Loading...
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(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, init_pkt);
58
59 ret = vmbus_sendpacket(dev->channel, init_pkt,
60 sizeof(struct nvsp_message),
61 (unsigned long)init_pkt, 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(ndev,
70 "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 ndev,
81 "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(w);
107 return;
108 }
109
110 rdev = nvdev->extension;
111 if (rdev) {
112 ret = rndis_set_subchannel(rdev->ndev, nvdev, NULL);
113 if (ret == 0) {
114 netif_device_attach(rdev->ndev);
115 } else {
116 /* fallback to only primary channel */
117 for (i = 1; i < nvdev->num_chn; i++)
118 netif_napi_del(&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(&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(nvdev->extension);
157
158 if (nvdev->recv_original_buf)
159 vfree(nvdev->recv_original_buf);
160 else
161 vfree(nvdev->recv_buf);
162
163 if (nvdev->send_original_buf)
164 vfree(nvdev->send_original_buf);
165 else
166 vfree(nvdev->send_buf);
167
168 bitmap_free(nvdev->send_section_map);
169
170 for (i = 0; i < VRSS_CHANNEL_MAX; i++) {
171 xdp_rxq_info_unreg(&nvdev->chan_table[i].xdp_rxq);
172 kfree(nvdev->chan_table[i].recv_buf);
173 vfree(nvdev->chan_table[i].mrc.slots);
174 }
175
176 kfree(nvdev);
177}
178
179static void free_netvsc_device_rcu(struct netvsc_device *nvdev)
180{
181 call_rcu(&nvdev->rcu, free_netvsc_device);
182}
183
184static void netvsc_revoke_recv_buf(struct hv_device *device,
185 struct netvsc_device *net_device,
186 struct net_device *ndev)
187{
188 struct nvsp_message *revoke_packet;
189 int ret;
190
191 /*
192 * If we got a section count, it means we received a
193 * SendReceiveBufferComplete msg (ie sent
194 * NvspMessage1TypeSendReceiveBuffer msg) therefore, we need
195 * to send a revoke msg here
196 */
197 if (net_device->recv_section_cnt) {
198 /* Send the revoke receive buffer */
199 revoke_packet = &net_device->revoke_packet;
200 memset(revoke_packet, 0, sizeof(struct nvsp_message));
201
202 revoke_packet->hdr.msg_type =
203 NVSP_MSG1_TYPE_REVOKE_RECV_BUF;
204 revoke_packet->msg.v1_msg.
205 revoke_recv_buf.id = NETVSC_RECEIVE_BUFFER_ID;
206
207 trace_nvsp_send(ndev, revoke_packet);
208
209 ret = vmbus_sendpacket(device->channel,
210 revoke_packet,
211 sizeof(struct nvsp_message),
212 VMBUS_RQST_ID_NO_RESPONSE,
213 VM_PKT_DATA_INBAND, 0);
214 /* If the failure is because the channel is rescinded;
215 * ignore the failure since we cannot send on a rescinded
216 * channel. This would allow us to properly cleanup
217 * even when the channel is rescinded.
218 */
219 if (device->channel->rescind)
220 ret = 0;
221 /*
222 * If we failed here, we might as well return and
223 * have a leak rather than continue and a bugchk
224 */
225 if (ret != 0) {
226 netdev_err(ndev, "unable to send "
227 "revoke receive buffer to netvsp\n");
228 return;
229 }
230 net_device->recv_section_cnt = 0;
231 }
232}
233
234static void netvsc_revoke_send_buf(struct hv_device *device,
235 struct netvsc_device *net_device,
236 struct net_device *ndev)
237{
238 struct nvsp_message *revoke_packet;
239 int ret;
240
241 /* Deal with the send buffer we may have setup.
242 * If we got a send section size, it means we received a
243 * NVSP_MSG1_TYPE_SEND_SEND_BUF_COMPLETE msg (ie sent
244 * NVSP_MSG1_TYPE_SEND_SEND_BUF msg) therefore, we need
245 * to send a revoke msg here
246 */
247 if (net_device->send_section_cnt) {
248 /* Send the revoke receive buffer */
249 revoke_packet = &net_device->revoke_packet;
250 memset(revoke_packet, 0, sizeof(struct nvsp_message));
251
252 revoke_packet->hdr.msg_type =
253 NVSP_MSG1_TYPE_REVOKE_SEND_BUF;
254 revoke_packet->msg.v1_msg.revoke_send_buf.id =
255 NETVSC_SEND_BUFFER_ID;
256
257 trace_nvsp_send(ndev, revoke_packet);
258
259 ret = vmbus_sendpacket(device->channel,
260 revoke_packet,
261 sizeof(struct nvsp_message),
262 VMBUS_RQST_ID_NO_RESPONSE,
263 VM_PKT_DATA_INBAND, 0);
264
265 /* If the failure is because the channel is rescinded;
266 * ignore the failure since we cannot send on a rescinded
267 * channel. This would allow us to properly cleanup
268 * even when the channel is rescinded.
269 */
270 if (device->channel->rescind)
271 ret = 0;
272
273 /* If we failed here, we might as well return and
274 * have a leak rather than continue and a bugchk
275 */
276 if (ret != 0) {
277 netdev_err(ndev, "unable to send "
278 "revoke send buffer to netvsp\n");
279 return;
280 }
281 net_device->send_section_cnt = 0;
282 }
283}
284
285static void netvsc_teardown_recv_gpadl(struct hv_device *device,
286 struct netvsc_device *net_device,
287 struct net_device *ndev)
288{
289 int ret;
290
291 if (net_device->recv_buf_gpadl_handle.gpadl_handle) {
292 ret = vmbus_teardown_gpadl(device->channel,
293 &net_device->recv_buf_gpadl_handle);
294
295 /* If we failed here, we might as well return and have a leak
296 * rather than continue and a bugchk
297 */
298 if (ret != 0) {
299 netdev_err(ndev,
300 "unable to teardown receive buffer's gpadl\n");
301 return;
302 }
303 }
304}
305
306static void netvsc_teardown_send_gpadl(struct hv_device *device,
307 struct netvsc_device *net_device,
308 struct net_device *ndev)
309{
310 int ret;
311
312 if (net_device->send_buf_gpadl_handle.gpadl_handle) {
313 ret = vmbus_teardown_gpadl(device->channel,
314 &net_device->send_buf_gpadl_handle);
315
316 /* If we failed here, we might as well return and have a leak
317 * rather than continue and a bugchk
318 */
319 if (ret != 0) {
320 netdev_err(ndev,
321 "unable to teardown send buffer's gpadl\n");
322 return;
323 }
324 }
325}
326
327int netvsc_alloc_recv_comp_ring(struct netvsc_device *net_device, u32 q_idx)
328{
329 struct netvsc_channel *nvchan = &net_device->chan_table[q_idx];
330 int node = cpu_to_node(nvchan->channel->target_cpu);
331 size_t size;
332
333 size = net_device->recv_completion_cnt * sizeof(struct recv_comp_data);
334 nvchan->mrc.slots = vzalloc_node(size, node);
335 if (!nvchan->mrc.slots)
336 nvchan->mrc.slots = vzalloc(size);
337
338 return nvchan->mrc.slots ? 0 : -ENOMEM;
339}
340
341static int netvsc_init_buf(struct hv_device *device,
342 struct netvsc_device *net_device,
343 const struct netvsc_device_info *device_info)
344{
345 struct nvsp_1_message_send_receive_buffer_complete *resp;
346 struct net_device *ndev = hv_get_drvdata(device);
347 struct nvsp_message *init_packet;
348 unsigned int buf_size;
349 int i, ret = 0;
350 void *vaddr;
351
352 /* Get receive buffer area. */
353 buf_size = device_info->recv_sections * device_info->recv_section_size;
354 buf_size = roundup(buf_size, PAGE_SIZE);
355
356 /* Legacy hosts only allow smaller receive buffer */
357 if (net_device->nvsp_version <= NVSP_PROTOCOL_VERSION_2)
358 buf_size = min_t(unsigned int, buf_size,
359 NETVSC_RECEIVE_BUFFER_SIZE_LEGACY);
360
361 net_device->recv_buf = vzalloc(buf_size);
362 if (!net_device->recv_buf) {
363 netdev_err(ndev,
364 "unable to allocate receive buffer of size %u\n",
365 buf_size);
366 ret = -ENOMEM;
367 goto cleanup;
368 }
369
370 net_device->recv_buf_size = buf_size;
371
372 /*
373 * Establish the gpadl handle for this buffer on this
374 * channel. Note: This call uses the vmbus connection rather
375 * than the channel to establish the gpadl handle.
376 */
377 ret = vmbus_establish_gpadl(device->channel, net_device->recv_buf,
378 buf_size,
379 &net_device->recv_buf_gpadl_handle);
380 if (ret != 0) {
381 netdev_err(ndev,
382 "unable to establish receive buffer's gpadl\n");
383 goto cleanup;
384 }
385
386 if (hv_isolation_type_snp()) {
387 vaddr = hv_map_memory(net_device->recv_buf, buf_size);
388 if (!vaddr) {
389 ret = -ENOMEM;
390 goto cleanup;
391 }
392
393 net_device->recv_original_buf = net_device->recv_buf;
394 net_device->recv_buf = vaddr;
395 }
396
397 /* Notify the NetVsp of the gpadl handle */
398 init_packet = &net_device->channel_init_pkt;
399 memset(init_packet, 0, sizeof(struct nvsp_message));
400 init_packet->hdr.msg_type = NVSP_MSG1_TYPE_SEND_RECV_BUF;
401 init_packet->msg.v1_msg.send_recv_buf.
402 gpadl_handle = net_device->recv_buf_gpadl_handle.gpadl_handle;
403 init_packet->msg.v1_msg.
404 send_recv_buf.id = NETVSC_RECEIVE_BUFFER_ID;
405
406 trace_nvsp_send(ndev, init_packet);
407
408 /* Send the gpadl notification request */
409 ret = vmbus_sendpacket(device->channel, init_packet,
410 sizeof(struct nvsp_message),
411 (unsigned long)init_packet,
412 VM_PKT_DATA_INBAND,
413 VMBUS_DATA_PACKET_FLAG_COMPLETION_REQUESTED);
414 if (ret != 0) {
415 netdev_err(ndev,
416 "unable to send receive buffer's gpadl to netvsp\n");
417 goto cleanup;
418 }
419
420 wait_for_completion(&net_device->channel_init_wait);
421
422 /* Check the response */
423 resp = &init_packet->msg.v1_msg.send_recv_buf_complete;
424 if (resp->status != NVSP_STAT_SUCCESS) {
425 netdev_err(ndev,
426 "Unable to complete receive buffer initialization with NetVsp - status %d\n",
427 resp->status);
428 ret = -EINVAL;
429 goto cleanup;
430 }
431
432 /* Parse the response */
433 netdev_dbg(ndev, "Receive sections: %u sub_allocs: size %u count: %u\n",
434 resp->num_sections, resp->sections[0].sub_alloc_size,
435 resp->sections[0].num_sub_allocs);
436
437 /* There should only be one section for the entire receive buffer */
438 if (resp->num_sections != 1 || resp->sections[0].offset != 0) {
439 ret = -EINVAL;
440 goto cleanup;
441 }
442
443 net_device->recv_section_size = resp->sections[0].sub_alloc_size;
444 net_device->recv_section_cnt = resp->sections[0].num_sub_allocs;
445
446 /* Ensure buffer will not overflow */
447 if (net_device->recv_section_size < NETVSC_MTU_MIN || (u64)net_device->recv_section_size *
448 (u64)net_device->recv_section_cnt > (u64)buf_size) {
449 netdev_err(ndev, "invalid recv_section_size %u\n",
450 net_device->recv_section_size);
451 ret = -EINVAL;
452 goto cleanup;
453 }
454
455 for (i = 0; i < VRSS_CHANNEL_MAX; i++) {
456 struct netvsc_channel *nvchan = &net_device->chan_table[i];
457
458 nvchan->recv_buf = kzalloc(net_device->recv_section_size, GFP_KERNEL);
459 if (nvchan->recv_buf == NULL) {
460 ret = -ENOMEM;
461 goto cleanup;
462 }
463 }
464
465 /* Setup receive completion ring.
