer(rq->xdp_prog); |
1667c08a TM |
737 | if (!xdp_prog) |
738 | return -ENXIO; | |
739 | ||
97c2c69e | 740 | sq = virtnet_xdp_get_sq(vi); |
1667c08a TM |
741 | |
742 | if (unlikely(flags & ~XDP_XMIT_FLAGS_MASK)) { | |
743 | ret = -EINVAL; | |
5b8f3c8d TM |
744 | goto out; |
745 | } | |
8dcc5b0a | 746 | |
735fc405 | 747 | /* Free up any pending old buffers before queueing new ones. */ |
5050471d | 748 | while ((ptr = virtqueue_get_buf(sq->vq, &len)) != NULL) { |
546f2897 TM |
749 | if (likely(is_xdp_frame(ptr))) { |
750 | struct xdp_frame *frame = ptr_to_xdp(ptr); | |
751 | ||
50bd14bc | 752 | bytes += xdp_get_frame_len(frame); |
546f2897 TM |
753 | xdp_return_frame(frame); |
754 | } else { | |
755 | struct sk_buff *skb = ptr; | |
756 | ||
757 | bytes += skb->len; | |
758 | napi_consume_skb(skb, false); | |
759 | } | |
760 | packets++; | |
5050471d | 761 | } |
735fc405 JDB |
762 | |
763 | for (i = 0; i < n; i++) { | |
764 | struct xdp_frame *xdpf = frames[i]; | |
765 | ||
fdc13979 LB |
766 | if (__virtnet_xdp_xmit_one(vi, sq, xdpf)) |
767 | break; | |
768 | nxmit++; | |
735fc405 | 769 | } |
fdc13979 | 770 | ret = nxmit; |
5d274cb4 | 771 | |
cd1c604a XZ |
772 | if (!is_xdp_raw_buffer_queue(vi, sq - vi->sq)) |
773 | check_sq_full_and_disable(vi, dev, sq); | |
774 | ||
461f03dc TM |
775 | if (flags & XDP_XMIT_FLUSH) { |
776 | if (virtqueue_kick_prepare(sq->vq) && virtqueue_notify(sq->vq)) | |
777 | kicks = 1; | |
778 | } | |
5b8f3c8d TM |
779 | out: |
780 | u64_stats_update_begin(&sq->stats.syncp); | |
546f2897 TM |
781 | sq->stats.bytes += bytes; |
782 | sq->stats.packets += packets; | |
5b8f3c8d | 783 | sq->stats.xdp_tx += n; |
fdc13979 | 784 | sq->stats.xdp_tx_drops += n - nxmit; |
461f03dc | 785 | sq->stats.kicks += kicks; |
5b8f3c8d | 786 | u64_stats_update_end(&sq->stats.syncp); |
5d274cb4 | 787 | |
97c2c69e | 788 | virtnet_xdp_put_sq(vi, sq); |
5b8f3c8d | 789 | return ret; |
186b3c99 JW |
790 | } |
791 | ||
f6b10209 JW |
792 | static unsigned int virtnet_get_headroom(struct virtnet_info *vi) |
793 | { | |
97c2c69e | 794 | return vi->xdp_enabled ? VIRTIO_XDP_HEADROOM : 0; |
f6b10209 JW |
795 | } |
796 | ||
4941d472 JW |
797 | /* We copy the packet for XDP in the following cases: |
798 | * | |
799 | * 1) Packet is scattered across multiple rx buffers. | |
800 | * 2) Headroom space is insufficient. | |
801 | * | |
802 | * This is inefficient but it's a temporary condition that | |
803 | * we hit right after XDP is enabled and until queue is refilled | |
804 | * with large buffers with sufficient headroom - so it should affect | |
805 | * at most queue size packets. | |
806 | * Afterwards, the conditions to enable | |
807 | * XDP should preclude the underlying device from sending packets | |
808 | * across multiple buffers (num_buf > 1), and we make sure buffers | |
809 | * have enough headroom. | |
810 | */ | |
811 | static struct page *xdp_linearize_page(struct receive_queue *rq, | |
981f14d4 | 812 | int *num_buf, |
4941d472 JW |
813 | struct page *p, |
814 | int offset, | |
815 | int page_off, | |
816 | unsigned int *len) | |
817 | { | |
853618d5 XZ |
818 | int tailroom = SKB_DATA_ALIGN(sizeof(struct skb_shared_info)); |
819 | struct page *page; | |
4941d472 | 820 | |
853618d5 XZ |
821 | if (page_off + *len + tailroom > PAGE_SIZE) |
822 | return NULL; | |
823 | ||
824 | page = alloc_page(GFP_ATOMIC); | |
4941d472 JW |
825 | if (!page) |
826 | return NULL; | |
827 | ||
828 | memcpy(page_address(page) + page_off, page_address(p) + offset, *len); | |
829 | page_off += *len; | |
830 | ||
831 | while (--*num_buf) { | |
832 | unsigned int buflen; | |
833 | void *buf; | |
834 | int off; | |
835 | ||
836 | buf = virtqueue_get_buf(rq->vq, &buflen); | |
837 | if (unlikely(!buf)) | |
838 | goto err_buf; | |
839 | ||
840 | p = virt_to_head_page(buf); | |
841 | off = buf - page_address(p); | |
842 | ||
843 | /* guard against a misconfigured or uncooperative backend that | |
844 | * is sending packet larger than the MTU. | |
845 | */ | |
3cc81a9a | 846 | if ((page_off + buflen + tailroom) > PAGE_SIZE) { |
4941d472 JW |
847 | put_page(p); |
848 | goto err_buf; | |
849 | } | |
850 | ||
851 | memcpy(page_address(page) + page_off, | |
852 | page_address(p) + off, buflen); | |
853 | page_off += buflen; | |
854 | put_page(p); | |
855 | } | |
856 | ||
857 | /* Headroom does not contribute to packet length */ | |
858 | *len = page_off - VIRTIO_XDP_HEADROOM; | |
859 | return page; | |
860 | err_buf: | |
861 | __free_pages(page, 0); | |
862 | return NULL; | |
863 | } | |
864 | ||
bb91accf JW |
865 | static struct sk_buff *receive_small(struct net_device *dev, |
866 | struct virtnet_info *vi, | |
867 | struct receive_queue *rq, | |
192f68cf | 868 | void *buf, void *ctx, |
186b3c99 | 869 | unsigned int len, |
7d9d60fd | 870 | unsigned int *xdp_xmit, |
d46eeeaf | 871 | struct virtnet_rq_stats *stats) |
f121159d | 872 | { |
f6b10209 | 873 | struct sk_buff *skb; |
bb91accf | 874 | struct bpf_prog *xdp_prog; |
4941d472 | 875 | unsigned int xdp_headroom = (unsigned long)ctx; |
f6b10209 JW |
876 | unsigned int header_offset = VIRTNET_RX_PAD + xdp_headroom; |
877 | unsigned int headroom = vi->hdr_len + header_offset; | |
878 | unsigned int buflen = SKB_DATA_ALIGN(GOOD_PACKET_LEN + headroom) + | |
879 | SKB_DATA_ALIGN(sizeof(struct skb_shared_info)); | |
4941d472 | 880 | struct page *page = virt_to_head_page(buf); |
11b7d897 | 881 | unsigned int delta = 0; |
4941d472 | 882 | struct page *xdp_page; |
11b7d897 | 883 | int err; |
503d539a | 884 | unsigned int metasize = 0; |
11b7d897 | 885 | |
012873d0 | 886 | len -= vi->hdr_len; |
d46eeeaf | 887 | stats->bytes += len; |
f121159d | 888 | |
ad993a95 XY |
889 | if (unlikely(len > GOOD_PACKET_LEN)) { |
890 | pr_debug("%s: rx error: len %u exceeds max size %d\n", | |
891 | dev->name, len, GOOD_PACKET_LEN); | |
892 | dev->stats.rx_length_errors++; | |
053c9e18 | 893 | goto err; |
ad993a95 | 894 | } |
6213f07c LR |
895 | |
896 | if (likely(!vi->xdp_enabled)) { | |
897 | xdp_prog = NULL; | |
898 | goto skip_xdp; | |
899 | } | |
900 | ||
bb91accf JW |
901 | rcu_read_lock(); |
902 | xdp_prog = rcu_dereference(rq->xdp_prog); | |
903 | if (xdp_prog) { | |
f6b10209 | 904 | struct virtio_net_hdr_mrg_rxbuf *hdr = buf + header_offset; |
44fa2dbd | 905 | struct xdp_frame *xdpf; |
0354e4d1 | 906 | struct xdp_buff xdp; |
f6b10209 | 907 | void *orig_data; |
bb91accf JW |
908 | u32 act; |
909 | ||
95dbe9e7 | 910 | if (unlikely(hdr->hdr.gso_type)) |
bb91accf | 911 | goto err_xdp; |
0354e4d1 | 912 | |
4941d472 JW |
913 | if (unlikely(xdp_headroom < virtnet_get_headroom(vi))) { |
914 | int offset = buf - page_address(page) + header_offset; | |
915 | unsigned int tlen = len + vi->hdr_len; | |
981f14d4 | 916 | int num_buf = 1; |
4941d472 JW |
917 | |
918 | xdp_headroom = virtnet_get_headroom(vi); | |
919 | header_offset = VIRTNET_RX_PAD + xdp_headroom; | |
920 | headroom = vi->hdr_len + header_offset; | |
921 | buflen = SKB_DATA_ALIGN(GOOD_PACKET_LEN + headroom) + | |
922 | SKB_DATA_ALIGN(sizeof(struct skb_shared_info)); | |
923 | xdp_page = xdp_linearize_page(rq, &num_buf, page, | |
924 | offset, header_offset, | |
925 | &tlen); | |
926 | if (!xdp_page) | |
927 | goto err_xdp; | |
928 | ||
929 | buf = page_address(xdp_page); | |
930 | put_page(page); | |
931 | page = xdp_page; | |
932 | } | |
933 | ||
43b5169d | 934 | xdp_init_buff(&xdp, buflen, &rq->xdp_rxq); |
be9df4af LB |
935 | xdp_prepare_buff(&xdp, buf + VIRTNET_RX_PAD + vi->hdr_len, |
936 | xdp_headroom, len, true); | |
f6b10209 | 937 | orig_data = xdp.data; |
0354e4d1 | 938 | act = bpf_prog_run_xdp(xdp_prog, &xdp); |
d46eeeaf | 939 | stats->xdp_packets++; |
0354e4d1 | 940 | |
bb91accf JW |
941 | switch (act) { |
942 | case XDP_PASS: | |
2de2f7f4 | 943 | /* Recalculate length in case bpf program changed it */ |
f6b10209 | 944 | delta = orig_data - xdp.data; |
6870de43 | 945 | len = xdp.data_end - xdp.data; |
503d539a | 946 | metasize = xdp.data - xdp.data_meta; |
bb91accf JW |
947 | break; |
948 | case XDP_TX: | |
d46eeeaf | 949 | stats->xdp_tx++; |
1b698fa5 | 950 | xdpf = xdp_convert_buff_to_frame(&xdp); |
44fa2dbd JDB |
951 | if (unlikely(!xdpf)) |
952 | goto err_xdp; | |
ca9e83b4 | 953 | err = virtnet_xdp_xmit(dev, 1, &xdpf, 0); |
fdc13979 LB |
954 | if (unlikely(!err)) { |
955 | xdp_return_frame_rx_napi(xdpf); | |
956 | } else if (unlikely(err < 0)) { | |
0354e4d1 | 957 | trace_xdp_exception(vi->dev, xdp_prog, act); |
11b7d897 JDB |
958 | goto err_xdp; |
959 | } | |
2471c75e | 960 | *xdp_xmit |= VIRTIO_XDP_TX; |
186b3c99 JW |
961 | rcu_read_unlock(); |
962 | goto xdp_xmit; | |
963 | case XDP_REDIRECT: | |
d46eeeaf | 964 | stats->xdp_redirects++; |
186b3c99 | 965 | err = xdp_do_redirect(dev, &xdp, xdp_prog); |
11b7d897 JDB |
966 | if (err) |
967 | goto err_xdp; | |
2471c75e | 968 | *xdp_xmit |= VIRTIO_XDP_REDIR; |
bb91accf JW |
969 | rcu_read_unlock(); |
970 | goto xdp_xmit; | |
bb91accf | 971 | default: |
c8064e5b | 972 | bpf_warn_invalid_xdp_action(vi->dev, xdp_prog, act); |
df561f66 | 973 | fallthrough; |
0354e4d1 JF |
974 | case XDP_ABORTED: |
975 | trace_xdp_exception(vi->dev, xdp_prog, act); | |
95efabf0 | 976 | goto err_xdp; |
0354e4d1 | 977 | case XDP_DROP: |
bb91accf JW |
978 | goto err_xdp; |
979 | } | |
980 | } | |
981 | rcu_read_unlock(); | |
982 | ||
6213f07c | 983 | skip_xdp: |
f6b10209 | 984 | skb = build_skb(buf, buflen); |
053c9e18 | 985 | if (!skb) |
f6b10209 | 986 | goto err; |
f6b10209 | 987 | skb_reserve(skb, headroom - delta); |
6870de43 | 988 | skb_put(skb, len); |
f1d4884d | 989 | if (!xdp_prog) { |
f6b10209 JW |
990 | buf += header_offset; |
991 | memcpy(skb_vnet_hdr(skb), buf, vi->hdr_len); | |
f1d4884d | 992 | } /* keep zeroed vnet hdr since XDP is loaded */ |
f6b10209 | 993 | |
503d539a YK |
994 | if (metasize) |
995 | skb_metadata_set(skb, metasize); | |
996 | ||
f121159d | 997 | return skb; |
bb91accf JW |
998 | |
999 | err_xdp: | |
1000 | rcu_read_unlock(); | |
d46eeeaf | 1001 | stats->xdp_drops++; |
053c9e18 | 1002 | err: |
d46eeeaf | 1003 | stats->drops++; |
4941d472 | 1004 | put_page(page); |
bb91accf JW |
1005 | xdp_xmit: |
1006 | return NULL; | |
f121159d MT |
1007 | } |
1008 | ||
1009 | static struct sk_buff *receive_big(struct net_device *dev, | |
946fa564 | 1010 | struct virtnet_info *vi, |
f121159d MT |
1011 | struct receive_queue *rq, |
1012 | void *buf, | |
7d9d60fd | 1013 | unsigned int len, |
d46eeeaf | 1014 | struct virtnet_rq_stats *stats) |
f121159d MT |
1015 | { |
1016 | struct page *page = buf; | |
503d539a | 1017 | struct sk_buff *skb = |
fa0f1ba7 | 1018 | page_to_skb(vi, rq, page, 0, len, PAGE_SIZE, 0); |
f600b690 | 1019 | |
d46eeeaf | 1020 | stats->bytes += len - vi->hdr_len; |
f121159d MT |
1021 | if (unlikely(!skb)) |
1022 | goto err; | |
1023 | ||
1024 | return skb; | |
1025 | ||
1026 | err: | |
d46eeeaf | 1027 | stats->drops++; |
f121159d MT |
1028 | give_pages(rq, page); |
1029 | return NULL; | |
1030 | } | |
1031 | ||
b26aa481 HQ |
1032 | /* Why not use xdp_build_skb_from_frame() ? |
1033 | * XDP core assumes that xdp frags are PAGE_SIZE in length, while in | |
1034 | * virtio-net there are 2 points that do not match its requirements: | |
1035 | * 1. The size of the prefilled buffer is not fixed before xdp is set. | |
1036 | * 2. xdp_build_skb_from_frame() does more checks that we don't need, | |
1037 | * like eth_type_trans() (which virtio-net does in receive_buf()). | |
1038 | */ | |
1039 | static struct sk_buff *build_skb_from_xdp_buff(struct net_device *dev, | |
1040 | struct virtnet_info *vi, | |
1041 | struct xdp_buff *xdp, | |
1042 | unsigned int xdp_frags_truesz) | |
1043 | { | |
1044 | struct skb_shared_info *sinfo = xdp_get_shared_info_from_buff(xdp); | |
1045 | unsigned int headroom, data_len; | |
1046 | struct sk_buff *skb; | |
1047 | int metasize; | |
1048 | u8 nr_frags; | |
1049 | ||
1050 | if (unlikely(xdp->data_end > xdp_data_hard_end(xdp))) { | |
1051 | pr_debug("Error building skb as missing reserved tailroom for xdp"); | |
1052 | return NULL; | |
1053 | } | |
1054 | ||
1055 | if (unlikely(xdp_buff_has_frags(xdp))) | |
1056 | nr_frags = sinfo->nr_frags; | |
1057 | ||
1058 | skb = build_skb(xdp->data_hard_start, xdp->frame_sz); | |
1059 | if (unlikely(!skb)) | |
1060 | return NULL; | |
1061 | ||
1062 | headroom = xdp->data - xdp->data_hard_start; | |
1063 | data_len = xdp->data_end - xdp->data; | |
1064 | skb_reserve(skb, headroom); | |
1065 | __skb_put(skb, data_len); | |
1066 | ||
1067 | metasize = xdp->data - xdp->data_meta; | |
1068 | metasize = metasize > 0 ? metasize : 0; | |
1069 | if (metasize) | |
1070 | skb_metadata_set(skb, metasize); | |
1071 | ||
1072 | if (unlikely(xdp_buff_has_frags(xdp))) | |
1073 | xdp_update_skb_shared_info(skb, nr_frags, | |
1074 | sinfo->xdp_frags_size, | |
1075 | xdp_frags_truesz, | |
1076 | xdp_buff_is_frag_pfmemalloc(xdp)); | |
1077 | ||
1078 | return skb; | |
1079 | } | |
1080 | ||
ef75cb51 HQ |
1081 | /* TODO: build xdp in big mode */ |
1082 | static int virtnet_build_xdp_buff_mrg(struct net_device *dev, | |
1083 | struct virtnet_info *vi, | |
1084 | struct receive_queue *rq, | |
1085 | struct xdp_buff *xdp, | |
1086 | void *buf, | |
1087 | unsigned int len, | |
1088 | unsigned int frame_sz, | |
981f14d4 | 1089 | int *num_buf, |
ef75cb51 HQ |
1090 | unsigned int *xdp_frags_truesize, |
1091 | struct virtnet_rq_stats *stats) | |
1092 | { | |
1093 | struct virtio_net_hdr_mrg_rxbuf *hdr = buf; | |
1094 | unsigned int headroom, tailroom, room; | |
1095 | unsigned int truesize, cur_frag_size; | |
1096 | struct skb_shared_info *shinfo; | |
1097 | unsigned int xdp_frags_truesz = 0; | |
1098 | struct page *page; | |
1099 | skb_frag_t *frag; | |
1100 | int offset; | |
1101 | void *ctx; | |
1102 | ||
1103 | xdp_init_buff(xdp, frame_sz, &rq->xdp_rxq); | |
1104 | xdp_prepare_buff(xdp, buf - VIRTIO_XDP_HEADROOM, | |
1105 | VIRTIO_XDP_HEADROOM + vi->hdr_len, len - vi->hdr_len, true); | |
1106 | ||
981f14d4 HQ |
1107 | if (!*num_buf) |
1108 | return 0; | |
1109 | ||
ef75cb51 HQ |
1110 | if (*num_buf > 1) { |
1111 | /* If we want to build multi-buffer xdp, we need | |
1112 | * to specify that the flags of xdp_buff have the | |
1113 | * XDP_FLAGS_HAS_FRAG bit. | |
1114 | */ | |
1115 | if (!xdp_buff_has_frags(xdp)) | |
1116 | xdp_buff_set_frags_flag(xdp); | |
1117 | ||
1118 | shinfo = xdp_get_shared_info_from_buff(xdp); | |
1119 | shinfo->nr_frags = 0; | |
1120 | shinfo->xdp_frags_size = 0; | |
1121 | } | |
1122 | ||
981f14d4 | 1123 | if (*num_buf > MAX_SKB_FRAGS + 1) |
ef75cb51 HQ |
1124 | return -EINVAL; |
1125 | ||
981f14d4 | 1126 | while (--*num_buf > 0) { |
ef75cb51 HQ |
1127 | buf = virtqueue_get_buf_ctx(rq->vq, &len, &ctx); |
1128 | if (unlikely(!buf)) { | |
1129 | pr_debug("%s: rx error: %d buffers out of %d missing\n", | |
1130 | dev->name, *num_buf, | |
1131 | virtio16_to_cpu(vi->vdev, hdr->num_buffers)); | |
1132 | dev->stats.rx_length_errors++; | |
1133 | return -EINVAL; | |
1134 | } | |
1135 | ||
1136 | stats->bytes += len; | |
1137 | page = virt_to_head_page(buf); | |
1138 | offset = buf - page_address(page); | |
1139 | ||
1140 | truesize = mergeable_ctx_to_truesize(ctx); | |
1141 | headroom = mergeable_ctx_to_headroom(ctx); | |
1142 | tailroom = headroom ? sizeof(struct skb_shared_info) : 0; | |
1143 | room = SKB_DATA_ALIGN(headroom + tailroom); | |
1144 | ||
1145 | cur_frag_size = truesize; | |
1146 | xdp_frags_truesz += cur_frag_size; | |
1147 | if (unlikely(len > truesize - room || cur_frag_size > PAGE_SIZE)) { | |
1148 | put_page(page); | |
1149 | pr_debug("%s: rx error: len %u exceeds truesize %lu\n", | |
1150 | dev->name, len, (unsigned long)(truesize - room)); | |
1151 | dev->stats.rx_length_errors++; | |
1152 | return -EINVAL; | |
1153 | } | |
1154 | ||
1155 | frag = &shinfo->frags[shinfo->nr_frags++]; | |
1156 | __skb_frag_set_page(frag, page); | |
1157 | skb_frag_off_set(frag, offset); | |
1158 | skb_frag_size_set(frag, len); | |
1159 | if (page_is_pfmemalloc(page)) | |
1160 | xdp_buff_set_frag_pfmemalloc(xdp); | |
1161 | ||
1162 | shinfo->xdp_frags_size += len; | |
1163 | } | |
1164 | ||
1165 | *xdp_frags_truesize = xdp_frags_truesz; | |
1166 | return 0; | |
1167 | } | |
1168 | ||
8fc3b9e9 | 1169 | static struct sk_buff *receive_mergeable(struct net_device *dev, |
fdd819b2 | 1170 | struct virtnet_info *vi, |
8fc3b9e9 | 1171 | struct receive_queue *rq, |
680557cf MT |
1172 | void *buf, |
1173 | void *ctx, | |
186b3c99 | 1174 | unsigned int len, |
7d9d60fd | 1175 | unsigned int *xdp_xmit, |
d46eeeaf | 1176 | struct virtnet_rq_stats *stats) |
9ab86bbc | 1177 | { |
012873d0 | 1178 | struct virtio_net_hdr_mrg_rxbuf *hdr = buf; |
981f14d4 | 1179 | int num_buf = virtio16_to_cpu(vi->vdev, hdr->num_buffers); |
8fc3b9e9 MT |
1180 | struct page *page = virt_to_head_page(buf); |
1181 | int offset = buf - page_address(page); | |
f600b690 JF |
1182 | struct sk_buff *head_skb, *curr_skb; |
1183 | struct bpf_prog *xdp_prog; | |
9ce6146e | 1184 | unsigned int truesize = mergeable_ctx_to_truesize(ctx); |
4941d472 | 1185 | unsigned int headroom = mergeable_ctx_to_headroom(ctx); |
ef75cb51 HQ |
1186 | unsigned int tailroom = headroom ? sizeof(struct skb_shared_info) : 0; |
1187 | unsigned int room = SKB_DATA_ALIGN(headroom + tailroom); | |
22174f79 | 1188 | unsigned int frame_sz, xdp_room; |
9ce6146e | 1189 | int err; |
f600b690 | 1190 | |
56434a01 | 1191 | head_skb = NULL; |
d46eeeaf | 1192 | stats->bytes += len - vi->hdr_len; |
56434a01 | 1193 | |
ef75cb51 | 1194 | if (unlikely(len > truesize - room)) { |
ad993a95 | 1195 | pr_debug("%s: rx error: len %u exceeds truesize %lu\n", |
ef75cb51 | 1196 | dev->name, len, (unsigned long)(truesize - room)); |
ad993a95 XY |
1197 | dev->stats.rx_length_errors++; |
1198 | goto err_skb; | |
1199 | } | |
6213f07c LR |
1200 | |
1201 | if (likely(!vi->xdp_enabled)) { | |
1202 | xdp_prog = NULL; | |
1203 | goto skip_xdp; | |
1204 | } | |
