Commit | Line | Data |
---|---|---|
fceaf24a | 1 | /* |
fceaf24a HJ |
2 | * Copyright (c) 2009, Microsoft Corporation. |
3 | * | |
4 | * This program is free software; you can redistribute it and/or modify it | |
5 | * under the terms and conditions of the GNU General Public License, | |
6 | * version 2, as published by the Free Software Foundation. | |
7 | * | |
8 | * This program is distributed in the hope it will be useful, but WITHOUT | |
9 | * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or | |
10 | * FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License for | |
11 | * more details. | |
12 | * | |
13 | * You should have received a copy of the GNU General Public License along with | |
adf8d3ff | 14 | * this program; if not, see <http://www.gnu.org/licenses/>. |
fceaf24a HJ |
15 | * |
16 | * Authors: | |
d0e94d17 | 17 | * Haiyang Zhang <haiyangz@microsoft.com> |
fceaf24a | 18 | * Hank Janssen <hjanssen@microsoft.com> |
fceaf24a | 19 | */ |
eb335bc4 HJ |
20 | #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt |
21 | ||
fceaf24a | 22 | #include <linux/init.h> |
9079ce69 | 23 | #include <linux/atomic.h> |
fceaf24a HJ |
24 | #include <linux/module.h> |
25 | #include <linux/highmem.h> | |
26 | #include <linux/device.h> | |
fceaf24a | 27 | #include <linux/io.h> |
fceaf24a HJ |
28 | #include <linux/delay.h> |
29 | #include <linux/netdevice.h> | |
30 | #include <linux/inetdevice.h> | |
31 | #include <linux/etherdevice.h> | |
32 | #include <linux/skbuff.h> | |
c802db11 | 33 | #include <linux/if_vlan.h> |
fceaf24a | 34 | #include <linux/in.h> |
5a0e3ad6 | 35 | #include <linux/slab.h> |
fceaf24a HJ |
36 | #include <net/arp.h> |
37 | #include <net/route.h> | |
38 | #include <net/sock.h> | |
39 | #include <net/pkt_sched.h> | |
3f335ea2 | 40 | |
5ca7252a | 41 | #include "hyperv_net.h" |
fceaf24a | 42 | |
fceaf24a | 43 | |
fa85a6c2 | 44 | #define RING_SIZE_MIN 64 |
27a70af3 | 45 | #define LINKCHANGE_INT (2 * HZ) |
a060679c | 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) | |
99c8da0f | 51 | static int ring_size = 128; |
450d7a4b SH |
52 | module_param(ring_size, int, S_IRUGO); |
53 | MODULE_PARM_DESC(ring_size, "Ring buffer size (# of pages)"); | |
fceaf24a | 54 | |
e01ec219 KS |
55 | static int max_num_vrss_chns = 8; |
56 | ||
3f300ff4 SX |
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 | ||
d426b2e3 HZ |
66 | static void do_set_multicast(struct work_struct *w) |
67 | { | |
792df872 WM |
68 | struct net_device_context *ndevctx = |
69 | container_of(w, struct net_device_context, work); | |
d426b2e3 HZ |
70 | struct netvsc_device *nvdev; |
71 | struct rndis_device *rdev; | |
72 | ||
73 | nvdev = hv_get_drvdata(ndevctx->device_ctx); | |
792df872 WM |
74 | if (nvdev == NULL || nvdev->ndev == NULL) |
75 | return; | |
d426b2e3 HZ |
76 | |
77 | rdev = nvdev->extension; | |
78 | if (rdev == NULL) | |
792df872 | 79 | return; |
d426b2e3 | 80 | |
792df872 | 81 | if (nvdev->ndev->flags & IFF_PROMISC) |
d426b2e3 HZ |
82 | rndis_filter_set_packet_filter(rdev, |
83 | NDIS_PACKET_TYPE_PROMISCUOUS); | |
84 | else | |
85 | rndis_filter_set_packet_filter(rdev, | |
86 | NDIS_PACKET_TYPE_BROADCAST | | |
87 | NDIS_PACKET_TYPE_ALL_MULTICAST | | |
88 | NDIS_PACKET_TYPE_DIRECTED); | |
d426b2e3 HZ |
89 | } |
90 | ||
4e9bfefa | 91 | static void netvsc_set_multicast_list(struct net_device *net) |
fceaf24a | 92 | { |
792df872 | 93 | struct net_device_context *net_device_ctx = netdev_priv(net); |
d426b2e3 | 94 | |
792df872 | 95 | schedule_work(&net_device_ctx->work); |
fceaf24a HJ |
96 | } |
97 | ||
fceaf24a HJ |
98 | static int netvsc_open(struct net_device *net) |
99 | { | |
fceaf24a | 100 | struct net_device_context *net_device_ctx = netdev_priv(net); |
6bad88da | 101 | struct hv_device *device_obj = net_device_ctx->device_ctx; |
891de74d HZ |
102 | struct netvsc_device *nvdev; |
103 | struct rndis_device *rdev; | |
02fafbc6 | 104 | int ret = 0; |
fceaf24a | 105 | |
891de74d HZ |
106 | netif_carrier_off(net); |
107 | ||
d515d0ff HZ |
108 | /* Open up the device */ |
109 | ret = rndis_filter_open(device_obj); | |
110 | if (ret != 0) { | |
111 | netdev_err(net, "unable to open device (ret %d).\n", ret); | |
112 | return ret; | |
fceaf24a HJ |
113 | } |
114 | ||
2de8530b | 115 | netif_tx_wake_all_queues(net); |
d515d0ff | 116 | |
891de74d HZ |
117 | nvdev = hv_get_drvdata(device_obj); |
118 | rdev = nvdev->extension; | |
119 | if (!rdev->link_state) | |
120 | netif_carrier_on(net); | |
121 | ||
fceaf24a HJ |
122 | return ret; |
123 | } | |
124 | ||
fceaf24a HJ |
125 | static int netvsc_close(struct net_device *net) |
126 | { | |
fceaf24a | 127 | struct net_device_context *net_device_ctx = netdev_priv(net); |
6bad88da | 128 | struct hv_device *device_obj = net_device_ctx->device_ctx; |
2de8530b | 129 | struct netvsc_device *nvdev = hv_get_drvdata(device_obj); |
02fafbc6 | 130 | int ret; |
2de8530b HZ |
131 | u32 aread, awrite, i, msec = 10, retry = 0, retry_max = 20; |
132 | struct vmbus_channel *chn; | |
fceaf24a | 133 | |
0a282538 | 134 | netif_tx_disable(net); |
fceaf24a | 135 | |
792df872 WM |
136 | /* Make sure netvsc_set_multicast_list doesn't re-enable filter! */ |
137 | cancel_work_sync(&net_device_ctx->work); | |
9c26aa0d | 138 | ret = rndis_filter_close(device_obj); |
2de8530b | 139 | if (ret != 0) { |
eb335bc4 | 140 | netdev_err(net, "unable to close device (ret %d).\n", ret); |
2de8530b HZ |
141 | return ret; |
142 | } | |
143 | ||
144 | /* Ensure pending bytes in ring are read */ | |
145 | while (true) { | |
146 | aread = 0; | |
147 | for (i = 0; i < nvdev->num_chn; i++) { | |
148 | chn = nvdev->chn_table[i]; | |
