?  search:
summary | log | graphiclog1 | graphiclog2 | commit | commitdiff | tree | refs | edit | fork
blame | history | raw | HEAD
Merge branch 'perf-urgent-for-linus' of git://git.kernel.org/pub/scm/linux/kernel...
[cris-mirror.git] / drivers / net / hyperv / netvsc_drv.c
blob3ba29fc80d057e744eef4ae98ad60315f742c4ce
1 /*
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
14 * this program; if not, see <http://www.gnu.org/licenses/>.
15 *
16 * Authors:
17 * Haiyang Zhang <haiyangz@microsoft.com>
18 * Hank Janssen <hjanssen@microsoft.com>
19 */
20 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
21
22 #include <linux/init.h>
23 #include <linux/atomic.h>
24 #include <linux/module.h>
25 #include <linux/highmem.h>
26 #include <linux/device.h>
27 #include <linux/io.h>
28 #include <linux/delay.h>
29 #include <linux/netdevice.h>
30 #include <linux/inetdevice.h>
31 #include <linux/etherdevice.h>
32 #include <linux/skbuff.h>
33 #include <linux/if_vlan.h>
34 #include <linux/in.h>
35 #include <linux/slab.h>
36 #include <net/arp.h>
37 #include <net/route.h>
38 #include <net/sock.h>
39 #include <net/pkt_sched.h>
40
41 #include "hyperv_net.h"
42
43
44 #define RING_SIZE_MIN 64
45 #define LINKCHANGE_INT (2 * HZ)
46 #define NETVSC_HW_FEATURES (NETIF_F_RXCSUM | \
47 NETIF_F_SG | \
48 NETIF_F_TSO | \
49 NETIF_F_TSO6 | \
50 NETIF_F_HW_CSUM)
51 static int ring_size = 128;
52 module_param(ring_size, int, S_IRUGO);
53 MODULE_PARM_DESC(ring_size, "Ring buffer size (# of pages)");
54
55 static int max_num_vrss_chns = 8;
56
57 static const u32 default_msg = NETIF_MSG_DRV | NETIF_MSG_PROBE |
58 NETIF_MSG_LINK | NETIF_MSG_IFUP |
59 NETIF_MSG_IFDOWN | NETIF_MSG_RX_ERR |
60 NETIF_MSG_TX_ERR;
61
62 static int debug = -1;
63 module_param(debug, int, S_IRUGO);
64 MODULE_PARM_DESC(debug, "Debug level (0=none,...,16=all)");
65
66 static void do_set_multicast(struct work_struct *w)
67 {
68 struct net_device_context *ndevctx =
69 container_of(w, struct net_device_context, work);
70 struct hv_device *device_obj = ndevctx->device_ctx;
71 struct net_device *ndev = hv_get_drvdata(device_obj);
72 struct netvsc_device *nvdev = ndevctx->nvdev;
73 struct rndis_device *rdev;
74
75 if (!nvdev)
76 return;
77
78 rdev = nvdev->extension;
79 if (rdev == NULL)
80 return;
81
82 if (ndev->flags & IFF_PROMISC)
83 rndis_filter_set_packet_filter(rdev,
84 NDIS_PACKET_TYPE_PROMISCUOUS);
85 else
86 rndis_filter_set_packet_filter(rdev,
87 NDIS_PACKET_TYPE_BROADCAST |
88 NDIS_PACKET_TYPE_ALL_MULTICAST |
89 NDIS_PACKET_TYPE_DIRECTED);
90 }
91
92 static void netvsc_set_multicast_list(struct net_device *net)
93 {
94 struct net_device_context *net_device_ctx = netdev_priv(net);
95
96 schedule_work(&net_device_ctx->work);
97 }
98
99 static int netvsc_open(struct net_device *net)
100 {
101 struct netvsc_device *nvdev = net_device_to_netvsc_device(net);
102 struct rndis_device *rdev;
103 int ret = 0;
104
105 netif_carrier_off(net);
106
107 /* Open up the device */
108 ret = rndis_filter_open(nvdev);
109 if (ret != 0) {
110 netdev_err(net, "unable to open device (ret %d).\n", ret);
111 return ret;
112 }
113
114 netif_tx_wake_all_queues(net);
115
116 rdev = nvdev->extension;
117 if (!rdev->link_state)
118 netif_carrier_on(net);
119
120 return ret;
121 }
122
123 static int netvsc_close(struct net_device *net)
124 {
125 struct net_device_context *net_device_ctx = netdev_priv(net);
126 struct netvsc_device *nvdev = net_device_ctx->nvdev;
127 int ret;
128 u32 aread, awrite, i, msec = 10, retry = 0, retry_max = 20;
129 struct vmbus_channel *chn;
130
131 netif_tx_disable(net);
132
133 /* Make sure netvsc_set_multicast_list doesn't re-enable filter! */
134 cancel_work_sync(&net_device_ctx->work);
135 ret = rndis_filter_close(nvdev);
136 if (ret != 0) {
137 netdev_err(net, "unable to close device (ret %d).\n", ret);
138 return ret;
139 }
140
141 /* Ensure pending bytes in ring are read */
142 while (true) {
143 aread = 0;
144 for (i = 0; i < nvdev->num_chn; i++) {
145 chn = nvdev->chn_table[i];
146 if (!chn)
147 continue;
148
149 hv_get_ringbuffer_availbytes(&chn->inbound, &aread,
150 &awrite);
151
152 if (aread)
153 break;
154
155 hv_get_ringbuffer_availbytes(&chn->outbound, &aread,
156 &awrite);
157
158 if (aread)
159 break;
160 }
161
162 retry++;
163 if (retry > retry_max || aread == 0)
164 break;
165
166 msleep(msec);
167
168 if (msec < 1000)
169 msec *= 2;
170 }
171
172 if (aread) {
173 netdev_err(net, "Ring buffer not empty after closing rndis\n");
174 ret = -ETIMEDOUT;
175 }
176
177 return ret;
178 }
179
180 static void *init_ppi_data(struct rndis_message *msg, u32 ppi_size,
181 int pkt_type)
182 {
183 struct rndis_packet *rndis_pkt;
184 struct rndis_per_packet_info *ppi;
185
186 rndis_pkt = &msg->msg.pkt;
187 rndis_pkt->data_offset += ppi_size;
188
189 ppi = (struct rndis_per_packet_info *)((void *)rndis_pkt +
190 rndis_pkt->per_pkt_info_offset + rndis_pkt->per_pkt_info_len);
191
192 ppi->size = ppi_size;
193 ppi->type = pkt_type;
194 ppi->ppi_offset = sizeof(struct rndis_per_packet_info);
195
196 rndis_pkt->per_pkt_info_len += ppi_size;
197
198 return ppi;
199 }
200
201 static u16 netvsc_select_queue(struct net_device *ndev, struct sk_buff *skb,
202 void *accel_priv, select_queue_fallback_t fallback)
203 {
204 struct net_device_context *net_device_ctx = netdev_priv(ndev);
205 struct netvsc_device *nvsc_dev = net_device_ctx->nvdev;
206 u32 hash;
207 u16 q_idx = 0;
208
209 if (nvsc_dev == NULL || ndev->real_num_tx_queues <= 1)
210 return 0;
211
212 hash = skb_get_hash(skb);
213 q_idx = nvsc_dev->send_table[hash % VRSS_SEND_TAB_SIZE] %
214 ndev->real_num_tx_queues;
215
216 if (!nvsc_dev->chn_table[q_idx])
217 q_idx = 0;
218
219 return q_idx;
220 }
221
222 static u32 fill_pg_buf(struct page *page, u32 offset, u32 len,
223 struct hv_page_buffer *pb)
224 {
225 int j = 0;
226
227 /* Deal with compund pages by ignoring unused part
228 * of the page.