1205 | ||
f600b690 JF |
1206 | rcu_read_lock(); |
1207 | xdp_prog = rcu_dereference(rq->xdp_prog); | |
1208 | if (xdp_prog) { | |
22174f79 HQ |
1209 | unsigned int xdp_frags_truesz = 0; |
1210 | struct skb_shared_info *shinfo; | |
44fa2dbd | 1211 | struct xdp_frame *xdpf; |
72979a6c | 1212 | struct page *xdp_page; |
0354e4d1 | 1213 | struct xdp_buff xdp; |
0354e4d1 | 1214 | void *data; |
f600b690 | 1215 | u32 act; |
22174f79 | 1216 | int i; |
f600b690 | 1217 | |
3d62b2a0 JW |
1218 | /* Transient failure which in theory could occur if |
1219 | * in-flight packets from before XDP was enabled reach | |
1220 | * the receive path after XDP is loaded. | |
1221 | */ | |
1222 | if (unlikely(hdr->hdr.gso_type)) | |
1223 | goto err_xdp; | |
1224 | ||
ef75cb51 HQ |
1225 | /* Now XDP core assumes frag size is PAGE_SIZE, but buffers |
1226 | * with headroom may add hole in truesize, which | |
1227 | * make their length exceed PAGE_SIZE. So we disabled the | |
1228 | * hole mechanism for xdp. See add_recvbuf_mergeable(). | |
9ce6146e | 1229 | */ |
ef75cb51 | 1230 | frame_sz = truesize; |
9ce6146e | 1231 | |
22174f79 HQ |
1232 | /* This happens when headroom is not enough because |
1233 | * of the buffer was prefilled before XDP is set. | |
1234 | * This should only happen for the first several packets. | |
1235 | * In fact, vq reset can be used here to help us clean up | |
1236 | * the prefilled buffers, but many existing devices do not | |
1237 | * support it, and we don't want to bother users who are | |
1238 | * using xdp normally. | |
3cc81a9a | 1239 | */ |
22174f79 HQ |
1240 | if (!xdp_prog->aux->xdp_has_frags && |
1241 | (num_buf > 1 || headroom < virtnet_get_headroom(vi))) { | |
72979a6c | 1242 | /* linearize data for XDP */ |
56a86f84 | 1243 | xdp_page = xdp_linearize_page(rq, &num_buf, |
4941d472 JW |
1244 | page, offset, |
1245 | VIRTIO_XDP_HEADROOM, | |
1246 | &len); | |
9ce6146e JDB |
1247 | frame_sz = PAGE_SIZE; |
1248 | ||
72979a6c JF |
1249 | if (!xdp_page) |
1250 | goto err_xdp; | |
2de2f7f4 | 1251 | offset = VIRTIO_XDP_HEADROOM; |
22174f79 HQ |
1252 | } else if (unlikely(headroom < virtnet_get_headroom(vi))) { |
1253 | xdp_room = SKB_DATA_ALIGN(VIRTIO_XDP_HEADROOM + | |
1254 | sizeof(struct skb_shared_info)); | |
1255 | if (len + xdp_room > PAGE_SIZE) | |
1256 | goto err_xdp; | |
1257 | ||
1258 | xdp_page = alloc_page(GFP_ATOMIC); | |
1259 | if (!xdp_page) | |
1260 | goto err_xdp; | |
1261 | ||
1262 | memcpy(page_address(xdp_page) + VIRTIO_XDP_HEADROOM, | |
1263 | page_address(page) + offset, len); | |
1264 | frame_sz = PAGE_SIZE; | |
1265 | offset = VIRTIO_XDP_HEADROOM; | |
72979a6c JF |
1266 | } else { |
1267 | xdp_page = page; | |
f600b690 JF |
1268 | } |
1269 | ||
0354e4d1 | 1270 | data = page_address(xdp_page) + offset; |
22174f79 HQ |
1271 | err = virtnet_build_xdp_buff_mrg(dev, vi, rq, &xdp, data, len, frame_sz, |
1272 | &num_buf, &xdp_frags_truesz, stats); | |
1273 | if (unlikely(err)) | |
1274 | goto err_xdp_frags; | |
754b8a21 | 1275 | |
0354e4d1 | 1276 | act = bpf_prog_run_xdp(xdp_prog, &xdp); |
d46eeeaf | 1277 | stats->xdp_packets++; |
0354e4d1 | 1278 | |
56434a01 JF |
1279 | switch (act) { |
1280 | case XDP_PASS: | |
1a3bd6ea XZ |
1281 | head_skb = build_skb_from_xdp_buff(dev, vi, &xdp, xdp_frags_truesz); |
1282 | if (unlikely(!head_skb)) | |
1283 | goto err_xdp_frags; | |
1284 | ||
fab89baf | 1285 | if (unlikely(xdp_page != page)) |
1830f893 | 1286 | put_page(page); |
fab89baf HQ |
1287 | rcu_read_unlock(); |
1288 | return head_skb; | |
56434a01 | 1289 | case XDP_TX: |
d46eeeaf | 1290 | stats->xdp_tx++; |
1b698fa5 | 1291 | xdpf = xdp_convert_buff_to_frame(&xdp); |
7a542bee | 1292 | if (unlikely(!xdpf)) { |
fab89baf HQ |
1293 | netdev_dbg(dev, "convert buff to frame failed for xdp\n"); |
1294 | goto err_xdp_frags; | |
7a542bee | 1295 | } |
ca9e83b4 | 1296 | err = virtnet_xdp_xmit(dev, 1, &xdpf, 0); |
fdc13979 LB |
1297 | if (unlikely(!err)) { |
1298 | xdp_return_frame_rx_napi(xdpf); | |
1299 | } else if (unlikely(err < 0)) { | |
0354e4d1 | 1300 | trace_xdp_exception(vi->dev, xdp_prog, act); |
fab89baf | 1301 | goto err_xdp_frags; |
11b7d897 | 1302 | } |
2471c75e | 1303 | *xdp_xmit |= VIRTIO_XDP_TX; |
72979a6c | 1304 | if (unlikely(xdp_page != page)) |
5d458a13 | 1305 | put_page(page); |
56434a01 JF |
1306 | rcu_read_unlock(); |
1307 | goto xdp_xmit; | |
3cc81a9a | 1308 | case XDP_REDIRECT: |
d46eeeaf | 1309 | stats->xdp_redirects++; |
3cc81a9a | 1310 | err = xdp_do_redirect(dev, &xdp, xdp_prog); |
fab89baf HQ |
1311 | if (err) |
1312 | goto err_xdp_frags; | |
2471c75e | 1313 | *xdp_xmit |= VIRTIO_XDP_REDIR; |
3cc81a9a | 1314 | if (unlikely(xdp_page != page)) |
6890418b | 1315 | put_page(page); |
3cc81a9a JW |
1316 | rcu_read_unlock(); |
1317 | goto xdp_xmit; | |
56434a01 | 1318 | default: |
c8064e5b | 1319 | bpf_warn_invalid_xdp_action(vi->dev, xdp_prog, act); |
df561f66 | 1320 | fallthrough; |
0354e4d1 JF |
1321 | case XDP_ABORTED: |
1322 | trace_xdp_exception(vi->dev, xdp_prog, act); | |
df561f66 | 1323 | fallthrough; |
0354e4d1 | 1324 | case XDP_DROP: |
fab89baf | 1325 | goto err_xdp_frags; |
56434a01 | 1326 | } |
22174f79 HQ |
1327 | err_xdp_frags: |
1328 | if (unlikely(xdp_page != page)) | |
1329 | __free_pages(xdp_page, 0); | |
1330 | ||
1331 | if (xdp_buff_has_frags(&xdp)) { | |
1332 | shinfo = xdp_get_shared_info_from_buff(&xdp); | |
1333 | for (i = 0; i < shinfo->nr_frags; i++) { | |
1334 | xdp_page = skb_frag_page(&shinfo->frags[i]); | |
1335 | put_page(xdp_page); | |
1336 | } | |
1337 | } | |
1338 | ||
1339 | goto err_xdp; | |
f600b690 JF |
1340 | } |
1341 | rcu_read_unlock(); | |
ab7db917 | 1342 | |
6213f07c | 1343 | skip_xdp: |
fa0f1ba7 | 1344 | head_skb = page_to_skb(vi, rq, page, offset, len, truesize, headroom); |
f600b690 | 1345 | curr_skb = head_skb; |
9ab86bbc | 1346 | |
8fc3b9e9 MT |
1347 | if (unlikely(!curr_skb)) |
1348 | goto err_skb; | |
9ab86bbc | 1349 | while (--num_buf) { |
8fc3b9e9 MT |
1350 | int num_skb_frags; |
1351 | ||
680557cf | 1352 | buf = virtqueue_get_buf_ctx(rq->vq, &len, &ctx); |
03e9f8a0 | 1353 | if (unlikely(!buf)) { |
8fc3b9e9 | 1354 | pr_debug("%s: rx error: %d buffers out of %d missing\n", |
fdd819b2 | 1355 | dev->name, num_buf, |
012873d0 MT |
1356 | virtio16_to_cpu(vi->vdev, |
1357 | hdr->num_buffers)); | |
8fc3b9e9 MT |
1358 | dev->stats.rx_length_errors++; |
1359 | goto err_buf; | |
3f2c31d9 | 1360 | } |
8fc3b9e9 | 1361 | |
d46eeeaf | 1362 | stats->bytes += len; |
8fc3b9e9 | 1363 | page = virt_to_head_page(buf); |
28b39bc7 JW |
1364 | |
1365 | truesize = mergeable_ctx_to_truesize(ctx); | |
ef75cb51 HQ |
1366 | headroom = mergeable_ctx_to_headroom(ctx); |
1367 | tailroom = headroom ? sizeof(struct skb_shared_info) : 0; | |
1368 | room = SKB_DATA_ALIGN(headroom + tailroom); | |
1369 | if (unlikely(len > truesize - room)) { | |
56da5fd0 | 1370 | pr_debug("%s: rx error: len %u exceeds truesize %lu\n", |
ef75cb51 | 1371 | dev->name, len, (unsigned long)(truesize - room)); |
680557cf MT |
1372 | dev->stats.rx_length_errors++; |
1373 | goto err_skb; | |
1374 | } | |
8fc3b9e9 MT |
1375 | |
1376 | num_skb_frags = skb_shinfo(curr_skb)->nr_frags; | |
2613af0e MD |
1377 | if (unlikely(num_skb_frags == MAX_SKB_FRAGS)) { |
1378 | struct sk_buff *nskb = alloc_skb(0, GFP_ATOMIC); | |
8fc3b9e9 MT |
1379 | |
1380 | if (unlikely(!nskb)) | |
1381 | goto err_skb; | |
2613af0e MD |
1382 | if (curr_skb == head_skb) |
1383 | skb_shinfo(curr_skb)->frag_list = nskb; | |
1384 | else | |
1385 | curr_skb->next = nskb; | |
1386 | curr_skb = nskb; | |
1387 | head_skb->truesize += nskb->truesize; | |
1388 | num_skb_frags = 0; | |
1389 | } | |
1390 | if (curr_skb != head_skb) { | |
1391 | head_skb->data_len += len; | |
1392 | head_skb->len += len; | |
fb51879d | 1393 | head_skb->truesize += truesize; |
2613af0e | 1394 | } |
8fc3b9e9 | 1395 | offset = buf - page_address(page); |
ba275241 JW |
1396 | if (skb_can_coalesce(curr_skb, num_skb_frags, page, offset)) { |
1397 | put_page(page); | |
1398 | skb_coalesce_rx_frag(curr_skb, num_skb_frags - 1, | |
fb51879d | 1399 | len, truesize); |
ba275241 JW |
1400 | } else { |
1401 | skb_add_rx_frag(curr_skb, num_skb_frags, page, | |
fb51879d | 1402 | offset, len, truesize); |
ba275241 | 1403 | } |
8fc3b9e9 MT |
1404 | } |
1405 | ||
5377d758 | 1406 | ewma_pkt_len_add(&rq->mrg_avg_pkt_len, head_skb->len); |
8fc3b9e9 MT |
1407 | return head_skb; |
1408 | ||
f600b690 JF |
1409 | err_xdp: |
1410 | rcu_read_unlock(); | |
d46eeeaf | 1411 | stats->xdp_drops++; |
8fc3b9e9 MT |
1412 | err_skb: |
1413 | put_page(page); | |
850e088d | 1414 | while (num_buf-- > 1) { |
680557cf MT |
1415 | buf = virtqueue_get_buf(rq->vq, &len); |
1416 | if (unlikely(!buf)) { | |
8fc3b9e9 MT |
1417 | pr_debug("%s: rx error: %d buffers missing\n", |
1418 | dev->name, num_buf); | |
1419 | dev->stats.rx_length_errors++; | |
1420 | break; | |
1421 | } | |
d46eeeaf | 1422 | stats->bytes += len; |
680557cf | 1423 | page = virt_to_head_page(buf); |
8fc3b9e9 | 1424 | put_page(page); |
9ab86bbc | 1425 | } |
8fc3b9e9 | 1426 | err_buf: |
d46eeeaf | 1427 | stats->drops++; |
8fc3b9e9 | 1428 | dev_kfree_skb(head_skb); |
56434a01 | 1429 | xdp_xmit: |
8fc3b9e9 | 1430 | return NULL; |
9ab86bbc SM |
1431 | } |
1432 | ||
91f41f01 AM |
1433 | static void virtio_skb_set_hash(const struct virtio_net_hdr_v1_hash *hdr_hash, |
1434 | struct sk_buff *skb) | |
1435 | { | |
1436 | enum pkt_hash_types rss_hash_type; | |
1437 | ||
1438 | if (!hdr_hash || !skb) | |
1439 | return; | |
1440 | ||
95bb6330 | 1441 | switch (__le16_to_cpu(hdr_hash->hash_report)) { |
91f41f01 AM |
1442 | case VIRTIO_NET_HASH_REPORT_TCPv4: |
1443 | case VIRTIO_NET_HASH_REPORT_UDPv4: | |
1444 | case VIRTIO_NET_HASH_REPORT_TCPv6: | |
1445 | case VIRTIO_NET_HASH_REPORT_UDPv6: | |
1446 | case VIRTIO_NET_HASH_REPORT_TCPv6_EX: | |
1447 | case VIRTIO_NET_HASH_REPORT_UDPv6_EX: | |
1448 | rss_hash_type = PKT_HASH_TYPE_L4; | |
1449 | break; | |
1450 | case VIRTIO_NET_HASH_REPORT_IPv4: | |
1451 | case VIRTIO_NET_HASH_REPORT_IPv6: | |
1452 | case VIRTIO_NET_HASH_REPORT_IPv6_EX: | |
1453 | rss_hash_type = PKT_HASH_TYPE_L3; | |
1454 | break; | |
1455 | case VIRTIO_NET_HASH_REPORT_NONE: | |
1456 | default: | |
1457 | rss_hash_type = PKT_HASH_TYPE_NONE; | |
1458 | } | |
95bb6330 | 1459 | skb_set_hash(skb, __le32_to_cpu(hdr_hash->hash_value), rss_hash_type); |
91f41f01 AM |
1460 | } |
1461 | ||
7d9d60fd TM |
1462 | static void receive_buf(struct virtnet_info *vi, struct receive_queue *rq, |
1463 | void *buf, unsigned int len, void **ctx, | |
a0929a44 | 1464 | unsigned int *xdp_xmit, |
d46eeeaf | 1465 | struct virtnet_rq_stats *stats) |
9ab86bbc | 1466 | { |
e9d7417b | 1467 | struct net_device *dev = vi->dev; |
9ab86bbc | 1468 | struct sk_buff *skb; |
012873d0 | 1469 | struct virtio_net_hdr_mrg_rxbuf *hdr; |
3f2c31d9 | 1470 | |
bcff3162 | 1471 | if (unlikely(len < vi->hdr_len + ETH_HLEN)) { |
9ab86bbc SM |
1472 | pr_debug("%s: short packet %i\n", dev->name, len); |
1473 | dev->stats.rx_length_errors++; | |
eb1d929f | 1474 | virtnet_rq_free_unused_buf(rq->vq, buf); |
7d9d60fd | 1475 | return; |
9ab86bbc | 1476 | } |
3f2c31d9 | 1477 | |
f121159d | 1478 | if (vi->mergeable_rx_bufs) |
7d9d60fd | 1479 | skb = receive_mergeable(dev, vi, rq, buf, ctx, len, xdp_xmit, |
a0929a44 | 1480 | stats); |
f121159d | 1481 | else if (vi->big_packets) |
a0929a44 | 1482 | skb = receive_big(dev, vi, rq, buf, len, stats); |
f121159d | 1483 | else |
a0929a44 | 1484 | skb = receive_small(dev, vi, rq, buf, ctx, len, xdp_xmit, stats); |
f121159d MT |
1485 | |
1486 | if (unlikely(!skb)) | |
7d9d60fd | 1487 | return; |
3f2c31d9 | 1488 | |
9ab86bbc | 1489 | hdr = skb_vnet_hdr(skb); |
91f41f01 AM |
1490 | if (dev->features & NETIF_F_RXHASH && vi->has_rss_hash_report) |
1491 | virtio_skb_set_hash((const struct virtio_net_hdr_v1_hash *)hdr, skb); | |
3fa2a1df | 1492 | |
e858fae2 | 1493 | if (hdr->hdr.flags & VIRTIO_NET_HDR_F_DATA_VALID) |
10a8d94a | 1494 | skb->ip_summed = CHECKSUM_UNNECESSARY; |
296f96fc | 1495 | |
e858fae2 MR |
1496 | if (virtio_net_hdr_to_skb(skb, &hdr->hdr, |
1497 | virtio_is_little_endian(vi->vdev))) { | |
1498 | net_warn_ratelimited("%s: bad gso: type: %u, size: %u\n", | |
1499 | dev->name, hdr->hdr.gso_type, | |
1500 | hdr->hdr.gso_size); | |
1501 | goto frame_err; | |
296f96fc RR |
1502 | } |
1503 | ||
133bbb18 | 1504 | skb_record_rx_queue(skb, vq2rxq(rq->vq)); |
d1dc06dc MR |
1505 | skb->protocol = eth_type_trans(skb, dev); |
1506 | pr_debug("Receiving skb proto 0x%04x len %i type %i\n", | |
1507 | ntohs(skb->protocol), skb->len, skb->pkt_type); | |
1508 | ||
0fbd050a | 1509 | napi_gro_receive(&rq->napi, skb); |
7d9d60fd | 1510 | return; |
296f96fc RR |
1511 | |
1512 | frame_err: | |
1513 | dev->stats.rx_frame_errors++; | |
296f96fc RR |
1514 | dev_kfree_skb(skb); |
1515 | } | |
1516 | ||
192f68cf JW |
1517 | /* Unlike mergeable buffers, all buffers are allocated to the |
1518 | * same size, except for the headroom. For this reason we do | |
1519 | * not need to use mergeable_len_to_ctx here - it is enough | |
1520 | * to store the headroom as the context ignoring the truesize. | |
1521 | */ | |
946fa564 MT |
1522 | static int add_recvbuf_small(struct virtnet_info *vi, struct receive_queue *rq, |
1523 | gfp_t gfp) | |
296f96fc | 1524 | { |
f6b10209 JW |
1525 | struct page_frag *alloc_frag = &rq->alloc_frag; |
1526 | char *buf; | |
2de2f7f4 | 1527 | unsigned int xdp_headroom = virtnet_get_headroom(vi); |
192f68cf | 1528 | void *ctx = (void *)(unsigned long)xdp_headroom; |
f6b10209 | 1529 | int len = vi->hdr_len + VIRTNET_RX_PAD + GOOD_PACKET_LEN + xdp_headroom; |
9ab86bbc | 1530 | int err; |
3f2c31d9 | 1531 | |
f6b10209 JW |
1532 | len = SKB_DATA_ALIGN(len) + |
1533 | SKB_DATA_ALIGN(sizeof(struct skb_shared_info)); | |
1534 | if (unlikely(!skb_page_frag_refill(len, alloc_frag, gfp))) | |
9ab86bbc | 1535 | return -ENOMEM; |
296f96fc | 1536 | |
f6b10209 JW |
1537 | buf = (char *)page_address(alloc_frag->page) + alloc_frag->offset; |
1538 | get_page(alloc_frag->page); | |
1539 | alloc_frag->offset += len; | |
1540 | sg_init_one(rq->sg, buf + VIRTNET_RX_PAD + xdp_headroom, | |
1541 | vi->hdr_len + GOOD_PACKET_LEN); | |
192f68cf | 1542 | err = virtqueue_add_inbuf_ctx(rq->vq, rq->sg, 1, buf, ctx, gfp); |
9ab86bbc | 1543 | if (err < 0) |
f6b10209 | 1544 | put_page(virt_to_head_page(buf)); |
9ab86bbc SM |
1545 | return err; |
1546 | } | |
97402b96 | 1547 | |
012873d0 MT |
1548 | static int add_recvbuf_big(struct virtnet_info *vi, struct receive_queue *rq, |
1549 | gfp_t gfp) | |
9ab86bbc | 1550 | { |
9ab86bbc SM |
1551 | struct page *first, *list = NULL; |
1552 | char *p; | |
1553 | int i, err, offset; | |
1554 | ||
4959aebb | 1555 | sg_init_table(rq->sg, vi->big_packets_num_skbfrags + 2); |
a5835440 | 1556 | |
4959aebb GL |
1557 | /* page in rq->sg[vi->big_packets_num_skbfrags + 1] is list tail */ |
1558 | for (i = vi->big_packets_num_skbfrags + 1; i > 1; --i) { | |
e9d7417b | 1559 | first = get_a_page(rq, gfp); |
9ab86bbc SM |
1560 | if (!first) { |
1561 | if (list) | |
e9d7417b | 1562 | give_pages(rq, list); |
9ab86bbc | 1563 | return -ENOMEM; |
97402b96 | 1564 | } |
e9d7417b | 1565 | sg_set_buf(&rq->sg[i], page_address(first), PAGE_SIZE); |
97402b96 | 1566 | |
9ab86bbc SM |
1567 | /* chain new page in list head to match sg */ |
1568 | first->private = (unsigned long)list; | |
1569 | list = first; | |
1570 | } | |
296f96fc | 1571 | |
e9d7417b | 1572 | first = get_a_page(rq, gfp); |
9ab86bbc | 1573 | if (!first) { |
e9d7417b | 1574 | give_pages(rq, list); |
9ab86bbc SM |
1575 | return -ENOMEM; |
1576 | } | |
1577 | p = page_address(first); | |
1578 | ||
e9d7417b | 1579 | /* rq->sg[0], rq->sg[1] share the same page */ |
012873d0 MT |
1580 | /* a separated rq->sg[0] for header - required in case !any_header_sg */ |
1581 | sg_set_buf(&rq->sg[0], p, vi->hdr_len); | |
9ab86bbc | 1582 | |
e9d7417b | 1583 | /* rq->sg[1] for data packet, from offset */ |
9ab86bbc | 1584 | offset = sizeof(struct padded_vnet_hdr); |
e9d7417b | 1585 | sg_set_buf(&rq->sg[1], p + offset, PAGE_SIZE - offset); |
9ab86bbc SM |
1586 | |
1587 | /* chain first in list head */ | |
1588 | first->private = (unsigned long)list; | |
4959aebb | 1589 | err = virtqueue_add_inbuf(rq->vq, rq->sg, vi->big_packets_num_skbfrags + 2, |
9dc7b9e4 | 1590 | first, gfp); |
9ab86bbc | 1591 | if (err < 0) |
e9d7417b | 1592 | give_pages(rq, first); |
9ab86bbc SM |
1593 | |
1594 | return err; | |
296f96fc RR |
1595 | } |
1596 | ||
d85b758f | 1597 | static unsigned int get_mergeable_buf_len(struct receive_queue *rq, |
3cc81a9a JW |
1598 | struct ewma_pkt_len *avg_pkt_len, |
1599 | unsigned int room) | |
3f2c31d9 | 1600 | { |
c1ddc42d AM |
1601 | struct virtnet_info *vi = rq->vq->vdev->priv; |
1602 | const size_t hdr_len = vi->hdr_len; | |
fbf28d78 MD |
1603 | unsigned int len; |
1604 | ||
3cc81a9a JW |
1605 | if (room) |
1606 | return PAGE_SIZE - room; | |
1607 | ||
1608 | len = hdr_len + clamp_t(unsigned int, ewma_pkt_len_read(avg_pkt_len), | |
f0c3192c | 1609 | rq->min_buf_len, PAGE_SIZE - hdr_len); |
3cc81a9a | 1610 | |
e377fcc8 | 1611 | return ALIGN(len, L1_CACHE_BYTES); |
fbf28d78 MD |
1612 | } |
1613 | ||
2de2f7f4 JF |
1614 | static int add_recvbuf_mergeable(struct virtnet_info *vi, |
1615 | struct receive_queue *rq, gfp_t gfp) | |
fbf28d78 | 1616 | { |
fb51879d | 1617 | struct page_frag *alloc_frag = &rq->alloc_frag; |
2de2f7f4 | 1618 | unsigned int headroom = virtnet_get_headroom(vi); |
3cc81a9a JW |
1619 | unsigned int tailroom = headroom ? sizeof(struct skb_shared_info) : 0; |
1620 | unsigned int room = SKB_DATA_ALIGN(headroom + tailroom); | |
fb51879d | 1621 | char *buf; |
680557cf | 1622 | void *ctx; |
3f2c31d9 | 1623 | int err; |
fb51879d | 1624 | unsigned int len, hole; |
3f2c31d9 | 1625 | |
3cc81a9a JW |
1626 | /* Extra tailroom is needed to satisfy XDP's assumption. This |
1627 | * means rx frags coalescing won't work, but consider we've | |
