]> git.proxmox.com Git - mirror_ubuntu-eoan-kernel.git/blame - drivers/net/hyperv/netvsc_drv.c
projects / mirror_ubuntu-eoan-kernel.git / blame
? search:
summary | shortlog | log | commit | commitdiff | tree
blob | blame (incremental) | history | HEAD
UBUNTU: Ubuntu-5.3.0-29.31
[mirror_ubuntu-eoan-kernel.git] / drivers / net / hyperv / netvsc_drv.c
CommitLineData
9952f691 1// SPDX-License-Identifier: GPL-2.0-only
fceaf24a 2/*
fceaf24a
HJ		 3 * Copyright (c) 2009, Microsoft Corporation.
4 *
fceaf24a 5 * Authors:
d0e94d17 6 * Haiyang Zhang <haiyangz@microsoft.com>
fceaf24a 7 * Hank Janssen <hjanssen@microsoft.com>
fceaf24a 8 */
eb335bc4
HJ		 9#define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
10
fceaf24a 11#include <linux/init.h>
9079ce69 12#include <linux/atomic.h>
fceaf24a
HJ		 13#include <linux/module.h>
14#include <linux/highmem.h>
15#include <linux/device.h>
fceaf24a 16#include <linux/io.h>
fceaf24a
HJ		 17#include <linux/delay.h>
18#include <linux/netdevice.h>
19#include <linux/inetdevice.h>
20#include <linux/etherdevice.h>
b93c1b5a 21#include <linux/pci.h>
fceaf24a 22#include <linux/skbuff.h>
c802db11 23#include <linux/if_vlan.h>
fceaf24a 24#include <linux/in.h>
5a0e3ad6 25#include <linux/slab.h>
27f5aa92 26#include <linux/rtnetlink.h>
0c195567 27#include <linux/netpoll.h>
27f5aa92 28
fceaf24a
HJ		 29#include <net/arp.h>
30#include <net/route.h>
31#include <net/sock.h>
32#include <net/pkt_sched.h>
8eb1b3c3
MK		 33#include <net/checksum.h>
34#include <net/ip6_checksum.h>
3f335ea2 35
5ca7252a 36#include "hyperv_net.h"
fceaf24a 37
7b2ee50c
SH		 38#define RING_SIZE_MIN 64
39#define RETRY_US_LO 5000
40#define RETRY_US_HI 10000
41#define RETRY_MAX 2000 /* >10 sec */
8b532797 42
27a70af3 43#define LINKCHANGE_INT (2 * HZ)
6123c668 44#define VF_TAKEOVER_INT (HZ / 10)
a50af86d 45
a7f99d0f 46static unsigned int ring_size __ro_after_init = 128;
d61e4038 47module_param(ring_size, uint, 0444);
450d7a4b 48MODULE_PARM_DESC(ring_size, "Ring buffer size (# of pages)");
a7f99d0f 49unsigned int netvsc_ring_bytes __ro_after_init;
fceaf24a 50
3f300ff4
SX		 51static const u32 default_msg = NETIF_MSG_DRV | NETIF_MSG_PROBE |
52 NETIF_MSG_LINK | NETIF_MSG_IFUP |
53 NETIF_MSG_IFDOWN | NETIF_MSG_RX_ERR |
54 NETIF_MSG_TX_ERR;
55
56static int debug = -1;
d61e4038 57module_param(debug, int, 0444);
3f300ff4
SX		 58MODULE_PARM_DESC(debug, "Debug level (0=none,...,16=all)");
59
7bf7bb37
SH		 60static LIST_HEAD(netvsc_dev_list);
61
bee9d41b 62static void netvsc_change_rx_flags(struct net_device *net, int change)
fceaf24a 63{
bee9d41b
SH		 64 struct net_device_context *ndev_ctx = netdev_priv(net);
65 struct net_device *vf_netdev = rtnl_dereference(ndev_ctx->vf_netdev);
66 int inc;
67
68 if (!vf_netdev)
69 return;
70
71 if (change & IFF_PROMISC) {
72 inc = (net->flags & IFF_PROMISC) ? 1 : -1;
73 dev_set_promiscuity(vf_netdev, inc);
74 }
75
76 if (change & IFF_ALLMULTI) {
77 inc = (net->flags & IFF_ALLMULTI) ? 1 : -1;
78 dev_set_allmulti(vf_netdev, inc);
79 }
80}
81
82static void netvsc_set_rx_mode(struct net_device *net)
83{
84 struct net_device_context *ndev_ctx = netdev_priv(net);
35a57b7f
SH		 85 struct net_device *vf_netdev;
86 struct netvsc_device *nvdev;
bee9d41b 87
35a57b7f
SH		 88 rcu_read_lock();
89 vf_netdev = rcu_dereference(ndev_ctx->vf_netdev);
bee9d41b
SH		 90 if (vf_netdev) {
91 dev_uc_sync(vf_netdev, net);
92 dev_mc_sync(vf_netdev, net);
93 }
d426b2e3 94
35a57b7f
SH		 95 nvdev = rcu_dereference(ndev_ctx->nvdev);
96 if (nvdev)
97 rndis_filter_update(nvdev);
98 rcu_read_unlock();
fceaf24a
HJ		 99}
100
1b704c4a
HZ		 101static void netvsc_tx_enable(struct netvsc_device *nvscdev,
102 struct net_device *ndev)
103{
104 nvscdev->tx_disable = false;
105 virt_wmb(); /* ensure queue wake up mechanism is on */
106
107 netif_tx_wake_all_queues(ndev);
108}
109
fceaf24a
HJ		 110static int netvsc_open(struct net_device *net)
111{
53fa1a6f 112 struct net_device_context *ndev_ctx = netdev_priv(net);
0c195567 113 struct net_device *vf_netdev = rtnl_dereference(ndev_ctx->vf_netdev);
79e8cbe7 114 struct netvsc_device *nvdev = rtnl_dereference(ndev_ctx->nvdev);
891de74d 115 struct rndis_device *rdev;
02fafbc6 116 int ret = 0;
fceaf24a 117
891de74d
HZ		 118 netif_carrier_off(net);
119
d515d0ff 120 /* Open up the device */
2f5fa6c8 121 ret = rndis_filter_open(nvdev);
d515d0ff
HZ		 122 if (ret != 0) {
123 netdev_err(net, "unable to open device (ret %d).\n", ret);
124 return ret;
fceaf24a
HJ		 125 }
126
891de74d 127 rdev = nvdev->extension;
52acf73b 128 if (!rdev->link_state) {
891de74d 129 netif_carrier_on(net);
1b704c4a 130 netvsc_tx_enable(nvdev, net);
52acf73b 131 }
891de74d 132
0c195567 133 if (vf_netdev) {
134 /* Setting synthetic device up transparently sets
135 * slave as up. If open fails, then slave will be
136 * still be offline (and not used).
137 */
00f54e68 138 ret = dev_open(vf_netdev, NULL);
0c195567 139 if (ret)
140 netdev_warn(net,
141 "unable to open slave: %s: %d\n",
142 vf_netdev->name, ret);
143 }
144 return 0;
fceaf24a
HJ		 145}
146
7b2ee50c 147static int netvsc_wait_until_empty(struct netvsc_device *nvdev)
fceaf24a 148{
7b2ee50c
SH		 149 unsigned int retry = 0;
150 int i;
2de8530b
HZ		 151
152 /* Ensure pending bytes in ring are read */
7b2ee50c
SH		 153 for (;;) {
154 u32 aread = 0;
155
2de8530b 156 for (i = 0; i < nvdev->num_chn; i++) {
7b2ee50c
SH		 157 struct vmbus_channel *chn
158 = nvdev->chan_table[i].channel;
159
2de8530b
HZ		 160 if (!chn)
161 continue;
162
7b2ee50c
SH		 163 /* make sure receive not running now */
164 napi_synchronize(&nvdev->chan_table[i].napi);
165
40975962 166 aread = hv_get_bytes_to_read(&chn->inbound);
2de8530b
HZ		 167 if (aread)
168 break;
169
40975962 170 aread = hv_get_bytes_to_read(&chn->outbound);
2de8530b
HZ		 171 if (aread)
172 break;
173 }
174
7b2ee50c
SH		 175 if (aread == 0)
176 return 0;
2de8530b 177
7b2ee50c
SH		 178 if (++retry > RETRY_MAX)
179 return -ETIMEDOUT;
2de8530b 180
7b2ee50c 181 usleep_range(RETRY_US_LO, RETRY_US_HI);
2de8530b 182 }
7b2ee50c 183}
2de8530b 184
1b704c4a
HZ		 185static void netvsc_tx_disable(struct netvsc_device *nvscdev,
186 struct net_device *ndev)
187{
188 if (nvscdev) {
189 nvscdev->tx_disable = true;
190 virt_wmb(); /* ensure txq will not wake up after stop */
191 }
192
193 netif_tx_disable(ndev);
194}
195
7b2ee50c
SH		 196static int netvsc_close(struct net_device *net)
197{
198 struct net_device_context *net_device_ctx = netdev_priv(net);
199 struct net_device *vf_netdev
200 = rtnl_dereference(net_device_ctx->vf_netdev);
201 struct netvsc_device *nvdev = rtnl_dereference(net_device_ctx->nvdev);
202 int ret;
203
1b704c4a 204 netvsc_tx_disable(nvdev, net);
7b2ee50c
SH		 205
206 /* No need to close rndis filter if it is removed already */
207 if (!nvdev)
208 return 0;
209
210 ret = rndis_filter_close(nvdev);
211 if (ret != 0) {
212 netdev_err(net, "unable to close device (ret %d).\n", ret);
213 return ret;
2de8530b 214 }
fceaf24a 215
7b2ee50c
SH		 216 ret = netvsc_wait_until_empty(nvdev);
217 if (ret)
218 netdev_err(net, "Ring buffer not empty after closing rndis\n");
219
0c195567 220 if (vf_netdev)
221 dev_close(vf_netdev);
222
fceaf24a
HJ		 223 return ret;
224}
225
f5a22550
SH		 226static inline void *init_ppi_data(struct rndis_message *msg,
227 u32 ppi_size, u32 pkt_type)
8a00251a 228{
f5a22550 229 struct rndis_packet *rndis_pkt = &msg->msg.pkt;
8a00251a
KS		 230 struct rndis_per_packet_info *ppi;
231
8a00251a 232 rndis_pkt->data_offset += ppi_size;
f5a22550
SH		 233 ppi = (void *)rndis_pkt + rndis_pkt->per_pkt_info_offset
234 + rndis_pkt->per_pkt_info_len;
8a00251a
KS		 235
236 ppi->size = ppi_size;
237 ppi->type = pkt_type;
e3a9667a 238 ppi->internal = 0;
8a00251a
KS		 239 ppi->ppi_offset = sizeof(struct rndis_per_packet_info);
240
241 rndis_pkt->per_pkt_info_len += ppi_size;
242
f5a22550 243 return ppi + 1;
8a00251a
KS		 244}
245
4823eb2f
HZ		 246/* Azure hosts don't support non-TCP port numbers in hashing for fragmented
247 * packets. We can use ethtool to change UDP hash level when necessary.
f72860af 248 */
4823eb2f
HZ		 249static inline u32 netvsc_get_hash(
250 struct sk_buff *skb,
251 const struct net_device_context *ndc)
f72860af
HZ		 252{
253 struct flow_keys flow;
486e3981 254 u32 hash, pkt_proto = 0;
f72860af
HZ		 255 static u32 hashrnd __read_mostly;
256
257 net_get_random_once(&hashrnd, sizeof(hashrnd));
258
259 if (!skb_flow_dissect_flow_keys(skb, &flow, 0))
260 return 0;
261
486e3981
HZ		 262 switch (flow.basic.ip_proto) {
263 case IPPROTO_TCP:
264 if (flow.basic.n_proto == htons(ETH_P_IP))
265 pkt_proto = HV_TCP4_L4HASH;
266 else if (flow.basic.n_proto == htons(ETH_P_IPV6))
267 pkt_proto = HV_TCP6_L4HASH;
268
269 break;
270
271 case IPPROTO_UDP:
272 if (flow.basic.n_proto == htons(ETH_P_IP))
273 pkt_proto = HV_UDP4_L4HASH;
274 else if (flow.basic.n_proto == htons(ETH_P_IPV6))
275 pkt_proto = HV_UDP6_L4HASH;
276
277 break;
278 }
279
280 if (pkt_proto & ndc->l4_hash) {
f72860af
HZ		 281 return skb_get_hash(skb);
282 } else {
283 if (flow.basic.n_proto == htons(ETH_P_IP))
284 hash = jhash2((u32 *)&flow.addrs.v4addrs, 2, hashrnd);
285 else if (flow.basic.n_proto == htons(ETH_P_IPV6))
286 hash = jhash2((u32 *)&flow.addrs.v6addrs, 8, hashrnd);
287 else
288 hash = 0;
289
290 skb_set_hash(skb, hash, PKT_HASH_TYPE_L3);
291 }
292
293 return hash;
294}
295
8db91f6a
HZ		 296static inline int netvsc_get_tx_queue(struct net_device *ndev,
297 struct sk_buff *skb, int old_idx)
298{
299 const struct net_device_context *ndc = netdev_priv(ndev);
300 struct sock *sk = skb->sk;
301 int q_idx;
302
39e91cfb
HZ		 303 q_idx = ndc->tx_table[netvsc_get_hash(skb, ndc) &
304 (VRSS_SEND_TAB_SIZE - 1)];
8db91f6a
HZ		 305
306 /* If queue index changed record the new value */
307 if (q_idx != old_idx &&
308 sk && sk_fullsock(sk) && rcu_access_pointer(sk->sk_dst_cache))
309 sk_tx_queue_set(sk, q_idx);
310
311 return q_idx;
312}
313
d8e18ee0 314/*
315 * Select queue for transmit.
316 *
317 * If a valid queue has already been assigned, then use that.
318 * Otherwise compute tx queue based on hash and the send table.
319 *
a350ecce 320 * This is basically similar to default (netdev_pick_tx) with the added step
d8e18ee0 321 * of using the host send_table when no other queue has been assigned.