466 * Add 1 to the recv_section_cnt because at least one entry in a
467 * ring buffer has to be empty.
468 */
469 net_device->recv_completion_cnt = net_device->recv_section_cnt + 1;
470 ret = netvsc_alloc_recv_comp_ring(net_device, 0);
471 if (ret)
472 goto cleanup;
473
474 /* Now setup the send buffer. */
475 buf_size = device_info->send_sections * device_info->send_section_size;
476 buf_size = round_up(buf_size, PAGE_SIZE);
477
478 net_device->send_buf = vzalloc(buf_size);
479 if (!net_device->send_buf) {
480 netdev_err(ndev, "unable to allocate send buffer of size %u\n",
481 buf_size);
482 ret = -ENOMEM;
483 goto cleanup;
484 }
485 net_device->send_buf_size = buf_size;
486
487 /* Establish the gpadl handle for this buffer on this
488 * channel. Note: This call uses the vmbus connection rather
489 * than the channel to establish the gpadl handle.
490 */
491 ret = vmbus_establish_gpadl(device->channel, net_device->send_buf,
492 buf_size,
493 &net_device->send_buf_gpadl_handle);
494 if (ret != 0) {
495 netdev_err(ndev,
496 "unable to establish send buffer's gpadl\n");
497 goto cleanup;
498 }
499
500 if (hv_isolation_type_snp()) {
501 vaddr = hv_map_memory(net_device->send_buf, buf_size);
502 if (!vaddr) {
503 ret = -ENOMEM;
504 goto cleanup;
505 }
506
507 net_device->send_original_buf = net_device->send_buf;
508 net_device->send_buf = vaddr;
509 }
510
511 /* Notify the NetVsp of the gpadl handle */
512 init_packet = &net_device->channel_init_pkt;
513 memset(init_packet, 0, sizeof(struct nvsp_message));
514 init_packet->hdr.msg_type = NVSP_MSG1_TYPE_SEND_SEND_BUF;
515 init_packet->msg.v1_msg.send_send_buf.gpadl_handle =
516 net_device->send_buf_gpadl_handle.gpadl_handle;
517 init_packet->msg.v1_msg.send_send_buf.id = NETVSC_SEND_BUFFER_ID;
518
519 trace_nvsp_send(ndev, init_packet);
520
521 /* Send the gpadl notification request */
522 ret = vmbus_sendpacket(device->channel, init_packet,
523 sizeof(struct nvsp_message),
524 (unsigned long)init_packet,
525 VM_PKT_DATA_INBAND,
526 VMBUS_DATA_PACKET_FLAG_COMPLETION_REQUESTED);
527 if (ret != 0) {
528 netdev_err(ndev,
529 "unable to send send buffer's gpadl to netvsp\n");
530 goto cleanup;
531 }
532
533 wait_for_completion(&net_device->channel_init_wait);
534
535 /* Check the response */
536 if (init_packet->msg.v1_msg.
537 send_send_buf_complete.status != NVSP_STAT_SUCCESS) {
538 netdev_err(ndev, "Unable to complete send buffer "
539 "initialization with NetVsp - status %d\n",
540 init_packet->msg.v1_msg.
541 send_send_buf_complete.status);
542 ret = -EINVAL;
543 goto cleanup;
544 }
545
546 /* Parse the response */
547 net_device->send_section_size = init_packet->msg.
548 v1_msg.send_send_buf_complete.section_size;
549 if (net_device->send_section_size < NETVSC_MTU_MIN) {
550 netdev_err(ndev, "invalid send_section_size %u\n",
551 net_device->send_section_size);
552 ret = -EINVAL;
553 goto cleanup;
554 }
555
556 /* Section count is simply the size divided by the section size. */
557 net_device->send_section_cnt = buf_size / net_device->send_section_size;
558
559 netdev_dbg(ndev, "Send section size: %d, Section count:%d\n",
560 net_device->send_section_size, net_device->send_section_cnt);
561
562 /* Setup state for managing the send buffer. */
563 net_device->send_section_map = bitmap_zalloc(net_device->send_section_cnt,
564 GFP_KERNEL);
565 if (!net_device->send_section_map) {
566 ret = -ENOMEM;
567 goto cleanup;
568 }
569
570 goto exit;
571
572cleanup:
573 netvsc_revoke_recv_buf(device, net_device, ndev);
574 netvsc_revoke_send_buf(device, net_device, ndev);
575 netvsc_teardown_recv_gpadl(device, net_device, ndev);
576 netvsc_teardown_send_gpadl(device, net_device, ndev);
577
578exit:
579 return ret;
580}
581
582/* Negotiate NVSP protocol version */
583static int negotiate_nvsp_ver(struct hv_device *device,
584 struct netvsc_device *net_device,
585 struct nvsp_message *init_packet,
586 u32 nvsp_ver)
587{
588 struct net_device *ndev = hv_get_drvdata(device);
589 int ret;
590
591 memset(init_packet, 0, sizeof(struct nvsp_message));
592 init_packet->hdr.msg_type = NVSP_MSG_TYPE_INIT;
593 init_packet->msg.init_msg.init.min_protocol_ver = nvsp_ver;
594 init_packet->msg.init_msg.init.max_protocol_ver = nvsp_ver;
595 trace_nvsp_send(ndev, init_packet);
596
597 /* Send the init request */
598 ret = vmbus_sendpacket(device->channel, init_packet,
599 sizeof(struct nvsp_message),
600 (unsigned long)init_packet,
601 VM_PKT_DATA_INBAND,
602 VMBUS_DATA_PACKET_FLAG_COMPLETION_REQUESTED);
603
604 if (ret != 0)
605 return ret;
606
607 wait_for_completion(&net_device->channel_init_wait);
608
609 if (init_packet->msg.init_msg.init_complete.status !=
610 NVSP_STAT_SUCCESS)
611 return -EINVAL;
612
613 if (nvsp_ver == NVSP_PROTOCOL_VERSION_1)
614 return 0;
615
616 /* NVSPv2 or later: Send NDIS config */
617 memset(init_packet, 0, sizeof(struct nvsp_message));
618 init_packet->hdr.msg_type = NVSP_MSG2_TYPE_SEND_NDIS_CONFIG;
619 init_packet->msg.v2_msg.send_ndis_config.mtu = ndev->mtu + ETH_HLEN;
620 init_packet->msg.v2_msg.send_ndis_config.capability.ieee8021q = 1;
621
622 if (nvsp_ver >= NVSP_PROTOCOL_VERSION_5) {
623 if (hv_is_isolation_supported())
624 netdev_info(ndev, "SR-IOV not advertised by guests on the host supporting isolation\n");
625 else
626 init_packet->msg.v2_msg.send_ndis_config.capability.sriov = 1;
627
628 /* Teaming bit is needed to receive link speed updates */
629 init_packet->msg.v2_msg.send_ndis_config.capability.teaming = 1;
630 }
631
632 if (nvsp_ver >= NVSP_PROTOCOL_VERSION_61)
633 init_packet->msg.v2_msg.send_ndis_config.capability.rsc = 1;
634
635 trace_nvsp_send(ndev, init_packet);
636
637 ret = vmbus_sendpacket(device->channel, init_packet,
638 sizeof(struct nvsp_message),
639 VMBUS_RQST_ID_NO_RESPONSE,
640 VM_PKT_DATA_INBAND, 0);
641
642 return ret;
643}
644
645static int netvsc_connect_vsp(struct hv_device *device,
646 struct netvsc_device *net_device,
647 const struct netvsc_device_info *device_info)
648{
649 struct net_device *ndev = hv_get_drvdata(device);
650 static const u32 ver_list[] = {
651 NVSP_PROTOCOL_VERSION_1, NVSP_PROTOCOL_VERSION_2,
652 NVSP_PROTOCOL_VERSION_4, NVSP_PROTOCOL_VERSION_5,
653 NVSP_PROTOCOL_VERSION_6, NVSP_PROTOCOL_VERSION_61
654 };
655 struct nvsp_message *init_packet;
656 int ndis_version, i, ret;
657
658 init_packet = &net_device->channel_init_pkt;
659
660 /* Negotiate the latest NVSP protocol supported */
661 for (i = ARRAY_SIZE(ver_list) - 1; i >= 0; i--)
662 if (negotiate_nvsp_ver(device, net_device, init_packet,
663 ver_list[i]) == 0) {
664 net_device->nvsp_version = ver_list[i];
665 break;
666 }
667
668 if (i < 0) {
669 ret = -EPROTO;
670 goto cleanup;
671 }
672
673 if (hv_is_isolation_supported() && net_device->nvsp_version < NVSP_PROTOCOL_VERSION_61) {
674 netdev_err(ndev, "Invalid NVSP version 0x%x (expected >= 0x%x) from the host supporting isolation\n",
675 net_device->nvsp_version, NVSP_PROTOCOL_VERSION_61);
676 ret = -EPROTO;
677 goto cleanup;
678 }
679
680 pr_debug("Negotiated NVSP version:%x\n", net_device->nvsp_version);
681
682 /* Send the ndis version */
683 memset(init_packet, 0, sizeof(struct nvsp_message));
684
685 if (net_device->nvsp_version <= NVSP_PROTOCOL_VERSION_4)
686 ndis_version = 0x00060001;
687 else
688 ndis_version = 0x0006001e;
689
690 init_packet->hdr.msg_type = NVSP_MSG1_TYPE_SEND_NDIS_VER;
691 init_packet->msg.v1_msg.
692 send_ndis_ver.ndis_major_ver =
693 (ndis_version & 0xFFFF0000) >> 16;
694 init_packet->msg.v1_msg.
695 send_ndis_ver.ndis_minor_ver =
696 ndis_version & 0xFFFF;
697
698 trace_nvsp_send(ndev, init_packet);
699
700 /* Send the init request */
701 ret = vmbus_sendpacket(device->channel, init_packet,
702 sizeof(struct nvsp_message),
703 VMBUS_RQST_ID_NO_RESPONSE,
704 VM_PKT_DATA_INBAND, 0);
705 if (ret != 0)
706 goto cleanup;
707
708
709 ret = netvsc_init_buf(device, net_device, device_info);
710
711cleanup:
712 return ret;
713}
714
715/*
716 * netvsc_device_remove - Callback when the root bus device is removed
717 */
718void netvsc_device_remove(struct hv_device *device)
719{
720 struct net_device *ndev = hv_get_drvdata(device);
721 struct net_device_context *net_device_ctx = netdev_priv(ndev);
722 struct netvsc_device *net_device
723 = rtnl_dereference(net_device_ctx->nvdev);
724 int i;
725
726 /*
727 * Revoke receive buffer. If host is pre-Win2016 then tear down
728 * receive buffer GPADL. Do the same for send buffer.