1628 | * disabled GSO for XDP, it won't be a big issue. | |
1629 | */ | |
1630 | len = get_mergeable_buf_len(rq, &rq->mrg_avg_pkt_len, room); | |
1631 | if (unlikely(!skb_page_frag_refill(len + room, alloc_frag, gfp))) | |
9ab86bbc | 1632 | return -ENOMEM; |
ab7db917 | 1633 | |
fb51879d | 1634 | buf = (char *)page_address(alloc_frag->page) + alloc_frag->offset; |
2de2f7f4 | 1635 | buf += headroom; /* advance address leaving hole at front of pkt */ |
fb51879d | 1636 | get_page(alloc_frag->page); |
3cc81a9a | 1637 | alloc_frag->offset += len + room; |
fb51879d | 1638 | hole = alloc_frag->size - alloc_frag->offset; |
3cc81a9a | 1639 | if (hole < len + room) { |
ab7db917 MD |
1640 | /* To avoid internal fragmentation, if there is very likely not |
1641 | * enough space for another buffer, add the remaining space to | |
1daa8790 | 1642 | * the current buffer. |
484beac2 HQ |
1643 | * XDP core assumes that frame_size of xdp_buff and the length |
1644 | * of the frag are PAGE_SIZE, so we disable the hole mechanism. | |
ab7db917 | 1645 | */ |
484beac2 HQ |
1646 | if (!headroom) |
1647 | len += hole; | |
fb51879d MD |
1648 | alloc_frag->offset += hole; |
1649 | } | |
3f2c31d9 | 1650 | |
fb51879d | 1651 | sg_init_one(rq->sg, buf, len); |
ef75cb51 | 1652 | ctx = mergeable_len_to_ctx(len + room, headroom); |
680557cf | 1653 | err = virtqueue_add_inbuf_ctx(rq->vq, rq->sg, 1, buf, ctx, gfp); |
9ab86bbc | 1654 | if (err < 0) |
2613af0e | 1655 | put_page(virt_to_head_page(buf)); |
3f2c31d9 | 1656 | |
9ab86bbc SM |
1657 | return err; |
1658 | } | |
3f2c31d9 | 1659 | |
b2baed69 RR |
1660 | /* |
1661 | * Returns false if we couldn't fill entirely (OOM). | |
1662 | * | |
1663 | * Normally run in the receive path, but can also be run from ndo_open | |
1664 | * before we're receiving packets, or from refill_work which is | |
1665 | * careful to disable receiving (using napi_disable). | |
1666 | */ | |
946fa564 MT |
1667 | static bool try_fill_recv(struct virtnet_info *vi, struct receive_queue *rq, |
1668 | gfp_t gfp) | |
9ab86bbc SM |
1669 | { |
1670 | int err; | |
1788f495 | 1671 | bool oom; |
3f2c31d9 | 1672 | |
9ab86bbc SM |
1673 | do { |
1674 | if (vi->mergeable_rx_bufs) | |
2de2f7f4 | 1675 | err = add_recvbuf_mergeable(vi, rq, gfp); |
9ab86bbc | 1676 | else if (vi->big_packets) |
012873d0 | 1677 | err = add_recvbuf_big(vi, rq, gfp); |
9ab86bbc | 1678 | else |
946fa564 | 1679 | err = add_recvbuf_small(vi, rq, gfp); |
3f2c31d9 | 1680 | |
1788f495 | 1681 | oom = err == -ENOMEM; |
9ed4cb07 | 1682 | if (err) |
3f2c31d9 | 1683 | break; |
b7dfde95 | 1684 | } while (rq->vq->num_free); |
461f03dc | 1685 | if (virtqueue_kick_prepare(rq->vq) && virtqueue_notify(rq->vq)) { |
01c32598 MT |
1686 | unsigned long flags; |
1687 | ||
1688 | flags = u64_stats_update_begin_irqsave(&rq->stats.syncp); | |
d46eeeaf | 1689 | rq->stats.kicks++; |
01c32598 | 1690 | u64_stats_update_end_irqrestore(&rq->stats.syncp, flags); |
461f03dc TM |
1691 | } |
1692 | ||
3161e453 | 1693 | return !oom; |
3f2c31d9 MM |
1694 | } |
1695 | ||
18445c4d | 1696 | static void skb_recv_done(struct virtqueue *rvq) |
296f96fc RR |
1697 | { |
1698 | struct virtnet_info *vi = rvq->vdev->priv; | |
986a4f4d | 1699 | struct receive_queue *rq = &vi->rq[vq2rxq(rvq)]; |
e9d7417b | 1700 | |
e4e8452a | 1701 | virtqueue_napi_schedule(&rq->napi, rvq); |
296f96fc RR |
1702 | } |
1703 | ||
e4e8452a | 1704 | static void virtnet_napi_enable(struct virtqueue *vq, struct napi_struct *napi) |
3e9d08ec | 1705 | { |
e4e8452a | 1706 | napi_enable(napi); |
3e9d08ec BR |
1707 | |
1708 | /* If all buffers were filled by other side before we napi_enabled, we | |
e4e8452a WB |
1709 | * won't get another interrupt, so process any outstanding packets now. |
1710 | * Call local_bh_enable after to trigger softIRQ processing. | |
1711 | */ | |
1712 | local_bh_disable(); | |
1713 | virtqueue_napi_schedule(napi, vq); | |
1714 | local_bh_enable(); | |
3e9d08ec BR |
1715 | } |
1716 | ||
b92f1e67 WB |
1717 | static void virtnet_napi_tx_enable(struct virtnet_info *vi, |
1718 | struct virtqueue *vq, | |
1719 | struct napi_struct *napi) | |
1720 | { | |
1721 | if (!napi->weight) | |
1722 | return; | |
1723 | ||
1724 | /* Tx napi touches cachelines on the cpu handling tx interrupts. Only | |
1725 | * enable the feature if this is likely affine with the transmit path. | |
1726 | */ | |
1727 | if (!vi->affinity_hint_set) { | |
1728 | napi->weight = 0; | |
1729 | return; | |
1730 | } | |
1731 | ||
1732 | return virtnet_napi_enable(vq, napi); | |
1733 | } | |
1734 | ||
78a57b48 WB |
1735 | static void virtnet_napi_tx_disable(struct napi_struct *napi) |
1736 | { | |
1737 | if (napi->weight) | |
1738 | napi_disable(napi); | |
1739 | } | |
1740 | ||
3161e453 RR |
1741 | static void refill_work(struct work_struct *work) |
1742 | { | |
e9d7417b JW |
1743 | struct virtnet_info *vi = |
1744 | container_of(work, struct virtnet_info, refill.work); | |
3161e453 | 1745 | bool still_empty; |
986a4f4d JW |
1746 | int i; |
1747 | ||
55257d72 | 1748 | for (i = 0; i < vi->curr_queue_pairs; i++) { |
986a4f4d | 1749 | struct receive_queue *rq = &vi->rq[i]; |
3161e453 | 1750 | |
986a4f4d | 1751 | napi_disable(&rq->napi); |
946fa564 | 1752 | still_empty = !try_fill_recv(vi, rq, GFP_KERNEL); |
e4e8452a | 1753 | virtnet_napi_enable(rq->vq, &rq->napi); |
3161e453 | 1754 | |
986a4f4d JW |
1755 | /* In theory, this can happen: if we don't get any buffers in |
1756 | * we will *never* try to fill again. | |
1757 | */ | |
1758 | if (still_empty) | |
1759 | schedule_delayed_work(&vi->refill, HZ/2); | |
1760 | } | |
3161e453 RR |
1761 | } |
1762 | ||
2471c75e JDB |
1763 | static int virtnet_receive(struct receive_queue *rq, int budget, |
1764 | unsigned int *xdp_xmit) | |
296f96fc | 1765 | { |
e9d7417b | 1766 | struct virtnet_info *vi = rq->vq->vdev->priv; |
d46eeeaf | 1767 | struct virtnet_rq_stats stats = {}; |
a0929a44 | 1768 | unsigned int len; |
9ab86bbc | 1769 | void *buf; |
a0929a44 | 1770 | int i; |
296f96fc | 1771 | |
192f68cf | 1772 | if (!vi->big_packets || vi->mergeable_rx_bufs) { |
680557cf MT |
1773 | void *ctx; |
1774 | ||
d46eeeaf | 1775 | while (stats.packets < budget && |
680557cf | 1776 | (buf = virtqueue_get_buf_ctx(rq->vq, &len, &ctx))) { |
a0929a44 | 1777 | receive_buf(vi, rq, buf, len, ctx, xdp_xmit, &stats); |
d46eeeaf | 1778 | stats.packets++; |
680557cf MT |
1779 | } |
1780 | } else { | |
d46eeeaf | 1781 | while (stats.packets < budget && |
680557cf | 1782 | (buf = virtqueue_get_buf(rq->vq, &len)) != NULL) { |
a0929a44 | 1783 | receive_buf(vi, rq, buf, len, NULL, xdp_xmit, &stats); |
d46eeeaf | 1784 | stats.packets++; |
680557cf | 1785 | } |
296f96fc RR |
1786 | } |
1787 | ||
718be6ba | 1788 | if (rq->vq->num_free > min((unsigned int)budget, virtqueue_get_vring_size(rq->vq)) / 2) { |
5a159128 JW |
1789 | if (!try_fill_recv(vi, rq, GFP_ATOMIC)) { |
1790 | spin_lock(&vi->refill_lock); | |
1791 | if (vi->refill_enabled) | |
1792 | schedule_delayed_work(&vi->refill, 0); | |
1793 | spin_unlock(&vi->refill_lock); | |
1794 | } | |
3161e453 | 1795 | } |
296f96fc | 1796 | |
d7dfc5cf | 1797 | u64_stats_update_begin(&rq->stats.syncp); |
a0929a44 TM |
1798 | for (i = 0; i < VIRTNET_RQ_STATS_LEN; i++) { |
1799 | size_t offset = virtnet_rq_stats_desc[i].offset; | |
1800 | u64 *item; | |
1801 | ||
d46eeeaf JW |
1802 | item = (u64 *)((u8 *)&rq->stats + offset); |
1803 | *item += *(u64 *)((u8 *)&stats + offset); | |
a0929a44 | 1804 | } |
d7dfc5cf | 1805 | u64_stats_update_end(&rq->stats.syncp); |
61845d20 | 1806 | |
d46eeeaf | 1807 | return stats.packets; |
2ffa7598 JW |
1808 | } |
1809 | ||
7b0411ef WB |
1810 | static void virtnet_poll_cleantx(struct receive_queue *rq) |
1811 | { | |
1812 | struct virtnet_info *vi = rq->vq->vdev->priv; | |
1813 | unsigned int index = vq2rxq(rq->vq); | |
1814 | struct send_queue *sq = &vi->sq[index]; | |
1815 | struct netdev_queue *txq = netdev_get_tx_queue(vi->dev, index); | |
1816 | ||
534da5e8 | 1817 | if (!sq->napi.weight || is_xdp_raw_buffer_queue(vi, index)) |
7b0411ef WB |
1818 | return; |
1819 | ||
1820 | if (__netif_tx_trylock(txq)) { | |
ebcce492 XZ |
1821 | if (sq->reset) { |
1822 | __netif_tx_unlock(txq); | |
1823 | return; | |
1824 | } | |
1825 | ||
a7766ef1 MT |
1826 | do { |
1827 | virtqueue_disable_cb(sq->vq); | |
1828 | free_old_xmit_skbs(sq, true); | |
1829 | } while (unlikely(!virtqueue_enable_cb_delayed(sq->vq))); | |
22bc63c5 MT |
1830 | |
1831 | if (sq->vq->num_free >= 2 + MAX_SKB_FRAGS) | |
1832 | netif_tx_wake_queue(txq); | |
1833 | ||
7b0411ef WB |
1834 | __netif_tx_unlock(txq); |
1835 | } | |
7b0411ef WB |
1836 | } |
1837 | ||
2ffa7598 JW |
1838 | static int virtnet_poll(struct napi_struct *napi, int budget) |
1839 | { | |
1840 | struct receive_queue *rq = | |
1841 | container_of(napi, struct receive_queue, napi); | |
9267c430 JW |
1842 | struct virtnet_info *vi = rq->vq->vdev->priv; |
1843 | struct send_queue *sq; | |
2a43565c | 1844 | unsigned int received; |
2471c75e | 1845 | unsigned int xdp_xmit = 0; |
2ffa7598 | 1846 | |
7b0411ef WB |
1847 | virtnet_poll_cleantx(rq); |
1848 | ||
186b3c99 | 1849 | received = virtnet_receive(rq, budget, &xdp_xmit); |
2ffa7598 | 1850 | |
ad7e615f MK |
1851 | if (xdp_xmit & VIRTIO_XDP_REDIR) |
1852 | xdp_do_flush(); | |
1853 | ||
8329d98e | 1854 | /* Out of packets? */ |
e4e8452a WB |
1855 | if (received < budget) |
1856 | virtqueue_napi_complete(napi, rq->vq, received); | |
296f96fc | 1857 | |
2471c75e | 1858 | if (xdp_xmit & VIRTIO_XDP_TX) { |
97c2c69e | 1859 | sq = virtnet_xdp_get_sq(vi); |
461f03dc TM |
1860 | if (virtqueue_kick_prepare(sq->vq) && virtqueue_notify(sq->vq)) { |
1861 | u64_stats_update_begin(&sq->stats.syncp); | |
1862 | sq->stats.kicks++; | |
1863 | u64_stats_update_end(&sq->stats.syncp); | |
1864 | } | |
97c2c69e | 1865 | virtnet_xdp_put_sq(vi, sq); |
9267c430 | 1866 | } |
186b3c99 | 1867 | |
296f96fc RR |
1868 | return received; |
1869 | } | |
1870 | ||
986a4f4d JW |
1871 | static int virtnet_open(struct net_device *dev) |
1872 | { | |
1873 | struct virtnet_info *vi = netdev_priv(dev); | |
754b8a21 | 1874 | int i, err; |
986a4f4d | 1875 | |
5a159128 JW |
1876 | enable_delayed_refill(vi); |
1877 | ||
e4166625 JW |
1878 | for (i = 0; i < vi->max_queue_pairs; i++) { |
1879 | if (i < vi->curr_queue_pairs) | |
1880 | /* Make sure we have some buffers: if oom use wq. */ | |
946fa564 | 1881 | if (!try_fill_recv(vi, &vi->rq[i], GFP_KERNEL)) |
e4166625 | 1882 | schedule_delayed_work(&vi->refill, 0); |
754b8a21 | 1883 | |
b02e5a0e | 1884 | err = xdp_rxq_info_reg(&vi->rq[i].xdp_rxq, dev, i, vi->rq[i].napi.napi_id); |
754b8a21 JDB |
1885 | if (err < 0) |
1886 | return err; | |
1887 | ||
8d5d8852 JDB |
1888 | err = xdp_rxq_info_reg_mem_model(&vi->rq[i].xdp_rxq, |
1889 | MEM_TYPE_PAGE_SHARED, NULL); | |
1890 | if (err < 0) { | |
1891 | xdp_rxq_info_unreg(&vi->rq[i].xdp_rxq); | |
1892 | return err; | |
1893 | } | |
1894 | ||
e4e8452a | 1895 | virtnet_napi_enable(vi->rq[i].vq, &vi->rq[i].napi); |
b92f1e67 | 1896 | virtnet_napi_tx_enable(vi, vi->sq[i].vq, &vi->sq[i].napi); |
986a4f4d JW |
1897 | } |
1898 | ||
1899 | return 0; | |
1900 | } | |
1901 | ||
b92f1e67 WB |
1902 | static int virtnet_poll_tx(struct napi_struct *napi, int budget) |
1903 | { | |
1904 | struct send_queue *sq = container_of(napi, struct send_queue, napi); | |
1905 | struct virtnet_info *vi = sq->vq->vdev->priv; | |
534da5e8 TM |
1906 | unsigned int index = vq2txq(sq->vq); |
1907 | struct netdev_queue *txq; | |
5a2f966d MT |
1908 | int opaque; |
1909 | bool done; | |
b92f1e67 | 1910 | |
534da5e8 TM |
1911 | if (unlikely(is_xdp_raw_buffer_queue(vi, index))) { |
1912 | /* We don't need to enable cb for XDP */ | |
1913 | napi_complete_done(napi, 0); | |
1914 | return 0; | |
1915 | } | |
1916 | ||
1917 | txq = netdev_get_tx_queue(vi->dev, index); | |
b92f1e67 | 1918 | __netif_tx_lock(txq, raw_smp_processor_id()); |
5a2f966d | 1919 | virtqueue_disable_cb(sq->vq); |
df133f3f | 1920 | free_old_xmit_skbs(sq, true); |
5a2f966d | 1921 | |
22bc63c5 MT |
1922 | if (sq->vq->num_free >= 2 + MAX_SKB_FRAGS) |
1923 | netif_tx_wake_queue(txq); | |
1924 | ||
5a2f966d MT |
1925 | opaque = virtqueue_enable_cb_prepare(sq->vq); |
1926 | ||
1927 | done = napi_complete_done(napi, 0); | |
1928 | ||
1929 | if (!done) | |
1930 | virtqueue_disable_cb(sq->vq); | |
1931 | ||
b92f1e67 WB |
1932 | __netif_tx_unlock(txq); |
1933 | ||
5a2f966d MT |
1934 | if (done) { |
1935 | if (unlikely(virtqueue_poll(sq->vq, opaque))) { | |
1936 | if (napi_schedule_prep(napi)) { | |
1937 | __netif_tx_lock(txq, raw_smp_processor_id()); | |
1938 | virtqueue_disable_cb(sq->vq); | |
1939 | __netif_tx_unlock(txq); | |
1940 | __napi_schedule(napi); | |
1941 | } | |
1942 | } | |
1943 | } | |
b92f1e67 | 1944 | |
b92f1e67 WB |
1945 | return 0; |
1946 | } | |
1947 | ||
e9d7417b | 1948 | static int xmit_skb(struct send_queue *sq, struct sk_buff *skb) |
296f96fc | 1949 | { |
012873d0 | 1950 | struct virtio_net_hdr_mrg_rxbuf *hdr; |
296f96fc | 1951 | const unsigned char *dest = ((struct ethhdr *)skb->data)->h_dest; |
e9d7417b | 1952 | struct virtnet_info *vi = sq->vq->vdev->priv; |
e2fcad58 | 1953 | int num_sg; |
012873d0 | 1954 | unsigned hdr_len = vi->hdr_len; |
e7428e95 | 1955 | bool can_push; |
296f96fc | 1956 | |
e174961c | 1957 | pr_debug("%s: xmit %p %pM\n", vi->dev->name, skb, dest); |
e7428e95 MT |
1958 | |
1959 | can_push = vi->any_header_sg && | |
1960 | !((unsigned long)skb->data & (__alignof__(*hdr) - 1)) && | |
1961 | !skb_header_cloned(skb) && skb_headroom(skb) >= hdr_len; | |
1962 | /* Even if we can, don't push here yet as this would skew | |
1963 | * csum_start offset below. */ | |
1964 | if (can_push) | |
012873d0 | 1965 | hdr = (struct virtio_net_hdr_mrg_rxbuf *)(skb->data - hdr_len); |
e7428e95 MT |
1966 | else |
1967 | hdr = skb_vnet_hdr(skb); | |
296f96fc | 1968 | |
e858fae2 | 1969 | if (virtio_net_hdr_from_skb(skb, &hdr->hdr, |
fd3a8862 WB |
1970 | virtio_is_little_endian(vi->vdev), false, |
1971 | 0)) | |
85eb1389 | 1972 | return -EPROTO; |
296f96fc | 1973 | |
3f2c31d9 | 1974 | if (vi->mergeable_rx_bufs) |
012873d0 | 1975 | hdr->num_buffers = 0; |
3f2c31d9 | 1976 | |
547c890c | 1977 | sg_init_table(sq->sg, skb_shinfo(skb)->nr_frags + (can_push ? 1 : 2)); |
e7428e95 MT |
1978 | if (can_push) { |
1979 | __skb_push(skb, hdr_len); | |
1980 | num_sg = skb_to_sgvec(skb, sq->sg, 0, skb->len); | |
e2fcad58 JD |
1981 | if (unlikely(num_sg < 0)) |
1982 | return num_sg; | |
e7428e95 MT |
1983 | /* Pull header back to avoid skew in tx bytes calculations. */ |
1984 | __skb_pull(skb, hdr_len); | |
1985 | } else { | |
1986 | sg_set_buf(sq->sg, hdr, hdr_len); | |
e2fcad58 JD |
1987 | num_sg = skb_to_sgvec(skb, sq->sg + 1, 0, skb->len); |
1988 | if (unlikely(num_sg < 0)) | |
1989 | return num_sg; | |
1990 | num_sg++; | |
e7428e95 | 1991 | } |
9dc7b9e4 | 1992 | return virtqueue_add_outbuf(sq->vq, sq->sg, num_sg, skb, GFP_ATOMIC); |
11a3a154 RR |
1993 | } |
1994 | ||
424efe9c | 1995 | static netdev_tx_t start_xmit(struct sk_buff *skb, struct net_device *dev) |
99ffc696 RR |
1996 | { |
1997 | struct virtnet_info *vi = netdev_priv(dev); | |
986a4f4d JW |
1998 | int qnum = skb_get_queue_mapping(skb); |
1999 | struct send_queue *sq = &vi->sq[qnum]; | |
9ed4cb07 | 2000 | int err; |
4b7fd2e6 | 2001 | struct netdev_queue *txq = netdev_get_tx_queue(dev, qnum); |
6b16f9ee | 2002 | bool kick = !netdev_xmit_more(); |
b92f1e67 | 2003 | bool use_napi = sq->napi.weight; |
2cb9c6ba | 2004 | |
2cb9c6ba | 2005 | /* Free up any pending old buffers before queueing new ones. */ |
a7766ef1 MT |
2006 | do { |
2007 | if (use_napi) | |
2008 | virtqueue_disable_cb(sq->vq); | |
2009 | ||
2010 | free_old_xmit_skbs(sq, false); | |
99ffc696 | 2011 | |
a7766ef1 MT |
2012 | } while (use_napi && kick && |
2013 | unlikely(!virtqueue_enable_cb_delayed(sq->vq))); | |
bdb12e0d | 2014 | |
074c3582 JK |
2015 | /* timestamp packet in software */ |
2016 | skb_tx_timestamp(skb); | |
2017 | ||
03f191ba | 2018 | /* Try to transmit */ |
b7dfde95 | 2019 | err = xmit_skb(sq, skb); |
48925e37 | 2020 | |
9ed4cb07 | 2021 | /* This should not happen! */ |
681daee2 | 2022 | if (unlikely(err)) { |
9ed4cb07 RR |
2023 | dev->stats.tx_fifo_errors++; |
2024 | if (net_ratelimit()) | |
2025 | dev_warn(&dev->dev, | |
7934b481 YS |
2026 | "Unexpected TXQ (%d) queue failure: %d\n", |
2027 | qnum, err); | |
58eba97d | 2028 | dev->stats.tx_dropped++; |
85e94525 | 2029 | dev_kfree_skb_any(skb); |
58eba97d | 2030 | return NETDEV_TX_OK; |
296f96fc | 2031 | } |
03f191ba | 2032 | |
48925e37 | 2033 | /* Don't wait up for transmitted skbs to be freed. */ |
b92f1e67 WB |
2034 | if (!use_napi) { |
2035 | skb_orphan(skb); | |
895b5c9f | 2036 | nf_reset_ct(skb); |
b92f1e67 | 2037 | } |
48925e37 | 2038 | |
b8ef4809 | 2039 | check_sq_full_and_disable(vi, dev, sq); |
48925e37 | 2040 | |
461f03dc TM |
2041 | if (kick || netif_xmit_stopped(txq)) { |
2042 | if (virtqueue_kick_prepare(sq->vq) && virtqueue_notify(sq->vq)) { | |
2043 | u64_stats_update_begin(&sq->stats.syncp); | |
2044 | sq->stats.kicks++; | |
2045 | u64_stats_update_end(&sq->stats.syncp); | |
2046 | } | |
2047 | } | |
296f96fc | 2048 | |
0b725a2c | 2049 | return NETDEV_TX_OK; |
c223a078 DM |
2050 | } |
2051 | ||
6a4763e2 XZ |
2052 | static int virtnet_rx_resize(struct virtnet_info *vi, |
2053 | struct receive_queue *rq, u32 ring_num) | |
2054 | { | |
2055 | bool running = netif_running(vi->dev); | |
2056 | int err, qindex; | |
2057 | ||
2058 | qindex = rq - vi->rq; | |
2059 | ||
2060 | if (running) | |
2061 | napi_disable(&rq->napi); | |
2062 | ||
2063 | err = virtqueue_resize(rq->vq, ring_num, virtnet_rq_free_unused_buf); | |
2064 | if (err) | |
2065 | netdev_err(vi->dev, "resize rx fail: rx queue index: %d err: %d\n", qindex, err); | |
2066 | ||
2067 | if (!try_fill_recv(vi, rq, GFP_KERNEL)) | |
2068 | schedule_delayed_work(&vi->refill, 0); | |