149 | if (!chn) | |
150 | continue; | |
151 | ||
152 | hv_get_ringbuffer_availbytes(&chn->inbound, &aread, | |
153 | &awrite); | |
154 | ||
155 | if (aread) | |
156 | break; | |
157 | ||
158 | hv_get_ringbuffer_availbytes(&chn->outbound, &aread, | |
159 | &awrite); | |
160 | ||
161 | if (aread) | |
162 | break; | |
163 | } | |
164 | ||
165 | retry++; | |
166 | if (retry > retry_max || aread == 0) | |
167 | break; | |
168 | ||
169 | msleep(msec); | |
170 | ||
171 | if (msec < 1000) | |
172 | msec *= 2; | |
173 | } | |
174 | ||
175 | if (aread) { | |
176 | netdev_err(net, "Ring buffer not empty after closing rndis\n"); | |
177 | ret = -ETIMEDOUT; | |
178 | } | |
fceaf24a | 179 | |
fceaf24a HJ |
180 | return ret; |
181 | } | |
182 | ||
8a00251a KS |
183 | static void *init_ppi_data(struct rndis_message *msg, u32 ppi_size, |
184 | int pkt_type) | |
185 | { | |
186 | struct rndis_packet *rndis_pkt; | |
187 | struct rndis_per_packet_info *ppi; | |
188 | ||
189 | rndis_pkt = &msg->msg.pkt; | |
190 | rndis_pkt->data_offset += ppi_size; | |
191 | ||
192 | ppi = (struct rndis_per_packet_info *)((void *)rndis_pkt + | |
193 | rndis_pkt->per_pkt_info_offset + rndis_pkt->per_pkt_info_len); | |
194 | ||
195 | ppi->size = ppi_size; | |
196 | ppi->type = pkt_type; | |
197 | ppi->ppi_offset = sizeof(struct rndis_per_packet_info); | |
198 | ||
199 | rndis_pkt->per_pkt_info_len += ppi_size; | |
200 | ||
201 | return ppi; | |
202 | } | |
203 | ||
5b54dac8 HZ |
204 | static u16 netvsc_select_queue(struct net_device *ndev, struct sk_buff *skb, |
205 | void *accel_priv, select_queue_fallback_t fallback) | |
206 | { | |
207 | struct net_device_context *net_device_ctx = netdev_priv(ndev); | |
208 | struct hv_device *hdev = net_device_ctx->device_ctx; | |
209 | struct netvsc_device *nvsc_dev = hv_get_drvdata(hdev); | |
210 | u32 hash; | |
211 | u16 q_idx = 0; | |
212 | ||
213 | if (nvsc_dev == NULL || ndev->real_num_tx_queues <= 1) | |
214 | return 0; | |
215 | ||
757647e1 VK |
216 | hash = skb_get_hash(skb); |
217 | q_idx = nvsc_dev->send_table[hash % VRSS_SEND_TAB_SIZE] % | |
218 | ndev->real_num_tx_queues; | |
5b54dac8 | 219 | |
8b9fbe1a VK |
220 | if (!nvsc_dev->chn_table[q_idx]) |
221 | q_idx = 0; | |
222 | ||
5b54dac8 HZ |
223 | return q_idx; |
224 | } | |
225 | ||
54a7357f KS |
226 | static u32 fill_pg_buf(struct page *page, u32 offset, u32 len, |
227 | struct hv_page_buffer *pb) | |
228 | { | |
229 | int j = 0; | |
230 | ||
231 | /* Deal with compund pages by ignoring unused part | |
232 | * of the page. | |
233 | */ | |
234 | page += (offset >> PAGE_SHIFT); | |
235 | offset &= ~PAGE_MASK; | |
236 | ||
237 | while (len > 0) { | |
238 | unsigned long bytes; | |
239 | ||
240 | bytes = PAGE_SIZE - offset; | |
241 | if (bytes > len) | |
242 | bytes = len; | |
243 | pb[j].pfn = page_to_pfn(page); | |
244 | pb[j].offset = offset; | |
245 | pb[j].len = bytes; | |
246 | ||
247 | offset += bytes; | |
248 | len -= bytes; | |
249 | ||
250 | if (offset == PAGE_SIZE && len) { | |
251 | page++; | |
252 | offset = 0; | |
253 | j++; | |
254 | } | |
255 | } | |
256 | ||
257 | return j + 1; | |
258 | } | |
259 | ||
8a00251a | 260 | static u32 init_page_array(void *hdr, u32 len, struct sk_buff *skb, |
a9f2e2d6 KS |
261 | struct hv_netvsc_packet *packet, |
262 | struct hv_page_buffer **page_buf) | |
54a7357f | 263 | { |
a9f2e2d6 | 264 | struct hv_page_buffer *pb = *page_buf; |
54a7357f KS |
265 | u32 slots_used = 0; |
266 | char *data = skb->data; | |
267 | int frags = skb_shinfo(skb)->nr_frags; | |
268 | int i; | |
269 | ||
270 | /* The packet is laid out thus: | |
aa0a34be | 271 | * 1. hdr: RNDIS header and PPI |
54a7357f KS |
272 | * 2. skb linear data |
273 | * 3. skb fragment data | |
274 | */ | |
275 | if (hdr != NULL) | |
276 | slots_used += fill_pg_buf(virt_to_page(hdr), | |
277 | offset_in_page(hdr), | |
278 | len, &pb[slots_used]); | |
279 | ||
aa0a34be HZ |
280 | packet->rmsg_size = len; |
281 | packet->rmsg_pgcnt = slots_used; | |
282 | ||
54a7357f KS |
283 | slots_used += fill_pg_buf(virt_to_page(data), |
284 | offset_in_page(data), | |
285 | skb_headlen(skb), &pb[slots_used]); | |
286 | ||
287 | for (i = 0; i < frags; i++) { | |
288 | skb_frag_t *frag = skb_shinfo(skb)->frags + i; | |
289 | ||
290 | slots_used += fill_pg_buf(skb_frag_page(frag), | |
291 | frag->page_offset, | |
292 | skb_frag_size(frag), &pb[slots_used]); | |
293 | } | |
8a00251a | 294 | return slots_used; |
54a7357f KS |
295 | } |
296 | ||
297 | static int count_skb_frag_slots(struct sk_buff *skb) | |
298 | { | |
299 | int i, frags = skb_shinfo(skb)->nr_frags; | |
300 | int pages = 0; | |
301 | ||
302 | for (i = 0; i < frags; i++) { | |
303 | skb_frag_t *frag = skb_shinfo(skb)->frags + i; | |
304 | unsigned long size = skb_frag_size(frag); | |
305 | unsigned long offset = frag->page_offset; | |
306 | ||
307 | /* Skip unused frames from start of page */ | |
308 | offset &= ~PAGE_MASK; | |
309 | pages += PFN_UP(offset + size); | |
310 | } | |
311 | return pages; | |
312 | } | |
313 | ||
314 | static int netvsc_get_slots(struct sk_buff *skb) | |
315 | { | |
316 | char *data = skb->data; | |
317 | unsigned int offset = offset_in_page(data); | |
318 | unsigned int len = skb_headlen(skb); | |
319 | int slots; | |
320 | int frag_slots; | |
321 | ||
322 | slots = DIV_ROUND_UP(offset + len, PAGE_SIZE); | |
323 | frag_slots = count_skb_frag_slots(skb); | |
324 | return slots + frag_slots; | |
325 | } | |
326 | ||
08cd04bf KS |
327 | static u32 get_net_transport_info(struct sk_buff *skb, u32 *trans_off) |
328 | { | |
329 | u32 ret_val = TRANSPORT_INFO_NOT_IP; | |
330 | ||