229 */
230 page += (offset >> PAGE_SHIFT);
231 offset &= ~PAGE_MASK;
232
233 while (len > 0) {
234 unsigned long bytes;
235
236 bytes = PAGE_SIZE - offset;
237 if (bytes > len)
238 bytes = len;
239 pb[j].pfn = page_to_pfn(page);
240 pb[j].offset = offset;
241 pb[j].len = bytes;
242
243 offset += bytes;
244 len -= bytes;
245
246 if (offset == PAGE_SIZE && len) {
247 page++;
248 offset = 0;
249 j++;
250 }
251 }
252
253 return j + 1;
254 }
255
256 static u32 init_page_array(void *hdr, u32 len, struct sk_buff *skb,
257 struct hv_netvsc_packet *packet,
258 struct hv_page_buffer **page_buf)
259 {
260 struct hv_page_buffer *pb = *page_buf;
261 u32 slots_used = 0;
262 char *data = skb->data;
263 int frags = skb_shinfo(skb)->nr_frags;
264 int i;
265
266 /* The packet is laid out thus:
267 * 1. hdr: RNDIS header and PPI
268 * 2. skb linear data
269 * 3. skb fragment data
270 */
271 if (hdr != NULL)
272 slots_used += fill_pg_buf(virt_to_page(hdr),
273 offset_in_page(hdr),
274 len, &pb[slots_used]);
275
276 packet->rmsg_size = len;
277 packet->rmsg_pgcnt = slots_used;
278
279 slots_used += fill_pg_buf(virt_to_page(data),
280 offset_in_page(data),
281 skb_headlen(skb), &pb[slots_used]);
282
283 for (i = 0; i < frags; i++) {
284 skb_frag_t *frag = skb_shinfo(skb)->frags + i;
285
286 slots_used += fill_pg_buf(skb_frag_page(frag),
287 frag->page_offset,
288 skb_frag_size(frag), &pb[slots_used]);
289 }
290 return slots_used;
291 }
292
293 static int count_skb_frag_slots(struct sk_buff *skb)
294 {
295 int i, frags = skb_shinfo(skb)->nr_frags;
296 int pages = 0;
297
298 for (i = 0; i < frags; i++) {
299 skb_frag_t *frag = skb_shinfo(skb)->frags + i;
300 unsigned long size = skb_frag_size(frag);
301 unsigned long offset = frag->page_offset;
302
303 /* Skip unused frames from start of page */
304 offset &= ~PAGE_MASK;
305 pages += PFN_UP(offset + size);
306 }
307 return pages;
308 }
309
310 static int netvsc_get_slots(struct sk_buff *skb)
311 {
312 char *data = skb->data;
313 unsigned int offset = offset_in_page(data);
314 unsigned int len = skb_headlen(skb);
315 int slots;
316 int frag_slots;
317
318 slots = DIV_ROUND_UP(offset + len, PAGE_SIZE);
319 frag_slots = count_skb_frag_slots(skb);
320 return slots + frag_slots;
321 }
322
323 static u32 get_net_transport_info(struct sk_buff *skb, u32 *trans_off)
324 {
325 u32 ret_val = TRANSPORT_INFO_NOT_IP;
326
327 if ((eth_hdr(skb)->h_proto != htons(ETH_P_IP)) &&
328 (eth_hdr(skb)->h_proto != htons(ETH_P_IPV6))) {
329 goto not_ip;
330 }
331
332 *trans_off = skb_transport_offset(skb);
333
334 if ((eth_hdr(skb)->h_proto == htons(ETH_P_IP))) {
335 struct iphdr *iphdr = ip_hdr(skb);
336
337 if (iphdr->protocol == IPPROTO_TCP)
338 ret_val = TRANSPORT_INFO_IPV4_TCP;
339 else if (iphdr->protocol == IPPROTO_UDP)
340 ret_val = TRANSPORT_INFO_IPV4_UDP;
341 } else {
342 if (ipv6_hdr(skb)->nexthdr == IPPROTO_TCP)
343 ret_val = TRANSPORT_INFO_IPV6_TCP;
344 else if (ipv6_hdr(skb)->nexthdr == IPPROTO_UDP)
345 ret_val = TRANSPORT_INFO_IPV6_UDP;
346 }
347
348 not_ip:
349 return ret_val;
350 }
351
352 static int netvsc_start_xmit(struct sk_buff *skb, struct net_device *net)
353 {
354 struct net_device_context *net_device_ctx = netdev_priv(net);
355 struct hv_netvsc_packet *packet = NULL;
356 int ret;
357 unsigned int num_data_pgs;
358 struct rndis_message *rndis_msg;
359 struct rndis_packet *rndis_pkt;
360 u32 rndis_msg_size;
361 bool isvlan;
362 bool linear = false;
363 struct rndis_per_packet_info *ppi;
364 struct ndis_tcp_ip_checksum_info *csum_info;
365 struct ndis_tcp_lso_info *lso_info;
366 int hdr_offset;
367 u32 net_trans_info;
368 u32 hash;
369 u32 skb_length;
370 struct hv_page_buffer page_buf[MAX_PAGE_BUFFER_COUNT];
371 struct hv_page_buffer *pb = page_buf;
372 struct netvsc_stats *tx_stats = this_cpu_ptr(net_device_ctx->tx_stats);
373
374 /* We will atmost need two pages to describe the rndis
375 * header. We can only transmit MAX_PAGE_BUFFER_COUNT number
376 * of pages in a single packet. If skb is scattered around
377 * more pages we try linearizing it.
378 */
379
380 check_size:
381 skb_length = skb->len;
382 num_data_pgs = netvsc_get_slots(skb) + 2;
383 if (num_data_pgs > MAX_PAGE_BUFFER_COUNT && linear) {
384 net_alert_ratelimited("packet too big: %u pages (%u bytes)\n",
385 num_data_pgs, skb->len);
386 ret = -EFAULT;
387 goto drop;
388 } else if (num_data_pgs > MAX_PAGE_BUFFER_COUNT) {
389 if (skb_linearize(skb)) {
390 net_alert_ratelimited("failed to linearize skb\n");
391 ret = -ENOMEM;
392 goto drop;
393 }
394 linear = true;
395 goto check_size;
396 }
397
398 /*
399 * Place the rndis header in the skb head room and
400 * the skb->cb will be used for hv_netvsc_packet
401 * structure.
