1// SPDX-License-Identifier: GPL-2.0-only
2/*
3 * Copyright (c) 2009, Microsoft Corporation.
4 *
5 * Authors:
6 * Haiyang Zhang <haiyangz@microsoft.com>
7 * Hank Janssen <hjanssen@microsoft.com>
8 */
9#define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
10
11#include <linux/kernel.h>
12#include <linux/sched.h>
13#include <linux/wait.h>
14#include <linux/mm.h>
15#include <linux/delay.h>
16#include <linux/io.h>
17#include <linux/slab.h>
18#include <linux/netdevice.h>
19#include <linux/if_ether.h>
20#include <linux/vmalloc.h>
21#include <linux/rtnetlink.h>
22#include <linux/prefetch.h>
23#include <linux/filter.h>
24
25#include <asm/sync_bitops.h>
26#include <asm/mshyperv.h>
27
28#include "hyperv_net.h"
29#include "netvsc_trace.h"
30
31/*
32 * Switch the data path from the synthetic interface to the VF
33 * interface.
34 */
35int netvsc_switch_datapath(struct net_device *ndev, bool vf)
36{
37 struct net_device_context *net_device_ctx = netdev_priv(dev: ndev);
38 struct hv_device *dev = net_device_ctx->device_ctx;
39 struct netvsc_device *nv_dev = rtnl_dereference(net_device_ctx->nvdev);
40 struct nvsp_message *init_pkt = &nv_dev->channel_init_pkt;
41 int ret, retry = 0;
42
43 /* Block sending traffic to VF if it's about to be gone */
44 if (!vf)
45 net_device_ctx->data_path_is_vf = vf;
46
47 memset(init_pkt, 0, sizeof(struct nvsp_message));
48 init_pkt->hdr.msg_type = NVSP_MSG4_TYPE_SWITCH_DATA_PATH;
49 if (vf)
50 init_pkt->msg.v4_msg.active_dp.active_datapath =
51 NVSP_DATAPATH_VF;
52 else
53 init_pkt->msg.v4_msg.active_dp.active_datapath =
54 NVSP_DATAPATH_SYNTHETIC;
55
56again:
57 trace_nvsp_send(ndev, msg: init_pkt);
58
59 ret = vmbus_sendpacket(channel: dev->channel, buffer: init_pkt,
60 bufferLen: sizeof(struct nvsp_message),
61 requestid: (unsigned long)init_pkt, type: VM_PKT_DATA_INBAND,
62 VMBUS_DATA_PACKET_FLAG_COMPLETION_REQUESTED);
63
64 /* If failed to switch to/from VF, let data_path_is_vf stay false,
65 * so we use synthetic path to send data.
66 */
67 if (ret) {
68 if (ret != -EAGAIN) {
69 netdev_err(dev: ndev,
70 format: "Unable to send sw datapath msg, err: %d\n",
71 ret);
72 return ret;
73 }
74
75 if (retry++ < RETRY_MAX) {
76 usleep_range(RETRY_US_LO, RETRY_US_HI);
77 goto again;
78 } else {
79 netdev_err(
80 dev: ndev,
81 format: "Retry failed to send sw datapath msg, err: %d\n",
82 ret);
83 return ret;
84 }
85 }
86
87 wait_for_completion(&nv_dev->channel_init_wait);
88 net_device_ctx->data_path_is_vf = vf;
89
90 return 0;
91}
92
93/* Worker to setup sub channels on initial setup
94 * Initial hotplug event occurs in softirq context
95 * and can't wait for channels.
96 */
97static void netvsc_subchan_work(struct work_struct *w)
98{
99 struct netvsc_device *nvdev =
100 container_of(w, struct netvsc_device, subchan_work);
101 struct rndis_device *rdev;
102 int i, ret;
103
104 /* Avoid deadlock with device removal already under RTNL */
105 if (!rtnl_trylock()) {
106 schedule_work(work: w);
107 return;
108 }
109
110 rdev = nvdev->extension;
111 if (rdev) {
112 ret = rndis_set_subchannel(ndev: rdev->ndev, nvdev, NULL);
113 if (ret == 0) {
114 netif_device_attach(dev: rdev->ndev);
115 } else {
116 /* fallback to only primary channel */
117 for (i = 1; i < nvdev->num_chn; i++)
118 netif_napi_del(napi: &nvdev->chan_table[i].napi);
119
120 nvdev->max_chn = 1;
121 nvdev->num_chn = 1;
122 }
123 }
124
125 rtnl_unlock();
126}
127
128static struct netvsc_device *alloc_net_device(void)
129{
130 struct netvsc_device *net_device;
131
132 net_device = kzalloc(sizeof(struct netvsc_device), GFP_KERNEL);
133 if (!net_device)
134 return NULL;
135
136 init_waitqueue_head(&net_device->wait_drain);
137 net_device->destroy = false;
138 net_device->tx_disable = true;
139
140 net_device->max_pkt = RNDIS_MAX_PKT_DEFAULT;
141 net_device->pkt_align = RNDIS_PKT_ALIGN_DEFAULT;
142
143 init_completion(x: &net_device->channel_init_wait);
144 init_waitqueue_head(&net_device->subchan_open);
145 INIT_WORK(&net_device->subchan_work, netvsc_subchan_work);
146
147 return net_device;
148}
149
150static void free_netvsc_device(struct rcu_head *head)
151{
152 struct netvsc_device *nvdev
153 = container_of(head, struct netvsc_device, rcu);
154 int i;
155
156 kfree(objp: nvdev->extension);
157
158 if (!nvdev->recv_buf_gpadl_handle.decrypted)
159 vfree(addr: nvdev->recv_buf);
160 if (!nvdev->send_buf_gpadl_handle.decrypted)
161 vfree(addr: nvdev->send_buf);
162 bitmap_free(bitmap: nvdev->send_section_map);
163
164 for (i = 0; i < VRSS_CHANNEL_MAX; i++) {
165 xdp_rxq_info_unreg(xdp_rxq: &nvdev->chan_table[i].xdp_rxq);
166 kfree(objp: nvdev->chan_table[i].recv_buf);
167 vfree(addr: nvdev->chan_table[i].mrc.slots);
168 }
169
170 kfree(objp: nvdev);
171}
172
173static void free_netvsc_device_rcu(struct netvsc_device *nvdev)
174{
175 call_rcu(head: &nvdev->rcu, func: free_netvsc_device);
176}
177
178static void netvsc_revoke_recv_buf(struct hv_device *device,
179 struct netvsc_device *net_device,
180 struct net_device *ndev)
181{
182 struct nvsp_message *revoke_packet;
183 int ret;
184
185 /*
186 * If we got a section count, it means we received a
187 * SendReceiveBufferComplete msg (ie sent
188 * NvspMessage1TypeSendReceiveBuffer msg) therefore, we need
189 * to send a revoke msg here
190 */
191 if (net_device->recv_section_cnt) {
192 /* Send the revoke receive buffer */
193 revoke_packet = &net_device->revoke_packet;
194 memset(revoke_packet, 0, sizeof(struct nvsp_message));
195
196 revoke_packet->hdr.msg_type =
197 NVSP_MSG1_TYPE_REVOKE_RECV_BUF;
198 revoke_packet->msg.v1_msg.
199 revoke_recv_buf.id = NETVSC_RECEIVE_BUFFER_ID;
200
201 trace_nvsp_send(ndev, msg: revoke_packet);
202
203 ret = vmbus_sendpacket(channel: device->channel,
204 buffer: revoke_packet,
205 bufferLen: sizeof(struct nvsp_message),
206 VMBUS_RQST_ID_NO_RESPONSE,
207 type: VM_PKT_DATA_INBAND, flags: 0);
208 /* If the failure is because the channel is rescinded;
209 * ignore the failure since we cannot send on a rescinded
210 * channel. This would allow us to properly cleanup
211 * even when the channel is rescinded.
212 */
213 if (device->channel->rescind)
214 ret = 0;
215 /*
216 * If we failed here, we might as well return and
217 * have a leak rather than continue and a bugchk
218 */
219 if (ret != 0) {
220 netdev_err(dev: ndev, format: "unable to send "
221 "revoke receive buffer to netvsp\n");
222 return;
223 }
224 net_device->recv_section_cnt = 0;
225 }
226}
227
228static void netvsc_revoke_send_buf(struct hv_device *device,
229 struct netvsc_device *net_device,
230 struct net_device *ndev)
231{
232 struct nvsp_message *revoke_packet;
233 int ret;
234
235 /* Deal with the send buffer we may have setup.
236 * If we got a send section size, it means we received a
237 * NVSP_MSG1_TYPE_SEND_SEND_BUF_COMPLETE msg (ie sent
238 * NVSP_MSG1_TYPE_SEND_SEND_BUF msg) therefore, we need
239 * to send a revoke msg here
240 */
241 if (net_device->send_section_cnt) {
242 /* Send the revoke receive buffer */
243 revoke_packet = &net_device->revoke_packet;
244 memset(revoke_packet, 0, sizeof(struct nvsp_message));
245
246 revoke_packet->hdr.msg_type =
247 NVSP_MSG1_TYPE_REVOKE_SEND_BUF;
248 revoke_packet->msg.v1_msg.revoke_send_buf.id =
249 NETVSC_SEND_BUFFER_ID;
250
251 trace_nvsp_send(ndev, msg: revoke_packet);
252
253 ret = vmbus_sendpacket(channel: device->channel,
254 buffer: revoke_packet,
255 bufferLen: sizeof(struct nvsp_message),
256 VMBUS_RQST_ID_NO_RESPONSE,
257 type: VM_PKT_DATA_INBAND, flags: 0);
258
259 /* If the failure is because the channel is rescinded;
260 * ignore the failure since we cannot send on a rescinded
261 * channel. This would allow us to properly cleanup
262 * even when the channel is rescinded.