331 | if ((eth_hdr(skb)->h_proto != htons(ETH_P_IP)) && | |
332 | (eth_hdr(skb)->h_proto != htons(ETH_P_IPV6))) { | |
333 | goto not_ip; | |
334 | } | |
335 | ||
336 | *trans_off = skb_transport_offset(skb); | |
337 | ||
338 | if ((eth_hdr(skb)->h_proto == htons(ETH_P_IP))) { | |
339 | struct iphdr *iphdr = ip_hdr(skb); | |
340 | ||
341 | if (iphdr->protocol == IPPROTO_TCP) | |
342 | ret_val = TRANSPORT_INFO_IPV4_TCP; | |
343 | else if (iphdr->protocol == IPPROTO_UDP) | |
344 | ret_val = TRANSPORT_INFO_IPV4_UDP; | |
345 | } else { | |
346 | if (ipv6_hdr(skb)->nexthdr == IPPROTO_TCP) | |
347 | ret_val = TRANSPORT_INFO_IPV6_TCP; | |
348 | else if (ipv6_hdr(skb)->nexthdr == IPPROTO_UDP) | |
349 | ret_val = TRANSPORT_INFO_IPV6_UDP; | |
350 | } | |
351 | ||
352 | not_ip: | |
353 | return ret_val; | |
354 | } | |
355 | ||
02fafbc6 | 356 | static int netvsc_start_xmit(struct sk_buff *skb, struct net_device *net) |
fceaf24a | 357 | { |
fceaf24a | 358 | struct net_device_context *net_device_ctx = netdev_priv(net); |
981a1bd8 | 359 | struct hv_netvsc_packet *packet = NULL; |
02fafbc6 | 360 | int ret; |
8a00251a KS |
361 | unsigned int num_data_pgs; |
362 | struct rndis_message *rndis_msg; | |
363 | struct rndis_packet *rndis_pkt; | |
364 | u32 rndis_msg_size; | |
365 | bool isvlan; | |
e88f7e07 | 366 | bool linear = false; |
8a00251a | 367 | struct rndis_per_packet_info *ppi; |
08cd04bf | 368 | struct ndis_tcp_ip_checksum_info *csum_info; |
77bf5487 | 369 | struct ndis_tcp_lso_info *lso_info; |
08cd04bf KS |
370 | int hdr_offset; |
371 | u32 net_trans_info; | |
307f0995 | 372 | u32 hash; |
e88f7e07 | 373 | u32 skb_length; |
b08cc791 | 374 | struct hv_page_buffer page_buf[MAX_PAGE_BUFFER_COUNT]; |
a9f2e2d6 | 375 | struct hv_page_buffer *pb = page_buf; |
7eafd9b4 | 376 | struct netvsc_stats *tx_stats = this_cpu_ptr(net_device_ctx->tx_stats); |
fceaf24a | 377 | |
54a7357f KS |
378 | /* We will atmost need two pages to describe the rndis |
379 | * header. We can only transmit MAX_PAGE_BUFFER_COUNT number | |
e88f7e07 VK |
380 | * of pages in a single packet. If skb is scattered around |
381 | * more pages we try linearizing it. | |
54a7357f | 382 | */ |
e88f7e07 VK |
383 | |
384 | check_size: | |
385 | skb_length = skb->len; | |
8a00251a | 386 | num_data_pgs = netvsc_get_slots(skb) + 2; |
e88f7e07 VK |
387 | if (num_data_pgs > MAX_PAGE_BUFFER_COUNT && linear) { |
388 | net_alert_ratelimited("packet too big: %u pages (%u bytes)\n", | |
389 | num_data_pgs, skb->len); | |
981a1bd8 VK |
390 | ret = -EFAULT; |
391 | goto drop; | |
e88f7e07 VK |
392 | } else if (num_data_pgs > MAX_PAGE_BUFFER_COUNT) { |
393 | if (skb_linearize(skb)) { | |
394 | net_alert_ratelimited("failed to linearize skb\n"); | |
395 | ret = -ENOMEM; | |
396 | goto drop; | |
397 | } | |
398 | linear = true; | |
399 | goto check_size; | |
54a7357f | 400 | } |
fceaf24a | 401 | |
c0eb4540 KS |
402 | /* |
403 | * Place the rndis header in the skb head room and | |
404 | * the skb->cb will be used for hv_netvsc_packet | |
405 | * structure. | |
406 | */ | |
407 | ret = skb_cow_head(skb, RNDIS_AND_PPI_SIZE); | |
b56fc3c5 KS |
408 | if (ret) { |
409 | netdev_err(net, "unable to alloc hv_netvsc_packet\n"); | |
410 | ret = -ENOMEM; | |
411 | goto drop; | |
fceaf24a | 412 | } |
c0eb4540 KS |
413 | /* Use the skb control buffer for building up the packet */ |
414 | BUILD_BUG_ON(sizeof(struct hv_netvsc_packet) > | |
415 | FIELD_SIZEOF(struct sk_buff, cb)); | |
416 | packet = (struct hv_netvsc_packet *)skb->cb; | |
fceaf24a | 417 | |
1f5f3a75 | 418 | |
5b54dac8 HZ |
419 | packet->q_idx = skb_get_queue_mapping(skb); |
420 | ||
4d447c9a | 421 | packet->total_data_buflen = skb->len; |
fceaf24a | 422 | |
c0eb4540 | 423 | rndis_msg = (struct rndis_message *)skb->head; |
b08cc791 | 424 | |
24476760 | 425 | memset(rndis_msg, 0, RNDIS_AND_PPI_SIZE); |
fceaf24a | 426 | |
760d1e36 | 427 | isvlan = skb->vlan_tci & VLAN_TAG_PRESENT; |
8a00251a KS |
428 | |
429 | /* Add the rndis header */ | |
8a00251a KS |
430 | rndis_msg->ndis_msg_type = RNDIS_MSG_PACKET; |
431 | rndis_msg->msg_len = packet->total_data_buflen; | |
432 | rndis_pkt = &rndis_msg->msg.pkt; | |
433 | rndis_pkt->data_offset = sizeof(struct rndis_packet); | |
434 | rndis_pkt->data_len = packet->total_data_buflen; | |
435 | rndis_pkt->per_pkt_info_offset = sizeof(struct rndis_packet); | |
436 | ||
437 | rndis_msg_size = RNDIS_MESSAGE_SIZE(struct rndis_packet); | |
438 | ||
307f0995 HZ |
439 | hash = skb_get_hash_raw(skb); |
440 | if (hash != 0 && net->real_num_tx_queues > 1) { | |
441 | rndis_msg_size += NDIS_HASH_PPI_SIZE; | |
442 | ppi = init_ppi_data(rndis_msg, NDIS_HASH_PPI_SIZE, | |
443 | NBL_HASH_VALUE); | |
444 | *(u32 *)((void *)ppi + ppi->ppi_offset) = hash; | |
445 | } | |
446 | ||
8a00251a KS |
447 | if (isvlan) { |
448 | struct ndis_pkt_8021q_info *vlan; | |
449 | ||
450 | rndis_msg_size += NDIS_VLAN_PPI_SIZE; | |
451 | ppi = init_ppi_data(rndis_msg, NDIS_VLAN_PPI_SIZE, | |
452 | IEEE_8021Q_INFO); | |
453 | vlan = (struct ndis_pkt_8021q_info *)((void *)ppi + | |
454 | ppi->ppi_offset); | |
760d1e36 KS |
455 | vlan->vlanid = skb->vlan_tci & VLAN_VID_MASK; |
456 | vlan->pri = (skb->vlan_tci & VLAN_PRIO_MASK) >> | |
8a00251a KS |
457 | VLAN_PRIO_SHIFT; |
458 | } | |
459 | ||
08cd04bf KS |
460 | net_trans_info = get_net_transport_info(skb, &hdr_offset); |
461 | if (net_trans_info == TRANSPORT_INFO_NOT_IP) | |
462 | goto do_send; | |
463 | ||
464 | /* | |
465 | * Setup the sendside checksum offload only if this is not a | |
466 | * GSO packet. | |
467 | */ | |
468 | if (skb_is_gso(skb)) | |
77bf5487 | 469 | goto do_lso; |
08cd04bf | 470 | |