402 */
403 ret = skb_cow_head(skb, RNDIS_AND_PPI_SIZE);
404 if (ret) {
405 netdev_err(net, "unable to alloc hv_netvsc_packet\n");
406 ret = -ENOMEM;
407 goto drop;
408 }
409 /* Use the skb control buffer for building up the packet */
410 BUILD_BUG_ON(sizeof(struct hv_netvsc_packet) >
411 FIELD_SIZEOF(struct sk_buff, cb));
412 packet = (struct hv_netvsc_packet *)skb->cb;
413
414
415 packet->q_idx = skb_get_queue_mapping(skb);
416
417 packet->total_data_buflen = skb->len;
418
419 rndis_msg = (struct rndis_message *)skb->head;
420
421 memset(rndis_msg, 0, RNDIS_AND_PPI_SIZE);
422
423 isvlan = skb->vlan_tci & VLAN_TAG_PRESENT;
424
425 /* Add the rndis header */
426 rndis_msg->ndis_msg_type = RNDIS_MSG_PACKET;
427 rndis_msg->msg_len = packet->total_data_buflen;
428 rndis_pkt = &rndis_msg->msg.pkt;
429 rndis_pkt->data_offset = sizeof(struct rndis_packet);
430 rndis_pkt->data_len = packet->total_data_buflen;
431 rndis_pkt->per_pkt_info_offset = sizeof(struct rndis_packet);
432
433 rndis_msg_size = RNDIS_MESSAGE_SIZE(struct rndis_packet);
434
435 hash = skb_get_hash_raw(skb);
436 if (hash != 0 && net->real_num_tx_queues > 1) {
437 rndis_msg_size += NDIS_HASH_PPI_SIZE;
438 ppi = init_ppi_data(rndis_msg, NDIS_HASH_PPI_SIZE,
439 NBL_HASH_VALUE);
440 *(u32 *)((void *)ppi + ppi->ppi_offset) = hash;
441 }
442
443 if (isvlan) {
444 struct ndis_pkt_8021q_info *vlan;
445
446 rndis_msg_size += NDIS_VLAN_PPI_SIZE;
447 ppi = init_ppi_data(rndis_msg, NDIS_VLAN_PPI_SIZE,
448 IEEE_8021Q_INFO);
449 vlan = (struct ndis_pkt_8021q_info *)((void *)ppi +
450 ppi->ppi_offset);
451 vlan->vlanid = skb->vlan_tci & VLAN_VID_MASK;
452 vlan->pri = (skb->vlan_tci & VLAN_PRIO_MASK) >>
453 VLAN_PRIO_SHIFT;
454 }
455
456 net_trans_info = get_net_transport_info(skb, &hdr_offset);
457 if (net_trans_info == TRANSPORT_INFO_NOT_IP)
458 goto do_send;
459
460 /*
461 * Setup the sendside checksum offload only if this is not a
462 * GSO packet.
463 */
464 if (skb_is_gso(skb))
465 goto do_lso;
466
467 if ((skb->ip_summed == CHECKSUM_NONE) ||
468 (skb->ip_summed == CHECKSUM_UNNECESSARY))
469 goto do_send;
470
471 rndis_msg_size += NDIS_CSUM_PPI_SIZE;
472 ppi = init_ppi_data(rndis_msg, NDIS_CSUM_PPI_SIZE,
473 TCPIP_CHKSUM_PKTINFO);
474
475 csum_info = (struct ndis_tcp_ip_checksum_info *)((void *)ppi +
476 ppi->ppi_offset);
477
478 if (net_trans_info & (INFO_IPV4 << 16))
479 csum_info->transmit.is_ipv4 = 1;
480 else
481 csum_info->transmit.is_ipv6 = 1;
482
483 if (net_trans_info & INFO_TCP) {
484 csum_info->transmit.tcp_checksum = 1;
485 csum_info->transmit.tcp_header_offset = hdr_offset;
486 } else if (net_trans_info & INFO_UDP) {
487 /* UDP checksum offload is not supported on ws2008r2.
488 * Furthermore, on ws2012 and ws2012r2, there are some
489 * issues with udp checksum offload from Linux guests.
490 * (these are host issues).
491 * For now compute the checksum here.
492 */
493 struct udphdr *uh;
494 u16 udp_len;
495
496 ret = skb_cow_head(skb, 0);
497 if (ret)
498 goto drop;
499
500 uh = udp_hdr(skb);
501 udp_len = ntohs(uh->len);
502 uh->check = 0;
503 uh->check = csum_tcpudp_magic(ip_hdr(skb)->saddr,
504 ip_hdr(skb)->daddr,
505 udp_len, IPPROTO_UDP,
506 csum_partial(uh, udp_len, 0));
507 if (uh->check == 0)
508 uh->check = CSUM_MANGLED_0;
509
510 csum_info->transmit.udp_checksum = 0;
511 }
512 goto do_send;
513
514 do_lso:
515 rndis_msg_size += NDIS_LSO_PPI_SIZE;
516 ppi = init_ppi_data(rndis_msg, NDIS_LSO_PPI_SIZE,
517 TCP_LARGESEND_PKTINFO);
518
519 lso_info = (struct ndis_tcp_lso_info *)((void *)ppi +
520 ppi->ppi_offset);
521
522 lso_info->lso_v2_transmit.type = NDIS_TCP_LARGE_SEND_OFFLOAD_V2_TYPE;
523 if (net_trans_info & (INFO_IPV4 << 16)) {
524 lso_info->lso_v2_transmit.ip_version =
525 NDIS_TCP_LARGE_SEND_OFFLOAD_IPV4;
526 ip_hdr(skb)->tot_len = 0;
527 ip_hdr(skb)->check = 0;
528 tcp_hdr(skb)->check =
529 ~csum_tcpudp_magic(ip_hdr(skb)->saddr,
530 ip_hdr(skb)->daddr, 0, IPPROTO_TCP, 0);
531 } else {
532 lso_info->lso_v2_transmit.ip_version =
533 NDIS_TCP_LARGE_SEND_OFFLOAD_IPV6;
534 ipv6_hdr(skb)->payload_len = 0;
535 tcp_hdr(skb)->check =
536 ~csum_ipv6_magic(&ipv6_hdr(skb)->saddr,
537 &ipv6_hdr(skb)->daddr, 0, IPPROTO_TCP, 0);
538 }
539 lso_info->lso_v2_transmit.tcp_header_offset = hdr_offset;
540 lso_info->lso_v2_transmit.mss = skb_shinfo(skb)->gso_size;
541
542 do_send:
543 /* Start filling in the page buffers with the rndis hdr */
544 rndis_msg->msg_len += rndis_msg_size;
545 packet->total_data_buflen = rndis_msg->msg_len;
546 packet->page_buf_cnt = init_page_array(rndis_msg, rndis_msg_size,
547 skb, packet, &pb);
548
549 /* timestamp packet in software */
550 skb_tx_timestamp(skb);
551 ret = netvsc_send(net_device_ctx->device_ctx, packet,
552 rndis_msg, &pb, skb);
553
554 drop:
555 if (ret == 0) {
556 u64_stats_update_begin(&tx_stats->syncp);
557 tx_stats->packets++;
558 tx_stats->bytes += skb_length;
559 u64_stats_update_end(&tx_stats->syncp);
560 } else {
561 if (ret != -EAGAIN) {
562 dev_kfree_skb_any(skb);
563 net->stats.tx_dropped++;
564 }
565 }
566
567 return (ret == -EAGAIN) ? NETDEV_TX_BUSY : NETDEV_TX_OK;
568 }
569
570 /*
571 * netvsc_linkstatus_callback - Link up/down notification
572 */
573 void netvsc_linkstatus_callback(struct hv_device *device_obj,
574 struct rndis_message *resp)
575 {
576 struct rndis_indicate_status *indicate = &resp->msg.indicate_status;
577 struct net_device *net;
578 struct net_device_context *ndev_ctx;
579 struct netvsc_reconfig *event;
580 unsigned long flags;
581
582 /* Handle link change statuses only */
583 if (indicate->status != RNDIS_STATUS_NETWORK_CHANGE &&
584 indicate->status != RNDIS_STATUS_MEDIA_CONNECT &&
585 indicate->status != RNDIS_STATUS_MEDIA_DISCONNECT)
586 return;
587
588 net = hv_get_drvdata(device_obj);
589
590 if (!net || net->reg_state != NETREG_REGISTERED)
591 return;
592
593 ndev_ctx = netdev_priv(net);
594
595 event = kzalloc(sizeof(*event), GFP_ATOMIC);
596 if (!event)
597 return;
598 event->event = indicate->status;
599
600 spin_lock_irqsave(&ndev_ctx->lock, flags);
601 list_add_tail(&event->list, &ndev_ctx->reconfig_events);
602 spin_unlock_irqrestore(&ndev_ctx->lock, flags);
603
604 schedule_delayed_work(&ndev_ctx->dwork, 0);
605 }
606
607
608 static struct sk_buff *netvsc_alloc_recv_skb(struct net_device *net,
609 struct hv_netvsc_packet *packet,
610 struct ndis_tcp_ip_checksum_info *csum_info,
611 void *data, u16 vlan_tci)
612 {
613 struct sk_buff *skb;
614
615 skb = netdev_alloc_skb_ip_align(net, packet->total_data_buflen);
616 if (!skb)
617 return skb;
618
619 /*
620 * Copy to skb. This copy is needed here since the memory pointed by
621 * hv_netvsc_packet cannot be deallocated
622 */
623 memcpy(skb_put(skb, packet->total_data_buflen), data,
624 packet->total_data_buflen);
625
626 skb->protocol = eth_type_trans(skb, net);
627 if (csum_info) {
628 /* We only look at the IP checksum here.