263 */
264 if (device->channel->rescind)
265 ret = 0;
266
267 /* If we failed here, we might as well return and
268 * have a leak rather than continue and a bugchk
269 */
270 if (ret != 0) {
271 netdev_err(dev: ndev, format: "unable to send "
272 "revoke send buffer to netvsp\n");
273 return;
274 }
275 net_device->send_section_cnt = 0;
276 }
277}
278
279static void netvsc_teardown_recv_gpadl(struct hv_device *device,
280 struct netvsc_device *net_device,
281 struct net_device *ndev)
282{
283 int ret;
284
285 if (net_device->recv_buf_gpadl_handle.gpadl_handle) {
286 ret = vmbus_teardown_gpadl(channel: device->channel,
287 gpadl: &net_device->recv_buf_gpadl_handle);
288
289 /* If we failed here, we might as well return and have a leak
290 * rather than continue and a bugchk
291 */
292 if (ret != 0) {
293 netdev_err(dev: ndev,
294 format: "unable to teardown receive buffer's gpadl\n");
295 return;
296 }
297 }
298}
299
300static void netvsc_teardown_send_gpadl(struct hv_device *device,
301 struct netvsc_device *net_device,
302 struct net_device *ndev)
303{
304 int ret;
305
306 if (net_device->send_buf_gpadl_handle.gpadl_handle) {
307 ret = vmbus_teardown_gpadl(channel: device->channel,
308 gpadl: &net_device->send_buf_gpadl_handle);
309
310 /* If we failed here, we might as well return and have a leak
311 * rather than continue and a bugchk
312 */
313 if (ret != 0) {
314 netdev_err(dev: ndev,
315 format: "unable to teardown send buffer's gpadl\n");
316 return;
317 }
318 }
319}
320
321int netvsc_alloc_recv_comp_ring(struct netvsc_device *net_device, u32 q_idx)
322{
323 struct netvsc_channel *nvchan = &net_device->chan_table[q_idx];
324 int node = cpu_to_node(cpu: nvchan->channel->target_cpu);
325 size_t size;
326
327 size = net_device->recv_completion_cnt * sizeof(struct recv_comp_data);
328 nvchan->mrc.slots = vzalloc_node(size, node);
329 if (!nvchan->mrc.slots)
330 nvchan->mrc.slots = vzalloc(size);
331
332 return nvchan->mrc.slots ? 0 : -ENOMEM;
333}
334
335static int netvsc_init_buf(struct hv_device *device,
336 struct netvsc_device *net_device,
337 const struct netvsc_device_info *device_info)
338{
339 struct nvsp_1_message_send_receive_buffer_complete *resp;
340 struct net_device *ndev = hv_get_drvdata(dev: device);
341 struct nvsp_message *init_packet;
342 unsigned int buf_size;
343 int i, ret = 0;
344
345 /* Get receive buffer area. */
346 buf_size = device_info->recv_sections * device_info->recv_section_size;
347 buf_size = roundup(buf_size, PAGE_SIZE);
348
349 /* Legacy hosts only allow smaller receive buffer */
350 if (net_device->nvsp_version <= NVSP_PROTOCOL_VERSION_2)
351 buf_size = min_t(unsigned int, buf_size,
352 NETVSC_RECEIVE_BUFFER_SIZE_LEGACY);
353
354 net_device->recv_buf = vzalloc(buf_size);
355 if (!net_device->recv_buf) {
356 netdev_err(dev: ndev,
357 format: "unable to allocate receive buffer of size %u\n",
358 buf_size);
359 ret = -ENOMEM;
360 goto cleanup;
361 }
362
363 net_device->recv_buf_size = buf_size;
364
365 /*
366 * Establish the gpadl handle for this buffer on this
367 * channel. Note: This call uses the vmbus connection rather
368 * than the channel to establish the gpadl handle.
369 */
370 ret = vmbus_establish_gpadl(channel: device->channel, kbuffer: net_device->recv_buf,
371 size: buf_size,
372 gpadl: &net_device->recv_buf_gpadl_handle);
373 if (ret != 0) {
374 netdev_err(dev: ndev,
375 format: "unable to establish receive buffer's gpadl\n");
376 goto cleanup;
377 }
378
379 /* Notify the NetVsp of the gpadl handle */
380 init_packet = &net_device->channel_init_pkt;
381 memset(init_packet, 0, sizeof(struct nvsp_message));
382 init_packet->hdr.msg_type = NVSP_MSG1_TYPE_SEND_RECV_BUF;
383 init_packet->msg.v1_msg.send_recv_buf.
384 gpadl_handle = net_device->recv_buf_gpadl_handle.gpadl_handle;
385 init_packet->msg.v1_msg.
386 send_recv_buf.id = NETVSC_RECEIVE_BUFFER_ID;
387
388 trace_nvsp_send(ndev, msg: init_packet);
389
390 /* Send the gpadl notification request */
391 ret = vmbus_sendpacket(channel: device->channel, buffer: init_packet,
392 bufferLen: sizeof(struct nvsp_message),
393 requestid: (unsigned long)init_packet,
394 type: VM_PKT_DATA_INBAND,
395 VMBUS_DATA_PACKET_FLAG_COMPLETION_REQUESTED);
396 if (ret != 0) {
397 netdev_err(dev: ndev,
398 format: "unable to send receive buffer's gpadl to netvsp\n");
399 goto cleanup;
400 }
401
402 wait_for_completion(&net_device->channel_init_wait);
403
404 /* Check the response */
405 resp = &init_packet->msg.v1_msg.send_recv_buf_complete;
406 if (resp->status != NVSP_STAT_SUCCESS) {
407 netdev_err(dev: ndev,
408 format: "Unable to complete receive buffer initialization with NetVsp - status %d\n",
409 resp->status);
410 ret = -EINVAL;
411 goto cleanup;
412 }
413
414 /* Parse the response */
415 netdev_dbg(ndev, "Receive sections: %u sub_allocs: size %u count: %u\n",
416 resp->num_sections, resp->sections[0].sub_alloc_size,
417 resp->sections[0].num_sub_allocs);
418
419 /* There should only be one section for the entire receive buffer */
420 if (resp->num_sections != 1 || resp->sections[0].offset != 0) {
421 ret = -EINVAL;
422 goto cleanup;
423 }
424
425 net_device->recv_section_size = resp->sections[0].sub_alloc_size;
426 net_device->recv_section_cnt = resp->sections[0].num_sub_allocs;
427
428 /* Ensure buffer will not overflow */
429 if (net_device->recv_section_size < NETVSC_MTU_MIN || (u64)net_device->recv_section_size *
430 (u64)net_device->recv_section_cnt > (u64)buf_size) {
431 netdev_err(dev: ndev, format: "invalid recv_section_size %u\n",
432 net_device->recv_section_size);
433 ret = -EINVAL;
434 goto cleanup;
435 }
436
437 for (i = 0; i < VRSS_CHANNEL_MAX; i++) {
438 struct netvsc_channel *nvchan = &net_device->chan_table[i];
439
440 nvchan->recv_buf = kzalloc(net_device->recv_section_size, GFP_KERNEL);
441 if (nvchan->recv_buf == NULL) {
442 ret = -ENOMEM;
443 goto cleanup;
444 }
445 }
446
447 /* Setup receive completion ring.
448 * Add 1 to the recv_section_cnt because at least one entry in a
449 * ring buffer has to be empty.
450 */
451 net_device->recv_completion_cnt = net_device->recv_section_cnt + 1;
452 ret = netvsc_alloc_recv_comp_ring(net_device, q_idx: 0);
453 if (ret)
454 goto cleanup;
455
456 /* Now setup the send buffer. */
457 buf_size = device_info->send_sections * device_info->send_section_size;
458 buf_size = round_up(buf_size, PAGE_SIZE);
459
460 net_device->send_buf = vzalloc(buf_size);
461 if (!net_device->send_buf) {
462 netdev_err(dev: ndev, format: "unable to allocate send buffer of size %u\n",
463 buf_size);
464 ret = -ENOMEM;
465 goto cleanup;
466 }
467 net_device->send_buf_size = buf_size;
468
469 /* Establish the gpadl handle for this buffer on this
470 * channel. Note: This call uses the vmbus connection rather
471 * than the channel to establish the gpadl handle.
472 */
473 ret = vmbus_establish_gpadl(channel: device->channel, kbuffer: net_device->send_buf,
474 size: buf_size,
475 gpadl: &net_device->send_buf_gpadl_handle);
476 if (ret != 0) {
477 netdev_err(dev: ndev,
478 format: "unable to establish send buffer's gpadl\n");
479 goto cleanup;
480 }
481
482 /* Notify the NetVsp of the gpadl handle */
483 init_packet = &net_device->channel_init_pkt;
484 memset(init_packet, 0, sizeof(struct nvsp_message));
485 init_packet->hdr.msg_type = NVSP_MSG1_TYPE_SEND_SEND_BUF;
486 init_packet->msg.v1_msg.send_send_buf.gpadl_handle =
487 net_device->send_buf_gpadl_handle.gpadl_handle;
488 init_packet->msg.v1_msg.send_send_buf.id = NETVSC_SEND_BUFFER_ID;
489
490 trace_nvsp_send(ndev, msg: init_packet);
491
492 /* Send the gpadl notification request */
493 ret = vmbus_sendpacket(channel: device->channel, buffer: init_packet,
494 bufferLen: sizeof(struct nvsp_message),
495 requestid: (unsigned long)init_packet,
496 type: VM_PKT_DATA_INBAND,
497 VMBUS_DATA_PACKET_FLAG_COMPLETION_REQUESTED);
498 if (ret != 0) {
499 netdev_err(dev: ndev,
500 format: "unable to send send buffer's gpadl to netvsp\n");
501 goto cleanup;
502 }
503
504 wait_for_completion(&net_device->channel_init_wait);
505
506 /* Check the response */
507 if (init_packet->msg.v1_msg.