729 */
730 netvsc_revoke_recv_buf(device, net_device, ndev);
731 if (vmbus_proto_version < VERSION_WIN10)
732 netvsc_teardown_recv_gpadl(device, net_device, ndev);
733
734 netvsc_revoke_send_buf(device, net_device, ndev);
735 if (vmbus_proto_version < VERSION_WIN10)
736 netvsc_teardown_send_gpadl(device, net_device, ndev);
737
738 RCU_INIT_POINTER(net_device_ctx->nvdev, NULL);
739
740 /* Disable NAPI and disassociate its context from the device. */
741 for (i = 0; i < net_device->num_chn; i++) {
742 /* See also vmbus_reset_channel_cb(). */
743 napi_disable(&net_device->chan_table[i].napi);
744 netif_napi_del(&net_device->chan_table[i].napi);
745 }
746
747 /*
748 * At this point, no one should be accessing net_device
749 * except in here
750 */
751 netdev_dbg(ndev, "net device safe to remove\n");
752
753 /* Now, we can close the channel safely */
754 vmbus_close(device->channel);
755
756 /*
757 * If host is Win2016 or higher then we do the GPADL tear down
758 * here after VMBus is closed.
759 */
760 if (vmbus_proto_version >= VERSION_WIN10) {
761 netvsc_teardown_recv_gpadl(device, net_device, ndev);
762 netvsc_teardown_send_gpadl(device, net_device, ndev);
763 }
764
765 if (net_device->recv_original_buf)
766 hv_unmap_memory(net_device->recv_buf);
767
768 if (net_device->send_original_buf)
769 hv_unmap_memory(net_device->send_buf);
770
771 /* Release all resources */
772 free_netvsc_device_rcu(net_device);
773}
774
775#define RING_AVAIL_PERCENT_HIWATER 20
776#define RING_AVAIL_PERCENT_LOWATER 10
777
778static inline void netvsc_free_send_slot(struct netvsc_device *net_device,
779 u32 index)
780{
781 sync_change_bit(index, net_device->send_section_map);
782}
783
784static void netvsc_send_tx_complete(struct net_device *ndev,
785 struct netvsc_device *net_device,
786 struct vmbus_channel *channel,
787 const struct vmpacket_descriptor *desc,
788 int budget)
789{
790 struct net_device_context *ndev_ctx = netdev_priv(ndev);
791 struct sk_buff *skb;
792 u16 q_idx = 0;
793 int queue_sends;
794 u64 cmd_rqst;
795
796 cmd_rqst = channel->request_addr_callback(channel, desc->trans_id);
797 if (cmd_rqst == VMBUS_RQST_ERROR) {
798 netdev_err(ndev, "Invalid transaction ID %llx\n", desc->trans_id);
799 return;
800 }
801
802 skb = (struct sk_buff *)(unsigned long)cmd_rqst;
803
804 /* Notify the layer above us */
805 if (likely(skb)) {
806 struct hv_netvsc_packet *packet
807 = (struct hv_netvsc_packet *)skb->cb;
808 u32 send_index = packet->send_buf_index;
809 struct netvsc_stats_tx *tx_stats;
810
811 if (send_index != NETVSC_INVALID_INDEX)
812 netvsc_free_send_slot(net_device, send_index);
813 q_idx = packet->q_idx;
814
815 tx_stats = &net_device->chan_table[q_idx].tx_stats;
816
817 u64_stats_update_begin(&tx_stats->syncp);
818 tx_stats->packets += packet->total_packets;
819 tx_stats->bytes += packet->total_bytes;
820 u64_stats_update_end(&tx_stats->syncp);
821
822 netvsc_dma_unmap(ndev_ctx->device_ctx, packet);
823 napi_consume_skb(skb, budget);
824 }
825
826 queue_sends =
827 atomic_dec_return(&net_device->chan_table[q_idx].queue_sends);
828
829 if (unlikely(net_device->destroy)) {
830 if (queue_sends == 0)
831 wake_up(&net_device->wait_drain);
832 } else {
833 struct netdev_queue *txq = netdev_get_tx_queue(ndev, q_idx);
834
835 if (netif_tx_queue_stopped(txq) && !net_device->tx_disable &&
836 (hv_get_avail_to_write_percent(&channel->outbound) >
837 RING_AVAIL_PERCENT_HIWATER || queue_sends < 1)) {
838 netif_tx_wake_queue(txq);
839 ndev_ctx->eth_stats.wake_queue++;
840 }
841 }
842}
843
844static void netvsc_send_completion(struct net_device *ndev,
845 struct netvsc_device *net_device,
846 struct vmbus_channel *incoming_channel,
847 const struct vmpacket_descriptor *desc,
848 int budget)
849{
850 const struct nvsp_message *nvsp_packet;
851 u32 msglen = hv_pkt_datalen(desc);
852 struct nvsp_message *pkt_rqst;
853 u64 cmd_rqst;
854
855 /* First check if this is a VMBUS completion without data payload */
856 if (!msglen) {
857 cmd_rqst = incoming_channel->request_addr_callback(incoming_channel,
858 desc->trans_id);
859 if (cmd_rqst == VMBUS_RQST_ERROR) {
860 netdev_err(ndev, "Invalid transaction ID %llx\n", desc->trans_id);
861 return;
862 }
863
864 pkt_rqst = (struct nvsp_message *)(uintptr_t)cmd_rqst;
865 switch (pkt_rqst->hdr.msg_type) {
866 case NVSP_MSG4_TYPE_SWITCH_DATA_PATH:
867 complete(&net_device->channel_init_wait);
868 break;
869
870 default:
871 netdev_err(ndev, "Unexpected VMBUS completion!!\n");
872 }
873 return;
874 }
875
876 /* Ensure packet is big enough to read header fields */
877 if (msglen < sizeof(struct nvsp_message_header)) {
878 netdev_err(ndev, "nvsp_message length too small: %u\n", msglen);
879 return;
880 }
881
882 nvsp_packet = hv_pkt_data(desc);
883 switch (nvsp_packet->hdr.msg_type) {
884 case NVSP_MSG_TYPE_INIT_COMPLETE:
885 if (msglen < sizeof(struct nvsp_message_header) +
886 sizeof(struct nvsp_message_init_complete)) {
887 netdev_err(ndev, "nvsp_msg length too small: %u\n",
888 msglen);
889 return;
890 }
891 fallthrough;
892
893 case NVSP_MSG1_TYPE_SEND_RECV_BUF_COMPLETE:
894 if (msglen < sizeof(struct nvsp_message_header) +
895 sizeof(struct nvsp_1_message_send_receive_buffer_complete)) {
896 netdev_err(ndev, "nvsp_msg1 length too small: %u\n",
897 msglen);
898 return;
899 }
900 fallthrough;
901
902 case NVSP_MSG1_TYPE_SEND_SEND_BUF_COMPLETE:
903 if (msglen < sizeof(struct nvsp_message_header) +
904 sizeof(struct nvsp_1_message_send_send_buffer_complete)) {
905 netdev_err(ndev, "nvsp_msg1 length too small: %u\n",
906 msglen);
907 return;
908 }
909 fallthrough;
910
911 case NVSP_MSG5_TYPE_SUBCHANNEL:
912 if (msglen < sizeof(struct nvsp_message_header) +
913 sizeof(struct nvsp_5_subchannel_complete)) {
914 netdev_err(ndev, "nvsp_msg5 length too small: %u\n",
915 msglen);
916 return;
917 }
918 /* Copy the response back */
919 memcpy(&net_device->channel_init_pkt, nvsp_packet,
920 sizeof(struct nvsp_message));
921 complete(&net_device->channel_init_wait);
922 break;
923
924 case NVSP_MSG1_TYPE_SEND_RNDIS_PKT_COMPLETE:
925 netvsc_send_tx_complete(ndev, net_device, incoming_channel,
926 desc, budget);
927 break;
928
929 default:
930 netdev_err(ndev,
931 "Unknown send completion type %d received!!\n",
932 nvsp_packet->hdr.msg_type);
933 }
934}
935
936static u32 netvsc_get_next_send_section(struct netvsc_device *net_device)
937{
938 unsigned long *map_addr = net_device->send_section_map;
939 unsigned int i;
940
941 for_each_clear_bit(i, map_addr, net_device->send_section_cnt) {
942 if (sync_test_and_set_bit(i, map_addr) == 0)
943 return i;
944 }
945
946 return NETVSC_INVALID_INDEX;
947}
948
949static void netvsc_copy_to_send_buf(struct netvsc_device *net_device,
950 unsigned int section_index,
951 u32 pend_size,
952 struct hv_netvsc_packet *packet,
953 struct rndis_message *rndis_msg,
954 struct hv_page_buffer *pb,
955 bool xmit_more)
956{
957 char *start = net_device->send_buf;
958 char *dest = start + (section_index * net_device->send_section_size)
959 + pend_size;
960 int i;
961 u32 padding = 0;
962 u32 page_count = packet->cp_partial ? packet->rmsg_pgcnt :
963 packet->page_buf_cnt;
964 u32 remain;
965
966 /* Add padding */
967 remain = packet->total_data_buflen & (net_device->pkt_align - 1);
968 if (xmit_more && remain) {
969 padding = net_device->pkt_align - remain;
970 rndis_msg->msg_len += padding;
971 packet->total_data_buflen += padding;
972 }
973
974 for (i = 0; i < page_count; i++) {
975 char *src = phys_to_virt(pb[i].pfn << HV_HYP_PAGE_SHIFT);
976 u32 offset = pb[i].offset;
977 u32 len = pb[i].len;
978
979 memcpy(dest, (src + offset), len);
980 dest += len;
981 }
982
983 if (padding)
984 memset(dest, 0, padding);
985}
986
987void netvsc_dma_unmap(struct hv_device *hv_dev,
988 struct hv_netvsc_packet *packet)
989{
990 int i;
991
992 if (!hv_is_isolation_supported())
993 return;
994
995 if (!packet->dma_range)
996 return;
997
998 for (i = 0; i < packet->page_buf_cnt; i++)
999 dma_unmap_single(&hv_dev->device, packet->dma_range[i].dma,
1000 packet->dma_range[i].mapping_size,
1001 DMA_TO_DEVICE);
1002
1003 kfree(packet->dma_range);
1004}
1005
1006/* netvsc_dma_map - Map swiotlb bounce buffer with data page of
1007 * packet sent by vmbus_sendpacket_pagebuffer() in the Isolation
1008 * VM.