2069 | ||
2070 | if (running) | |
2071 | virtnet_napi_enable(rq->vq, &rq->napi); | |
2072 | return err; | |
2073 | } | |
2074 | ||
ebcce492 XZ |
2075 | static int virtnet_tx_resize(struct virtnet_info *vi, |
2076 | struct send_queue *sq, u32 ring_num) | |
2077 | { | |
2078 | bool running = netif_running(vi->dev); | |
2079 | struct netdev_queue *txq; | |
2080 | int err, qindex; | |
2081 | ||
2082 | qindex = sq - vi->sq; | |
2083 | ||
2084 | if (running) | |
2085 | virtnet_napi_tx_disable(&sq->napi); | |
2086 | ||
2087 | txq = netdev_get_tx_queue(vi->dev, qindex); | |
2088 | ||
2089 | /* 1. wait all ximt complete | |
2090 | * 2. fix the race of netif_stop_subqueue() vs netif_start_subqueue() | |
2091 | */ | |
2092 | __netif_tx_lock_bh(txq); | |
2093 | ||
2094 | /* Prevent rx poll from accessing sq. */ | |
2095 | sq->reset = true; | |
2096 | ||
2097 | /* Prevent the upper layer from trying to send packets. */ | |
2098 | netif_stop_subqueue(vi->dev, qindex); | |
2099 | ||
2100 | __netif_tx_unlock_bh(txq); | |
2101 | ||
2102 | err = virtqueue_resize(sq->vq, ring_num, virtnet_sq_free_unused_buf); | |
2103 | if (err) | |
2104 | netdev_err(vi->dev, "resize tx fail: tx queue index: %d err: %d\n", qindex, err); | |
2105 | ||
2106 | __netif_tx_lock_bh(txq); | |
2107 | sq->reset = false; | |
2108 | netif_tx_wake_queue(txq); | |
2109 | __netif_tx_unlock_bh(txq); | |
2110 | ||
2111 | if (running) | |
2112 | virtnet_napi_tx_enable(vi, sq->vq, &sq->napi); | |
2113 | return err; | |
2114 | } | |
2115 | ||
40cbfc37 AK |
2116 | /* |
2117 | * Send command via the control virtqueue and check status. Commands | |
2118 | * supported by the hypervisor, as indicated by feature bits, should | |
788a8b6d | 2119 | * never fail unless improperly formatted. |
40cbfc37 AK |
2120 | */ |
2121 | static bool virtnet_send_command(struct virtnet_info *vi, u8 class, u8 cmd, | |
d24bae32 | 2122 | struct scatterlist *out) |
40cbfc37 | 2123 | { |
f7bc9594 | 2124 | struct scatterlist *sgs[4], hdr, stat; |
d24bae32 | 2125 | unsigned out_num = 0, tmp; |
222722bc | 2126 | int ret; |
40cbfc37 AK |
2127 | |
2128 | /* Caller should know better */ | |
f7bc9594 | 2129 | BUG_ON(!virtio_has_feature(vi->vdev, VIRTIO_NET_F_CTRL_VQ)); |
40cbfc37 | 2130 | |
12e57169 MT |
2131 | vi->ctrl->status = ~0; |
2132 | vi->ctrl->hdr.class = class; | |
2133 | vi->ctrl->hdr.cmd = cmd; | |
f7bc9594 | 2134 | /* Add header */ |
12e57169 | 2135 | sg_init_one(&hdr, &vi->ctrl->hdr, sizeof(vi->ctrl->hdr)); |
f7bc9594 | 2136 | sgs[out_num++] = &hdr; |
40cbfc37 | 2137 | |
f7bc9594 RR |
2138 | if (out) |
2139 | sgs[out_num++] = out; | |
40cbfc37 | 2140 | |
f7bc9594 | 2141 | /* Add return status. */ |
12e57169 | 2142 | sg_init_one(&stat, &vi->ctrl->status, sizeof(vi->ctrl->status)); |
d24bae32 | 2143 | sgs[out_num] = &stat; |
40cbfc37 | 2144 | |
d24bae32 | 2145 | BUG_ON(out_num + 1 > ARRAY_SIZE(sgs)); |
222722bc YW |
2146 | ret = virtqueue_add_sgs(vi->cvq, sgs, out_num, 1, vi, GFP_ATOMIC); |
2147 | if (ret < 0) { | |
2148 | dev_warn(&vi->vdev->dev, | |
2149 | "Failed to add sgs for command vq: %d\n.", ret); | |
2150 | return false; | |
2151 | } | |
40cbfc37 | 2152 | |
67975901 | 2153 | if (unlikely(!virtqueue_kick(vi->cvq))) |
12e57169 | 2154 | return vi->ctrl->status == VIRTIO_NET_OK; |
40cbfc37 AK |
2155 | |
2156 | /* Spin for a response, the kick causes an ioport write, trapping | |
2157 | * into the hypervisor, so the request should be handled immediately. | |
2158 | */ | |
047b9b94 HG |
2159 | while (!virtqueue_get_buf(vi->cvq, &tmp) && |
2160 | !virtqueue_is_broken(vi->cvq)) | |
40cbfc37 AK |
2161 | cpu_relax(); |
2162 | ||
12e57169 | 2163 | return vi->ctrl->status == VIRTIO_NET_OK; |
40cbfc37 AK |
2164 | } |
2165 | ||
9c46f6d4 AW |
2166 | static int virtnet_set_mac_address(struct net_device *dev, void *p) |
2167 | { | |
2168 | struct virtnet_info *vi = netdev_priv(dev); | |
2169 | struct virtio_device *vdev = vi->vdev; | |
f2f2c8b4 | 2170 | int ret; |
e37e2ff3 | 2171 | struct sockaddr *addr; |
7e58d5ae | 2172 | struct scatterlist sg; |
9c46f6d4 | 2173 | |
ba5e4426 SS |
2174 | if (virtio_has_feature(vi->vdev, VIRTIO_NET_F_STANDBY)) |
2175 | return -EOPNOTSUPP; | |
2176 | ||
801822d1 | 2177 | addr = kmemdup(p, sizeof(*addr), GFP_KERNEL); |
e37e2ff3 AL |
2178 | if (!addr) |
2179 | return -ENOMEM; | |
e37e2ff3 AL |
2180 | |
2181 | ret = eth_prepare_mac_addr_change(dev, addr); | |
f2f2c8b4 | 2182 | if (ret) |
e37e2ff3 | 2183 | goto out; |
9c46f6d4 | 2184 | |
7e58d5ae AK |
2185 | if (virtio_has_feature(vdev, VIRTIO_NET_F_CTRL_MAC_ADDR)) { |
2186 | sg_init_one(&sg, addr->sa_data, dev->addr_len); | |
2187 | if (!virtnet_send_command(vi, VIRTIO_NET_CTRL_MAC, | |
d24bae32 | 2188 | VIRTIO_NET_CTRL_MAC_ADDR_SET, &sg)) { |
7e58d5ae AK |
2189 | dev_warn(&vdev->dev, |
2190 | "Failed to set mac address by vq command.\n"); | |
e37e2ff3 AL |
2191 | ret = -EINVAL; |
2192 | goto out; | |
7e58d5ae | 2193 | } |
7e93a02f MT |
2194 | } else if (virtio_has_feature(vdev, VIRTIO_NET_F_MAC) && |
2195 | !virtio_has_feature(vdev, VIRTIO_F_VERSION_1)) { | |
855e0c52 RR |
2196 | unsigned int i; |
2197 | ||
2198 | /* Naturally, this has an atomicity problem. */ | |
2199 | for (i = 0; i < dev->addr_len; i++) | |
2200 | virtio_cwrite8(vdev, | |
2201 | offsetof(struct virtio_net_config, mac) + | |
2202 | i, addr->sa_data[i]); | |
7e58d5ae AK |
2203 | } |
2204 | ||
2205 | eth_commit_mac_addr_change(dev, p); | |
e37e2ff3 | 2206 | ret = 0; |
9c46f6d4 | 2207 | |
e37e2ff3 AL |
2208 | out: |
2209 | kfree(addr); | |
2210 | return ret; | |
9c46f6d4 AW |
2211 | } |
2212 | ||
bc1f4470 | 2213 | static void virtnet_stats(struct net_device *dev, |
2214 | struct rtnl_link_stats64 *tot) | |
3fa2a1df | 2215 | { |
2216 | struct virtnet_info *vi = netdev_priv(dev); | |
3fa2a1df | 2217 | unsigned int start; |
d7dfc5cf | 2218 | int i; |
3fa2a1df | 2219 | |
d7dfc5cf | 2220 | for (i = 0; i < vi->max_queue_pairs; i++) { |
a520794b | 2221 | u64 tpackets, tbytes, terrors, rpackets, rbytes, rdrops; |
d7dfc5cf TM |
2222 | struct receive_queue *rq = &vi->rq[i]; |
2223 | struct send_queue *sq = &vi->sq[i]; | |
3fa2a1df | 2224 | |
2225 | do { | |
068c38ad | 2226 | start = u64_stats_fetch_begin(&sq->stats.syncp); |
d7dfc5cf TM |
2227 | tpackets = sq->stats.packets; |
2228 | tbytes = sq->stats.bytes; | |
a520794b | 2229 | terrors = sq->stats.tx_timeouts; |
068c38ad | 2230 | } while (u64_stats_fetch_retry(&sq->stats.syncp, start)); |
83a27052 ED |
2231 | |
2232 | do { | |
068c38ad | 2233 | start = u64_stats_fetch_begin(&rq->stats.syncp); |
d46eeeaf JW |
2234 | rpackets = rq->stats.packets; |
2235 | rbytes = rq->stats.bytes; | |
2236 | rdrops = rq->stats.drops; | |
068c38ad | 2237 | } while (u64_stats_fetch_retry(&rq->stats.syncp, start)); |
3fa2a1df | 2238 | |
2239 | tot->rx_packets += rpackets; | |
2240 | tot->tx_packets += tpackets; | |
2241 | tot->rx_bytes += rbytes; | |
2242 | tot->tx_bytes += tbytes; | |
2c4a2f7d | 2243 | tot->rx_dropped += rdrops; |
a520794b | 2244 | tot->tx_errors += terrors; |
3fa2a1df | 2245 | } |
2246 | ||
2247 | tot->tx_dropped = dev->stats.tx_dropped; | |
021ac8d3 | 2248 | tot->tx_fifo_errors = dev->stats.tx_fifo_errors; |
3fa2a1df | 2249 | tot->rx_length_errors = dev->stats.rx_length_errors; |
2250 | tot->rx_frame_errors = dev->stats.rx_frame_errors; | |
3fa2a1df | 2251 | } |
2252 | ||
586d17c5 JW |
2253 | static void virtnet_ack_link_announce(struct virtnet_info *vi) |
2254 | { | |
2255 | rtnl_lock(); | |
2256 | if (!virtnet_send_command(vi, VIRTIO_NET_CTRL_ANNOUNCE, | |
d24bae32 | 2257 | VIRTIO_NET_CTRL_ANNOUNCE_ACK, NULL)) |
586d17c5 JW |
2258 | dev_warn(&vi->dev->dev, "Failed to ack link announce.\n"); |
2259 | rtnl_unlock(); | |
2260 | } | |
2261 | ||
47315329 | 2262 | static int _virtnet_set_queues(struct virtnet_info *vi, u16 queue_pairs) |
986a4f4d JW |
2263 | { |
2264 | struct scatterlist sg; | |
986a4f4d JW |
2265 | struct net_device *dev = vi->dev; |
2266 | ||
2267 | if (!vi->has_cvq || !virtio_has_feature(vi->vdev, VIRTIO_NET_F_MQ)) | |
2268 | return 0; | |
2269 | ||
12e57169 MT |
2270 | vi->ctrl->mq.virtqueue_pairs = cpu_to_virtio16(vi->vdev, queue_pairs); |
2271 | sg_init_one(&sg, &vi->ctrl->mq, sizeof(vi->ctrl->mq)); | |
986a4f4d JW |
2272 | |
2273 | if (!virtnet_send_command(vi, VIRTIO_NET_CTRL_MQ, | |
d24bae32 | 2274 | VIRTIO_NET_CTRL_MQ_VQ_PAIRS_SET, &sg)) { |
986a4f4d JW |
2275 | dev_warn(&dev->dev, "Fail to set num of queue pairs to %d\n", |
2276 | queue_pairs); | |
2277 | return -EINVAL; | |
55257d72 | 2278 | } else { |
986a4f4d | 2279 | vi->curr_queue_pairs = queue_pairs; |
35ed159b JW |
2280 | /* virtnet_open() will refill when device is going to up. */ |
2281 | if (dev->flags & IFF_UP) | |
2282 | schedule_delayed_work(&vi->refill, 0); | |
55257d72 | 2283 | } |
986a4f4d JW |
2284 | |
2285 | return 0; | |
2286 | } | |
2287 | ||
47315329 JF |
2288 | static int virtnet_set_queues(struct virtnet_info *vi, u16 queue_pairs) |
2289 | { | |
2290 | int err; | |
2291 | ||
2292 | rtnl_lock(); | |
2293 | err = _virtnet_set_queues(vi, queue_pairs); | |
2294 | rtnl_unlock(); | |
2295 | return err; | |
2296 | } | |
2297 | ||
296f96fc RR |
2298 | static int virtnet_close(struct net_device *dev) |
2299 | { | |
2300 | struct virtnet_info *vi = netdev_priv(dev); | |
986a4f4d | 2301 | int i; |
296f96fc | 2302 | |
5a159128 JW |
2303 | /* Make sure NAPI doesn't schedule refill work */ |
2304 | disable_delayed_refill(vi); | |
b2baed69 RR |
2305 | /* Make sure refill_work doesn't re-enable napi! */ |
2306 | cancel_delayed_work_sync(&vi->refill); | |
986a4f4d | 2307 | |
b92f1e67 | 2308 | for (i = 0; i < vi->max_queue_pairs; i++) { |
27369c9c | 2309 | virtnet_napi_tx_disable(&vi->sq[i].napi); |
986a4f4d | 2310 | napi_disable(&vi->rq[i].napi); |
63b11404 | 2311 | xdp_rxq_info_unreg(&vi->rq[i].xdp_rxq); |
b92f1e67 | 2312 | } |
296f96fc | 2313 | |
296f96fc RR |
2314 | return 0; |
2315 | } | |
2316 | ||
2af7698e AW |
2317 | static void virtnet_set_rx_mode(struct net_device *dev) |
2318 | { | |
2319 | struct virtnet_info *vi = netdev_priv(dev); | |
f565a7c2 | 2320 | struct scatterlist sg[2]; |
f565a7c2 | 2321 | struct virtio_net_ctrl_mac *mac_data; |
ccffad25 | 2322 | struct netdev_hw_addr *ha; |
32e7bfc4 | 2323 | int uc_count; |
4cd24eaf | 2324 | int mc_count; |
f565a7c2 AW |
2325 | void *buf; |
2326 | int i; | |
2af7698e | 2327 | |
788a8b6d | 2328 | /* We can't dynamically set ndo_set_rx_mode, so return gracefully */ |
2af7698e AW |
2329 | if (!virtio_has_feature(vi->vdev, VIRTIO_NET_F_CTRL_RX)) |
2330 | return; | |
2331 | ||
12e57169 MT |
2332 | vi->ctrl->promisc = ((dev->flags & IFF_PROMISC) != 0); |
2333 | vi->ctrl->allmulti = ((dev->flags & IFF_ALLMULTI) != 0); | |
2af7698e | 2334 | |
12e57169 | 2335 | sg_init_one(sg, &vi->ctrl->promisc, sizeof(vi->ctrl->promisc)); |
2af7698e AW |
2336 | |
2337 | if (!virtnet_send_command(vi, VIRTIO_NET_CTRL_RX, | |
d24bae32 | 2338 | VIRTIO_NET_CTRL_RX_PROMISC, sg)) |
2af7698e | 2339 | dev_warn(&dev->dev, "Failed to %sable promisc mode.\n", |
12e57169 | 2340 | vi->ctrl->promisc ? "en" : "dis"); |
2af7698e | 2341 | |
12e57169 | 2342 | sg_init_one(sg, &vi->ctrl->allmulti, sizeof(vi->ctrl->allmulti)); |
2af7698e AW |
2343 | |
2344 | if (!virtnet_send_command(vi, VIRTIO_NET_CTRL_RX, | |
d24bae32 | 2345 | VIRTIO_NET_CTRL_RX_ALLMULTI, sg)) |
2af7698e | 2346 | dev_warn(&dev->dev, "Failed to %sable allmulti mode.\n", |
12e57169 | 2347 | vi->ctrl->allmulti ? "en" : "dis"); |
f565a7c2 | 2348 | |
32e7bfc4 | 2349 | uc_count = netdev_uc_count(dev); |
4cd24eaf | 2350 | mc_count = netdev_mc_count(dev); |
f565a7c2 | 2351 | /* MAC filter - use one buffer for both lists */ |
4cd24eaf JP |
2352 | buf = kzalloc(((uc_count + mc_count) * ETH_ALEN) + |
2353 | (2 * sizeof(mac_data->entries)), GFP_ATOMIC); | |
2354 | mac_data = buf; | |
e68ed8f0 | 2355 | if (!buf) |
f565a7c2 | 2356 | return; |
f565a7c2 | 2357 | |
23e258e1 AW |
2358 | sg_init_table(sg, 2); |
2359 | ||
f565a7c2 | 2360 | /* Store the unicast list and count in the front of the buffer */ |
fdd819b2 | 2361 | mac_data->entries = cpu_to_virtio32(vi->vdev, uc_count); |
ccffad25 | 2362 | i = 0; |
32e7bfc4 | 2363 | netdev_for_each_uc_addr(ha, dev) |
ccffad25 | 2364 | memcpy(&mac_data->macs[i++][0], ha->addr, ETH_ALEN); |
f565a7c2 AW |
2365 | |
2366 | sg_set_buf(&sg[0], mac_data, | |
32e7bfc4 | 2367 | sizeof(mac_data->entries) + (uc_count * ETH_ALEN)); |
f565a7c2 AW |
2368 | |
2369 | /* multicast list and count fill the end */ | |
32e7bfc4 | 2370 | mac_data = (void *)&mac_data->macs[uc_count][0]; |
f565a7c2 | 2371 | |
fdd819b2 | 2372 | mac_data->entries = cpu_to_virtio32(vi->vdev, mc_count); |
567ec874 | 2373 | i = 0; |
22bedad3 JP |
2374 | netdev_for_each_mc_addr(ha, dev) |
2375 | memcpy(&mac_data->macs[i++][0], ha->addr, ETH_ALEN); | |
f565a7c2 AW |
2376 | |
2377 | sg_set_buf(&sg[1], mac_data, | |
4cd24eaf | 2378 | sizeof(mac_data->entries) + (mc_count * ETH_ALEN)); |
f565a7c2 AW |
2379 | |
2380 | if (!virtnet_send_command(vi, VIRTIO_NET_CTRL_MAC, | |
d24bae32 | 2381 | VIRTIO_NET_CTRL_MAC_TABLE_SET, sg)) |
99e872ae | 2382 | dev_warn(&dev->dev, "Failed to set MAC filter table.\n"); |
f565a7c2 AW |
2383 | |
2384 | kfree(buf); | |
2af7698e AW |
2385 | } |
2386 | ||
80d5c368 PM |
2387 | static int virtnet_vlan_rx_add_vid(struct net_device *dev, |
2388 | __be16 proto, u16 vid) | |
0bde9569 AW |
2389 | { |
2390 | struct virtnet_info *vi = netdev_priv(dev); | |
2391 | struct scatterlist sg; | |
2392 | ||
d7fad4c8 | 2393 | vi->ctrl->vid = cpu_to_virtio16(vi->vdev, vid); |
12e57169 | 2394 | sg_init_one(&sg, &vi->ctrl->vid, sizeof(vi->ctrl->vid)); |
0bde9569 AW |
2395 | |
2396 | if (!virtnet_send_command(vi, VIRTIO_NET_CTRL_VLAN, | |
d24bae32 | 2397 | VIRTIO_NET_CTRL_VLAN_ADD, &sg)) |
0bde9569 | 2398 | dev_warn(&dev->dev, "Failed to add VLAN ID %d.\n", vid); |
8e586137 | 2399 | return 0; |
0bde9569 AW |
2400 | } |
2401 | ||
80d5c368 PM |
2402 | static int virtnet_vlan_rx_kill_vid(struct net_device *dev, |
2403 | __be16 proto, u16 vid) | |
0bde9569 AW |
2404 | { |
2405 | struct virtnet_info *vi = netdev_priv(dev); | |
2406 | struct scatterlist sg; | |
2407 | ||
d7fad4c8 | 2408 | vi->ctrl->vid = cpu_to_virtio16(vi->vdev, vid); |
12e57169 | 2409 | sg_init_one(&sg, &vi->ctrl->vid, sizeof(vi->ctrl->vid)); |
0bde9569 AW |
2410 | |
2411 | if (!virtnet_send_command(vi, VIRTIO_NET_CTRL_VLAN, | |
d24bae32 | 2412 | VIRTIO_NET_CTRL_VLAN_DEL, &sg)) |
0bde9569 | 2413 | dev_warn(&dev->dev, "Failed to kill VLAN ID %d.\n", vid); |
8e586137 | 2414 | return 0; |
0bde9569 AW |
2415 | } |
2416 | ||
310974fa | 2417 | static void virtnet_clean_affinity(struct virtnet_info *vi) |
986a4f4d JW |
2418 | { |
2419 | int i; | |
2420 | ||
8898c21c WG |
2421 | if (vi->affinity_hint_set) { |
2422 | for (i = 0; i < vi->max_queue_pairs; i++) { | |
19e226e8 CR |
2423 | virtqueue_set_affinity(vi->rq[i].vq, NULL); |
2424 | virtqueue_set_affinity(vi->sq[i].vq, NULL); | |
47be2479 WG |
2425 | } |
2426 | ||
8898c21c WG |
2427 | vi->affinity_hint_set = false; |
2428 | } | |
8898c21c | 2429 | } |
47be2479 | 2430 | |
8898c21c WG |
2431 | static void virtnet_set_affinity(struct virtnet_info *vi) |
2432 | { | |
2ca653d6 CR |
2433 | cpumask_var_t mask; |
2434 | int stragglers; | |
2435 | int group_size; | |
2436 | int i, j, cpu; | |
2437 | int num_cpu; | |
2438 | int stride; | |
2439 | ||
2440 | if (!zalloc_cpumask_var(&mask, GFP_KERNEL)) { | |
310974fa | 2441 | virtnet_clean_affinity(vi); |
8898c21c | 2442 | return; |
986a4f4d JW |
2443 | } |
2444 | ||
2ca653d6 CR |
2445 | num_cpu = num_online_cpus(); |
2446 | stride = max_t(int, num_cpu / vi->curr_queue_pairs, 1); | |
2447 | stragglers = num_cpu >= vi->curr_queue_pairs ? | |
2448 | num_cpu % vi->curr_queue_pairs : | |
2449 | 0; | |
9b51d9d8 | 2450 | cpu = cpumask_first(cpu_online_mask); |
4d99f660 | 2451 | |
2ca653d6 CR |
2452 | for (i = 0; i < vi->curr_queue_pairs; i++) { |
2453 | group_size = stride + (i < stragglers ? 1 : 0); | |
2454 | ||
2455 | for (j = 0; j < group_size; j++) { | |
2456 | cpumask_set_cpu(cpu, mask); | |
2457 | cpu = cpumask_next_wrap(cpu, cpu_online_mask, | |
2458 | nr_cpu_ids, false); | |
2459 | } | |
2460 | virtqueue_set_affinity(vi->rq[i].vq, mask); | |
2461 | virtqueue_set_affinity(vi->sq[i].vq, mask); | |
044ab86d | 2462 | __netif_set_xps_queue(vi->dev, cpumask_bits(mask), i, XPS_CPUS); |
2ca653d6 | 2463 | cpumask_clear(mask); |
986a4f4d JW |
2464 | } |
2465 | ||
8898c21c | 2466 | vi->affinity_hint_set = true; |
2ca653d6 | 2467 | free_cpumask_var(mask); |
986a4f4d JW |
2468 | } |
2469 | ||
8017c279 | 2470 | static int virtnet_cpu_online(unsigned int cpu, struct hlist_node *node) |
8de4b2f3 | 2471 | { |
8017c279 SAS |
2472 | struct virtnet_info *vi = hlist_entry_safe(node, struct virtnet_info, |
2473 | node); | |
2474 | virtnet_set_affinity(vi); | |
2475 | return 0; | |
2476 | } | |
8de4b2f3 | 2477 | |
8017c279 SAS |
2478 | static int virtnet_cpu_dead(unsigned int cpu, struct hlist_node *node) |
2479 | { | |
2480 | struct virtnet_info *vi = hlist_entry_safe(node, struct virtnet_info, | |
2481 | node_dead); | |
2482 | virtnet_set_affinity(vi); | |
2483 | return 0; | |
2484 | } | |
3ab098df | 2485 | |
8017c279 SAS |
2486 | static int virtnet_cpu_down_prep(unsigned int cpu, struct hlist_node *node) |
2487 | { | |
2488 | struct virtnet_info *vi = hlist_entry_safe(node, struct virtnet_info, | |
2489 | node); | |
2490 | ||
310974fa | 2491 | virtnet_clean_affinity(vi); |
8017c279 SAS |
2492 | return 0; |
2493 | } | |
2494 | ||
2495 | static enum cpuhp_state virtionet_online; | |
2496 | ||
2497 | static int virtnet_cpu_notif_add(struct virtnet_info *vi) | |
2498 | { | |
2499 | int ret; | |
2500 | ||
2501 | ret = cpuhp_state_add_instance_nocalls(virtionet_online, &vi->node); | |
2502 | if (ret) | |
2503 | return ret; | |
2504 | ret = cpuhp_state_add_instance_nocalls(CPUHP_VIRT_NET_DEAD, | |
2505 | &vi->node_dead); | |
2506 | if (!ret) | |