508 send_send_buf_complete.status != NVSP_STAT_SUCCESS) {
509 netdev_err(dev: ndev, format: "Unable to complete send buffer "
510 "initialization with NetVsp - status %d\n",
511 init_packet->msg.v1_msg.
512 send_send_buf_complete.status);
513 ret = -EINVAL;
514 goto cleanup;
515 }
516
517 /* Parse the response */
518 net_device->send_section_size = init_packet->msg.
519 v1_msg.send_send_buf_complete.section_size;
520 if (net_device->send_section_size < NETVSC_MTU_MIN) {
521 netdev_err(dev: ndev, format: "invalid send_section_size %u\n",
522 net_device->send_section_size);
523 ret = -EINVAL;
524 goto cleanup;
525 }
526
527 /* Section count is simply the size divided by the section size. */
528 net_device->send_section_cnt = buf_size / net_device->send_section_size;
529
530 netdev_dbg(ndev, "Send section size: %d, Section count:%d\n",
531 net_device->send_section_size, net_device->send_section_cnt);
532
533 /* Setup state for managing the send buffer. */
534 net_device->send_section_map = bitmap_zalloc(nbits: net_device->send_section_cnt,
535 GFP_KERNEL);
536 if (!net_device->send_section_map) {
537 ret = -ENOMEM;
538 goto cleanup;
539 }
540
541 goto exit;
542
543cleanup:
544 netvsc_revoke_recv_buf(device, net_device, ndev);
545 netvsc_revoke_send_buf(device, net_device, ndev);
546 netvsc_teardown_recv_gpadl(device, net_device, ndev);
547 netvsc_teardown_send_gpadl(device, net_device, ndev);
548
549exit:
550 return ret;
551}
552
553/* Negotiate NVSP protocol version */
554static int negotiate_nvsp_ver(struct hv_device *device,
555 struct netvsc_device *net_device,
556 struct nvsp_message *init_packet,
557 u32 nvsp_ver)
558{
559 struct net_device *ndev = hv_get_drvdata(dev: device);
560 int ret;
561
562 memset(init_packet, 0, sizeof(struct nvsp_message));
563 init_packet->hdr.msg_type = NVSP_MSG_TYPE_INIT;
564 init_packet->msg.init_msg.init.min_protocol_ver = nvsp_ver;
565 init_packet->msg.init_msg.init.max_protocol_ver = nvsp_ver;
566 trace_nvsp_send(ndev, msg: init_packet);
567
568 /* Send the init request */
569 ret = vmbus_sendpacket(channel: device->channel, buffer: init_packet,
570 bufferLen: sizeof(struct nvsp_message),
571 requestid: (unsigned long)init_packet,
572 type: VM_PKT_DATA_INBAND,
573 VMBUS_DATA_PACKET_FLAG_COMPLETION_REQUESTED);
574
575 if (ret != 0)
576 return ret;
577
578 wait_for_completion(&net_device->channel_init_wait);
579
580 if (init_packet->msg.init_msg.init_complete.status !=
581 NVSP_STAT_SUCCESS)
582 return -EINVAL;
583
584 if (nvsp_ver == NVSP_PROTOCOL_VERSION_1)
585 return 0;
586
587 /* NVSPv2 or later: Send NDIS config */
588 memset(init_packet, 0, sizeof(struct nvsp_message));
589 init_packet->hdr.msg_type = NVSP_MSG2_TYPE_SEND_NDIS_CONFIG;
590 init_packet->msg.v2_msg.send_ndis_config.mtu = ndev->mtu + ETH_HLEN;
591 init_packet->msg.v2_msg.send_ndis_config.capability.ieee8021q = 1;
592
593 if (nvsp_ver >= NVSP_PROTOCOL_VERSION_5) {
594 if (hv_is_isolation_supported())
595 netdev_info(dev: ndev, format: "SR-IOV not advertised by guests on the host supporting isolation\n");
596 else
597 init_packet->msg.v2_msg.send_ndis_config.capability.sriov = 1;
598
599 /* Teaming bit is needed to receive link speed updates */
600 init_packet->msg.v2_msg.send_ndis_config.capability.teaming = 1;
601 }
602
603 if (nvsp_ver >= NVSP_PROTOCOL_VERSION_61)
604 init_packet->msg.v2_msg.send_ndis_config.capability.rsc = 1;
605
606 trace_nvsp_send(ndev, msg: init_packet);
607
608 ret = vmbus_sendpacket(channel: device->channel, buffer: init_packet,
609 bufferLen: sizeof(struct nvsp_message),
610 VMBUS_RQST_ID_NO_RESPONSE,
611 type: VM_PKT_DATA_INBAND, flags: 0);
612
613 return ret;
614}
615
616static int netvsc_connect_vsp(struct hv_device *device,
617 struct netvsc_device *net_device,
618 const struct netvsc_device_info *device_info)
619{
620 struct net_device *ndev = hv_get_drvdata(dev: device);
621 static const u32 ver_list[] = {
622 NVSP_PROTOCOL_VERSION_1, NVSP_PROTOCOL_VERSION_2,
623 NVSP_PROTOCOL_VERSION_4, NVSP_PROTOCOL_VERSION_5,
624 NVSP_PROTOCOL_VERSION_6, NVSP_PROTOCOL_VERSION_61
625 };
626 struct nvsp_message *init_packet;
627 int ndis_version, i, ret;
628
629 init_packet = &net_device->channel_init_pkt;
630
631 /* Negotiate the latest NVSP protocol supported */
632 for (i = ARRAY_SIZE(ver_list) - 1; i >= 0; i--)
633 if (negotiate_nvsp_ver(device, net_device, init_packet,
634 nvsp_ver: ver_list[i]) == 0) {
635 net_device->nvsp_version = ver_list[i];
636 break;
637 }
638
639 if (i < 0) {
640 ret = -EPROTO;
641 goto cleanup;
642 }
643
644 if (hv_is_isolation_supported() && net_device->nvsp_version < NVSP_PROTOCOL_VERSION_61) {
645 netdev_err(dev: ndev, format: "Invalid NVSP version 0x%x (expected >= 0x%x) from the host supporting isolation\n",
646 net_device->nvsp_version, NVSP_PROTOCOL_VERSION_61);
647 ret = -EPROTO;
648 goto cleanup;
649 }
650
651 pr_debug("Negotiated NVSP version:%x\n", net_device->nvsp_version);
652
653 /* Send the ndis version */
654 memset(init_packet, 0, sizeof(struct nvsp_message));
655
656 if (net_device->nvsp_version <= NVSP_PROTOCOL_VERSION_4)
657 ndis_version = 0x00060001;
658 else
659 ndis_version = 0x0006001e;
660
661 init_packet->hdr.msg_type = NVSP_MSG1_TYPE_SEND_NDIS_VER;
662 init_packet->msg.v1_msg.
663 send_ndis_ver.ndis_major_ver =
664 (ndis_version & 0xFFFF0000) >> 16;
665 init_packet->msg.v1_msg.
666 send_ndis_ver.ndis_minor_ver =
667 ndis_version & 0xFFFF;
668
669 trace_nvsp_send(ndev, msg: init_packet);
670
671 /* Send the init request */
672 ret = vmbus_sendpacket(channel: device->channel, buffer: init_packet,
673 bufferLen: sizeof(struct nvsp_message),
674 VMBUS_RQST_ID_NO_RESPONSE,
675 type: VM_PKT_DATA_INBAND, flags: 0);
676 if (ret != 0)
677 goto cleanup;
678
679
680 ret = netvsc_init_buf(device, net_device, device_info);
681
682cleanup:
683 return ret;
684}
685
686/*
687 * netvsc_device_remove - Callback when the root bus device is removed
688 */
689void netvsc_device_remove(struct hv_device *device)
690{
691 struct net_device *ndev = hv_get_drvdata(dev: device);
692 struct net_device_context *net_device_ctx = netdev_priv(dev: ndev);
693 struct netvsc_device *net_device
694 = rtnl_dereference(net_device_ctx->nvdev);
695 int i;
696
697 /*
698 * Revoke receive buffer. If host is pre-Win2016 then tear down
699 * receive buffer GPADL. Do the same for send buffer.
700 */
701 netvsc_revoke_recv_buf(device, net_device, ndev);
702 if (vmbus_proto_version < VERSION_WIN10)
703 netvsc_teardown_recv_gpadl(device, net_device, ndev);
704
705 netvsc_revoke_send_buf(device, net_device, ndev);
706 if (vmbus_proto_version < VERSION_WIN10)
707 netvsc_teardown_send_gpadl(device, net_device, ndev);
708
709 RCU_INIT_POINTER(net_device_ctx->nvdev, NULL);
710
711 /* Disable NAPI and disassociate its context from the device. */
712 for (i = 0; i < net_device->num_chn; i++) {
713 /* See also vmbus_reset_channel_cb(). */
714 /* only disable enabled NAPI channel */
715 if (i < ndev->real_num_rx_queues) {
716 netif_queue_set_napi(dev: ndev, queue_index: i, type: NETDEV_QUEUE_TYPE_TX,
717 NULL);
718 netif_queue_set_napi(dev: ndev, queue_index: i, type: NETDEV_QUEUE_TYPE_RX,
719 NULL);
720 napi_disable(n: &net_device->chan_table[i].napi);
721 }
722
723 netif_napi_del(napi: &net_device->chan_table[i].napi);
724 }
725
726 /*
727 * At this point, no one should be accessing net_device
728 * except in here
729 */
730 netdev_dbg(ndev, "net device safe to remove\n");
731
732 /* Now, we can close the channel safely */
733 vmbus_close(channel: device->channel);
734
735 /*
736 * If host is Win2016 or higher then we do the GPADL tear down
737 * here after VMBus is closed.