322 *
323 * TODO support XPS - but get_xps_queue not exported
324 */
0c195567 325static u16 netvsc_pick_tx(struct net_device *ndev, struct sk_buff *skb)
5b54dac8 326{
8db91f6a
HZ		 327 int q_idx = sk_tx_queue_get(skb->sk);
328
0c195567 329 if (q_idx < 0 || skb->ooo_okay || q_idx >= ndev->real_num_tx_queues) {
8db91f6a
HZ		 330 /* If forwarding a packet, we use the recorded queue when
331 * available for better cache locality.
332 */
333 if (skb_rx_queue_recorded(skb))
334 q_idx = skb_get_rx_queue(skb);
335 else
336 q_idx = netvsc_get_tx_queue(ndev, skb, q_idx);
d8e18ee0 337 }
5b54dac8
HZ		 338
339 return q_idx;
340}
341
0c195567 342static u16 netvsc_select_queue(struct net_device *ndev, struct sk_buff *skb,
a350ecce 343 struct net_device *sb_dev)
0c195567 344{
345 struct net_device_context *ndc = netdev_priv(ndev);
346 struct net_device *vf_netdev;
347 u16 txq;
348
349 rcu_read_lock();
350 vf_netdev = rcu_dereference(ndc->vf_netdev);
351 if (vf_netdev) {
b3bf5666
SH		 352 const struct net_device_ops *vf_ops = vf_netdev->netdev_ops;
353
354 if (vf_ops->ndo_select_queue)
a350ecce 355 txq = vf_ops->ndo_select_queue(vf_netdev, skb, sb_dev);
b3bf5666 356 else
a350ecce 357 txq = netdev_pick_tx(vf_netdev, skb, NULL);
b3bf5666
SH		 358
359 /* Record the queue selected by VF so that it can be
360 * used for common case where VF has more queues than
361 * the synthetic device.
362 */
363 qdisc_skb_cb(skb)->slave_dev_queue_mapping = txq;
0c195567 364 } else {
365 txq = netvsc_pick_tx(ndev, skb);
366 }
367 rcu_read_unlock();
368
369 while (unlikely(txq >= ndev->real_num_tx_queues))
370 txq -= ndev->real_num_tx_queues;
371
372 return txq;
373}
374
54a7357f 375static u32 fill_pg_buf(struct page *page, u32 offset, u32 len,
89bb42b1 376 struct hv_page_buffer *pb)
54a7357f
KS		 377{
378 int j = 0;
379
52d3b494 380 /* Deal with compound pages by ignoring unused part
54a7357f
KS		 381 * of the page.
382 */
383 page += (offset >> PAGE_SHIFT);
384 offset &= ~PAGE_MASK;
385
386 while (len > 0) {
387 unsigned long bytes;
388
389 bytes = PAGE_SIZE - offset;
390 if (bytes > len)
391 bytes = len;
392 pb[j].pfn = page_to_pfn(page);
393 pb[j].offset = offset;
394 pb[j].len = bytes;
395
396 offset += bytes;
397 len -= bytes;
398
399 if (offset == PAGE_SIZE && len) {
400 page++;
401 offset = 0;
402 j++;
403 }
404 }
405
406 return j + 1;
407}
408
8a00251a 409static u32 init_page_array(void *hdr, u32 len, struct sk_buff *skb,
a9f2e2d6 410 struct hv_netvsc_packet *packet,
02b6de01 411 struct hv_page_buffer *pb)
54a7357f
KS		 412{
413 u32 slots_used = 0;
414 char *data = skb->data;
415 int frags = skb_shinfo(skb)->nr_frags;
416 int i;
417
418 /* The packet is laid out thus:
aa0a34be 419 * 1. hdr: RNDIS header and PPI
54a7357f
KS		 420 * 2. skb linear data
421 * 3. skb fragment data
422 */
ea5a32c0 423 slots_used += fill_pg_buf(virt_to_page(hdr),
424 offset_in_page(hdr),
425 len, &pb[slots_used]);
54a7357f 426
aa0a34be
HZ		 427 packet->rmsg_size = len;
428 packet->rmsg_pgcnt = slots_used;
429
54a7357f
KS		 430 slots_used += fill_pg_buf(virt_to_page(data),
431 offset_in_page(data),
432 skb_headlen(skb), &pb[slots_used]);
433
434 for (i = 0; i < frags; i++) {
435 skb_frag_t *frag = skb_shinfo(skb)->frags + i;
436
437 slots_used += fill_pg_buf(skb_frag_page(frag),
438 frag->page_offset,
439 skb_frag_size(frag), &pb[slots_used]);
440 }
8a00251a 441 return slots_used;
54a7357f
KS		 442}
443
80d887db 444static int count_skb_frag_slots(struct sk_buff *skb)
445{
446 int i, frags = skb_shinfo(skb)->nr_frags;
447 int pages = 0;
448
449 for (i = 0; i < frags; i++) {
450 skb_frag_t *frag = skb_shinfo(skb)->frags + i;
451 unsigned long size = skb_frag_size(frag);
452 unsigned long offset = frag->page_offset;
453
454 /* Skip unused frames from start of page */
455 offset &= ~PAGE_MASK;
456 pages += PFN_UP(offset + size);
457 }
458 return pages;
459}
460
461static int netvsc_get_slots(struct sk_buff *skb)
54a7357f 462{
80d887db 463 char *data = skb->data;
464 unsigned int offset = offset_in_page(data);
465 unsigned int len = skb_headlen(skb);
466 int slots;
467 int frag_slots;
468
469 slots = DIV_ROUND_UP(offset + len, PAGE_SIZE);
470 frag_slots = count_skb_frag_slots(skb);
471 return slots + frag_slots;
54a7357f
KS		 472}
473
23312a3b 474static u32 net_checksum_info(struct sk_buff *skb)
08cd04bf 475{
23312a3b 476 if (skb->protocol == htons(ETH_P_IP)) {
477 struct iphdr *ip = ip_hdr(skb);
08cd04bf 478
23312a3b 479 if (ip->protocol == IPPROTO_TCP)
480 return TRANSPORT_INFO_IPV4_TCP;
481 else if (ip->protocol == IPPROTO_UDP)
482 return TRANSPORT_INFO_IPV4_UDP;
08cd04bf 483 } else {
23312a3b 484 struct ipv6hdr *ip6 = ipv6_hdr(skb);
485
486 if (ip6->nexthdr == IPPROTO_TCP)
487 return TRANSPORT_INFO_IPV6_TCP;
37b9dfa0 488 else if (ip6->nexthdr == IPPROTO_UDP)
23312a3b 489 return TRANSPORT_INFO_IPV6_UDP;
08cd04bf
KS		 490 }
491
23312a3b 492 return TRANSPORT_INFO_NOT_IP;
08cd04bf
KS		 493}
494
0c195567 495/* Send skb on the slave VF device. */
496static int netvsc_vf_xmit(struct net_device *net, struct net_device *vf_netdev,
497 struct sk_buff *skb)
498{
499 struct net_device_context *ndev_ctx = netdev_priv(net);
500 unsigned int len = skb->len;
501 int rc;
502
503 skb->dev = vf_netdev;
504 skb->queue_mapping = qdisc_skb_cb(skb)->slave_dev_queue_mapping;
505
506 rc = dev_queue_xmit(skb);
507 if (likely(rc == NET_XMIT_SUCCESS || rc == NET_XMIT_CN)) {
508 struct netvsc_vf_pcpu_stats *pcpu_stats
509 = this_cpu_ptr(ndev_ctx->vf_stats);
510
511 u64_stats_update_begin(&pcpu_stats->syncp);
512 pcpu_stats->tx_packets++;
513 pcpu_stats->tx_bytes += len;
514 u64_stats_update_end(&pcpu_stats->syncp);
515 } else {
516 this_cpu_inc(ndev_ctx->vf_stats->tx_dropped);
517 }
518
519 return rc;
520}
521
02fafbc6 522static int netvsc_start_xmit(struct sk_buff *skb, struct net_device *net)
fceaf24a 523{
fceaf24a 524 struct net_device_context *net_device_ctx = netdev_priv(net);
981a1bd8 525 struct hv_netvsc_packet *packet = NULL;
02fafbc6 526 int ret;
8a00251a
KS		 527 unsigned int num_data_pgs;
528 struct rndis_message *rndis_msg;
0c195567 529 struct net_device *vf_netdev;
8a00251a 530 u32 rndis_msg_size;
307f0995 531 u32 hash;
02b6de01 532 struct hv_page_buffer pb[MAX_PAGE_BUFFER_COUNT];
fceaf24a 533
0c195567 534 /* if VF is present and up then redirect packets
535 * already called with rcu_read_lock_bh
536 */
537 vf_netdev = rcu_dereference_bh(net_device_ctx->vf_netdev);
538 if (vf_netdev && netif_running(vf_netdev) &&
539 !netpoll_tx_running(net))
540 return netvsc_vf_xmit(net, vf_netdev, skb);
541
80d887db 542 /* We will atmost need two pages to describe the rndis
543 * header. We can only transmit MAX_PAGE_BUFFER_COUNT number
e88f7e07
VK		 544 * of pages in a single packet. If skb is scattered around
545 * more pages we try linearizing it.
54a7357f 546 */
80d887db 547
548 num_data_pgs = netvsc_get_slots(skb) + 2;
549
0ab05141 550 if (unlikely(num_data_pgs > MAX_PAGE_BUFFER_COUNT)) {
4323b47c
SH		 551 ++net_device_ctx->eth_stats.tx_scattered;
552
553 if (skb_linearize(skb))
554 goto no_memory;
0ab05141 555
80d887db 556 num_data_pgs = netvsc_get_slots(skb) + 2;
0ab05141 557 if (num_data_pgs > MAX_PAGE_BUFFER_COUNT) {
4323b47c 558 ++net_device_ctx->eth_stats.tx_too_big;
0ab05141
SH		 559 goto drop;
560 }
54a7357f 561 }
fceaf24a 562
c0eb4540
KS		 563 /*
564 * Place the rndis header in the skb head room and
565 * the skb->cb will be used for hv_netvsc_packet
566 * structure.
567 */
568 ret = skb_cow_head(skb, RNDIS_AND_PPI_SIZE);
4323b47c
SH		 569 if (ret)
570 goto no_memory;
571
c0eb4540
KS		 572 /* Use the skb control buffer for building up the packet */
573 BUILD_BUG_ON(sizeof(struct hv_netvsc_packet) >
574 FIELD_SIZEOF(struct sk_buff, cb));
575 packet = (struct hv_netvsc_packet *)skb->cb;
fceaf24a 576
5b54dac8
HZ		 577 packet->q_idx = skb_get_queue_mapping(skb);
578
4d447c9a 579 packet->total_data_buflen = skb->len;
793e3955 580 packet->total_bytes = skb->len;
581 packet->total_packets = 1;
fceaf24a 582
c0eb4540 583 rndis_msg = (struct rndis_message *)skb->head;
b08cc791 584
8a00251a 585 /* Add the rndis header */
8a00251a
KS		 586 rndis_msg->ndis_msg_type = RNDIS_MSG_PACKET;
587 rndis_msg->msg_len = packet->total_data_buflen;
f5a22550
SH		 588
589 rndis_msg->msg.pkt = (struct rndis_packet) {
590 .data_offset = sizeof(struct rndis_packet),
591 .data_len = packet->total_data_buflen,
592 .per_pkt_info_offset = sizeof(struct rndis_packet),
593 };
8a00251a
KS		 594
595 rndis_msg_size = RNDIS_MESSAGE_SIZE(struct rndis_packet);
596
307f0995
HZ		 597 hash = skb_get_hash_raw(skb);
598 if (hash != 0 && net->real_num_tx_queues > 1) {
f5a22550
SH		 599 u32 *hash_info;
600
307f0995 601 rndis_msg_size += NDIS_HASH_PPI_SIZE;
f5a22550
SH		 602 hash_info = init_ppi_data(rndis_msg, NDIS_HASH_PPI_SIZE,
603 NBL_HASH_VALUE);
604 *hash_info = hash;
307f0995
HZ		 605 }
606
0ab05141 607 if (skb_vlan_tag_present(skb)) {
8a00251a
KS		 608 struct ndis_pkt_8021q_info *vlan;
609
610 rndis_msg_size += NDIS_VLAN_PPI_SIZE;
f5a22550
SH		 611 vlan = init_ppi_data(rndis_msg, NDIS_VLAN_PPI_SIZE,
612 IEEE_8021Q_INFO);
00f5024e 613
f5a22550 614 vlan->value = 0;
98ba780e
MM		 615 vlan->vlanid = skb_vlan_tag_get_id(skb);
616 vlan->cfi = skb_vlan_tag_get_cfi(skb);
617 vlan->pri = skb_vlan_tag_get_prio(skb);
8a00251a
KS		 618 }
619
23312a3b 620 if (skb_is_gso(skb)) {
0ab05141
SH		 621 struct ndis_tcp_lso_info *lso_info;
622