629 * Should we be dropping the packet if checksum
630 * failed? How do we deal with other checksums - TCP/UDP?
631 */
632 if (csum_info->receive.ip_checksum_succeeded)
633 skb->ip_summed = CHECKSUM_UNNECESSARY;
634 else
635 skb->ip_summed = CHECKSUM_NONE;
636 }
637
638 if (vlan_tci & VLAN_TAG_PRESENT)
639 __vlan_hwaccel_put_tag(skb, htons(ETH_P_8021Q),
640 vlan_tci);
641
642 return skb;
643 }
644
645 /*
646 * netvsc_recv_callback - Callback when we receive a packet from the
647 * "wire" on the specified device.
648 */
649 int netvsc_recv_callback(struct hv_device *device_obj,
650 struct hv_netvsc_packet *packet,
651 void **data,
652 struct ndis_tcp_ip_checksum_info *csum_info,
653 struct vmbus_channel *channel,
654 u16 vlan_tci)
655 {
656 struct net_device *net = hv_get_drvdata(device_obj);
657 struct net_device_context *net_device_ctx = netdev_priv(net);
658 struct sk_buff *skb;
659 struct sk_buff *vf_skb;
660 struct netvsc_stats *rx_stats;
661 u32 bytes_recvd = packet->total_data_buflen;
662 int ret = 0;
663
664 if (!net || net->reg_state != NETREG_REGISTERED)
665 return NVSP_STAT_FAIL;
666
667 if (READ_ONCE(net_device_ctx->vf_inject)) {
668 atomic_inc(&net_device_ctx->vf_use_cnt);
669 if (!READ_ONCE(net_device_ctx->vf_inject)) {
670 /*
671 * We raced; just move on.
672 */
673 atomic_dec(&net_device_ctx->vf_use_cnt);
674 goto vf_injection_done;
675 }
676
677 /*
678 * Inject this packet into the VF inerface.
679 * On Hyper-V, multicast and brodcast packets
680 * are only delivered on the synthetic interface
681 * (after subjecting these to policy filters on
682 * the host). Deliver these via the VF interface
683 * in the guest.
684 */
685 vf_skb = netvsc_alloc_recv_skb(net_device_ctx->vf_netdev,
686 packet, csum_info, *data,
687 vlan_tci);
688 if (vf_skb != NULL) {
689 ++net_device_ctx->vf_netdev->stats.rx_packets;
690 net_device_ctx->vf_netdev->stats.rx_bytes +=
691 bytes_recvd;
692 netif_receive_skb(vf_skb);
693 } else {
694 ++net->stats.rx_dropped;
695 ret = NVSP_STAT_FAIL;
696 }
697 atomic_dec(&net_device_ctx->vf_use_cnt);
698 return ret;
699 }
700
701 vf_injection_done:
702 rx_stats = this_cpu_ptr(net_device_ctx->rx_stats);
703
704 /* Allocate a skb - TODO direct I/O to pages? */
705 skb = netvsc_alloc_recv_skb(net, packet, csum_info, *data, vlan_tci);
706 if (unlikely(!skb)) {
707 ++net->stats.rx_dropped;
708 return NVSP_STAT_FAIL;
709 }
710 skb_record_rx_queue(skb, channel->
711 offermsg.offer.sub_channel_index);
712
713 u64_stats_update_begin(&rx_stats->syncp);
714 rx_stats->packets++;
715 rx_stats->bytes += packet->total_data_buflen;
716 u64_stats_update_end(&rx_stats->syncp);
717
718 /*
719 * Pass the skb back up. Network stack will deallocate the skb when it
720 * is done.
721 * TODO - use NAPI?