738 */
739 if (vmbus_proto_version >= VERSION_WIN10) {
740 netvsc_teardown_recv_gpadl(device, net_device, ndev);
741 netvsc_teardown_send_gpadl(device, net_device, ndev);
742 }
743
744 /* Release all resources */
745 free_netvsc_device_rcu(nvdev: net_device);
746}
747
748#define RING_AVAIL_PERCENT_HIWATER 20
749#define RING_AVAIL_PERCENT_LOWATER 10
750
751static inline void netvsc_free_send_slot(struct netvsc_device *net_device,
752 u32 index)
753{
754 sync_change_bit(nr: index, addr: net_device->send_section_map);
755}
756
757static void netvsc_send_tx_complete(struct net_device *ndev,
758 struct netvsc_device *net_device,
759 struct vmbus_channel *channel,
760 const struct vmpacket_descriptor *desc,
761 int budget)
762{
763 struct net_device_context *ndev_ctx = netdev_priv(dev: ndev);
764 struct sk_buff *skb;
765 u16 q_idx = 0;
766 int queue_sends;
767 u64 cmd_rqst;
768
769 cmd_rqst = channel->request_addr_callback(channel, desc->trans_id);
770 if (cmd_rqst == VMBUS_RQST_ERROR) {
771 netdev_err(dev: ndev, format: "Invalid transaction ID %llx\n", desc->trans_id);
772 return;
773 }
774
775 skb = (struct sk_buff *)(unsigned long)cmd_rqst;
776
777 /* Notify the layer above us */
778 if (likely(skb)) {
779 struct hv_netvsc_packet *packet
780 = (struct hv_netvsc_packet *)skb->cb;
781 u32 send_index = packet->send_buf_index;
782 struct netvsc_stats_tx *tx_stats;
783
784 if (send_index != NETVSC_INVALID_INDEX)
785 netvsc_free_send_slot(net_device, index: send_index);
786 q_idx = packet->q_idx;
787
788 tx_stats = &net_device->chan_table[q_idx].tx_stats;
789
790 u64_stats_update_begin(syncp: &tx_stats->syncp);
791 tx_stats->packets += packet->total_packets;
792 tx_stats->bytes += packet->total_bytes;
793 u64_stats_update_end(syncp: &tx_stats->syncp);
794
795 netvsc_dma_unmap(hv_dev: ndev_ctx->device_ctx, packet);
796 napi_consume_skb(skb, budget);
797 }
798
799 queue_sends =
800 atomic_dec_return(v: &net_device->chan_table[q_idx].queue_sends);
801
802 if (unlikely(net_device->destroy)) {
803 if (queue_sends == 0)
804 wake_up(&net_device->wait_drain);
805 } else {
806 struct netdev_queue *txq = netdev_get_tx_queue(dev: ndev, index: q_idx);
807
808 if (netif_tx_queue_stopped(dev_queue: txq) && !net_device->tx_disable &&
809 (hv_get_avail_to_write_percent(rbi: &channel->outbound) >
810 RING_AVAIL_PERCENT_HIWATER || queue_sends < 1)) {
811 netif_tx_wake_queue(dev_queue: txq);
812 ndev_ctx->eth_stats.wake_queue++;
813 }
814 }
815}
816
817static void netvsc_send_completion(struct net_device *ndev,
818 struct netvsc_device *net_device,
819 struct vmbus_channel *incoming_channel,
820 const struct vmpacket_descriptor *desc,
821 int budget)
822{
823 const struct nvsp_message *nvsp_packet;
824 u32 msglen = hv_pkt_datalen(desc);
825 struct nvsp_message *pkt_rqst;
826 u64 cmd_rqst;
827 u32 status;
828
829 /* First check if this is a VMBUS completion without data payload */
830 if (!msglen) {
831 cmd_rqst = incoming_channel->request_addr_callback(incoming_channel,
832 desc->trans_id);
833 if (cmd_rqst == VMBUS_RQST_ERROR) {
834 netdev_err(dev: ndev, format: "Invalid transaction ID %llx\n", desc->trans_id);
835 return;
836 }
837
838 pkt_rqst = (struct nvsp_message *)(uintptr_t)cmd_rqst;
839 switch (pkt_rqst->hdr.msg_type) {
840 case NVSP_MSG4_TYPE_SWITCH_DATA_PATH:
841 complete(&net_device->channel_init_wait);
842 break;
843
844 default:
845 netdev_err(dev: ndev, format: "Unexpected VMBUS completion!!\n");
846 }
847 return;
848 }
849
850 /* Ensure packet is big enough to read header fields */
851 if (msglen < sizeof(struct nvsp_message_header)) {
852 netdev_err(dev: ndev, format: "nvsp_message length too small: %u\n", msglen);
853 return;
854 }
855
856 nvsp_packet = hv_pkt_data(desc);
857 switch (nvsp_packet->hdr.msg_type) {
858 case NVSP_MSG_TYPE_INIT_COMPLETE:
859 if (msglen < sizeof(struct nvsp_message_header) +
860 sizeof(struct nvsp_message_init_complete)) {
861 netdev_err(dev: ndev, format: "nvsp_msg length too small: %u\n",
862 msglen);
863 return;
864 }
865 break;
866
867 case NVSP_MSG1_TYPE_SEND_RECV_BUF_COMPLETE:
868 if (msglen < sizeof(struct nvsp_message_header) +
869 struct_size_t(struct nvsp_1_message_send_receive_buffer_complete,
870 sections, 1)) {
871 netdev_err(dev: ndev, format: "nvsp_msg1 length too small: %u\n",
872 msglen);
873 return;
874 }
875 break;
876
877 case NVSP_MSG1_TYPE_SEND_SEND_BUF_COMPLETE:
878 if (msglen < sizeof(struct nvsp_message_header) +
879 sizeof(struct nvsp_1_message_send_send_buffer_complete)) {
880 netdev_err(dev: ndev, format: "nvsp_msg1 length too small: %u\n",
881 msglen);
882 return;
883 }
884 break;
885
886 case NVSP_MSG5_TYPE_SUBCHANNEL:
887 if (msglen < sizeof(struct nvsp_message_header) +
888 sizeof(struct nvsp_5_subchannel_complete)) {
889 netdev_err(dev: ndev, format: "nvsp_msg5 length too small: %u\n",
890 msglen);
891 return;
892 }
893 break;
894
895 case NVSP_MSG1_TYPE_SEND_RNDIS_PKT_COMPLETE:
896 if (msglen < sizeof(struct nvsp_message_header) +
897 sizeof(struct nvsp_1_message_send_rndis_packet_complete)) {
898 if (net_ratelimit())
899 netdev_err(dev: ndev, format: "nvsp_rndis_pkt_complete length too small: %u\n",
900 msglen);
901 return;
902 }
903
904 /* If status indicates an error, output a message so we know
905 * there's a problem. But process the completion anyway so the
906 * resources are released.
907 */
908 status = nvsp_packet->msg.v1_msg.send_rndis_pkt_complete.status;
909 if (status != NVSP_STAT_SUCCESS && net_ratelimit())
910 netdev_err(dev: ndev, format: "nvsp_rndis_pkt_complete error status: %x\n",
911 status);
912
913 netvsc_send_tx_complete(ndev, net_device, channel: incoming_channel,
914 desc, budget);
915 return;
916
917 default:
918 netdev_err(dev: ndev,
919 format: "Unknown send completion type %d received!!\n",
920 nvsp_packet->hdr.msg_type);
921 return;
922 }
923
924 /* Copy the response back */
925 memcpy(&net_device->channel_init_pkt, nvsp_packet,
926 sizeof(struct nvsp_message));
927 complete(&net_device->channel_init_wait);
928}
929
930static u32 netvsc_get_next_send_section(struct netvsc_device *net_device)
931{
932 unsigned long *map_addr = net_device->send_section_map;
933 unsigned int i;
934
935 for_each_clear_bit(i, map_addr, net_device->send_section_cnt) {
936 if (sync_test_and_set_bit(nr: i, addr: map_addr) == 0)
937 return i;
938 }
939
940 return NETVSC_INVALID_INDEX;
941}
942
943static void netvsc_copy_to_send_buf(struct netvsc_device *net_device,
944 unsigned int section_index,
945 u32 pend_size,
946 struct hv_netvsc_packet *packet,
947 struct rndis_message *rndis_msg,
948 struct hv_page_buffer *pb,
949 bool xmit_more)
950{
951 char *start = net_device->send_buf;
952 char *dest = start + (section_index * net_device->send_section_size)
953 + pend_size;
954 int i;
955 u32 padding = 0;
956 u32 page_count = packet->cp_partial ? 1 : packet->page_buf_cnt;
957 u32 remain;
958
959 /* Add padding */
960 remain = packet->total_data_buflen & (net_device->pkt_align - 1);
961 if (xmit_more && remain) {
962 padding = net_device->pkt_align - remain;
963 rndis_msg->msg_len += padding;
964 packet->total_data_buflen += padding;
965 }
966
967 for (i = 0; i < page_count; i++) {
968 char *src = phys_to_virt(address: pb[i].pfn << HV_HYP_PAGE_SHIFT);
969 u32 offset = pb[i].offset;
970 u32 len = pb[i].len;
971
972 memcpy(dest, (src + offset), len);
973 dest += len;
974 }
975
976 if (padding)
977 memset(dest, 0, padding);
978}
979
980void netvsc_dma_unmap(struct hv_device *hv_dev,
981 struct hv_netvsc_packet *packet)
982{
983 int i;
984
985 if (!hv_is_isolation_supported())
986 return;
987
988 if (!packet->dma_range)
989 return;
990
991 for (i = 0; i < packet->page_buf_cnt; i++)
992 dma_unmap_single(&hv_dev->device, packet->dma_range[i].dma,
993 packet->dma_range[i].mapping_size,
994 DMA_TO_DEVICE);
995
996 kfree(objp: packet->dma_range);
997}
998
999/* netvsc_dma_map - Map swiotlb bounce buffer with data page of
1000 * packet sent by vmbus_sendpacket_pagebuffer() in the Isolation
1001 * VM.