1009 *
1010 * In isolation VM, netvsc send buffer has been marked visible to
1011 * host and so the data copied to send buffer doesn't need to use
1012 * bounce buffer. The data pages handled by vmbus_sendpacket_pagebuffer()
1013 * may not be copied to send buffer and so these pages need to be
1014 * mapped with swiotlb bounce buffer. netvsc_dma_map() is to do
1015 * that. The pfns in the struct hv_page_buffer need to be converted
1016 * to bounce buffer's pfn. The loop here is necessary because the
1017 * entries in the page buffer array are not necessarily full
1018 * pages of data. Each entry in the array has a separate offset and
1019 * len that may be non-zero, even for entries in the middle of the
1020 * array. And the entries are not physically contiguous. So each
1021 * entry must be individually mapped rather than as a contiguous unit.
1022 * So not use dma_map_sg() here.
1023 */
1024static int netvsc_dma_map(struct hv_device *hv_dev,
1025 struct hv_netvsc_packet *packet,
1026 struct hv_page_buffer *pb)
1027{
1028 u32 page_count = packet->page_buf_cnt;
1029 dma_addr_t dma;
1030 int i;
1031
1032 if (!hv_is_isolation_supported())
1033 return 0;
1034
1035 packet->dma_range = kcalloc(page_count,
1036 sizeof(*packet->dma_range),
1037 GFP_ATOMIC);
1038 if (!packet->dma_range)
1039 return -ENOMEM;
1040
1041 for (i = 0; i < page_count; i++) {
1042 char *src = phys_to_virt((pb[i].pfn << HV_HYP_PAGE_SHIFT)
1043 + pb[i].offset);
1044 u32 len = pb[i].len;
1045
1046 dma = dma_map_single(&hv_dev->device, src, len,
1047 DMA_TO_DEVICE);
1048 if (dma_mapping_error(&hv_dev->device, dma)) {
1049 kfree(packet->dma_range);
1050 return -ENOMEM;
1051 }
1052
1053 /* pb[].offset and pb[].len are not changed during dma mapping
1054 * and so not reassign.
1055 */
1056 packet->dma_range[i].dma = dma;
1057 packet->dma_range[i].mapping_size = len;
1058 pb[i].pfn = dma >> HV_HYP_PAGE_SHIFT;
1059 }
1060
1061 return 0;
1062}
1063
1064static inline int netvsc_send_pkt(
1065 struct hv_device *device,
1066 struct hv_netvsc_packet *packet,
1067 struct netvsc_device *net_device,
1068 struct hv_page_buffer *pb,
1069 struct sk_buff *skb)
1070{
1071 struct nvsp_message nvmsg;
1072 struct nvsp_1_message_send_rndis_packet *rpkt =
1073 &nvmsg.msg.v1_msg.send_rndis_pkt;
1074 struct netvsc_channel * const nvchan =
1075 &net_device->chan_table[packet->q_idx];
1076 struct vmbus_channel *out_channel = nvchan->channel;
1077 struct net_device *ndev = hv_get_drvdata(device);
1078 struct net_device_context *ndev_ctx = netdev_priv(ndev);
1079 struct netdev_queue *txq = netdev_get_tx_queue(ndev, packet->q_idx);
1080 u64 req_id;
1081 int ret;
1082 u32 ring_avail = hv_get_avail_to_write_percent(&out_channel->outbound);
1083
1084 memset(&nvmsg, 0, sizeof(struct nvsp_message));
1085 nvmsg.hdr.msg_type = NVSP_MSG1_TYPE_SEND_RNDIS_PKT;
1086 if (skb)
1087 rpkt->channel_type = 0; /* 0 is RMC_DATA */
1088 else
1089 rpkt->channel_type = 1; /* 1 is RMC_CONTROL */
1090
1091 rpkt->send_buf_section_index = packet->send_buf_index;
1092 if (packet->send_buf_index == NETVSC_INVALID_INDEX)
1093 rpkt->send_buf_section_size = 0;
1094 else
1095 rpkt->send_buf_section_size = packet->total_data_buflen;
1096
1097 req_id = (ulong)skb;
1098
1099 if (out_channel->rescind)
1100 return -ENODEV;
1101
1102 trace_nvsp_send_pkt(ndev, out_channel, rpkt);
1103
1104 packet->dma_range = NULL;
1105 if (packet->page_buf_cnt) {
1106 if (packet->cp_partial)
1107 pb += packet->rmsg_pgcnt;
1108
1109 ret = netvsc_dma_map(ndev_ctx->device_ctx, packet, pb);
1110 if (ret) {
1111 ret = -EAGAIN;
1112 goto exit;
1113 }
1114
1115 ret = vmbus_sendpacket_pagebuffer(out_channel,
1116 pb, packet->page_buf_cnt,
1117 &nvmsg, sizeof(nvmsg),
1118 req_id);
1119
1120 if (ret)
1121 netvsc_dma_unmap(ndev_ctx->device_ctx, packet);
1122 } else {
1123 ret = vmbus_sendpacket(out_channel,
1124 &nvmsg, sizeof(nvmsg),
1125 req_id, VM_PKT_DATA_INBAND,
1126 VMBUS_DATA_PACKET_FLAG_COMPLETION_REQUESTED);
1127 }
1128
1129exit:
1130 if (ret == 0) {
1131 atomic_inc_return(&nvchan->queue_sends);
1132
1133 if (ring_avail < RING_AVAIL_PERCENT_LOWATER) {
1134 netif_tx_stop_queue(txq);
1135 ndev_ctx->eth_stats.stop_queue++;
1136 }
1137 } else if (ret == -EAGAIN) {
1138 netif_tx_stop_queue(txq);
1139 ndev_ctx->eth_stats.stop_queue++;
1140 } else {
1141 netdev_err(ndev,
1142 "Unable to send packet pages %u len %u, ret %d\n",
1143 packet->page_buf_cnt, packet->total_data_buflen,
1144 ret);
1145 }
1146
1147 if (netif_tx_queue_stopped(txq) &&
1148 atomic_read(&nvchan->queue_sends) < 1 &&
1149 !net_device->tx_disable) {
1150 netif_tx_wake_queue(txq);
1151 ndev_ctx->eth_stats.wake_queue++;
1152 if (ret == -EAGAIN)
1153 ret = -ENOSPC;
1154 }
1155
1156 return ret;
1157}
1158
1159/* Move packet out of multi send data (msd), and clear msd */
1160static inline void move_pkt_msd(struct hv_netvsc_packet **msd_send,
1161 struct sk_buff **msd_skb,
1162 struct multi_send_data *msdp)
1163{
1164 *msd_skb = msdp->skb;
1165 *msd_send = msdp->pkt;
1166 msdp->skb = NULL;
1167 msdp->pkt = NULL;
1168 msdp->count = 0;
1169}
1170
1171/* RCU already held by caller */
1172/* Batching/bouncing logic is designed to attempt to optimize
1173 * performance.
1174 *
1175 * For small, non-LSO packets we copy the packet to a send buffer
1176 * which is pre-registered with the Hyper-V side. This enables the
1177 * hypervisor to avoid remapping the aperture to access the packet
1178 * descriptor and data.
1179 *
1180 * If we already started using a buffer and the netdev is transmitting
1181 * a burst of packets, keep on copying into the buffer until it is
1182 * full or we are done collecting a burst. If there is an existing
1183 * buffer with space for the RNDIS descriptor but not the packet, copy
1184 * the RNDIS descriptor to the buffer, keeping the packet in place.
1185 *
1186 * If we do batching and send more than one packet using a single
1187 * NetVSC message, free the SKBs of the packets copied, except for the
1188 * last packet. This is done to streamline the handling of the case
1189 * where the last packet only had the RNDIS descriptor copied to the
1190 * send buffer, with the data pointers included in the NetVSC message.
1191 */
1192int netvsc_send(struct net_device *ndev,
1193 struct hv_netvsc_packet *packet,
1194 struct rndis_message *rndis_msg,
1195 struct hv_page_buffer *pb,
1196 struct sk_buff *skb,
1197 bool xdp_tx)
1198{
1199 struct net_device_context *ndev_ctx = netdev_priv(ndev);
1200 struct netvsc_device *net_device
1201 = rcu_dereference_bh(ndev_ctx->nvdev);
1202 struct hv_device *device = ndev_ctx->device_ctx;
1203 int ret = 0;
1204 struct netvsc_channel *nvchan;
1205 u32 pktlen = packet->total_data_buflen, msd_len = 0;
1206 unsigned int section_index = NETVSC_INVALID_INDEX;
1207 struct multi_send_data *msdp;
1208 struct hv_netvsc_packet *msd_send = NULL, *cur_send = NULL;
1209 struct sk_buff *msd_skb = NULL;
1210 bool try_batch, xmit_more;
1211
1212 /* If device is rescinded, return error and packet will get dropped. */
1213 if (unlikely(!net_device || net_device->destroy))
1214 return -ENODEV;
1215
1216 nvchan = &net_device->chan_table[packet->q_idx];
1217 packet->send_buf_index = NETVSC_INVALID_INDEX;
1218 packet->cp_partial = false;
1219
1220 /* Send a control message or XDP packet directly without accessing
1221 * msd (Multi-Send Data) field which may be changed during data packet
1222 * processing.