22041fb0 KS |
471 | if ((skb->ip_summed == CHECKSUM_NONE) || |
472 | (skb->ip_summed == CHECKSUM_UNNECESSARY)) | |
473 | goto do_send; | |
474 | ||
08cd04bf KS |
475 | rndis_msg_size += NDIS_CSUM_PPI_SIZE; |
476 | ppi = init_ppi_data(rndis_msg, NDIS_CSUM_PPI_SIZE, | |
477 | TCPIP_CHKSUM_PKTINFO); | |
478 | ||
479 | csum_info = (struct ndis_tcp_ip_checksum_info *)((void *)ppi + | |
480 | ppi->ppi_offset); | |
481 | ||
482 | if (net_trans_info & (INFO_IPV4 << 16)) | |
483 | csum_info->transmit.is_ipv4 = 1; | |
484 | else | |
485 | csum_info->transmit.is_ipv6 = 1; | |
486 | ||
487 | if (net_trans_info & INFO_TCP) { | |
488 | csum_info->transmit.tcp_checksum = 1; | |
489 | csum_info->transmit.tcp_header_offset = hdr_offset; | |
490 | } else if (net_trans_info & INFO_UDP) { | |
af9893a3 KS |
491 | /* UDP checksum offload is not supported on ws2008r2. |
492 | * Furthermore, on ws2012 and ws2012r2, there are some | |
493 | * issues with udp checksum offload from Linux guests. | |
494 | * (these are host issues). | |
495 | * For now compute the checksum here. | |
496 | */ | |
497 | struct udphdr *uh; | |
498 | u16 udp_len; | |
499 | ||
500 | ret = skb_cow_head(skb, 0); | |
501 | if (ret) | |
502 | goto drop; | |
503 | ||
504 | uh = udp_hdr(skb); | |
505 | udp_len = ntohs(uh->len); | |
506 | uh->check = 0; | |
507 | uh->check = csum_tcpudp_magic(ip_hdr(skb)->saddr, | |
508 | ip_hdr(skb)->daddr, | |
509 | udp_len, IPPROTO_UDP, | |
510 | csum_partial(uh, udp_len, 0)); | |
511 | if (uh->check == 0) | |
512 | uh->check = CSUM_MANGLED_0; | |
513 | ||
514 | csum_info->transmit.udp_checksum = 0; | |
08cd04bf | 515 | } |
77bf5487 KS |
516 | goto do_send; |
517 | ||
518 | do_lso: | |
519 | rndis_msg_size += NDIS_LSO_PPI_SIZE; | |
520 | ppi = init_ppi_data(rndis_msg, NDIS_LSO_PPI_SIZE, | |
521 | TCP_LARGESEND_PKTINFO); | |
522 | ||
523 | lso_info = (struct ndis_tcp_lso_info *)((void *)ppi + | |
524 | ppi->ppi_offset); | |
525 | ||
526 | lso_info->lso_v2_transmit.type = NDIS_TCP_LARGE_SEND_OFFLOAD_V2_TYPE; | |
527 | if (net_trans_info & (INFO_IPV4 << 16)) { | |
528 | lso_info->lso_v2_transmit.ip_version = | |
529 | NDIS_TCP_LARGE_SEND_OFFLOAD_IPV4; | |
530 | ip_hdr(skb)->tot_len = 0; | |
531 | ip_hdr(skb)->check = 0; | |
532 | tcp_hdr(skb)->check = | |
533 | ~csum_tcpudp_magic(ip_hdr(skb)->saddr, | |
534 | ip_hdr(skb)->daddr, 0, IPPROTO_TCP, 0); | |
535 | } else { | |
536 | lso_info->lso_v2_transmit.ip_version = | |
537 | NDIS_TCP_LARGE_SEND_OFFLOAD_IPV6; | |
538 | ipv6_hdr(skb)->payload_len = 0; | |
539 | tcp_hdr(skb)->check = | |
540 | ~csum_ipv6_magic(&ipv6_hdr(skb)->saddr, | |
541 | &ipv6_hdr(skb)->daddr, 0, IPPROTO_TCP, 0); | |
542 | } | |
543 | lso_info->lso_v2_transmit.tcp_header_offset = hdr_offset; | |
544 | lso_info->lso_v2_transmit.mss = skb_shinfo(skb)->gso_size; | |
08cd04bf KS |
545 | |
546 | do_send: | |
8a00251a KS |
547 | /* Start filling in the page buffers with the rndis hdr */ |
548 | rndis_msg->msg_len += rndis_msg_size; | |
942396b0 | 549 | packet->total_data_buflen = rndis_msg->msg_len; |
8a00251a | 550 | packet->page_buf_cnt = init_page_array(rndis_msg, rndis_msg_size, |
a9f2e2d6 | 551 | skb, packet, &pb); |
8a00251a | 552 | |
76d13b56 | 553 | /* timestamp packet in software */ |
554 | skb_tx_timestamp(skb); | |
3a3d9a0a KS |
555 | ret = netvsc_send(net_device_ctx->device_ctx, packet, |
556 | rndis_msg, &pb, skb); | |
8a00251a | 557 | |
af9893a3 | 558 | drop: |
02fafbc6 | 559 | if (ret == 0) { |
4b02b58b | 560 | u64_stats_update_begin(&tx_stats->syncp); |
7eafd9b4 | 561 | tx_stats->packets++; |
562 | tx_stats->bytes += skb_length; | |
4b02b58b | 563 | u64_stats_update_end(&tx_stats->syncp); |
b220f5f9 | 564 | } else { |
33be96e4 HZ |
565 | if (ret != -EAGAIN) { |
566 | dev_kfree_skb_any(skb); | |
567 | net->stats.tx_dropped++; | |
568 | } | |
fceaf24a HJ |
569 | } |
570 | ||
33be96e4 | 571 | return (ret == -EAGAIN) ? NETDEV_TX_BUSY : NETDEV_TX_OK; |
fceaf24a HJ |
572 | } |
573 | ||
3e189519 | 574 | /* |
02fafbc6 GKH |
575 | * netvsc_linkstatus_callback - Link up/down notification |
576 | */ | |
90ef117a | 577 | void netvsc_linkstatus_callback(struct hv_device *device_obj, |
3a494e71 | 578 | struct rndis_message *resp) |
fceaf24a | 579 | { |
3a494e71 | 580 | struct rndis_indicate_status *indicate = &resp->msg.indicate_status; |
2ddd5e5f | 581 | struct net_device *net; |
c996edcf | 582 | struct net_device_context *ndev_ctx; |
2ddd5e5f | 583 | struct netvsc_device *net_device; |
27a70af3 VK |
584 | struct netvsc_reconfig *event; |
585 | unsigned long flags; | |
891de74d | 586 | |
27a70af3 VK |
587 | /* Handle link change statuses only */ |
588 | if (indicate->status != RNDIS_STATUS_NETWORK_CHANGE && | |
589 | indicate->status != RNDIS_STATUS_MEDIA_CONNECT && | |
590 | indicate->status != RNDIS_STATUS_MEDIA_DISCONNECT) | |
3a494e71 | 591 | return; |
891de74d | 592 | |
27a70af3 | 593 | net_device = hv_get_drvdata(device_obj); |
2ddd5e5f | 594 | net = net_device->ndev; |
fceaf24a | 595 | |
891de74d | 596 | if (!net || net->reg_state != NETREG_REGISTERED) |
fceaf24a | 597 | return; |
fceaf24a | 598 | |
891de74d | 599 | ndev_ctx = netdev_priv(net); |
27a70af3 VK |
600 | |
601 | event = kzalloc(sizeof(*event), GFP_ATOMIC); | |
602 | if (!event) | |
603 | return; | |
604 | event->event = indicate->status; | |
605 | ||
606 | spin_lock_irqsave(&ndev_ctx->lock, flags); | |
607 | list_add_tail(&event->list, &ndev_ctx->reconfig_events); | |
608 | spin_unlock_irqrestore(&ndev_ctx->lock, flags); | |
609 | ||
610 | schedule_delayed_work(&ndev_ctx->dwork, 0); | |
fceaf24a HJ |
611 | } |
612 | ||
3e189519 HJ |
613 | /* |