1002 *
1003 * In isolation VM, netvsc send buffer has been marked visible to
1004 * host and so the data copied to send buffer doesn't need to use
1005 * bounce buffer. The data pages handled by vmbus_sendpacket_pagebuffer()
1006 * may not be copied to send buffer and so these pages need to be
1007 * mapped with swiotlb bounce buffer. netvsc_dma_map() is to do
1008 * that. The pfns in the struct hv_page_buffer need to be converted
1009 * to bounce buffer's pfn. The loop here is necessary because the
1010 * entries in the page buffer array are not necessarily full
1011 * pages of data. Each entry in the array has a separate offset and
1012 * len that may be non-zero, even for entries in the middle of the
1013 * array. And the entries are not physically contiguous. So each
1014 * entry must be individually mapped rather than as a contiguous unit.
1015 * So not use dma_map_sg() here.
1016 */
1017static int netvsc_dma_map(struct hv_device *hv_dev,
1018 struct hv_netvsc_packet *packet,
1019 struct hv_page_buffer *pb)
1020{
1021 u32 page_count = packet->page_buf_cnt;
1022 dma_addr_t dma;
1023 int i;
1024
1025 if (!hv_is_isolation_supported())
1026 return 0;
1027
1028 packet->dma_range = kcalloc(page_count,
1029 sizeof(*packet->dma_range),
1030 GFP_ATOMIC);
1031 if (!packet->dma_range)
1032 return -ENOMEM;
1033
1034 for (i = 0; i < page_count; i++) {
1035 char *src = phys_to_virt(address: (pb[i].pfn << HV_HYP_PAGE_SHIFT)
1036 + pb[i].offset);
1037 u32 len = pb[i].len;
1038
1039 dma = dma_map_single(&hv_dev->device, src, len,
1040 DMA_TO_DEVICE);
1041 if (dma_mapping_error(dev: &hv_dev->device, dma_addr: dma)) {
1042 kfree(objp: packet->dma_range);
1043 return -ENOMEM;
1044 }
1045
1046 /* pb[].offset and pb[].len are not changed during dma mapping
1047 * and so not reassign.
1048 */
1049 packet->dma_range[i].dma = dma;
1050 packet->dma_range[i].mapping_size = len;
1051 pb[i].pfn = dma >> HV_HYP_PAGE_SHIFT;
1052 }
1053
1054 return 0;
1055}
1056
1057/* Build an "array" of mpb entries describing the data to be transferred
1058 * over VMBus. After the desc header fields, each "array" entry is variable
1059 * size, and each entry starts after the end of the previous entry. The
1060 * "offset" and "len" fields for each entry imply the size of the entry.
1061 *
1062 * The pfns are in HV_HYP_PAGE_SIZE, because all communication with Hyper-V
1063 * uses that granularity, even if the system page size of the guest is larger.
1064 * Each entry in the input "pb" array must describe a contiguous range of
1065 * guest physical memory so that the pfns are sequential if the range crosses
1066 * a page boundary. The offset field must be < HV_HYP_PAGE_SIZE.
1067 */
1068static inline void netvsc_build_mpb_array(struct hv_page_buffer *pb,
1069 u32 page_buffer_count,
1070 struct vmbus_packet_mpb_array *desc,
1071 u32 *desc_size)
1072{
1073 struct hv_mpb_array *mpb_entry = &desc->range;
1074 int i, j;
1075
1076 for (i = 0; i < page_buffer_count; i++) {
1077 u32 offset = pb[i].offset;
1078 u32 len = pb[i].len;
1079
1080 mpb_entry->offset = offset;
1081 mpb_entry->len = len;
1082
1083 for (j = 0; j < HVPFN_UP(offset + len); j++)
1084 mpb_entry->pfn_array[j] = pb[i].pfn + j;
1085
1086 mpb_entry = (struct hv_mpb_array *)&mpb_entry->pfn_array[j];
1087 }
1088
1089 desc->rangecount = page_buffer_count;
1090 *desc_size = (char *)mpb_entry - (char *)desc;
1091}
1092
1093static inline int netvsc_send_pkt(
1094 struct hv_device *device,
1095 struct hv_netvsc_packet *packet,
1096 struct netvsc_device *net_device,
1097 struct hv_page_buffer *pb,
1098 struct sk_buff *skb)
1099{
1100 struct nvsp_message nvmsg;
1101 struct nvsp_1_message_send_rndis_packet *rpkt =
1102 &nvmsg.msg.v1_msg.send_rndis_pkt;
1103 struct netvsc_channel * const nvchan =
1104 &net_device->chan_table[packet->q_idx];
1105 struct vmbus_channel *out_channel = nvchan->channel;
1106 struct net_device *ndev = hv_get_drvdata(dev: device);
1107 struct net_device_context *ndev_ctx = netdev_priv(dev: ndev);
1108 struct netdev_queue *txq = netdev_get_tx_queue(dev: ndev, index: packet->q_idx);
1109 u64 req_id;
1110 int ret;
1111 u32 ring_avail = hv_get_avail_to_write_percent(rbi: &out_channel->outbound);
1112
1113 memset(&nvmsg, 0, sizeof(struct nvsp_message));
1114 nvmsg.hdr.msg_type = NVSP_MSG1_TYPE_SEND_RNDIS_PKT;
1115 if (skb)
1116 rpkt->channel_type = 0; /* 0 is RMC_DATA */
1117 else
1118 rpkt->channel_type = 1; /* 1 is RMC_CONTROL */
1119
1120 rpkt->send_buf_section_index = packet->send_buf_index;
1121 if (packet->send_buf_index == NETVSC_INVALID_INDEX)
1122 rpkt->send_buf_section_size = 0;
1123 else
1124 rpkt->send_buf_section_size = packet->total_data_buflen;
1125
1126 req_id = (ulong)skb;
1127
1128 if (out_channel->rescind)
1129 return -ENODEV;
1130
1131 trace_nvsp_send_pkt(ndev, chan: out_channel, rpkt);
1132
1133 packet->dma_range = NULL;
1134 if (packet->page_buf_cnt) {
1135 struct vmbus_channel_packet_page_buffer desc;
1136 u32 desc_size;
1137
1138 if (packet->cp_partial)
1139 pb++;
1140
1141 ret = netvsc_dma_map(hv_dev: ndev_ctx->device_ctx, packet, pb);
1142 if (ret) {
1143 ret = -EAGAIN;
1144 goto exit;
1145 }
1146
1147 netvsc_build_mpb_array(pb, page_buffer_count: packet->page_buf_cnt,
1148 desc: (struct vmbus_packet_mpb_array *)&desc,
1149 desc_size: &desc_size);
1150 ret = vmbus_sendpacket_mpb_desc(channel: out_channel,
1151 mpb: (struct vmbus_packet_mpb_array *)&desc,
1152 desc_size, buffer: &nvmsg, bufferlen: sizeof(nvmsg), requestid: req_id);
1153 if (ret)
1154 netvsc_dma_unmap(hv_dev: ndev_ctx->device_ctx, packet);
1155 } else {
1156 ret = vmbus_sendpacket(channel: out_channel,
1157 buffer: &nvmsg, bufferLen: sizeof(nvmsg),
1158 requestid: req_id, type: VM_PKT_DATA_INBAND,
1159 VMBUS_DATA_PACKET_FLAG_COMPLETION_REQUESTED);
1160 }
1161
1162exit:
1163 if (ret == 0) {
1164 atomic_inc_return(v: &nvchan->queue_sends);
1165
1166 if (ring_avail < RING_AVAIL_PERCENT_LOWATER) {
1167 netif_tx_stop_queue(dev_queue: txq);
1168 ndev_ctx->eth_stats.stop_queue++;
1169 }
1170 } else if (ret == -EAGAIN) {
1171 netif_tx_stop_queue(dev_queue: txq);
1172 ndev_ctx->eth_stats.stop_queue++;
1173 } else {
1174 netdev_err(dev: ndev,
1175 format: "Unable to send packet pages %u len %u, ret %d\n",
1176 packet->page_buf_cnt, packet->total_data_buflen,
1177 ret);
1178 }
1179
1180 if (netif_tx_queue_stopped(dev_queue: txq) &&
1181 atomic_read(v: &nvchan->queue_sends) < 1 &&
1182 !net_device->tx_disable) {
1183 netif_tx_wake_queue(dev_queue: txq);
1184 ndev_ctx->eth_stats.wake_queue++;
1185 if (ret == -EAGAIN)
1186 ret = -ENOSPC;
1187 }
1188
1189 return ret;
1190}
1191
1192/* Move packet out of multi send data (msd), and clear msd */
1193static inline void move_pkt_msd(struct hv_netvsc_packet **msd_send,
1194 struct sk_buff **msd_skb,
1195 struct multi_send_data *msdp)
1196{
1197 *msd_skb = msdp->skb;
1198 *msd_send = msdp->pkt;
1199 msdp->skb = NULL;
1200 msdp->pkt = NULL;
1201 msdp->count = 0;
1202}
1203
1204/* RCU already held by caller */
1205/* Batching/bouncing logic is designed to attempt to optimize
1206 * performance.