1223 */
1224 if (!skb || xdp_tx)
1225 return netvsc_send_pkt(device, packet, net_device, pb, skb);
1226
1227 /* batch packets in send buffer if possible */
1228 msdp = &nvchan->msd;
1229 if (msdp->pkt)
1230 msd_len = msdp->pkt->total_data_buflen;
1231
1232 try_batch = msd_len > 0 && msdp->count < net_device->max_pkt;
1233 if (try_batch && msd_len + pktlen + net_device->pkt_align <
1234 net_device->send_section_size) {
1235 section_index = msdp->pkt->send_buf_index;
1236
1237 } else if (try_batch && msd_len + packet->rmsg_size <
1238 net_device->send_section_size) {
1239 section_index = msdp->pkt->send_buf_index;
1240 packet->cp_partial = true;
1241
1242 } else if (pktlen + net_device->pkt_align <
1243 net_device->send_section_size) {
1244 section_index = netvsc_get_next_send_section(net_device);
1245 if (unlikely(section_index == NETVSC_INVALID_INDEX)) {
1246 ++ndev_ctx->eth_stats.tx_send_full;
1247 } else {
1248 move_pkt_msd(&msd_send, &msd_skb, msdp);
1249 msd_len = 0;
1250 }
1251 }
1252
1253 /* Keep aggregating only if stack says more data is coming
1254 * and not doing mixed modes send and not flow blocked
1255 */
1256 xmit_more = netdev_xmit_more() &&
1257 !packet->cp_partial &&
1258 !netif_xmit_stopped(netdev_get_tx_queue(ndev, packet->q_idx));
1259
1260 if (section_index != NETVSC_INVALID_INDEX) {
1261 netvsc_copy_to_send_buf(net_device,
1262 section_index, msd_len,
1263 packet, rndis_msg, pb, xmit_more);
1264
1265 packet->send_buf_index = section_index;
1266
1267 if (packet->cp_partial) {
1268 packet->page_buf_cnt -= packet->rmsg_pgcnt;
1269 packet->total_data_buflen = msd_len + packet->rmsg_size;
1270 } else {
1271 packet->page_buf_cnt = 0;
1272 packet->total_data_buflen += msd_len;
1273 }
1274
1275 if (msdp->pkt) {
1276 packet->total_packets += msdp->pkt->total_packets;
1277 packet->total_bytes += msdp->pkt->total_bytes;
1278 }
1279
1280 if (msdp->skb)
1281 dev_consume_skb_any(msdp->skb);
1282
1283 if (xmit_more) {
1284 msdp->skb = skb;
1285 msdp->pkt = packet;
1286 msdp->count++;
1287 } else {
1288 cur_send = packet;
1289 msdp->skb = NULL;
1290 msdp->pkt = NULL;
1291 msdp->count = 0;
1292 }
1293 } else {
1294 move_pkt_msd(&msd_send, &msd_skb, msdp);
1295 cur_send = packet;
1296 }
1297
1298 if (msd_send) {
1299 int m_ret = netvsc_send_pkt(device, msd_send, net_device,
1300 NULL, msd_skb);
1301
1302 if (m_ret != 0) {
1303 netvsc_free_send_slot(net_device,
1304 msd_send->send_buf_index);
1305 dev_kfree_skb_any(msd_skb);
1306 }
1307 }
1308
1309 if (cur_send)
1310 ret = netvsc_send_pkt(device, cur_send, net_device, pb, skb);
1311
1312 if (ret != 0 && section_index != NETVSC_INVALID_INDEX)
1313 netvsc_free_send_slot(net_device, section_index);
1314
1315 return ret;
1316}
1317
1318/* Send pending recv completions */
1319static int send_recv_completions(struct net_device *ndev,
1320 struct netvsc_device *nvdev,
1321 struct netvsc_channel *nvchan)
1322{
1323 struct multi_recv_comp *mrc = &nvchan->mrc;
1324 struct recv_comp_msg {
1325 struct nvsp_message_header hdr;
1326 u32 status;
1327 } __packed;
1328 struct recv_comp_msg msg = {
1329 .hdr.msg_type = NVSP_MSG1_TYPE_SEND_RNDIS_PKT_COMPLETE,
1330 };
1331 int ret;
1332
1333 while (mrc->first != mrc->next) {
1334 const struct recv_comp_data *rcd
1335 = mrc->slots + mrc->first;
1336
1337 msg.status = rcd->status;
1338 ret = vmbus_sendpacket(nvchan->channel, &msg, sizeof(msg),
1339 rcd->tid, VM_PKT_COMP, 0);
1340 if (unlikely(ret)) {
1341 struct net_device_context *ndev_ctx = netdev_priv(ndev);
1342
1343 ++ndev_ctx->eth_stats.rx_comp_busy;
1344 return ret;
1345 }
1346
1347 if (++mrc->first == nvdev->recv_completion_cnt)
1348 mrc->first = 0;
1349 }
1350
1351 /* receive completion ring has been emptied */
1352 if (unlikely(nvdev->destroy))
1353 wake_up(&nvdev->wait_drain);
1354
1355 return 0;
1356}
1357
1358/* Count how many receive completions are outstanding */
1359static void recv_comp_slot_avail(const struct netvsc_device *nvdev,
1360 const struct multi_recv_comp *mrc,
1361 u32 *filled, u32 *avail)
1362{
1363 u32 count = nvdev->recv_completion_cnt;
1364
1365 if (mrc->next >= mrc->first)
1366 *filled = mrc->next - mrc->first;
1367 else
1368 *filled = (count - mrc->first) + mrc->next;
1369
1370 *avail = count - *filled - 1;
1371}
1372
1373/* Add receive complete to ring to send to host. */
1374static void enq_receive_complete(struct net_device *ndev,
1375 struct netvsc_device *nvdev, u16 q_idx,
1376 u64 tid, u32 status)
1377{
1378 struct netvsc_channel *nvchan = &nvdev->chan_table[q_idx];
1379 struct multi_recv_comp *mrc = &nvchan->mrc;
1380 struct recv_comp_data *rcd;
1381 u32 filled, avail;
1382
1383 recv_comp_slot_avail(nvdev, mrc, &filled, &avail);
1384
1385 if (unlikely(filled > NAPI_POLL_WEIGHT)) {
1386 send_recv_completions(ndev, nvdev, nvchan);
1387 recv_comp_slot_avail(nvdev, mrc, &filled, &avail);
1388 }
1389
1390 if (unlikely(!avail)) {
1391 netdev_err(ndev, "Recv_comp full buf q:%hd, tid:%llx\n",
1392 q_idx, tid);
1393 return;
1394 }
1395
1396 rcd = mrc->slots + mrc->next;
1397 rcd->tid = tid;
1398 rcd->status = status;
1399
1400 if (++mrc->next == nvdev->recv_completion_cnt)
1401 mrc->next = 0;
1402}
1403
1404static int netvsc_receive(struct net_device *ndev,
1405 struct netvsc_device *net_device,
1406 struct netvsc_channel *nvchan,
1407 const struct vmpacket_descriptor *desc)
1408{
1409 struct net_device_context *net_device_ctx = netdev_priv(ndev);
1410 struct vmbus_channel *channel = nvchan->channel;
1411 const struct vmtransfer_page_packet_header *vmxferpage_packet
1412 = container_of(desc, const struct vmtransfer_page_packet_header, d);
1413 const struct nvsp_message *nvsp = hv_pkt_data(desc);
1414 u32 msglen = hv_pkt_datalen(desc);
1415 u16 q_idx = channel->offermsg.offer.sub_channel_index;
1416 char *recv_buf = net_device->recv_buf;
1417 u32 status = NVSP_STAT_SUCCESS;
1418 int i;
1419 int count = 0;
1420
1421 /* Ensure packet is big enough to read header fields */
1422 if (msglen < sizeof(struct nvsp_message_header)) {
1423 netif_err(net_device_ctx, rx_err, ndev,
1424 "invalid nvsp header, length too small: %u\n",
1425 msglen);
1426 return 0;
1427 }
1428
1429 /* Make sure this is a valid nvsp packet */
1430 if (unlikely(nvsp->hdr.msg_type != NVSP_MSG1_TYPE_SEND_RNDIS_PKT)) {
1431 netif_err(net_device_ctx, rx_err, ndev,
1432 "Unknown nvsp packet type received %u\n",
1433 nvsp->hdr.msg_type);
1434 return 0;
1435 }
1436
1437 /* Validate xfer page pkt header */
1438 if ((desc->offset8 << 3) < sizeof(struct vmtransfer_page_packet_header)) {
1439 netif_err(net_device_ctx, rx_err, ndev,
1440 "Invalid xfer page pkt, offset too small: %u\n",
1441 desc->offset8 << 3);
1442 return 0;
1443 }
1444
1445 if (unlikely(vmxferpage_packet->xfer_pageset_id != NETVSC_RECEIVE_BUFFER_ID)) {
1446 netif_err(net_device_ctx, rx_err, ndev,
1447 "Invalid xfer page set id - expecting %x got %x\n",
1448 NETVSC_RECEIVE_BUFFER_ID,
1449 vmxferpage_packet->xfer_pageset_id);
1450 return 0;
1451 }
1452
1453 count = vmxferpage_packet->range_cnt;
1454
1455 /* Check count for a valid value */
1456 if (NETVSC_XFER_HEADER_SIZE(count) > desc->offset8 << 3) {
1457 netif_err(net_device_ctx, rx_err, ndev,
1458 "Range count is not valid: %d\n",
1459 count);
1460 return 0;
1461 }
1462
1463 /* Each range represents 1 RNDIS pkt that contains 1 ethernet frame */
1464 for (i = 0; i < count; i++) {
1465 u32 offset = vmxferpage_packet->ranges[i].byte_offset;
1466 u32 buflen = vmxferpage_packet->ranges[i].byte_count;
1467 void *data;
1468 int ret;
1469
1470 if (unlikely(offset > net_device->recv_buf_size ||
1471 buflen > net_device->recv_buf_size - offset)) {
1472 nvchan->rsc.cnt = 0;
1473 status = NVSP_STAT_FAIL;
1474 netif_err(net_device_ctx, rx_err, ndev,
1475 "Packet offset:%u + len:%u too big\n",
1476 offset, buflen);
1477
1478 continue;
1479 }
1480
1481 /* We're going to copy (sections of) the packet into nvchan->recv_buf;
1482 * make sure that nvchan->recv_buf is large enough to hold the packet.
1483 */
1484 if (unlikely(buflen > net_device->recv_section_size)) {
1485 nvchan->rsc.cnt = 0;
1486 status = NVSP_STAT_FAIL;
1487 netif_err(net_device_ctx, rx_err, ndev,
1488 "Packet too big: buflen=%u recv_section_size=%u\n",
1489 buflen, net_device->recv_section_size);
1490
1491 continue;
1492 }
1493
1494 data = recv_buf + offset;
1495
1496 nvchan->rsc.is_last = (i == count - 1);
1497
1498 trace_rndis_recv(ndev, q_idx, data);
1499
1500 /* Pass it to the upper layer */
1501 ret = rndis_filter_receive(ndev, net_device,
1502 nvchan, data, buflen);
1503
1504 if (unlikely(ret != NVSP_STAT_SUCCESS)) {
1505 /* Drop incomplete packet */
1506 nvchan->rsc.cnt = 0;
1507 status = NVSP_STAT_FAIL;
1508 }
1509 }
1510
1511 enq_receive_complete(ndev, net_device, q_idx,
1512 vmxferpage_packet->d.trans_id, status);
1513
1514 return count;
1515}
1516
1517static void netvsc_send_table(struct net_device *ndev,
1518 struct netvsc_device *nvscdev,
1519 const struct nvsp_message *nvmsg,
1520 u32 msglen)
1521{
1522 struct net_device_context *net_device_ctx = netdev_priv(ndev);
1523 u32 count, offset, *tab;
1524 int i;
1525
1526 /* Ensure packet is big enough to read send_table fields */
1527 if (msglen < sizeof(struct nvsp_message_header) +
1528 sizeof(struct nvsp_5_send_indirect_table)) {
1529 netdev_err(ndev, "nvsp_v5_msg length too small: %u\n", msglen);
1530 return;
1531 }
1532
1533 count = nvmsg->msg.v5_msg.send_table.count;
1534 offset = nvmsg->msg.v5_msg.send_table.offset;
1535
1536 if (count != VRSS_SEND_TAB_SIZE) {
1537 netdev_err(ndev, "Received wrong send-table size:%u\n", count);
1538 return;
1539 }
1540
1541 /* If negotiated version <= NVSP_PROTOCOL_VERSION_6, the offset may be
1542 * wrong due to a host bug. So fix the offset here.