614 | * netvsc_recv_callback - Callback when we receive a packet from the | |
615 | * "wire" on the specified device. | |
02fafbc6 | 616 | */ |
f79adf8f | 617 | int netvsc_recv_callback(struct hv_device *device_obj, |
e3d605ed | 618 | struct hv_netvsc_packet *packet, |
c4b20c63 | 619 | void **data, |
25b85ee8 | 620 | struct ndis_tcp_ip_checksum_info *csum_info, |
760d1e36 KS |
621 | struct vmbus_channel *channel, |
622 | u16 vlan_tci) | |
fceaf24a | 623 | { |
6f4c4446 | 624 | struct net_device *net; |
7eafd9b4 | 625 | struct net_device_context *net_device_ctx; |
fceaf24a | 626 | struct sk_buff *skb; |
7eafd9b4 | 627 | struct netvsc_stats *rx_stats; |
fceaf24a | 628 | |
6f4c4446 | 629 | net = ((struct netvsc_device *)hv_get_drvdata(device_obj))->ndev; |
a68f9614 | 630 | if (!net || net->reg_state != NETREG_REGISTERED) { |
10082f98 | 631 | return NVSP_STAT_FAIL; |
fceaf24a | 632 | } |
7eafd9b4 | 633 | net_device_ctx = netdev_priv(net); |
634 | rx_stats = this_cpu_ptr(net_device_ctx->rx_stats); | |
fceaf24a | 635 | |
9495c282 | 636 | /* Allocate a skb - TODO direct I/O to pages? */ |
72a2f5bd | 637 | skb = netdev_alloc_skb_ip_align(net, packet->total_data_buflen); |
9495c282 SH |
638 | if (unlikely(!skb)) { |
639 | ++net->stats.rx_dropped; | |
10082f98 | 640 | return NVSP_STAT_FAIL; |
9495c282 | 641 | } |
fceaf24a | 642 | |
02fafbc6 GKH |
643 | /* |
644 | * Copy to skb. This copy is needed here since the memory pointed by | |
645 | * hv_netvsc_packet cannot be deallocated | |
646 | */ | |
c4b20c63 | 647 | memcpy(skb_put(skb, packet->total_data_buflen), *data, |
45326342 | 648 | packet->total_data_buflen); |
fceaf24a HJ |
649 | |
650 | skb->protocol = eth_type_trans(skb, net); | |
e3d605ed KS |
651 | if (csum_info) { |
652 | /* We only look at the IP checksum here. | |
653 | * Should we be dropping the packet if checksum | |
654 | * failed? How do we deal with other checksums - TCP/UDP? | |
655 | */ | |
656 | if (csum_info->receive.ip_checksum_succeeded) | |
657 | skb->ip_summed = CHECKSUM_UNNECESSARY; | |
658 | else | |
659 | skb->ip_summed = CHECKSUM_NONE; | |
660 | } | |
661 | ||
760d1e36 | 662 | if (vlan_tci & VLAN_TAG_PRESENT) |
93725cbd | 663 | __vlan_hwaccel_put_tag(skb, htons(ETH_P_8021Q), |
760d1e36 | 664 | vlan_tci); |
fceaf24a | 665 | |
25b85ee8 | 666 | skb_record_rx_queue(skb, channel-> |
e565e803 | 667 | offermsg.offer.sub_channel_index); |
5b54dac8 | 668 | |
4b02b58b | 669 | u64_stats_update_begin(&rx_stats->syncp); |
7eafd9b4 | 670 | rx_stats->packets++; |
671 | rx_stats->bytes += packet->total_data_buflen; | |
4b02b58b | 672 | u64_stats_update_end(&rx_stats->syncp); |
9495c282 | 673 | |
02fafbc6 GKH |
674 | /* |
675 | * Pass the skb back up. Network stack will deallocate the skb when it | |
9495c282 SH |
676 | * is done. |
677 | * TODO - use NAPI? | |
02fafbc6 | 678 | */ |
9495c282 | 679 | netif_rx(skb); |
fceaf24a | 680 | |
fceaf24a HJ |
681 | return 0; |
682 | } | |
683 | ||
f82f4ad7 SH |
684 | static void netvsc_get_drvinfo(struct net_device *net, |
685 | struct ethtool_drvinfo *info) | |
686 | { | |
7826d43f | 687 | strlcpy(info->driver, KBUILD_MODNAME, sizeof(info->driver)); |
7826d43f | 688 | strlcpy(info->fw_version, "N/A", sizeof(info->fw_version)); |
f82f4ad7 SH |
689 | } |
690 | ||
59995370 AS |
691 | static void netvsc_get_channels(struct net_device *net, |
692 | struct ethtool_channels *channel) | |
693 | { | |
694 | struct net_device_context *net_device_ctx = netdev_priv(net); | |
695 | struct hv_device *dev = net_device_ctx->device_ctx; | |
696 | struct netvsc_device *nvdev = hv_get_drvdata(dev); | |
697 | ||
698 | if (nvdev) { | |
699 | channel->max_combined = nvdev->max_chn; | |
700 | channel->combined_count = nvdev->num_chn; | |
701 | } | |
702 | } | |
703 | ||
b5960e6e AS |
704 | static int netvsc_set_channels(struct net_device *net, |
705 | struct ethtool_channels *channels) | |
706 | { | |
707 | struct net_device_context *net_device_ctx = netdev_priv(net); | |
708 | struct hv_device *dev = net_device_ctx->device_ctx; | |
709 | struct netvsc_device *nvdev = hv_get_drvdata(dev); | |
710 | struct netvsc_device_info device_info; | |
954591b9 AS |
711 | u32 num_chn; |
712 | u32 max_chn; | |
b5960e6e AS |
713 | int ret = 0; |
714 | bool recovering = false; | |
715 | ||
716 | if (!nvdev || nvdev->destroy) | |
717 | return -ENODEV; | |
718 | ||
954591b9 AS |
719 | num_chn = nvdev->num_chn; |
720 | max_chn = min_t(u32, nvdev->max_chn, num_online_cpus()); | |
721 | ||
b5960e6e AS |
722 | if (nvdev->nvsp_version < NVSP_PROTOCOL_VERSION_5) { |
723 | pr_info("vRSS unsupported before NVSP Version 5\n"); | |
724 | return -EINVAL; | |
725 | } | |
726 | ||
727 | /* We do not support rx, tx, or other */ | |
728 | if (!channels || | |
729 | channels->rx_count || | |
730 | channels->tx_count || | |
731 | channels->other_count || | |
732 | (channels->combined_count < 1)) | |
733 | return -EINVAL; | |
734 | ||
735 | if (channels->combined_count > max_chn) { | |
736 | pr_info("combined channels too high, using %d\n", max_chn); | |
737 | channels->combined_count = max_chn; | |
738 | } | |
739 | ||
740 | ret = netvsc_close(net); | |
741 | if (ret) | |
742 | goto out; | |
743 | ||
744 | do_set: | |
745 | nvdev->start_remove = true; | |
746 | rndis_filter_device_remove(dev); | |
747 | ||
748 | nvdev->num_chn = channels->combined_count; | |
749 | ||
750 | net_device_ctx->device_ctx = dev; | |
751 | hv_set_drvdata(dev, net); | |
752 | ||
753 | memset(&device_info, 0, sizeof(device_info)); | |
754 | device_info.num_chn = nvdev->num_chn; /* passed to RNDIS */ | |
755 | device_info.ring_size = ring_size; | |