722 */
723 netif_rx(skb);
724
725 return 0;
726 }
727
728 static void netvsc_get_drvinfo(struct net_device *net,
729 struct ethtool_drvinfo *info)
730 {
731 strlcpy(info->driver, KBUILD_MODNAME, sizeof(info->driver));
732 strlcpy(info->fw_version, "N/A", sizeof(info->fw_version));
733 }
734
735 static void netvsc_get_channels(struct net_device *net,
736 struct ethtool_channels *channel)
737 {
738 struct net_device_context *net_device_ctx = netdev_priv(net);
739 struct netvsc_device *nvdev = net_device_ctx->nvdev;
740
741 if (nvdev) {
742 channel->max_combined = nvdev->max_chn;
743 channel->combined_count = nvdev->num_chn;
744 }
745 }
746
747 static int netvsc_set_channels(struct net_device *net,
748 struct ethtool_channels *channels)
749 {
750 struct net_device_context *net_device_ctx = netdev_priv(net);
751 struct hv_device *dev = net_device_ctx->device_ctx;
752 struct netvsc_device *nvdev = net_device_ctx->nvdev;
753 struct netvsc_device_info device_info;
754 u32 num_chn;
755 u32 max_chn;
756 int ret = 0;
757 bool recovering = false;
758
759 if (net_device_ctx->start_remove || !nvdev || nvdev->destroy)
760 return -ENODEV;
761
762 num_chn = nvdev->num_chn;
763 max_chn = min_t(u32, nvdev->max_chn, num_online_cpus());
764
765 if (nvdev->nvsp_version < NVSP_PROTOCOL_VERSION_5) {
766 pr_info("vRSS unsupported before NVSP Version 5\n");
767 return -EINVAL;
768 }
769
770 /* We do not support rx, tx, or other */
771 if (!channels ||
772 channels->rx_count ||
773 channels->tx_count ||
774 channels->other_count ||
775 (channels->combined_count < 1))
776 return -EINVAL;
777
778 if (channels->combined_count > max_chn) {
779 pr_info("combined channels too high, using %d\n", max_chn);
780 channels->combined_count = max_chn;
781 }
782
783 ret = netvsc_close(net);
784 if (ret)
785 goto out;
786
787 do_set:
788 net_device_ctx->start_remove = true;
789 rndis_filter_device_remove(dev);
790
791 nvdev->num_chn = channels->combined_count;
792
793 memset(&device_info, 0, sizeof(device_info));
794 device_info.num_chn = nvdev->num_chn; /* passed to RNDIS */
795 device_info.ring_size = ring_size;
796 device_info.max_num_vrss_chns = max_num_vrss_chns;
797
798 ret = rndis_filter_device_add(dev, &device_info);
799 if (ret) {
800 if (recovering) {
801 netdev_err(net, "unable to add netvsc device (ret %d)\n", ret);
802 return ret;
803 }
804 goto recover;
805 }
806
807 nvdev = net_device_ctx->nvdev;
808
809 ret = netif_set_real_num_tx_queues(net, nvdev->num_chn);
810 if (ret) {
811 if (recovering) {
812 netdev_err(net, "could not set tx queue count (ret %d)\n", ret);
813 return ret;
814 }
815 goto recover;
816 }
817
818 ret = netif_set_real_num_rx_queues(net, nvdev->num_chn);
819 if (ret) {
820 if (recovering) {
821 netdev_err(net, "could not set rx queue count (ret %d)\n", ret);
822 return ret;
823 }
824 goto recover;
825 }
826
827 out:
828 netvsc_open(net);
829 net_device_ctx->start_remove = false;
830 /* We may have missed link change notifications */
831 schedule_delayed_work(&net_device_ctx->dwork, 0);
832
833 return ret;
834
835 recover:
836 /* If the above failed, we attempt to recover through the same
837 * process but with the original number of channels.
838 */
839 netdev_err(net, "could not set channels, recovering\n");
840 recovering = true;
841 channels->combined_count = num_chn;
842 goto do_set;
843 }
844
845 static bool netvsc_validate_ethtool_ss_cmd(const struct ethtool_cmd *cmd)
846 {
847 struct ethtool_cmd diff1 = *cmd;
848 struct ethtool_cmd diff2 = {};
849
850 ethtool_cmd_speed_set(&diff1, 0);
851 diff1.duplex = 0;
852 /* advertising and cmd are usually set */
853 diff1.advertising = 0;
854 diff1.cmd = 0;
855 /* We set port to PORT_OTHER */
856 diff2.port = PORT_OTHER;
857
858 return !memcmp(&diff1, &diff2, sizeof(diff1));
859 }
860
861 static void netvsc_init_settings(struct net_device *dev)
862 {
863 struct net_device_context *ndc = netdev_priv(dev);
864
865 ndc->speed = SPEED_UNKNOWN;
866 ndc->duplex = DUPLEX_UNKNOWN;
867 }
868
869 static int netvsc_get_settings(struct net_device *dev, struct ethtool_cmd *cmd)
870 {
871 struct net_device_context *ndc = netdev_priv(dev);
872
873 ethtool_cmd_speed_set(cmd, ndc->speed);
874 cmd->duplex = ndc->duplex;
875 cmd->port = PORT_OTHER;
876
877 return 0;
878 }
879
880 static int netvsc_set_settings(struct net_device *dev, struct ethtool_cmd *cmd)
881 {
882 struct net_device_context *ndc = netdev_priv(dev);
883 u32 speed;
884
885 speed = ethtool_cmd_speed(cmd);
886 if (!ethtool_validate_speed(speed) ||
887 !ethtool_validate_duplex(cmd->duplex) ||
888 !netvsc_validate_ethtool_ss_cmd(cmd))
889 return -EINVAL;
890
891 ndc->speed = speed;
892 ndc->duplex = cmd->duplex;
893
894 return 0;
895 }
896
897 static int netvsc_change_mtu(struct net_device *ndev, int mtu)
898 {
899 struct net_device_context *ndevctx = netdev_priv(ndev);
900 struct netvsc_device *nvdev = ndevctx->nvdev;
901 struct hv_device *hdev = ndevctx->device_ctx;
902 struct netvsc_device_info device_info;
903 int limit = ETH_DATA_LEN;
904 u32 num_chn;
905 int ret = 0;
906
907 if (ndevctx->start_remove || !nvdev || nvdev->destroy)
908 return -ENODEV;
909
910 if (nvdev->nvsp_version >= NVSP_PROTOCOL_VERSION_2)
911 limit = NETVSC_MTU - ETH_HLEN;
912
913 if (mtu < NETVSC_MTU_MIN || mtu > limit)
914 return -EINVAL;
915
916 ret = netvsc_close(ndev);
917 if (ret)
918 goto out;
919
920 num_chn = nvdev->num_chn;
921
922 ndevctx->start_remove = true;
923 rndis_filter_device_remove(hdev);
924
925 ndev->mtu = mtu;
926
927 memset(&device_info, 0, sizeof(device_info));
928 device_info.ring_size = ring_size;
929 device_info.num_chn = num_chn;
930 device_info.max_num_vrss_chns = max_num_vrss_chns;
931 rndis_filter_device_add(hdev, &device_info);
932
933 out:
934 netvsc_open(ndev);
935 ndevctx->start_remove = false;
936
937 /* We may have missed link change notifications */
938 schedule_delayed_work(&ndevctx->dwork, 0);
939
940 return ret;
941 }
942
943 static struct rtnl_link_stats64 *netvsc_get_stats64(struct net_device *net,
944 struct rtnl_link_stats64 *t)
945 {
946 struct net_device_context *ndev_ctx = netdev_priv(net);
947 int cpu;
948
949 for_each_possible_cpu(cpu) {
950 struct netvsc_stats *tx_stats = per_cpu_ptr(ndev_ctx->tx_stats,
951 cpu);
952 struct netvsc_stats *rx_stats = per_cpu_ptr(ndev_ctx->rx_stats,
953 cpu);
954 u64 tx_packets, tx_bytes, rx_packets, rx_bytes;
955 unsigned int start;
956
957 do {
958 start = u64_stats_fetch_begin_irq(&tx_stats->syncp);
959 tx_packets = tx_stats->packets;
960 tx_bytes = tx_stats->bytes;
961 } while (u64_stats_fetch_retry_irq(&tx_stats->syncp, start));
962
963 do {
964 start = u64_stats_fetch_begin_irq(&rx_stats->syncp);
965 rx_packets = rx_stats->packets;
966 rx_bytes = rx_stats->bytes;
967 } while (u64_stats_fetch_retry_irq(&rx_stats->syncp, start));
968
969 t->tx_bytes += tx_bytes;
970 t->tx_packets += tx_packets;
971 t->rx_bytes += rx_bytes;
972 t->rx_packets += rx_packets;
973 }
974
975 t->tx_dropped = net->stats.tx_dropped;
976 t->tx_errors = net->stats.tx_dropped;
977
978 t->rx_dropped = net->stats.rx_dropped;
979 t->rx_errors = net->stats.rx_errors;
980
981 return t;
982 }
983
984 static int netvsc_set_mac_addr(struct net_device *ndev, void *p)
985 {
986 struct sockaddr *addr = p;
987 char save_adr[ETH_ALEN];
988 unsigned char save_aatype;
989 int err;
990
991 memcpy(save_adr, ndev->dev_addr, ETH_ALEN);
992 save_aatype = ndev->addr_assign_type;
993
994 err = eth_mac_addr(ndev, p);
995 if (err != 0)
996 return err;
997
998 err = rndis_filter_set_device_mac(ndev, addr->sa_data);
999 if (err != 0) {
1000 /* roll back to saved MAC */
1001 memcpy(ndev->dev_addr, save_adr, ETH_ALEN);
1002 ndev->addr_assign_type = save_aatype;
1003 }
1004
1005 return err;
1006 }
1007
1008 #ifdef CONFIG_NET_POLL_CONTROLLER
1009 static void netvsc_poll_controller(struct net_device *net)
1010 {
1011 /* As netvsc_start_xmit() works synchronous we don't have to
1012 * trigger anything here.
