]>
Commit | Line | Data |
---|---|---|
fceaf24a | 1 | /* |
fceaf24a HJ |
2 | * Copyright (c) 2009, Microsoft Corporation. |
3 | * | |
4 | * This program is free software; you can redistribute it and/or modify it | |
5 | * under the terms and conditions of the GNU General Public License, | |
6 | * version 2, as published by the Free Software Foundation. | |
7 | * | |
8 | * This program is distributed in the hope it will be useful, but WITHOUT | |
9 | * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or | |
10 | * FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License for | |
11 | * more details. | |
12 | * | |
13 | * You should have received a copy of the GNU General Public License along with | |
adf8d3ff | 14 | * this program; if not, see <http://www.gnu.org/licenses/>. |
fceaf24a HJ |
15 | * |
16 | * Authors: | |
d0e94d17 | 17 | * Haiyang Zhang <haiyangz@microsoft.com> |
fceaf24a | 18 | * Hank Janssen <hjanssen@microsoft.com> |
fceaf24a | 19 | */ |
eb335bc4 HJ |
20 | #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt |
21 | ||
fceaf24a | 22 | #include <linux/init.h> |
9079ce69 | 23 | #include <linux/atomic.h> |
fceaf24a HJ |
24 | #include <linux/module.h> |
25 | #include <linux/highmem.h> | |
26 | #include <linux/device.h> | |
fceaf24a | 27 | #include <linux/io.h> |
fceaf24a HJ |
28 | #include <linux/delay.h> |
29 | #include <linux/netdevice.h> | |
30 | #include <linux/inetdevice.h> | |
31 | #include <linux/etherdevice.h> | |
b93c1b5a | 32 | #include <linux/pci.h> |
fceaf24a | 33 | #include <linux/skbuff.h> |
c802db11 | 34 | #include <linux/if_vlan.h> |
fceaf24a | 35 | #include <linux/in.h> |
5a0e3ad6 | 36 | #include <linux/slab.h> |
27f5aa92 | 37 | #include <linux/rtnetlink.h> |
0c195567 | 38 | #include <linux/netpoll.h> |
27f5aa92 | 39 | |
fceaf24a HJ |
40 | #include <net/arp.h> |
41 | #include <net/route.h> | |
42 | #include <net/sock.h> | |
43 | #include <net/pkt_sched.h> | |
8eb1b3c3 MK |
44 | #include <net/checksum.h> |
45 | #include <net/ip6_checksum.h> | |
3f335ea2 | 46 | |
5ca7252a | 47 | #include "hyperv_net.h" |
fceaf24a | 48 | |
7b2ee50c SH |
49 | #define RING_SIZE_MIN 64 |
50 | #define RETRY_US_LO 5000 | |
51 | #define RETRY_US_HI 10000 | |
52 | #define RETRY_MAX 2000 /* >10 sec */ | |
8b532797 | 53 | |
27a70af3 | 54 | #define LINKCHANGE_INT (2 * HZ) |
6123c668 | 55 | #define VF_TAKEOVER_INT (HZ / 10) |
a50af86d | 56 | |
a7f99d0f | 57 | static unsigned int ring_size __ro_after_init = 128; |
d61e4038 | 58 | module_param(ring_size, uint, 0444); |
450d7a4b | 59 | MODULE_PARM_DESC(ring_size, "Ring buffer size (# of pages)"); |
a7f99d0f | 60 | unsigned int netvsc_ring_bytes __ro_after_init; |
fceaf24a | 61 | |
3f300ff4 SX |
62 | static const u32 default_msg = NETIF_MSG_DRV | NETIF_MSG_PROBE | |
63 | NETIF_MSG_LINK | NETIF_MSG_IFUP | | |
64 | NETIF_MSG_IFDOWN | NETIF_MSG_RX_ERR | | |
65 | NETIF_MSG_TX_ERR; | |
66 | ||
67 | static int debug = -1; | |
d61e4038 | 68 | module_param(debug, int, 0444); |
3f300ff4 SX |
69 | MODULE_PARM_DESC(debug, "Debug level (0=none,...,16=all)"); |
70 | ||
7bf7bb37 SH |
71 | static LIST_HEAD(netvsc_dev_list); |
72 | ||
bee9d41b | 73 | static void netvsc_change_rx_flags(struct net_device *net, int change) |
fceaf24a | 74 | { |
bee9d41b SH |
75 | struct net_device_context *ndev_ctx = netdev_priv(net); |
76 | struct net_device *vf_netdev = rtnl_dereference(ndev_ctx->vf_netdev); | |
77 | int inc; | |
78 | ||
79 | if (!vf_netdev) | |
80 | return; | |
81 | ||
82 | if (change & IFF_PROMISC) { | |
83 | inc = (net->flags & IFF_PROMISC) ? 1 : -1; | |
84 | dev_set_promiscuity(vf_netdev, inc); | |
85 | } | |
86 | ||
87 | if (change & IFF_ALLMULTI) { | |
88 | inc = (net->flags & IFF_ALLMULTI) ? 1 : -1; | |
89 | dev_set_allmulti(vf_netdev, inc); | |
90 | } | |
91 | } | |
92 | ||
93 | static void netvsc_set_rx_mode(struct net_device *net) | |
94 | { | |
95 | struct net_device_context *ndev_ctx = netdev_priv(net); | |
35a57b7f SH |
96 | struct net_device *vf_netdev; |
97 | struct netvsc_device *nvdev; | |
bee9d41b | 98 | |
35a57b7f SH |
99 | rcu_read_lock(); |
100 | vf_netdev = rcu_dereference(ndev_ctx->vf_netdev); | |
bee9d41b SH |
101 | if (vf_netdev) { |
102 | dev_uc_sync(vf_netdev, net); | |
103 | dev_mc_sync(vf_netdev, net); | |
104 | } | |
d426b2e3 | 105 | |
35a57b7f SH |
106 | nvdev = rcu_dereference(ndev_ctx->nvdev); |
107 | if (nvdev) | |
108 | rndis_filter_update(nvdev); | |
109 | rcu_read_unlock(); | |
fceaf24a HJ |
110 | } |
111 | ||
fceaf24a HJ |
112 | static int netvsc_open(struct net_device *net) |
113 | { | |
53fa1a6f | 114 | struct net_device_context *ndev_ctx = netdev_priv(net); |
0c195567 | 115 | struct net_device *vf_netdev = rtnl_dereference(ndev_ctx->vf_netdev); |
79e8cbe7 | 116 | struct netvsc_device *nvdev = rtnl_dereference(ndev_ctx->nvdev); |
891de74d | 117 | struct rndis_device *rdev; |
02fafbc6 | 118 | int ret = 0; |
fceaf24a | 119 | |
891de74d HZ |
120 | netif_carrier_off(net); |
121 | ||
d515d0ff | 122 | /* Open up the device */ |
2f5fa6c8 | 123 | ret = rndis_filter_open(nvdev); |
d515d0ff HZ |
124 | if (ret != 0) { |
125 | netdev_err(net, "unable to open device (ret %d).\n", ret); | |
126 | return ret; | |
fceaf24a HJ |
127 | } |
128 | ||
891de74d | 129 | rdev = nvdev->extension; |
52acf73b | 130 | if (!rdev->link_state) { |
891de74d | 131 | netif_carrier_on(net); |
52acf73b DC |
132 | netif_tx_wake_all_queues(net); |
133 | } | |
891de74d | 134 | |
0c195567 | 135 | if (vf_netdev) { |
136 | /* Setting synthetic device up transparently sets | |
137 | * slave as up. If open fails, then slave will be | |
138 | * still be offline (and not used). | |
139 | */ | |
00f54e68 | 140 | ret = dev_open(vf_netdev, NULL); |
0c195567 | 141 | if (ret) |
142 | netdev_warn(net, | |
143 | "unable to open slave: %s: %d\n", | |
144 | vf_netdev->name, ret); | |
145 | } | |
146 | return 0; | |
fceaf24a HJ |
147 | } |
148 | ||
7b2ee50c | 149 | static int netvsc_wait_until_empty(struct netvsc_device *nvdev) |
fceaf24a | 150 | { |
7b2ee50c SH |
151 | unsigned int retry = 0; |
152 | int i; | |
2de8530b HZ |
153 | |
154 | /* Ensure pending bytes in ring are read */ | |
7b2ee50c SH |
155 | for (;;) { |
156 | u32 aread = 0; | |
157 | ||
2de8530b | 158 | for (i = 0; i < nvdev->num_chn; i++) { |
7b2ee50c SH |
159 | struct vmbus_channel *chn |
160 | = nvdev->chan_table[i].channel; | |
161 | ||
2de8530b HZ |
162 | if (!chn) |
163 | continue; | |
164 | ||
7b2ee50c SH |
165 | /* make sure receive not running now */ |
166 | napi_synchronize(&nvdev->chan_table[i].napi); | |
167 | ||
40975962 | 168 | aread = hv_get_bytes_to_read(&chn->inbound); |
2de8530b HZ |
169 | if (aread) |
170 | break; | |
171 | ||
40975962 | 172 | aread = hv_get_bytes_to_read(&chn->outbound); |
2de8530b HZ |
173 | if (aread) |
174 | break; | |
175 | } | |
176 | ||
7b2ee50c SH |
177 | if (aread == 0) |
178 | return 0; | |
2de8530b | 179 | |
7b2ee50c SH |
180 | if (++retry > RETRY_MAX) |
181 | return -ETIMEDOUT; | |
2de8530b | 182 | |
7b2ee50c | 183 | usleep_range(RETRY_US_LO, RETRY_US_HI); |
2de8530b | 184 | } |
7b2ee50c | 185 | } |
2de8530b | 186 | |
7b2ee50c SH |
187 | static int netvsc_close(struct net_device *net) |
188 | { | |
189 | struct net_device_context *net_device_ctx = netdev_priv(net); | |
190 | struct net_device *vf_netdev | |
191 | = rtnl_dereference(net_device_ctx->vf_netdev); | |
192 | struct netvsc_device *nvdev = rtnl_dereference(net_device_ctx->nvdev); | |
193 | int ret; | |
194 | ||
195 | netif_tx_disable(net); | |
196 | ||
197 | /* No need to close rndis filter if it is removed already */ | |
198 | if (!nvdev) | |
199 | return 0; | |
200 | ||
201 | ret = rndis_filter_close(nvdev); | |
202 | if (ret != 0) { | |
203 | netdev_err(net, "unable to close device (ret %d).\n", ret); | |
204 | return ret; | |
2de8530b | 205 | } |
fceaf24a | 206 | |
7b2ee50c SH |
207 | ret = netvsc_wait_until_empty(nvdev); |
208 | if (ret) | |
209 | netdev_err(net, "Ring buffer not empty after closing rndis\n"); | |
210 | ||
0c195567 | 211 | if (vf_netdev) |
212 | dev_close(vf_netdev); | |
213 | ||
fceaf24a HJ |
214 | return ret; |
215 | } | |
216 | ||
f5a22550 SH |
217 | static inline void *init_ppi_data(struct rndis_message *msg, |
218 | u32 ppi_size, u32 pkt_type) | |
8a00251a | 219 | { |
f5a22550 | 220 | struct rndis_packet *rndis_pkt = &msg->msg.pkt; |
8a00251a KS |
221 | struct rndis_per_packet_info *ppi; |
222 | ||
8a00251a | 223 | rndis_pkt->data_offset += ppi_size; |
f5a22550 SH |
224 | ppi = (void *)rndis_pkt + rndis_pkt->per_pkt_info_offset |
225 | + rndis_pkt->per_pkt_info_len; | |
8a00251a KS |
226 | |
227 | ppi->size = ppi_size; | |
228 | ppi->type = pkt_type; | |
e3a9667a | 229 | ppi->internal = 0; |
8a00251a KS |
230 | ppi->ppi_offset = sizeof(struct rndis_per_packet_info); |
231 | ||
232 | rndis_pkt->per_pkt_info_len += ppi_size; | |
233 | ||
f5a22550 | 234 | return ppi + 1; |
8a00251a KS |
235 | } |
236 | ||
4823eb2f HZ |
237 | /* Azure hosts don't support non-TCP port numbers in hashing for fragmented |
238 | * packets. We can use ethtool to change UDP hash level when necessary. | |
f72860af | 239 | */ |
4823eb2f HZ |
240 | static inline u32 netvsc_get_hash( |
241 | struct sk_buff *skb, | |
242 | const struct net_device_context *ndc) | |
f72860af HZ |
243 | { |
244 | struct flow_keys flow; | |
486e3981 | 245 | u32 hash, pkt_proto = 0; |
f72860af HZ |
246 | static u32 hashrnd __read_mostly; |
247 | ||
248 | net_get_random_once(&hashrnd, sizeof(hashrnd)); | |
249 | ||
250 | if (!skb_flow_dissect_flow_keys(skb, &flow, 0)) | |
251 | return 0; | |
252 | ||
486e3981 HZ |
253 | switch (flow.basic.ip_proto) { |
254 | case IPPROTO_TCP: | |
255 | if (flow.basic.n_proto == htons(ETH_P_IP)) | |
256 | pkt_proto = HV_TCP4_L4HASH; | |
257 | else if (flow.basic.n_proto == htons(ETH_P_IPV6)) | |
258 | pkt_proto = HV_TCP6_L4HASH; | |
259 | ||
260 | break; | |
261 | ||
262 | case IPPROTO_UDP: | |
263 | if (flow.basic.n_proto == htons(ETH_P_IP)) | |
264 | pkt_proto = HV_UDP4_L4HASH; | |
265 | else if (flow.basic.n_proto == htons(ETH_P_IPV6)) | |
266 | pkt_proto = HV_UDP6_L4HASH; | |
267 | ||
268 | break; | |
269 | } | |
270 | ||
271 | if (pkt_proto & ndc->l4_hash) { | |
f72860af HZ |
272 | return skb_get_hash(skb); |
273 | } else { | |
274 | if (flow.basic.n_proto == htons(ETH_P_IP)) | |
275 | hash = jhash2((u32 *)&flow.addrs.v4addrs, 2, hashrnd); | |
276 | else if (flow.basic.n_proto == htons(ETH_P_IPV6)) | |
277 | hash = jhash2((u32 *)&flow.addrs.v6addrs, 8, hashrnd); | |
278 | else | |
279 | hash = 0; | |
280 | ||
281 | skb_set_hash(skb, hash, PKT_HASH_TYPE_L3); | |
282 | } | |
283 | ||
284 | return hash; | |
285 | } | |
286 | ||
8db91f6a HZ |
287 | static inline int netvsc_get_tx_queue(struct net_device *ndev, |
288 | struct sk_buff *skb, int old_idx) | |
289 | { | |
290 | const struct net_device_context *ndc = netdev_priv(ndev); | |
291 | struct sock *sk = skb->sk; | |
292 | int q_idx; | |
293 | ||
39e91cfb HZ |
294 | q_idx = ndc->tx_table[netvsc_get_hash(skb, ndc) & |
295 | (VRSS_SEND_TAB_SIZE - 1)]; | |
8db91f6a HZ |
296 | |
297 | /* If queue index changed record the new value */ | |
298 | if (q_idx != old_idx && | |
299 | sk && sk_fullsock(sk) && rcu_access_pointer(sk->sk_dst_cache)) | |
300 | sk_tx_queue_set(sk, q_idx); | |
301 | ||
302 | return q_idx; | |
303 | } | |
304 | ||
d8e18ee0 | 305 | /* |
306 | * Select queue for transmit. | |
307 | * | |
308 | * If a valid queue has already been assigned, then use that. | |
309 | * Otherwise compute tx queue based on hash and the send table. | |
310 | * | |
311 | * This is basically similar to default (__netdev_pick_tx) with the added step | |
312 | * of using the host send_table when no other queue has been assigned. | |
313 | * | |
314 | * TODO support XPS - but get_xps_queue not exported | |
315 | */ | |