1207 *
1208 * For small, non-LSO packets we copy the packet to a send buffer
1209 * which is pre-registered with the Hyper-V side. This enables the
1210 * hypervisor to avoid remapping the aperture to access the packet
1211 * descriptor and data.
1212 *
1213 * If we already started using a buffer and the netdev is transmitting
1214 * a burst of packets, keep on copying into the buffer until it is
1215 * full or we are done collecting a burst. If there is an existing
1216 * buffer with space for the RNDIS descriptor but not the packet, copy
1217 * the RNDIS descriptor to the buffer, keeping the packet in place.
1218 *
1219 * If we do batching and send more than one packet using a single
1220 * NetVSC message, free the SKBs of the packets copied, except for the
1221 * last packet. This is done to streamline the handling of the case
1222 * where the last packet only had the RNDIS descriptor copied to the
1223 * send buffer, with the data pointers included in the NetVSC message.
1224 */
1225int netvsc_send(struct net_device *ndev,
1226 struct hv_netvsc_packet *packet,
1227 struct rndis_message *rndis_msg,
1228 struct hv_page_buffer *pb,
1229 struct sk_buff *skb,
1230 bool xdp_tx)
1231{
1232 struct net_device_context *ndev_ctx = netdev_priv(dev: ndev);
1233 struct netvsc_device *net_device
1234 = rcu_dereference_bh(ndev_ctx->nvdev);
1235 struct hv_device *device = ndev_ctx->device_ctx;
1236 int ret = 0;
1237 struct netvsc_channel *nvchan;
1238 u32 pktlen = packet->total_data_buflen, msd_len = 0;
1239 unsigned int section_index = NETVSC_INVALID_INDEX;
1240 struct multi_send_data *msdp;
1241 struct hv_netvsc_packet *msd_send = NULL, *cur_send = NULL;
1242 struct sk_buff *msd_skb = NULL;
1243 bool try_batch, xmit_more;
1244
1245 /* If device is rescinded, return error and packet will get dropped. */
1246 if (unlikely(!net_device || net_device->destroy))
1247 return -ENODEV;
1248
1249 nvchan = &net_device->chan_table[packet->q_idx];
1250 packet->send_buf_index = NETVSC_INVALID_INDEX;
1251 packet->cp_partial = false;
1252
1253 /* Send a control message or XDP packet directly without accessing
1254 * msd (Multi-Send Data) field which may be changed during data packet
1255 * processing.
1256 */
1257 if (!skb || xdp_tx)
1258 return netvsc_send_pkt(device, packet, net_device, pb, skb);
1259
1260 /* batch packets in send buffer if possible */
1261 msdp = &nvchan->msd;
1262 if (msdp->pkt)
1263 msd_len = msdp->pkt->total_data_buflen;
1264
1265 try_batch = msd_len > 0 && msdp->count < net_device->max_pkt;
1266 if (try_batch && msd_len + pktlen + net_device->pkt_align <
1267 net_device->send_section_size) {
1268 section_index = msdp->pkt->send_buf_index;
1269
1270 } else if (try_batch && msd_len + packet->rmsg_size <
1271 net_device->send_section_size) {
1272 section_index = msdp->pkt->send_buf_index;
1273 packet->cp_partial = true;
1274
1275 } else if (pktlen + net_device->pkt_align <
1276 net_device->send_section_size) {
1277 section_index = netvsc_get_next_send_section(net_device);
1278 if (unlikely(section_index == NETVSC_INVALID_INDEX)) {
1279 ++ndev_ctx->eth_stats.tx_send_full;
1280 } else {
1281 move_pkt_msd(msd_send: &msd_send, msd_skb: &msd_skb, msdp);
1282 msd_len = 0;
1283 }
1284 }
1285
1286 /* Keep aggregating only if stack says more data is coming
1287 * and not doing mixed modes send and not flow blocked
1288 */
1289 xmit_more = netdev_xmit_more() &&
1290 !packet->cp_partial &&
1291 !netif_xmit_stopped(dev_queue: netdev_get_tx_queue(dev: ndev, index: packet->q_idx));
1292
1293 if (section_index != NETVSC_INVALID_INDEX) {
1294 netvsc_copy_to_send_buf(net_device,
1295 section_index, pend_size: msd_len,
1296 packet, rndis_msg, pb, xmit_more);
1297
1298 packet->send_buf_index = section_index;
1299
1300 if (packet->cp_partial) {
1301 packet->page_buf_cnt--;
1302 packet->total_data_buflen = msd_len + packet->rmsg_size;
1303 } else {
1304 packet->page_buf_cnt = 0;
1305 packet->total_data_buflen += msd_len;
1306 }
1307
1308 if (msdp->pkt) {
1309 packet->total_packets += msdp->pkt->total_packets;
1310 packet->total_bytes += msdp->pkt->total_bytes;
1311 }
1312
1313 if (msdp->skb)
1314 dev_consume_skb_any(skb: msdp->skb);
1315
1316 if (xmit_more) {
1317 msdp->skb = skb;
1318 msdp->pkt = packet;
1319 msdp->count++;
1320 } else {
1321 cur_send = packet;
1322 msdp->skb = NULL;
1323 msdp->pkt = NULL;
1324 msdp->count = 0;
1325 }
1326 } else {
1327 move_pkt_msd(msd_send: &msd_send, msd_skb: &msd_skb, msdp);
1328 cur_send = packet;
1329 }
1330
1331 if (msd_send) {
1332 int m_ret = netvsc_send_pkt(device, packet: msd_send, net_device,
1333 NULL, skb: msd_skb);
1334
1335 if (m_ret != 0) {
1336 netvsc_free_send_slot(net_device,
1337 index: msd_send->send_buf_index);
1338 dev_kfree_skb_any(skb: msd_skb);
1339 }
1340 }
1341
1342 if (cur_send)
1343 ret = netvsc_send_pkt(device, packet: cur_send, net_device, pb, skb);
1344
1345 if (ret != 0 && section_index != NETVSC_INVALID_INDEX)
1346 netvsc_free_send_slot(net_device, index: section_index);
1347
1348 return ret;
1349}
1350
1351/* Send pending recv completions */
1352static int send_recv_completions(struct net_device *ndev,
1353 struct netvsc_device *nvdev,
1354 struct netvsc_channel *nvchan)
1355{
1356 struct multi_recv_comp *mrc = &nvchan->mrc;
1357 struct recv_comp_msg {
1358 struct nvsp_message_header hdr;
1359 u32 status;
1360 } __packed;
1361 struct recv_comp_msg msg = {
1362 .hdr.msg_type = NVSP_MSG1_TYPE_SEND_RNDIS_PKT_COMPLETE,
1363 };
1364 int ret;
1365
1366 while (mrc->first != mrc->next) {
1367 const struct recv_comp_data *rcd
1368 = mrc->slots + mrc->first;
1369
1370 msg.status = rcd->status;
1371 ret = vmbus_sendpacket(channel: nvchan->channel, buffer: &msg, bufferLen: sizeof(msg),
1372 requestid: rcd->tid, type: VM_PKT_COMP, flags: 0);
1373 if (unlikely(ret)) {
1374 struct net_device_context *ndev_ctx = netdev_priv(dev: ndev);
1375
1376 ++ndev_ctx->eth_stats.rx_comp_busy;
1377 return ret;
1378 }
1379
1380 if (++mrc->first == nvdev->recv_completion_cnt)
1381 mrc->first = 0;
1382 }
1383
1384 /* receive completion ring has been emptied */
1385 if (unlikely(nvdev->destroy))
1386 wake_up(&nvdev->wait_drain);
1387
1388 return 0;
1389}
1390
1391/* Count how many receive completions are outstanding */
1392static void recv_comp_slot_avail(const struct netvsc_device *nvdev,
1393 const struct multi_recv_comp *mrc,
1394 u32 *filled, u32 *avail)
1395{
1396 u32 count = nvdev->recv_completion_cnt;
1397
1398 if (mrc->next >= mrc->first)
1399 *filled = mrc->next - mrc->first;
1400 else
1401 *filled = (count - mrc->first) + mrc->next;
1402
1403 *avail = count - *filled - 1;
1404}
1405
1406/* Add receive complete to ring to send to host. */
1407static void enq_receive_complete(struct net_device *ndev,
1408 struct netvsc_device *nvdev, u16 q_idx,
1409 u64 tid, u32 status)
1410{
1411 struct netvsc_channel *nvchan = &nvdev->chan_table[q_idx];
1412 struct multi_recv_comp *mrc = &nvchan->mrc;
1413 struct recv_comp_data *rcd;
1414 u32 filled, avail;
1415
1416 recv_comp_slot_avail(nvdev, mrc, filled: &filled, avail: &avail);
1417
1418 if (unlikely(filled > NAPI_POLL_WEIGHT)) {
1419 send_recv_completions(ndev, nvdev, nvchan);
1420 recv_comp_slot_avail(nvdev, mrc, filled: &filled, avail: &avail);
1421 }
1422
1423 if (unlikely(!avail)) {
1424 netdev_err(dev: ndev, format: "Recv_comp full buf q:%hd, tid:%llx\n",
1425 q_idx, tid);
1426 return;
1427 }
1428
1429 rcd = mrc->slots + mrc->next;
1430 rcd->tid = tid;
1431 rcd->status = status;
1432
1433 if (++mrc->next == nvdev->recv_completion_cnt)
1434 mrc->next = 0;
1435}
1436
1437static int netvsc_receive(struct net_device *ndev,
1438 struct netvsc_device *net_device,
1439 struct netvsc_channel *nvchan,
1440 const struct vmpacket_descriptor *desc)
1441{
1442 struct net_device_context *net_device_ctx = netdev_priv(dev: ndev);
1443 struct vmbus_channel *channel = nvchan->channel;
1444 const struct vmtransfer_page_packet_header *vmxferpage_packet