2507 | return ret; | |
2508 | cpuhp_state_remove_instance_nocalls(virtionet_online, &vi->node); | |
2509 | return ret; | |
2510 | } | |
2511 | ||
2512 | static void virtnet_cpu_notif_remove(struct virtnet_info *vi) | |
2513 | { | |
2514 | cpuhp_state_remove_instance_nocalls(virtionet_online, &vi->node); | |
2515 | cpuhp_state_remove_instance_nocalls(CPUHP_VIRT_NET_DEAD, | |
2516 | &vi->node_dead); | |
986a4f4d JW |
2517 | } |
2518 | ||
8f9f4668 | 2519 | static void virtnet_get_ringparam(struct net_device *dev, |
74624944 HC |
2520 | struct ethtool_ringparam *ring, |
2521 | struct kernel_ethtool_ringparam *kernel_ring, | |
2522 | struct netlink_ext_ack *extack) | |
8f9f4668 RJ |
2523 | { |
2524 | struct virtnet_info *vi = netdev_priv(dev); | |
2525 | ||
8597b5dd XZ |
2526 | ring->rx_max_pending = vi->rq[0].vq->num_max; |
2527 | ring->tx_max_pending = vi->sq[0].vq->num_max; | |
2528 | ring->rx_pending = virtqueue_get_vring_size(vi->rq[0].vq); | |
2529 | ring->tx_pending = virtqueue_get_vring_size(vi->sq[0].vq); | |
8f9f4668 RJ |
2530 | } |
2531 | ||
a335b33f XZ |
2532 | static int virtnet_set_ringparam(struct net_device *dev, |
2533 | struct ethtool_ringparam *ring, | |
2534 | struct kernel_ethtool_ringparam *kernel_ring, | |
2535 | struct netlink_ext_ack *extack) | |
2536 | { | |
2537 | struct virtnet_info *vi = netdev_priv(dev); | |
2538 | u32 rx_pending, tx_pending; | |
2539 | struct receive_queue *rq; | |
2540 | struct send_queue *sq; | |
2541 | int i, err; | |
2542 | ||
2543 | if (ring->rx_mini_pending || ring->rx_jumbo_pending) | |
2544 | return -EINVAL; | |
2545 | ||
2546 | rx_pending = virtqueue_get_vring_size(vi->rq[0].vq); | |
2547 | tx_pending = virtqueue_get_vring_size(vi->sq[0].vq); | |
2548 | ||
2549 | if (ring->rx_pending == rx_pending && | |
2550 | ring->tx_pending == tx_pending) | |
2551 | return 0; | |
2552 | ||
2553 | if (ring->rx_pending > vi->rq[0].vq->num_max) | |
2554 | return -EINVAL; | |
2555 | ||
2556 | if (ring->tx_pending > vi->sq[0].vq->num_max) | |
2557 | return -EINVAL; | |
2558 | ||
2559 | for (i = 0; i < vi->max_queue_pairs; i++) { | |
2560 | rq = vi->rq + i; | |
2561 | sq = vi->sq + i; | |
2562 | ||
2563 | if (ring->tx_pending != tx_pending) { | |
2564 | err = virtnet_tx_resize(vi, sq, ring->tx_pending); | |
2565 | if (err) | |
2566 | return err; | |
2567 | } | |
2568 | ||
2569 | if (ring->rx_pending != rx_pending) { | |
2570 | err = virtnet_rx_resize(vi, rq, ring->rx_pending); | |
2571 | if (err) | |
2572 | return err; | |
2573 | } | |
2574 | } | |
2575 | ||
2576 | return 0; | |
8f9f4668 RJ |
2577 | } |
2578 | ||
c7114b12 AM |
2579 | static bool virtnet_commit_rss_command(struct virtnet_info *vi) |
2580 | { | |
2581 | struct net_device *dev = vi->dev; | |
2582 | struct scatterlist sgs[4]; | |
2583 | unsigned int sg_buf_size; | |
2584 | ||
2585 | /* prepare sgs */ | |
2586 | sg_init_table(sgs, 4); | |
2587 | ||
2588 | sg_buf_size = offsetof(struct virtio_net_ctrl_rss, indirection_table); | |
2589 | sg_set_buf(&sgs[0], &vi->ctrl->rss, sg_buf_size); | |
2590 | ||
2591 | sg_buf_size = sizeof(uint16_t) * (vi->ctrl->rss.indirection_table_mask + 1); | |
2592 | sg_set_buf(&sgs[1], vi->ctrl->rss.indirection_table, sg_buf_size); | |
2593 | ||
2594 | sg_buf_size = offsetof(struct virtio_net_ctrl_rss, key) | |
2595 | - offsetof(struct virtio_net_ctrl_rss, max_tx_vq); | |
2596 | sg_set_buf(&sgs[2], &vi->ctrl->rss.max_tx_vq, sg_buf_size); | |
2597 | ||
2598 | sg_buf_size = vi->rss_key_size; | |
2599 | sg_set_buf(&sgs[3], vi->ctrl->rss.key, sg_buf_size); | |
2600 | ||
2601 | if (!virtnet_send_command(vi, VIRTIO_NET_CTRL_MQ, | |
91f41f01 AM |
2602 | vi->has_rss ? VIRTIO_NET_CTRL_MQ_RSS_CONFIG |
2603 | : VIRTIO_NET_CTRL_MQ_HASH_CONFIG, sgs)) { | |
c7114b12 AM |
2604 | dev_warn(&dev->dev, "VIRTIONET issue with committing RSS sgs\n"); |
2605 | return false; | |
2606 | } | |
2607 | return true; | |
2608 | } | |
2609 | ||
2610 | static void virtnet_init_default_rss(struct virtnet_info *vi) | |
2611 | { | |
2612 | u32 indir_val = 0; | |
2613 | int i = 0; | |
2614 | ||
2615 | vi->ctrl->rss.hash_types = vi->rss_hash_types_supported; | |
c1170820 | 2616 | vi->rss_hash_types_saved = vi->rss_hash_types_supported; |
c7114b12 AM |
2617 | vi->ctrl->rss.indirection_table_mask = vi->rss_indir_table_size |
2618 | ? vi->rss_indir_table_size - 1 : 0; | |
2619 | vi->ctrl->rss.unclassified_queue = 0; | |
2620 | ||
2621 | for (; i < vi->rss_indir_table_size; ++i) { | |
2622 | indir_val = ethtool_rxfh_indir_default(i, vi->curr_queue_pairs); | |
2623 | vi->ctrl->rss.indirection_table[i] = indir_val; | |
2624 | } | |
2625 | ||
2626 | vi->ctrl->rss.max_tx_vq = vi->curr_queue_pairs; | |
2627 | vi->ctrl->rss.hash_key_length = vi->rss_key_size; | |
2628 | ||
2629 | netdev_rss_key_fill(vi->ctrl->rss.key, vi->rss_key_size); | |
2630 | } | |
2631 | ||
c1170820 AM |
2632 | static void virtnet_get_hashflow(const struct virtnet_info *vi, struct ethtool_rxnfc *info) |
2633 | { | |
2634 | info->data = 0; | |
2635 | switch (info->flow_type) { | |
2636 | case TCP_V4_FLOW: | |
2637 | if (vi->rss_hash_types_saved & VIRTIO_NET_RSS_HASH_TYPE_TCPv4) { | |
2638 | info->data = RXH_IP_SRC | RXH_IP_DST | | |
2639 | RXH_L4_B_0_1 | RXH_L4_B_2_3; | |
2640 | } else if (vi->rss_hash_types_saved & VIRTIO_NET_RSS_HASH_TYPE_IPv4) { | |
2641 | info->data = RXH_IP_SRC | RXH_IP_DST; | |
2642 | } | |
2643 | break; | |
2644 | case TCP_V6_FLOW: | |
2645 | if (vi->rss_hash_types_saved & VIRTIO_NET_RSS_HASH_TYPE_TCPv6) { | |
2646 | info->data = RXH_IP_SRC | RXH_IP_DST | | |
2647 | RXH_L4_B_0_1 | RXH_L4_B_2_3; | |
2648 | } else if (vi->rss_hash_types_saved & VIRTIO_NET_RSS_HASH_TYPE_IPv6) { | |
2649 | info->data = RXH_IP_SRC | RXH_IP_DST; | |
2650 | } | |
2651 | break; | |
2652 | case UDP_V4_FLOW: | |
2653 | if (vi->rss_hash_types_saved & VIRTIO_NET_RSS_HASH_TYPE_UDPv4) { | |
2654 | info->data = RXH_IP_SRC | RXH_IP_DST | | |
2655 | RXH_L4_B_0_1 | RXH_L4_B_2_3; | |
2656 | } else if (vi->rss_hash_types_saved & VIRTIO_NET_RSS_HASH_TYPE_IPv4) { | |
2657 | info->data = RXH_IP_SRC | RXH_IP_DST; | |
2658 | } | |
2659 | break; | |
2660 | case UDP_V6_FLOW: | |
2661 | if (vi->rss_hash_types_saved & VIRTIO_NET_RSS_HASH_TYPE_UDPv6) { | |
2662 | info->data = RXH_IP_SRC | RXH_IP_DST | | |
2663 | RXH_L4_B_0_1 | RXH_L4_B_2_3; | |
2664 | } else if (vi->rss_hash_types_saved & VIRTIO_NET_RSS_HASH_TYPE_IPv6) { | |
2665 | info->data = RXH_IP_SRC | RXH_IP_DST; | |
2666 | } | |
2667 | break; | |
2668 | case IPV4_FLOW: | |
2669 | if (vi->rss_hash_types_saved & VIRTIO_NET_RSS_HASH_TYPE_IPv4) | |
2670 | info->data = RXH_IP_SRC | RXH_IP_DST; | |
2671 | ||
2672 | break; | |
2673 | case IPV6_FLOW: | |
2674 | if (vi->rss_hash_types_saved & VIRTIO_NET_RSS_HASH_TYPE_IPv6) | |
2675 | info->data = RXH_IP_SRC | RXH_IP_DST; | |
2676 | ||
2677 | break; | |
2678 | default: | |
2679 | info->data = 0; | |
2680 | break; | |
2681 | } | |
2682 | } | |
2683 | ||
2684 | static bool virtnet_set_hashflow(struct virtnet_info *vi, struct ethtool_rxnfc *info) | |
2685 | { | |
2686 | u32 new_hashtypes = vi->rss_hash_types_saved; | |
2687 | bool is_disable = info->data & RXH_DISCARD; | |
2688 | bool is_l4 = info->data == (RXH_IP_SRC | RXH_IP_DST | RXH_L4_B_0_1 | RXH_L4_B_2_3); | |
2689 | ||
2690 | /* supports only 'sd', 'sdfn' and 'r' */ | |
2691 | if (!((info->data == (RXH_IP_SRC | RXH_IP_DST)) | is_l4 | is_disable)) | |
2692 | return false; | |
2693 | ||
2694 | switch (info->flow_type) { | |
2695 | case TCP_V4_FLOW: | |
2696 | new_hashtypes &= ~(VIRTIO_NET_RSS_HASH_TYPE_IPv4 | VIRTIO_NET_RSS_HASH_TYPE_TCPv4); | |
2697 | if (!is_disable) | |
2698 | new_hashtypes |= VIRTIO_NET_RSS_HASH_TYPE_IPv4 | |
2699 | | (is_l4 ? VIRTIO_NET_RSS_HASH_TYPE_TCPv4 : 0); | |
2700 | break; | |
2701 | case UDP_V4_FLOW: | |
2702 | new_hashtypes &= ~(VIRTIO_NET_RSS_HASH_TYPE_IPv4 | VIRTIO_NET_RSS_HASH_TYPE_UDPv4); | |
2703 | if (!is_disable) | |
2704 | new_hashtypes |= VIRTIO_NET_RSS_HASH_TYPE_IPv4 | |
2705 | | (is_l4 ? VIRTIO_NET_RSS_HASH_TYPE_UDPv4 : 0); | |
2706 | break; | |
2707 | case IPV4_FLOW: | |
2708 | new_hashtypes &= ~VIRTIO_NET_RSS_HASH_TYPE_IPv4; | |
2709 | if (!is_disable) | |
2710 | new_hashtypes = VIRTIO_NET_RSS_HASH_TYPE_IPv4; | |
2711 | break; | |
2712 | case TCP_V6_FLOW: | |
2713 | new_hashtypes &= ~(VIRTIO_NET_RSS_HASH_TYPE_IPv6 | VIRTIO_NET_RSS_HASH_TYPE_TCPv6); | |
2714 | if (!is_disable) | |
2715 | new_hashtypes |= VIRTIO_NET_RSS_HASH_TYPE_IPv6 | |
2716 | | (is_l4 ? VIRTIO_NET_RSS_HASH_TYPE_TCPv6 : 0); | |
2717 | break; | |
2718 | case UDP_V6_FLOW: | |
2719 | new_hashtypes &= ~(VIRTIO_NET_RSS_HASH_TYPE_IPv6 | VIRTIO_NET_RSS_HASH_TYPE_UDPv6); | |
2720 | if (!is_disable) | |
2721 | new_hashtypes |= VIRTIO_NET_RSS_HASH_TYPE_IPv6 | |
2722 | | (is_l4 ? VIRTIO_NET_RSS_HASH_TYPE_UDPv6 : 0); | |
2723 | break; | |
2724 | case IPV6_FLOW: | |
2725 | new_hashtypes &= ~VIRTIO_NET_RSS_HASH_TYPE_IPv6; | |
2726 | if (!is_disable) | |
2727 | new_hashtypes = VIRTIO_NET_RSS_HASH_TYPE_IPv6; | |
2728 | break; | |
2729 | default: | |
2730 | /* unsupported flow */ | |
2731 | return false; | |
2732 | } | |
2733 | ||
2734 | /* if unsupported hashtype was set */ | |
2735 | if (new_hashtypes != (new_hashtypes & vi->rss_hash_types_supported)) | |
2736 | return false; | |
2737 | ||
2738 | if (new_hashtypes != vi->rss_hash_types_saved) { | |
2739 | vi->rss_hash_types_saved = new_hashtypes; | |
2740 | vi->ctrl->rss.hash_types = vi->rss_hash_types_saved; | |
2741 | if (vi->dev->features & NETIF_F_RXHASH) | |
2742 | return virtnet_commit_rss_command(vi); | |
2743 | } | |
2744 | ||
2745 | return true; | |
2746 | } | |
66846048 RJ |
2747 | |
2748 | static void virtnet_get_drvinfo(struct net_device *dev, | |
2749 | struct ethtool_drvinfo *info) | |
2750 | { | |
2751 | struct virtnet_info *vi = netdev_priv(dev); | |
2752 | struct virtio_device *vdev = vi->vdev; | |
2753 | ||
fb3ceec1 WS |
2754 | strscpy(info->driver, KBUILD_MODNAME, sizeof(info->driver)); |
2755 | strscpy(info->version, VIRTNET_DRIVER_VERSION, sizeof(info->version)); | |
2756 | strscpy(info->bus_info, virtio_bus_name(vdev), sizeof(info->bus_info)); | |
66846048 RJ |
2757 | |
2758 | } | |
2759 | ||
d73bcd2c JW |
2760 | /* TODO: Eliminate OOO packets during switching */ |
2761 | static int virtnet_set_channels(struct net_device *dev, | |
2762 | struct ethtool_channels *channels) | |
2763 | { | |
2764 | struct virtnet_info *vi = netdev_priv(dev); | |
2765 | u16 queue_pairs = channels->combined_count; | |
2766 | int err; | |
2767 | ||
2768 | /* We don't support separate rx/tx channels. | |
2769 | * We don't allow setting 'other' channels. | |
2770 | */ | |
2771 | if (channels->rx_count || channels->tx_count || channels->other_count) | |
2772 | return -EINVAL; | |
2773 | ||
c18e9cd6 | 2774 | if (queue_pairs > vi->max_queue_pairs || queue_pairs == 0) |
d73bcd2c JW |
2775 | return -EINVAL; |
2776 | ||
f600b690 JF |
2777 | /* For now we don't support modifying channels while XDP is loaded |
2778 | * also when XDP is loaded all RX queues have XDP programs so we only | |
2779 | * need to check a single RX queue. | |
2780 | */ | |
2781 | if (vi->rq[0].xdp_prog) | |
2782 | return -EINVAL; | |
2783 | ||
a0d1d0f4 | 2784 | cpus_read_lock(); |
47315329 | 2785 | err = _virtnet_set_queues(vi, queue_pairs); |
de33212f | 2786 | if (err) { |
a0d1d0f4 | 2787 | cpus_read_unlock(); |
de33212f | 2788 | goto err; |
d73bcd2c | 2789 | } |
de33212f | 2790 | virtnet_set_affinity(vi); |
a0d1d0f4 | 2791 | cpus_read_unlock(); |
d73bcd2c | 2792 | |
de33212f JD |
2793 | netif_set_real_num_tx_queues(dev, queue_pairs); |
2794 | netif_set_real_num_rx_queues(dev, queue_pairs); | |
2795 | err: | |
d73bcd2c JW |
2796 | return err; |
2797 | } | |
2798 | ||
d7dfc5cf TM |
2799 | static void virtnet_get_strings(struct net_device *dev, u32 stringset, u8 *data) |
2800 | { | |
2801 | struct virtnet_info *vi = netdev_priv(dev); | |
d7dfc5cf | 2802 | unsigned int i, j; |
d7a9a01b | 2803 | u8 *p = data; |
d7dfc5cf TM |
2804 | |
2805 | switch (stringset) { | |
2806 | case ETH_SS_STATS: | |
2807 | for (i = 0; i < vi->curr_queue_pairs; i++) { | |
d7a9a01b AD |
2808 | for (j = 0; j < VIRTNET_RQ_STATS_LEN; j++) |
2809 | ethtool_sprintf(&p, "rx_queue_%u_%s", i, | |
2810 | virtnet_rq_stats_desc[j].desc); | |
d7dfc5cf TM |
2811 | } |
2812 | ||
2813 | for (i = 0; i < vi->curr_queue_pairs; i++) { | |
d7a9a01b AD |
2814 | for (j = 0; j < VIRTNET_SQ_STATS_LEN; j++) |
2815 | ethtool_sprintf(&p, "tx_queue_%u_%s", i, | |
2816 | virtnet_sq_stats_desc[j].desc); | |
d7dfc5cf TM |
2817 | } |
2818 | break; | |
2819 | } | |
2820 | } | |
2821 | ||
2822 | static int virtnet_get_sset_count(struct net_device *dev, int sset) | |
2823 | { | |
2824 | struct virtnet_info *vi = netdev_priv(dev); | |
2825 | ||
2826 | switch (sset) { | |
2827 | case ETH_SS_STATS: | |
2828 | return vi->curr_queue_pairs * (VIRTNET_RQ_STATS_LEN + | |
2829 | VIRTNET_SQ_STATS_LEN); | |
2830 | default: | |
2831 | return -EOPNOTSUPP; | |
2832 | } | |
2833 | } | |
2834 | ||
2835 | static void virtnet_get_ethtool_stats(struct net_device *dev, | |
2836 | struct ethtool_stats *stats, u64 *data) | |
2837 | { | |
2838 | struct virtnet_info *vi = netdev_priv(dev); | |
2839 | unsigned int idx = 0, start, i, j; | |
2840 | const u8 *stats_base; | |
2841 | size_t offset; | |
2842 | ||
2843 | for (i = 0; i < vi->curr_queue_pairs; i++) { | |
2844 | struct receive_queue *rq = &vi->rq[i]; | |
2845 | ||
d46eeeaf | 2846 | stats_base = (u8 *)&rq->stats; |
d7dfc5cf | 2847 | do { |
068c38ad | 2848 | start = u64_stats_fetch_begin(&rq->stats.syncp); |
d7dfc5cf TM |
2849 | for (j = 0; j < VIRTNET_RQ_STATS_LEN; j++) { |
2850 | offset = virtnet_rq_stats_desc[j].offset; | |
2851 | data[idx + j] = *(u64 *)(stats_base + offset); | |
2852 | } | |
068c38ad | 2853 | } while (u64_stats_fetch_retry(&rq->stats.syncp, start)); |
d7dfc5cf TM |
2854 | idx += VIRTNET_RQ_STATS_LEN; |
2855 | } | |
2856 | ||
2857 | for (i = 0; i < vi->curr_queue_pairs; i++) { | |
2858 | struct send_queue *sq = &vi->sq[i]; | |
2859 | ||
2860 | stats_base = (u8 *)&sq->stats; | |
2861 | do { | |
068c38ad | 2862 | start = u64_stats_fetch_begin(&sq->stats.syncp); |
d7dfc5cf TM |
2863 | for (j = 0; j < VIRTNET_SQ_STATS_LEN; j++) { |
2864 | offset = virtnet_sq_stats_desc[j].offset; | |
2865 | data[idx + j] = *(u64 *)(stats_base + offset); | |
2866 | } | |
068c38ad | 2867 | } while (u64_stats_fetch_retry(&sq->stats.syncp, start)); |
d7dfc5cf TM |
2868 | idx += VIRTNET_SQ_STATS_LEN; |
2869 | } | |
2870 | } | |
2871 | ||
d73bcd2c JW |
2872 | static void virtnet_get_channels(struct net_device *dev, |
2873 | struct ethtool_channels *channels) | |
2874 | { | |
2875 | struct virtnet_info *vi = netdev_priv(dev); | |
2876 | ||
2877 | channels->combined_count = vi->curr_queue_pairs; | |
2878 | channels->max_combined = vi->max_queue_pairs; | |
2879 | channels->max_other = 0; | |
2880 | channels->rx_count = 0; | |
2881 | channels->tx_count = 0; | |
2882 | channels->other_count = 0; | |
2883 | } | |
2884 | ||
ebb6b4b1 PR |
2885 | static int virtnet_set_link_ksettings(struct net_device *dev, |
2886 | const struct ethtool_link_ksettings *cmd) | |
16032be5 NA |
2887 | { |
2888 | struct virtnet_info *vi = netdev_priv(dev); | |
16032be5 | 2889 | |
9aedc6e2 CF |
2890 | return ethtool_virtdev_set_link_ksettings(dev, cmd, |
2891 | &vi->speed, &vi->duplex); | |
16032be5 NA |
2892 | } |
2893 | ||
ebb6b4b1 PR |
2894 | static int virtnet_get_link_ksettings(struct net_device *dev, |
2895 | struct ethtool_link_ksettings *cmd) | |
16032be5 NA |
2896 | { |
2897 | struct virtnet_info *vi = netdev_priv(dev); | |
2898 | ||
ebb6b4b1 PR |
2899 | cmd->base.speed = vi->speed; |
2900 | cmd->base.duplex = vi->duplex; | |
2901 | cmd->base.port = PORT_OTHER; | |
16032be5 NA |
2902 | |
2903 | return 0; | |
2904 | } | |
2905 | ||
699b045a AK |
2906 | static int virtnet_send_notf_coal_cmds(struct virtnet_info *vi, |
2907 | struct ethtool_coalesce *ec) | |
2908 | { | |
2909 | struct scatterlist sgs_tx, sgs_rx; | |
2910 | struct virtio_net_ctrl_coal_tx coal_tx; | |
2911 | struct virtio_net_ctrl_coal_rx coal_rx; | |
2912 | ||
2913 | coal_tx.tx_usecs = cpu_to_le32(ec->tx_coalesce_usecs); | |
2914 | coal_tx.tx_max_packets = cpu_to_le32(ec->tx_max_coalesced_frames); | |
2915 | sg_init_one(&sgs_tx, &coal_tx, sizeof(coal_tx)); | |
2916 | ||
2917 | if (!virtnet_send_command(vi, VIRTIO_NET_CTRL_NOTF_COAL, | |
2918 | VIRTIO_NET_CTRL_NOTF_COAL_TX_SET, | |
2919 | &sgs_tx)) | |
2920 | return -EINVAL; | |
2921 | ||
2922 | /* Save parameters */ | |
2923 | vi->tx_usecs = ec->tx_coalesce_usecs; | |
2924 | vi->tx_max_packets = ec->tx_max_coalesced_frames; | |
2925 | ||
2926 | coal_rx.rx_usecs = cpu_to_le32(ec->rx_coalesce_usecs); | |
2927 | coal_rx.rx_max_packets = cpu_to_le32(ec->rx_max_coalesced_frames); | |
2928 | sg_init_one(&sgs_rx, &coal_rx, sizeof(coal_rx)); | |
2929 | ||
2930 | if (!virtnet_send_command(vi, VIRTIO_NET_CTRL_NOTF_COAL, | |
2931 | VIRTIO_NET_CTRL_NOTF_COAL_RX_SET, | |
2932 | &sgs_rx)) | |
2933 | return -EINVAL; | |
2934 | ||
2935 | /* Save parameters */ | |
2936 | vi->rx_usecs = ec->rx_coalesce_usecs; | |
2937 | vi->rx_max_packets = ec->rx_max_coalesced_frames; | |
2938 | ||
2939 | return 0; | |
2940 | } | |
2941 | ||
2942 | static int virtnet_coal_params_supported(struct ethtool_coalesce *ec) | |
2943 | { | |
2944 | /* usecs coalescing is supported only if VIRTIO_NET_F_NOTF_COAL | |
2945 | * feature is negotiated. | |
2946 | */ | |
2947 | if (ec->rx_coalesce_usecs || ec->tx_coalesce_usecs) | |
2948 | return -EOPNOTSUPP; | |
2949 | ||
2950 | if (ec->tx_max_coalesced_frames > 1 || | |
2951 | ec->rx_max_coalesced_frames != 1) | |
2952 | return -EINVAL; | |
2953 | ||
2954 | return 0; | |
2955 | } | |
2956 | ||
0c465be1 | 2957 | static int virtnet_set_coalesce(struct net_device *dev, |
f3ccfda1 YM |
2958 | struct ethtool_coalesce *ec, |
2959 | struct kernel_ethtool_coalesce *kernel_coal, | |