0c195567 | 316 | static u16 netvsc_pick_tx(struct net_device *ndev, struct sk_buff *skb) |
5b54dac8 | 317 | { |
8db91f6a HZ |
318 | int q_idx = sk_tx_queue_get(skb->sk); |
319 | ||
0c195567 | 320 | if (q_idx < 0 || skb->ooo_okay || q_idx >= ndev->real_num_tx_queues) { |
8db91f6a HZ |
321 | /* If forwarding a packet, we use the recorded queue when |
322 | * available for better cache locality. | |
323 | */ | |
324 | if (skb_rx_queue_recorded(skb)) | |
325 | q_idx = skb_get_rx_queue(skb); | |
326 | else | |
327 | q_idx = netvsc_get_tx_queue(ndev, skb, q_idx); | |
d8e18ee0 | 328 | } |
5b54dac8 HZ |
329 | |
330 | return q_idx; | |
331 | } | |
332 | ||
0c195567 | 333 | static u16 netvsc_select_queue(struct net_device *ndev, struct sk_buff *skb, |
4f49dec9 | 334 | struct net_device *sb_dev, |
0c195567 | 335 | select_queue_fallback_t fallback) |
336 | { | |
337 | struct net_device_context *ndc = netdev_priv(ndev); | |
338 | struct net_device *vf_netdev; | |
339 | u16 txq; | |
340 | ||
341 | rcu_read_lock(); | |
342 | vf_netdev = rcu_dereference(ndc->vf_netdev); | |
343 | if (vf_netdev) { | |
b3bf5666 SH |
344 | const struct net_device_ops *vf_ops = vf_netdev->netdev_ops; |
345 | ||
346 | if (vf_ops->ndo_select_queue) | |
347 | txq = vf_ops->ndo_select_queue(vf_netdev, skb, | |
4f49dec9 | 348 | sb_dev, fallback); |
b3bf5666 | 349 | else |
8ec56fc3 | 350 | txq = fallback(vf_netdev, skb, NULL); |
b3bf5666 SH |
351 | |
352 | /* Record the queue selected by VF so that it can be | |
353 | * used for common case where VF has more queues than | |
354 | * the synthetic device. | |
355 | */ | |
356 | qdisc_skb_cb(skb)->slave_dev_queue_mapping = txq; | |
0c195567 | 357 | } else { |
358 | txq = netvsc_pick_tx(ndev, skb); | |
359 | } | |
360 | rcu_read_unlock(); | |
361 | ||
362 | while (unlikely(txq >= ndev->real_num_tx_queues)) | |
363 | txq -= ndev->real_num_tx_queues; | |
364 | ||
365 | return txq; | |
366 | } | |
367 | ||
54a7357f | 368 | static u32 fill_pg_buf(struct page *page, u32 offset, u32 len, |
89bb42b1 | 369 | struct hv_page_buffer *pb) |
54a7357f KS |
370 | { |
371 | int j = 0; | |
372 | ||
52d3b494 | 373 | /* Deal with compound pages by ignoring unused part |
54a7357f KS |
374 | * of the page. |
375 | */ | |
376 | page += (offset >> PAGE_SHIFT); | |
377 | offset &= ~PAGE_MASK; | |
378 | ||
379 | while (len > 0) { | |
380 | unsigned long bytes; | |
381 | ||
382 | bytes = PAGE_SIZE - offset; | |
383 | if (bytes > len) | |
384 | bytes = len; | |
385 | pb[j].pfn = page_to_pfn(page); | |
386 | pb[j].offset = offset; | |
387 | pb[j].len = bytes; | |
388 | ||
389 | offset += bytes; | |
390 | len -= bytes; | |
391 | ||
392 | if (offset == PAGE_SIZE && len) { | |
393 | page++; | |
394 | offset = 0; | |
395 | j++; | |
396 | } | |
397 | } | |
398 | ||
399 | return j + 1; | |
400 | } | |
401 | ||
8a00251a | 402 | static u32 init_page_array(void *hdr, u32 len, struct sk_buff *skb, |
a9f2e2d6 | 403 | struct hv_netvsc_packet *packet, |
02b6de01 | 404 | struct hv_page_buffer *pb) |
54a7357f KS |
405 | { |
406 | u32 slots_used = 0; | |
407 | char *data = skb->data; | |
408 | int frags = skb_shinfo(skb)->nr_frags; | |
409 | int i; | |
410 | ||
411 | /* The packet is laid out thus: | |
aa0a34be | 412 | * 1. hdr: RNDIS header and PPI |
54a7357f KS |
413 | * 2. skb linear data |
414 | * 3. skb fragment data | |
415 | */ | |
ea5a32c0 | 416 | slots_used += fill_pg_buf(virt_to_page(hdr), |
417 | offset_in_page(hdr), | |
418 | len, &pb[slots_used]); | |
54a7357f | 419 | |
aa0a34be HZ |
420 | packet->rmsg_size = len; |
421 | packet->rmsg_pgcnt = slots_used; | |
422 | ||
54a7357f KS |
423 | slots_used += fill_pg_buf(virt_to_page(data), |
424 | offset_in_page(data), | |
425 | skb_headlen(skb), &pb[slots_used]); | |
426 | ||
427 | for (i = 0; i < frags; i++) { | |
428 | skb_frag_t *frag = skb_shinfo(skb)->frags + i; | |
429 | ||
430 | slots_used += fill_pg_buf(skb_frag_page(frag), | |
431 | frag->page_offset, | |
432 | skb_frag_size(frag), &pb[slots_used]); | |
433 | } | |
8a00251a | 434 | return slots_used; |
54a7357f KS |
435 | } |
436 | ||
80d887db | 437 | static int count_skb_frag_slots(struct sk_buff *skb) |
438 | { | |
439 | int i, frags = skb_shinfo(skb)->nr_frags; | |
440 | int pages = 0; | |
441 | ||
442 | for (i = 0; i < frags; i++) { | |
443 | skb_frag_t *frag = skb_shinfo(skb)->frags + i; | |
444 | unsigned long size = skb_frag_size(frag); | |
445 | unsigned long offset = frag->page_offset; | |
446 | ||
447 | /* Skip unused frames from start of page */ | |
448 | offset &= ~PAGE_MASK; | |
449 | pages += PFN_UP(offset + size); | |
450 | } | |
451 | return pages; | |
452 | } | |
453 | ||
454 | static int netvsc_get_slots(struct sk_buff *skb) | |
54a7357f | 455 | { |
80d887db | 456 | char *data = skb->data; |
457 | unsigned int offset = offset_in_page(data); | |
458 | unsigned int len = skb_headlen(skb); | |
459 | int slots; | |
460 | int frag_slots; | |
461 | ||
462 | slots = DIV_ROUND_UP(offset + len, PAGE_SIZE); | |
463 | frag_slots = count_skb_frag_slots(skb); | |
464 | return slots + frag_slots; | |
54a7357f KS |
465 | } |
466 | ||
23312a3b | 467 | static u32 net_checksum_info(struct sk_buff *skb) |
08cd04bf | 468 | { |
23312a3b | 469 | if (skb->protocol == htons(ETH_P_IP)) { |
470 | struct iphdr *ip = ip_hdr(skb); | |
08cd04bf | 471 | |
23312a3b | 472 | if (ip->protocol == IPPROTO_TCP) |
473 | return TRANSPORT_INFO_IPV4_TCP; | |
474 | else if (ip->protocol == IPPROTO_UDP) | |
475 | return TRANSPORT_INFO_IPV4_UDP; | |
08cd04bf | 476 | } else { |
23312a3b | 477 | struct ipv6hdr *ip6 = ipv6_hdr(skb); |
478 | ||
479 | if (ip6->nexthdr == IPPROTO_TCP) | |
480 | return TRANSPORT_INFO_IPV6_TCP; | |
37b9dfa0 | 481 | else if (ip6->nexthdr == IPPROTO_UDP) |
23312a3b | 482 | return TRANSPORT_INFO_IPV6_UDP; |
08cd04bf KS |
483 | } |
484 | ||
23312a3b | 485 | return TRANSPORT_INFO_NOT_IP; |
08cd04bf KS |
486 | } |
487 | ||
0c195567 | 488 | /* Send skb on the slave VF device. */ |
489 | static int netvsc_vf_xmit(struct net_device *net, struct net_device *vf_netdev, | |
490 | struct sk_buff *skb) | |
491 | { | |
492 | struct net_device_context *ndev_ctx = netdev_priv(net); | |
493 | unsigned int len = skb->len; | |
494 | int rc; | |
495 | ||
496 | skb->dev = vf_netdev; | |
497 | skb->queue_mapping = qdisc_skb_cb(skb)->slave_dev_queue_mapping; | |
498 | ||
499 | rc = dev_queue_xmit(skb); | |
500 | if (likely(rc == NET_XMIT_SUCCESS || rc == NET_XMIT_CN)) { | |
501 | struct netvsc_vf_pcpu_stats *pcpu_stats | |
502 | = this_cpu_ptr(ndev_ctx->vf_stats); | |
503 | ||
504 | u64_stats_update_begin(&pcpu_stats->syncp); | |
505 | pcpu_stats->tx_packets++; | |
506 | pcpu_stats->tx_bytes += len; | |
507 | u64_stats_update_end(&pcpu_stats->syncp); | |
508 | } else { | |
509 | this_cpu_inc(ndev_ctx->vf_stats->tx_dropped); | |
510 | } | |
511 | ||
512 | return rc; | |
513 | } | |
514 | ||
02fafbc6 | 515 | static int netvsc_start_xmit(struct sk_buff *skb, struct net_device *net) |
fceaf24a | 516 | { |
fceaf24a | 517 | struct net_device_context *net_device_ctx = netdev_priv(net); |
981a1bd8 | 518 | struct hv_netvsc_packet *packet = NULL; |
02fafbc6 | 519 | int ret; |
8a00251a KS |
520 | unsigned int num_data_pgs; |
521 | struct rndis_message *rndis_msg; | |
0c195567 | 522 | struct net_device *vf_netdev; |
8a00251a | 523 | u32 rndis_msg_size; |
307f0995 | 524 | u32 hash; |
02b6de01 | 525 | struct hv_page_buffer pb[MAX_PAGE_BUFFER_COUNT]; |
fceaf24a | 526 | |
0c195567 | 527 | /* if VF is present and up then redirect packets |
528 | * already called with rcu_read_lock_bh | |
529 | */ | |
530 | vf_netdev = rcu_dereference_bh(net_device_ctx->vf_netdev); | |
531 | if (vf_netdev && netif_running(vf_netdev) && | |
532 | !netpoll_tx_running(net)) | |
533 | return netvsc_vf_xmit(net, vf_netdev, skb); | |
534 | ||
80d887db | 535 | /* We will atmost need two pages to describe the rndis |
536 | * header. We can only transmit MAX_PAGE_BUFFER_COUNT number | |
e88f7e07 VK |
537 | * of pages in a single packet. If skb is scattered around |
538 | * more pages we try linearizing it. | |
54a7357f | 539 | */ |
80d887db | 540 | |
541 | num_data_pgs = netvsc_get_slots(skb) + 2; | |
542 | ||
0ab05141 | 543 | if (unlikely(num_data_pgs > MAX_PAGE_BUFFER_COUNT)) { |
4323b47c SH |
544 | ++net_device_ctx->eth_stats.tx_scattered; |
545 | ||
546 | if (skb_linearize(skb)) | |
547 | goto no_memory; | |
0ab05141 | 548 | |
80d887db | 549 | num_data_pgs = netvsc_get_slots(skb) + 2; |
0ab05141 | 550 | if (num_data_pgs > MAX_PAGE_BUFFER_COUNT) { |
4323b47c | 551 | ++net_device_ctx->eth_stats.tx_too_big; |
0ab05141 SH |
552 | goto drop; |
553 | } | |
54a7357f | 554 | } |
fceaf24a | 555 | |
c0eb4540 KS |
556 | /* |
557 | * Place the rndis header in the skb head room and | |
558 | * the skb->cb will be used for hv_netvsc_packet | |
559 | * structure. | |
560 | */ | |
561 | ret = skb_cow_head(skb, RNDIS_AND_PPI_SIZE); | |
4323b47c SH |
562 | if (ret) |
563 | goto no_memory; | |
564 | ||
c0eb4540 KS |
565 | /* Use the skb control buffer for building up the packet */ |
566 | BUILD_BUG_ON(sizeof(struct hv_netvsc_packet) > | |
567 | FIELD_SIZEOF(struct sk_buff, cb)); | |
568 | packet = (struct hv_netvsc_packet *)skb->cb; | |
fceaf24a | 569 | |
5b54dac8 HZ |
570 | packet->q_idx = skb_get_queue_mapping(skb); |
571 | ||
4d447c9a | 572 | packet->total_data_buflen = skb->len; |
793e3955 | 573 | packet->total_bytes = skb->len; |
574 | packet->total_packets = 1; | |
fceaf24a | 575 | |
c0eb4540 | 576 | rndis_msg = (struct rndis_message *)skb->head; |
b08cc791 | 577 | |
8a00251a | 578 | /* Add the rndis header */ |
8a00251a KS |
579 | rndis_msg->ndis_msg_type = RNDIS_MSG_PACKET; |
580 | rndis_msg->msg_len = packet->total_data_buflen; | |
f5a22550 SH |
581 | |
582 | rndis_msg->msg.pkt = (struct rndis_packet) { | |
583 | .data_offset = sizeof(struct rndis_packet), | |
584 | .data_len = packet->total_data_buflen, | |
585 | .per_pkt_info_offset = sizeof(struct rndis_packet), | |
586 | }; | |
8a00251a KS |
587 | |
588 | rndis_msg_size = RNDIS_MESSAGE_SIZE(struct rndis_packet); | |
589 | ||
307f0995 HZ |
590 | hash = skb_get_hash_raw(skb); |
591 | if (hash != 0 && net->real_num_tx_queues > 1) { | |
f5a22550 SH |
592 | u32 *hash_info; |
593 | ||
307f0995 | 594 | rndis_msg_size += NDIS_HASH_PPI_SIZE; |
f5a22550 SH |
595 | hash_info = init_ppi_data(rndis_msg, NDIS_HASH_PPI_SIZE, |
596 | NBL_HASH_VALUE); | |
597 | *hash_info = hash; | |
307f0995 HZ |
598 | } |
599 | ||
0ab05141 | 600 | if (skb_vlan_tag_present(skb)) { |
8a00251a KS |
601 | struct ndis_pkt_8021q_info *vlan; |
602 | ||
603 | rndis_msg_size += NDIS_VLAN_PPI_SIZE; | |
f5a22550 SH |
604 | vlan = init_ppi_data(rndis_msg, NDIS_VLAN_PPI_SIZE, |
605 | IEEE_8021Q_INFO); | |
00f5024e | 606 | |
f5a22550 | 607 | vlan->value = 0; |
98ba780e MM |
608 | vlan->vlanid = skb_vlan_tag_get_id(skb); |
609 | vlan->cfi = skb_vlan_tag_get_cfi(skb); | |
610 | vlan->pri = skb_vlan_tag_get_prio(skb); | |
8a00251a KS |
611 | } |
612 | ||
23312a3b | 613 | if (skb_is_gso(skb)) { |
0ab05141 SH |
614 | struct ndis_tcp_lso_info *lso_info; |
615 | ||
616 | rndis_msg_size += NDIS_LSO_PPI_SIZE; | |
f5a22550 SH |
617 | lso_info = init_ppi_data(rndis_msg, NDIS_LSO_PPI_SIZE, |
618 | TCP_LARGESEND_PKTINFO); | |
0ab05141 | 619 | |
f5a22550 | 620 | lso_info->value = 0; |
0ab05141 | 621 | lso_info->lso_v2_transmit.type = NDIS_TCP_LARGE_SEND_OFFLOAD_V2_TYPE; |
23312a3b | 622 | if (skb->protocol == htons(ETH_P_IP)) { |
0ab05141 SH |
623 | lso_info->lso_v2_transmit.ip_version = |
624 | NDIS_TCP_LARGE_SEND_OFFLOAD_IPV4; | |
625 | ip_hdr(skb)->tot_len = 0; | |
626 | ip_hdr(skb)->check = 0; | |
627 | tcp_hdr(skb)->check = | |
628 | ~csum_tcpudp_magic(ip_hdr(skb)->saddr, | |
629 | ip_hdr(skb)->daddr, 0, IPPROTO_TCP, 0); | |
630 | } else { | |