1445 = container_of(desc, const struct vmtransfer_page_packet_header, d);
1446 const struct nvsp_message *nvsp = hv_pkt_data(desc);
1447 u32 msglen = hv_pkt_datalen(desc);
1448 u16 q_idx = channel->offermsg.offer.sub_channel_index;
1449 char *recv_buf = net_device->recv_buf;
1450 u32 status = NVSP_STAT_SUCCESS;
1451 int i;
1452 int count = 0;
1453
1454 /* Ensure packet is big enough to read header fields */
1455 if (msglen < sizeof(struct nvsp_message_header)) {
1456 netif_err(net_device_ctx, rx_err, ndev,
1457 "invalid nvsp header, length too small: %u\n",
1458 msglen);
1459 return 0;
1460 }
1461
1462 /* Make sure this is a valid nvsp packet */
1463 if (unlikely(nvsp->hdr.msg_type != NVSP_MSG1_TYPE_SEND_RNDIS_PKT)) {
1464 netif_err(net_device_ctx, rx_err, ndev,
1465 "Unknown nvsp packet type received %u\n",
1466 nvsp->hdr.msg_type);
1467 return 0;
1468 }
1469
1470 /* Validate xfer page pkt header */
1471 if ((desc->offset8 << 3) < sizeof(struct vmtransfer_page_packet_header)) {
1472 netif_err(net_device_ctx, rx_err, ndev,
1473 "Invalid xfer page pkt, offset too small: %u\n",
1474 desc->offset8 << 3);
1475 return 0;
1476 }
1477
1478 if (unlikely(vmxferpage_packet->xfer_pageset_id != NETVSC_RECEIVE_BUFFER_ID)) {
1479 netif_err(net_device_ctx, rx_err, ndev,
1480 "Invalid xfer page set id - expecting %x got %x\n",
1481 NETVSC_RECEIVE_BUFFER_ID,
1482 vmxferpage_packet->xfer_pageset_id);
1483 return 0;
1484 }
1485
1486 count = vmxferpage_packet->range_cnt;
1487
1488 /* Check count for a valid value */
1489 if (NETVSC_XFER_HEADER_SIZE(count) > desc->offset8 << 3) {
1490 netif_err(net_device_ctx, rx_err, ndev,
1491 "Range count is not valid: %d\n",
1492 count);
1493 return 0;
1494 }
1495
1496 /* Each range represents 1 RNDIS pkt that contains 1 ethernet frame */
1497 for (i = 0; i < count; i++) {
1498 u32 offset = vmxferpage_packet->ranges[i].byte_offset;
1499 u32 buflen = vmxferpage_packet->ranges[i].byte_count;
1500 void *data;
1501 int ret;
1502
1503 if (unlikely(offset > net_device->recv_buf_size ||
1504 buflen > net_device->recv_buf_size - offset)) {
1505 nvchan->rsc.cnt = 0;
1506 status = NVSP_STAT_FAIL;
1507 netif_err(net_device_ctx, rx_err, ndev,
1508 "Packet offset:%u + len:%u too big\n",
1509 offset, buflen);
1510
1511 continue;
1512 }
1513
1514 /* We're going to copy (sections of) the packet into nvchan->recv_buf;
1515 * make sure that nvchan->recv_buf is large enough to hold the packet.
1516 */
1517 if (unlikely(buflen > net_device->recv_section_size)) {
1518 nvchan->rsc.cnt = 0;
1519 status = NVSP_STAT_FAIL;
1520 netif_err(net_device_ctx, rx_err, ndev,
1521 "Packet too big: buflen=%u recv_section_size=%u\n",
1522 buflen, net_device->recv_section_size);
1523
1524 continue;
1525 }
1526
1527 data = recv_buf + offset;
1528
1529 nvchan->rsc.is_last = (i == count - 1);
1530
1531 trace_rndis_recv(ndev, q: q_idx, msg: data);
1532
1533 /* Pass it to the upper layer */
1534 ret = rndis_filter_receive(ndev, net_dev: net_device,
1535 nvchan, data, buflen);
1536
1537 if (unlikely(ret != NVSP_STAT_SUCCESS)) {
1538 /* Drop incomplete packet */
1539 nvchan->rsc.cnt = 0;
1540 status = NVSP_STAT_FAIL;
1541 }
1542 }
1543
1544 enq_receive_complete(ndev, nvdev: net_device, q_idx,
1545 tid: vmxferpage_packet->d.trans_id, status);
1546
1547 return count;
1548}
1549
1550static void netvsc_send_table(struct net_device *ndev,
1551 struct netvsc_device *nvscdev,
1552 const struct nvsp_message *nvmsg,
1553 u32 msglen)
1554{
1555 struct net_device_context *net_device_ctx = netdev_priv(dev: ndev);
1556 u32 count, offset, *tab;
1557 int i;
1558
1559 /* Ensure packet is big enough to read send_table fields */
1560 if (msglen < sizeof(struct nvsp_message_header) +
1561 sizeof(struct nvsp_5_send_indirect_table)) {
1562 netdev_err(dev: ndev, format: "nvsp_v5_msg length too small: %u\n", msglen);
1563 return;
1564 }
1565
1566 count = nvmsg->msg.v5_msg.send_table.count;
1567 offset = nvmsg->msg.v5_msg.send_table.offset;
1568
1569 if (count != VRSS_SEND_TAB_SIZE) {
1570 netdev_err(dev: ndev, format: "Received wrong send-table size:%u\n", count);
1571 return;
1572 }
1573
1574 /* If negotiated version <= NVSP_PROTOCOL_VERSION_6, the offset may be
1575 * wrong due to a host bug. So fix the offset here.
1576 */
1577 if (nvscdev->nvsp_version <= NVSP_PROTOCOL_VERSION_6 &&
1578 msglen >= sizeof(struct nvsp_message_header) +
1579 sizeof(union nvsp_6_message_uber) + count * sizeof(u32))
1580 offset = sizeof(struct nvsp_message_header) +
1581 sizeof(union nvsp_6_message_uber);
1582
1583 /* Boundary check for all versions */
1584 if (msglen < count * sizeof(u32) || offset > msglen - count * sizeof(u32)) {
1585 netdev_err(dev: ndev, format: "Received send-table offset too big:%u\n",
1586 offset);
1587 return;
1588 }
1589
1590 tab = (void *)nvmsg + offset;
1591
1592 for (i = 0; i < count; i++)
1593 net_device_ctx->tx_table[i] = tab[i];
1594}
1595
1596static void netvsc_send_vf(struct net_device *ndev,
1597 const struct nvsp_message *nvmsg,
1598 u32 msglen)
1599{
1600 struct net_device_context *net_device_ctx = netdev_priv(dev: ndev);
1601
1602 /* Ensure packet is big enough to read its fields */
1603 if (msglen < sizeof(struct nvsp_message_header) +
1604 sizeof(struct nvsp_4_send_vf_association)) {
1605 netdev_err(dev: ndev, format: "nvsp_v4_msg length too small: %u\n", msglen);
1606 return;
1607 }
1608
1609 net_device_ctx->vf_alloc = nvmsg->msg.v4_msg.vf_assoc.allocated;
1610 net_device_ctx->vf_serial = nvmsg->msg.v4_msg.vf_assoc.serial;
1611
1612 if (net_device_ctx->vf_alloc)
1613 complete(&net_device_ctx->vf_add);
1614
1615 netdev_info(dev: ndev, format: "VF slot %u %s\n",
1616 net_device_ctx->vf_serial,
1617 net_device_ctx->vf_alloc ? "added" : "removed");
1618}
1619
1620static void netvsc_receive_inband(struct net_device *ndev,
1621 struct netvsc_device *nvscdev,
1622 const struct vmpacket_descriptor *desc)
1623{
1624 const struct nvsp_message *nvmsg = hv_pkt_data(desc);
1625 u32 msglen = hv_pkt_datalen(desc);
1626
1627 /* Ensure packet is big enough to read header fields */
1628 if (msglen < sizeof(struct nvsp_message_header)) {
1629 netdev_err(dev: ndev, format: "inband nvsp_message length too small: %u\n", msglen);
1630 return;
1631 }
1632
1633 switch (nvmsg->hdr.msg_type) {
1634 case NVSP_MSG5_TYPE_SEND_INDIRECTION_TABLE:
1635 netvsc_send_table(ndev, nvscdev, nvmsg, msglen);
1636 break;
1637
1638 case NVSP_MSG4_TYPE_SEND_VF_ASSOCIATION:
1639 if (hv_is_isolation_supported())
1640 netdev_err(dev: ndev, format: "Ignore VF_ASSOCIATION msg from the host supporting isolation\n");
1641 else
1642 netvsc_send_vf(ndev, nvmsg, msglen);
1643 break;
1644 }
1645}
1646
1647static int netvsc_process_raw_pkt(struct hv_device *device,
1648 struct netvsc_channel *nvchan,
1649 struct netvsc_device *net_device,
1650 struct net_device *ndev,
1651 const struct vmpacket_descriptor *desc,
1652 int budget)
1653{
1654 struct vmbus_channel *channel = nvchan->channel;
1655 const struct nvsp_message *nvmsg = hv_pkt_data(desc);
1656
1657 trace_nvsp_recv(ndev, chan: channel, msg: nvmsg);
1658
1659 switch (desc->type) {
1660 case VM_PKT_COMP:
1661 netvsc_send_completion(ndev, net_device, incoming_channel: channel, desc, budget);
1662 break;
1663
1664 case VM_PKT_DATA_USING_XFER_PAGES:
1665 return netvsc_receive(ndev, net_device, nvchan, desc);
1666
1667 case VM_PKT_DATA_INBAND:
1668 netvsc_receive_inband(ndev, nvscdev: net_device, desc);
1669 break;
1670
1671 default:
1672 netdev_err(dev: ndev, format: "unhandled packet type %d, tid %llx\n",
1673 desc->type, desc->trans_id);
1674 break;
1675 }
1676
1677 return 0;
1678}
1679
1680static struct hv_device *netvsc_channel_to_device(struct vmbus_channel *channel)
1681{
1682 struct vmbus_channel *primary = channel->primary_channel;
1683
1684 return primary ? primary->device_obj : channel->device_obj;
1685}
1686
1687/* Network processing softirq
1688 * Process data in incoming ring buffer from host
1689 * Stops when ring is empty or budget is met or exceeded.