1543 */
1544 if (nvscdev->nvsp_version <= NVSP_PROTOCOL_VERSION_6 &&
1545 msglen >= sizeof(struct nvsp_message_header) +
1546 sizeof(union nvsp_6_message_uber) + count * sizeof(u32))
1547 offset = sizeof(struct nvsp_message_header) +
1548 sizeof(union nvsp_6_message_uber);
1549
1550 /* Boundary check for all versions */
1551 if (msglen < count * sizeof(u32) || offset > msglen - count * sizeof(u32)) {
1552 netdev_err(ndev, "Received send-table offset too big:%u\n",
1553 offset);
1554 return;
1555 }
1556
1557 tab = (void *)nvmsg + offset;
1558
1559 for (i = 0; i < count; i++)
1560 net_device_ctx->tx_table[i] = tab[i];
1561}
1562
1563static void netvsc_send_vf(struct net_device *ndev,
1564 const struct nvsp_message *nvmsg,
1565 u32 msglen)
1566{
1567 struct net_device_context *net_device_ctx = netdev_priv(ndev);
1568
1569 /* Ensure packet is big enough to read its fields */
1570 if (msglen < sizeof(struct nvsp_message_header) +
1571 sizeof(struct nvsp_4_send_vf_association)) {
1572 netdev_err(ndev, "nvsp_v4_msg length too small: %u\n", msglen);
1573 return;
1574 }
1575
1576 net_device_ctx->vf_alloc = nvmsg->msg.v4_msg.vf_assoc.allocated;
1577 net_device_ctx->vf_serial = nvmsg->msg.v4_msg.vf_assoc.serial;
1578
1579 if (net_device_ctx->vf_alloc)
1580 complete(&net_device_ctx->vf_add);
1581
1582 netdev_info(ndev, "VF slot %u %s\n",
1583 net_device_ctx->vf_serial,
1584 net_device_ctx->vf_alloc ? "added" : "removed");
1585}
1586
1587static void netvsc_receive_inband(struct net_device *ndev,
1588 struct netvsc_device *nvscdev,
1589 const struct vmpacket_descriptor *desc)
1590{
1591 const struct nvsp_message *nvmsg = hv_pkt_data(desc);
1592 u32 msglen = hv_pkt_datalen(desc);
1593
1594 /* Ensure packet is big enough to read header fields */
1595 if (msglen < sizeof(struct nvsp_message_header)) {
1596 netdev_err(ndev, "inband nvsp_message length too small: %u\n", msglen);
1597 return;
1598 }
1599
1600 switch (nvmsg->hdr.msg_type) {
1601 case NVSP_MSG5_TYPE_SEND_INDIRECTION_TABLE:
1602 netvsc_send_table(ndev, nvscdev, nvmsg, msglen);
1603 break;
1604
1605 case NVSP_MSG4_TYPE_SEND_VF_ASSOCIATION:
1606 if (hv_is_isolation_supported())
1607 netdev_err(ndev, "Ignore VF_ASSOCIATION msg from the host supporting isolation\n");
1608 else
1609 netvsc_send_vf(ndev, nvmsg, msglen);
1610 break;
1611 }
1612}
1613
1614static int netvsc_process_raw_pkt(struct hv_device *device,
1615 struct netvsc_channel *nvchan,
1616 struct netvsc_device *net_device,
1617 struct net_device *ndev,
1618 const struct vmpacket_descriptor *desc,
1619 int budget)
1620{
1621 struct vmbus_channel *channel = nvchan->channel;
1622 const struct nvsp_message *nvmsg = hv_pkt_data(desc);
1623
1624 trace_nvsp_recv(ndev, channel, nvmsg);
1625
1626 switch (desc->type) {
1627 case VM_PKT_COMP:
1628 netvsc_send_completion(ndev, net_device, channel, desc, budget);
1629 break;
1630
1631 case VM_PKT_DATA_USING_XFER_PAGES:
1632 return netvsc_receive(ndev, net_device, nvchan, desc);
1633
1634 case VM_PKT_DATA_INBAND:
1635 netvsc_receive_inband(ndev, net_device, desc);
1636 break;
1637
1638 default:
1639 netdev_err(ndev, "unhandled packet type %d, tid %llx\n",
1640 desc->type, desc->trans_id);
1641 break;
1642 }
1643
1644 return 0;
1645}
1646
1647static struct hv_device *netvsc_channel_to_device(struct vmbus_channel *channel)
1648{
1649 struct vmbus_channel *primary = channel->primary_channel;
1650
1651 return primary ? primary->device_obj : channel->device_obj;
1652}
1653
1654/* Network processing softirq
1655 * Process data in incoming ring buffer from host
1656 * Stops when ring is empty or budget is met or exceeded.
1657 */
1658int netvsc_poll(struct napi_struct *napi, int budget)
1659{
1660 struct netvsc_channel *nvchan
1661 = container_of(napi, struct netvsc_channel, napi);
1662 struct netvsc_device *net_device = nvchan->net_device;
1663 struct vmbus_channel *channel = nvchan->channel;
1664 struct hv_device *device = netvsc_channel_to_device(channel);
1665 struct net_device *ndev = hv_get_drvdata(device);
1666 int work_done = 0;
1667 int ret;
1668
1669 /* If starting a new interval */
1670 if (!nvchan->desc)
1671 nvchan->desc = hv_pkt_iter_first(channel);
1672
1673 nvchan->xdp_flush = false;
1674
1675 while (nvchan->desc && work_done < budget) {
1676 work_done += netvsc_process_raw_pkt(device, nvchan, net_device,
1677 ndev, nvchan->desc, budget);
1678 nvchan->desc = hv_pkt_iter_next(channel, nvchan->desc);
1679 }
1680
1681 if (nvchan->xdp_flush)
1682 xdp_do_flush();
1683
1684 /* Send any pending receive completions */
1685 ret = send_recv_completions(ndev, net_device, nvchan);
1686
1687 /* If it did not exhaust NAPI budget this time
1688 * and not doing busy poll
1689 * then re-enable host interrupts
1690 * and reschedule if ring is not empty
1691 * or sending receive completion failed.
1692 */
1693 if (work_done < budget &&
1694 napi_complete_done(napi, work_done) &&
1695 (ret || hv_end_read(&channel->inbound)) &&
1696 napi_schedule_prep(napi)) {
1697 hv_begin_read(&channel->inbound);
1698 __napi_schedule(napi);
1699 }
1700
1701 /* Driver may overshoot since multiple packets per descriptor */
1702 return min(work_done, budget);
1703}
1704
1705/* Call back when data is available in host ring buffer.
1706 * Processing is deferred until network softirq (NAPI)
1707 */
1708void netvsc_channel_cb(void *context)
1709{
1710 struct netvsc_channel *nvchan = context;
1711 struct vmbus_channel *channel = nvchan->channel;
1712 struct hv_ring_buffer_info *rbi = &channel->inbound;
1713
1714 /* preload first vmpacket descriptor */
1715 prefetch(hv_get_ring_buffer(rbi) + rbi->priv_read_index);
1716
1717 if (napi_schedule_prep(&nvchan->napi)) {
1718 /* disable interrupts from host */
1719 hv_begin_read(rbi);
1720
1721 __napi_schedule_irqoff(&nvchan->napi);
1722 }
1723}
1724
1725/*
1726 * netvsc_device_add - Callback when the device belonging to this
1727 * driver is added
1728 */
1729struct netvsc_device *netvsc_device_add(struct hv_device *device,
1730 const struct netvsc_device_info *device_info)
1731{
1732 int i, ret = 0;
1733 struct netvsc_device *net_device;
1734 struct net_device *ndev = hv_get_drvdata(device);
1735 struct net_device_context *net_device_ctx = netdev_priv(ndev);
1736
1737 net_device = alloc_net_device();
1738 if (!net_device)
1739 return ERR_PTR(-ENOMEM);
1740
1741 for (i = 0; i < VRSS_SEND_TAB_SIZE; i++)
1742 net_device_ctx->tx_table[i] = 0;
1743
1744 /* Because the device uses NAPI, all the interrupt batching and
1745 * control is done via Net softirq, not the channel handling
1746 */
1747 set_channel_read_mode(device->channel, HV_CALL_ISR);
1748
1749 /* If we're reopening the device we may have multiple queues, fill the
1750 * chn_table with the default channel to use it before subchannels are
1751 * opened.
1752 * Initialize the channel state before we open;
1753 * we can be interrupted as soon as we open the channel.
1754 */
1755
1756 for (i = 0; i < VRSS_CHANNEL_MAX; i++) {
1757 struct netvsc_channel *nvchan = &net_device->chan_table[i];
1758
1759 nvchan->channel = device->channel;
1760 nvchan->net_device = net_device;
1761 u64_stats_init(&nvchan->tx_stats.syncp);
1762 u64_stats_init(&nvchan->rx_stats.syncp);
1763
1764 ret = xdp_rxq_info_reg(&nvchan->xdp_rxq, ndev, i, 0);
1765
1766 if (ret) {
1767 netdev_err(ndev, "xdp_rxq_info_reg fail: %d\n", ret);
1768 goto cleanup2;
1769 }
1770
1771 ret = xdp_rxq_info_reg_mem_model(&nvchan->xdp_rxq,
1772 MEM_TYPE_PAGE_SHARED, NULL);
1773
1774 if (ret) {
1775 netdev_err(ndev, "xdp reg_mem_model fail: %d\n", ret);
1776 goto cleanup2;
1777 }
1778 }
1779
1780 /* Enable NAPI handler before init callbacks */
1781 netif_napi_add(ndev, &net_device->chan_table[0].napi, netvsc_poll);
1782
1783 /* Open the channel */
1784 device->channel->next_request_id_callback = vmbus_next_request_id;
1785 device->channel->request_addr_callback = vmbus_request_addr;
1786 device->channel->rqstor_size = netvsc_rqstor_size(netvsc_ring_bytes);
1787 device->channel->max_pkt_size = NETVSC_MAX_PKT_SIZE;
1788
1789 ret = vmbus_open(device->channel, netvsc_ring_bytes,
1790 netvsc_ring_bytes, NULL, 0,
1791 netvsc_channel_cb, net_device->chan_table);
1792
1793 if (ret != 0) {
1794 netdev_err(ndev, "unable to open channel: %d\n", ret);
1795 goto cleanup;
1796 }
1797
1798 /* Channel is opened */
1799 netdev_dbg(ndev, "hv_netvsc channel opened successfully\n");
1800
1801 napi_enable(&net_device->chan_table[0].napi);
1802
1803 /* Connect with the NetVsp */
1804 ret = netvsc_connect_vsp(device, net_device, device_info);
1805 if (ret != 0) {
1806 netdev_err(ndev,
1807 "unable to connect to NetVSP - %d\n", ret);
1808 goto close;
1809 }
1810
1811 /* Writing nvdev pointer unlocks netvsc_send(), make sure chn_table is
1812 * populated.
1813 */
1814 rcu_assign_pointer(net_device_ctx->nvdev, net_device);
1815
1816 return net_device;
1817
1818close:
1819 RCU_INIT_POINTER(net_device_ctx->nvdev, NULL);
1820 napi_disable(&net_device->chan_table[0].napi);
1821
1822 /* Now, we can close the channel safely */
1823 vmbus_close(device->channel);
1824
1825cleanup:
1826 netif_napi_del(&net_device->chan_table[0].napi);
1827
1828cleanup2:
1829 if (net_device->recv_original_buf)
1830 hv_unmap_memory(net_device->recv_buf);
1831
1832 if (net_device->send_original_buf)
1833 hv_unmap_memory(net_device->send_buf);
1834
1835 free_netvsc_device(&net_device->rcu);
1836
1837 return ERR_PTR(ret);
1838}
1/*
2 * Copyright (c) 2009, Microsoft Corporation.
3 *
4 * This program is free software; you can redistribute it and/or modify it
5 * under the terms and conditions of the GNU General Public License,
6 * version 2, as published by the Free Software Foundation.
7 *
8 * This program is distributed in the hope it will be useful, but WITHOUT
9 * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
10 * FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License for
11 * more details.
12 *
13 * You should have received a copy of the GNU General Public License along with
14 * this program; if not, write to the Free Software Foundation, Inc., 59 Temple
15 * Place - Suite 330, Boston, MA 02111-1307 USA.