623 rndis_msg_size += NDIS_LSO_PPI_SIZE;
f5a22550
SH		 624 lso_info = init_ppi_data(rndis_msg, NDIS_LSO_PPI_SIZE,
625 TCP_LARGESEND_PKTINFO);
0ab05141 626
f5a22550 627 lso_info->value = 0;
0ab05141 628 lso_info->lso_v2_transmit.type = NDIS_TCP_LARGE_SEND_OFFLOAD_V2_TYPE;
23312a3b 629 if (skb->protocol == htons(ETH_P_IP)) {
0ab05141
SH		 630 lso_info->lso_v2_transmit.ip_version =
631 NDIS_TCP_LARGE_SEND_OFFLOAD_IPV4;
632 ip_hdr(skb)->tot_len = 0;
633 ip_hdr(skb)->check = 0;
634 tcp_hdr(skb)->check =
635 ~csum_tcpudp_magic(ip_hdr(skb)->saddr,
636 ip_hdr(skb)->daddr, 0, IPPROTO_TCP, 0);
637 } else {
638 lso_info->lso_v2_transmit.ip_version =
639 NDIS_TCP_LARGE_SEND_OFFLOAD_IPV6;
640 ipv6_hdr(skb)->payload_len = 0;
641 tcp_hdr(skb)->check =
642 ~csum_ipv6_magic(&ipv6_hdr(skb)->saddr,
643 &ipv6_hdr(skb)->daddr, 0, IPPROTO_TCP, 0);
644 }
23312a3b 645 lso_info->lso_v2_transmit.tcp_header_offset = skb_transport_offset(skb);
0ab05141 646 lso_info->lso_v2_transmit.mss = skb_shinfo(skb)->gso_size;
ad19bc8a 647 } else if (skb->ip_summed == CHECKSUM_PARTIAL) {
23312a3b 648 if (net_checksum_info(skb) & net_device_ctx->tx_checksum_mask) {
649 struct ndis_tcp_ip_checksum_info *csum_info;
650
ad19bc8a 651 rndis_msg_size += NDIS_CSUM_PPI_SIZE;
f5a22550
SH		 652 csum_info = init_ppi_data(rndis_msg, NDIS_CSUM_PPI_SIZE,
653 TCPIP_CHKSUM_PKTINFO);
ad19bc8a 654
f5a22550 655 csum_info->value = 0;
23312a3b 656 csum_info->transmit.tcp_header_offset = skb_transport_offset(skb);
657
658 if (skb->protocol == htons(ETH_P_IP)) {
ad19bc8a 659 csum_info->transmit.is_ipv4 = 1;
23312a3b 660
661 if (ip_hdr(skb)->protocol == IPPROTO_TCP)
662 csum_info->transmit.tcp_checksum = 1;
663 else
664 csum_info->transmit.udp_checksum = 1;
665 } else {
ad19bc8a 666 csum_info->transmit.is_ipv6 = 1;
667
23312a3b 668 if (ipv6_hdr(skb)->nexthdr == IPPROTO_TCP)
669 csum_info->transmit.tcp_checksum = 1;
670 else
671 csum_info->transmit.udp_checksum = 1;
672 }
ad19bc8a 673 } else {
23312a3b 674 /* Can't do offload of this type of checksum */
ad19bc8a 675 if (skb_checksum_help(skb))
676 goto drop;
677 }
08cd04bf
KS		 678 }
679
8a00251a
KS		 680 /* Start filling in the page buffers with the rndis hdr */
681 rndis_msg->msg_len += rndis_msg_size;
942396b0 682 packet->total_data_buflen = rndis_msg->msg_len;
8a00251a 683 packet->page_buf_cnt = init_page_array(rndis_msg, rndis_msg_size,
02b6de01 684 skb, packet, pb);
8a00251a 685
76d13b56 686 /* timestamp packet in software */
687 skb_tx_timestamp(skb);
2a926f79 688
cfd8afd9 689 ret = netvsc_send(net, packet, rndis_msg, pb, skb);
793e3955 690 if (likely(ret == 0))
0ab05141 691 return NETDEV_TX_OK;
4323b47c
SH		 692
693 if (ret == -EAGAIN) {
694 ++net_device_ctx->eth_stats.tx_busy;
0ab05141 695 return NETDEV_TX_BUSY;
4323b47c
SH		 696 }
697
698 if (ret == -ENOSPC)
699 ++net_device_ctx->eth_stats.tx_no_space;
0ab05141
SH		 700
701drop:
702 dev_kfree_skb_any(skb);
703 net->stats.tx_dropped++;
fceaf24a 704
0ab05141 705 return NETDEV_TX_OK;
4323b47c
SH		 706
707no_memory:
708 ++net_device_ctx->eth_stats.tx_no_memory;
709 goto drop;
fceaf24a 710}
89bb42b1 711
3e189519 712/*
02fafbc6
GKH		 713 * netvsc_linkstatus_callback - Link up/down notification
714 */
79cf1bae 715void netvsc_linkstatus_callback(struct net_device *net,
3a494e71 716 struct rndis_message *resp)
fceaf24a 717{
3a494e71 718 struct rndis_indicate_status *indicate = &resp->msg.indicate_status;
79cf1bae 719 struct net_device_context *ndev_ctx = netdev_priv(net);
27a70af3
VK		 720 struct netvsc_reconfig *event;
721 unsigned long flags;
891de74d 722
7f5d5af0
HZ		 723 /* Update the physical link speed when changing to another vSwitch */
724 if (indicate->status == RNDIS_STATUS_LINK_SPEED_CHANGE) {
725 u32 speed;
726
89bb42b1 727 speed = *(u32 *)((void *)indicate
728 + indicate->status_buf_offset) / 10000;
7f5d5af0
HZ		 729 ndev_ctx->speed = speed;
730 return;
731 }
732
733 /* Handle these link change statuses below */
27a70af3
VK		 734 if (indicate->status != RNDIS_STATUS_NETWORK_CHANGE &&
735 indicate->status != RNDIS_STATUS_MEDIA_CONNECT &&
736 indicate->status != RNDIS_STATUS_MEDIA_DISCONNECT)
3a494e71 737 return;
891de74d 738
7f5d5af0 739 if (net->reg_state != NETREG_REGISTERED)
fceaf24a 740 return;
fceaf24a 741
27a70af3
VK		 742 event = kzalloc(sizeof(*event), GFP_ATOMIC);
743 if (!event)
744 return;
745 event->event = indicate->status;
746
747 spin_lock_irqsave(&ndev_ctx->lock, flags);
748 list_add_tail(&event->list, &ndev_ctx->reconfig_events);
749 spin_unlock_irqrestore(&ndev_ctx->lock, flags);
750
751 schedule_delayed_work(&ndev_ctx->dwork, 0);
fceaf24a
HJ		 752}
753
bf48648d
HZ		 754static void netvsc_comp_ipcsum(struct sk_buff *skb)
755{
756 struct iphdr *iph = (struct iphdr *)skb->data;
757
758 iph->check = 0;
759 iph->check = ip_fast_csum(iph, iph->ihl);
760}
761
84bf9cef 762static struct sk_buff *netvsc_alloc_recv_skb(struct net_device *net,
c8e4eff4 763 struct netvsc_channel *nvchan)
fceaf24a 764{
c8e4eff4
HZ		 765 struct napi_struct *napi = &nvchan->napi;
766 const struct ndis_pkt_8021q_info *vlan = nvchan->rsc.vlan;
767 const struct ndis_tcp_ip_checksum_info *csum_info =
768 nvchan->rsc.csum_info;
fceaf24a 769 struct sk_buff *skb;
c8e4eff4 770 int i;
fceaf24a 771
c8e4eff4 772 skb = napi_alloc_skb(napi, nvchan->rsc.pktlen);
84bf9cef
KS		 773 if (!skb)
774 return skb;
fceaf24a 775
02fafbc6
GKH		 776 /*
777 * Copy to skb. This copy is needed here since the memory pointed by
778 * hv_netvsc_packet cannot be deallocated
779 */
c8e4eff4
HZ		 780 for (i = 0; i < nvchan->rsc.cnt; i++)
781 skb_put_data(skb, nvchan->rsc.data[i], nvchan->rsc.len[i]);
fceaf24a
HJ		 782
783 skb->protocol = eth_type_trans(skb, net);
e52fed71
SH		 784
785 /* skb is already created with CHECKSUM_NONE */
786 skb_checksum_none_assert(skb);
787
bf48648d
HZ		 788 /* Incoming packets may have IP header checksum verified by the host.
789 * They may not have IP header checksum computed after coalescing.
790 * We compute it here if the flags are set, because on Linux, the IP
791 * checksum is always checked.
792 */
793 if (csum_info && csum_info->receive.ip_checksum_value_invalid &&
794 csum_info->receive.ip_checksum_succeeded &&
795 skb->protocol == htons(ETH_P_IP))
796 netvsc_comp_ipcsum(skb);
797
798 /* Do L4 checksum offload if enabled and present.
e52fed71
SH		 799 */
800 if (csum_info && (net->features & NETIF_F_RXCSUM)) {
801 if (csum_info->receive.tcp_checksum_succeeded ||
802 csum_info->receive.udp_checksum_succeeded)
e3d605ed 803 skb->ip_summed = CHECKSUM_UNNECESSARY;
e3d605ed
KS		 804 }
805
dc54a08c 806 if (vlan) {
98ba780e
MM		 807 u16 vlan_tci = vlan->vlanid | (vlan->pri << VLAN_PRIO_SHIFT) |
808 (vlan->cfi ? VLAN_CFI_MASK : 0);
dc54a08c 809
93725cbd 810 __vlan_hwaccel_put_tag(skb, htons(ETH_P_8021Q),
760d1e36 811 vlan_tci);
dc54a08c 812 }
fceaf24a 813
84bf9cef
KS		 814 return skb;
815}
816
817/*
818 * netvsc_recv_callback - Callback when we receive a packet from the
819 * "wire" on the specified device.
820 */
dc54a08c 821int netvsc_recv_callback(struct net_device *net,
345ac089 822 struct netvsc_device *net_device,
c8e4eff4 823 struct netvsc_channel *nvchan)
84bf9cef 824{
3d541ac5 825 struct net_device_context *net_device_ctx = netdev_priv(net);
c8e4eff4 826 struct vmbus_channel *channel = nvchan->channel;
742fe54c 827 u16 q_idx = channel->offermsg.offer.sub_channel_index;
84bf9cef 828 struct sk_buff *skb;
84bf9cef 829 struct netvsc_stats *rx_stats;
84bf9cef 830
9cbcc428 831 if (net->reg_state != NETREG_REGISTERED)
84bf9cef
KS		 832 return NVSP_STAT_FAIL;
833
84bf9cef 834 /* Allocate a skb - TODO direct I/O to pages? */
c8e4eff4
HZ		 835 skb = netvsc_alloc_recv_skb(net, nvchan);
836
84bf9cef 837 if (unlikely(!skb)) {
f61a9d62 838 ++net_device_ctx->eth_stats.rx_no_memory;
84bf9cef
KS		 839 return NVSP_STAT_FAIL;
840 }
5b54dac8 841
0c195567 842 skb_record_rx_queue(skb, q_idx);
9cbcc428
SH		 843
844 /*
845 * Even if injecting the packet, record the statistics
846 * on the synthetic device because modifying the VF device
847 * statistics will not work correctly.
848 */
742fe54c 849 rx_stats = &nvchan->rx_stats;
4b02b58b 850 u64_stats_update_begin(&rx_stats->syncp);
7eafd9b4 851 rx_stats->packets++;
c8e4eff4 852 rx_stats->bytes += nvchan->rsc.pktlen;
f7ad75b7
SH		 853
854 if (skb->pkt_type == PACKET_BROADCAST)
855 ++rx_stats->broadcast;
856 else if (skb->pkt_type == PACKET_MULTICAST)
857 ++rx_stats->multicast;
4b02b58b 858 u64_stats_update_end(&rx_stats->syncp);
9495c282 859
742fe54c 860 napi_gro_receive(&nvchan->napi, skb);
5c71dadb 861 return NVSP_STAT_SUCCESS;
fceaf24a
HJ		 862}
863
f82f4ad7
SH		 864static void netvsc_get_drvinfo(struct net_device *net,
865 struct ethtool_drvinfo *info)
866{
7826d43f 867 strlcpy(info->driver, KBUILD_MODNAME, sizeof(info->driver));
7826d43f 868 strlcpy(info->fw_version, "N/A", sizeof(info->fw_version));
f82f4ad7
SH		 869}
870
59995370
AS		 871static void netvsc_get_channels(struct net_device *net,
872 struct ethtool_channels *channel)
873{
874 struct net_device_context *net_device_ctx = netdev_priv(net);
545a8e79 875 struct netvsc_device *nvdev = rtnl_dereference(net_device_ctx->nvdev);
59995370
AS		 876
877 if (nvdev) {
878 channel->max_combined = nvdev->max_chn;
879 channel->combined_count = nvdev->num_chn;
880 }
881}
882
7c9f335a
HZ		 883/* Alloc struct netvsc_device_info, and initialize it from either existing
884 * struct netvsc_device, or from default values.