1013 */
1014 }
1015 #endif
1016
1017 static const struct ethtool_ops ethtool_ops = {
1018 .get_drvinfo = netvsc_get_drvinfo,
1019 .get_link = ethtool_op_get_link,
1020 .get_channels = netvsc_get_channels,
1021 .set_channels = netvsc_set_channels,
1022 .get_ts_info = ethtool_op_get_ts_info,
1023 .get_settings = netvsc_get_settings,
1024 .set_settings = netvsc_set_settings,
1025 };
1026
1027 static const struct net_device_ops device_ops = {
1028 .ndo_open = netvsc_open,
1029 .ndo_stop = netvsc_close,
1030 .ndo_start_xmit = netvsc_start_xmit,
1031 .ndo_set_rx_mode = netvsc_set_multicast_list,
1032 .ndo_change_mtu = netvsc_change_mtu,
1033 .ndo_validate_addr = eth_validate_addr,
1034 .ndo_set_mac_address = netvsc_set_mac_addr,
1035 .ndo_select_queue = netvsc_select_queue,
1036 .ndo_get_stats64 = netvsc_get_stats64,
1037 #ifdef CONFIG_NET_POLL_CONTROLLER
1038 .ndo_poll_controller = netvsc_poll_controller,
1039 #endif
1040 };
1041
1042 /*
1043 * Handle link status changes. For RNDIS_STATUS_NETWORK_CHANGE emulate link
1044 * down/up sequence. In case of RNDIS_STATUS_MEDIA_CONNECT when carrier is
1045 * present send GARP packet to network peers with netif_notify_peers().
1046 */
1047 static void netvsc_link_change(struct work_struct *w)
1048 {
1049 struct net_device_context *ndev_ctx =
1050 container_of(w, struct net_device_context, dwork.work);
1051 struct hv_device *device_obj = ndev_ctx->device_ctx;
1052 struct net_device *net = hv_get_drvdata(device_obj);
1053 struct netvsc_device *net_device;
1054 struct rndis_device *rdev;
1055 struct netvsc_reconfig *event = NULL;
1056 bool notify = false, reschedule = false;
1057 unsigned long flags, next_reconfig, delay;
1058
1059 rtnl_lock();
1060 if (ndev_ctx->start_remove)
1061 goto out_unlock;
1062
1063 net_device = ndev_ctx->nvdev;
1064 rdev = net_device->extension;
1065
1066 next_reconfig = ndev_ctx->last_reconfig + LINKCHANGE_INT;
1067 if (time_is_after_jiffies(next_reconfig)) {
1068 /* link_watch only sends one notification with current state
1069 * per second, avoid doing reconfig more frequently. Handle
1070 * wrap around.
1071 */
1072 delay = next_reconfig - jiffies;
1073 delay = delay < LINKCHANGE_INT ? delay : LINKCHANGE_INT;
1074 schedule_delayed_work(&ndev_ctx->dwork, delay);
1075 goto out_unlock;
1076 }
1077 ndev_ctx->last_reconfig = jiffies;
1078
1079 spin_lock_irqsave(&ndev_ctx->lock, flags);
1080 if (!list_empty(&ndev_ctx->reconfig_events)) {
1081 event = list_first_entry(&ndev_ctx->reconfig_events,
1082 struct netvsc_reconfig, list);
1083 list_del(&event->list);
1084 reschedule = !list_empty(&ndev_ctx->reconfig_events);
1085 }
1086 spin_unlock_irqrestore(&ndev_ctx->lock, flags);
1087
1088 if (!event)
1089 goto out_unlock;
1090
1091 switch (event->event) {
1092 /* Only the following events are possible due to the check in
1093 * netvsc_linkstatus_callback()
1094 */
1095 case RNDIS_STATUS_MEDIA_CONNECT:
1096 if (rdev->link_state) {
1097 rdev->link_state = false;
1098 netif_carrier_on(net);
1099 netif_tx_wake_all_queues(net);
1100 } else {
1101 notify = true;
1102 }
1103 kfree(event);
1104 break;
1105 case RNDIS_STATUS_MEDIA_DISCONNECT:
1106 if (!rdev->link_state) {
1107 rdev->link_state = true;
1108 netif_carrier_off(net);
1109 netif_tx_stop_all_queues(net);
1110 }
1111 kfree(event);
1112 break;
1113 case RNDIS_STATUS_NETWORK_CHANGE:
1114 /* Only makes sense if carrier is present */
1115 if (!rdev->link_state) {
1116 rdev->link_state = true;
1117 netif_carrier_off(net);
1118 netif_tx_stop_all_queues(net);
1119 event->event = RNDIS_STATUS_MEDIA_CONNECT;
1120 spin_lock_irqsave(&ndev_ctx->lock, flags);
1121 list_add(&event->list, &ndev_ctx->reconfig_events);
1122 spin_unlock_irqrestore(&ndev_ctx->lock, flags);
1123 reschedule = true;
1124 }
1125 break;
1126 }
1127
1128 rtnl_unlock();
1129
1130 if (notify)
1131 netdev_notify_peers(net);
1132
1133 /* link_watch only sends one notification with current state per
1134 * second, handle next reconfig event in 2 seconds.