2960 | struct netlink_ext_ack *extack) | |
0c465be1 | 2961 | { |
0c465be1 | 2962 | struct virtnet_info *vi = netdev_priv(dev); |
699b045a AK |
2963 | int ret, i, napi_weight; |
2964 | bool update_napi = false; | |
0c465be1 | 2965 | |
699b045a | 2966 | /* Can't change NAPI weight if the link is up */ |
0c465be1 | 2967 | napi_weight = ec->tx_max_coalesced_frames ? NAPI_POLL_WEIGHT : 0; |
0c465be1 JW |
2968 | if (napi_weight ^ vi->sq[0].napi.weight) { |
2969 | if (dev->flags & IFF_UP) | |
2970 | return -EBUSY; | |
699b045a AK |
2971 | else |
2972 | update_napi = true; | |
2973 | } | |
2974 | ||
2975 | if (virtio_has_feature(vi->vdev, VIRTIO_NET_F_NOTF_COAL)) | |
2976 | ret = virtnet_send_notf_coal_cmds(vi, ec); | |
2977 | else | |
2978 | ret = virtnet_coal_params_supported(ec); | |
2979 | ||
2980 | if (ret) | |
2981 | return ret; | |
2982 | ||
2983 | if (update_napi) { | |
0c465be1 JW |
2984 | for (i = 0; i < vi->max_queue_pairs; i++) |
2985 | vi->sq[i].napi.weight = napi_weight; | |
2986 | } | |
2987 | ||
699b045a | 2988 | return ret; |
0c465be1 JW |
2989 | } |
2990 | ||
2991 | static int virtnet_get_coalesce(struct net_device *dev, | |
f3ccfda1 YM |
2992 | struct ethtool_coalesce *ec, |
2993 | struct kernel_ethtool_coalesce *kernel_coal, | |
2994 | struct netlink_ext_ack *extack) | |
0c465be1 | 2995 | { |
0c465be1 JW |
2996 | struct virtnet_info *vi = netdev_priv(dev); |
2997 | ||
699b045a AK |
2998 | if (virtio_has_feature(vi->vdev, VIRTIO_NET_F_NOTF_COAL)) { |
2999 | ec->rx_coalesce_usecs = vi->rx_usecs; | |
3000 | ec->tx_coalesce_usecs = vi->tx_usecs; | |
3001 | ec->tx_max_coalesced_frames = vi->tx_max_packets; | |
3002 | ec->rx_max_coalesced_frames = vi->rx_max_packets; | |
3003 | } else { | |
3004 | ec->rx_max_coalesced_frames = 1; | |
0c465be1 | 3005 | |
699b045a AK |
3006 | if (vi->sq[0].napi.weight) |
3007 | ec->tx_max_coalesced_frames = 1; | |
3008 | } | |
0c465be1 JW |
3009 | |
3010 | return 0; | |
3011 | } | |
3012 | ||
16032be5 NA |
3013 | static void virtnet_init_settings(struct net_device *dev) |
3014 | { | |
3015 | struct virtnet_info *vi = netdev_priv(dev); | |
3016 | ||
3017 | vi->speed = SPEED_UNKNOWN; | |
3018 | vi->duplex = DUPLEX_UNKNOWN; | |
3019 | } | |
3020 | ||
faa9b39f JB |
3021 | static void virtnet_update_settings(struct virtnet_info *vi) |
3022 | { | |
3023 | u32 speed; | |
3024 | u8 duplex; | |
3025 | ||
3026 | if (!virtio_has_feature(vi->vdev, VIRTIO_NET_F_SPEED_DUPLEX)) | |
3027 | return; | |
3028 | ||
64ffa39d MT |
3029 | virtio_cread_le(vi->vdev, struct virtio_net_config, speed, &speed); |
3030 | ||
faa9b39f JB |
3031 | if (ethtool_validate_speed(speed)) |
3032 | vi->speed = speed; | |
64ffa39d MT |
3033 | |
3034 | virtio_cread_le(vi->vdev, struct virtio_net_config, duplex, &duplex); | |
3035 | ||
faa9b39f JB |
3036 | if (ethtool_validate_duplex(duplex)) |
3037 | vi->duplex = duplex; | |
3038 | } | |
3039 | ||
c7114b12 AM |
3040 | static u32 virtnet_get_rxfh_key_size(struct net_device *dev) |
3041 | { | |
3042 | return ((struct virtnet_info *)netdev_priv(dev))->rss_key_size; | |
3043 | } | |
3044 | ||
3045 | static u32 virtnet_get_rxfh_indir_size(struct net_device *dev) | |
3046 | { | |
3047 | return ((struct virtnet_info *)netdev_priv(dev))->rss_indir_table_size; | |
3048 | } | |
3049 | ||
3050 | static int virtnet_get_rxfh(struct net_device *dev, u32 *indir, u8 *key, u8 *hfunc) | |
3051 | { | |
3052 | struct virtnet_info *vi = netdev_priv(dev); | |
3053 | int i; | |
3054 | ||
3055 | if (indir) { | |
3056 | for (i = 0; i < vi->rss_indir_table_size; ++i) | |
3057 | indir[i] = vi->ctrl->rss.indirection_table[i]; | |
3058 | } | |
3059 | ||
3060 | if (key) | |
3061 | memcpy(key, vi->ctrl->rss.key, vi->rss_key_size); | |
3062 | ||
3063 | if (hfunc) | |
3064 | *hfunc = ETH_RSS_HASH_TOP; | |
3065 | ||
3066 | return 0; | |
3067 | } | |
3068 | ||
3069 | static int virtnet_set_rxfh(struct net_device *dev, const u32 *indir, const u8 *key, const u8 hfunc) | |
3070 | { | |
3071 | struct virtnet_info *vi = netdev_priv(dev); | |
3072 | int i; | |
3073 | ||
3074 | if (hfunc != ETH_RSS_HASH_NO_CHANGE && hfunc != ETH_RSS_HASH_TOP) | |
3075 | return -EOPNOTSUPP; | |
3076 | ||
3077 | if (indir) { | |
3078 | for (i = 0; i < vi->rss_indir_table_size; ++i) | |
3079 | vi->ctrl->rss.indirection_table[i] = indir[i]; | |
3080 | } | |
3081 | if (key) | |
3082 | memcpy(vi->ctrl->rss.key, key, vi->rss_key_size); | |
3083 | ||
3084 | virtnet_commit_rss_command(vi); | |
3085 | ||
3086 | return 0; | |
3087 | } | |
3088 | ||
3089 | static int virtnet_get_rxnfc(struct net_device *dev, struct ethtool_rxnfc *info, u32 *rule_locs) | |
3090 | { | |
3091 | struct virtnet_info *vi = netdev_priv(dev); | |
3092 | int rc = 0; | |
3093 | ||
3094 | switch (info->cmd) { | |
3095 | case ETHTOOL_GRXRINGS: | |
3096 | info->data = vi->curr_queue_pairs; | |
c1170820 AM |
3097 | break; |
3098 | case ETHTOOL_GRXFH: | |
3099 | virtnet_get_hashflow(vi, info); | |
3100 | break; | |
3101 | default: | |
3102 | rc = -EOPNOTSUPP; | |
3103 | } | |
3104 | ||
3105 | return rc; | |
3106 | } | |
3107 | ||
3108 | static int virtnet_set_rxnfc(struct net_device *dev, struct ethtool_rxnfc *info) | |
3109 | { | |
3110 | struct virtnet_info *vi = netdev_priv(dev); | |
3111 | int rc = 0; | |
3112 | ||
3113 | switch (info->cmd) { | |
3114 | case ETHTOOL_SRXFH: | |
3115 | if (!virtnet_set_hashflow(vi, info)) | |
3116 | rc = -EINVAL; | |
3117 | ||
c7114b12 AM |
3118 | break; |
3119 | default: | |
3120 | rc = -EOPNOTSUPP; | |
3121 | } | |
3122 | ||
3123 | return rc; | |
3124 | } | |
3125 | ||
0fc0b732 | 3126 | static const struct ethtool_ops virtnet_ethtool_ops = { |
699b045a AK |
3127 | .supported_coalesce_params = ETHTOOL_COALESCE_MAX_FRAMES | |
3128 | ETHTOOL_COALESCE_USECS, | |
66846048 | 3129 | .get_drvinfo = virtnet_get_drvinfo, |
9f4d26d0 | 3130 | .get_link = ethtool_op_get_link, |
8f9f4668 | 3131 | .get_ringparam = virtnet_get_ringparam, |
a335b33f | 3132 | .set_ringparam = virtnet_set_ringparam, |
d7dfc5cf TM |
3133 | .get_strings = virtnet_get_strings, |
3134 | .get_sset_count = virtnet_get_sset_count, | |
3135 | .get_ethtool_stats = virtnet_get_ethtool_stats, | |
d73bcd2c JW |
3136 | .set_channels = virtnet_set_channels, |
3137 | .get_channels = virtnet_get_channels, | |
074c3582 | 3138 | .get_ts_info = ethtool_op_get_ts_info, |
ebb6b4b1 PR |
3139 | .get_link_ksettings = virtnet_get_link_ksettings, |
3140 | .set_link_ksettings = virtnet_set_link_ksettings, | |
0c465be1 JW |
3141 | .set_coalesce = virtnet_set_coalesce, |
3142 | .get_coalesce = virtnet_get_coalesce, | |
c7114b12 AM |
3143 | .get_rxfh_key_size = virtnet_get_rxfh_key_size, |
3144 | .get_rxfh_indir_size = virtnet_get_rxfh_indir_size, | |
3145 | .get_rxfh = virtnet_get_rxfh, | |
3146 | .set_rxfh = virtnet_set_rxfh, | |
3147 | .get_rxnfc = virtnet_get_rxnfc, | |
c1170820 | 3148 | .set_rxnfc = virtnet_set_rxnfc, |
a9ea3fc6 HX |
3149 | }; |
3150 | ||
9fe7bfce JF |
3151 | static void virtnet_freeze_down(struct virtio_device *vdev) |
3152 | { | |
3153 | struct virtnet_info *vi = vdev->priv; | |
9fe7bfce JF |
3154 | |
3155 | /* Make sure no work handler is accessing the device */ | |
3156 | flush_work(&vi->config_work); | |
3157 | ||
05c998b7 | 3158 | netif_tx_lock_bh(vi->dev); |
9fe7bfce | 3159 | netif_device_detach(vi->dev); |
05c998b7 | 3160 | netif_tx_unlock_bh(vi->dev); |
8af52fe9 SG |
3161 | if (netif_running(vi->dev)) |
3162 | virtnet_close(vi->dev); | |
9fe7bfce JF |
3163 | } |
3164 | ||
3165 | static int init_vqs(struct virtnet_info *vi); | |
3166 | ||
3167 | static int virtnet_restore_up(struct virtio_device *vdev) | |
3168 | { | |
3169 | struct virtnet_info *vi = vdev->priv; | |
8af52fe9 | 3170 | int err; |
9fe7bfce JF |
3171 | |
3172 | err = init_vqs(vi); | |
3173 | if (err) | |
3174 | return err; | |
3175 | ||
3176 | virtio_device_ready(vdev); | |
3177 | ||
5a159128 JW |
3178 | enable_delayed_refill(vi); |
3179 | ||
9fe7bfce | 3180 | if (netif_running(vi->dev)) { |
8af52fe9 SG |
3181 | err = virtnet_open(vi->dev); |
3182 | if (err) | |
3183 | return err; | |
9fe7bfce JF |
3184 | } |
3185 | ||
05c998b7 | 3186 | netif_tx_lock_bh(vi->dev); |
9fe7bfce | 3187 | netif_device_attach(vi->dev); |
05c998b7 | 3188 | netif_tx_unlock_bh(vi->dev); |
9fe7bfce JF |
3189 | return err; |
3190 | } | |
3191 | ||
3f93522f JW |
3192 | static int virtnet_set_guest_offloads(struct virtnet_info *vi, u64 offloads) |
3193 | { | |
3194 | struct scatterlist sg; | |
12e57169 | 3195 | vi->ctrl->offloads = cpu_to_virtio64(vi->vdev, offloads); |
3f93522f | 3196 | |
12e57169 | 3197 | sg_init_one(&sg, &vi->ctrl->offloads, sizeof(vi->ctrl->offloads)); |
3f93522f JW |
3198 | |
3199 | if (!virtnet_send_command(vi, VIRTIO_NET_CTRL_GUEST_OFFLOADS, | |
3200 | VIRTIO_NET_CTRL_GUEST_OFFLOADS_SET, &sg)) { | |
7934b481 | 3201 | dev_warn(&vi->dev->dev, "Fail to set guest offload.\n"); |
3f93522f JW |
3202 | return -EINVAL; |
3203 | } | |
3204 | ||
3205 | return 0; | |
3206 | } | |
3207 | ||
3208 | static int virtnet_clear_guest_offloads(struct virtnet_info *vi) | |
3209 | { | |
3210 | u64 offloads = 0; | |
3211 | ||
3212 | if (!vi->guest_offloads) | |
3213 | return 0; | |
3214 | ||
3f93522f JW |
3215 | return virtnet_set_guest_offloads(vi, offloads); |
3216 | } | |
3217 | ||
3218 | static int virtnet_restore_guest_offloads(struct virtnet_info *vi) | |
3219 | { | |
3220 | u64 offloads = vi->guest_offloads; | |
3221 | ||
3222 | if (!vi->guest_offloads) | |
3223 | return 0; | |
3f93522f JW |
3224 | |
3225 | return virtnet_set_guest_offloads(vi, offloads); | |
3226 | } | |
3227 | ||
9861ce03 JK |
3228 | static int virtnet_xdp_set(struct net_device *dev, struct bpf_prog *prog, |
3229 | struct netlink_ext_ack *extack) | |
f600b690 | 3230 | { |
e814b958 HQ |
3231 | unsigned int room = SKB_DATA_ALIGN(VIRTIO_XDP_HEADROOM + |
3232 | sizeof(struct skb_shared_info)); | |
3233 | unsigned int max_sz = PAGE_SIZE - room - ETH_HLEN; | |
f600b690 JF |
3234 | struct virtnet_info *vi = netdev_priv(dev); |
3235 | struct bpf_prog *old_prog; | |
017b29c3 | 3236 | u16 xdp_qp = 0, curr_qp; |
672aafd5 | 3237 | int i, err; |
f600b690 | 3238 | |
3f93522f JW |
3239 | if (!virtio_has_feature(vi->vdev, VIRTIO_NET_F_CTRL_GUEST_OFFLOADS) |
3240 | && (virtio_has_feature(vi->vdev, VIRTIO_NET_F_GUEST_TSO4) || | |
3241 | virtio_has_feature(vi->vdev, VIRTIO_NET_F_GUEST_TSO6) || | |
3242 | virtio_has_feature(vi->vdev, VIRTIO_NET_F_GUEST_ECN) || | |
18ba58e1 | 3243 | virtio_has_feature(vi->vdev, VIRTIO_NET_F_GUEST_UFO) || |
418044e1 AM |
3244 | virtio_has_feature(vi->vdev, VIRTIO_NET_F_GUEST_CSUM) || |
3245 | virtio_has_feature(vi->vdev, VIRTIO_NET_F_GUEST_USO4) || | |
3246 | virtio_has_feature(vi->vdev, VIRTIO_NET_F_GUEST_USO6))) { | |
dbcf24d1 | 3247 | NL_SET_ERR_MSG_MOD(extack, "Can't set XDP while host is implementing GRO_HW/CSUM, disable GRO_HW/CSUM first"); |
f600b690 JF |
3248 | return -EOPNOTSUPP; |
3249 | } | |
3250 | ||
3251 | if (vi->mergeable_rx_bufs && !vi->any_header_sg) { | |
4d463c4d | 3252 | NL_SET_ERR_MSG_MOD(extack, "XDP expects header/data in single page, any_header_sg required"); |
f600b690 JF |
3253 | return -EINVAL; |
3254 | } | |
3255 | ||
8d9bc36d HQ |
3256 | if (prog && !prog->aux->xdp_has_frags && dev->mtu > max_sz) { |
3257 | NL_SET_ERR_MSG_MOD(extack, "MTU too large to enable XDP without frags"); | |
3258 | netdev_warn(dev, "single-buffer XDP requires MTU less than %u\n", max_sz); | |
f600b690 JF |
3259 | return -EINVAL; |
3260 | } | |
3261 | ||
672aafd5 JF |
3262 | curr_qp = vi->curr_queue_pairs - vi->xdp_queue_pairs; |
3263 | if (prog) | |
3264 | xdp_qp = nr_cpu_ids; | |
3265 | ||
3266 | /* XDP requires extra queues for XDP_TX */ | |
3267 | if (curr_qp + xdp_qp > vi->max_queue_pairs) { | |
9ce4e3d6 XZ |
3268 | netdev_warn_once(dev, "XDP request %i queues but max is %i. XDP_TX and XDP_REDIRECT will operate in a slower locked tx mode.\n", |
3269 | curr_qp + xdp_qp, vi->max_queue_pairs); | |
97c2c69e | 3270 | xdp_qp = 0; |
672aafd5 JF |
3271 | } |
3272 | ||
03aa6d34 TM |
3273 | old_prog = rtnl_dereference(vi->rq[0].xdp_prog); |
3274 | if (!prog && !old_prog) | |
3275 | return 0; | |
3276 | ||
85192dbf AN |
3277 | if (prog) |
3278 | bpf_prog_add(prog, vi->max_queue_pairs - 1); | |
2de2f7f4 | 3279 | |
4941d472 | 3280 | /* Make sure NAPI is not using any XDP TX queues for RX. */ |
534da5e8 TM |
3281 | if (netif_running(dev)) { |
3282 | for (i = 0; i < vi->max_queue_pairs; i++) { | |
4e09ff53 | 3283 | napi_disable(&vi->rq[i].napi); |
534da5e8 TM |
3284 | virtnet_napi_tx_disable(&vi->sq[i].napi); |
3285 | } | |
3286 | } | |
f600b690 | 3287 | |
03aa6d34 TM |
3288 | if (!prog) { |
3289 | for (i = 0; i < vi->max_queue_pairs; i++) { | |
3290 | rcu_assign_pointer(vi->rq[i].xdp_prog, prog); | |
3291 | if (i == 0) | |
3292 | virtnet_restore_guest_offloads(vi); | |
3293 | } | |
3294 | synchronize_net(); | |
3295 | } | |
f600b690 | 3296 | |
4941d472 JW |
3297 | err = _virtnet_set_queues(vi, curr_qp + xdp_qp); |
3298 | if (err) | |
3299 | goto err; | |
188313c1 | 3300 | netif_set_real_num_rx_queues(dev, curr_qp + xdp_qp); |
4941d472 | 3301 | vi->xdp_queue_pairs = xdp_qp; |
672aafd5 | 3302 | |
03aa6d34 | 3303 | if (prog) { |
97c2c69e | 3304 | vi->xdp_enabled = true; |
03aa6d34 TM |
3305 | for (i = 0; i < vi->max_queue_pairs; i++) { |
3306 | rcu_assign_pointer(vi->rq[i].xdp_prog, prog); | |
3307 | if (i == 0 && !old_prog) | |
3f93522f | 3308 | virtnet_clear_guest_offloads(vi); |
3f93522f | 3309 | } |
66c0e13a | 3310 | if (!old_prog) |
30bbf891 | 3311 | xdp_features_set_redirect_target(dev, true); |
97c2c69e | 3312 | } else { |
66c0e13a | 3313 | xdp_features_clear_redirect_target(dev); |
97c2c69e | 3314 | vi->xdp_enabled = false; |
03aa6d34 TM |
3315 | } |
3316 | ||
3317 | for (i = 0; i < vi->max_queue_pairs; i++) { | |
f600b690 JF |
3318 | if (old_prog) |
3319 | bpf_prog_put(old_prog); | |
534da5e8 | 3320 | if (netif_running(dev)) { |
4e09ff53 | 3321 | virtnet_napi_enable(vi->rq[i].vq, &vi->rq[i].napi); |
534da5e8 TM |
3322 | virtnet_napi_tx_enable(vi, vi->sq[i].vq, |
3323 | &vi->sq[i].napi); | |
3324 | } | |
f600b690 JF |
3325 | } |
3326 | ||
3327 | return 0; | |
2de2f7f4 | 3328 | |
4941d472 | 3329 | err: |
03aa6d34 TM |
3330 | if (!prog) { |
3331 | virtnet_clear_guest_offloads(vi); | |
3332 | for (i = 0; i < vi->max_queue_pairs; i++) | |
3333 | rcu_assign_pointer(vi->rq[i].xdp_prog, old_prog); | |
3334 | } | |
3335 | ||
8be4d9a4 | 3336 | if (netif_running(dev)) { |
534da5e8 | 3337 | for (i = 0; i < vi->max_queue_pairs; i++) { |
8be4d9a4 | 3338 | virtnet_napi_enable(vi->rq[i].vq, &vi->rq[i].napi); |
534da5e8 TM |
3339 | virtnet_napi_tx_enable(vi, vi->sq[i].vq, |
3340 | &vi->sq[i].napi); | |
3341 | } | |
8be4d9a4 | 3342 | } |
2de2f7f4 JF |
3343 | if (prog) |
3344 | bpf_prog_sub(prog, vi->max_queue_pairs - 1); | |
3345 | return err; | |
f600b690 JF |
3346 | } |
3347 | ||
f4e63525 | 3348 | static int virtnet_xdp(struct net_device *dev, struct netdev_bpf *xdp) |
f600b690 JF |
3349 | { |
3350 | switch (xdp->command) { | |
3351 | case XDP_SETUP_PROG: | |
9861ce03 | 3352 | return virtnet_xdp_set(dev, xdp->prog, xdp->extack); |
f600b690 JF |
3353 | default: |
3354 | return -EINVAL; | |
3355 | } | |
3356 | } | |
3357 | ||
ba5e4426 SS |
3358 | static int virtnet_get_phys_port_name(struct net_device *dev, char *buf, |
3359 | size_t len) | |
3360 | { | |
3361 | struct virtnet_info *vi = netdev_priv(dev); | |
3362 | int ret; | |
3363 | ||
3364 | if (!virtio_has_feature(vi->vdev, VIRTIO_NET_F_STANDBY)) | |
3365 | return -EOPNOTSUPP; | |
3366 | ||
3367 | ret = snprintf(buf, len, "sby"); | |
3368 | if (ret >= len) | |
3369 | return -EOPNOTSUPP; | |
3370 | ||
3371 | return 0; | |
3372 | } | |
3373 | ||
a02e8964 WB |
3374 | static int virtnet_set_features(struct net_device *dev, |
3375 | netdev_features_t features) | |
3376 | { | |
3377 | struct virtnet_info *vi = netdev_priv(dev); | |
cf8691cb | 3378 | u64 offloads; |
a02e8964 WB |
3379 | int err; |
3380 | ||
dbcf24d1 | 3381 | if ((dev->features ^ features) & NETIF_F_GRO_HW) { |
97c2c69e | 3382 | if (vi->xdp_enabled) |
cf8691cb MT |
3383 | return -EBUSY; |
3384 | ||
dbcf24d1 | 3385 | if (features & NETIF_F_GRO_HW) |
cf8691cb | 3386 | offloads = vi->guest_offloads_capable; |
a02e8964 | 3387 | else |
cf8691cb | 3388 | offloads = vi->guest_offloads_capable & |
dbcf24d1 | 3389 | ~GUEST_OFFLOAD_GRO_HW_MASK; |
a02e8964 | 3390 | |
cf8691cb MT |
3391 | err = virtnet_set_guest_offloads(vi, offloads); |
3392 | if (err) | |
3393 | return err; | |
3394 | vi->guest_offloads = offloads; | |
a02e8964 WB |
3395 | } |
3396 | ||
c7114b12 AM |
3397 | if ((dev->features ^ features) & NETIF_F_RXHASH) { |
3398 | if (features & NETIF_F_RXHASH) | |
c1170820 | 3399 | vi->ctrl->rss.hash_types = vi->rss_hash_types_saved; |
c7114b12 AM |
3400 | else |
3401 | vi->ctrl->rss.hash_types = VIRTIO_NET_HASH_REPORT_NONE; | |
3402 | ||
3403 | if (!virtnet_commit_rss_command(vi)) | |
3404 | return -EINVAL; | |
3405 | } | |
3406 | ||
a02e8964 WB |
3407 | return 0; |
3408 | } | |
3409 | ||
a520794b TL |
3410 | static void virtnet_tx_timeout(struct net_device *dev, unsigned int txqueue) |
3411 | { | |
3412 | struct virtnet_info *priv = netdev_priv(dev); | |
3413 | struct send_queue *sq = &priv->sq[txqueue]; | |
3414 | struct netdev_queue *txq = netdev_get_tx_queue(dev, txqueue); | |
3415 | ||
3416 | u64_stats_update_begin(&sq->stats.syncp); | |
3417 | sq->stats.tx_timeouts++; | |
3418 | u64_stats_update_end(&sq->stats.syncp); | |
3419 | ||
3420 | netdev_err(dev, "TX timeout on queue: %u, sq: %s, vq: 0x%x, name: %s, %u usecs ago\n", | |
3421 | txqueue, sq->name, sq->vq->index, sq->vq->name, | |
5337824f | 3422 | jiffies_to_usecs(jiffies - READ_ONCE(txq->trans_start))); |
a520794b TL |
3423 | } |
3424 | ||
76288b4e SH |
3425 | static const struct net_device_ops virtnet_netdev = { |
3426 | .ndo_open = virtnet_open, | |
3427 | .ndo_stop = virtnet_close, | |