885 */
886static struct netvsc_device_info *netvsc_devinfo_get
887 (struct netvsc_device *nvdev)
888{
889 struct netvsc_device_info *dev_info;
890
891 dev_info = kzalloc(sizeof(*dev_info), GFP_ATOMIC);
892
893 if (!dev_info)
894 return NULL;
895
896 if (nvdev) {
897 dev_info->num_chn = nvdev->num_chn;
898 dev_info->send_sections = nvdev->send_section_cnt;
899 dev_info->send_section_size = nvdev->send_section_size;
900 dev_info->recv_sections = nvdev->recv_section_cnt;
901 dev_info->recv_section_size = nvdev->recv_section_size;
17d91256
HZ		 902
903 memcpy(dev_info->rss_key, nvdev->extension->rss_key,
904 NETVSC_HASH_KEYLEN);
7c9f335a
HZ		 905 } else {
906 dev_info->num_chn = VRSS_CHANNEL_DEFAULT;
907 dev_info->send_sections = NETVSC_DEFAULT_TX;
908 dev_info->send_section_size = NETVSC_SEND_SECTION_SIZE;
909 dev_info->recv_sections = NETVSC_DEFAULT_RX;
910 dev_info->recv_section_size = NETVSC_RECV_SECTION_SIZE;
911 }
912
913 return dev_info;
914}
915
7b2ee50c
SH		 916static int netvsc_detach(struct net_device *ndev,
917 struct netvsc_device *nvdev)
918{
919 struct net_device_context *ndev_ctx = netdev_priv(ndev);
920 struct hv_device *hdev = ndev_ctx->device_ctx;
921 int ret;
922
923 /* Don't try continuing to try and setup sub channels */
924 if (cancel_work_sync(&nvdev->subchan_work))
925 nvdev->num_chn = 1;
926
927 /* If device was up (receiving) then shutdown */
928 if (netif_running(ndev)) {
1b704c4a 929 netvsc_tx_disable(nvdev, ndev);
7b2ee50c
SH		 930
931 ret = rndis_filter_close(nvdev);
932 if (ret) {
933 netdev_err(ndev,
934 "unable to close device (ret %d).\n", ret);
935 return ret;
936 }
937
938 ret = netvsc_wait_until_empty(nvdev);
939 if (ret) {
940 netdev_err(ndev,
941 "Ring buffer not empty after closing rndis\n");
942 return ret;
943 }
944 }
945
946 netif_device_detach(ndev);
947
948 rndis_filter_device_remove(hdev, nvdev);
949
950 return 0;
951}
952
953static int netvsc_attach(struct net_device *ndev,
954 struct netvsc_device_info *dev_info)
955{
956 struct net_device_context *ndev_ctx = netdev_priv(ndev);
957 struct hv_device *hdev = ndev_ctx->device_ctx;
958 struct netvsc_device *nvdev;
959 struct rndis_device *rdev;
960 int ret;
961
962 nvdev = rndis_filter_device_add(hdev, dev_info);
963 if (IS_ERR(nvdev))
964 return PTR_ERR(nvdev);
965
3ffe64f1 966 if (nvdev->num_chn > 1) {
17d91256 967 ret = rndis_set_subchannel(ndev, nvdev, dev_info);
3ffe64f1
SH		 968
969 /* if unavailable, just proceed with one queue */
970 if (ret) {
971 nvdev->max_chn = 1;
972 nvdev->num_chn = 1;
973 }
974 }
975
976 /* In any case device is now ready */
977 netif_device_attach(ndev);
7b2ee50c 978
3ffe64f1 979 /* Note: enable and attach happen when sub-channels setup */
7b2ee50c
SH		 980 netif_carrier_off(ndev);
981
982 if (netif_running(ndev)) {
983 ret = rndis_filter_open(nvdev);
984 if (ret)
cc29a1d8 985 goto err;
7b2ee50c
SH		 986
987 rdev = nvdev->extension;
988 if (!rdev->link_state)
989 netif_carrier_on(ndev);
990 }
991
992 return 0;
cc29a1d8
HZ		 993
994err:
995 netif_device_detach(ndev);
996
997 rndis_filter_device_remove(hdev, nvdev);
998
999 return ret;
7b2ee50c
SH		 1000}
1001
b5960e6e
AS		 1002static int netvsc_set_channels(struct net_device *net,
1003 struct ethtool_channels *channels)
1004{
1005 struct net_device_context *net_device_ctx = netdev_priv(net);
545a8e79 1006 struct netvsc_device *nvdev = rtnl_dereference(net_device_ctx->nvdev);
7ca45933 1007 unsigned int orig, count = channels->combined_count;
7c9f335a 1008 struct netvsc_device_info *device_info;
7b2ee50c 1009 int ret;
2b01888d 1010
1011 /* We do not support separate count for rx, tx, or other */
1012 if (count == 0 ||
1013 channels->rx_count || channels->tx_count || channels->other_count)
1014 return -EINVAL;
1015
a0be450e 1016 if (!nvdev || nvdev->destroy)
b5960e6e
AS		 1017 return -ENODEV;
1018
2b01888d 1019 if (nvdev->nvsp_version < NVSP_PROTOCOL_VERSION_5)
b5960e6e 1020 return -EINVAL;
b5960e6e 1021
2b01888d 1022 if (count > nvdev->max_chn)
b5960e6e
AS		 1023 return -EINVAL;
1024
7ca45933 1025 orig = nvdev->num_chn;
b5960e6e 1026
7c9f335a
HZ		 1027 device_info = netvsc_devinfo_get(nvdev);
1028
1029 if (!device_info)
1030 return -ENOMEM;
1031
1032 device_info->num_chn = count;
8b532797 1033
7b2ee50c
SH		 1034 ret = netvsc_detach(net, nvdev);
1035 if (ret)
7c9f335a 1036 goto out;
7ca45933 1037
7c9f335a 1038 ret = netvsc_attach(net, device_info);
7b2ee50c 1039 if (ret) {
7c9f335a
HZ		 1040 device_info->num_chn = orig;
1041 if (netvsc_attach(net, device_info))
7b2ee50c 1042 netdev_err(net, "restoring channel setting failed\n");
7ca45933 1043 }
b5960e6e 1044
7c9f335a
HZ		 1045out:
1046 kfree(device_info);
b5960e6e 1047 return ret;
b5960e6e
AS		 1048}
1049
5e8456fd
PR		 1050static bool
1051netvsc_validate_ethtool_ss_cmd(const struct ethtool_link_ksettings *cmd)
49eb9389 1052{
5e8456fd
PR		 1053 struct ethtool_link_ksettings diff1 = *cmd;
1054 struct ethtool_link_ksettings diff2 = {};
49eb9389 1055
5e8456fd
PR		 1056 diff1.base.speed = 0;
1057 diff1.base.duplex = 0;
49eb9389 1058 /* advertising and cmd are usually set */
5e8456fd
PR		 1059 ethtool_link_ksettings_zero_link_mode(&diff1, advertising);
1060 diff1.base.cmd = 0;
49eb9389 1061 /* We set port to PORT_OTHER */
5e8456fd 1062 diff2.base.port = PORT_OTHER;
49eb9389 1063
1064 return !memcmp(&diff1, &diff2, sizeof(diff1));
1065}
1066
1067static void netvsc_init_settings(struct net_device *dev)
1068{
1069 struct net_device_context *ndc = netdev_priv(dev);
1070
486e3981 1071 ndc->l4_hash = HV_DEFAULT_L4HASH;
4823eb2f 1072
49eb9389 1073 ndc->speed = SPEED_UNKNOWN;
f3c9d40e 1074 ndc->duplex = DUPLEX_FULL;
d6792a5a
HZ		 1075
1076 dev->features = NETIF_F_LRO;
49eb9389 1077}
1078
5e8456fd
PR		 1079static int netvsc_get_link_ksettings(struct net_device *dev,
1080 struct ethtool_link_ksettings *cmd)
49eb9389 1081{
1082 struct net_device_context *ndc = netdev_priv(dev);
1083
5e8456fd
PR		 1084 cmd->base.speed = ndc->speed;
1085 cmd->base.duplex = ndc->duplex;
1086 cmd->base.port = PORT_OTHER;
49eb9389 1087
1088 return 0;
1089}
1090
5e8456fd
PR		 1091static int netvsc_set_link_ksettings(struct net_device *dev,
1092 const struct ethtool_link_ksettings *cmd)
49eb9389 1093{
1094 struct net_device_context *ndc = netdev_priv(dev);
1095 u32 speed;
1096
5e8456fd 1097 speed = cmd->base.speed;
49eb9389 1098 if (!ethtool_validate_speed(speed) ||
5e8456fd 1099 !ethtool_validate_duplex(cmd->base.duplex) ||
49eb9389 1100 !netvsc_validate_ethtool_ss_cmd(cmd))
1101 return -EINVAL;
1102
1103 ndc->speed = speed;
5e8456fd 1104 ndc->duplex = cmd->base.duplex;
49eb9389 1105
1106 return 0;
1107}
1108
4d447c9a
HZ		 1109static int netvsc_change_mtu(struct net_device *ndev, int mtu)
1110{
1111 struct net_device_context *ndevctx = netdev_priv(ndev);
0c195567 1112 struct net_device *vf_netdev = rtnl_dereference(ndevctx->vf_netdev);
545a8e79 1113 struct netvsc_device *nvdev = rtnl_dereference(ndevctx->nvdev);
9749fed5 1114 int orig_mtu = ndev->mtu;
7c9f335a 1115 struct netvsc_device_info *device_info;
9749fed5 1116 int ret = 0;
4d447c9a 1117
a0be450e 1118 if (!nvdev || nvdev->destroy)
4d447c9a
HZ		 1119 return -ENODEV;
1120
7c9f335a
HZ		 1121 device_info = netvsc_devinfo_get(nvdev);
1122
1123 if (!device_info)
1124 return -ENOMEM;
1125
0c195567 1126 /* Change MTU of underlying VF netdev first. */
1127 if (vf_netdev) {
1128 ret = dev_set_mtu(vf_netdev, mtu);
1129 if (ret)
7c9f335a 1130 goto out;
0c195567 1131 }
1132
7b2ee50c
SH		 1133 ret = netvsc_detach(ndev, nvdev);
1134 if (ret)
1135 goto rollback_vf;
4d447c9a
HZ		 1136
1137 ndev->mtu = mtu;
1138
7c9f335a
HZ		 1139 ret = netvsc_attach(ndev, device_info);
1140 if (!ret)
1141 goto out;
ea383bf1 1142
7b2ee50c
SH		 1143 /* Attempt rollback to original MTU */
1144 ndev->mtu = orig_mtu;
163891d7 1145
7c9f335a 1146 if (netvsc_attach(ndev, device_info))
7b2ee50c
SH		 1147 netdev_err(ndev, "restoring mtu failed\n");
1148rollback_vf:
1149 if (vf_netdev)
1150 dev_set_mtu(vf_netdev, orig_mtu);
1bdcec8a 1151
7c9f335a
HZ		 1152out:
1153 kfree(device_info);
9749fed5 1154 return ret;
4d447c9a
HZ		 1155}
1156
0c195567 1157static void netvsc_get_vf_stats(struct net_device *net,
1158 struct netvsc_vf_pcpu_stats *tot)
1159{
1160 struct net_device_context *ndev_ctx = netdev_priv(net);
1161 int i;
1162
1163 memset(tot, 0, sizeof(*tot));
1164
1165 for_each_possible_cpu(i) {
1166 const struct netvsc_vf_pcpu_stats *stats
1167 = per_cpu_ptr(ndev_ctx->vf_stats, i);
1168 u64 rx_packets, rx_bytes, tx_packets, tx_bytes;
1169 unsigned int start;
1170
1171 do {
1172 start = u64_stats_fetch_begin_irq(&stats->syncp);
1173 rx_packets = stats->rx_packets;
1174 tx_packets = stats->tx_packets;
1175 rx_bytes = stats->rx_bytes;
1176 tx_bytes = stats->tx_bytes;
1177 } while (u64_stats_fetch_retry_irq(&stats->syncp, start));
1178
1179 tot->rx_packets += rx_packets;
1180 tot->tx_packets += tx_packets;
1181 tot->rx_bytes += rx_bytes;
1182 tot->tx_bytes += tx_bytes;
1183 tot->tx_dropped += stats->tx_dropped;
1184 }
1185}
1186
6ae74671
YR		 1187static void netvsc_get_pcpu_stats(struct net_device *net,
1188 struct netvsc_ethtool_pcpu_stats *pcpu_tot)
1189{
1190 struct net_device_context *ndev_ctx = netdev_priv(net);
1191 struct netvsc_device *nvdev = rcu_dereference_rtnl(ndev_ctx->nvdev);
1192 int i;
1193
1194 /* fetch percpu stats of vf */
1195 for_each_possible_cpu(i) {
1196 const struct netvsc_vf_pcpu_stats *stats =
1197 per_cpu_ptr(ndev_ctx->vf_stats, i);
1198 struct netvsc_ethtool_pcpu_stats *this_tot = &pcpu_tot[i];
1199 unsigned int start;
1200
1201 do {
1202 start = u64_stats_fetch_begin_irq(&stats->syncp);
1203 this_tot->vf_rx_packets = stats->rx_packets;
1204 this_tot->vf_tx_packets = stats->tx_packets;
1205 this_tot->vf_rx_bytes = stats->rx_bytes;
1206 this_tot->vf_tx_bytes = stats->tx_bytes;
1207 } while (u64_stats_fetch_retry_irq(&stats->syncp, start));
1208 this_tot->rx_packets = this_tot->vf_rx_packets;
1209 this_tot->tx_packets = this_tot->vf_tx_packets;
1210 this_tot->rx_bytes = this_tot->vf_rx_bytes;
1211 this_tot->tx_bytes = this_tot->vf_tx_bytes;
1212 }
1213
1214 /* fetch percpu stats of netvsc */
1215 for (i = 0; i < nvdev->num_chn; i++) {
1216 const struct netvsc_channel *nvchan = &nvdev->chan_table[i];
1217 const struct netvsc_stats *stats;
1218 struct netvsc_ethtool_pcpu_stats *this_tot =
1219 &pcpu_tot[nvchan->channel->target_cpu];
1220 u64 packets, bytes;
1221 unsigned int start;
1222
1223 stats = &nvchan->tx_stats;
1224 do {
1225 start = u64_stats_fetch_begin_irq(&stats->syncp);
1226 packets = stats->packets;
1227 bytes = stats->bytes;
1228 } while (u64_stats_fetch_retry_irq(&stats->syncp, start));
1229
1230 this_tot->tx_bytes += bytes;
1231 this_tot->tx_packets += packets;
1232
1233 stats = &nvchan->rx_stats;
1234 do {
1235 start = u64_stats_fetch_begin_irq(&stats->syncp);
1236 packets = stats->packets;
1237 bytes = stats->bytes;