631 | lso_info->lso_v2_transmit.ip_version = | |
632 | NDIS_TCP_LARGE_SEND_OFFLOAD_IPV6; | |
633 | ipv6_hdr(skb)->payload_len = 0; | |
634 | tcp_hdr(skb)->check = | |
635 | ~csum_ipv6_magic(&ipv6_hdr(skb)->saddr, | |
636 | &ipv6_hdr(skb)->daddr, 0, IPPROTO_TCP, 0); | |
637 | } | |
23312a3b | 638 | lso_info->lso_v2_transmit.tcp_header_offset = skb_transport_offset(skb); |
0ab05141 | 639 | lso_info->lso_v2_transmit.mss = skb_shinfo(skb)->gso_size; |
ad19bc8a | 640 | } else if (skb->ip_summed == CHECKSUM_PARTIAL) { |
23312a3b | 641 | if (net_checksum_info(skb) & net_device_ctx->tx_checksum_mask) { |
642 | struct ndis_tcp_ip_checksum_info *csum_info; | |
643 | ||
ad19bc8a | 644 | rndis_msg_size += NDIS_CSUM_PPI_SIZE; |
f5a22550 SH |
645 | csum_info = init_ppi_data(rndis_msg, NDIS_CSUM_PPI_SIZE, |
646 | TCPIP_CHKSUM_PKTINFO); | |
ad19bc8a | 647 | |
f5a22550 | 648 | csum_info->value = 0; |
23312a3b | 649 | csum_info->transmit.tcp_header_offset = skb_transport_offset(skb); |
650 | ||
651 | if (skb->protocol == htons(ETH_P_IP)) { | |
ad19bc8a | 652 | csum_info->transmit.is_ipv4 = 1; |
23312a3b | 653 | |
654 | if (ip_hdr(skb)->protocol == IPPROTO_TCP) | |
655 | csum_info->transmit.tcp_checksum = 1; | |
656 | else | |
657 | csum_info->transmit.udp_checksum = 1; | |
658 | } else { | |
ad19bc8a | 659 | csum_info->transmit.is_ipv6 = 1; |
660 | ||
23312a3b | 661 | if (ipv6_hdr(skb)->nexthdr == IPPROTO_TCP) |
662 | csum_info->transmit.tcp_checksum = 1; | |
663 | else | |
664 | csum_info->transmit.udp_checksum = 1; | |
665 | } | |
ad19bc8a | 666 | } else { |
23312a3b | 667 | /* Can't do offload of this type of checksum */ |
ad19bc8a | 668 | if (skb_checksum_help(skb)) |
669 | goto drop; | |
670 | } | |
08cd04bf KS |
671 | } |
672 | ||
8a00251a KS |
673 | /* Start filling in the page buffers with the rndis hdr */ |
674 | rndis_msg->msg_len += rndis_msg_size; | |
942396b0 | 675 | packet->total_data_buflen = rndis_msg->msg_len; |
8a00251a | 676 | packet->page_buf_cnt = init_page_array(rndis_msg, rndis_msg_size, |
02b6de01 | 677 | skb, packet, pb); |
8a00251a | 678 | |
76d13b56 | 679 | /* timestamp packet in software */ |
680 | skb_tx_timestamp(skb); | |
2a926f79 | 681 | |
cfd8afd9 | 682 | ret = netvsc_send(net, packet, rndis_msg, pb, skb); |
793e3955 | 683 | if (likely(ret == 0)) |
0ab05141 | 684 | return NETDEV_TX_OK; |
4323b47c SH |
685 | |
686 | if (ret == -EAGAIN) { | |
687 | ++net_device_ctx->eth_stats.tx_busy; | |
0ab05141 | 688 | return NETDEV_TX_BUSY; |
4323b47c SH |
689 | } |
690 | ||
691 | if (ret == -ENOSPC) | |
692 | ++net_device_ctx->eth_stats.tx_no_space; | |
0ab05141 SH |
693 | |
694 | drop: | |
695 | dev_kfree_skb_any(skb); | |
696 | net->stats.tx_dropped++; | |
fceaf24a | 697 | |
0ab05141 | 698 | return NETDEV_TX_OK; |
4323b47c SH |
699 | |
700 | no_memory: | |
701 | ++net_device_ctx->eth_stats.tx_no_memory; | |
702 | goto drop; | |
fceaf24a | 703 | } |
89bb42b1 | 704 | |
3e189519 | 705 | /* |
02fafbc6 GKH |
706 | * netvsc_linkstatus_callback - Link up/down notification |
707 | */ | |
79cf1bae | 708 | void netvsc_linkstatus_callback(struct net_device *net, |
3a494e71 | 709 | struct rndis_message *resp) |
fceaf24a | 710 | { |
3a494e71 | 711 | struct rndis_indicate_status *indicate = &resp->msg.indicate_status; |
79cf1bae | 712 | struct net_device_context *ndev_ctx = netdev_priv(net); |
27a70af3 VK |
713 | struct netvsc_reconfig *event; |
714 | unsigned long flags; | |
891de74d | 715 | |
7f5d5af0 HZ |
716 | /* Update the physical link speed when changing to another vSwitch */ |
717 | if (indicate->status == RNDIS_STATUS_LINK_SPEED_CHANGE) { | |
718 | u32 speed; | |
719 | ||
89bb42b1 | 720 | speed = *(u32 *)((void *)indicate |
721 | + indicate->status_buf_offset) / 10000; | |
7f5d5af0 HZ |
722 | ndev_ctx->speed = speed; |
723 | return; | |
724 | } | |
725 | ||
726 | /* Handle these link change statuses below */ | |
27a70af3 VK |
727 | if (indicate->status != RNDIS_STATUS_NETWORK_CHANGE && |
728 | indicate->status != RNDIS_STATUS_MEDIA_CONNECT && | |
729 | indicate->status != RNDIS_STATUS_MEDIA_DISCONNECT) | |
3a494e71 | 730 | return; |
891de74d | 731 | |
7f5d5af0 | 732 | if (net->reg_state != NETREG_REGISTERED) |
fceaf24a | 733 | return; |
fceaf24a | 734 | |
27a70af3 VK |
735 | event = kzalloc(sizeof(*event), GFP_ATOMIC); |
736 | if (!event) | |
737 | return; | |
738 | event->event = indicate->status; | |
739 | ||
740 | spin_lock_irqsave(&ndev_ctx->lock, flags); | |
741 | list_add_tail(&event->list, &ndev_ctx->reconfig_events); | |
742 | spin_unlock_irqrestore(&ndev_ctx->lock, flags); | |
743 | ||
744 | schedule_delayed_work(&ndev_ctx->dwork, 0); | |
fceaf24a HJ |
745 | } |
746 | ||
84bf9cef | 747 | static struct sk_buff *netvsc_alloc_recv_skb(struct net_device *net, |
c8e4eff4 | 748 | struct netvsc_channel *nvchan) |
fceaf24a | 749 | { |
c8e4eff4 HZ |
750 | struct napi_struct *napi = &nvchan->napi; |
751 | const struct ndis_pkt_8021q_info *vlan = nvchan->rsc.vlan; | |
752 | const struct ndis_tcp_ip_checksum_info *csum_info = | |
753 | nvchan->rsc.csum_info; | |
fceaf24a | 754 | struct sk_buff *skb; |
c8e4eff4 | 755 | int i; |
fceaf24a | 756 | |
c8e4eff4 | 757 | skb = napi_alloc_skb(napi, nvchan->rsc.pktlen); |
84bf9cef KS |
758 | if (!skb) |
759 | return skb; | |
fceaf24a | 760 | |
02fafbc6 GKH |
761 | /* |
762 | * Copy to skb. This copy is needed here since the memory pointed by | |
763 | * hv_netvsc_packet cannot be deallocated | |
764 | */ | |
c8e4eff4 HZ |
765 | for (i = 0; i < nvchan->rsc.cnt; i++) |
766 | skb_put_data(skb, nvchan->rsc.data[i], nvchan->rsc.len[i]); | |
fceaf24a HJ |
767 | |
768 | skb->protocol = eth_type_trans(skb, net); | |
e52fed71 SH |
769 | |
770 | /* skb is already created with CHECKSUM_NONE */ | |
771 | skb_checksum_none_assert(skb); | |
772 | ||
773 | /* | |
774 | * In Linux, the IP checksum is always checked. | |
775 | * Do L4 checksum offload if enabled and present. | |
776 | */ | |
777 | if (csum_info && (net->features & NETIF_F_RXCSUM)) { | |
778 | if (csum_info->receive.tcp_checksum_succeeded || | |
779 | csum_info->receive.udp_checksum_succeeded) | |
e3d605ed | 780 | skb->ip_summed = CHECKSUM_UNNECESSARY; |
e3d605ed KS |
781 | } |
782 | ||
dc54a08c | 783 | if (vlan) { |
98ba780e MM |
784 | u16 vlan_tci = vlan->vlanid | (vlan->pri << VLAN_PRIO_SHIFT) | |
785 | (vlan->cfi ? VLAN_CFI_MASK : 0); | |
dc54a08c | 786 | |
93725cbd | 787 | __vlan_hwaccel_put_tag(skb, htons(ETH_P_8021Q), |
760d1e36 | 788 | vlan_tci); |
dc54a08c | 789 | } |
fceaf24a | 790 | |
84bf9cef KS |
791 | return skb; |
792 | } | |
793 | ||
794 | /* | |
795 | * netvsc_recv_callback - Callback when we receive a packet from the | |
796 | * "wire" on the specified device. | |
797 | */ | |
dc54a08c | 798 | int netvsc_recv_callback(struct net_device *net, |
345ac089 | 799 | struct netvsc_device *net_device, |
c8e4eff4 | 800 | struct netvsc_channel *nvchan) |
84bf9cef | 801 | { |
3d541ac5 | 802 | struct net_device_context *net_device_ctx = netdev_priv(net); |
c8e4eff4 | 803 | struct vmbus_channel *channel = nvchan->channel; |
742fe54c | 804 | u16 q_idx = channel->offermsg.offer.sub_channel_index; |
84bf9cef | 805 | struct sk_buff *skb; |
84bf9cef | 806 | struct netvsc_stats *rx_stats; |
84bf9cef | 807 | |
9cbcc428 | 808 | if (net->reg_state != NETREG_REGISTERED) |
84bf9cef KS |
809 | return NVSP_STAT_FAIL; |
810 | ||
84bf9cef | 811 | /* Allocate a skb - TODO direct I/O to pages? */ |
c8e4eff4 HZ |
812 | skb = netvsc_alloc_recv_skb(net, nvchan); |
813 | ||
84bf9cef | 814 | if (unlikely(!skb)) { |
f61a9d62 | 815 | ++net_device_ctx->eth_stats.rx_no_memory; |
0719e72c | 816 | rcu_read_unlock(); |
84bf9cef KS |
817 | return NVSP_STAT_FAIL; |
818 | } | |
5b54dac8 | 819 | |
0c195567 | 820 | skb_record_rx_queue(skb, q_idx); |
9cbcc428 SH |
821 | |
822 | /* | |
823 | * Even if injecting the packet, record the statistics | |
824 | * on the synthetic device because modifying the VF device | |
825 | * statistics will not work correctly. | |
826 | */ | |
742fe54c | 827 | rx_stats = &nvchan->rx_stats; |
4b02b58b | 828 | u64_stats_update_begin(&rx_stats->syncp); |
7eafd9b4 | 829 | rx_stats->packets++; |
c8e4eff4 | 830 | rx_stats->bytes += nvchan->rsc.pktlen; |
f7ad75b7 SH |
831 | |
832 | if (skb->pkt_type == PACKET_BROADCAST) | |
833 | ++rx_stats->broadcast; | |
834 | else if (skb->pkt_type == PACKET_MULTICAST) | |
835 | ++rx_stats->multicast; | |
4b02b58b | 836 | u64_stats_update_end(&rx_stats->syncp); |
9495c282 | 837 | |
742fe54c | 838 | napi_gro_receive(&nvchan->napi, skb); |
5c71dadb | 839 | return NVSP_STAT_SUCCESS; |
fceaf24a HJ |
840 | } |
841 | ||
f82f4ad7 SH |
842 | static void netvsc_get_drvinfo(struct net_device *net, |
843 | struct ethtool_drvinfo *info) | |
844 | { | |
7826d43f | 845 | strlcpy(info->driver, KBUILD_MODNAME, sizeof(info->driver)); |
7826d43f | 846 | strlcpy(info->fw_version, "N/A", sizeof(info->fw_version)); |
f82f4ad7 SH |
847 | } |
848 | ||
59995370 AS |
849 | static void netvsc_get_channels(struct net_device *net, |
850 | struct ethtool_channels *channel) | |
851 | { | |
852 | struct net_device_context *net_device_ctx = netdev_priv(net); | |
545a8e79 | 853 | struct netvsc_device *nvdev = rtnl_dereference(net_device_ctx->nvdev); |
59995370 AS |
854 | |
855 | if (nvdev) { | |
856 | channel->max_combined = nvdev->max_chn; | |
857 | channel->combined_count = nvdev->num_chn; | |
858 | } | |
859 | } | |
860 | ||
7c9f335a HZ |
861 | /* Alloc struct netvsc_device_info, and initialize it from either existing |
862 | * struct netvsc_device, or from default values. | |
863 | */ | |
864 | static struct netvsc_device_info *netvsc_devinfo_get | |
865 | (struct netvsc_device *nvdev) | |
866 | { | |
867 | struct netvsc_device_info *dev_info; | |
868 | ||
869 | dev_info = kzalloc(sizeof(*dev_info), GFP_ATOMIC); | |
870 | ||
871 | if (!dev_info) | |
872 | return NULL; | |
873 | ||
874 | if (nvdev) { | |
875 | dev_info->num_chn = nvdev->num_chn; | |
876 | dev_info->send_sections = nvdev->send_section_cnt; | |
877 | dev_info->send_section_size = nvdev->send_section_size; | |
878 | dev_info->recv_sections = nvdev->recv_section_cnt; | |
879 | dev_info->recv_section_size = nvdev->recv_section_size; | |
17d91256 HZ |
880 | |
881 | memcpy(dev_info->rss_key, nvdev->extension->rss_key, | |
882 | NETVSC_HASH_KEYLEN); | |
7c9f335a HZ |
883 | } else { |
884 | dev_info->num_chn = VRSS_CHANNEL_DEFAULT; | |
885 | dev_info->send_sections = NETVSC_DEFAULT_TX; | |
886 | dev_info->send_section_size = NETVSC_SEND_SECTION_SIZE; | |
887 | dev_info->recv_sections = NETVSC_DEFAULT_RX; | |
888 | dev_info->recv_section_size = NETVSC_RECV_SECTION_SIZE; | |
889 | } | |
890 | ||
891 | return dev_info; | |
892 | } | |
893 | ||
7b2ee50c SH |
894 | static int netvsc_detach(struct net_device *ndev, |
895 | struct netvsc_device *nvdev) | |
896 | { | |
897 | struct net_device_context *ndev_ctx = netdev_priv(ndev); | |
898 | struct hv_device *hdev = ndev_ctx->device_ctx; | |
899 | int ret; | |
900 | ||
901 | /* Don't try continuing to try and setup sub channels */ | |
902 | if (cancel_work_sync(&nvdev->subchan_work)) | |
903 | nvdev->num_chn = 1; | |
904 | ||
905 | /* If device was up (receiving) then shutdown */ | |
906 | if (netif_running(ndev)) { | |
907 | netif_tx_disable(ndev); | |
908 | ||
909 | ret = rndis_filter_close(nvdev); | |
910 | if (ret) { | |
911 | netdev_err(ndev, | |
912 | "unable to close device (ret %d).\n", ret); | |
913 | return ret; | |
914 | } | |
915 | ||
916 | ret = netvsc_wait_until_empty(nvdev); | |
917 | if (ret) { | |
918 | netdev_err(ndev, | |
919 | "Ring buffer not empty after closing rndis\n"); | |
920 | return ret; | |
921 | } | |
922 | } | |
923 | ||
924 | netif_device_detach(ndev); | |
925 | ||
926 | rndis_filter_device_remove(hdev, nvdev); | |
927 | ||
928 | return 0; | |