3428 | .ndo_start_xmit = start_xmit, | |
3429 | .ndo_validate_addr = eth_validate_addr, | |
9c46f6d4 | 3430 | .ndo_set_mac_address = virtnet_set_mac_address, |
2af7698e | 3431 | .ndo_set_rx_mode = virtnet_set_rx_mode, |
3fa2a1df | 3432 | .ndo_get_stats64 = virtnet_stats, |
1824a989 AW |
3433 | .ndo_vlan_rx_add_vid = virtnet_vlan_rx_add_vid, |
3434 | .ndo_vlan_rx_kill_vid = virtnet_vlan_rx_kill_vid, | |
f4e63525 | 3435 | .ndo_bpf = virtnet_xdp, |
186b3c99 | 3436 | .ndo_xdp_xmit = virtnet_xdp_xmit, |
2836b4f2 | 3437 | .ndo_features_check = passthru_features_check, |
ba5e4426 | 3438 | .ndo_get_phys_port_name = virtnet_get_phys_port_name, |
a02e8964 | 3439 | .ndo_set_features = virtnet_set_features, |
a520794b | 3440 | .ndo_tx_timeout = virtnet_tx_timeout, |
76288b4e SH |
3441 | }; |
3442 | ||
586d17c5 | 3443 | static void virtnet_config_changed_work(struct work_struct *work) |
9f4d26d0 | 3444 | { |
586d17c5 JW |
3445 | struct virtnet_info *vi = |
3446 | container_of(work, struct virtnet_info, config_work); | |
9f4d26d0 MM |
3447 | u16 v; |
3448 | ||
855e0c52 RR |
3449 | if (virtio_cread_feature(vi->vdev, VIRTIO_NET_F_STATUS, |
3450 | struct virtio_net_config, status, &v) < 0) | |
507613bf | 3451 | return; |
586d17c5 JW |
3452 | |
3453 | if (v & VIRTIO_NET_S_ANNOUNCE) { | |
ee89bab1 | 3454 | netdev_notify_peers(vi->dev); |
586d17c5 JW |
3455 | virtnet_ack_link_announce(vi); |
3456 | } | |
9f4d26d0 MM |
3457 | |
3458 | /* Ignore unknown (future) status bits */ | |
3459 | v &= VIRTIO_NET_S_LINK_UP; | |
3460 | ||
3461 | if (vi->status == v) | |
507613bf | 3462 | return; |
9f4d26d0 MM |
3463 | |
3464 | vi->status = v; | |
3465 | ||
3466 | if (vi->status & VIRTIO_NET_S_LINK_UP) { | |
faa9b39f | 3467 | virtnet_update_settings(vi); |
9f4d26d0 | 3468 | netif_carrier_on(vi->dev); |
986a4f4d | 3469 | netif_tx_wake_all_queues(vi->dev); |
9f4d26d0 MM |
3470 | } else { |
3471 | netif_carrier_off(vi->dev); | |
986a4f4d | 3472 | netif_tx_stop_all_queues(vi->dev); |
9f4d26d0 MM |
3473 | } |
3474 | } | |
3475 | ||
3476 | static void virtnet_config_changed(struct virtio_device *vdev) | |
3477 | { | |
3478 | struct virtnet_info *vi = vdev->priv; | |
3479 | ||
3b07e9ca | 3480 | schedule_work(&vi->config_work); |
9f4d26d0 MM |
3481 | } |
3482 | ||
986a4f4d JW |
3483 | static void virtnet_free_queues(struct virtnet_info *vi) |
3484 | { | |
d4fb84ee AV |
3485 | int i; |
3486 | ||
ab3971b1 | 3487 | for (i = 0; i < vi->max_queue_pairs; i++) { |
5198d545 JK |
3488 | __netif_napi_del(&vi->rq[i].napi); |
3489 | __netif_napi_del(&vi->sq[i].napi); | |
ab3971b1 | 3490 | } |
d4fb84ee | 3491 | |
5198d545 | 3492 | /* We called __netif_napi_del(), |
963abe5c ED |
3493 | * we need to respect an RCU grace period before freeing vi->rq |
3494 | */ | |
3495 | synchronize_net(); | |
3496 | ||
986a4f4d JW |
3497 | kfree(vi->rq); |
3498 | kfree(vi->sq); | |
12e57169 | 3499 | kfree(vi->ctrl); |
986a4f4d JW |
3500 | } |
3501 | ||
47315329 | 3502 | static void _free_receive_bufs(struct virtnet_info *vi) |
986a4f4d | 3503 | { |
f600b690 | 3504 | struct bpf_prog *old_prog; |
986a4f4d JW |
3505 | int i; |
3506 | ||
3507 | for (i = 0; i < vi->max_queue_pairs; i++) { | |
3508 | while (vi->rq[i].pages) | |
3509 | __free_pages(get_a_page(&vi->rq[i], GFP_KERNEL), 0); | |
f600b690 JF |
3510 | |
3511 | old_prog = rtnl_dereference(vi->rq[i].xdp_prog); | |
3512 | RCU_INIT_POINTER(vi->rq[i].xdp_prog, NULL); | |
3513 | if (old_prog) | |
3514 | bpf_prog_put(old_prog); | |
986a4f4d | 3515 | } |
47315329 JF |
3516 | } |
3517 | ||
3518 | static void free_receive_bufs(struct virtnet_info *vi) | |
3519 | { | |
3520 | rtnl_lock(); | |
3521 | _free_receive_bufs(vi); | |
f600b690 | 3522 | rtnl_unlock(); |
986a4f4d JW |
3523 | } |
3524 | ||
fb51879d MD |
3525 | static void free_receive_page_frags(struct virtnet_info *vi) |
3526 | { | |
3527 | int i; | |
3528 | for (i = 0; i < vi->max_queue_pairs; i++) | |
3529 | if (vi->rq[i].alloc_frag.page) | |
3530 | put_page(vi->rq[i].alloc_frag.page); | |
3531 | } | |
3532 | ||
6e345f8c XZ |
3533 | static void virtnet_sq_free_unused_buf(struct virtqueue *vq, void *buf) |
3534 | { | |
3535 | if (!is_xdp_frame(buf)) | |
3536 | dev_kfree_skb(buf); | |
3537 | else | |
3538 | xdp_return_frame(ptr_to_xdp(buf)); | |
3539 | } | |
3540 | ||
3541 | static void virtnet_rq_free_unused_buf(struct virtqueue *vq, void *buf) | |
3542 | { | |
3543 | struct virtnet_info *vi = vq->vdev->priv; | |
3544 | int i = vq2rxq(vq); | |
3545 | ||
3546 | if (vi->mergeable_rx_bufs) | |
3547 | put_page(virt_to_head_page(buf)); | |
3548 | else if (vi->big_packets) | |
3549 | give_pages(&vi->rq[i], buf); | |
3550 | else | |
3551 | put_page(virt_to_head_page(buf)); | |
3552 | } | |
3553 | ||
986a4f4d JW |
3554 | static void free_unused_bufs(struct virtnet_info *vi) |
3555 | { | |
3556 | void *buf; | |
3557 | int i; | |
3558 | ||
3559 | for (i = 0; i < vi->max_queue_pairs; i++) { | |
3560 | struct virtqueue *vq = vi->sq[i].vq; | |
6e345f8c XZ |
3561 | while ((buf = virtqueue_detach_unused_buf(vq)) != NULL) |
3562 | virtnet_sq_free_unused_buf(vq, buf); | |
f8bb5104 | 3563 | cond_resched(); |
986a4f4d JW |
3564 | } |
3565 | ||
3566 | for (i = 0; i < vi->max_queue_pairs; i++) { | |
3567 | struct virtqueue *vq = vi->rq[i].vq; | |
6e345f8c XZ |
3568 | while ((buf = virtqueue_detach_unused_buf(vq)) != NULL) |
3569 | virtnet_rq_free_unused_buf(vq, buf); | |
f8bb5104 | 3570 | cond_resched(); |
986a4f4d JW |
3571 | } |
3572 | } | |
3573 | ||
e9d7417b JW |
3574 | static void virtnet_del_vqs(struct virtnet_info *vi) |
3575 | { | |
3576 | struct virtio_device *vdev = vi->vdev; | |
3577 | ||
310974fa | 3578 | virtnet_clean_affinity(vi); |
986a4f4d | 3579 | |
e9d7417b | 3580 | vdev->config->del_vqs(vdev); |
986a4f4d JW |
3581 | |
3582 | virtnet_free_queues(vi); | |
e9d7417b JW |
3583 | } |
3584 | ||
d85b758f MT |
3585 | /* How large should a single buffer be so a queue full of these can fit at |
3586 | * least one full packet? | |
3587 | * Logic below assumes the mergeable buffer header is used. | |
3588 | */ | |
3589 | static unsigned int mergeable_min_buf_len(struct virtnet_info *vi, struct virtqueue *vq) | |
3590 | { | |
c1ddc42d | 3591 | const unsigned int hdr_len = vi->hdr_len; |
d85b758f MT |
3592 | unsigned int rq_size = virtqueue_get_vring_size(vq); |
3593 | unsigned int packet_len = vi->big_packets ? IP_MAX_MTU : vi->dev->max_mtu; | |
3594 | unsigned int buf_len = hdr_len + ETH_HLEN + VLAN_HLEN + packet_len; | |
3595 | unsigned int min_buf_len = DIV_ROUND_UP(buf_len, rq_size); | |
3596 | ||
f0c3192c MT |
3597 | return max(max(min_buf_len, hdr_len) - hdr_len, |
3598 | (unsigned int)GOOD_PACKET_LEN); | |
d85b758f MT |
3599 | } |
3600 | ||
986a4f4d | 3601 | static int virtnet_find_vqs(struct virtnet_info *vi) |
3f9c10b0 | 3602 | { |
986a4f4d JW |
3603 | vq_callback_t **callbacks; |
3604 | struct virtqueue **vqs; | |
3605 | int ret = -ENOMEM; | |
3606 | int i, total_vqs; | |
3607 | const char **names; | |
d45b897b | 3608 | bool *ctx; |
986a4f4d JW |
3609 | |
3610 | /* We expect 1 RX virtqueue followed by 1 TX virtqueue, followed by | |
3611 | * possible N-1 RX/TX queue pairs used in multiqueue mode, followed by | |
3612 | * possible control vq. | |
3613 | */ | |
3614 | total_vqs = vi->max_queue_pairs * 2 + | |
3615 | virtio_has_feature(vi->vdev, VIRTIO_NET_F_CTRL_VQ); | |
3616 | ||
3617 | /* Allocate space for find_vqs parameters */ | |
6396bb22 | 3618 | vqs = kcalloc(total_vqs, sizeof(*vqs), GFP_KERNEL); |
986a4f4d JW |
3619 | if (!vqs) |
3620 | goto err_vq; | |
6da2ec56 | 3621 | callbacks = kmalloc_array(total_vqs, sizeof(*callbacks), GFP_KERNEL); |
986a4f4d JW |
3622 | if (!callbacks) |
3623 | goto err_callback; | |
6da2ec56 | 3624 | names = kmalloc_array(total_vqs, sizeof(*names), GFP_KERNEL); |
986a4f4d JW |
3625 | if (!names) |
3626 | goto err_names; | |
192f68cf | 3627 | if (!vi->big_packets || vi->mergeable_rx_bufs) { |
6396bb22 | 3628 | ctx = kcalloc(total_vqs, sizeof(*ctx), GFP_KERNEL); |
d45b897b MT |
3629 | if (!ctx) |
3630 | goto err_ctx; | |
3631 | } else { | |
3632 | ctx = NULL; | |
3633 | } | |
986a4f4d JW |
3634 | |
3635 | /* Parameters for control virtqueue, if any */ | |
3636 | if (vi->has_cvq) { | |
3637 | callbacks[total_vqs - 1] = NULL; | |
3638 | names[total_vqs - 1] = "control"; | |
3639 | } | |
3f9c10b0 | 3640 | |
986a4f4d JW |
3641 | /* Allocate/initialize parameters for send/receive virtqueues */ |
3642 | for (i = 0; i < vi->max_queue_pairs; i++) { | |
3643 | callbacks[rxq2vq(i)] = skb_recv_done; | |
3644 | callbacks[txq2vq(i)] = skb_xmit_done; | |
3645 | sprintf(vi->rq[i].name, "input.%d", i); | |
3646 | sprintf(vi->sq[i].name, "output.%d", i); | |
3647 | names[rxq2vq(i)] = vi->rq[i].name; | |
3648 | names[txq2vq(i)] = vi->sq[i].name; | |
d45b897b MT |
3649 | if (ctx) |
3650 | ctx[rxq2vq(i)] = true; | |
986a4f4d | 3651 | } |
3f9c10b0 | 3652 | |
2e9ca760 MT |
3653 | ret = virtio_find_vqs_ctx(vi->vdev, total_vqs, vqs, callbacks, |
3654 | names, ctx, NULL); | |
986a4f4d JW |
3655 | if (ret) |
3656 | goto err_find; | |
3f9c10b0 | 3657 | |
986a4f4d JW |
3658 | if (vi->has_cvq) { |
3659 | vi->cvq = vqs[total_vqs - 1]; | |
3f9c10b0 | 3660 | if (virtio_has_feature(vi->vdev, VIRTIO_NET_F_CTRL_VLAN)) |
f646968f | 3661 | vi->dev->features |= NETIF_F_HW_VLAN_CTAG_FILTER; |
3f9c10b0 | 3662 | } |
986a4f4d JW |
3663 | |
3664 | for (i = 0; i < vi->max_queue_pairs; i++) { | |
3665 | vi->rq[i].vq = vqs[rxq2vq(i)]; | |
d85b758f | 3666 | vi->rq[i].min_buf_len = mergeable_min_buf_len(vi, vi->rq[i].vq); |
986a4f4d JW |
3667 | vi->sq[i].vq = vqs[txq2vq(i)]; |
3668 | } | |
3669 | ||
2fa3c8a8 | 3670 | /* run here: ret == 0. */ |
986a4f4d | 3671 | |
986a4f4d JW |
3672 | |
3673 | err_find: | |
d45b897b MT |
3674 | kfree(ctx); |
3675 | err_ctx: | |
986a4f4d JW |
3676 | kfree(names); |
3677 | err_names: | |
3678 | kfree(callbacks); | |
3679 | err_callback: | |
3680 | kfree(vqs); | |
3681 | err_vq: | |
3682 | return ret; | |
3683 | } | |
3684 | ||
3685 | static int virtnet_alloc_queues(struct virtnet_info *vi) | |
3686 | { | |
3687 | int i; | |
3688 | ||
122b84a1 MG |
3689 | if (vi->has_cvq) { |
3690 | vi->ctrl = kzalloc(sizeof(*vi->ctrl), GFP_KERNEL); | |
3691 | if (!vi->ctrl) | |
3692 | goto err_ctrl; | |
3693 | } else { | |
3694 | vi->ctrl = NULL; | |
3695 | } | |
6396bb22 | 3696 | vi->sq = kcalloc(vi->max_queue_pairs, sizeof(*vi->sq), GFP_KERNEL); |
986a4f4d JW |
3697 | if (!vi->sq) |
3698 | goto err_sq; | |
6396bb22 | 3699 | vi->rq = kcalloc(vi->max_queue_pairs, sizeof(*vi->rq), GFP_KERNEL); |
008d4278 | 3700 | if (!vi->rq) |
986a4f4d JW |
3701 | goto err_rq; |
3702 | ||
3703 | INIT_DELAYED_WORK(&vi->refill, refill_work); | |
3704 | for (i = 0; i < vi->max_queue_pairs; i++) { | |
3705 | vi->rq[i].pages = NULL; | |
d484735d JK |
3706 | netif_napi_add_weight(vi->dev, &vi->rq[i].napi, virtnet_poll, |
3707 | napi_weight); | |
8d602e1a JK |
3708 | netif_napi_add_tx_weight(vi->dev, &vi->sq[i].napi, |
3709 | virtnet_poll_tx, | |
3710 | napi_tx ? napi_weight : 0); | |
986a4f4d JW |
3711 | |
3712 | sg_init_table(vi->rq[i].sg, ARRAY_SIZE(vi->rq[i].sg)); | |
5377d758 | 3713 | ewma_pkt_len_init(&vi->rq[i].mrg_avg_pkt_len); |
986a4f4d | 3714 | sg_init_table(vi->sq[i].sg, ARRAY_SIZE(vi->sq[i].sg)); |
d7dfc5cf TM |
3715 | |
3716 | u64_stats_init(&vi->rq[i].stats.syncp); | |
3717 | u64_stats_init(&vi->sq[i].stats.syncp); | |
986a4f4d JW |
3718 | } |
3719 | ||
3720 | return 0; | |
3721 | ||
3722 | err_rq: | |
3723 | kfree(vi->sq); | |
3724 | err_sq: | |
12e57169 MT |
3725 | kfree(vi->ctrl); |
3726 | err_ctrl: | |
986a4f4d JW |
3727 | return -ENOMEM; |
3728 | } | |
3729 | ||
3730 | static int init_vqs(struct virtnet_info *vi) | |
3731 | { | |
3732 | int ret; | |
3733 | ||
3734 | /* Allocate send & receive queues */ | |
3735 | ret = virtnet_alloc_queues(vi); | |
3736 | if (ret) | |
3737 | goto err; | |
3738 | ||
3739 | ret = virtnet_find_vqs(vi); | |
3740 | if (ret) | |
3741 | goto err_free; | |
3742 | ||
a0d1d0f4 | 3743 | cpus_read_lock(); |
8898c21c | 3744 | virtnet_set_affinity(vi); |
a0d1d0f4 | 3745 | cpus_read_unlock(); |
47be2479 | 3746 | |
986a4f4d JW |
3747 | return 0; |
3748 | ||
3749 | err_free: | |
3750 | virtnet_free_queues(vi); | |
3751 | err: | |
3752 | return ret; | |
3f9c10b0 AS |
3753 | } |
3754 | ||
fbf28d78 MD |
3755 | #ifdef CONFIG_SYSFS |
3756 | static ssize_t mergeable_rx_buffer_size_show(struct netdev_rx_queue *queue, | |
718ad681 | 3757 | char *buf) |
fbf28d78 MD |
3758 | { |
3759 | struct virtnet_info *vi = netdev_priv(queue->dev); | |
3760 | unsigned int queue_index = get_netdev_rx_queue_index(queue); | |
3cc81a9a JW |
3761 | unsigned int headroom = virtnet_get_headroom(vi); |
3762 | unsigned int tailroom = headroom ? sizeof(struct skb_shared_info) : 0; | |
5377d758 | 3763 | struct ewma_pkt_len *avg; |
fbf28d78 MD |
3764 | |
3765 | BUG_ON(queue_index >= vi->max_queue_pairs); | |
3766 | avg = &vi->rq[queue_index].mrg_avg_pkt_len; | |
d85b758f | 3767 | return sprintf(buf, "%u\n", |
3cc81a9a JW |
3768 | get_mergeable_buf_len(&vi->rq[queue_index], avg, |
3769 | SKB_DATA_ALIGN(headroom + tailroom))); | |
fbf28d78 MD |
3770 | } |
3771 | ||
3772 | static struct rx_queue_attribute mergeable_rx_buffer_size_attribute = | |
3773 | __ATTR_RO(mergeable_rx_buffer_size); | |
3774 | ||
3775 | static struct attribute *virtio_net_mrg_rx_attrs[] = { | |
3776 | &mergeable_rx_buffer_size_attribute.attr, | |
3777 | NULL | |
3778 | }; | |
3779 | ||
3780 | static const struct attribute_group virtio_net_mrg_rx_group = { | |
3781 | .name = "virtio_net", | |
3782 | .attrs = virtio_net_mrg_rx_attrs | |
3783 | }; | |
3784 | #endif | |
3785 | ||
892d6eb1 JW |
3786 | static bool virtnet_fail_on_feature(struct virtio_device *vdev, |
3787 | unsigned int fbit, | |
3788 | const char *fname, const char *dname) | |
3789 | { | |
3790 | if (!virtio_has_feature(vdev, fbit)) | |
3791 | return false; | |
3792 | ||
3793 | dev_err(&vdev->dev, "device advertises feature %s but not %s", | |
3794 | fname, dname); | |
3795 | ||
3796 | return true; | |
3797 | } | |
3798 | ||
3799 | #define VIRTNET_FAIL_ON(vdev, fbit, dbit) \ | |
3800 | virtnet_fail_on_feature(vdev, fbit, #fbit, dbit) | |
3801 | ||
3802 | static bool virtnet_validate_features(struct virtio_device *vdev) | |
3803 | { | |
3804 | if (!virtio_has_feature(vdev, VIRTIO_NET_F_CTRL_VQ) && | |
3805 | (VIRTNET_FAIL_ON(vdev, VIRTIO_NET_F_CTRL_RX, | |
3806 | "VIRTIO_NET_F_CTRL_VQ") || | |
3807 | VIRTNET_FAIL_ON(vdev, VIRTIO_NET_F_CTRL_VLAN, | |
3808 | "VIRTIO_NET_F_CTRL_VQ") || | |
3809 | VIRTNET_FAIL_ON(vdev, VIRTIO_NET_F_GUEST_ANNOUNCE, | |
3810 | "VIRTIO_NET_F_CTRL_VQ") || | |
3811 | VIRTNET_FAIL_ON(vdev, VIRTIO_NET_F_MQ, "VIRTIO_NET_F_CTRL_VQ") || | |
3812 | VIRTNET_FAIL_ON(vdev, VIRTIO_NET_F_CTRL_MAC_ADDR, | |
c7114b12 AM |
3813 | "VIRTIO_NET_F_CTRL_VQ") || |
3814 | VIRTNET_FAIL_ON(vdev, VIRTIO_NET_F_RSS, | |
91f41f01 AM |
3815 | "VIRTIO_NET_F_CTRL_VQ") || |
3816 | VIRTNET_FAIL_ON(vdev, VIRTIO_NET_F_HASH_REPORT, | |
699b045a AK |
3817 | "VIRTIO_NET_F_CTRL_VQ") || |
3818 | VIRTNET_FAIL_ON(vdev, VIRTIO_NET_F_NOTF_COAL, | |
892d6eb1 JW |
3819 | "VIRTIO_NET_F_CTRL_VQ"))) { |
3820 | return false; | |
3821 | } | |
3822 | ||
3823 | return true; | |
3824 | } | |
3825 | ||
d0c2c997 JW |
3826 | #define MIN_MTU ETH_MIN_MTU |
3827 | #define MAX_MTU ETH_MAX_MTU | |
3828 | ||
fe36cbe0 | 3829 | static int virtnet_validate(struct virtio_device *vdev) |
296f96fc | 3830 | { |
6ba42248 MT |
3831 | if (!vdev->config->get) { |
3832 | dev_err(&vdev->dev, "%s failure: config access disabled\n", | |
3833 | __func__); | |
3834 | return -EINVAL; | |
3835 | } | |
3836 | ||
892d6eb1 JW |
3837 | if (!virtnet_validate_features(vdev)) |
3838 | return -EINVAL; | |
3839 | ||
fe36cbe0 MT |
3840 | if (virtio_has_feature(vdev, VIRTIO_NET_F_MTU)) { |
3841 | int mtu = virtio_cread16(vdev, | |
3842 | offsetof(struct virtio_net_config, | |
3843 | mtu)); | |
3844 | if (mtu < MIN_MTU) | |
3845 | __virtio_clear_bit(vdev, VIRTIO_NET_F_MTU); | |
3846 | } | |
3847 | ||
7c06458c LV |
3848 | if (virtio_has_feature(vdev, VIRTIO_NET_F_STANDBY) && |
3849 | !virtio_has_feature(vdev, VIRTIO_NET_F_MAC)) { | |
3850 | dev_warn(&vdev->dev, "device advertises feature VIRTIO_NET_F_STANDBY but not VIRTIO_NET_F_MAC, disabling standby"); | |
3851 | __virtio_clear_bit(vdev, VIRTIO_NET_F_STANDBY); | |
3852 | } | |
3853 | ||
fe36cbe0 MT |
3854 | return 0; |
3855 | } | |
3856 | ||
46cd26f4 GL |
3857 | static bool virtnet_check_guest_gso(const struct virtnet_info *vi) |
3858 | { | |
3859 | return virtio_has_feature(vi->vdev, VIRTIO_NET_F_GUEST_TSO4) || | |
3860 | virtio_has_feature(vi->vdev, VIRTIO_NET_F_GUEST_TSO6) || | |
3861 | virtio_has_feature(vi->vdev, VIRTIO_NET_F_GUEST_ECN) || | |
418044e1 AM |
3862 | virtio_has_feature(vi->vdev, VIRTIO_NET_F_GUEST_UFO) || |
3863 | (virtio_has_feature(vi->vdev, VIRTIO_NET_F_GUEST_USO4) && | |
3864 | virtio_has_feature(vi->vdev, VIRTIO_NET_F_GUEST_USO6)); | |
46cd26f4 GL |
3865 | } |
3866 | ||
4959aebb GL |
3867 | static void virtnet_set_big_packets(struct virtnet_info *vi, const int mtu) |
3868 | { | |
3869 | bool guest_gso = virtnet_check_guest_gso(vi); | |
3870 | ||
3871 | /* If device can receive ANY guest GSO packets, regardless of mtu, | |
3872 | * allocate packets of maximum size, otherwise limit it to only | |
3873 | * mtu size worth only. | |
3874 | */ | |
3875 | if (mtu > ETH_DATA_LEN || guest_gso) { | |
3876 | vi->big_packets = true; | |
3877 | vi->big_packets_num_skbfrags = guest_gso ? MAX_SKB_FRAGS : DIV_ROUND_UP(mtu, PAGE_SIZE); | |
3878 | } | |
3879 | } | |
3880 | ||
fe36cbe0 MT |
3881 | static int virtnet_probe(struct virtio_device *vdev) |
3882 | { | |
d7dfc5cf | 3883 | int i, err = -ENOMEM; |
fe36cbe0 MT |
3884 | struct net_device *dev; |
3885 | struct virtnet_info *vi; | |
3886 | u16 max_queue_pairs; | |
4959aebb | 3887 | int mtu = 0; |
fe36cbe0 | 3888 | |
c7114b12 AM |