756 | device_info.max_num_vrss_chns = max_num_vrss_chns; | |
757 | ||
758 | ret = rndis_filter_device_add(dev, &device_info); | |
759 | if (ret) { | |
760 | if (recovering) { | |
761 | netdev_err(net, "unable to add netvsc device (ret %d)\n", ret); | |
762 | return ret; | |
763 | } | |
764 | goto recover; | |
765 | } | |
766 | ||
767 | nvdev = hv_get_drvdata(dev); | |
768 | ||
769 | ret = netif_set_real_num_tx_queues(net, nvdev->num_chn); | |
770 | if (ret) { | |
771 | if (recovering) { | |
772 | netdev_err(net, "could not set tx queue count (ret %d)\n", ret); | |
773 | return ret; | |
774 | } | |
775 | goto recover; | |
776 | } | |
777 | ||
778 | ret = netif_set_real_num_rx_queues(net, nvdev->num_chn); | |
779 | if (ret) { | |
780 | if (recovering) { | |
781 | netdev_err(net, "could not set rx queue count (ret %d)\n", ret); | |
782 | return ret; | |
783 | } | |
784 | goto recover; | |
785 | } | |
786 | ||
787 | out: | |
788 | netvsc_open(net); | |
789 | ||
790 | return ret; | |
791 | ||
792 | recover: | |
793 | /* If the above failed, we attempt to recover through the same | |
794 | * process but with the original number of channels. | |
795 | */ | |
796 | netdev_err(net, "could not set channels, recovering\n"); | |
797 | recovering = true; | |
798 | channels->combined_count = num_chn; | |
799 | goto do_set; | |
800 | } | |
801 | ||
4d447c9a HZ |
802 | static int netvsc_change_mtu(struct net_device *ndev, int mtu) |
803 | { | |
804 | struct net_device_context *ndevctx = netdev_priv(ndev); | |
805 | struct hv_device *hdev = ndevctx->device_ctx; | |
806 | struct netvsc_device *nvdev = hv_get_drvdata(hdev); | |
807 | struct netvsc_device_info device_info; | |
808 | int limit = ETH_DATA_LEN; | |
2de8530b | 809 | int ret = 0; |
4d447c9a HZ |
810 | |
811 | if (nvdev == NULL || nvdev->destroy) | |
812 | return -ENODEV; | |
813 | ||
a1eabb01 | 814 | if (nvdev->nvsp_version >= NVSP_PROTOCOL_VERSION_2) |
4d3c9d37 | 815 | limit = NETVSC_MTU - ETH_HLEN; |
4d447c9a | 816 | |
f9cbce34 | 817 | if (mtu < NETVSC_MTU_MIN || mtu > limit) |
4d447c9a HZ |
818 | return -EINVAL; |
819 | ||
2de8530b HZ |
820 | ret = netvsc_close(ndev); |
821 | if (ret) | |
822 | goto out; | |
823 | ||
4d447c9a | 824 | nvdev->start_remove = true; |
4d447c9a HZ |
825 | rndis_filter_device_remove(hdev); |
826 | ||
827 | ndev->mtu = mtu; | |
828 | ||
829 | ndevctx->device_ctx = hdev; | |
830 | hv_set_drvdata(hdev, ndev); | |
8ebdcc52 AS |
831 | |
832 | memset(&device_info, 0, sizeof(device_info)); | |
4d447c9a | 833 | device_info.ring_size = ring_size; |
b5960e6e | 834 | device_info.num_chn = nvdev->num_chn; |
e01ec219 | 835 | device_info.max_num_vrss_chns = max_num_vrss_chns; |
4d447c9a | 836 | rndis_filter_device_add(hdev, &device_info); |
4d447c9a | 837 | |
2de8530b HZ |
838 | out: |
839 | netvsc_open(ndev); | |
840 | ||
841 | return ret; | |
4d447c9a HZ |
842 | } |
843 | ||
7eafd9b4 | 844 | static struct rtnl_link_stats64 *netvsc_get_stats64(struct net_device *net, |
845 | struct rtnl_link_stats64 *t) | |
846 | { | |
847 | struct net_device_context *ndev_ctx = netdev_priv(net); | |
848 | int cpu; | |
849 | ||
850 | for_each_possible_cpu(cpu) { | |
851 | struct netvsc_stats *tx_stats = per_cpu_ptr(ndev_ctx->tx_stats, | |
852 | cpu); | |
853 | struct netvsc_stats *rx_stats = per_cpu_ptr(ndev_ctx->rx_stats, | |
854 | cpu); | |
855 | u64 tx_packets, tx_bytes, rx_packets, rx_bytes; | |
856 | unsigned int start; | |
857 | ||
858 | do { | |
4b02b58b | 859 | start = u64_stats_fetch_begin_irq(&tx_stats->syncp); |
7eafd9b4 | 860 | tx_packets = tx_stats->packets; |
861 | tx_bytes = tx_stats->bytes; | |
4b02b58b | 862 | } while (u64_stats_fetch_retry_irq(&tx_stats->syncp, start)); |
7eafd9b4 | 863 | |
864 | do { | |
4b02b58b | 865 | start = u64_stats_fetch_begin_irq(&rx_stats->syncp); |
7eafd9b4 | 866 | rx_packets = rx_stats->packets; |
867 | rx_bytes = rx_stats->bytes; | |
4b02b58b | 868 | } while (u64_stats_fetch_retry_irq(&rx_stats->syncp, start)); |
7eafd9b4 | 869 | |
870 | t->tx_bytes += tx_bytes; | |
871 | t->tx_packets += tx_packets; | |
872 | t->rx_bytes += rx_bytes; | |
873 | t->rx_packets += rx_packets; | |
874 | } | |
875 | ||
876 | t->tx_dropped = net->stats.tx_dropped; | |
877 | t->tx_errors = net->stats.tx_dropped; | |
878 | ||
879 | t->rx_dropped = net->stats.rx_dropped; | |
880 | t->rx_errors = net->stats.rx_errors; | |
881 | ||
882 | return t; | |
883 | } | |
1ce09e89 HZ |
884 | |
885 | static int netvsc_set_mac_addr(struct net_device *ndev, void *p) | |
886 | { | |
887 | struct net_device_context *ndevctx = netdev_priv(ndev); | |
888 | struct hv_device *hdev = ndevctx->device_ctx; | |
889 | struct sockaddr *addr = p; | |
9a4c831e | 890 | char save_adr[ETH_ALEN]; |
1ce09e89 HZ |
891 | unsigned char save_aatype; |
892 | int err; | |
893 | ||
894 | memcpy(save_adr, ndev->dev_addr, ETH_ALEN); | |
895 | save_aatype = ndev->addr_assign_type; | |
896 | ||
897 | err = eth_mac_addr(ndev, p); | |
898 | if (err != 0) | |
899 | return err; | |
900 | ||
901 | err = rndis_filter_set_device_mac(hdev, addr->sa_data); | |
902 | if (err != 0) { | |
903 | /* roll back to saved MAC */ | |
904 | memcpy(ndev->dev_addr, save_adr, ETH_ALEN); | |
905 | ndev->addr_assign_type = save_aatype; | |
906 | } | |
907 | ||
908 | return err; | |
909 | } | |
910 | ||
316158fe RW |
911 | #ifdef CONFIG_NET_POLL_CONTROLLER |
912 | static void netvsc_poll_controller(struct net_device *net) | |
913 | { | |
914 | /* As netvsc_start_xmit() works synchronous we don't have to | |
915 | * trigger anything here. | |
916 | */ | |
917 | } | |
918 | #endif | |
1ce09e89 | 919 | |
f82f4ad7 SH |