929 | } | |
930 | ||
931 | static int netvsc_attach(struct net_device *ndev, | |
932 | struct netvsc_device_info *dev_info) | |
933 | { | |
934 | struct net_device_context *ndev_ctx = netdev_priv(ndev); | |
935 | struct hv_device *hdev = ndev_ctx->device_ctx; | |
936 | struct netvsc_device *nvdev; | |
937 | struct rndis_device *rdev; | |
938 | int ret; | |
939 | ||
940 | nvdev = rndis_filter_device_add(hdev, dev_info); | |
941 | if (IS_ERR(nvdev)) | |
942 | return PTR_ERR(nvdev); | |
943 | ||
3ffe64f1 | 944 | if (nvdev->num_chn > 1) { |
17d91256 | 945 | ret = rndis_set_subchannel(ndev, nvdev, dev_info); |
3ffe64f1 SH |
946 | |
947 | /* if unavailable, just proceed with one queue */ | |
948 | if (ret) { | |
949 | nvdev->max_chn = 1; | |
950 | nvdev->num_chn = 1; | |
951 | } | |
952 | } | |
953 | ||
954 | /* In any case device is now ready */ | |
955 | netif_device_attach(ndev); | |
7b2ee50c | 956 | |
3ffe64f1 | 957 | /* Note: enable and attach happen when sub-channels setup */ |
7b2ee50c SH |
958 | netif_carrier_off(ndev); |
959 | ||
960 | if (netif_running(ndev)) { | |
961 | ret = rndis_filter_open(nvdev); | |
962 | if (ret) | |
963 | return ret; | |
964 | ||
965 | rdev = nvdev->extension; | |
966 | if (!rdev->link_state) | |
967 | netif_carrier_on(ndev); | |
968 | } | |
969 | ||
970 | return 0; | |
971 | } | |
972 | ||
b5960e6e AS |
973 | static int netvsc_set_channels(struct net_device *net, |
974 | struct ethtool_channels *channels) | |
975 | { | |
976 | struct net_device_context *net_device_ctx = netdev_priv(net); | |
545a8e79 | 977 | struct netvsc_device *nvdev = rtnl_dereference(net_device_ctx->nvdev); |
7ca45933 | 978 | unsigned int orig, count = channels->combined_count; |
7c9f335a | 979 | struct netvsc_device_info *device_info; |
7b2ee50c | 980 | int ret; |
2b01888d | 981 | |
982 | /* We do not support separate count for rx, tx, or other */ | |
983 | if (count == 0 || | |
984 | channels->rx_count || channels->tx_count || channels->other_count) | |
985 | return -EINVAL; | |
986 | ||
a0be450e | 987 | if (!nvdev || nvdev->destroy) |
b5960e6e AS |
988 | return -ENODEV; |
989 | ||
2b01888d | 990 | if (nvdev->nvsp_version < NVSP_PROTOCOL_VERSION_5) |
b5960e6e | 991 | return -EINVAL; |
b5960e6e | 992 | |
2b01888d | 993 | if (count > nvdev->max_chn) |
b5960e6e AS |
994 | return -EINVAL; |
995 | ||
7ca45933 | 996 | orig = nvdev->num_chn; |
b5960e6e | 997 | |
7c9f335a HZ |
998 | device_info = netvsc_devinfo_get(nvdev); |
999 | ||
1000 | if (!device_info) | |
1001 | return -ENOMEM; | |
1002 | ||
1003 | device_info->num_chn = count; | |
8b532797 | 1004 | |
7b2ee50c SH |
1005 | ret = netvsc_detach(net, nvdev); |
1006 | if (ret) | |
7c9f335a | 1007 | goto out; |
7ca45933 | 1008 | |
7c9f335a | 1009 | ret = netvsc_attach(net, device_info); |
7b2ee50c | 1010 | if (ret) { |
7c9f335a HZ |
1011 | device_info->num_chn = orig; |
1012 | if (netvsc_attach(net, device_info)) | |
7b2ee50c | 1013 | netdev_err(net, "restoring channel setting failed\n"); |
7ca45933 | 1014 | } |
b5960e6e | 1015 | |
7c9f335a HZ |
1016 | out: |
1017 | kfree(device_info); | |
b5960e6e | 1018 | return ret; |
b5960e6e AS |
1019 | } |
1020 | ||
5e8456fd PR |
1021 | static bool |
1022 | netvsc_validate_ethtool_ss_cmd(const struct ethtool_link_ksettings *cmd) | |
49eb9389 | 1023 | { |
5e8456fd PR |
1024 | struct ethtool_link_ksettings diff1 = *cmd; |
1025 | struct ethtool_link_ksettings diff2 = {}; | |
49eb9389 | 1026 | |
5e8456fd PR |
1027 | diff1.base.speed = 0; |
1028 | diff1.base.duplex = 0; | |
49eb9389 | 1029 | /* advertising and cmd are usually set */ |
5e8456fd PR |
1030 | ethtool_link_ksettings_zero_link_mode(&diff1, advertising); |
1031 | diff1.base.cmd = 0; | |
49eb9389 | 1032 | /* We set port to PORT_OTHER */ |
5e8456fd | 1033 | diff2.base.port = PORT_OTHER; |
49eb9389 | 1034 | |
1035 | return !memcmp(&diff1, &diff2, sizeof(diff1)); | |
1036 | } | |
1037 | ||
1038 | static void netvsc_init_settings(struct net_device *dev) | |
1039 | { | |
1040 | struct net_device_context *ndc = netdev_priv(dev); | |
1041 | ||
486e3981 | 1042 | ndc->l4_hash = HV_DEFAULT_L4HASH; |
4823eb2f | 1043 | |
49eb9389 | 1044 | ndc->speed = SPEED_UNKNOWN; |
f3c9d40e | 1045 | ndc->duplex = DUPLEX_FULL; |
d6792a5a HZ |
1046 | |
1047 | dev->features = NETIF_F_LRO; | |
49eb9389 | 1048 | } |
1049 | ||
5e8456fd PR |
1050 | static int netvsc_get_link_ksettings(struct net_device *dev, |
1051 | struct ethtool_link_ksettings *cmd) | |
49eb9389 | 1052 | { |
1053 | struct net_device_context *ndc = netdev_priv(dev); | |
1054 | ||
5e8456fd PR |
1055 | cmd->base.speed = ndc->speed; |
1056 | cmd->base.duplex = ndc->duplex; | |
1057 | cmd->base.port = PORT_OTHER; | |
49eb9389 | 1058 | |
1059 | return 0; | |
1060 | } | |
1061 | ||
5e8456fd PR |
1062 | static int netvsc_set_link_ksettings(struct net_device *dev, |
1063 | const struct ethtool_link_ksettings *cmd) | |
49eb9389 | 1064 | { |
1065 | struct net_device_context *ndc = netdev_priv(dev); | |
1066 | u32 speed; | |
1067 | ||
5e8456fd | 1068 | speed = cmd->base.speed; |
49eb9389 | 1069 | if (!ethtool_validate_speed(speed) || |
5e8456fd | 1070 | !ethtool_validate_duplex(cmd->base.duplex) || |
49eb9389 | 1071 | !netvsc_validate_ethtool_ss_cmd(cmd)) |
1072 | return -EINVAL; | |
1073 | ||
1074 | ndc->speed = speed; | |
5e8456fd | 1075 | ndc->duplex = cmd->base.duplex; |
49eb9389 | 1076 | |
1077 | return 0; | |
1078 | } | |
1079 | ||
4d447c9a HZ |
1080 | static int netvsc_change_mtu(struct net_device *ndev, int mtu) |
1081 | { | |
1082 | struct net_device_context *ndevctx = netdev_priv(ndev); | |
0c195567 | 1083 | struct net_device *vf_netdev = rtnl_dereference(ndevctx->vf_netdev); |
545a8e79 | 1084 | struct netvsc_device *nvdev = rtnl_dereference(ndevctx->nvdev); |
9749fed5 | 1085 | int orig_mtu = ndev->mtu; |
7c9f335a | 1086 | struct netvsc_device_info *device_info; |
9749fed5 | 1087 | int ret = 0; |
4d447c9a | 1088 | |
a0be450e | 1089 | if (!nvdev || nvdev->destroy) |
4d447c9a HZ |
1090 | return -ENODEV; |
1091 | ||
7c9f335a HZ |
1092 | device_info = netvsc_devinfo_get(nvdev); |
1093 | ||
1094 | if (!device_info) | |
1095 | return -ENOMEM; | |
1096 | ||
0c195567 | 1097 | /* Change MTU of underlying VF netdev first. */ |
1098 | if (vf_netdev) { | |
1099 | ret = dev_set_mtu(vf_netdev, mtu); | |
1100 | if (ret) | |
7c9f335a | 1101 | goto out; |
0c195567 | 1102 | } |
1103 | ||
7b2ee50c SH |
1104 | ret = netvsc_detach(ndev, nvdev); |
1105 | if (ret) | |
1106 | goto rollback_vf; | |
4d447c9a HZ |
1107 | |
1108 | ndev->mtu = mtu; | |
1109 | ||
7c9f335a HZ |
1110 | ret = netvsc_attach(ndev, device_info); |
1111 | if (!ret) | |
1112 | goto out; | |
ea383bf1 | 1113 | |
7b2ee50c SH |
1114 | /* Attempt rollback to original MTU */ |
1115 | ndev->mtu = orig_mtu; | |
163891d7 | 1116 | |
7c9f335a | 1117 | if (netvsc_attach(ndev, device_info)) |
7b2ee50c SH |
1118 | netdev_err(ndev, "restoring mtu failed\n"); |
1119 | rollback_vf: | |
1120 | if (vf_netdev) | |
1121 | dev_set_mtu(vf_netdev, orig_mtu); | |
1bdcec8a | 1122 | |
7c9f335a HZ |
1123 | out: |
1124 | kfree(device_info); | |
9749fed5 | 1125 | return ret; |
4d447c9a HZ |
1126 | } |
1127 | ||
0c195567 | 1128 | static void netvsc_get_vf_stats(struct net_device *net, |
1129 | struct netvsc_vf_pcpu_stats *tot) | |
1130 | { | |
1131 | struct net_device_context *ndev_ctx = netdev_priv(net); | |
1132 | int i; | |
1133 | ||
1134 | memset(tot, 0, sizeof(*tot)); | |
1135 | ||
1136 | for_each_possible_cpu(i) { | |
1137 | const struct netvsc_vf_pcpu_stats *stats | |
1138 | = per_cpu_ptr(ndev_ctx->vf_stats, i); | |
1139 | u64 rx_packets, rx_bytes, tx_packets, tx_bytes; | |
1140 | unsigned int start; | |
1141 | ||
1142 | do { | |
1143 | start = u64_stats_fetch_begin_irq(&stats->syncp); | |
1144 | rx_packets = stats->rx_packets; | |
1145 | tx_packets = stats->tx_packets; | |
1146 | rx_bytes = stats->rx_bytes; | |
1147 | tx_bytes = stats->tx_bytes; | |
1148 | } while (u64_stats_fetch_retry_irq(&stats->syncp, start)); | |
1149 | ||
1150 | tot->rx_packets += rx_packets; | |
1151 | tot->tx_packets += tx_packets; | |
1152 | tot->rx_bytes += rx_bytes; | |
1153 | tot->tx_bytes += tx_bytes; | |
1154 | tot->tx_dropped += stats->tx_dropped; | |
1155 | } | |
1156 | } | |
1157 | ||
6ae74671 YR |
1158 | static void netvsc_get_pcpu_stats(struct net_device *net, |
1159 | struct netvsc_ethtool_pcpu_stats *pcpu_tot) | |
1160 | { | |
1161 | struct net_device_context *ndev_ctx = netdev_priv(net); | |
1162 | struct netvsc_device *nvdev = rcu_dereference_rtnl(ndev_ctx->nvdev); | |
1163 | int i; | |
1164 | ||
1165 | /* fetch percpu stats of vf */ | |
1166 | for_each_possible_cpu(i) { | |
1167 | const struct netvsc_vf_pcpu_stats *stats = | |
1168 | per_cpu_ptr(ndev_ctx->vf_stats, i); | |
1169 | struct netvsc_ethtool_pcpu_stats *this_tot = &pcpu_tot[i]; | |
1170 | unsigned int start; | |
1171 | ||
1172 | do { | |
1173 | start = u64_stats_fetch_begin_irq(&stats->syncp); | |
1174 | this_tot->vf_rx_packets = stats->rx_packets; | |
1175 | this_tot->vf_tx_packets = stats->tx_packets; | |
1176 | this_tot->vf_rx_bytes = stats->rx_bytes; | |
1177 | this_tot->vf_tx_bytes = stats->tx_bytes; | |
1178 | } while (u64_stats_fetch_retry_irq(&stats->syncp, start)); | |
1179 | this_tot->rx_packets = this_tot->vf_rx_packets; | |
1180 | this_tot->tx_packets = this_tot->vf_tx_packets; | |
1181 | this_tot->rx_bytes = this_tot->vf_rx_bytes; | |
1182 | this_tot->tx_bytes = this_tot->vf_tx_bytes; | |
1183 | } | |
1184 | ||
1185 | /* fetch percpu stats of netvsc */ | |
1186 | for (i = 0; i < nvdev->num_chn; i++) { | |
1187 | const struct netvsc_channel *nvchan = &nvdev->chan_table[i]; | |
1188 | const struct netvsc_stats *stats; | |
1189 | struct netvsc_ethtool_pcpu_stats *this_tot = | |
1190 | &pcpu_tot[nvchan->channel->target_cpu]; | |
1191 | u64 packets, bytes; | |
1192 | unsigned int start; | |
1193 | ||
1194 | stats = &nvchan->tx_stats; | |
1195 | do { | |
1196 | start = u64_stats_fetch_begin_irq(&stats->syncp); | |
1197 | packets = stats->packets; | |
1198 | bytes = stats->bytes; | |
1199 | } while (u64_stats_fetch_retry_irq(&stats->syncp, start)); | |
1200 | ||
1201 | this_tot->tx_bytes += bytes; | |
1202 | this_tot->tx_packets += packets; | |
1203 | ||
1204 | stats = &nvchan->rx_stats; | |
1205 | do { | |
1206 | start = u64_stats_fetch_begin_irq(&stats->syncp); | |
1207 | packets = stats->packets; | |
1208 | bytes = stats->bytes; | |
1209 | } while (u64_stats_fetch_retry_irq(&stats->syncp, start)); | |
1210 | ||
1211 | this_tot->rx_bytes += bytes; | |
1212 | this_tot->rx_packets += packets; | |
1213 | } | |
1214 | } | |
1215 | ||
bc1f4470 | 1216 | static void netvsc_get_stats64(struct net_device *net, |
1217 | struct rtnl_link_stats64 *t) | |
7eafd9b4 | 1218 | { |
1219 | struct net_device_context *ndev_ctx = netdev_priv(net); | |
776e726b | 1220 | struct netvsc_device *nvdev = rcu_dereference_rtnl(ndev_ctx->nvdev); |
0c195567 | 1221 | struct netvsc_vf_pcpu_stats vf_tot; |
89bb42b1 | 1222 | int i; |
6c80f3fc SX |
1223 | |
1224 | if (!nvdev) | |
1225 | return; | |
1226 | ||
0c195567 | 1227 | netdev_stats_to_stats64(t, &net->stats); |
1228 | ||
1229 | netvsc_get_vf_stats(net, &vf_tot); | |
1230 | t->rx_packets += vf_tot.rx_packets; | |
1231 | t->tx_packets += vf_tot.tx_packets; | |
1232 | t->rx_bytes += vf_tot.rx_bytes; | |
1233 | t->tx_bytes += vf_tot.tx_bytes; | |
1234 | t->tx_dropped += vf_tot.tx_dropped; | |
1235 | ||
6c80f3fc SX |
1236 | for (i = 0; i < nvdev->num_chn; i++) { |
1237 | const struct netvsc_channel *nvchan = &nvdev->chan_table[i]; | |
1238 | const struct netvsc_stats *stats; | |
1239 | u64 packets, bytes, multicast; | |
7eafd9b4 | 1240 | unsigned int start; |
1241 | ||
6c80f3fc | 1242 | stats = &nvchan->tx_stats; |
7eafd9b4 | 1243 | do { |
6c80f3fc SX |