16 *
17 * Authors:
18 * Haiyang Zhang <haiyangz@microsoft.com>
19 * Hank Janssen <hjanssen@microsoft.com>
20 */
21#define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
22
23#include <linux/kernel.h>
24#include <linux/sched.h>
25#include <linux/wait.h>
26#include <linux/mm.h>
27#include <linux/delay.h>
28#include <linux/io.h>
29#include <linux/slab.h>
30#include <linux/netdevice.h>
31#include <linux/if_ether.h>
32
33#include "hyperv_net.h"
34
35
36static struct netvsc_device *alloc_net_device(struct hv_device *device)
37{
38 struct netvsc_device *net_device;
39 struct net_device *ndev = hv_get_drvdata(device);
40
41 net_device = kzalloc(sizeof(struct netvsc_device), GFP_KERNEL);
42 if (!net_device)
43 return NULL;
44
45 init_waitqueue_head(&net_device->wait_drain);
46 net_device->start_remove = false;
47 net_device->destroy = false;
48 net_device->dev = device;
49 net_device->ndev = ndev;
50
51 hv_set_drvdata(device, net_device);
52 return net_device;
53}
54
55static struct netvsc_device *get_outbound_net_device(struct hv_device *device)
56{
57 struct netvsc_device *net_device;
58
59 net_device = hv_get_drvdata(device);
60 if (net_device && net_device->destroy)
61 net_device = NULL;
62
63 return net_device;
64}
65
66static struct netvsc_device *get_inbound_net_device(struct hv_device *device)
67{
68 struct netvsc_device *net_device;
69
70 net_device = hv_get_drvdata(device);
71
72 if (!net_device)
73 goto get_in_err;
74
75 if (net_device->destroy &&
76 atomic_read(&net_device->num_outstanding_sends) == 0)
77 net_device = NULL;
78
79get_in_err:
80 return net_device;
81}
82
83
84static int netvsc_destroy_recv_buf(struct netvsc_device *net_device)
85{
86 struct nvsp_message *revoke_packet;
87 int ret = 0;
88 struct net_device *ndev = net_device->ndev;
89
90 /*
91 * If we got a section count, it means we received a
92 * SendReceiveBufferComplete msg (ie sent
93 * NvspMessage1TypeSendReceiveBuffer msg) therefore, we need
94 * to send a revoke msg here
95 */
96 if (net_device->recv_section_cnt) {
97 /* Send the revoke receive buffer */
98 revoke_packet = &net_device->revoke_packet;
99 memset(revoke_packet, 0, sizeof(struct nvsp_message));
100
101 revoke_packet->hdr.msg_type =
102 NVSP_MSG1_TYPE_REVOKE_RECV_BUF;
103 revoke_packet->msg.v1_msg.
104 revoke_recv_buf.id = NETVSC_RECEIVE_BUFFER_ID;
105
106 ret = vmbus_sendpacket(net_device->dev->channel,
107 revoke_packet,
108 sizeof(struct nvsp_message),
109 (unsigned long)revoke_packet,
110 VM_PKT_DATA_INBAND, 0);
111 /*
112 * If we failed here, we might as well return and
113 * have a leak rather than continue and a bugchk
114 */
115 if (ret != 0) {
116 netdev_err(ndev, "unable to send "
117 "revoke receive buffer to netvsp\n");
118 return ret;
119 }
120 }
121
122 /* Teardown the gpadl on the vsp end */
123 if (net_device->recv_buf_gpadl_handle) {
124 ret = vmbus_teardown_gpadl(net_device->dev->channel,
125 net_device->recv_buf_gpadl_handle);
126
127 /* If we failed here, we might as well return and have a leak
128 * rather than continue and a bugchk
129 */
130 if (ret != 0) {
131 netdev_err(ndev,
132 "unable to teardown receive buffer's gpadl\n");
133 return ret;
134 }
135 net_device->recv_buf_gpadl_handle = 0;
136 }
137
138 if (net_device->recv_buf) {
139 /* Free up the receive buffer */
140 free_pages((unsigned long)net_device->recv_buf,
141 get_order(net_device->recv_buf_size));
142 net_device->recv_buf = NULL;
143 }
144
145 if (net_device->recv_section) {
146 net_device->recv_section_cnt = 0;
147 kfree(net_device->recv_section);
148 net_device->recv_section = NULL;
149 }
150
151 return ret;
152}
153
154static int netvsc_init_recv_buf(struct hv_device *device)
155{
156 int ret = 0;
157 int t;
158 struct netvsc_device *net_device;
159 struct nvsp_message *init_packet;
160 struct net_device *ndev;
161
162 net_device = get_outbound_net_device(device);
163 if (!net_device)
164 return -ENODEV;
165 ndev = net_device->ndev;
166
167 net_device->recv_buf =
168 (void *)__get_free_pages(GFP_KERNEL|__GFP_ZERO,
169 get_order(net_device->recv_buf_size));
170 if (!net_device->recv_buf) {
171 netdev_err(ndev, "unable to allocate receive "
172 "buffer of size %d\n", net_device->recv_buf_size);
173 ret = -ENOMEM;
174 goto cleanup;
175 }
176
177 /*
178 * Establish the gpadl handle for this buffer on this
179 * channel. Note: This call uses the vmbus connection rather
180 * than the channel to establish the gpadl handle.
181 */
182 ret = vmbus_establish_gpadl(device->channel, net_device->recv_buf,
183 net_device->recv_buf_size,
184 &net_device->recv_buf_gpadl_handle);
185 if (ret != 0) {
186 netdev_err(ndev,
187 "unable to establish receive buffer's gpadl\n");
188 goto cleanup;
189 }
190
191
192 /* Notify the NetVsp of the gpadl handle */
193 init_packet = &net_device->channel_init_pkt;
194
195 memset(init_packet, 0, sizeof(struct nvsp_message));
196
197 init_packet->hdr.msg_type = NVSP_MSG1_TYPE_SEND_RECV_BUF;
198 init_packet->msg.v1_msg.send_recv_buf.
199 gpadl_handle = net_device->recv_buf_gpadl_handle;
200 init_packet->msg.v1_msg.
201 send_recv_buf.id = NETVSC_RECEIVE_BUFFER_ID;
202
203 /* Send the gpadl notification request */
204 ret = vmbus_sendpacket(device->channel, init_packet,
205 sizeof(struct nvsp_message),
206 (unsigned long)init_packet,
207 VM_PKT_DATA_INBAND,
208 VMBUS_DATA_PACKET_FLAG_COMPLETION_REQUESTED);
209 if (ret != 0) {
210 netdev_err(ndev,
211 "unable to send receive buffer's gpadl to netvsp\n");
212 goto cleanup;
213 }
214
215 t = wait_for_completion_timeout(&net_device->channel_init_wait, 5*HZ);
216 BUG_ON(t == 0);
217
218
219 /* Check the response */
220 if (init_packet->msg.v1_msg.
221 send_recv_buf_complete.status != NVSP_STAT_SUCCESS) {
222 netdev_err(ndev, "Unable to complete receive buffer "
223 "initialization with NetVsp - status %d\n",
224 init_packet->msg.v1_msg.
225 send_recv_buf_complete.status);
226 ret = -EINVAL;
227 goto cleanup;
228 }
229
230 /* Parse the response */
231
232 net_device->recv_section_cnt = init_packet->msg.
233 v1_msg.send_recv_buf_complete.num_sections;
234
235 net_device->recv_section = kmemdup(
236 init_packet->msg.v1_msg.send_recv_buf_complete.sections,
237 net_device->recv_section_cnt *
238 sizeof(struct nvsp_1_receive_buffer_section),
239 GFP_KERNEL);
240 if (net_device->recv_section == NULL) {
241 ret = -EINVAL;
242 goto cleanup;
243 }
244
245 /*
246 * For 1st release, there should only be 1 section that represents the
247 * entire receive buffer
248 */
249 if (net_device->recv_section_cnt != 1 ||
250 net_device->recv_section->offset != 0) {
251 ret = -EINVAL;
252 goto cleanup;
253 }
254
255 goto exit;
256
257cleanup:
258 netvsc_destroy_recv_buf(net_device);
259
260exit:
261 return ret;
262}
263
264
265/* Negotiate NVSP protocol version */
266static int negotiate_nvsp_ver(struct hv_device *device,
267 struct netvsc_device *net_device,
268 struct nvsp_message *init_packet,
269 u32 nvsp_ver)
270{
271 int ret, t;
272
273 memset(init_packet, 0, sizeof(struct nvsp_message));
274 init_packet->hdr.msg_type = NVSP_MSG_TYPE_INIT;
275 init_packet->msg.init_msg.init.min_protocol_ver = nvsp_ver;
276 init_packet->msg.init_msg.init.max_protocol_ver = nvsp_ver;
277
278 /* Send the init request */
279 ret = vmbus_sendpacket(device->channel, init_packet,
280 sizeof(struct nvsp_message),
281 (unsigned long)init_packet,
282 VM_PKT_DATA_INBAND,
283 VMBUS_DATA_PACKET_FLAG_COMPLETION_REQUESTED);
284
285 if (ret != 0)
286 return ret;
287
288 t = wait_for_completion_timeout(&net_device->channel_init_wait, 5*HZ);
289
290 if (t == 0)
291 return -ETIMEDOUT;
292
293 if (init_packet->msg.init_msg.init_complete.status !=
294 NVSP_STAT_SUCCESS)
295 return -EINVAL;
296
297 if (nvsp_ver != NVSP_PROTOCOL_VERSION_2)
298 return 0;
299
300 /* NVSPv2 only: Send NDIS config */
301 memset(init_packet, 0, sizeof(struct nvsp_message));
302 init_packet->hdr.msg_type = NVSP_MSG2_TYPE_SEND_NDIS_CONFIG;
303 init_packet->msg.v2_msg.send_ndis_config.mtu = net_device->ndev->mtu;
304 init_packet->msg.v2_msg.send_ndis_config.capability.ieee8021q = 1;
305
306 ret = vmbus_sendpacket(device->channel, init_packet,
307 sizeof(struct nvsp_message),
308 (unsigned long)init_packet,
309 VM_PKT_DATA_INBAND, 0);
310
311 return ret;
312}
313
314static int netvsc_connect_vsp(struct hv_device *device)
315{
316 int ret;
317 struct netvsc_device *net_device;
318 struct nvsp_message *init_packet;
319 int ndis_version;
320 struct net_device *ndev;
321
322 net_device = get_outbound_net_device(device);
323 if (!net_device)
324 return -ENODEV;
325 ndev = net_device->ndev;
326
327 init_packet = &net_device->channel_init_pkt;
328
329 /* Negotiate the latest NVSP protocol supported */
330 if (negotiate_nvsp_ver(device, net_device, init_packet,
331 NVSP_PROTOCOL_VERSION_2) == 0) {
332 net_device->nvsp_version = NVSP_PROTOCOL_VERSION_2;
333 } else if (negotiate_nvsp_ver(device, net_device, init_packet,
334 NVSP_PROTOCOL_VERSION_1) == 0) {
335 net_device->nvsp_version = NVSP_PROTOCOL_VERSION_1;
336 } else {
337 ret = -EPROTO;
338 goto cleanup;
339 }
340
341 pr_debug("Negotiated NVSP version:%x\n", net_device->nvsp_version);
342
343 /* Send the ndis version */
344 memset(init_packet, 0, sizeof(struct nvsp_message));
345
346 ndis_version = 0x00050001;
347
348 init_packet->hdr.msg_type = NVSP_MSG1_TYPE_SEND_NDIS_VER;
349 init_packet->msg.v1_msg.