920 | static const struct ethtool_ops ethtool_ops = { |
921 | .get_drvinfo = netvsc_get_drvinfo, | |
f82f4ad7 | 922 | .get_link = ethtool_op_get_link, |
59995370 | 923 | .get_channels = netvsc_get_channels, |
b5960e6e | 924 | .set_channels = netvsc_set_channels, |
76d13b56 | 925 | .get_ts_info = ethtool_op_get_ts_info, |
f82f4ad7 SH |
926 | }; |
927 | ||
df2fff28 GKH |
928 | static const struct net_device_ops device_ops = { |
929 | .ndo_open = netvsc_open, | |
930 | .ndo_stop = netvsc_close, | |
931 | .ndo_start_xmit = netvsc_start_xmit, | |
afc4b13d | 932 | .ndo_set_rx_mode = netvsc_set_multicast_list, |
4d447c9a | 933 | .ndo_change_mtu = netvsc_change_mtu, |
b681b588 | 934 | .ndo_validate_addr = eth_validate_addr, |
1ce09e89 | 935 | .ndo_set_mac_address = netvsc_set_mac_addr, |
5b54dac8 | 936 | .ndo_select_queue = netvsc_select_queue, |
7eafd9b4 | 937 | .ndo_get_stats64 = netvsc_get_stats64, |
316158fe RW |
938 | #ifdef CONFIG_NET_POLL_CONTROLLER |
939 | .ndo_poll_controller = netvsc_poll_controller, | |
940 | #endif | |
df2fff28 GKH |
941 | }; |
942 | ||
c996edcf | 943 | /* |
27a70af3 VK |
944 | * Handle link status changes. For RNDIS_STATUS_NETWORK_CHANGE emulate link |
945 | * down/up sequence. In case of RNDIS_STATUS_MEDIA_CONNECT when carrier is | |
946 | * present send GARP packet to network peers with netif_notify_peers(). | |
c996edcf | 947 | */ |
891de74d | 948 | static void netvsc_link_change(struct work_struct *w) |
c996edcf HZ |
949 | { |
950 | struct net_device_context *ndev_ctx; | |
951 | struct net_device *net; | |
2ddd5e5f | 952 | struct netvsc_device *net_device; |
891de74d | 953 | struct rndis_device *rdev; |
27a70af3 VK |
954 | struct netvsc_reconfig *event = NULL; |
955 | bool notify = false, reschedule = false; | |
956 | unsigned long flags, next_reconfig, delay; | |
c996edcf | 957 | |
122a5f64 | 958 | ndev_ctx = container_of(w, struct net_device_context, dwork.work); |
2ddd5e5f | 959 | net_device = hv_get_drvdata(ndev_ctx->device_ctx); |
891de74d | 960 | rdev = net_device->extension; |
2ddd5e5f | 961 | net = net_device->ndev; |
891de74d | 962 | |
27a70af3 VK |
963 | next_reconfig = ndev_ctx->last_reconfig + LINKCHANGE_INT; |
964 | if (time_is_after_jiffies(next_reconfig)) { | |
965 | /* link_watch only sends one notification with current state | |
966 | * per second, avoid doing reconfig more frequently. Handle | |
967 | * wrap around. | |
968 | */ | |
969 | delay = next_reconfig - jiffies; | |
970 | delay = delay < LINKCHANGE_INT ? delay : LINKCHANGE_INT; | |
971 | schedule_delayed_work(&ndev_ctx->dwork, delay); | |
972 | return; | |
973 | } | |
974 | ndev_ctx->last_reconfig = jiffies; | |
975 | ||
976 | spin_lock_irqsave(&ndev_ctx->lock, flags); | |
977 | if (!list_empty(&ndev_ctx->reconfig_events)) { | |
978 | event = list_first_entry(&ndev_ctx->reconfig_events, | |
979 | struct netvsc_reconfig, list); | |
980 | list_del(&event->list); | |
981 | reschedule = !list_empty(&ndev_ctx->reconfig_events); | |
982 | } | |
983 | spin_unlock_irqrestore(&ndev_ctx->lock, flags); | |
984 | ||
985 | if (!event) | |
986 | return; | |
987 | ||
988 | rtnl_lock(); | |
989 | ||
990 | switch (event->event) { | |
991 | /* Only the following events are possible due to the check in | |
992 | * netvsc_linkstatus_callback() | |
993 | */ | |
994 | case RNDIS_STATUS_MEDIA_CONNECT: | |
995 | if (rdev->link_state) { | |
996 | rdev->link_state = false; | |
997 | netif_carrier_on(net); | |
998 | netif_tx_wake_all_queues(net); | |
999 | } else { | |
1000 | notify = true; | |
1001 | } | |
1002 | kfree(event); | |
1003 | break; | |
1004 | case RNDIS_STATUS_MEDIA_DISCONNECT: | |
1005 | if (!rdev->link_state) { | |
1006 | rdev->link_state = true; | |
1007 | netif_carrier_off(net); | |
1008 | netif_tx_stop_all_queues(net); | |
1009 | } | |
1010 | kfree(event); | |
1011 | break; | |
1012 | case RNDIS_STATUS_NETWORK_CHANGE: | |
1013 | /* Only makes sense if carrier is present */ | |
1014 | if (!rdev->link_state) { | |
1015 | rdev->link_state = true; | |
1016 | netif_carrier_off(net); | |
1017 | netif_tx_stop_all_queues(net); | |
1018 | event->event = RNDIS_STATUS_MEDIA_CONNECT; | |
1019 | spin_lock_irqsave(&ndev_ctx->lock, flags); | |
1020 | list_add_tail(&event->list, &ndev_ctx->reconfig_events); | |
1021 | spin_unlock_irqrestore(&ndev_ctx->lock, flags); | |
1022 | reschedule = true; | |
3a494e71 | 1023 | } |
27a70af3 | 1024 | break; |
891de74d HZ |
1025 | } |
1026 | ||
1027 | rtnl_unlock(); | |
1028 | ||
1029 | if (notify) | |
1030 | netdev_notify_peers(net); | |
27a70af3 VK |
1031 | |
1032 | /* link_watch only sends one notification with current state per | |
1033 | * second, handle next reconfig event in 2 seconds. | |
1034 | */ | |
1035 | if (reschedule) | |
1036 | schedule_delayed_work(&ndev_ctx->dwork, LINKCHANGE_INT); | |
c996edcf HZ |
1037 | } |
1038 | ||
7eafd9b4 | 1039 | static void netvsc_free_netdev(struct net_device *netdev) |
1040 | { | |
1041 | struct net_device_context *net_device_ctx = netdev_priv(netdev); | |
1042 | ||
1043 | free_percpu(net_device_ctx->tx_stats); | |
1044 | free_percpu(net_device_ctx->rx_stats); | |
1045 | free_netdev(netdev); | |
1046 | } | |
c996edcf | 1047 | |
84946899 S |
1048 | static int netvsc_probe(struct hv_device *dev, |
1049 | const struct hv_vmbus_device_id *dev_id) | |
df2fff28 | 1050 | { |
df2fff28 GKH |
1051 | struct net_device *net = NULL; |
1052 | struct net_device_context *net_device_ctx; | |
1053 | struct netvsc_device_info device_info; | |
5b54dac8 | 1054 | struct netvsc_device *nvdev; |
df2fff28 GKH |
1055 | int ret; |
1056 | ||
5b54dac8 HZ |
1057 | net = alloc_etherdev_mq(sizeof(struct net_device_context), |