1244 | start = u64_stats_fetch_begin_irq(&stats->syncp); |
1245 | packets = stats->packets; | |
1246 | bytes = stats->bytes; | |
1247 | } while (u64_stats_fetch_retry_irq(&stats->syncp, start)); | |
1248 | ||
1249 | t->tx_bytes += bytes; | |
1250 | t->tx_packets += packets; | |
7eafd9b4 | 1251 | |
6c80f3fc | 1252 | stats = &nvchan->rx_stats; |
7eafd9b4 | 1253 | do { |
6c80f3fc SX |
1254 | start = u64_stats_fetch_begin_irq(&stats->syncp); |
1255 | packets = stats->packets; | |
1256 | bytes = stats->bytes; | |
1257 | multicast = stats->multicast + stats->broadcast; | |
1258 | } while (u64_stats_fetch_retry_irq(&stats->syncp, start)); | |
1259 | ||
1260 | t->rx_bytes += bytes; | |
1261 | t->rx_packets += packets; | |
1262 | t->multicast += multicast; | |
7eafd9b4 | 1263 | } |
7eafd9b4 | 1264 | } |
1ce09e89 HZ |
1265 | |
1266 | static int netvsc_set_mac_addr(struct net_device *ndev, void *p) | |
1267 | { | |
867047c4 | 1268 | struct net_device_context *ndc = netdev_priv(ndev); |
16ba3266 | 1269 | struct net_device *vf_netdev = rtnl_dereference(ndc->vf_netdev); |
867047c4 | 1270 | struct netvsc_device *nvdev = rtnl_dereference(ndc->nvdev); |
1ce09e89 | 1271 | struct sockaddr *addr = p; |
1ce09e89 HZ |
1272 | int err; |
1273 | ||
16ba3266 | 1274 | err = eth_prepare_mac_addr_change(ndev, p); |
1275 | if (err) | |
1ce09e89 HZ |
1276 | return err; |
1277 | ||
867047c4 | 1278 | if (!nvdev) |
1279 | return -ENODEV; | |
1280 | ||
16ba3266 | 1281 | if (vf_netdev) { |
3a37a963 | 1282 | err = dev_set_mac_address(vf_netdev, addr, NULL); |
16ba3266 | 1283 | if (err) |
1284 | return err; | |
1285 | } | |
1286 | ||
867047c4 | 1287 | err = rndis_filter_set_device_mac(nvdev, addr->sa_data); |
16ba3266 | 1288 | if (!err) { |
1289 | eth_commit_mac_addr_change(ndev, p); | |
1290 | } else if (vf_netdev) { | |
1291 | /* rollback change on VF */ | |
1292 | memcpy(addr->sa_data, ndev->dev_addr, ETH_ALEN); | |
3a37a963 | 1293 | dev_set_mac_address(vf_netdev, addr, NULL); |
1ce09e89 HZ |
1294 | } |
1295 | ||
1296 | return err; | |
1297 | } | |
1298 | ||
4323b47c SH |
1299 | static const struct { |
1300 | char name[ETH_GSTRING_LEN]; | |
1301 | u16 offset; | |
1302 | } netvsc_stats[] = { | |
1303 | { "tx_scattered", offsetof(struct netvsc_ethtool_stats, tx_scattered) }, | |
f61a9d62 | 1304 | { "tx_no_memory", offsetof(struct netvsc_ethtool_stats, tx_no_memory) }, |
4323b47c SH |
1305 | { "tx_no_space", offsetof(struct netvsc_ethtool_stats, tx_no_space) }, |
1306 | { "tx_too_big", offsetof(struct netvsc_ethtool_stats, tx_too_big) }, | |
1307 | { "tx_busy", offsetof(struct netvsc_ethtool_stats, tx_busy) }, | |
cad5c197 | 1308 | { "tx_send_full", offsetof(struct netvsc_ethtool_stats, tx_send_full) }, |
1309 | { "rx_comp_busy", offsetof(struct netvsc_ethtool_stats, rx_comp_busy) }, | |
f61a9d62 | 1310 | { "rx_no_memory", offsetof(struct netvsc_ethtool_stats, rx_no_memory) }, |
09af87d1 SX |
1311 | { "stop_queue", offsetof(struct netvsc_ethtool_stats, stop_queue) }, |
1312 | { "wake_queue", offsetof(struct netvsc_ethtool_stats, wake_queue) }, | |
6ae74671 YR |
1313 | }, pcpu_stats[] = { |
1314 | { "cpu%u_rx_packets", | |
1315 | offsetof(struct netvsc_ethtool_pcpu_stats, rx_packets) }, | |
1316 | { "cpu%u_rx_bytes", | |
1317 | offsetof(struct netvsc_ethtool_pcpu_stats, rx_bytes) }, | |
1318 | { "cpu%u_tx_packets", | |
1319 | offsetof(struct netvsc_ethtool_pcpu_stats, tx_packets) }, | |
1320 | { "cpu%u_tx_bytes", | |
1321 | offsetof(struct netvsc_ethtool_pcpu_stats, tx_bytes) }, | |
1322 | { "cpu%u_vf_rx_packets", | |
1323 | offsetof(struct netvsc_ethtool_pcpu_stats, vf_rx_packets) }, | |
1324 | { "cpu%u_vf_rx_bytes", | |
1325 | offsetof(struct netvsc_ethtool_pcpu_stats, vf_rx_bytes) }, | |
1326 | { "cpu%u_vf_tx_packets", | |
1327 | offsetof(struct netvsc_ethtool_pcpu_stats, vf_tx_packets) }, | |
1328 | { "cpu%u_vf_tx_bytes", | |
1329 | offsetof(struct netvsc_ethtool_pcpu_stats, vf_tx_bytes) }, | |
0c195567 | 1330 | }, vf_stats[] = { |
1331 | { "vf_rx_packets", offsetof(struct netvsc_vf_pcpu_stats, rx_packets) }, | |
1332 | { "vf_rx_bytes", offsetof(struct netvsc_vf_pcpu_stats, rx_bytes) }, | |
1333 | { "vf_tx_packets", offsetof(struct netvsc_vf_pcpu_stats, tx_packets) }, | |
1334 | { "vf_tx_bytes", offsetof(struct netvsc_vf_pcpu_stats, tx_bytes) }, | |
1335 | { "vf_tx_dropped", offsetof(struct netvsc_vf_pcpu_stats, tx_dropped) }, | |
4323b47c SH |
1336 | }; |
1337 | ||
6c80f3fc | 1338 | #define NETVSC_GLOBAL_STATS_LEN ARRAY_SIZE(netvsc_stats) |
0c195567 | 1339 | #define NETVSC_VF_STATS_LEN ARRAY_SIZE(vf_stats) |
6c80f3fc | 1340 | |
6ae74671 YR |
1341 | /* statistics per queue (rx/tx packets/bytes) */ |
1342 | #define NETVSC_PCPU_STATS_LEN (num_present_cpus() * ARRAY_SIZE(pcpu_stats)) | |
1343 | ||
6c80f3fc SX |
1344 | /* 4 statistics per queue (rx/tx packets/bytes) */ |
1345 | #define NETVSC_QUEUE_STATS_LEN(dev) ((dev)->num_chn * 4) | |
1346 | ||
4323b47c SH |
1347 | static int netvsc_get_sset_count(struct net_device *dev, int string_set) |
1348 | { | |
6c80f3fc | 1349 | struct net_device_context *ndc = netdev_priv(dev); |
fbd4c7e7 | 1350 | struct netvsc_device *nvdev = rtnl_dereference(ndc->nvdev); |
545a8e79 | 1351 | |
1352 | if (!nvdev) | |
1353 | return -ENODEV; | |
6c80f3fc | 1354 | |
4323b47c SH |
1355 | switch (string_set) { |
1356 | case ETH_SS_STATS: | |
0c195567 | 1357 | return NETVSC_GLOBAL_STATS_LEN |
1358 | + NETVSC_VF_STATS_LEN | |
6ae74671 YR |
1359 | + NETVSC_QUEUE_STATS_LEN(nvdev) |
1360 | + NETVSC_PCPU_STATS_LEN; | |
4323b47c SH |
1361 | default: |
1362 | return -EINVAL; | |
1363 | } | |
1364 | } | |
1365 | ||
1366 | static void netvsc_get_ethtool_stats(struct net_device *dev, | |
1367 | struct ethtool_stats *stats, u64 *data) | |
1368 | { | |
1369 | struct net_device_context *ndc = netdev_priv(dev); | |
867047c4 | 1370 | struct netvsc_device *nvdev = rtnl_dereference(ndc->nvdev); |
4323b47c | 1371 | const void *nds = &ndc->eth_stats; |
6c80f3fc | 1372 | const struct netvsc_stats *qstats; |
0c195567 | 1373 | struct netvsc_vf_pcpu_stats sum; |
6ae74671 | 1374 | struct netvsc_ethtool_pcpu_stats *pcpu_sum; |
6c80f3fc SX |
1375 | unsigned int start; |
1376 | u64 packets, bytes; | |
6ae74671 | 1377 | int i, j, cpu; |
4323b47c | 1378 | |
545a8e79 | 1379 | if (!nvdev) |
1380 | return; | |
1381 | ||
6c80f3fc | 1382 | for (i = 0; i < NETVSC_GLOBAL_STATS_LEN; i++) |
4323b47c | 1383 | data[i] = *(unsigned long *)(nds + netvsc_stats[i].offset); |
6c80f3fc | 1384 | |
0c195567 | 1385 | netvsc_get_vf_stats(dev, &sum); |
1386 | for (j = 0; j < NETVSC_VF_STATS_LEN; j++) | |
1387 | data[i++] = *(u64 *)((void *)&sum + vf_stats[j].offset); | |
1388 | ||
6c80f3fc SX |
1389 | for (j = 0; j < nvdev->num_chn; j++) { |
1390 | qstats = &nvdev->chan_table[j].tx_stats; | |
1391 | ||
1392 | do { | |
1393 | start = u64_stats_fetch_begin_irq(&qstats->syncp); | |
1394 | packets = qstats->packets; | |
1395 | bytes = qstats->bytes; | |
1396 | } while (u64_stats_fetch_retry_irq(&qstats->syncp, start)); | |
1397 | data[i++] = packets; | |
1398 | data[i++] = bytes; | |
1399 | ||
1400 | qstats = &nvdev->chan_table[j].rx_stats; | |
1401 | do { | |
1402 | start = u64_stats_fetch_begin_irq(&qstats->syncp); | |
1403 | packets = qstats->packets; | |
1404 | bytes = qstats->bytes; | |
1405 | } while (u64_stats_fetch_retry_irq(&qstats->syncp, start)); | |
1406 | data[i++] = packets; | |
1407 | data[i++] = bytes; | |
1408 | } | |
6ae74671 YR |
1409 | |
1410 | pcpu_sum = kvmalloc_array(num_possible_cpus(), | |
1411 | sizeof(struct netvsc_ethtool_pcpu_stats), | |
1412 | GFP_KERNEL); | |
1413 | netvsc_get_pcpu_stats(dev, pcpu_sum); | |
1414 | for_each_present_cpu(cpu) { | |
1415 | struct netvsc_ethtool_pcpu_stats *this_sum = &pcpu_sum[cpu]; | |
1416 | ||
1417 | for (j = 0; j < ARRAY_SIZE(pcpu_stats); j++) | |
1418 | data[i++] = *(u64 *)((void *)this_sum | |
1419 | + pcpu_stats[j].offset); | |
1420 | } | |
1421 | kvfree(pcpu_sum); | |
4323b47c SH |
1422 | } |
1423 | ||
1424 | static void netvsc_get_strings(struct net_device *dev, u32 stringset, u8 *data) | |
1425 | { | |
6c80f3fc | 1426 | struct net_device_context *ndc = netdev_priv(dev); |
867047c4 | 1427 | struct netvsc_device *nvdev = rtnl_dereference(ndc->nvdev); |
6c80f3fc | 1428 | u8 *p = data; |
6ae74671 | 1429 | int i, cpu; |
4323b47c | 1430 | |
545a8e79 | 1431 | if (!nvdev) |
1432 | return; | |
1433 | ||
4323b47c SH |
1434 | switch (stringset) { |
1435 | case ETH_SS_STATS: | |
0c195567 | 1436 | for (i = 0; i < ARRAY_SIZE(netvsc_stats); i++) { |
1437 | memcpy(p, netvsc_stats[i].name, ETH_GSTRING_LEN); | |
1438 | p += ETH_GSTRING_LEN; | |
1439 | } | |
1440 | ||
1441 | for (i = 0; i < ARRAY_SIZE(vf_stats); i++) { | |
1442 | memcpy(p, vf_stats[i].name, ETH_GSTRING_LEN); | |
1443 | p += ETH_GSTRING_LEN; | |
1444 | } | |
6c80f3fc | 1445 | |
6c80f3fc SX |
1446 | for (i = 0; i < nvdev->num_chn; i++) { |
1447 | sprintf(p, "tx_queue_%u_packets", i); | |
1448 | p += ETH_GSTRING_LEN; | |
1449 | sprintf(p, "tx_queue_%u_bytes", i); | |
1450 | p += ETH_GSTRING_LEN; | |
1451 | sprintf(p, "rx_queue_%u_packets", i); | |
1452 | p += ETH_GSTRING_LEN; | |
1453 | sprintf(p, "rx_queue_%u_bytes", i); | |
1454 | p += ETH_GSTRING_LEN; | |
1455 | } | |
1456 | ||
6ae74671 YR |
1457 | for_each_present_cpu(cpu) { |
1458 | for (i = 0; i < ARRAY_SIZE(pcpu_stats); i++) { | |
1459 | sprintf(p, pcpu_stats[i].name, cpu); | |
1460 | p += ETH_GSTRING_LEN; | |
1461 | } | |
1462 | } | |
1463 | ||
4323b47c SH |
1464 | break; |
1465 | } | |
1466 | } | |
1467 | ||
b5a5dc8d | 1468 | static int |
4823eb2f HZ |
1469 | netvsc_get_rss_hash_opts(struct net_device_context *ndc, |
1470 | struct ethtool_rxnfc *info) | |
b5a5dc8d | 1471 | { |
486e3981 HZ |
1472 | const u32 l4_flag = RXH_L4_B_0_1 | RXH_L4_B_2_3; |
1473 | ||
b5a5dc8d | 1474 | info->data = RXH_IP_SRC | RXH_IP_DST; |
1475 | ||
1476 | switch (info->flow_type) { | |
1477 | case TCP_V4_FLOW: | |
0518ec4f HZ |
1478 | if (ndc->l4_hash & HV_TCP4_L4HASH) |
1479 | info->data |= l4_flag; | |
1480 | ||
1481 | break; | |
1482 | ||
b5a5dc8d | 1483 | case TCP_V6_FLOW: |
0518ec4f HZ |
1484 | if (ndc->l4_hash & HV_TCP6_L4HASH) |
1485 | info->data |= l4_flag; | |
1486 | ||
4823eb2f HZ |
1487 | break; |
1488 | ||
b5a5dc8d | 1489 | case UDP_V4_FLOW: |
486e3981 HZ |
1490 | if (ndc->l4_hash & HV_UDP4_L4HASH) |
1491 | info->data |= l4_flag; | |
4823eb2f HZ |
1492 | |
1493 | break; | |
1494 | ||
b5a5dc8d | 1495 | case UDP_V6_FLOW: |
486e3981 HZ |
1496 | if (ndc->l4_hash & HV_UDP6_L4HASH) |
1497 | info->data |= l4_flag; | |
4823eb2f HZ |
1498 | |
1499 | break; | |
1500 | ||
b5a5dc8d | 1501 | case IPV4_FLOW: |
1502 | case IPV6_FLOW: | |
1503 | break; | |
1504 | default: | |
1505 | info->data = 0; | |
1506 | break; | |
1507 | } | |
1508 | ||
1509 | return 0; | |
1510 | } | |
1511 | ||
b448f4e8 | 1512 | static int |
1513 | netvsc_get_rxnfc(struct net_device *dev, struct ethtool_rxnfc *info, | |
1514 | u32 *rules) | |
1515 | { | |
1516 | struct net_device_context *ndc = netdev_priv(dev); | |
867047c4 | 1517 | struct netvsc_device *nvdev = rtnl_dereference(ndc->nvdev); |
545a8e79 | 1518 | |
1519 | if (!nvdev) | |
1520 | return -ENODEV; | |
b448f4e8 | 1521 | |
1522 | switch (info->cmd) { | |
1523 | case ETHTOOL_GRXRINGS: | |
1524 | info->data = nvdev->num_chn; | |
1525 | return 0; | |
b5a5dc8d | 1526 | |
1527 | case ETHTOOL_GRXFH: | |
4823eb2f | 1528 | return netvsc_get_rss_hash_opts(ndc, info); |
b448f4e8 | 1529 | } |
1530 | return -EOPNOTSUPP; | |
1531 | } | |
1532 | ||
4823eb2f HZ |
1533 | static int netvsc_set_rss_hash_opts(struct net_device_context *ndc, |
1534 | struct ethtool_rxnfc *info) | |
1535 | { | |
1536 | if (info->data == (RXH_IP_SRC | RXH_IP_DST | | |
1537 | RXH_L4_B_0_1 | RXH_L4_B_2_3)) { | |
486e3981 | 1538 | switch (info->flow_type) { |
0518ec4f HZ |
1539 | case TCP_V4_FLOW: |