3889 | /* Find if host supports multiqueue/rss virtio_net device */ |
3890 | max_queue_pairs = 1; | |
3891 | if (virtio_has_feature(vdev, VIRTIO_NET_F_MQ) || virtio_has_feature(vdev, VIRTIO_NET_F_RSS)) | |
3892 | max_queue_pairs = | |
3893 | virtio_cread16(vdev, offsetof(struct virtio_net_config, max_virtqueue_pairs)); | |
986a4f4d JW |
3894 | |
3895 | /* We need at least 2 queue's */ | |
c7114b12 | 3896 | if (max_queue_pairs < VIRTIO_NET_CTRL_MQ_VQ_PAIRS_MIN || |
986a4f4d JW |
3897 | max_queue_pairs > VIRTIO_NET_CTRL_MQ_VQ_PAIRS_MAX || |
3898 | !virtio_has_feature(vdev, VIRTIO_NET_F_CTRL_VQ)) | |
3899 | max_queue_pairs = 1; | |
296f96fc RR |
3900 | |
3901 | /* Allocate ourselves a network device with room for our info */ | |
986a4f4d | 3902 | dev = alloc_etherdev_mq(sizeof(struct virtnet_info), max_queue_pairs); |
296f96fc RR |
3903 | if (!dev) |
3904 | return -ENOMEM; | |
3905 | ||
3906 | /* Set up network device as normal. */ | |
ab5bd583 XZ |
3907 | dev->priv_flags |= IFF_UNICAST_FLT | IFF_LIVE_ADDR_CHANGE | |
3908 | IFF_TX_SKB_NO_LINEAR; | |
76288b4e | 3909 | dev->netdev_ops = &virtnet_netdev; |
296f96fc | 3910 | dev->features = NETIF_F_HIGHDMA; |
3fa2a1df | 3911 | |
7ad24ea4 | 3912 | dev->ethtool_ops = &virtnet_ethtool_ops; |
296f96fc RR |
3913 | SET_NETDEV_DEV(dev, &vdev->dev); |
3914 | ||
3915 | /* Do we support "hardware" checksums? */ | |
98e778c9 | 3916 | if (virtio_has_feature(vdev, VIRTIO_NET_F_CSUM)) { |
296f96fc | 3917 | /* This opens up the world of extra features. */ |
48900cb6 | 3918 | dev->hw_features |= NETIF_F_HW_CSUM | NETIF_F_SG; |
98e778c9 | 3919 | if (csum) |
48900cb6 | 3920 | dev->features |= NETIF_F_HW_CSUM | NETIF_F_SG; |
98e778c9 MM |
3921 | |
3922 | if (virtio_has_feature(vdev, VIRTIO_NET_F_GSO)) { | |
e078de03 | 3923 | dev->hw_features |= NETIF_F_TSO |
34a48579 RR |
3924 | | NETIF_F_TSO_ECN | NETIF_F_TSO6; |
3925 | } | |
5539ae96 | 3926 | /* Individual feature bits: what can host handle? */ |
98e778c9 MM |
3927 | if (virtio_has_feature(vdev, VIRTIO_NET_F_HOST_TSO4)) |
3928 | dev->hw_features |= NETIF_F_TSO; | |
3929 | if (virtio_has_feature(vdev, VIRTIO_NET_F_HOST_TSO6)) | |
3930 | dev->hw_features |= NETIF_F_TSO6; | |
3931 | if (virtio_has_feature(vdev, VIRTIO_NET_F_HOST_ECN)) | |
3932 | dev->hw_features |= NETIF_F_TSO_ECN; | |
418044e1 AM |
3933 | if (virtio_has_feature(vdev, VIRTIO_NET_F_HOST_USO)) |
3934 | dev->hw_features |= NETIF_F_GSO_UDP_L4; | |
98e778c9 | 3935 | |
41f2f127 JW |
3936 | dev->features |= NETIF_F_GSO_ROBUST; |
3937 | ||
98e778c9 | 3938 | if (gso) |
e078de03 | 3939 | dev->features |= dev->hw_features & NETIF_F_ALL_TSO; |
98e778c9 | 3940 | /* (!csum && gso) case will be fixed by register_netdev() */ |
296f96fc | 3941 | } |
4f49129b TH |
3942 | if (virtio_has_feature(vdev, VIRTIO_NET_F_GUEST_CSUM)) |
3943 | dev->features |= NETIF_F_RXCSUM; | |
a02e8964 WB |
3944 | if (virtio_has_feature(vdev, VIRTIO_NET_F_GUEST_TSO4) || |
3945 | virtio_has_feature(vdev, VIRTIO_NET_F_GUEST_TSO6)) | |
dbcf24d1 | 3946 | dev->features |= NETIF_F_GRO_HW; |
cf8691cb | 3947 | if (virtio_has_feature(vdev, VIRTIO_NET_F_CTRL_GUEST_OFFLOADS)) |
dbcf24d1 | 3948 | dev->hw_features |= NETIF_F_GRO_HW; |
296f96fc | 3949 | |
4fda8302 | 3950 | dev->vlan_features = dev->features; |
66c0e13a | 3951 | dev->xdp_features = NETDEV_XDP_ACT_BASIC | NETDEV_XDP_ACT_REDIRECT; |
4fda8302 | 3952 | |
d0c2c997 JW |
3953 | /* MTU range: 68 - 65535 */ |
3954 | dev->min_mtu = MIN_MTU; | |
3955 | dev->max_mtu = MAX_MTU; | |
3956 | ||
296f96fc | 3957 | /* Configuration may specify what MAC to use. Otherwise random. */ |
f2edaa4a JK |
3958 | if (virtio_has_feature(vdev, VIRTIO_NET_F_MAC)) { |
3959 | u8 addr[ETH_ALEN]; | |
3960 | ||
855e0c52 RR |
3961 | virtio_cread_bytes(vdev, |
3962 | offsetof(struct virtio_net_config, mac), | |
f2edaa4a JK |
3963 | addr, ETH_ALEN); |
3964 | eth_hw_addr_set(dev, addr); | |
3965 | } else { | |
f2cedb63 | 3966 | eth_hw_addr_random(dev); |
9f62d221 LV |
3967 | dev_info(&vdev->dev, "Assigned random MAC address %pM\n", |
3968 | dev->dev_addr); | |
f2edaa4a | 3969 | } |
296f96fc RR |
3970 | |
3971 | /* Set up our device-specific information */ | |
3972 | vi = netdev_priv(dev); | |
296f96fc RR |
3973 | vi->dev = dev; |
3974 | vi->vdev = vdev; | |
d9d5dcc8 | 3975 | vdev->priv = vi; |
827da44c | 3976 | |
586d17c5 | 3977 | INIT_WORK(&vi->config_work, virtnet_config_changed_work); |
5a159128 | 3978 | spin_lock_init(&vi->refill_lock); |
296f96fc | 3979 | |
30bbf891 | 3980 | if (virtio_has_feature(vdev, VIRTIO_NET_F_MRG_RXBUF)) { |
3f2c31d9 | 3981 | vi->mergeable_rx_bufs = true; |
30bbf891 LB |
3982 | dev->xdp_features |= NETDEV_XDP_ACT_RX_SG; |
3983 | } | |
3f2c31d9 | 3984 | |
699b045a AK |
3985 | if (virtio_has_feature(vi->vdev, VIRTIO_NET_F_NOTF_COAL)) { |
3986 | vi->rx_usecs = 0; | |
3987 | vi->tx_usecs = 0; | |
3988 | vi->tx_max_packets = 0; | |
3989 | vi->rx_max_packets = 0; | |
3990 | } | |
3991 | ||
91f41f01 AM |
3992 | if (virtio_has_feature(vdev, VIRTIO_NET_F_HASH_REPORT)) |
3993 | vi->has_rss_hash_report = true; | |
3994 | ||
3995 | if (virtio_has_feature(vdev, VIRTIO_NET_F_RSS)) | |
c7114b12 | 3996 | vi->has_rss = true; |
91f41f01 AM |
3997 | |
3998 | if (vi->has_rss || vi->has_rss_hash_report) { | |
c7114b12 AM |
3999 | vi->rss_indir_table_size = |
4000 | virtio_cread16(vdev, offsetof(struct virtio_net_config, | |
4001 | rss_max_indirection_table_length)); | |
4002 | vi->rss_key_size = | |
4003 | virtio_cread8(vdev, offsetof(struct virtio_net_config, rss_max_key_size)); | |
4004 | ||
4005 | vi->rss_hash_types_supported = | |
4006 | virtio_cread32(vdev, offsetof(struct virtio_net_config, supported_hash_types)); | |
4007 | vi->rss_hash_types_supported &= | |
4008 | ~(VIRTIO_NET_RSS_HASH_TYPE_IP_EX | | |
4009 | VIRTIO_NET_RSS_HASH_TYPE_TCP_EX | | |
4010 | VIRTIO_NET_RSS_HASH_TYPE_UDP_EX); | |
4011 | ||
4012 | dev->hw_features |= NETIF_F_RXHASH; | |
4013 | } | |
91f41f01 AM |
4014 | |
4015 | if (vi->has_rss_hash_report) | |
4016 | vi->hdr_len = sizeof(struct virtio_net_hdr_v1_hash); | |
4017 | else if (virtio_has_feature(vdev, VIRTIO_NET_F_MRG_RXBUF) || | |
4018 | virtio_has_feature(vdev, VIRTIO_F_VERSION_1)) | |
012873d0 MT |
4019 | vi->hdr_len = sizeof(struct virtio_net_hdr_mrg_rxbuf); |
4020 | else | |
4021 | vi->hdr_len = sizeof(struct virtio_net_hdr); | |
4022 | ||
75993300 MT |
4023 | if (virtio_has_feature(vdev, VIRTIO_F_ANY_LAYOUT) || |
4024 | virtio_has_feature(vdev, VIRTIO_F_VERSION_1)) | |
e7428e95 MT |
4025 | vi->any_header_sg = true; |
4026 | ||
986a4f4d JW |
4027 | if (virtio_has_feature(vdev, VIRTIO_NET_F_CTRL_VQ)) |
4028 | vi->has_cvq = true; | |
4029 | ||
14de9d11 AC |
4030 | if (virtio_has_feature(vdev, VIRTIO_NET_F_MTU)) { |
4031 | mtu = virtio_cread16(vdev, | |
4032 | offsetof(struct virtio_net_config, | |
4033 | mtu)); | |
93a205ee | 4034 | if (mtu < dev->min_mtu) { |
fe36cbe0 MT |
4035 | /* Should never trigger: MTU was previously validated |
4036 | * in virtnet_validate. | |
4037 | */ | |
7934b481 YS |
4038 | dev_err(&vdev->dev, |
4039 | "device MTU appears to have changed it is now %d < %d", | |
4040 | mtu, dev->min_mtu); | |
411ea23a | 4041 | err = -EINVAL; |
d7dfc5cf | 4042 | goto free; |
93a205ee | 4043 | } |
2e123b44 | 4044 | |
fe36cbe0 MT |
4045 | dev->mtu = mtu; |
4046 | dev->max_mtu = mtu; | |
14de9d11 AC |
4047 | } |
4048 | ||
4959aebb GL |
4049 | virtnet_set_big_packets(vi, mtu); |
4050 | ||
012873d0 MT |
4051 | if (vi->any_header_sg) |
4052 | dev->needed_headroom = vi->hdr_len; | |
6ebbc1a6 | 4053 | |
44900010 JW |
4054 | /* Enable multiqueue by default */ |
4055 | if (num_online_cpus() >= max_queue_pairs) | |
4056 | vi->curr_queue_pairs = max_queue_pairs; | |
4057 | else | |
4058 | vi->curr_queue_pairs = num_online_cpus(); | |
986a4f4d JW |
4059 | vi->max_queue_pairs = max_queue_pairs; |
4060 | ||
4061 | /* Allocate/initialize the rx/tx queues, and invoke find_vqs */ | |
3f9c10b0 | 4062 | err = init_vqs(vi); |
d2a7ddda | 4063 | if (err) |
d7dfc5cf | 4064 | goto free; |
296f96fc | 4065 | |
fbf28d78 MD |
4066 | #ifdef CONFIG_SYSFS |
4067 | if (vi->mergeable_rx_bufs) | |
4068 | dev->sysfs_rx_queue_group = &virtio_net_mrg_rx_group; | |
4069 | #endif | |
0f13b66b ZYW |
4070 | netif_set_real_num_tx_queues(dev, vi->curr_queue_pairs); |
4071 | netif_set_real_num_rx_queues(dev, vi->curr_queue_pairs); | |
986a4f4d | 4072 | |
2e9ca760 MT |
4073 | virtnet_init_settings(dev); |
4074 | ||
ba5e4426 SS |
4075 | if (virtio_has_feature(vdev, VIRTIO_NET_F_STANDBY)) { |
4076 | vi->failover = net_failover_create(vi->dev); | |
4b8e6ac4 WY |
4077 | if (IS_ERR(vi->failover)) { |
4078 | err = PTR_ERR(vi->failover); | |
ba5e4426 | 4079 | goto free_vqs; |
4b8e6ac4 | 4080 | } |
ba5e4426 SS |
4081 | } |
4082 | ||
91f41f01 | 4083 | if (vi->has_rss || vi->has_rss_hash_report) |
c7114b12 AM |
4084 | virtnet_init_default_rss(vi); |
4085 | ||
50c0ada6 JW |
4086 | /* serialize netdev register + virtio_device_ready() with ndo_open() */ |
4087 | rtnl_lock(); | |
4088 | ||
4089 | err = register_netdevice(dev); | |
296f96fc RR |
4090 | if (err) { |
4091 | pr_debug("virtio_net: registering device failed\n"); | |
50c0ada6 | 4092 | rtnl_unlock(); |
ba5e4426 | 4093 | goto free_failover; |
296f96fc | 4094 | } |
b3369c1f | 4095 | |
4baf1e33 MT |
4096 | virtio_device_ready(vdev); |
4097 | ||
9f62d221 LV |
4098 | /* a random MAC address has been assigned, notify the device. |
4099 | * We don't fail probe if VIRTIO_NET_F_CTRL_MAC_ADDR is not there | |
4100 | * because many devices work fine without getting MAC explicitly | |
4101 | */ | |
4102 | if (!virtio_has_feature(vdev, VIRTIO_NET_F_MAC) && | |
4103 | virtio_has_feature(vi->vdev, VIRTIO_NET_F_CTRL_MAC_ADDR)) { | |
4104 | struct scatterlist sg; | |
4105 | ||
4106 | sg_init_one(&sg, dev->dev_addr, dev->addr_len); | |
4107 | if (!virtnet_send_command(vi, VIRTIO_NET_CTRL_MAC, | |
4108 | VIRTIO_NET_CTRL_MAC_ADDR_SET, &sg)) { | |
4109 | pr_debug("virtio_net: setting MAC address failed\n"); | |
4110 | rtnl_unlock(); | |
4111 | err = -EINVAL; | |
4112 | goto free_unregister_netdev; | |
4113 | } | |
4114 | } | |
4115 | ||
50c0ada6 JW |
4116 | rtnl_unlock(); |
4117 | ||
8017c279 | 4118 | err = virtnet_cpu_notif_add(vi); |
8de4b2f3 WG |
4119 | if (err) { |
4120 | pr_debug("virtio_net: registering cpu notifier failed\n"); | |
f00e35e2 | 4121 | goto free_unregister_netdev; |
8de4b2f3 WG |
4122 | } |
4123 | ||
a220871b | 4124 | virtnet_set_queues(vi, vi->curr_queue_pairs); |
44900010 | 4125 | |
167c25e4 JW |
4126 | /* Assume link up if device can't report link status, |
4127 | otherwise get link status from config. */ | |
bda7fab5 | 4128 | netif_carrier_off(dev); |
167c25e4 | 4129 | if (virtio_has_feature(vi->vdev, VIRTIO_NET_F_STATUS)) { |
3b07e9ca | 4130 | schedule_work(&vi->config_work); |
167c25e4 JW |
4131 | } else { |
4132 | vi->status = VIRTIO_NET_S_LINK_UP; | |
faa9b39f | 4133 | virtnet_update_settings(vi); |
167c25e4 JW |
4134 | netif_carrier_on(dev); |
4135 | } | |
9f4d26d0 | 4136 | |
3f93522f JW |
4137 | for (i = 0; i < ARRAY_SIZE(guest_offloads); i++) |
4138 | if (virtio_has_feature(vi->vdev, guest_offloads[i])) | |
4139 | set_bit(guest_offloads[i], &vi->guest_offloads); | |
a02e8964 | 4140 | vi->guest_offloads_capable = vi->guest_offloads; |
3f93522f | 4141 | |
986a4f4d JW |
4142 | pr_debug("virtnet: registered device %s with %d RX and TX vq's\n", |
4143 | dev->name, max_queue_pairs); | |
4144 | ||
296f96fc RR |
4145 | return 0; |
4146 | ||
f00e35e2 | 4147 | free_unregister_netdev: |
b3369c1f | 4148 | unregister_netdev(dev); |
ba5e4426 SS |
4149 | free_failover: |
4150 | net_failover_destroy(vi->failover); | |
d2a7ddda | 4151 | free_vqs: |
b0686565 | 4152 | virtio_reset_device(vdev); |
986a4f4d | 4153 | cancel_delayed_work_sync(&vi->refill); |
fb51879d | 4154 | free_receive_page_frags(vi); |
e9d7417b | 4155 | virtnet_del_vqs(vi); |
296f96fc RR |
4156 | free: |
4157 | free_netdev(dev); | |
4158 | return err; | |
4159 | } | |
4160 | ||
04486ed0 | 4161 | static void remove_vq_common(struct virtnet_info *vi) |
296f96fc | 4162 | { |
d9679d00 | 4163 | virtio_reset_device(vi->vdev); |
830a8a97 SM |
4164 | |
4165 | /* Free unused buffers in both send and recv, if any. */ | |
9ab86bbc | 4166 | free_unused_bufs(vi); |
fb6813f4 | 4167 | |
986a4f4d | 4168 | free_receive_bufs(vi); |
d2a7ddda | 4169 | |
fb51879d MD |
4170 | free_receive_page_frags(vi); |
4171 | ||
986a4f4d | 4172 | virtnet_del_vqs(vi); |
04486ed0 AS |
4173 | } |
4174 | ||
8cc085d6 | 4175 | static void virtnet_remove(struct virtio_device *vdev) |
04486ed0 AS |
4176 | { |
4177 | struct virtnet_info *vi = vdev->priv; | |
4178 | ||
8017c279 | 4179 | virtnet_cpu_notif_remove(vi); |
8de4b2f3 | 4180 | |
102a2786 MT |
4181 | /* Make sure no work handler is accessing the device. */ |
4182 | flush_work(&vi->config_work); | |
586d17c5 | 4183 | |
04486ed0 AS |
4184 | unregister_netdev(vi->dev); |
4185 | ||
ba5e4426 SS |
4186 | net_failover_destroy(vi->failover); |
4187 | ||
04486ed0 | 4188 | remove_vq_common(vi); |
fb6813f4 | 4189 | |
74b2553f | 4190 | free_netdev(vi->dev); |
296f96fc RR |
4191 | } |
4192 | ||
67a75194 | 4193 | static __maybe_unused int virtnet_freeze(struct virtio_device *vdev) |
0741bcb5 AS |
4194 | { |
4195 | struct virtnet_info *vi = vdev->priv; | |
4196 | ||
8017c279 | 4197 | virtnet_cpu_notif_remove(vi); |
9fe7bfce | 4198 | virtnet_freeze_down(vdev); |
0741bcb5 AS |
4199 | remove_vq_common(vi); |
4200 | ||
4201 | return 0; | |
4202 | } | |
4203 | ||
67a75194 | 4204 | static __maybe_unused int virtnet_restore(struct virtio_device *vdev) |
0741bcb5 AS |
4205 | { |
4206 | struct virtnet_info *vi = vdev->priv; | |
9fe7bfce | 4207 | int err; |
0741bcb5 | 4208 | |
9fe7bfce | 4209 | err = virtnet_restore_up(vdev); |
0741bcb5 AS |
4210 | if (err) |
4211 | return err; | |
986a4f4d JW |
4212 | virtnet_set_queues(vi, vi->curr_queue_pairs); |
4213 | ||
8017c279 | 4214 | err = virtnet_cpu_notif_add(vi); |
3f2869ca XY |
4215 | if (err) { |
4216 | virtnet_freeze_down(vdev); | |
4217 | remove_vq_common(vi); | |
ec9debbd | 4218 | return err; |
3f2869ca | 4219 | } |
ec9debbd | 4220 | |
0741bcb5 AS |
4221 | return 0; |
4222 | } | |
0741bcb5 | 4223 | |
296f96fc RR |
4224 | static struct virtio_device_id id_table[] = { |
4225 | { VIRTIO_ID_NET, VIRTIO_DEV_ANY_ID }, | |
4226 | { 0 }, | |
4227 | }; | |
4228 | ||
f3358507 MT |
4229 | #define VIRTNET_FEATURES \ |
4230 | VIRTIO_NET_F_CSUM, VIRTIO_NET_F_GUEST_CSUM, \ | |
4231 | VIRTIO_NET_F_MAC, \ | |
4232 | VIRTIO_NET_F_HOST_TSO4, VIRTIO_NET_F_HOST_UFO, VIRTIO_NET_F_HOST_TSO6, \ | |
4233 | VIRTIO_NET_F_HOST_ECN, VIRTIO_NET_F_GUEST_TSO4, VIRTIO_NET_F_GUEST_TSO6, \ | |
4234 | VIRTIO_NET_F_GUEST_ECN, VIRTIO_NET_F_GUEST_UFO, \ | |
418044e1 | 4235 | VIRTIO_NET_F_HOST_USO, VIRTIO_NET_F_GUEST_USO4, VIRTIO_NET_F_GUEST_USO6, \ |
f3358507 MT |
4236 | VIRTIO_NET_F_MRG_RXBUF, VIRTIO_NET_F_STATUS, VIRTIO_NET_F_CTRL_VQ, \ |
4237 | VIRTIO_NET_F_CTRL_RX, VIRTIO_NET_F_CTRL_VLAN, \ | |
4238 | VIRTIO_NET_F_GUEST_ANNOUNCE, VIRTIO_NET_F_MQ, \ | |
4239 | VIRTIO_NET_F_CTRL_MAC_ADDR, \ | |
faa9b39f | 4240 | VIRTIO_NET_F_MTU, VIRTIO_NET_F_CTRL_GUEST_OFFLOADS, \ |
c7114b12 | 4241 | VIRTIO_NET_F_SPEED_DUPLEX, VIRTIO_NET_F_STANDBY, \ |
be50da3e JP |
4242 | VIRTIO_NET_F_RSS, VIRTIO_NET_F_HASH_REPORT, VIRTIO_NET_F_NOTF_COAL, \ |
4243 | VIRTIO_NET_F_GUEST_HDRLEN | |
f3358507 | 4244 | |
c45a6816 | 4245 | static unsigned int features[] = { |
f3358507 MT |
4246 | VIRTNET_FEATURES, |
4247 | }; | |
4248 | ||
4249 | static unsigned int features_legacy[] = { | |
4250 | VIRTNET_FEATURES, | |
4251 | VIRTIO_NET_F_GSO, | |
e7428e95 | 4252 | VIRTIO_F_ANY_LAYOUT, |
c45a6816 RR |
4253 | }; |
4254 | ||
22402529 | 4255 | static struct virtio_driver virtio_net_driver = { |
c45a6816 RR |
4256 | .feature_table = features, |
4257 | .feature_table_size = ARRAY_SIZE(features), | |
f3358507 MT |
4258 | .feature_table_legacy = features_legacy, |
4259 | .feature_table_size_legacy = ARRAY_SIZE(features_legacy), | |
296f96fc RR |
4260 | .driver.name = KBUILD_MODNAME, |
4261 | .driver.owner = THIS_MODULE, | |
4262 | .id_table = id_table, | |
fe36cbe0 | 4263 | .validate = virtnet_validate, |
296f96fc | 4264 | .probe = virtnet_probe, |
8cc085d6 | 4265 | .remove = virtnet_remove, |
9f4d26d0 | 4266 | .config_changed = virtnet_config_changed, |
89107000 | 4267 | #ifdef CONFIG_PM_SLEEP |
0741bcb5 AS |
4268 | .freeze = virtnet_freeze, |
4269 | .restore = virtnet_restore, | |
4270 | #endif | |
296f96fc RR |
4271 | }; |
4272 | ||
8017c279 SAS |
4273 | static __init int virtio_net_driver_init(void) |
4274 | { | |
4275 | int ret; | |
4276 | ||
73c1b41e | 4277 | ret = cpuhp_setup_state_multi(CPUHP_AP_ONLINE_DYN, "virtio/net:online", |
8017c279 SAS |
4278 | virtnet_cpu_online, |
4279 | virtnet_cpu_down_prep); | |
4280 | if (ret < 0) | |
4281 | goto out; | |
4282 | virtionet_online = ret; | |
73c1b41e | 4283 | ret = cpuhp_setup_state_multi(CPUHP_VIRT_NET_DEAD, "virtio/net:dead", |
8017c279 SAS |
4284 | NULL, virtnet_cpu_dead); |
4285 | if (ret) | |
4286 | goto err_dead; | |
4f50ef15 | 4287 | ret = register_virtio_driver(&virtio_net_driver); |
8017c279 SAS |
4288 | if (ret) |
4289 | goto err_virtio; | |
4290 | return 0; | |
4291 | err_virtio: | |
4292 | cpuhp_remove_multi_state(CPUHP_VIRT_NET_DEAD); | |
4293 | err_dead: | |
4294 | cpuhp_remove_multi_state(virtionet_online); | |
4295 | out: | |
4296 | return ret; | |
4297 | } | |
4298 | module_init(virtio_net_driver_init); | |
4299 | ||
4300 | static __exit void virtio_net_driver_exit(void) | |
4301 | { | |
cfa0ebc9 | 4302 | unregister_virtio_driver(&virtio_net_driver); |
8017c279 SAS |
4303 | cpuhp_remove_multi_state(CPUHP_VIRT_NET_DEAD); |
4304 | cpuhp_remove_multi_state(virtionet_online); | |
8017c279 SAS |
4305 | } |
4306 | module_exit(virtio_net_driver_exit); | |
296f96fc RR |
4307 | |
4308 | MODULE_DEVICE_TABLE(virtio, id_table); | |
4309 | MODULE_DESCRIPTION("Virtio network driver"); | |
4310 | MODULE_LICENSE("GPL"); |