1238 } while (u64_stats_fetch_retry_irq(&stats->syncp, start));
1239
1240 this_tot->rx_bytes += bytes;
1241 this_tot->rx_packets += packets;
1242 }
1243}
1244
bc1f4470 1245static void netvsc_get_stats64(struct net_device *net,
1246 struct rtnl_link_stats64 *t)
7eafd9b4 1247{
1248 struct net_device_context *ndev_ctx = netdev_priv(net);
6d0d779d 1249 struct netvsc_device *nvdev;
0c195567 1250 struct netvsc_vf_pcpu_stats vf_tot;
89bb42b1 1251 int i;
6c80f3fc 1252
6d0d779d
DC		 1253 rcu_read_lock();
1254
1255 nvdev = rcu_dereference(ndev_ctx->nvdev);
6c80f3fc 1256 if (!nvdev)
6d0d779d 1257 goto out;
6c80f3fc 1258
0c195567 1259 netdev_stats_to_stats64(t, &net->stats);
1260
1261 netvsc_get_vf_stats(net, &vf_tot);
1262 t->rx_packets += vf_tot.rx_packets;
1263 t->tx_packets += vf_tot.tx_packets;
1264 t->rx_bytes += vf_tot.rx_bytes;
1265 t->tx_bytes += vf_tot.tx_bytes;
1266 t->tx_dropped += vf_tot.tx_dropped;
1267
6c80f3fc
SX		 1268 for (i = 0; i < nvdev->num_chn; i++) {
1269 const struct netvsc_channel *nvchan = &nvdev->chan_table[i];
1270 const struct netvsc_stats *stats;
1271 u64 packets, bytes, multicast;
7eafd9b4 1272 unsigned int start;
1273
6c80f3fc 1274 stats = &nvchan->tx_stats;
7eafd9b4 1275 do {
6c80f3fc
SX		 1276 start = u64_stats_fetch_begin_irq(&stats->syncp);
1277 packets = stats->packets;
1278 bytes = stats->bytes;
1279 } while (u64_stats_fetch_retry_irq(&stats->syncp, start));
1280
1281 t->tx_bytes += bytes;
1282 t->tx_packets += packets;
7eafd9b4 1283
6c80f3fc 1284 stats = &nvchan->rx_stats;
7eafd9b4 1285 do {
6c80f3fc
SX		 1286 start = u64_stats_fetch_begin_irq(&stats->syncp);
1287 packets = stats->packets;
1288 bytes = stats->bytes;
1289 multicast = stats->multicast + stats->broadcast;
1290 } while (u64_stats_fetch_retry_irq(&stats->syncp, start));
1291
1292 t->rx_bytes += bytes;
1293 t->rx_packets += packets;
1294 t->multicast += multicast;
7eafd9b4 1295 }
6d0d779d
DC		 1296out:
1297 rcu_read_unlock();
7eafd9b4 1298}
1ce09e89
HZ		 1299
1300static int netvsc_set_mac_addr(struct net_device *ndev, void *p)
1301{
867047c4 1302 struct net_device_context *ndc = netdev_priv(ndev);
16ba3266 1303 struct net_device *vf_netdev = rtnl_dereference(ndc->vf_netdev);
867047c4 1304 struct netvsc_device *nvdev = rtnl_dereference(ndc->nvdev);
1ce09e89 1305 struct sockaddr *addr = p;
1ce09e89
HZ		 1306 int err;
1307
16ba3266 1308 err = eth_prepare_mac_addr_change(ndev, p);
1309 if (err)
1ce09e89
HZ		 1310 return err;
1311
867047c4 1312 if (!nvdev)
1313 return -ENODEV;
1314
16ba3266 1315 if (vf_netdev) {
3a37a963 1316 err = dev_set_mac_address(vf_netdev, addr, NULL);
16ba3266 1317 if (err)
1318 return err;
1319 }
1320
867047c4 1321 err = rndis_filter_set_device_mac(nvdev, addr->sa_data);
16ba3266 1322 if (!err) {
1323 eth_commit_mac_addr_change(ndev, p);
1324 } else if (vf_netdev) {
1325 /* rollback change on VF */
1326 memcpy(addr->sa_data, ndev->dev_addr, ETH_ALEN);
3a37a963 1327 dev_set_mac_address(vf_netdev, addr, NULL);
1ce09e89
HZ		 1328 }
1329
1330 return err;
1331}
1332
4323b47c
SH		 1333static const struct {
1334 char name[ETH_GSTRING_LEN];
1335 u16 offset;
1336} netvsc_stats[] = {
1337 { "tx_scattered", offsetof(struct netvsc_ethtool_stats, tx_scattered) },
f61a9d62 1338 { "tx_no_memory", offsetof(struct netvsc_ethtool_stats, tx_no_memory) },
4323b47c
SH		 1339 { "tx_no_space", offsetof(struct netvsc_ethtool_stats, tx_no_space) },
1340 { "tx_too_big", offsetof(struct netvsc_ethtool_stats, tx_too_big) },
1341 { "tx_busy", offsetof(struct netvsc_ethtool_stats, tx_busy) },
cad5c197 1342 { "tx_send_full", offsetof(struct netvsc_ethtool_stats, tx_send_full) },
1343 { "rx_comp_busy", offsetof(struct netvsc_ethtool_stats, rx_comp_busy) },
f61a9d62 1344 { "rx_no_memory", offsetof(struct netvsc_ethtool_stats, rx_no_memory) },
09af87d1
SX		 1345 { "stop_queue", offsetof(struct netvsc_ethtool_stats, stop_queue) },
1346 { "wake_queue", offsetof(struct netvsc_ethtool_stats, wake_queue) },
6ae74671
YR		 1347}, pcpu_stats[] = {
1348 { "cpu%u_rx_packets",
1349 offsetof(struct netvsc_ethtool_pcpu_stats, rx_packets) },
1350 { "cpu%u_rx_bytes",
1351 offsetof(struct netvsc_ethtool_pcpu_stats, rx_bytes) },
1352 { "cpu%u_tx_packets",
1353 offsetof(struct netvsc_ethtool_pcpu_stats, tx_packets) },
1354 { "cpu%u_tx_bytes",
1355 offsetof(struct netvsc_ethtool_pcpu_stats, tx_bytes) },
1356 { "cpu%u_vf_rx_packets",
1357 offsetof(struct netvsc_ethtool_pcpu_stats, vf_rx_packets) },
1358 { "cpu%u_vf_rx_bytes",
1359 offsetof(struct netvsc_ethtool_pcpu_stats, vf_rx_bytes) },
1360 { "cpu%u_vf_tx_packets",
1361 offsetof(struct netvsc_ethtool_pcpu_stats, vf_tx_packets) },
1362 { "cpu%u_vf_tx_bytes",
1363 offsetof(struct netvsc_ethtool_pcpu_stats, vf_tx_bytes) },
0c195567 1364}, vf_stats[] = {
1365 { "vf_rx_packets", offsetof(struct netvsc_vf_pcpu_stats, rx_packets) },
1366 { "vf_rx_bytes", offsetof(struct netvsc_vf_pcpu_stats, rx_bytes) },
1367 { "vf_tx_packets", offsetof(struct netvsc_vf_pcpu_stats, tx_packets) },
1368 { "vf_tx_bytes", offsetof(struct netvsc_vf_pcpu_stats, tx_bytes) },
1369 { "vf_tx_dropped", offsetof(struct netvsc_vf_pcpu_stats, tx_dropped) },
4323b47c
SH		 1370};
1371
6c80f3fc 1372#define NETVSC_GLOBAL_STATS_LEN ARRAY_SIZE(netvsc_stats)
0c195567 1373#define NETVSC_VF_STATS_LEN ARRAY_SIZE(vf_stats)
6c80f3fc 1374
6ae74671
YR		 1375/* statistics per queue (rx/tx packets/bytes) */
1376#define NETVSC_PCPU_STATS_LEN (num_present_cpus() * ARRAY_SIZE(pcpu_stats))
1377
6c80f3fc
SX		 1378/* 4 statistics per queue (rx/tx packets/bytes) */
1379#define NETVSC_QUEUE_STATS_LEN(dev) ((dev)->num_chn * 4)
1380
4323b47c
SH		 1381static int netvsc_get_sset_count(struct net_device *dev, int string_set)
1382{
6c80f3fc 1383 struct net_device_context *ndc = netdev_priv(dev);
fbd4c7e7 1384 struct netvsc_device *nvdev = rtnl_dereference(ndc->nvdev);
545a8e79 1385
1386 if (!nvdev)
1387 return -ENODEV;
6c80f3fc 1388
4323b47c
SH		 1389 switch (string_set) {
1390 case ETH_SS_STATS:
0c195567 1391 return NETVSC_GLOBAL_STATS_LEN
1392 + NETVSC_VF_STATS_LEN
6ae74671
YR		 1393 + NETVSC_QUEUE_STATS_LEN(nvdev)
1394 + NETVSC_PCPU_STATS_LEN;
4323b47c
SH		 1395 default:
1396 return -EINVAL;
1397 }
1398}
1399
1400static void netvsc_get_ethtool_stats(struct net_device *dev,
1401 struct ethtool_stats *stats, u64 *data)
1402{
1403 struct net_device_context *ndc = netdev_priv(dev);
867047c4 1404 struct netvsc_device *nvdev = rtnl_dereference(ndc->nvdev);
4323b47c 1405 const void *nds = &ndc->eth_stats;
6c80f3fc 1406 const struct netvsc_stats *qstats;
0c195567 1407 struct netvsc_vf_pcpu_stats sum;
6ae74671 1408 struct netvsc_ethtool_pcpu_stats *pcpu_sum;
6c80f3fc
SX		 1409 unsigned int start;
1410 u64 packets, bytes;
6ae74671 1411 int i, j, cpu;
4323b47c 1412
545a8e79 1413 if (!nvdev)
1414 return;
1415
6c80f3fc 1416 for (i = 0; i < NETVSC_GLOBAL_STATS_LEN; i++)
4323b47c 1417 data[i] = *(unsigned long *)(nds + netvsc_stats[i].offset);
6c80f3fc 1418
0c195567 1419 netvsc_get_vf_stats(dev, &sum);
1420 for (j = 0; j < NETVSC_VF_STATS_LEN; j++)
1421 data[i++] = *(u64 *)((void *)&sum + vf_stats[j].offset);
1422
6c80f3fc
SX		 1423 for (j = 0; j < nvdev->num_chn; j++) {
1424 qstats = &nvdev->chan_table[j].tx_stats;
1425
1426 do {
1427 start = u64_stats_fetch_begin_irq(&qstats->syncp);
1428 packets = qstats->packets;
1429 bytes = qstats->bytes;
1430 } while (u64_stats_fetch_retry_irq(&qstats->syncp, start));
1431 data[i++] = packets;
1432 data[i++] = bytes;
1433
1434 qstats = &nvdev->chan_table[j].rx_stats;
1435 do {
1436 start = u64_stats_fetch_begin_irq(&qstats->syncp);
1437 packets = qstats->packets;
1438 bytes = qstats->bytes;
1439 } while (u64_stats_fetch_retry_irq(&qstats->syncp, start));
1440 data[i++] = packets;
1441 data[i++] = bytes;
1442 }
6ae74671
YR		 1443
1444 pcpu_sum = kvmalloc_array(num_possible_cpus(),
1445 sizeof(struct netvsc_ethtool_pcpu_stats),
1446 GFP_KERNEL);
1447 netvsc_get_pcpu_stats(dev, pcpu_sum);
1448 for_each_present_cpu(cpu) {
1449 struct netvsc_ethtool_pcpu_stats *this_sum = &pcpu_sum[cpu];
1450
1451 for (j = 0; j < ARRAY_SIZE(pcpu_stats); j++)
1452 data[i++] = *(u64 *)((void *)this_sum
1453 + pcpu_stats[j].offset);
1454 }
1455 kvfree(pcpu_sum);
4323b47c
SH		 1456}
1457
1458static void netvsc_get_strings(struct net_device *dev, u32 stringset, u8 *data)
1459{
6c80f3fc 1460 struct net_device_context *ndc = netdev_priv(dev);
867047c4 1461 struct netvsc_device *nvdev = rtnl_dereference(ndc->nvdev);
6c80f3fc 1462 u8 *p = data;
6ae74671 1463 int i, cpu;
4323b47c 1464
545a8e79 1465 if (!nvdev)
1466 return;
1467
4323b47c
SH		 1468 switch (stringset) {
1469 case ETH_SS_STATS:
0c195567 1470 for (i = 0; i < ARRAY_SIZE(netvsc_stats); i++) {
1471 memcpy(p, netvsc_stats[i].name, ETH_GSTRING_LEN);
1472 p += ETH_GSTRING_LEN;
1473 }
1474
1475 for (i = 0; i < ARRAY_SIZE(vf_stats); i++) {
1476 memcpy(p, vf_stats[i].name, ETH_GSTRING_LEN);
1477 p += ETH_GSTRING_LEN;
1478 }
6c80f3fc 1479
6c80f3fc
SX		 1480 for (i = 0; i < nvdev->num_chn; i++) {
1481 sprintf(p, "tx_queue_%u_packets", i);
1482 p += ETH_GSTRING_LEN;
1483 sprintf(p, "tx_queue_%u_bytes", i);
1484 p += ETH_GSTRING_LEN;
1485 sprintf(p, "rx_queue_%u_packets", i);
1486 p += ETH_GSTRING_LEN;
1487 sprintf(p, "rx_queue_%u_bytes", i);
1488 p += ETH_GSTRING_LEN;
1489 }
1490
6ae74671
YR		 1491 for_each_present_cpu(cpu) {
1492 for (i = 0; i < ARRAY_SIZE(pcpu_stats); i++) {
1493 sprintf(p, pcpu_stats[i].name, cpu);
1494 p += ETH_GSTRING_LEN;
1495 }
1496 }
1497
4323b47c
SH		 1498 break;
1499 }
1500}
1501
b5a5dc8d 1502static int
4823eb2f
HZ		 1503netvsc_get_rss_hash_opts(struct net_device_context *ndc,
1504 struct ethtool_rxnfc *info)
b5a5dc8d 1505{
486e3981
HZ		 1506 const u32 l4_flag = RXH_L4_B_0_1 | RXH_L4_B_2_3;
1507
b5a5dc8d 1508 info->data = RXH_IP_SRC | RXH_IP_DST;
1509
1510 switch (info->flow_type) {
1511 case TCP_V4_FLOW:
0518ec4f
HZ		 1512 if (ndc->l4_hash & HV_TCP4_L4HASH)
1513 info->data |= l4_flag;
1514
1515 break;
1516
b5a5dc8d 1517 case TCP_V6_FLOW:
0518ec4f
HZ		 1518 if (ndc->l4_hash & HV_TCP6_L4HASH)
1519 info->data |= l4_flag;
1520
4823eb2f
HZ		 1521 break;
1522
b5a5dc8d 1523 case UDP_V4_FLOW:
486e3981
HZ		 1524 if (ndc->l4_hash & HV_UDP4_L4HASH)
1525 info->data |= l4_flag;
4823eb2f
HZ		 1526
1527 break;
1528
b5a5dc8d 1529 case UDP_V6_FLOW:
486e3981
HZ		 1530 if (ndc->l4_hash & HV_UDP6_L4HASH)
1531 info->data |= l4_flag;
4823eb2f
HZ		 1532
1533 break;
1534
b5a5dc8d 1535 case IPV4_FLOW:
1536 case IPV6_FLOW:
1537 break;
1538 default:
1539 info->data = 0;
1540 break;
1541 }
1542
1543 return 0;
1544}
1545
b448f4e8 1546static int
1547netvsc_get_rxnfc(struct net_device *dev, struct ethtool_rxnfc *info,
1548 u32 *rules)
1549{
1550 struct net_device_context *ndc = netdev_priv(dev);
867047c4 1551 struct netvsc_device *nvdev = rtnl_dereference(ndc->nvdev);
545a8e79 1552
1553 if (!nvdev)
1554 return -ENODEV;
b448f4e8 1555
1556 switch (info->cmd) {
1557 case ETHTOOL_GRXRINGS:
1558 info->data = nvdev->num_chn;
1559 return 0;
b5a5dc8d 1560
1561 case ETHTOOL_GRXFH:
4823eb2f 1562 return netvsc_get_rss_hash_opts(ndc, info);
b448f4e8 1563 }
1564 return -EOPNOTSUPP;
1565}
1566
4823eb2f
HZ		 1567static int netvsc_set_rss_hash_opts(struct net_device_context *ndc,
1568 struct ethtool_rxnfc *info)
1569{
1570 if (info->data == (RXH_IP_SRC | RXH_IP_DST |
1571 RXH_L4_B_0_1 | RXH_L4_B_2_3)) {
486e3981 1572 switch (info->flow_type) {
0518ec4f
HZ		 1573 case TCP_V4_FLOW:
1574 ndc->l4_hash |= HV_TCP4_L4HASH;
1575 break;
1576
1577 case TCP_V6_FLOW:
1578 ndc->l4_hash |= HV_TCP6_L4HASH;
1579 break;
1580
486e3981
HZ		 1581 case UDP_V4_FLOW:
1582 ndc->l4_hash |= HV_UDP4_L4HASH;
1583 break;
1584
1585 case UDP_V6_FLOW:
1586 ndc->l4_hash |= HV_UDP6_L4HASH;
1587 break;
1588
1589 default:
4823eb2f 1590 return -EOPNOTSUPP;
486e3981 1591 }
4823eb2f
HZ		 1592
1593 return 0;
1594 }
1595
1596 if (info->data == (RXH_IP_SRC | RXH_IP_DST)) {
486e3981 1597 switch (info->flow_type) {
0518ec4f
HZ		 1598 case TCP_V4_FLOW:
1599 ndc->l4_hash &= ~HV_TCP4_L4HASH;
1600 break;
1601
1602 case TCP_V6_FLOW:
1603 ndc->l4_hash &= ~HV_TCP6_L4HASH;
1604 break;
1605
486e3981
HZ		 1606 case UDP_V4_FLOW:
1607 ndc->l4_hash &= ~HV_UDP4_L4HASH;
1608 break;
1609
1610 case UDP_V6_FLOW:
1611 ndc->l4_hash &= ~HV_UDP6_L4HASH;
1612 break;
1613
1614 default:
4823eb2f 1615 return -EOPNOTSUPP;
486e3981 1616 }
4823eb2f
HZ		 1617
1618 return 0;
1619 }
1620
1621 return -EOPNOTSUPP;
1622}
1623
1624static int
1625netvsc_set_rxnfc(struct net_device *ndev, struct ethtool_rxnfc *info)
1626{
1627 struct net_device_context *ndc = netdev_priv(ndev);
1628
1629 if (info->cmd == ETHTOOL_SRXFH)
1630 return netvsc_set_rss_hash_opts(ndc, info);
1631
1632 return -EOPNOTSUPP;
1633}
1634
962f3fee 1635static u32 netvsc_get_rxfh_key_size(struct net_device *dev)
1636{
1637 return NETVSC_HASH_KEYLEN;
1638}
1639
1640static u32 netvsc_rss_indir_size(struct net_device *dev)
1641{
ff4a4419 1642 return ITAB_NUM;
962f3fee 1643}
1644
1645static int netvsc_get_rxfh(struct net_device *dev, u32 *indir, u8 *key,
1646 u8 *hfunc)
1647{
1648 struct net_device_context *ndc = netdev_priv(dev);
867047c4 1649 struct netvsc_device *ndev = rtnl_dereference(ndc->nvdev);
eb996edb 1650 struct rndis_device *rndis_dev;
ff4a4419 1651 int i;
962f3fee 1652
545a8e79 1653 if (!ndev)
1654 return -ENODEV;
1655
962f3fee 1656 if (hfunc)
1657 *hfunc = ETH_RSS_HASH_TOP; /* Toeplitz */
1658
eb996edb 1659 rndis_dev = ndev->extension;
ff4a4419 1660 if (indir) {
1661 for (i = 0; i < ITAB_NUM; i++)
47371300 1662 indir[i] = rndis_dev->rx_table[i];
ff4a4419 1663 }
1664
962f3fee 1665 if (key)
1666 memcpy(key, rndis_dev->rss_key, NETVSC_HASH_KEYLEN);
1667
1668 return 0;
1669}
1670
1671static int netvsc_set_rxfh(struct net_device *dev, const u32 *indir,
1672 const u8 *key, const u8 hfunc)
1673{
1674 struct net_device_context *ndc = netdev_priv(dev);
545a8e79 1675 struct netvsc_device *ndev = rtnl_dereference(ndc->nvdev);
eb996edb 1676 struct rndis_device *rndis_dev;
ff4a4419 1677 int i;
962f3fee 1678
545a8e79 1679 if (!ndev)
1680 return -ENODEV;
1681
962f3fee 1682 if (hfunc != ETH_RSS_HASH_NO_CHANGE && hfunc != ETH_RSS_HASH_TOP)
1683 return -EOPNOTSUPP;
1684
eb996edb 1685 rndis_dev = ndev->extension;
ff4a4419 1686 if (indir) {
1687 for (i = 0; i < ITAB_NUM; i++)
db3cd7af 1688 if (indir[i] >= ndev->num_chn)
ff4a4419 1689 return -EINVAL;
1690
1691 for (i = 0; i < ITAB_NUM; i++)
47371300 1692 rndis_dev->rx_table[i] = indir[i];
ff4a4419 1693 }
1694
1695 if (!key) {
1696 if (!indir)
1697 return 0;
1698
1699 key = rndis_dev->rss_key;
1700 }
962f3fee 1701
715e2ec5 1702 return rndis_filter_set_rss_param(rndis_dev, key);
962f3fee 1703}
1704
8b532797 1705/* Hyper-V RNDIS protocol does not have ring in the HW sense.
1706 * It does have pre-allocated receive area which is divided into sections.
1707 */
1708static void __netvsc_get_ringparam(struct netvsc_device *nvdev,
1709 struct ethtool_ringparam *ring)
1710{
1711 u32 max_buf_size;
1712
1713 ring->rx_pending = nvdev->recv_section_cnt;
1714 ring->tx_pending = nvdev->send_section_cnt;
1715
1716 if (nvdev->nvsp_version <= NVSP_PROTOCOL_VERSION_2)
1717 max_buf_size = NETVSC_RECEIVE_BUFFER_SIZE_LEGACY;
1718 else
1719 max_buf_size = NETVSC_RECEIVE_BUFFER_SIZE;
1720
1721 ring->rx_max_pending = max_buf_size / nvdev->recv_section_size;
1722 ring->tx_max_pending = NETVSC_SEND_BUFFER_SIZE
1723 / nvdev->send_section_size;
1724}
1725
1726static void netvsc_get_ringparam(struct net_device *ndev,
1727 struct ethtool_ringparam *ring)
1728{
1729 struct net_device_context *ndevctx = netdev_priv(ndev);
1730 struct netvsc_device *nvdev = rtnl_dereference(ndevctx->nvdev);
1731
1732 if (!nvdev)
1733 return;
1734
1735 __netvsc_get_ringparam(nvdev, ring);
1736}
1737
1738static int netvsc_set_ringparam(struct net_device *ndev,
1739 struct ethtool_ringparam *ring)
1740{
1741 struct net_device_context *ndevctx = netdev_priv(ndev);
1742 struct netvsc_device *nvdev = rtnl_dereference(ndevctx->nvdev);
7c9f335a 1743 struct netvsc_device_info *device_info;
8b532797 1744 struct ethtool_ringparam orig;
1745 u32 new_tx, new_rx;
8b532797 1746 int ret = 0;
1747
1748 if (!nvdev || nvdev->destroy)
1749 return -ENODEV;
1750
1751 memset(&orig, 0, sizeof(orig));
1752 __netvsc_get_ringparam(nvdev, &orig);
1753
1754 new_tx = clamp_t(u32, ring->tx_pending,
1755 NETVSC_MIN_TX_SECTIONS, orig.tx_max_pending);
1756 new_rx = clamp_t(u32, ring->rx_pending,
1757 NETVSC_MIN_RX_SECTIONS, orig.rx_max_pending);
1758
1759 if (new_tx == orig.tx_pending &&
1760 new_rx == orig.rx_pending)
1761 return 0; /* no change */
1762
7c9f335a
HZ		 1763 device_info = netvsc_devinfo_get(nvdev);
1764
1765 if (!device_info)
1766 return -ENOMEM;
1767
1768 device_info->send_sections = new_tx;
1769 device_info->recv_sections = new_rx;
8b532797 1770
7b2ee50c
SH		 1771 ret = netvsc_detach(ndev, nvdev);
1772 if (ret)
7c9f335a 1773 goto out;
8b532797 1774
7c9f335a 1775 ret = netvsc_attach(ndev, device_info);
7b2ee50c 1776 if (ret) {
7c9f335a
HZ		 1777 device_info->send_sections = orig.tx_pending;
1778 device_info->recv_sections = orig.rx_pending;
8b532797 1779
7c9f335a 1780 if (netvsc_attach(ndev, device_info))
7b2ee50c
SH		 1781 netdev_err(ndev, "restoring ringparam failed");
1782 }
8b532797 1783
7c9f335a
HZ		 1784out:
1785 kfree(device_info);
8b532797 1786 return ret;
1787}
1788
d6792a5a
HZ		 1789static int netvsc_set_features(struct net_device *ndev,
1790 netdev_features_t features)
1791{
1792 netdev_features_t change = features ^ ndev->features;
1793 struct net_device_context *ndevctx = netdev_priv(ndev);
1794 struct netvsc_device *nvdev = rtnl_dereference(ndevctx->nvdev);
1795 struct ndis_offload_params offloads;
1796
1797 if (!nvdev || nvdev->destroy)
1798 return -ENODEV;
1799
1800 if (!(change & NETIF_F_LRO))
1801 return 0;
1802
1803 memset(&offloads, 0, sizeof(struct ndis_offload_params));
1804
1805 if (features & NETIF_F_LRO) {
1806 offloads.rsc_ip_v4 = NDIS_OFFLOAD_PARAMETERS_RSC_ENABLED;
1807 offloads.rsc_ip_v6 = NDIS_OFFLOAD_PARAMETERS_RSC_ENABLED;
1808 } else {
1809 offloads.rsc_ip_v4 = NDIS_OFFLOAD_PARAMETERS_RSC_DISABLED;
1810 offloads.rsc_ip_v6 = NDIS_OFFLOAD_PARAMETERS_RSC_DISABLED;
1811 }
1812
1813 return rndis_filter_set_offload_params(ndev, nvdev, &offloads);
1814}
1815
273de02a
HZ		 1816static u32 netvsc_get_msglevel(struct net_device *ndev)
1817{
1818 struct net_device_context *ndev_ctx = netdev_priv(ndev);
1819
1820 return ndev_ctx->msg_enable;
1821}
1822
1823static void netvsc_set_msglevel(struct net_device *ndev, u32 val)
1824{
1825 struct net_device_context *ndev_ctx = netdev_priv(ndev);
1826
1827 ndev_ctx->msg_enable = val;
1828}
1829
f82f4ad7
SH		 1830static const struct ethtool_ops ethtool_ops = {
1831 .get_drvinfo = netvsc_get_drvinfo,
273de02a
HZ		 1832 .get_msglevel = netvsc_get_msglevel,
1833 .set_msglevel = netvsc_set_msglevel,
f82f4ad7 1834 .get_link = ethtool_op_get_link,
4323b47c
SH		 1835 .get_ethtool_stats = netvsc_get_ethtool_stats,
1836 .get_sset_count = netvsc_get_sset_count,
1837 .get_strings = netvsc_get_strings,
59995370 1838 .get_channels = netvsc_get_channels,
b5960e6e 1839 .set_channels = netvsc_set_channels,
76d13b56 1840 .get_ts_info = ethtool_op_get_ts_info,
b448f4e8 1841 .get_rxnfc = netvsc_get_rxnfc,
4823eb2f 1842 .set_rxnfc = netvsc_set_rxnfc,
962f3fee 1843 .get_rxfh_key_size = netvsc_get_rxfh_key_size,
1844 .get_rxfh_indir_size = netvsc_rss_indir_size,
1845 .get_rxfh = netvsc_get_rxfh,
1846 .set_rxfh = netvsc_set_rxfh,
5e8456fd
PR		 1847 .get_link_ksettings = netvsc_get_link_ksettings,
1848 .set_link_ksettings = netvsc_set_link_ksettings,
8b532797 1849 .get_ringparam = netvsc_get_ringparam,
1850 .set_ringparam = netvsc_set_ringparam,
f82f4ad7
SH		 1851};
1852
df2fff28
GKH		 1853static const struct net_device_ops device_ops = {
1854 .ndo_open = netvsc_open,
1855 .ndo_stop = netvsc_close,
1856 .ndo_start_xmit = netvsc_start_xmit,
bee9d41b
SH		 1857 .ndo_change_rx_flags = netvsc_change_rx_flags,
1858 .ndo_set_rx_mode = netvsc_set_rx_mode,
d6792a5a 1859 .ndo_set_features = netvsc_set_features,
4d447c9a 1860 .ndo_change_mtu = netvsc_change_mtu,
b681b588 1861 .ndo_validate_addr = eth_validate_addr,
1ce09e89 1862 .ndo_set_mac_address = netvsc_set_mac_addr,
5b54dac8 1863 .ndo_select_queue = netvsc_select_queue,
7eafd9b4 1864 .ndo_get_stats64 = netvsc_get_stats64,
df2fff28
GKH		 1865};
1866
c996edcf 1867/*
27a70af3
VK		 1868 * Handle link status changes. For RNDIS_STATUS_NETWORK_CHANGE emulate link
1869 * down/up sequence. In case of RNDIS_STATUS_MEDIA_CONNECT when carrier is
1870 * present send GARP packet to network peers with netif_notify_peers().
c996edcf 1871 */
891de74d 1872static void netvsc_link_change(struct work_struct *w)
c996edcf 1873{
0a1275ca
VK		 1874 struct net_device_context *ndev_ctx =
1875 container_of(w, struct net_device_context, dwork.work);
1876 struct hv_device *device_obj = ndev_ctx->device_ctx;
1877 struct net_device *net = hv_get_drvdata(device_obj);
2ddd5e5f 1878 struct netvsc_device *net_device;
891de74d 1879 struct rndis_device *rdev;
27a70af3
VK		 1880 struct netvsc_reconfig *event = NULL;
1881 bool notify = false, reschedule = false;
1882 unsigned long flags, next_reconfig, delay;
c996edcf 1883
9b4e946c 1884 /* if changes are happening, comeback later */
1885 if (!rtnl_trylock()) {
1886 schedule_delayed_work(&ndev_ctx->dwork, LINKCHANGE_INT);
1887 return;
1888 }
1889
a0be450e 1890 net_device = rtnl_dereference(ndev_ctx->nvdev);
1891 if (!net_device)
1bdcec8a
VK		 1892 goto out_unlock;
1893
891de74d 1894 rdev = net_device->extension;
891de74d 1895
27a70af3
VK		 1896 next_reconfig = ndev_ctx->last_reconfig + LINKCHANGE_INT;
1897 if (time_is_after_jiffies(next_reconfig)) {
1898 /* link_watch only sends one notification with current state
1899 * per second, avoid doing reconfig more frequently. Handle
1900 * wrap around.