1540 | ndc->l4_hash |= HV_TCP4_L4HASH; | |
1541 | break; | |
1542 | ||
1543 | case TCP_V6_FLOW: | |
1544 | ndc->l4_hash |= HV_TCP6_L4HASH; | |
1545 | break; | |
1546 | ||
486e3981 HZ |
1547 | case UDP_V4_FLOW: |
1548 | ndc->l4_hash |= HV_UDP4_L4HASH; | |
1549 | break; | |
1550 | ||
1551 | case UDP_V6_FLOW: | |
1552 | ndc->l4_hash |= HV_UDP6_L4HASH; | |
1553 | break; | |
1554 | ||
1555 | default: | |
4823eb2f | 1556 | return -EOPNOTSUPP; |
486e3981 | 1557 | } |
4823eb2f HZ |
1558 | |
1559 | return 0; | |
1560 | } | |
1561 | ||
1562 | if (info->data == (RXH_IP_SRC | RXH_IP_DST)) { | |
486e3981 | 1563 | switch (info->flow_type) { |
0518ec4f HZ |
1564 | case TCP_V4_FLOW: |
1565 | ndc->l4_hash &= ~HV_TCP4_L4HASH; | |
1566 | break; | |
1567 | ||
1568 | case TCP_V6_FLOW: | |
1569 | ndc->l4_hash &= ~HV_TCP6_L4HASH; | |
1570 | break; | |
1571 | ||
486e3981 HZ |
1572 | case UDP_V4_FLOW: |
1573 | ndc->l4_hash &= ~HV_UDP4_L4HASH; | |
1574 | break; | |
1575 | ||
1576 | case UDP_V6_FLOW: | |
1577 | ndc->l4_hash &= ~HV_UDP6_L4HASH; | |
1578 | break; | |
1579 | ||
1580 | default: | |
4823eb2f | 1581 | return -EOPNOTSUPP; |
486e3981 | 1582 | } |
4823eb2f HZ |
1583 | |
1584 | return 0; | |
1585 | } | |
1586 | ||
1587 | return -EOPNOTSUPP; | |
1588 | } | |
1589 | ||
1590 | static int | |
1591 | netvsc_set_rxnfc(struct net_device *ndev, struct ethtool_rxnfc *info) | |
1592 | { | |
1593 | struct net_device_context *ndc = netdev_priv(ndev); | |
1594 | ||
1595 | if (info->cmd == ETHTOOL_SRXFH) | |
1596 | return netvsc_set_rss_hash_opts(ndc, info); | |
1597 | ||
1598 | return -EOPNOTSUPP; | |
1599 | } | |
1600 | ||
962f3fee | 1601 | static u32 netvsc_get_rxfh_key_size(struct net_device *dev) |
1602 | { | |
1603 | return NETVSC_HASH_KEYLEN; | |
1604 | } | |
1605 | ||
1606 | static u32 netvsc_rss_indir_size(struct net_device *dev) | |
1607 | { | |
ff4a4419 | 1608 | return ITAB_NUM; |
962f3fee | 1609 | } |
1610 | ||
1611 | static int netvsc_get_rxfh(struct net_device *dev, u32 *indir, u8 *key, | |
1612 | u8 *hfunc) | |
1613 | { | |
1614 | struct net_device_context *ndc = netdev_priv(dev); | |
867047c4 | 1615 | struct netvsc_device *ndev = rtnl_dereference(ndc->nvdev); |
eb996edb | 1616 | struct rndis_device *rndis_dev; |
ff4a4419 | 1617 | int i; |
962f3fee | 1618 | |
545a8e79 | 1619 | if (!ndev) |
1620 | return -ENODEV; | |
1621 | ||
962f3fee | 1622 | if (hfunc) |
1623 | *hfunc = ETH_RSS_HASH_TOP; /* Toeplitz */ | |
1624 | ||
eb996edb | 1625 | rndis_dev = ndev->extension; |
ff4a4419 | 1626 | if (indir) { |
1627 | for (i = 0; i < ITAB_NUM; i++) | |
47371300 | 1628 | indir[i] = rndis_dev->rx_table[i]; |
ff4a4419 | 1629 | } |
1630 | ||
962f3fee | 1631 | if (key) |
1632 | memcpy(key, rndis_dev->rss_key, NETVSC_HASH_KEYLEN); | |
1633 | ||
1634 | return 0; | |
1635 | } | |
1636 | ||
1637 | static int netvsc_set_rxfh(struct net_device *dev, const u32 *indir, | |
1638 | const u8 *key, const u8 hfunc) | |
1639 | { | |
1640 | struct net_device_context *ndc = netdev_priv(dev); | |
545a8e79 | 1641 | struct netvsc_device *ndev = rtnl_dereference(ndc->nvdev); |
eb996edb | 1642 | struct rndis_device *rndis_dev; |
ff4a4419 | 1643 | int i; |
962f3fee | 1644 | |
545a8e79 | 1645 | if (!ndev) |
1646 | return -ENODEV; | |
1647 | ||
962f3fee | 1648 | if (hfunc != ETH_RSS_HASH_NO_CHANGE && hfunc != ETH_RSS_HASH_TOP) |
1649 | return -EOPNOTSUPP; | |
1650 | ||
eb996edb | 1651 | rndis_dev = ndev->extension; |
ff4a4419 | 1652 | if (indir) { |
1653 | for (i = 0; i < ITAB_NUM; i++) | |
db3cd7af | 1654 | if (indir[i] >= ndev->num_chn) |
ff4a4419 | 1655 | return -EINVAL; |
1656 | ||
1657 | for (i = 0; i < ITAB_NUM; i++) | |
47371300 | 1658 | rndis_dev->rx_table[i] = indir[i]; |
ff4a4419 | 1659 | } |
1660 | ||
1661 | if (!key) { | |
1662 | if (!indir) | |
1663 | return 0; | |
1664 | ||
1665 | key = rndis_dev->rss_key; | |
1666 | } | |
962f3fee | 1667 | |
715e2ec5 | 1668 | return rndis_filter_set_rss_param(rndis_dev, key); |
962f3fee | 1669 | } |
1670 | ||
8b532797 | 1671 | /* Hyper-V RNDIS protocol does not have ring in the HW sense. |
1672 | * It does have pre-allocated receive area which is divided into sections. | |
1673 | */ | |
1674 | static void __netvsc_get_ringparam(struct netvsc_device *nvdev, | |
1675 | struct ethtool_ringparam *ring) | |
1676 | { | |
1677 | u32 max_buf_size; | |
1678 | ||
1679 | ring->rx_pending = nvdev->recv_section_cnt; | |
1680 | ring->tx_pending = nvdev->send_section_cnt; | |
1681 | ||
1682 | if (nvdev->nvsp_version <= NVSP_PROTOCOL_VERSION_2) | |
1683 | max_buf_size = NETVSC_RECEIVE_BUFFER_SIZE_LEGACY; | |
1684 | else | |
1685 | max_buf_size = NETVSC_RECEIVE_BUFFER_SIZE; | |
1686 | ||
1687 | ring->rx_max_pending = max_buf_size / nvdev->recv_section_size; | |
1688 | ring->tx_max_pending = NETVSC_SEND_BUFFER_SIZE | |
1689 | / nvdev->send_section_size; | |
1690 | } | |
1691 | ||
1692 | static void netvsc_get_ringparam(struct net_device *ndev, | |
1693 | struct ethtool_ringparam *ring) | |
1694 | { | |
1695 | struct net_device_context *ndevctx = netdev_priv(ndev); | |
1696 | struct netvsc_device *nvdev = rtnl_dereference(ndevctx->nvdev); | |
1697 | ||
1698 | if (!nvdev) | |
1699 | return; | |
1700 | ||
1701 | __netvsc_get_ringparam(nvdev, ring); | |
1702 | } | |
1703 | ||
1704 | static int netvsc_set_ringparam(struct net_device *ndev, | |
1705 | struct ethtool_ringparam *ring) | |
1706 | { | |
1707 | struct net_device_context *ndevctx = netdev_priv(ndev); | |
1708 | struct netvsc_device *nvdev = rtnl_dereference(ndevctx->nvdev); | |
7c9f335a | 1709 | struct netvsc_device_info *device_info; |
8b532797 | 1710 | struct ethtool_ringparam orig; |
1711 | u32 new_tx, new_rx; | |
8b532797 | 1712 | int ret = 0; |
1713 | ||
1714 | if (!nvdev || nvdev->destroy) | |
1715 | return -ENODEV; | |
1716 | ||
1717 | memset(&orig, 0, sizeof(orig)); | |
1718 | __netvsc_get_ringparam(nvdev, &orig); | |
1719 | ||
1720 | new_tx = clamp_t(u32, ring->tx_pending, | |
1721 | NETVSC_MIN_TX_SECTIONS, orig.tx_max_pending); | |
1722 | new_rx = clamp_t(u32, ring->rx_pending, | |
1723 | NETVSC_MIN_RX_SECTIONS, orig.rx_max_pending); | |
1724 | ||
1725 | if (new_tx == orig.tx_pending && | |
1726 | new_rx == orig.rx_pending) | |
1727 | return 0; /* no change */ | |
1728 | ||
7c9f335a HZ |
1729 | device_info = netvsc_devinfo_get(nvdev); |
1730 | ||
1731 | if (!device_info) | |
1732 | return -ENOMEM; | |
1733 | ||
1734 | device_info->send_sections = new_tx; | |
1735 | device_info->recv_sections = new_rx; | |
8b532797 | 1736 | |
7b2ee50c SH |
1737 | ret = netvsc_detach(ndev, nvdev); |
1738 | if (ret) | |
7c9f335a | 1739 | goto out; |
8b532797 | 1740 | |
7c9f335a | 1741 | ret = netvsc_attach(ndev, device_info); |
7b2ee50c | 1742 | if (ret) { |
7c9f335a HZ |
1743 | device_info->send_sections = orig.tx_pending; |
1744 | device_info->recv_sections = orig.rx_pending; | |
8b532797 | 1745 | |
7c9f335a | 1746 | if (netvsc_attach(ndev, device_info)) |
7b2ee50c SH |
1747 | netdev_err(ndev, "restoring ringparam failed"); |
1748 | } | |
8b532797 | 1749 | |
7c9f335a HZ |
1750 | out: |
1751 | kfree(device_info); | |
8b532797 | 1752 | return ret; |
1753 | } | |
1754 | ||
d6792a5a HZ |
1755 | static int netvsc_set_features(struct net_device *ndev, |
1756 | netdev_features_t features) | |
1757 | { | |
1758 | netdev_features_t change = features ^ ndev->features; | |
1759 | struct net_device_context *ndevctx = netdev_priv(ndev); | |
1760 | struct netvsc_device *nvdev = rtnl_dereference(ndevctx->nvdev); | |
1761 | struct ndis_offload_params offloads; | |
1762 | ||
1763 | if (!nvdev || nvdev->destroy) | |
1764 | return -ENODEV; | |
1765 | ||
1766 | if (!(change & NETIF_F_LRO)) | |
1767 | return 0; | |
1768 | ||
1769 | memset(&offloads, 0, sizeof(struct ndis_offload_params)); | |
1770 | ||
1771 | if (features & NETIF_F_LRO) { | |
1772 | offloads.rsc_ip_v4 = NDIS_OFFLOAD_PARAMETERS_RSC_ENABLED; | |
1773 | offloads.rsc_ip_v6 = NDIS_OFFLOAD_PARAMETERS_RSC_ENABLED; | |
1774 | } else { | |
1775 | offloads.rsc_ip_v4 = NDIS_OFFLOAD_PARAMETERS_RSC_DISABLED; | |
1776 | offloads.rsc_ip_v6 = NDIS_OFFLOAD_PARAMETERS_RSC_DISABLED; | |
1777 | } | |
1778 | ||
1779 | return rndis_filter_set_offload_params(ndev, nvdev, &offloads); | |
1780 | } | |
1781 | ||
273de02a HZ |
1782 | static u32 netvsc_get_msglevel(struct net_device *ndev) |
1783 | { | |
1784 | struct net_device_context *ndev_ctx = netdev_priv(ndev); | |
1785 | ||
1786 | return ndev_ctx->msg_enable; | |
1787 | } | |
1788 | ||
1789 | static void netvsc_set_msglevel(struct net_device *ndev, u32 val) | |
1790 | { | |
1791 | struct net_device_context *ndev_ctx = netdev_priv(ndev); | |
1792 | ||
1793 | ndev_ctx->msg_enable = val; | |
1794 | } | |
1795 | ||
f82f4ad7 SH |
1796 | static const struct ethtool_ops ethtool_ops = { |
1797 | .get_drvinfo = netvsc_get_drvinfo, | |
273de02a HZ |
1798 | .get_msglevel = netvsc_get_msglevel, |
1799 | .set_msglevel = netvsc_set_msglevel, | |
f82f4ad7 | 1800 | .get_link = ethtool_op_get_link, |
4323b47c SH |
1801 | .get_ethtool_stats = netvsc_get_ethtool_stats, |
1802 | .get_sset_count = netvsc_get_sset_count, | |
1803 | .get_strings = netvsc_get_strings, | |
59995370 | 1804 | .get_channels = netvsc_get_channels, |
b5960e6e | 1805 | .set_channels = netvsc_set_channels, |
76d13b56 | 1806 | .get_ts_info = ethtool_op_get_ts_info, |
b448f4e8 | 1807 | .get_rxnfc = netvsc_get_rxnfc, |
4823eb2f | 1808 | .set_rxnfc = netvsc_set_rxnfc, |
962f3fee | 1809 | .get_rxfh_key_size = netvsc_get_rxfh_key_size, |
1810 | .get_rxfh_indir_size = netvsc_rss_indir_size, | |
1811 | .get_rxfh = netvsc_get_rxfh, | |
1812 | .set_rxfh = netvsc_set_rxfh, | |
5e8456fd PR |
1813 | .get_link_ksettings = netvsc_get_link_ksettings, |
1814 | .set_link_ksettings = netvsc_set_link_ksettings, | |
8b532797 | 1815 | .get_ringparam = netvsc_get_ringparam, |
1816 | .set_ringparam = netvsc_set_ringparam, | |
f82f4ad7 SH |
1817 | }; |
1818 | ||
df2fff28 GKH |
1819 | static const struct net_device_ops device_ops = { |
1820 | .ndo_open = netvsc_open, | |
1821 | .ndo_stop = netvsc_close, | |
1822 | .ndo_start_xmit = netvsc_start_xmit, | |
bee9d41b SH |
1823 | .ndo_change_rx_flags = netvsc_change_rx_flags, |
1824 | .ndo_set_rx_mode = netvsc_set_rx_mode, | |
d6792a5a | 1825 | .ndo_set_features = netvsc_set_features, |
4d447c9a | 1826 | .ndo_change_mtu = netvsc_change_mtu, |
b681b588 | 1827 | .ndo_validate_addr = eth_validate_addr, |
1ce09e89 | 1828 | .ndo_set_mac_address = netvsc_set_mac_addr, |
5b54dac8 | 1829 | .ndo_select_queue = netvsc_select_queue, |
7eafd9b4 | 1830 | .ndo_get_stats64 = netvsc_get_stats64, |
df2fff28 GKH |
1831 | }; |
1832 | ||
c996edcf | 1833 | /* |
27a70af3 VK |
1834 | * Handle link status changes. For RNDIS_STATUS_NETWORK_CHANGE emulate link |
1835 | * down/up sequence. In case of RNDIS_STATUS_MEDIA_CONNECT when carrier is | |
1836 | * present send GARP packet to network peers with netif_notify_peers(). | |
c996edcf | 1837 | */ |
891de74d | 1838 | static void netvsc_link_change(struct work_struct *w) |
c996edcf | 1839 | { |
0a1275ca VK |
1840 | struct net_device_context *ndev_ctx = |
1841 | container_of(w, struct net_device_context, dwork.work); | |
1842 | struct hv_device *device_obj = ndev_ctx->device_ctx; | |
1843 | struct net_device *net = hv_get_drvdata(device_obj); | |
2ddd5e5f | 1844 | struct netvsc_device *net_device; |
891de74d | 1845 | struct rndis_device *rdev; |
27a70af3 VK |
1846 | struct netvsc_reconfig *event = NULL; |
1847 | bool notify = false, reschedule = false; | |
1848 | unsigned long flags, next_reconfig, delay; | |
c996edcf | 1849 | |
9b4e946c | 1850 | /* if changes are happening, comeback later */ |
1851 | if (!rtnl_trylock()) { | |
1852 | schedule_delayed_work(&ndev_ctx->dwork, LINKCHANGE_INT); | |
1853 | return; | |
1854 | } | |
1855 | ||
a0be450e | 1856 | net_device = rtnl_dereference(ndev_ctx->nvdev); |