1690 */
1691int netvsc_poll(struct napi_struct *napi, int budget)
1692{
1693 struct netvsc_channel *nvchan
1694 = container_of(napi, struct netvsc_channel, napi);
1695 struct netvsc_device *net_device = nvchan->net_device;
1696 struct vmbus_channel *channel = nvchan->channel;
1697 struct hv_device *device = netvsc_channel_to_device(channel);
1698 struct net_device *ndev = hv_get_drvdata(dev: device);
1699 int work_done = 0;
1700 int ret;
1701
1702 /* If starting a new interval */
1703 if (!nvchan->desc)
1704 nvchan->desc = hv_pkt_iter_first(channel);
1705
1706 nvchan->xdp_flush = false;
1707
1708 while (nvchan->desc && work_done < budget) {
1709 work_done += netvsc_process_raw_pkt(device, nvchan, net_device,
1710 ndev, desc: nvchan->desc, budget);
1711 nvchan->desc = hv_pkt_iter_next(channel, pkt: nvchan->desc);
1712 }
1713
1714 if (nvchan->xdp_flush)
1715 xdp_do_flush();
1716
1717 /* Send any pending receive completions */
1718 ret = send_recv_completions(ndev, nvdev: net_device, nvchan);
1719
1720 /* If it did not exhaust NAPI budget this time
1721 * and not doing busy poll
1722 * then re-enable host interrupts
1723 * and reschedule if ring is not empty
1724 * or sending receive completion failed.
1725 */
1726 if (work_done < budget &&
1727 napi_complete_done(n: napi, work_done) &&
1728 (ret || hv_end_read(rbi: &channel->inbound)) &&
1729 napi_schedule_prep(n: napi)) {
1730 hv_begin_read(rbi: &channel->inbound);
1731 __napi_schedule(n: napi);
1732 }
1733
1734 /* Driver may overshoot since multiple packets per descriptor */
1735 return min(work_done, budget);
1736}
1737
1738/* Call back when data is available in host ring buffer.
1739 * Processing is deferred until network softirq (NAPI)
1740 */
1741void netvsc_channel_cb(void *context)
1742{
1743 struct netvsc_channel *nvchan = context;
1744 struct vmbus_channel *channel = nvchan->channel;
1745 struct hv_ring_buffer_info *rbi = &channel->inbound;
1746
1747 /* preload first vmpacket descriptor */
1748 prefetch(hv_get_ring_buffer(rbi) + rbi->priv_read_index);
1749
1750 if (napi_schedule_prep(n: &nvchan->napi)) {
1751 /* disable interrupts from host */
1752 hv_begin_read(rbi);
1753
1754 __napi_schedule_irqoff(n: &nvchan->napi);
1755 }
1756}
1757
1758/*
1759 * netvsc_device_add - Callback when the device belonging to this
1760 * driver is added
1761 */
1762struct netvsc_device *netvsc_device_add(struct hv_device *device,
1763 const struct netvsc_device_info *device_info)
1764{
1765 int i, ret = 0;
1766 struct netvsc_device *net_device;
1767 struct net_device *ndev = hv_get_drvdata(dev: device);
1768 struct net_device_context *net_device_ctx = netdev_priv(dev: ndev);
1769
1770 net_device = alloc_net_device();
1771 if (!net_device)
1772 return ERR_PTR(error: -ENOMEM);
1773
1774 for (i = 0; i < VRSS_SEND_TAB_SIZE; i++)
1775 net_device_ctx->tx_table[i] = 0;
1776
1777 /* Because the device uses NAPI, all the interrupt batching and
1778 * control is done via Net softirq, not the channel handling
1779 */
1780 set_channel_read_mode(c: device->channel, mode: HV_CALL_ISR);
1781
1782 /* If we're reopening the device we may have multiple queues, fill the
1783 * chn_table with the default channel to use it before subchannels are
1784 * opened.
1785 * Initialize the channel state before we open;
1786 * we can be interrupted as soon as we open the channel.
1787 */
1788
1789 for (i = 0; i < VRSS_CHANNEL_MAX; i++) {
1790 struct netvsc_channel *nvchan = &net_device->chan_table[i];
1791
1792 nvchan->channel = device->channel;
1793 nvchan->net_device = net_device;
1794 u64_stats_init(syncp: &nvchan->tx_stats.syncp);
1795 u64_stats_init(syncp: &nvchan->rx_stats.syncp);
1796
1797 ret = xdp_rxq_info_reg(xdp_rxq: &nvchan->xdp_rxq, dev: ndev, queue_index: i, napi_id: 0);
1798
1799 if (ret) {
1800 netdev_err(dev: ndev, format: "xdp_rxq_info_reg fail: %d\n", ret);
1801 goto cleanup2;
1802 }
1803
1804 ret = xdp_rxq_info_reg_mem_model(xdp_rxq: &nvchan->xdp_rxq,
1805 type: MEM_TYPE_PAGE_SHARED, NULL);
1806
1807 if (ret) {
1808 netdev_err(dev: ndev, format: "xdp reg_mem_model fail: %d\n", ret);
1809 goto cleanup2;
1810 }
1811 }
1812
1813 /* Enable NAPI handler before init callbacks */
1814 netif_napi_add(dev: ndev, napi: &net_device->chan_table[0].napi, poll: netvsc_poll);
1815 napi_enable(n: &net_device->chan_table[0].napi);
1816 netif_queue_set_napi(dev: ndev, queue_index: 0, type: NETDEV_QUEUE_TYPE_RX,
1817 napi: &net_device->chan_table[0].napi);
1818 netif_queue_set_napi(dev: ndev, queue_index: 0, type: NETDEV_QUEUE_TYPE_TX,
1819 napi: &net_device->chan_table[0].napi);
1820
1821 /* Open the channel */
1822 device->channel->next_request_id_callback = vmbus_next_request_id;
1823 device->channel->request_addr_callback = vmbus_request_addr;
1824 device->channel->rqstor_size = netvsc_rqstor_size(ringbytes: netvsc_ring_bytes);
1825 device->channel->max_pkt_size = NETVSC_MAX_PKT_SIZE;
1826
1827 ret = vmbus_open(channel: device->channel, send_ringbuffersize: netvsc_ring_bytes,
1828 recv_ringbuffersize: netvsc_ring_bytes, NULL, userdatalen: 0,
1829 onchannel_callback: netvsc_channel_cb, context: net_device->chan_table);
1830
1831 if (ret != 0) {
1832 netdev_err(dev: ndev, format: "unable to open channel: %d\n", ret);
1833 goto cleanup;
1834 }
1835
1836 /* Channel is opened */
1837 netdev_dbg(ndev, "hv_netvsc channel opened successfully\n");
1838
1839 /* Connect with the NetVsp */
1840 ret = netvsc_connect_vsp(device, net_device, device_info);
1841 if (ret != 0) {
1842 netdev_err(dev: ndev,
1843 format: "unable to connect to NetVSP - %d\n", ret);
1844 goto close;
1845 }
1846
1847 /* Writing nvdev pointer unlocks netvsc_send(), make sure chn_table is
1848 * populated.
1849 */
1850 rcu_assign_pointer(net_device_ctx->nvdev, net_device);
1851
1852 return net_device;
1853
1854close:
1855 RCU_INIT_POINTER(net_device_ctx->nvdev, NULL);
1856
1857 /* Now, we can close the channel safely */
1858 vmbus_close(channel: device->channel);
1859
1860cleanup:
1861 netif_queue_set_napi(dev: ndev, queue_index: 0, type: NETDEV_QUEUE_TYPE_TX, NULL);
1862 netif_queue_set_napi(dev: ndev, queue_index: 0, type: NETDEV_QUEUE_TYPE_RX, NULL);
1863 napi_disable(n: &net_device->chan_table[0].napi);
1864 netif_napi_del(napi: &net_device->chan_table[0].napi);
1865
1866cleanup2:
1867 free_netvsc_device(head: &net_device->rcu);
1868
1869 return ERR_PTR(error: ret);
1870}
1871

source code of linux/drivers/net/hyperv/netvsc.c