1901 */
1902 delay = next_reconfig - jiffies;
1903 delay = delay < LINKCHANGE_INT ? delay : LINKCHANGE_INT;
1904 schedule_delayed_work(&ndev_ctx->dwork, delay);
1bdcec8a 1905 goto out_unlock;
27a70af3
VK		 1906 }
1907 ndev_ctx->last_reconfig = jiffies;
1908
1909 spin_lock_irqsave(&ndev_ctx->lock, flags);
1910 if (!list_empty(&ndev_ctx->reconfig_events)) {
1911 event = list_first_entry(&ndev_ctx->reconfig_events,
1912 struct netvsc_reconfig, list);
1913 list_del(&event->list);
1914 reschedule = !list_empty(&ndev_ctx->reconfig_events);
1915 }
1916 spin_unlock_irqrestore(&ndev_ctx->lock, flags);
1917
1918 if (!event)
1bdcec8a 1919 goto out_unlock;
27a70af3
VK		 1920
1921 switch (event->event) {
1922 /* Only the following events are possible due to the check in
1923 * netvsc_linkstatus_callback()
1924 */
1925 case RNDIS_STATUS_MEDIA_CONNECT:
1926 if (rdev->link_state) {
1927 rdev->link_state = false;
0c195567 1928 netif_carrier_on(net);
1b704c4a 1929 netvsc_tx_enable(net_device, net);
27a70af3
VK		 1930 } else {
1931 notify = true;
1932 }
1933 kfree(event);
1934 break;
1935 case RNDIS_STATUS_MEDIA_DISCONNECT:
1936 if (!rdev->link_state) {
1937 rdev->link_state = true;
1938 netif_carrier_off(net);
1b704c4a 1939 netvsc_tx_disable(net_device, net);
27a70af3
VK		 1940 }
1941 kfree(event);
1942 break;
1943 case RNDIS_STATUS_NETWORK_CHANGE:
1944 /* Only makes sense if carrier is present */
1945 if (!rdev->link_state) {
1946 rdev->link_state = true;
1947 netif_carrier_off(net);
1b704c4a 1948 netvsc_tx_disable(net_device, net);
27a70af3
VK		 1949 event->event = RNDIS_STATUS_MEDIA_CONNECT;
1950 spin_lock_irqsave(&ndev_ctx->lock, flags);
15cfd407 1951 list_add(&event->list, &ndev_ctx->reconfig_events);
27a70af3
VK		 1952 spin_unlock_irqrestore(&ndev_ctx->lock, flags);
1953 reschedule = true;
3a494e71 1954 }
27a70af3 1955 break;
891de74d
HZ		 1956 }
1957
1958 rtnl_unlock();
1959
1960 if (notify)
1961 netdev_notify_peers(net);
27a70af3
VK		 1962
1963 /* link_watch only sends one notification with current state per
1964 * second, handle next reconfig event in 2 seconds.
1965 */
1966 if (reschedule)
1967 schedule_delayed_work(&ndev_ctx->dwork, LINKCHANGE_INT);
1bdcec8a
VK		 1968
1969 return;
1970
1971out_unlock:
1972 rtnl_unlock();
c996edcf
HZ		 1973}
1974
8cde8f0c
SH		 1975static struct net_device *get_netvsc_byref(struct net_device *vf_netdev)
1976{
7bf7bb37 1977 struct net_device_context *net_device_ctx;
8cde8f0c
SH		 1978 struct net_device *dev;
1979
7bf7bb37
SH		 1980 dev = netdev_master_upper_dev_get(vf_netdev);
1981 if (!dev || dev->netdev_ops != &device_ops)
1982 return NULL; /* not a netvsc device */
8cde8f0c 1983
7bf7bb37
SH		 1984 net_device_ctx = netdev_priv(dev);
1985 if (!rtnl_dereference(net_device_ctx->nvdev))
1986 return NULL; /* device is removed */
8cde8f0c 1987
7bf7bb37 1988 return dev;
8cde8f0c
SH		 1989}
1990
0c195567 1991/* Called when VF is injecting data into network stack.
1992 * Change the associated network device from VF to netvsc.
1993 * note: already called with rcu_read_lock
1994 */
1995static rx_handler_result_t netvsc_vf_handle_frame(struct sk_buff **pskb)
1996{
1997 struct sk_buff *skb = *pskb;
1998 struct net_device *ndev = rcu_dereference(skb->dev->rx_handler_data);
1999 struct net_device_context *ndev_ctx = netdev_priv(ndev);
2000 struct netvsc_vf_pcpu_stats *pcpu_stats
2001 = this_cpu_ptr(ndev_ctx->vf_stats);
2002
996ed047
SH		 2003 skb = skb_share_check(skb, GFP_ATOMIC);
2004 if (unlikely(!skb))
2005 return RX_HANDLER_CONSUMED;
2006
2007 *pskb = skb;
2008
0c195567 2009 skb->dev = ndev;
2010
2011 u64_stats_update_begin(&pcpu_stats->syncp);
2012 pcpu_stats->rx_packets++;
2013 pcpu_stats->rx_bytes += skb->len;
2014 u64_stats_update_end(&pcpu_stats->syncp);
2015
2016 return RX_HANDLER_ANOTHER;
2017}
2018
8cde8f0c
SH		 2019static int netvsc_vf_join(struct net_device *vf_netdev,
2020 struct net_device *ndev)
2021{
2022 struct net_device_context *ndev_ctx = netdev_priv(ndev);
2023 int ret;
2024
2025 ret = netdev_rx_handler_register(vf_netdev,
2026 netvsc_vf_handle_frame, ndev);
2027 if (ret != 0) {
2028 netdev_err(vf_netdev,
2029 "can not register netvsc VF receive handler (err = %d)\n",
2030 ret);
2031 goto rx_handler_failed;
2032 }
2033
2034 ret = netdev_master_upper_dev_link(vf_netdev, ndev,
2035 NULL, NULL, NULL);
2036 if (ret != 0) {
2037 netdev_err(vf_netdev,
2038 "can not set master device %s (err = %d)\n",
2039 ndev->name, ret);
2040 goto upper_link_failed;
2041 }
2042
2043 /* set slave flag before open to prevent IPv6 addrconf */
2044 vf_netdev->flags |= IFF_SLAVE;
2045
2046 schedule_delayed_work(&ndev_ctx->vf_takeover, VF_TAKEOVER_INT);
2047
2048 call_netdevice_notifiers(NETDEV_JOIN, vf_netdev);
2049
2050 netdev_info(vf_netdev, "joined to %s\n", ndev->name);
2051 return 0;
2052
2053upper_link_failed:
2054 netdev_rx_handler_unregister(vf_netdev);
2055rx_handler_failed:
2056 return ret;
2057}
2058
0c195567 2059static void __netvsc_vf_setup(struct net_device *ndev,
2060 struct net_device *vf_netdev)
2061{
2062 int ret;
2063
0c195567 2064 /* Align MTU of VF with master */
2065 ret = dev_set_mtu(vf_netdev, ndev->mtu);
2066 if (ret)
2067 netdev_warn(vf_netdev,
2068 "unable to change mtu to %u\n", ndev->mtu);
2069
bee9d41b 2070 /* set multicast etc flags on VF */
567c5e13 2071 dev_change_flags(vf_netdev, ndev->flags | IFF_SLAVE, NULL);
b0dee791
SH		 2072
2073 /* sync address list from ndev to VF */
2074 netif_addr_lock_bh(ndev);
bee9d41b
SH		 2075 dev_uc_sync(vf_netdev, ndev);
2076 dev_mc_sync(vf_netdev, ndev);
b0dee791 2077 netif_addr_unlock_bh(ndev);
bee9d41b 2078
0c195567 2079 if (netif_running(ndev)) {
00f54e68 2080 ret = dev_open(vf_netdev, NULL);
0c195567 2081 if (ret)
2082 netdev_warn(vf_netdev,
2083 "unable to open: %d\n", ret);
2084 }
2085}
2086
2087/* Setup VF as slave of the synthetic device.
2088 * Runs in workqueue to avoid recursion in netlink callbacks.
2089 */
2090static void netvsc_vf_setup(struct work_struct *w)
2091{
2092 struct net_device_context *ndev_ctx
6123c668 2093 = container_of(w, struct net_device_context, vf_takeover.work);
0c195567 2094 struct net_device *ndev = hv_get_drvdata(ndev_ctx->device_ctx);
2095 struct net_device *vf_netdev;
2096
fb84af8a 2097 if (!rtnl_trylock()) {
6123c668 2098 schedule_delayed_work(&ndev_ctx->vf_takeover, 0);
fb84af8a 2099 return;
2100 }
2101
0c195567 2102 vf_netdev = rtnl_dereference(ndev_ctx->vf_netdev);
2103 if (vf_netdev)
2104 __netvsc_vf_setup(ndev, vf_netdev);
2105
2106 rtnl_unlock();
2107}
2108
00547955
HZ		 2109/* Find netvsc by VF serial number.
2110 * The PCI hyperv controller records the serial number as the slot kobj name.
00d7ddba
SH		 2111 */
2112static struct net_device *get_netvsc_byslot(const struct net_device *vf_netdev)
2113{
2114 struct device *parent = vf_netdev->dev.parent;
2115 struct net_device_context *ndev_ctx;
2116 struct pci_dev *pdev;
00547955 2117 u32 serial;
00d7ddba
SH		 2118
2119 if (!parent || !dev_is_pci(parent))
2120 return NULL; /* not a PCI device */
2121
2122 pdev = to_pci_dev(parent);
2123 if (!pdev->slot) {
2124 netdev_notice(vf_netdev, "no PCI slot information\n");
2125 return NULL;
2126 }
2127
00547955
HZ		 2128 if (kstrtou32(pci_slot_name(pdev->slot), 10, &serial)) {
2129 netdev_notice(vf_netdev, "Invalid vf serial:%s\n",
2130 pci_slot_name(pdev->slot));
2131 return NULL;
2132 }
2133
00d7ddba
SH		 2134 list_for_each_entry(ndev_ctx, &netvsc_dev_list, list) {
2135 if (!ndev_ctx->vf_alloc)
2136 continue;
2137
00547955 2138 if (ndev_ctx->vf_serial == serial)
00d7ddba
SH		 2139 return hv_get_drvdata(ndev_ctx->device_ctx);
2140 }
2141
2142 netdev_notice(vf_netdev,
00547955 2143 "no netdev found for vf serial:%u\n", serial);
00d7ddba
SH		 2144 return NULL;
2145}
2146
8cde8f0c 2147static int netvsc_register_vf(struct net_device *vf_netdev)
84bf9cef 2148{
0a1275ca 2149 struct net_device_context *net_device_ctx;
84bf9cef 2150 struct netvsc_device *netvsc_dev;
00d7ddba 2151 struct net_device *ndev;
c0a41b88 2152 int ret;
84bf9cef 2153
8cde8f0c
SH		 2154 if (vf_netdev->addr_len != ETH_ALEN)
2155 return NOTIFY_DONE;
2156
00d7ddba 2157 ndev = get_netvsc_byslot(vf_netdev);
8cde8f0c
SH		 2158 if (!ndev)
2159 return NOTIFY_DONE;
2160
0a1275ca 2161 net_device_ctx = netdev_priv(ndev);
545a8e79 2162 netvsc_dev = rtnl_dereference(net_device_ctx->nvdev);
f207c10d 2163 if (!netvsc_dev || rtnl_dereference(net_device_ctx->vf_netdev))
8cde8f0c 2164 return NOTIFY_DONE;
0c195567 2165
52d3b494 2166 /* if synthetic interface is a different namespace,
c0a41b88
SH		 2167 * then move the VF to that namespace; join will be
2168 * done again in that context.