1857 | if (!net_device) | |
1bdcec8a VK |
1858 | goto out_unlock; |
1859 | ||
891de74d | 1860 | rdev = net_device->extension; |
891de74d | 1861 | |
27a70af3 VK |
1862 | next_reconfig = ndev_ctx->last_reconfig + LINKCHANGE_INT; |
1863 | if (time_is_after_jiffies(next_reconfig)) { | |
1864 | /* link_watch only sends one notification with current state | |
1865 | * per second, avoid doing reconfig more frequently. Handle | |
1866 | * wrap around. | |
1867 | */ | |
1868 | delay = next_reconfig - jiffies; | |
1869 | delay = delay < LINKCHANGE_INT ? delay : LINKCHANGE_INT; | |
1870 | schedule_delayed_work(&ndev_ctx->dwork, delay); | |
1bdcec8a | 1871 | goto out_unlock; |
27a70af3 VK |
1872 | } |
1873 | ndev_ctx->last_reconfig = jiffies; | |
1874 | ||
1875 | spin_lock_irqsave(&ndev_ctx->lock, flags); | |
1876 | if (!list_empty(&ndev_ctx->reconfig_events)) { | |
1877 | event = list_first_entry(&ndev_ctx->reconfig_events, | |
1878 | struct netvsc_reconfig, list); | |
1879 | list_del(&event->list); | |
1880 | reschedule = !list_empty(&ndev_ctx->reconfig_events); | |
1881 | } | |
1882 | spin_unlock_irqrestore(&ndev_ctx->lock, flags); | |
1883 | ||
1884 | if (!event) | |
1bdcec8a | 1885 | goto out_unlock; |
27a70af3 VK |
1886 | |
1887 | switch (event->event) { | |
1888 | /* Only the following events are possible due to the check in | |
1889 | * netvsc_linkstatus_callback() | |
1890 | */ | |
1891 | case RNDIS_STATUS_MEDIA_CONNECT: | |
1892 | if (rdev->link_state) { | |
1893 | rdev->link_state = false; | |
0c195567 | 1894 | netif_carrier_on(net); |
27a70af3 VK |
1895 | netif_tx_wake_all_queues(net); |
1896 | } else { | |
1897 | notify = true; | |
1898 | } | |
1899 | kfree(event); | |
1900 | break; | |
1901 | case RNDIS_STATUS_MEDIA_DISCONNECT: | |
1902 | if (!rdev->link_state) { | |
1903 | rdev->link_state = true; | |
1904 | netif_carrier_off(net); | |
1905 | netif_tx_stop_all_queues(net); | |
1906 | } | |
1907 | kfree(event); | |
1908 | break; | |
1909 | case RNDIS_STATUS_NETWORK_CHANGE: | |
1910 | /* Only makes sense if carrier is present */ | |
1911 | if (!rdev->link_state) { | |
1912 | rdev->link_state = true; | |
1913 | netif_carrier_off(net); | |
1914 | netif_tx_stop_all_queues(net); | |
1915 | event->event = RNDIS_STATUS_MEDIA_CONNECT; | |
1916 | spin_lock_irqsave(&ndev_ctx->lock, flags); | |
15cfd407 | 1917 | list_add(&event->list, &ndev_ctx->reconfig_events); |
27a70af3 VK |
1918 | spin_unlock_irqrestore(&ndev_ctx->lock, flags); |
1919 | reschedule = true; | |
3a494e71 | 1920 | } |
27a70af3 | 1921 | break; |
891de74d HZ |
1922 | } |
1923 | ||
1924 | rtnl_unlock(); | |
1925 | ||
1926 | if (notify) | |
1927 | netdev_notify_peers(net); | |
27a70af3 VK |
1928 | |
1929 | /* link_watch only sends one notification with current state per | |
1930 | * second, handle next reconfig event in 2 seconds. | |
1931 | */ | |
1932 | if (reschedule) | |
1933 | schedule_delayed_work(&ndev_ctx->dwork, LINKCHANGE_INT); | |
1bdcec8a VK |
1934 | |
1935 | return; | |
1936 | ||
1937 | out_unlock: | |
1938 | rtnl_unlock(); | |
c996edcf HZ |
1939 | } |
1940 | ||
8cde8f0c SH |
1941 | static struct net_device *get_netvsc_byref(struct net_device *vf_netdev) |
1942 | { | |
7bf7bb37 | 1943 | struct net_device_context *net_device_ctx; |
8cde8f0c SH |
1944 | struct net_device *dev; |
1945 | ||
7bf7bb37 SH |
1946 | dev = netdev_master_upper_dev_get(vf_netdev); |
1947 | if (!dev || dev->netdev_ops != &device_ops) | |
1948 | return NULL; /* not a netvsc device */ | |
8cde8f0c | 1949 | |
7bf7bb37 SH |
1950 | net_device_ctx = netdev_priv(dev); |
1951 | if (!rtnl_dereference(net_device_ctx->nvdev)) | |
1952 | return NULL; /* device is removed */ | |
8cde8f0c | 1953 | |
7bf7bb37 | 1954 | return dev; |
8cde8f0c SH |
1955 | } |
1956 | ||
0c195567 | 1957 | /* Called when VF is injecting data into network stack. |
1958 | * Change the associated network device from VF to netvsc. | |
1959 | * note: already called with rcu_read_lock | |
1960 | */ | |
1961 | static rx_handler_result_t netvsc_vf_handle_frame(struct sk_buff **pskb) | |
1962 | { | |
1963 | struct sk_buff *skb = *pskb; | |
1964 | struct net_device *ndev = rcu_dereference(skb->dev->rx_handler_data); | |
1965 | struct net_device_context *ndev_ctx = netdev_priv(ndev); | |
1966 | struct netvsc_vf_pcpu_stats *pcpu_stats | |
1967 | = this_cpu_ptr(ndev_ctx->vf_stats); | |
1968 | ||
1969 | skb->dev = ndev; | |
1970 | ||
1971 | u64_stats_update_begin(&pcpu_stats->syncp); | |
1972 | pcpu_stats->rx_packets++; | |
1973 | pcpu_stats->rx_bytes += skb->len; | |
1974 | u64_stats_update_end(&pcpu_stats->syncp); | |
1975 | ||
1976 | return RX_HANDLER_ANOTHER; | |
1977 | } | |
1978 | ||
8cde8f0c SH |
1979 | static int netvsc_vf_join(struct net_device *vf_netdev, |
1980 | struct net_device *ndev) | |
1981 | { | |
1982 | struct net_device_context *ndev_ctx = netdev_priv(ndev); | |
1983 | int ret; | |
1984 | ||
1985 | ret = netdev_rx_handler_register(vf_netdev, | |
1986 | netvsc_vf_handle_frame, ndev); | |
1987 | if (ret != 0) { | |
1988 | netdev_err(vf_netdev, | |
1989 | "can not register netvsc VF receive handler (err = %d)\n", | |
1990 | ret); | |
1991 | goto rx_handler_failed; | |
1992 | } | |
1993 | ||
1994 | ret = netdev_master_upper_dev_link(vf_netdev, ndev, | |
1995 | NULL, NULL, NULL); | |
1996 | if (ret != 0) { | |
1997 | netdev_err(vf_netdev, | |
1998 | "can not set master device %s (err = %d)\n", | |
1999 | ndev->name, ret); | |
2000 | goto upper_link_failed; | |
2001 | } | |
2002 | ||
2003 | /* set slave flag before open to prevent IPv6 addrconf */ | |
2004 | vf_netdev->flags |= IFF_SLAVE; | |
2005 | ||
2006 | schedule_delayed_work(&ndev_ctx->vf_takeover, VF_TAKEOVER_INT); | |
2007 | ||
2008 | call_netdevice_notifiers(NETDEV_JOIN, vf_netdev); | |
2009 | ||
2010 | netdev_info(vf_netdev, "joined to %s\n", ndev->name); | |
2011 | return 0; | |
2012 | ||
2013 | upper_link_failed: | |
2014 | netdev_rx_handler_unregister(vf_netdev); | |
2015 | rx_handler_failed: | |
2016 | return ret; | |
2017 | } | |
2018 | ||
0c195567 | 2019 | static void __netvsc_vf_setup(struct net_device *ndev, |
2020 | struct net_device *vf_netdev) | |
2021 | { | |
2022 | int ret; | |
2023 | ||
0c195567 | 2024 | /* Align MTU of VF with master */ |
2025 | ret = dev_set_mtu(vf_netdev, ndev->mtu); | |
2026 | if (ret) | |
2027 | netdev_warn(vf_netdev, | |
2028 | "unable to change mtu to %u\n", ndev->mtu); | |
2029 | ||
bee9d41b | 2030 | /* set multicast etc flags on VF */ |
567c5e13 | 2031 | dev_change_flags(vf_netdev, ndev->flags | IFF_SLAVE, NULL); |
b0dee791 SH |
2032 | |
2033 | /* sync address list from ndev to VF */ | |
2034 | netif_addr_lock_bh(ndev); | |
bee9d41b SH |
2035 | dev_uc_sync(vf_netdev, ndev); |
2036 | dev_mc_sync(vf_netdev, ndev); | |
b0dee791 | 2037 | netif_addr_unlock_bh(ndev); |
bee9d41b | 2038 | |
0c195567 | 2039 | if (netif_running(ndev)) { |
00f54e68 | 2040 | ret = dev_open(vf_netdev, NULL); |
0c195567 | 2041 | if (ret) |
2042 | netdev_warn(vf_netdev, | |
2043 | "unable to open: %d\n", ret); | |
2044 | } | |
2045 | } | |
2046 | ||
2047 | /* Setup VF as slave of the synthetic device. | |
2048 | * Runs in workqueue to avoid recursion in netlink callbacks. | |
2049 | */ | |
2050 | static void netvsc_vf_setup(struct work_struct *w) | |
2051 | { | |
2052 | struct net_device_context *ndev_ctx | |
6123c668 | 2053 | = container_of(w, struct net_device_context, vf_takeover.work); |
0c195567 | 2054 | struct net_device *ndev = hv_get_drvdata(ndev_ctx->device_ctx); |
2055 | struct net_device *vf_netdev; | |
2056 | ||
fb84af8a | 2057 | if (!rtnl_trylock()) { |
6123c668 | 2058 | schedule_delayed_work(&ndev_ctx->vf_takeover, 0); |
fb84af8a | 2059 | return; |
2060 | } | |
2061 | ||
0c195567 | 2062 | vf_netdev = rtnl_dereference(ndev_ctx->vf_netdev); |
2063 | if (vf_netdev) | |
2064 | __netvsc_vf_setup(ndev, vf_netdev); | |
2065 | ||
2066 | rtnl_unlock(); | |
2067 | } | |
2068 | ||
00547955 HZ |
2069 | /* Find netvsc by VF serial number. |
2070 | * The PCI hyperv controller records the serial number as the slot kobj name. | |
00d7ddba SH |
2071 | */ |
2072 | static struct net_device *get_netvsc_byslot(const struct net_device *vf_netdev) | |
2073 | { | |
2074 | struct device *parent = vf_netdev->dev.parent; | |
2075 | struct net_device_context *ndev_ctx; | |
2076 | struct pci_dev *pdev; | |
00547955 | 2077 | u32 serial; |
00d7ddba SH |
2078 | |
2079 | if (!parent || !dev_is_pci(parent)) | |
2080 | return NULL; /* not a PCI device */ | |
2081 | ||
2082 | pdev = to_pci_dev(parent); | |
2083 | if (!pdev->slot) { | |
2084 | netdev_notice(vf_netdev, "no PCI slot information\n"); | |
2085 | return NULL; | |
2086 | } | |
2087 | ||
00547955 HZ |
2088 | if (kstrtou32(pci_slot_name(pdev->slot), 10, &serial)) { |
2089 | netdev_notice(vf_netdev, "Invalid vf serial:%s\n", | |
2090 | pci_slot_name(pdev->slot)); | |
2091 | return NULL; | |
2092 | } | |
2093 | ||
00d7ddba SH |
2094 | list_for_each_entry(ndev_ctx, &netvsc_dev_list, list) { |
2095 | if (!ndev_ctx->vf_alloc) | |
2096 | continue; | |
2097 | ||
00547955 | 2098 | if (ndev_ctx->vf_serial == serial) |
00d7ddba SH |
2099 | return hv_get_drvdata(ndev_ctx->device_ctx); |
2100 | } | |
2101 | ||
2102 | netdev_notice(vf_netdev, | |
00547955 | 2103 | "no netdev found for vf serial:%u\n", serial); |
00d7ddba SH |
2104 | return NULL; |
2105 | } | |
2106 | ||
8cde8f0c | 2107 | static int netvsc_register_vf(struct net_device *vf_netdev) |
84bf9cef | 2108 | { |
0a1275ca | 2109 | struct net_device_context *net_device_ctx; |
84bf9cef | 2110 | struct netvsc_device *netvsc_dev; |
00d7ddba | 2111 | struct net_device *ndev; |
c0a41b88 | 2112 | int ret; |
84bf9cef | 2113 | |
8cde8f0c SH |
2114 | if (vf_netdev->addr_len != ETH_ALEN) |
2115 | return NOTIFY_DONE; | |
2116 | ||
00d7ddba | 2117 | ndev = get_netvsc_byslot(vf_netdev); |
8cde8f0c SH |
2118 | if (!ndev) |
2119 | return NOTIFY_DONE; | |
2120 | ||
0a1275ca | 2121 | net_device_ctx = netdev_priv(ndev); |
545a8e79 | 2122 | netvsc_dev = rtnl_dereference(net_device_ctx->nvdev); |
f207c10d | 2123 | if (!netvsc_dev || rtnl_dereference(net_device_ctx->vf_netdev)) |
8cde8f0c | 2124 | return NOTIFY_DONE; |
0c195567 | 2125 | |
52d3b494 | 2126 | /* if synthetic interface is a different namespace, |
c0a41b88 SH |
2127 | * then move the VF to that namespace; join will be |
2128 | * done again in that context. | |
2129 | */ | |
2130 | if (!net_eq(dev_net(ndev), dev_net(vf_netdev))) { | |
2131 | ret = dev_change_net_namespace(vf_netdev, | |
2132 | dev_net(ndev), "eth%d"); | |
2133 | if (ret) | |
2134 | netdev_err(vf_netdev, | |
2135 | "could not move to same namespace as %s: %d\n", | |
2136 | ndev->name, ret); | |
2137 | else | |
2138 | netdev_info(vf_netdev, | |
2139 | "VF moved to namespace with: %s\n", | |
2140 | ndev->name); | |
8cde8f0c | 2141 | return NOTIFY_DONE; |
c0a41b88 | 2142 | } |
1ff78076 | 2143 | |
8cde8f0c | 2144 | netdev_info(ndev, "VF registering: %s\n", vf_netdev->name); |
0c195567 | 2145 | |
c0a41b88 SH |
2146 | if (netvsc_vf_join(vf_netdev, ndev) != 0) |
2147 | return NOTIFY_DONE; | |
2148 | ||
07d0f000 | 2149 | dev_hold(vf_netdev); |
8cde8f0c SH |
2150 | rcu_assign_pointer(net_device_ctx->vf_netdev, vf_netdev); |
2151 | return NOTIFY_OK; | |
84bf9cef KS |
2152 | } |
2153 | ||
9a0c48df | 2154 | /* VF up/down change detected, schedule to change data path */ |
8cde8f0c | 2155 | static int netvsc_vf_changed(struct net_device *vf_netdev) |
84bf9cef | 2156 | { |
7b83f520 | 2157 | struct net_device_context *net_device_ctx; |
84bf9cef | 2158 | struct netvsc_device *netvsc_dev; |
8cde8f0c | 2159 | struct net_device *ndev; |
9a0c48df | 2160 | bool vf_is_up = netif_running(vf_netdev); |
fb84af8a | 2161 | |
8cde8f0c SH |
2162 | ndev = get_netvsc_byref(vf_netdev); |
2163 | if (!ndev) | |