1135 */
1136 if (reschedule)
1137 schedule_delayed_work(&ndev_ctx->dwork, LINKCHANGE_INT);
1138
1139 return;
1140
1141 out_unlock:
1142 rtnl_unlock();
1143 }
1144
1145 static void netvsc_free_netdev(struct net_device *netdev)
1146 {
1147 struct net_device_context *net_device_ctx = netdev_priv(netdev);
1148
1149 free_percpu(net_device_ctx->tx_stats);
1150 free_percpu(net_device_ctx->rx_stats);
1151 free_netdev(netdev);
1152 }
1153
1154 static struct net_device *get_netvsc_net_device(char *mac)
1155 {
1156 struct net_device *dev, *found = NULL;
1157 int rtnl_locked;
1158
1159 rtnl_locked = rtnl_trylock();
1160
1161 for_each_netdev(&init_net, dev) {
1162 if (memcmp(dev->dev_addr, mac, ETH_ALEN) == 0) {
1163 if (dev->netdev_ops != &device_ops)
1164 continue;
1165 found = dev;
1166 break;
1167 }
1168 }
1169 if (rtnl_locked)
1170 rtnl_unlock();
1171
1172 return found;
1173 }
1174
1175 static int netvsc_register_vf(struct net_device *vf_netdev)
1176 {
1177 struct net_device *ndev;
1178 struct net_device_context *net_device_ctx;
1179 struct netvsc_device *netvsc_dev;
1180 const struct ethtool_ops *eth_ops = vf_netdev->ethtool_ops;
1181
1182 if (eth_ops == NULL || eth_ops == &ethtool_ops)
1183 return NOTIFY_DONE;
1184
1185 /*
1186 * We will use the MAC address to locate the synthetic interface to
1187 * associate with the VF interface. If we don't find a matching
1188 * synthetic interface, move on.
1189 */
1190 ndev = get_netvsc_net_device(vf_netdev->dev_addr);
1191 if (!ndev)
1192 return NOTIFY_DONE;
1193
1194 net_device_ctx = netdev_priv(ndev);
1195 netvsc_dev = net_device_ctx->nvdev;
1196 if (!netvsc_dev || net_device_ctx->vf_netdev)
1197 return NOTIFY_DONE;
1198
1199 netdev_info(ndev, "VF registering: %s\n", vf_netdev->name);
1200 /*
1201 * Take a reference on the module.
1202 */
1203 try_module_get(THIS_MODULE);
1204 net_device_ctx->vf_netdev = vf_netdev;
1205 return NOTIFY_OK;
1206 }
1207
1208 static void netvsc_inject_enable(struct net_device_context *net_device_ctx)
1209 {
1210 net_device_ctx->vf_inject = true;
1211 }
1212
1213 static void netvsc_inject_disable(struct net_device_context *net_device_ctx)
1214 {
1215 net_device_ctx->vf_inject = false;
1216
1217 /* Wait for currently active users to drain out. */
1218 while (atomic_read(&net_device_ctx->vf_use_cnt) != 0)
1219 udelay(50);
1220 }
1221
1222 static int netvsc_vf_up(struct net_device *vf_netdev)
1223 {
1224 struct net_device *ndev;
1225 struct netvsc_device *netvsc_dev;
1226 const struct ethtool_ops *eth_ops = vf_netdev->ethtool_ops;
1227 struct net_device_context *net_device_ctx;
1228
1229 if (eth_ops == &ethtool_ops)
1230 return NOTIFY_DONE;
1231
1232 ndev = get_netvsc_net_device(vf_netdev->dev_addr);
1233 if (!ndev)
1234 return NOTIFY_DONE;
1235
1236 net_device_ctx = netdev_priv(ndev);
1237 netvsc_dev = net_device_ctx->nvdev;
1238
1239 if (!netvsc_dev || !net_device_ctx->vf_netdev)
1240 return NOTIFY_DONE;
1241
1242 netdev_info(ndev, "VF up: %s\n", vf_netdev->name);
1243 netvsc_inject_enable(net_device_ctx);
1244
1245 /*
1246 * Open the device before switching data path.
1247 */
1248 rndis_filter_open(netvsc_dev);
1249
1250 /*
1251 * notify the host to switch the data path.
1252 */
1253 netvsc_switch_datapath(ndev, true);
1254 netdev_info(ndev, "Data path switched to VF: %s\n", vf_netdev->name);
1255
1256 netif_carrier_off(ndev);
1257
1258 /* Now notify peers through VF device. */
1259 call_netdevice_notifiers(NETDEV_NOTIFY_PEERS, vf_netdev);
1260
1261 return NOTIFY_OK;
1262 }
1263
1264
1265 static int netvsc_vf_down(struct net_device *vf_netdev)
1266 {
1267 struct net_device *ndev;
1268 struct netvsc_device *netvsc_dev;
1269 struct net_device_context *net_device_ctx;
1270 const struct ethtool_ops *eth_ops = vf_netdev->ethtool_ops;
1271
1272 if (eth_ops == &ethtool_ops)
1273 return NOTIFY_DONE;
1274
1275 ndev = get_netvsc_net_device(vf_netdev->dev_addr);
1276 if (!ndev)
1277 return NOTIFY_DONE;
1278
1279 net_device_ctx = netdev_priv(ndev);
1280 netvsc_dev = net_device_ctx->nvdev;
1281
1282 if (!netvsc_dev || !net_device_ctx->vf_netdev)
1283 return NOTIFY_DONE;
1284
1285 netdev_info(ndev, "VF down: %s\n", vf_netdev->name);
1286 netvsc_inject_disable(net_device_ctx);
1287 netvsc_switch_datapath(ndev, false);
1288 netdev_info(ndev, "Data path switched from VF: %s\n", vf_netdev->name);
1289 rndis_filter_close(netvsc_dev);
1290 netif_carrier_on(ndev);
1291
1292 /* Now notify peers through netvsc device. */
1293 call_netdevice_notifiers(NETDEV_NOTIFY_PEERS, ndev);
1294
1295 return NOTIFY_OK;
1296 }
1297
1298
1299 static int netvsc_unregister_vf(struct net_device *vf_netdev)
1300 {
1301 struct net_device *ndev;
1302 struct netvsc_device *netvsc_dev;
1303 const struct ethtool_ops *eth_ops = vf_netdev->ethtool_ops;
1304 struct net_device_context *net_device_ctx;
1305
1306 if (eth_ops == &ethtool_ops)
1307 return NOTIFY_DONE;
1308
1309 ndev = get_netvsc_net_device(vf_netdev->dev_addr);
1310 if (!ndev)
1311 return NOTIFY_DONE;
1312
1313 net_device_ctx = netdev_priv(ndev);
1314 netvsc_dev = net_device_ctx->nvdev;
1315 if (!netvsc_dev || !net_device_ctx->vf_netdev)
1316 return NOTIFY_DONE;
1317 netdev_info(ndev, "VF unregistering: %s\n", vf_netdev->name);
1318 netvsc_inject_disable(net_device_ctx);
1319 net_device_ctx->vf_netdev = NULL;
1320 module_put(THIS_MODULE);
1321 return NOTIFY_OK;
1322 }
1323
1324 static int netvsc_probe(struct hv_device *dev,
1325 const struct hv_vmbus_device_id *dev_id)
1326 {
1327 struct net_device *net = NULL;
1328 struct net_device_context *net_device_ctx;
1329 struct netvsc_device_info device_info;
1330 struct netvsc_device *nvdev;
1331 int ret;
1332
1333 net = alloc_etherdev_mq(sizeof(struct net_device_context),
1334 num_online_cpus());
1335 if (!net)
1336 return -ENOMEM;
1337