2169 */
2170 if (!net_eq(dev_net(ndev), dev_net(vf_netdev))) {
2171 ret = dev_change_net_namespace(vf_netdev,
2172 dev_net(ndev), "eth%d");
2173 if (ret)
2174 netdev_err(vf_netdev,
2175 "could not move to same namespace as %s: %d\n",
2176 ndev->name, ret);
2177 else
2178 netdev_info(vf_netdev,
2179 "VF moved to namespace with: %s\n",
2180 ndev->name);
8cde8f0c 2181 return NOTIFY_DONE;
c0a41b88 2182 }
1ff78076 2183
8cde8f0c 2184 netdev_info(ndev, "VF registering: %s\n", vf_netdev->name);
0c195567 2185
c0a41b88
SH		 2186 if (netvsc_vf_join(vf_netdev, ndev) != 0)
2187 return NOTIFY_DONE;
2188
07d0f000 2189 dev_hold(vf_netdev);
8cde8f0c
SH		 2190 rcu_assign_pointer(net_device_ctx->vf_netdev, vf_netdev);
2191 return NOTIFY_OK;
84bf9cef
KS		 2192}
2193
9a0c48df 2194/* VF up/down change detected, schedule to change data path */
8cde8f0c 2195static int netvsc_vf_changed(struct net_device *vf_netdev)
84bf9cef 2196{
7b83f520 2197 struct net_device_context *net_device_ctx;
84bf9cef 2198 struct netvsc_device *netvsc_dev;
8cde8f0c 2199 struct net_device *ndev;
9a0c48df 2200 bool vf_is_up = netif_running(vf_netdev);
fb84af8a 2201
8cde8f0c
SH		 2202 ndev = get_netvsc_byref(vf_netdev);
2203 if (!ndev)
2204 return NOTIFY_DONE;
2205
7b83f520 2206 net_device_ctx = netdev_priv(ndev);
2207 netvsc_dev = rtnl_dereference(net_device_ctx->nvdev);
0c195567 2208 if (!netvsc_dev)
8cde8f0c 2209 return NOTIFY_DONE;
84bf9cef 2210
9a0c48df
SH		 2211 netvsc_switch_datapath(ndev, vf_is_up);
2212 netdev_info(ndev, "Data path switched %s VF: %s\n",
2213 vf_is_up ? "to" : "from", vf_netdev->name);
84bf9cef 2214
8cde8f0c 2215 return NOTIFY_OK;
84bf9cef
KS		 2216}
2217
8cde8f0c 2218static int netvsc_unregister_vf(struct net_device *vf_netdev)
84bf9cef 2219{
8cde8f0c 2220 struct net_device *ndev;
0a1275ca 2221 struct net_device_context *net_device_ctx;
84bf9cef 2222
8cde8f0c
SH		 2223 ndev = get_netvsc_byref(vf_netdev);
2224 if (!ndev)
2225 return NOTIFY_DONE;
1ff78076
SS		 2226
2227 net_device_ctx = netdev_priv(ndev);
8cde8f0c 2228 cancel_delayed_work_sync(&net_device_ctx->vf_takeover);
1ff78076 2229
0a1275ca 2230 netdev_info(ndev, "VF unregistering: %s\n", vf_netdev->name);
f207c10d 2231
8cde8f0c
SH		 2232 netdev_rx_handler_unregister(vf_netdev);
2233 netdev_upper_dev_unlink(vf_netdev, ndev);
f207c10d 2234 RCU_INIT_POINTER(net_device_ctx->vf_netdev, NULL);
07d0f000 2235 dev_put(vf_netdev);
ec158f77 2236
8cde8f0c 2237 return NOTIFY_OK;
84bf9cef
KS		 2238}
2239
84946899
S		 2240static int netvsc_probe(struct hv_device *dev,
2241 const struct hv_vmbus_device_id *dev_id)
df2fff28 2242{
df2fff28
GKH		 2243 struct net_device *net = NULL;
2244 struct net_device_context *net_device_ctx;
7c9f335a 2245 struct netvsc_device_info *device_info = NULL;
5b54dac8 2246 struct netvsc_device *nvdev;
0c195567 2247 int ret = -ENOMEM;
df2fff28 2248
5b54dac8 2249 net = alloc_etherdev_mq(sizeof(struct net_device_context),
2b01888d 2250 VRSS_CHANNEL_MAX);
df2fff28 2251 if (!net)
0c195567 2252 goto no_net;
df2fff28 2253
1b07da51
HZ		 2254 netif_carrier_off(net);
2255
b37879e6
HZ		 2256 netvsc_init_settings(net);
2257
df2fff28 2258 net_device_ctx = netdev_priv(net);
9efd21e1 2259 net_device_ctx->device_ctx = dev;
3f300ff4
SX		 2260 net_device_ctx->msg_enable = netif_msg_init(debug, default_msg);
2261 if (netif_msg_probe(net_device_ctx))
2262 netdev_dbg(net, "netvsc msg_enable: %d\n",
2263 net_device_ctx->msg_enable);
2264
2ddd5e5f 2265 hv_set_drvdata(dev, net);
f580aec4 2266
891de74d 2267 INIT_DELAYED_WORK(&net_device_ctx->dwork, netvsc_link_change);
df2fff28 2268
27a70af3
VK		 2269 spin_lock_init(&net_device_ctx->lock);
2270 INIT_LIST_HEAD(&net_device_ctx->reconfig_events);
6123c668 2271 INIT_DELAYED_WORK(&net_device_ctx->vf_takeover, netvsc_vf_setup);
0c195567 2272
2273 net_device_ctx->vf_stats
2274 = netdev_alloc_pcpu_stats(struct netvsc_vf_pcpu_stats);
2275 if (!net_device_ctx->vf_stats)
2276 goto no_stats;
27a70af3 2277
df2fff28 2278 net->netdev_ops = &device_ops;
7ad24ea4 2279 net->ethtool_ops = &ethtool_ops;
9efd21e1 2280 SET_NETDEV_DEV(net, &dev->device);
df2fff28 2281
14a03cf8
VK		 2282 /* We always need headroom for rndis header */
2283 net->needed_headroom = RNDIS_AND_PPI_SIZE;
2284
6450f8f2
HZ		 2285 /* Initialize the number of queues to be 1, we may change it if more
2286 * channels are offered later.
2287 */
2288 netif_set_real_num_tx_queues(net, 1);
2289 netif_set_real_num_rx_queues(net, 1);
2290
692e084e 2291 /* Notify the netvsc driver of the new device */
7c9f335a
HZ		 2292 device_info = netvsc_devinfo_get(NULL);
2293
2294 if (!device_info) {
2295 ret = -ENOMEM;
2296 goto devinfo_failed;
2297 }
2298
2299 nvdev = rndis_filter_device_add(dev, device_info);
9749fed5 2300 if (IS_ERR(nvdev)) {
2301 ret = PTR_ERR(nvdev);
692e084e 2302 netdev_err(net, "unable to add netvsc device (ret %d)\n", ret);
0c195567 2303 goto rndis_failed;
df2fff28 2304 }
0c195567 2305
7c9f335a 2306 memcpy(net->dev_addr, device_info->mac_adr, ETH_ALEN);
692e084e 2307
e04e7a7b
DC		 2308 /* We must get rtnl lock before scheduling nvdev->subchan_work,
2309 * otherwise netvsc_subchan_work() can get rtnl lock first and wait
2310 * all subchannels to show up, but that may not happen because
2311 * netvsc_probe() can't get rtnl lock and as a result vmbus_onoffer()
2312 * -> ... -> device_add() -> ... -> __device_attach() can't get
2313 * the device lock, so all the subchannels can't be processed --
52d3b494 2314 * finally netvsc_subchan_work() hangs forever.
e04e7a7b
DC		 2315 */
2316 rtnl_lock();
2317
3ffe64f1
SH		 2318 if (nvdev->num_chn > 1)
2319 schedule_work(&nvdev->subchan_work);
2320
aefd80e8 2321 /* hw_features computed in rndis_netdev_set_hwcaps() */
23312a3b 2322 net->features = net->hw_features |
2323 NETIF_F_HIGHDMA | NETIF_F_SG |
2324 NETIF_F_HW_VLAN_CTAG_TX | NETIF_F_HW_VLAN_CTAG_RX;
2325 net->vlan_features = net->features;
2326
9749fed5 2327 netdev_lockdep_set_classes(net);
2328
d0c2c997
JW		 2329 /* MTU range: 68 - 1500 or 65521 */
2330 net->min_mtu = NETVSC_MTU_MIN;
2331 if (nvdev->nvsp_version >= NVSP_PROTOCOL_VERSION_2)
2332 net->max_mtu = NETVSC_MTU - ETH_HLEN;
2333 else
2334 net->max_mtu = ETH_DATA_LEN;
2335
7bf7bb37 2336 ret = register_netdevice(net);
a68f9614
HZ		 2337 if (ret != 0) {
2338 pr_err("Unable to register netdev.\n");
0c195567 2339 goto register_failed;
a68f9614
HZ		 2340 }
2341
7bf7bb37
SH		 2342 list_add(&net_device_ctx->list, &netvsc_dev_list);
2343 rtnl_unlock();
7c9f335a
HZ		 2344
2345 kfree(device_info);
7bf7bb37 2346 return 0;
0c195567 2347
2348register_failed:
7bf7bb37 2349 rtnl_unlock();
0c195567 2350 rndis_filter_device_remove(dev, nvdev);
2351rndis_failed:
7c9f335a
HZ		 2352 kfree(device_info);
2353devinfo_failed:
0c195567 2354 free_percpu(net_device_ctx->vf_stats);
2355no_stats:
2356 hv_set_drvdata(dev, NULL);
2357 free_netdev(net);
2358no_net:
2359 return ret;
df2fff28
GKH		 2360}
2361
415b023a 2362static int netvsc_remove(struct hv_device *dev)
df2fff28 2363{
122a5f64 2364 struct net_device_context *ndev_ctx;
7b2ee50c
SH		 2365 struct net_device *vf_netdev, *net;
2366 struct netvsc_device *nvdev;
2ddd5e5f 2367
3d541ac5 2368 net = hv_get_drvdata(dev);
df2fff28 2369 if (net == NULL) {
415b023a 2370 dev_err(&dev->device, "No net device to remove\n");
df2fff28
GKH		 2371 return 0;
2372 }
2373
122a5f64 2374 ndev_ctx = netdev_priv(net);
3d541ac5 2375
122a5f64
HZ		 2376 cancel_delayed_work_sync(&ndev_ctx->dwork);
2377
018349d7
SH		 2378 rtnl_lock();
2379 nvdev = rtnl_dereference(ndev_ctx->nvdev);
2380 if (nvdev)
7b2ee50c
SH		 2381 cancel_work_sync(&nvdev->subchan_work);
2382
df2fff28
GKH		 2383 /*
2384 * Call to the vsc driver to let it know that the device is being
a0be450e 2385 * removed. Also blocks mtu and channel changes.
df2fff28 2386 */
ec158f77
SH		 2387 vf_netdev = rtnl_dereference(ndev_ctx->vf_netdev);
2388 if (vf_netdev)
8cde8f0c 2389 netvsc_unregister_vf(vf_netdev);
ec158f77 2390
7b2ee50c
SH		 2391 if (nvdev)
2392 rndis_filter_device_remove(dev, nvdev);
2393
8195b139 2394 unregister_netdevice(net);
7bf7bb37 2395 list_del(&ndev_ctx->list);
8195b139 2396
a0be450e 2397 rtnl_unlock();
2398
3d541ac5
VK		 2399 hv_set_drvdata(dev, NULL);
2400
0c195567 2401 free_percpu(ndev_ctx->vf_stats);
6c80f3fc 2402 free_netdev(net);
df06bcff 2403 return 0;
df2fff28
GKH		 2404}
2405
345c4cc3 2406static const struct hv_vmbus_device_id id_table[] = {
c45cf2d4 2407 /* Network guid */
8f505944 2408 { HV_NIC_GUID, },
c45cf2d4 2409 { },
345c4cc3
S		 2410};
2411
2412MODULE_DEVICE_TABLE(vmbus, id_table);
2413
f1542a66 2414/* The one and only one */
fde0ef9b 2415static struct hv_driver netvsc_drv = {
d31b20fc 2416 .name = KBUILD_MODNAME,
345c4cc3 2417 .id_table = id_table,
fde0ef9b
S		 2418 .probe = netvsc_probe,
2419 .remove = netvsc_remove,
af0a5646 2420 .driver = {
9a33629b 2421 .probe_type = PROBE_FORCE_SYNCHRONOUS,
af0a5646 2422 },
d4890970 2423};
f1542a66 2424
8cde8f0c
SH		 2425/*
2426 * On Hyper-V, every VF interface is matched with a corresponding
2427 * synthetic interface. The synthetic interface is presented first
2428 * to the guest. When the corresponding VF instance is registered,
2429 * we will take care of switching the data path.
2430 */
2431static int netvsc_netdev_event(struct notifier_block *this,
2432 unsigned long event, void *ptr)
2433{
2434 struct net_device *event_dev = netdev_notifier_info_to_dev(ptr);
2435
2436 /* Skip our own events */
2437 if (event_dev->netdev_ops == &device_ops)
2438 return NOTIFY_DONE;
2439
2440 /* Avoid non-Ethernet type devices */
2441 if (event_dev->type != ARPHRD_ETHER)
2442 return NOTIFY_DONE;
2443
2444 /* Avoid Vlan dev with same MAC registering as VF */
2445 if (is_vlan_dev(event_dev))
2446 return NOTIFY_DONE;
2447
2448 /* Avoid Bonding master dev with same MAC registering as VF */
2449 if ((event_dev->priv_flags & IFF_BONDING) &&
2450 (event_dev->flags & IFF_MASTER))
2451 return NOTIFY_DONE;
2452
2453 switch (event) {
2454 case NETDEV_REGISTER:
2455 return netvsc_register_vf(event_dev);
2456 case NETDEV_UNREGISTER:
2457 return netvsc_unregister_vf(event_dev);
2458 case NETDEV_UP:
2459 case NETDEV_DOWN:
2460 return netvsc_vf_changed(event_dev);
2461 default:
2462 return NOTIFY_DONE;
2463 }
2464}
2465
2466static struct notifier_block netvsc_netdev_notifier = {
2467 .notifier_call = netvsc_netdev_event,
2468};
2469
a9869c94 2470static void __exit netvsc_drv_exit(void)
fceaf24a 2471{
8cde8f0c 2472 unregister_netdevice_notifier(&netvsc_netdev_notifier);
768fa219 2473 vmbus_driver_unregister(&netvsc_drv);
fceaf24a
HJ		 2474}
2475
1fde28cf 2476static int __init netvsc_drv_init(void)
df2fff28 2477{
84bf9cef
KS		 2478 int ret;
2479
fa85a6c2
HZ		 2480 if (ring_size < RING_SIZE_MIN) {
2481 ring_size = RING_SIZE_MIN;
a7f99d0f 2482 pr_info("Increased ring_size to %u (min allowed)\n",
fa85a6c2
HZ		 2483 ring_size);
2484 }
a7f99d0f 2485 netvsc_ring_bytes = ring_size * PAGE_SIZE;
84bf9cef 2486
a7f99d0f 2487 ret = vmbus_driver_register(&netvsc_drv);
84bf9cef
KS		 2488 if (ret)
2489 return ret;
2490
8cde8f0c 2491 register_netdevice_notifier(&netvsc_netdev_notifier);
84bf9cef 2492 return 0;
df2fff28
GKH		 2493}
2494
26c14cc1 2495MODULE_LICENSE("GPL");
7880fc54 2496MODULE_DESCRIPTION("Microsoft Hyper-V network driver");
fceaf24a 2497
1fde28cf 2498module_init(netvsc_drv_init);
a9869c94 2499module_exit(netvsc_drv_exit);
Ubuntu Eoan Linux kernel mirror