2164 | return NOTIFY_DONE; | |
2165 | ||
7b83f520 | 2166 | net_device_ctx = netdev_priv(ndev); |
2167 | netvsc_dev = rtnl_dereference(net_device_ctx->nvdev); | |
0c195567 | 2168 | if (!netvsc_dev) |
8cde8f0c | 2169 | return NOTIFY_DONE; |
84bf9cef | 2170 | |
9a0c48df SH |
2171 | netvsc_switch_datapath(ndev, vf_is_up); |
2172 | netdev_info(ndev, "Data path switched %s VF: %s\n", | |
2173 | vf_is_up ? "to" : "from", vf_netdev->name); | |
84bf9cef | 2174 | |
8cde8f0c | 2175 | return NOTIFY_OK; |
84bf9cef KS |
2176 | } |
2177 | ||
8cde8f0c | 2178 | static int netvsc_unregister_vf(struct net_device *vf_netdev) |
84bf9cef | 2179 | { |
8cde8f0c | 2180 | struct net_device *ndev; |
0a1275ca | 2181 | struct net_device_context *net_device_ctx; |
84bf9cef | 2182 | |
8cde8f0c SH |
2183 | ndev = get_netvsc_byref(vf_netdev); |
2184 | if (!ndev) | |
2185 | return NOTIFY_DONE; | |
1ff78076 SS |
2186 | |
2187 | net_device_ctx = netdev_priv(ndev); | |
8cde8f0c | 2188 | cancel_delayed_work_sync(&net_device_ctx->vf_takeover); |
1ff78076 | 2189 | |
0a1275ca | 2190 | netdev_info(ndev, "VF unregistering: %s\n", vf_netdev->name); |
f207c10d | 2191 | |
8cde8f0c SH |
2192 | netdev_rx_handler_unregister(vf_netdev); |
2193 | netdev_upper_dev_unlink(vf_netdev, ndev); | |
f207c10d | 2194 | RCU_INIT_POINTER(net_device_ctx->vf_netdev, NULL); |
07d0f000 | 2195 | dev_put(vf_netdev); |
ec158f77 | 2196 | |
8cde8f0c | 2197 | return NOTIFY_OK; |
84bf9cef KS |
2198 | } |
2199 | ||
84946899 S |
2200 | static int netvsc_probe(struct hv_device *dev, |
2201 | const struct hv_vmbus_device_id *dev_id) | |
df2fff28 | 2202 | { |
df2fff28 GKH |
2203 | struct net_device *net = NULL; |
2204 | struct net_device_context *net_device_ctx; | |
7c9f335a | 2205 | struct netvsc_device_info *device_info = NULL; |
5b54dac8 | 2206 | struct netvsc_device *nvdev; |
0c195567 | 2207 | int ret = -ENOMEM; |
df2fff28 | 2208 | |
5b54dac8 | 2209 | net = alloc_etherdev_mq(sizeof(struct net_device_context), |
2b01888d | 2210 | VRSS_CHANNEL_MAX); |
df2fff28 | 2211 | if (!net) |
0c195567 | 2212 | goto no_net; |
df2fff28 | 2213 | |
1b07da51 HZ |
2214 | netif_carrier_off(net); |
2215 | ||
b37879e6 HZ |
2216 | netvsc_init_settings(net); |
2217 | ||
df2fff28 | 2218 | net_device_ctx = netdev_priv(net); |
9efd21e1 | 2219 | net_device_ctx->device_ctx = dev; |
3f300ff4 SX |
2220 | net_device_ctx->msg_enable = netif_msg_init(debug, default_msg); |
2221 | if (netif_msg_probe(net_device_ctx)) | |
2222 | netdev_dbg(net, "netvsc msg_enable: %d\n", | |
2223 | net_device_ctx->msg_enable); | |
2224 | ||
2ddd5e5f | 2225 | hv_set_drvdata(dev, net); |
f580aec4 | 2226 | |
891de74d | 2227 | INIT_DELAYED_WORK(&net_device_ctx->dwork, netvsc_link_change); |
df2fff28 | 2228 | |
27a70af3 VK |
2229 | spin_lock_init(&net_device_ctx->lock); |
2230 | INIT_LIST_HEAD(&net_device_ctx->reconfig_events); | |
6123c668 | 2231 | INIT_DELAYED_WORK(&net_device_ctx->vf_takeover, netvsc_vf_setup); |
0c195567 | 2232 | |
2233 | net_device_ctx->vf_stats | |
2234 | = netdev_alloc_pcpu_stats(struct netvsc_vf_pcpu_stats); | |
2235 | if (!net_device_ctx->vf_stats) | |
2236 | goto no_stats; | |
27a70af3 | 2237 | |
df2fff28 | 2238 | net->netdev_ops = &device_ops; |
7ad24ea4 | 2239 | net->ethtool_ops = ðtool_ops; |
9efd21e1 | 2240 | SET_NETDEV_DEV(net, &dev->device); |
df2fff28 | 2241 | |
14a03cf8 VK |
2242 | /* We always need headroom for rndis header */ |
2243 | net->needed_headroom = RNDIS_AND_PPI_SIZE; | |
2244 | ||
6450f8f2 HZ |
2245 | /* Initialize the number of queues to be 1, we may change it if more |
2246 | * channels are offered later. | |
2247 | */ | |
2248 | netif_set_real_num_tx_queues(net, 1); | |
2249 | netif_set_real_num_rx_queues(net, 1); | |
2250 | ||
692e084e | 2251 | /* Notify the netvsc driver of the new device */ |
7c9f335a HZ |
2252 | device_info = netvsc_devinfo_get(NULL); |
2253 | ||
2254 | if (!device_info) { | |
2255 | ret = -ENOMEM; | |
2256 | goto devinfo_failed; | |
2257 | } | |
2258 | ||
2259 | nvdev = rndis_filter_device_add(dev, device_info); | |
9749fed5 | 2260 | if (IS_ERR(nvdev)) { |
2261 | ret = PTR_ERR(nvdev); | |
692e084e | 2262 | netdev_err(net, "unable to add netvsc device (ret %d)\n", ret); |
0c195567 | 2263 | goto rndis_failed; |
df2fff28 | 2264 | } |
0c195567 | 2265 | |
7c9f335a | 2266 | memcpy(net->dev_addr, device_info->mac_adr, ETH_ALEN); |
692e084e | 2267 | |
e04e7a7b DC |
2268 | /* We must get rtnl lock before scheduling nvdev->subchan_work, |
2269 | * otherwise netvsc_subchan_work() can get rtnl lock first and wait | |
2270 | * all subchannels to show up, but that may not happen because | |
2271 | * netvsc_probe() can't get rtnl lock and as a result vmbus_onoffer() | |
2272 | * -> ... -> device_add() -> ... -> __device_attach() can't get | |
2273 | * the device lock, so all the subchannels can't be processed -- | |
52d3b494 | 2274 | * finally netvsc_subchan_work() hangs forever. |
e04e7a7b DC |
2275 | */ |
2276 | rtnl_lock(); | |
2277 | ||
3ffe64f1 SH |
2278 | if (nvdev->num_chn > 1) |
2279 | schedule_work(&nvdev->subchan_work); | |
2280 | ||
aefd80e8 | 2281 | /* hw_features computed in rndis_netdev_set_hwcaps() */ |
23312a3b | 2282 | net->features = net->hw_features | |
2283 | NETIF_F_HIGHDMA | NETIF_F_SG | | |
2284 | NETIF_F_HW_VLAN_CTAG_TX | NETIF_F_HW_VLAN_CTAG_RX; | |
2285 | net->vlan_features = net->features; | |
2286 | ||
9749fed5 | 2287 | netdev_lockdep_set_classes(net); |
2288 | ||
d0c2c997 JW |
2289 | /* MTU range: 68 - 1500 or 65521 */ |
2290 | net->min_mtu = NETVSC_MTU_MIN; | |
2291 | if (nvdev->nvsp_version >= NVSP_PROTOCOL_VERSION_2) | |
2292 | net->max_mtu = NETVSC_MTU - ETH_HLEN; | |
2293 | else | |
2294 | net->max_mtu = ETH_DATA_LEN; | |
2295 | ||
7bf7bb37 | 2296 | ret = register_netdevice(net); |
a68f9614 HZ |
2297 | if (ret != 0) { |
2298 | pr_err("Unable to register netdev.\n"); | |
0c195567 | 2299 | goto register_failed; |
a68f9614 HZ |
2300 | } |
2301 | ||
7bf7bb37 SH |
2302 | list_add(&net_device_ctx->list, &netvsc_dev_list); |
2303 | rtnl_unlock(); | |
7c9f335a HZ |
2304 | |
2305 | kfree(device_info); | |
7bf7bb37 | 2306 | return 0; |
0c195567 | 2307 | |
2308 | register_failed: | |
7bf7bb37 | 2309 | rtnl_unlock(); |
0c195567 | 2310 | rndis_filter_device_remove(dev, nvdev); |
2311 | rndis_failed: | |
7c9f335a HZ |
2312 | kfree(device_info); |
2313 | devinfo_failed: | |
0c195567 | 2314 | free_percpu(net_device_ctx->vf_stats); |
2315 | no_stats: | |
2316 | hv_set_drvdata(dev, NULL); | |
2317 | free_netdev(net); | |
2318 | no_net: | |
2319 | return ret; | |
df2fff28 GKH |
2320 | } |
2321 | ||
415b023a | 2322 | static int netvsc_remove(struct hv_device *dev) |
df2fff28 | 2323 | { |
122a5f64 | 2324 | struct net_device_context *ndev_ctx; |
7b2ee50c SH |
2325 | struct net_device *vf_netdev, *net; |
2326 | struct netvsc_device *nvdev; | |
2ddd5e5f | 2327 | |
3d541ac5 | 2328 | net = hv_get_drvdata(dev); |
df2fff28 | 2329 | if (net == NULL) { |
415b023a | 2330 | dev_err(&dev->device, "No net device to remove\n"); |
df2fff28 GKH |
2331 | return 0; |
2332 | } | |
2333 | ||
122a5f64 | 2334 | ndev_ctx = netdev_priv(net); |
3d541ac5 | 2335 | |
122a5f64 HZ |
2336 | cancel_delayed_work_sync(&ndev_ctx->dwork); |
2337 | ||
018349d7 SH |
2338 | rtnl_lock(); |
2339 | nvdev = rtnl_dereference(ndev_ctx->nvdev); | |
2340 | if (nvdev) | |
7b2ee50c SH |
2341 | cancel_work_sync(&nvdev->subchan_work); |
2342 | ||
df2fff28 GKH |
2343 | /* |
2344 | * Call to the vsc driver to let it know that the device is being | |
a0be450e | 2345 | * removed. Also blocks mtu and channel changes. |
df2fff28 | 2346 | */ |
ec158f77 SH |
2347 | vf_netdev = rtnl_dereference(ndev_ctx->vf_netdev); |
2348 | if (vf_netdev) | |
8cde8f0c | 2349 | netvsc_unregister_vf(vf_netdev); |
ec158f77 | 2350 | |
7b2ee50c SH |
2351 | if (nvdev) |
2352 | rndis_filter_device_remove(dev, nvdev); | |
2353 | ||
8195b139 | 2354 | unregister_netdevice(net); |
7bf7bb37 | 2355 | list_del(&ndev_ctx->list); |
8195b139 | 2356 | |
a0be450e | 2357 | rtnl_unlock(); |
2358 | ||
3d541ac5 VK |
2359 | hv_set_drvdata(dev, NULL); |
2360 | ||
0c195567 | 2361 | free_percpu(ndev_ctx->vf_stats); |
6c80f3fc | 2362 | free_netdev(net); |
df06bcff | 2363 | return 0; |
df2fff28 GKH |
2364 | } |
2365 | ||
345c4cc3 | 2366 | static const struct hv_vmbus_device_id id_table[] = { |
c45cf2d4 | 2367 | /* Network guid */ |
8f505944 | 2368 | { HV_NIC_GUID, }, |
c45cf2d4 | 2369 | { }, |
345c4cc3 S |
2370 | }; |
2371 | ||
2372 | MODULE_DEVICE_TABLE(vmbus, id_table); | |
2373 | ||
f1542a66 | 2374 | /* The one and only one */ |
fde0ef9b | 2375 | static struct hv_driver netvsc_drv = { |
d31b20fc | 2376 | .name = KBUILD_MODNAME, |
345c4cc3 | 2377 | .id_table = id_table, |
fde0ef9b S |
2378 | .probe = netvsc_probe, |
2379 | .remove = netvsc_remove, | |
af0a5646 AV |
2380 | .driver = { |
2381 | .probe_type = PROBE_PREFER_ASYNCHRONOUS, | |
2382 | }, | |
d4890970 | 2383 | }; |
f1542a66 | 2384 | |
8cde8f0c SH |
2385 | /* |
2386 | * On Hyper-V, every VF interface is matched with a corresponding | |
2387 | * synthetic interface. The synthetic interface is presented first | |
2388 | * to the guest. When the corresponding VF instance is registered, | |
2389 | * we will take care of switching the data path. | |
2390 | */ | |
2391 | static int netvsc_netdev_event(struct notifier_block *this, | |
2392 | unsigned long event, void *ptr) | |
2393 | { | |
2394 | struct net_device *event_dev = netdev_notifier_info_to_dev(ptr); | |
2395 | ||
2396 | /* Skip our own events */ | |
2397 | if (event_dev->netdev_ops == &device_ops) | |
2398 | return NOTIFY_DONE; | |
2399 | ||
2400 | /* Avoid non-Ethernet type devices */ | |
2401 | if (event_dev->type != ARPHRD_ETHER) | |
2402 | return NOTIFY_DONE; | |
2403 | ||
2404 | /* Avoid Vlan dev with same MAC registering as VF */ | |
2405 | if (is_vlan_dev(event_dev)) | |
2406 | return NOTIFY_DONE; | |
2407 | ||
2408 | /* Avoid Bonding master dev with same MAC registering as VF */ | |
2409 | if ((event_dev->priv_flags & IFF_BONDING) && | |
2410 | (event_dev->flags & IFF_MASTER)) | |
2411 | return NOTIFY_DONE; | |
2412 | ||
2413 | switch (event) { | |
2414 | case NETDEV_REGISTER: | |
2415 | return netvsc_register_vf(event_dev); | |
2416 | case NETDEV_UNREGISTER: | |
2417 | return netvsc_unregister_vf(event_dev); | |
2418 | case NETDEV_UP: | |
2419 | case NETDEV_DOWN: | |
2420 | return netvsc_vf_changed(event_dev); | |
2421 | default: | |
2422 | return NOTIFY_DONE; | |
2423 | } | |
2424 | } | |
2425 | ||
2426 | static struct notifier_block netvsc_netdev_notifier = { | |
2427 | .notifier_call = netvsc_netdev_event, | |
2428 | }; | |
2429 | ||
a9869c94 | 2430 | static void __exit netvsc_drv_exit(void) |
fceaf24a | 2431 | { |
8cde8f0c | 2432 | unregister_netdevice_notifier(&netvsc_netdev_notifier); |
768fa219 | 2433 | vmbus_driver_unregister(&netvsc_drv); |
fceaf24a HJ |
2434 | } |
2435 | ||
1fde28cf | 2436 | static int __init netvsc_drv_init(void) |
df2fff28 | 2437 | { |
84bf9cef KS |
2438 | int ret; |
2439 | ||
fa85a6c2 HZ |
2440 | if (ring_size < RING_SIZE_MIN) { |
2441 | ring_size = RING_SIZE_MIN; | |
a7f99d0f | 2442 | pr_info("Increased ring_size to %u (min allowed)\n", |
fa85a6c2 HZ |
2443 | ring_size); |
2444 | } | |
a7f99d0f | 2445 | netvsc_ring_bytes = ring_size * PAGE_SIZE; |
84bf9cef | 2446 | |
a7f99d0f | 2447 | ret = vmbus_driver_register(&netvsc_drv); |
84bf9cef KS |
2448 | if (ret) |
2449 | return ret; | |
2450 | ||
8cde8f0c | 2451 | register_netdevice_notifier(&netvsc_netdev_notifier); |
84bf9cef | 2452 | return 0; |
df2fff28 GKH |
2453 | } |
2454 | ||
26c14cc1 | 2455 | MODULE_LICENSE("GPL"); |
7880fc54 | 2456 | MODULE_DESCRIPTION("Microsoft Hyper-V network driver"); |
fceaf24a | 2457 | |
1fde28cf | 2458 | module_init(netvsc_drv_init); |
a9869c94 | 2459 | module_exit(netvsc_drv_exit); |