1058 | num_online_cpus()); | |
df2fff28 | 1059 | if (!net) |
51a805d0 | 1060 | return -ENOMEM; |
df2fff28 | 1061 | |
1b07da51 HZ |
1062 | netif_carrier_off(net); |
1063 | ||
df2fff28 | 1064 | net_device_ctx = netdev_priv(net); |
9efd21e1 | 1065 | net_device_ctx->device_ctx = dev; |
3f300ff4 SX |
1066 | net_device_ctx->msg_enable = netif_msg_init(debug, default_msg); |
1067 | if (netif_msg_probe(net_device_ctx)) | |
1068 | netdev_dbg(net, "netvsc msg_enable: %d\n", | |
1069 | net_device_ctx->msg_enable); | |
1070 | ||
7eafd9b4 | 1071 | net_device_ctx->tx_stats = netdev_alloc_pcpu_stats(struct netvsc_stats); |
1072 | if (!net_device_ctx->tx_stats) { | |
1073 | free_netdev(net); | |
1074 | return -ENOMEM; | |
1075 | } | |
1076 | net_device_ctx->rx_stats = netdev_alloc_pcpu_stats(struct netvsc_stats); | |
1077 | if (!net_device_ctx->rx_stats) { | |
1078 | free_percpu(net_device_ctx->tx_stats); | |
1079 | free_netdev(net); | |
1080 | return -ENOMEM; | |
1081 | } | |
1082 | ||
2ddd5e5f | 1083 | hv_set_drvdata(dev, net); |
891de74d | 1084 | INIT_DELAYED_WORK(&net_device_ctx->dwork, netvsc_link_change); |
792df872 | 1085 | INIT_WORK(&net_device_ctx->work, do_set_multicast); |
df2fff28 | 1086 | |
27a70af3 VK |
1087 | spin_lock_init(&net_device_ctx->lock); |
1088 | INIT_LIST_HEAD(&net_device_ctx->reconfig_events); | |
1089 | ||
df2fff28 GKH |
1090 | net->netdev_ops = &device_ops; |
1091 | ||
a060679c | 1092 | net->hw_features = NETVSC_HW_FEATURES; |
1093 | net->features = NETVSC_HW_FEATURES | NETIF_F_HW_VLAN_CTAG_TX; | |
6048718d | 1094 | |
7ad24ea4 | 1095 | net->ethtool_ops = ðtool_ops; |
9efd21e1 | 1096 | SET_NETDEV_DEV(net, &dev->device); |
df2fff28 | 1097 | |
14a03cf8 VK |
1098 | /* We always need headroom for rndis header */ |
1099 | net->needed_headroom = RNDIS_AND_PPI_SIZE; | |
1100 | ||
692e084e | 1101 | /* Notify the netvsc driver of the new device */ |
8ebdcc52 | 1102 | memset(&device_info, 0, sizeof(device_info)); |
692e084e | 1103 | device_info.ring_size = ring_size; |
e01ec219 | 1104 | device_info.max_num_vrss_chns = max_num_vrss_chns; |
692e084e HZ |
1105 | ret = rndis_filter_device_add(dev, &device_info); |
1106 | if (ret != 0) { | |
1107 | netdev_err(net, "unable to add netvsc device (ret %d)\n", ret); | |
7eafd9b4 | 1108 | netvsc_free_netdev(net); |
2ddd5e5f | 1109 | hv_set_drvdata(dev, NULL); |
692e084e | 1110 | return ret; |
df2fff28 | 1111 | } |
692e084e HZ |
1112 | memcpy(net->dev_addr, device_info.mac_adr, ETH_ALEN); |
1113 | ||
5b54dac8 HZ |
1114 | nvdev = hv_get_drvdata(dev); |
1115 | netif_set_real_num_tx_queues(net, nvdev->num_chn); | |
1116 | netif_set_real_num_rx_queues(net, nvdev->num_chn); | |
5b54dac8 | 1117 | |
a68f9614 HZ |
1118 | ret = register_netdev(net); |
1119 | if (ret != 0) { | |
1120 | pr_err("Unable to register netdev.\n"); | |
1121 | rndis_filter_device_remove(dev); | |
7eafd9b4 | 1122 | netvsc_free_netdev(net); |
a68f9614 HZ |
1123 | } |
1124 | ||
df2fff28 GKH |
1125 | return ret; |
1126 | } | |
1127 | ||
415b023a | 1128 | static int netvsc_remove(struct hv_device *dev) |
df2fff28 | 1129 | { |
2ddd5e5f | 1130 | struct net_device *net; |
122a5f64 | 1131 | struct net_device_context *ndev_ctx; |
2ddd5e5f S |
1132 | struct netvsc_device *net_device; |
1133 | ||
1134 | net_device = hv_get_drvdata(dev); | |
1135 | net = net_device->ndev; | |
df2fff28 | 1136 | |
df2fff28 | 1137 | if (net == NULL) { |
415b023a | 1138 | dev_err(&dev->device, "No net device to remove\n"); |
df2fff28 GKH |
1139 | return 0; |
1140 | } | |
1141 | ||
4d447c9a HZ |
1142 | net_device->start_remove = true; |
1143 | ||
122a5f64 HZ |
1144 | ndev_ctx = netdev_priv(net); |
1145 | cancel_delayed_work_sync(&ndev_ctx->dwork); | |
792df872 | 1146 | cancel_work_sync(&ndev_ctx->work); |
122a5f64 | 1147 | |
df2fff28 | 1148 | /* Stop outbound asap */ |
0a282538 | 1149 | netif_tx_disable(net); |
df2fff28 GKH |
1150 | |
1151 | unregister_netdev(net); | |
1152 | ||
1153 | /* | |
1154 | * Call to the vsc driver to let it know that the device is being | |
1155 | * removed | |
1156 | */ | |
df06bcff | 1157 | rndis_filter_device_remove(dev); |
df2fff28 | 1158 | |
7eafd9b4 | 1159 | netvsc_free_netdev(net); |
df06bcff | 1160 | return 0; |
df2fff28 GKH |
1161 | } |
1162 | ||
345c4cc3 | 1163 | static const struct hv_vmbus_device_id id_table[] = { |
c45cf2d4 | 1164 | /* Network guid */ |
8f505944 | 1165 | { HV_NIC_GUID, }, |
c45cf2d4 | 1166 | { }, |
345c4cc3 S |
1167 | }; |
1168 | ||
1169 | MODULE_DEVICE_TABLE(vmbus, id_table); | |
1170 | ||
f1542a66 | 1171 | /* The one and only one */ |
fde0ef9b | 1172 | static struct hv_driver netvsc_drv = { |
d31b20fc | 1173 | .name = KBUILD_MODNAME, |
345c4cc3 | 1174 | .id_table = id_table, |
fde0ef9b S |
1175 | .probe = netvsc_probe, |
1176 | .remove = netvsc_remove, | |
d4890970 | 1177 | }; |
f1542a66 | 1178 | |
a9869c94 | 1179 | static void __exit netvsc_drv_exit(void) |
fceaf24a | 1180 | { |
768fa219 | 1181 | vmbus_driver_unregister(&netvsc_drv); |
fceaf24a HJ |
1182 | } |
1183 | ||
1fde28cf | 1184 | static int __init netvsc_drv_init(void) |
df2fff28 | 1185 | { |
fa85a6c2 HZ |
1186 | if (ring_size < RING_SIZE_MIN) { |
1187 | ring_size = RING_SIZE_MIN; | |
1188 | pr_info("Increased ring_size to %d (min allowed)\n", | |
1189 | ring_size); | |
1190 | } | |
768fa219 | 1191 | return vmbus_driver_register(&netvsc_drv); |
df2fff28 GKH |
1192 | } |
1193 | ||
26c14cc1 | 1194 | MODULE_LICENSE("GPL"); |
7880fc54 | 1195 | MODULE_DESCRIPTION("Microsoft Hyper-V network driver"); |
fceaf24a | 1196 | |
1fde28cf | 1197 | module_init(netvsc_drv_init); |
a9869c94 | 1198 | module_exit(netvsc_drv_exit); |