1338 netif_carrier_off(net);
1339
1340 net_device_ctx = netdev_priv(net);
1341 net_device_ctx->device_ctx = dev;
1342 net_device_ctx->msg_enable = netif_msg_init(debug, default_msg);
1343 if (netif_msg_probe(net_device_ctx))
1344 netdev_dbg(net, "netvsc msg_enable: %d\n",
1345 net_device_ctx->msg_enable);
1346
1347 net_device_ctx->tx_stats = netdev_alloc_pcpu_stats(struct netvsc_stats);
1348 if (!net_device_ctx->tx_stats) {
1349 free_netdev(net);
1350 return -ENOMEM;
1351 }
1352 net_device_ctx->rx_stats = netdev_alloc_pcpu_stats(struct netvsc_stats);
1353 if (!net_device_ctx->rx_stats) {
1354 free_percpu(net_device_ctx->tx_stats);
1355 free_netdev(net);
1356 return -ENOMEM;
1357 }
1358
1359 hv_set_drvdata(dev, net);
1360
1361 net_device_ctx->start_remove = false;
1362
1363 INIT_DELAYED_WORK(&net_device_ctx->dwork, netvsc_link_change);
1364 INIT_WORK(&net_device_ctx->work, do_set_multicast);
1365
1366 spin_lock_init(&net_device_ctx->lock);
1367 INIT_LIST_HEAD(&net_device_ctx->reconfig_events);
1368
1369 atomic_set(&net_device_ctx->vf_use_cnt, 0);
1370 net_device_ctx->vf_netdev = NULL;
1371 net_device_ctx->vf_inject = false;
1372
1373 net->netdev_ops = &device_ops;
1374
1375 net->hw_features = NETVSC_HW_FEATURES;
1376 net->features = NETVSC_HW_FEATURES | NETIF_F_HW_VLAN_CTAG_TX;
1377
1378 net->ethtool_ops = &ethtool_ops;
1379 SET_NETDEV_DEV(net, &dev->device);
1380
1381 /* We always need headroom for rndis header */
1382 net->needed_headroom = RNDIS_AND_PPI_SIZE;
1383
1384 /* Notify the netvsc driver of the new device */
1385 memset(&device_info, 0, sizeof(device_info));
1386 device_info.ring_size = ring_size;
1387 device_info.max_num_vrss_chns = max_num_vrss_chns;
1388 ret = rndis_filter_device_add(dev, &device_info);
1389 if (ret != 0) {
1390 netdev_err(net, "unable to add netvsc device (ret %d)\n", ret);
1391 netvsc_free_netdev(net);
1392 hv_set_drvdata(dev, NULL);
1393 return ret;
1394 }
1395 memcpy(net->dev_addr, device_info.mac_adr, ETH_ALEN);
1396
1397 nvdev = net_device_ctx->nvdev;
1398 netif_set_real_num_tx_queues(net, nvdev->num_chn);
1399 netif_set_real_num_rx_queues(net, nvdev->num_chn);
1400
1401 netvsc_init_settings(net);
1402
1403 ret = register_netdev(net);
1404 if (ret != 0) {
1405 pr_err("Unable to register netdev.\n");
1406 rndis_filter_device_remove(dev);
1407 netvsc_free_netdev(net);
1408 }
1409
1410 return ret;
1411 }
1412
1413 static int netvsc_remove(struct hv_device *dev)
1414 {
1415 struct net_device *net;
1416 struct net_device_context *ndev_ctx;
1417 struct netvsc_device *net_device;
1418
1419 net = hv_get_drvdata(dev);
1420
1421 if (net == NULL) {
1422 dev_err(&dev->device, "No net device to remove\n");
1423 return 0;
1424 }
1425
1426
1427 ndev_ctx = netdev_priv(net);
1428 net_device = ndev_ctx->nvdev;
1429
1430 /* Avoid racing with netvsc_change_mtu()/netvsc_set_channels()
1431 * removing the device.
1432 */
1433 rtnl_lock();
1434 ndev_ctx->start_remove = true;
1435 rtnl_unlock();
1436
1437 cancel_delayed_work_sync(&ndev_ctx->dwork);
1438 cancel_work_sync(&ndev_ctx->work);
1439
1440 /* Stop outbound asap */
1441 netif_tx_disable(net);
1442
1443 unregister_netdev(net);
1444
1445 /*
1446 * Call to the vsc driver to let it know that the device is being
1447 * removed
1448 */
1449 rndis_filter_device_remove(dev);
1450
1451 hv_set_drvdata(dev, NULL);
1452
1453 netvsc_free_netdev(net);
1454 return 0;
1455 }
1456
1457 static const struct hv_vmbus_device_id id_table[] = {
1458 /* Network guid */
1459 { HV_NIC_GUID, },
1460 { },
1461 };
1462
1463 MODULE_DEVICE_TABLE(vmbus, id_table);
1464
1465 /* The one and only one */
1466 static struct hv_driver netvsc_drv = {
1467 .name = KBUILD_MODNAME,
1468 .id_table = id_table,
1469 .probe = netvsc_probe,
1470 .remove = netvsc_remove,
1471 };
1472
1473
1474 /*
1475 * On Hyper-V, every VF interface is matched with a corresponding
1476 * synthetic interface. The synthetic interface is presented first
1477 * to the guest. When the corresponding VF instance is registered,
1478 * we will take care of switching the data path.
1479 */
1480 static int netvsc_netdev_event(struct notifier_block *this,
1481 unsigned long event, void *ptr)
1482 {
1483 struct net_device *event_dev = netdev_notifier_info_to_dev(ptr);
1484
1485 /* Avoid Vlan dev with same MAC registering as VF */
1486 if (event_dev->priv_flags & IFF_802_1Q_VLAN)
1487 return NOTIFY_DONE;
1488
1489 /* Avoid Bonding master dev with same MAC registering as VF */
1490 if (event_dev->priv_flags & IFF_BONDING &&
1491 event_dev->flags & IFF_MASTER)
1492 return NOTIFY_DONE;
1493
1494 switch (event) {
1495 case NETDEV_REGISTER:
1496 return netvsc_register_vf(event_dev);
1497 case NETDEV_UNREGISTER:
1498 return netvsc_unregister_vf(event_dev);
1499 case NETDEV_UP:
1500 return netvsc_vf_up(event_dev);
1501 case NETDEV_DOWN:
1502 return netvsc_vf_down(event_dev);
1503 default:
1504 return NOTIFY_DONE;
1505 }
1506 }
1507
1508 static struct notifier_block netvsc_netdev_notifier = {
1509 .notifier_call = netvsc_netdev_event,
1510 };
1511
1512 static void __exit netvsc_drv_exit(void)
1513 {
1514 unregister_netdevice_notifier(&netvsc_netdev_notifier);
1515 vmbus_driver_unregister(&netvsc_drv);
1516 }
1517
1518 static int __init netvsc_drv_init(void)
1519 {
1520 int ret;
1521
1522 if (ring_size < RING_SIZE_MIN) {
1523 ring_size = RING_SIZE_MIN;
1524 pr_info("Increased ring_size to %d (min allowed)\n",
1525 ring_size);
1526 }
1527 ret = vmbus_driver_register(&netvsc_drv);
1528
1529 if (ret)
1530 return ret;
1531
1532 register_netdevice_notifier(&netvsc_netdev_notifier);
1533 return 0;
1534 }
1535
1536 MODULE_LICENSE("GPL");
1537 MODULE_DESCRIPTION("Microsoft Hyper-V network driver");
1538
1539 module_init(netvsc_drv_init);
1540 module_exit(netvsc_drv_exit);
CRIS linux kernel tree