350 send_ndis_ver.ndis_major_ver =
351 (ndis_version & 0xFFFF0000) >> 16;
352 init_packet->msg.v1_msg.
353 send_ndis_ver.ndis_minor_ver =
354 ndis_version & 0xFFFF;
355
356 /* Send the init request */
357 ret = vmbus_sendpacket(device->channel, init_packet,
358 sizeof(struct nvsp_message),
359 (unsigned long)init_packet,
360 VM_PKT_DATA_INBAND, 0);
361 if (ret != 0)
362 goto cleanup;
363
364 /* Post the big receive buffer to NetVSP */
365 ret = netvsc_init_recv_buf(device);
366
367cleanup:
368 return ret;
369}
370
371static void netvsc_disconnect_vsp(struct netvsc_device *net_device)
372{
373 netvsc_destroy_recv_buf(net_device);
374}
375
376/*
377 * netvsc_device_remove - Callback when the root bus device is removed
378 */
379int netvsc_device_remove(struct hv_device *device)
380{
381 struct netvsc_device *net_device;
382 struct hv_netvsc_packet *netvsc_packet, *pos;
383 unsigned long flags;
384
385 net_device = hv_get_drvdata(device);
386 spin_lock_irqsave(&device->channel->inbound_lock, flags);
387 net_device->destroy = true;
388 spin_unlock_irqrestore(&device->channel->inbound_lock, flags);
389
390 /* Wait for all send completions */
391 wait_event(net_device->wait_drain,
392 atomic_read(&net_device->num_outstanding_sends) == 0);
393
394 netvsc_disconnect_vsp(net_device);
395
396 /*
397 * Since we have already drained, we don't need to busy wait
398 * as was done in final_release_stor_device()
399 * Note that we cannot set the ext pointer to NULL until
400 * we have drained - to drain the outgoing packets, we need to
401 * allow incoming packets.
402 */
403
404 spin_lock_irqsave(&device->channel->inbound_lock, flags);
405 hv_set_drvdata(device, NULL);
406 spin_unlock_irqrestore(&device->channel->inbound_lock, flags);
407
408 /*
409 * At this point, no one should be accessing net_device
410 * except in here
411 */
412 dev_notice(&device->device, "net device safe to remove\n");
413
414 /* Now, we can close the channel safely */
415 vmbus_close(device->channel);
416
417 /* Release all resources */
418 list_for_each_entry_safe(netvsc_packet, pos,
419 &net_device->recv_pkt_list, list_ent) {
420 list_del(&netvsc_packet->list_ent);
421 kfree(netvsc_packet);
422 }
423
424 kfree(net_device);
425 return 0;
426}
427
428
429#define RING_AVAIL_PERCENT_HIWATER 20
430#define RING_AVAIL_PERCENT_LOWATER 10
431
432/*
433 * Get the percentage of available bytes to write in the ring.
434 * The return value is in range from 0 to 100.
435 */
436static inline u32 hv_ringbuf_avail_percent(
437 struct hv_ring_buffer_info *ring_info)
438{
439 u32 avail_read, avail_write;
440
441 hv_get_ringbuffer_availbytes(ring_info, &avail_read, &avail_write);
442
443 return avail_write * 100 / ring_info->ring_datasize;
444}
445
446static void netvsc_send_completion(struct hv_device *device,
447 struct vmpacket_descriptor *packet)
448{
449 struct netvsc_device *net_device;
450 struct nvsp_message *nvsp_packet;
451 struct hv_netvsc_packet *nvsc_packet;
452 struct net_device *ndev;
453
454 net_device = get_inbound_net_device(device);
455 if (!net_device)
456 return;
457 ndev = net_device->ndev;
458
459 nvsp_packet = (struct nvsp_message *)((unsigned long)packet +
460 (packet->offset8 << 3));
461
462 if ((nvsp_packet->hdr.msg_type == NVSP_MSG_TYPE_INIT_COMPLETE) ||
463 (nvsp_packet->hdr.msg_type ==
464 NVSP_MSG1_TYPE_SEND_RECV_BUF_COMPLETE) ||
465 (nvsp_packet->hdr.msg_type ==
466 NVSP_MSG1_TYPE_SEND_SEND_BUF_COMPLETE)) {
467 /* Copy the response back */
468 memcpy(&net_device->channel_init_pkt, nvsp_packet,
469 sizeof(struct nvsp_message));
470 complete(&net_device->channel_init_wait);
471 } else if (nvsp_packet->hdr.msg_type ==
472 NVSP_MSG1_TYPE_SEND_RNDIS_PKT_COMPLETE) {
473 int num_outstanding_sends;
474
475 /* Get the send context */
476 nvsc_packet = (struct hv_netvsc_packet *)(unsigned long)
477 packet->trans_id;
478
479 /* Notify the layer above us */
480 nvsc_packet->completion.send.send_completion(
481 nvsc_packet->completion.send.send_completion_ctx);
482
483 num_outstanding_sends =
484 atomic_dec_return(&net_device->num_outstanding_sends);
485
486 if (net_device->destroy && num_outstanding_sends == 0)
487 wake_up(&net_device->wait_drain);
488
489 if (netif_queue_stopped(ndev) && !net_device->start_remove &&
490 (hv_ringbuf_avail_percent(&device->channel->outbound)
491 > RING_AVAIL_PERCENT_HIWATER ||
492 num_outstanding_sends < 1))
493 netif_wake_queue(ndev);
494 } else {
495 netdev_err(ndev, "Unknown send completion packet type- "
496 "%d received!!\n", nvsp_packet->hdr.msg_type);
497 }
498
499}
500
501int netvsc_send(struct hv_device *device,
502 struct hv_netvsc_packet *packet)
503{
504 struct netvsc_device *net_device;
505 int ret = 0;
506 struct nvsp_message sendMessage;
507 struct net_device *ndev;
508
509 net_device = get_outbound_net_device(device);
510 if (!net_device)
511 return -ENODEV;
512 ndev = net_device->ndev;
513
514 sendMessage.hdr.msg_type = NVSP_MSG1_TYPE_SEND_RNDIS_PKT;
515 if (packet->is_data_pkt) {
516 /* 0 is RMC_DATA; */
517 sendMessage.msg.v1_msg.send_rndis_pkt.channel_type = 0;
518 } else {
519 /* 1 is RMC_CONTROL; */
520 sendMessage.msg.v1_msg.send_rndis_pkt.channel_type = 1;
521 }
522
523 /* Not using send buffer section */
524 sendMessage.msg.v1_msg.send_rndis_pkt.send_buf_section_index =
525 0xFFFFFFFF;
526 sendMessage.msg.v1_msg.send_rndis_pkt.send_buf_section_size = 0;
527
528 if (packet->page_buf_cnt) {
529 ret = vmbus_sendpacket_pagebuffer(device->channel,
530 packet->page_buf,
531 packet->page_buf_cnt,
532 &sendMessage,
533 sizeof(struct nvsp_message),
534 (unsigned long)packet);
535 } else {
536 ret = vmbus_sendpacket(device->channel, &sendMessage,
537 sizeof(struct nvsp_message),
538 (unsigned long)packet,
539 VM_PKT_DATA_INBAND,
540 VMBUS_DATA_PACKET_FLAG_COMPLETION_REQUESTED);
541
542 }
543
544 if (ret == 0) {
545 atomic_inc(&net_device->num_outstanding_sends);
546 if (hv_ringbuf_avail_percent(&device->channel->outbound) <
547 RING_AVAIL_PERCENT_LOWATER) {
548 netif_stop_queue(ndev);
549 if (atomic_read(&net_device->
550 num_outstanding_sends) < 1)
551 netif_wake_queue(ndev);
552 }
553 } else if (ret == -EAGAIN) {
554 netif_stop_queue(ndev);
555 if (atomic_read(&net_device->num_outstanding_sends) < 1) {
556 netif_wake_queue(ndev);
557 ret = -ENOSPC;
558 }
559 } else {
560 netdev_err(ndev, "Unable to send packet %p ret %d\n",
561 packet, ret);
562 }
563
564 return ret;
565}
566
567static void netvsc_send_recv_completion(struct hv_device *device,
568 u64 transaction_id)
569{
570 struct nvsp_message recvcompMessage;
571 int retries = 0;
572 int ret;
573 struct net_device *ndev;
574 struct netvsc_device *net_device = hv_get_drvdata(device);
575
576 ndev = net_device->ndev;
577
578 recvcompMessage.hdr.msg_type =
579 NVSP_MSG1_TYPE_SEND_RNDIS_PKT_COMPLETE;
580
581 /* FIXME: Pass in the status */
582 recvcompMessage.msg.v1_msg.send_rndis_pkt_complete.status =
583 NVSP_STAT_SUCCESS;
584
585retry_send_cmplt:
586 /* Send the completion */
587 ret = vmbus_sendpacket(device->channel, &recvcompMessage,
588 sizeof(struct nvsp_message), transaction_id,
589 VM_PKT_COMP, 0);
590 if (ret == 0) {
591 /* success */
592 /* no-op */
593 } else if (ret == -EAGAIN) {
594 /* no more room...wait a bit and attempt to retry 3 times */
595 retries++;
596 netdev_err(ndev, "unable to send receive completion pkt"
597 " (tid %llx)...retrying %d\n", transaction_id, retries);
598
599 if (retries < 4) {
600 udelay(100);
601 goto retry_send_cmplt;
602 } else {
603 netdev_err(ndev, "unable to send receive "
604 "completion pkt (tid %llx)...give up retrying\n",
605 transaction_id);
606 }
607 } else {
608 netdev_err(ndev, "unable to send receive "
609 "completion pkt - %llx\n", transaction_id);
610 }
611}
612
613/* Send a receive completion packet to RNDIS device (ie NetVsp) */
614static void netvsc_receive_completion(void *context)
615{
616 struct hv_netvsc_packet *packet = context;
617 struct hv_device *device = (struct hv_device *)packet->device;
618 struct netvsc_device *net_device;
619 u64 transaction_id = 0;
620 bool fsend_receive_comp = false;
621 unsigned long flags;
622 struct net_device *ndev;
623
624 /*
625 * Even though it seems logical to do a GetOutboundNetDevice() here to
626 * send out receive completion, we are using GetInboundNetDevice()
627 * since we may have disable outbound traffic already.
628 */
629 net_device = get_inbound_net_device(device);
630 if (!net_device)
631 return;
632 ndev = net_device->ndev;
633
634 /* Overloading use of the lock. */
635 spin_lock_irqsave(&net_device->recv_pkt_list_lock, flags);
636
637 packet->xfer_page_pkt->count--;
638
639 /*
640 * Last one in the line that represent 1 xfer page packet.
641 * Return the xfer page packet itself to the freelist
642 */
643 if (packet->xfer_page_pkt->count == 0) {
644 fsend_receive_comp = true;
645 transaction_id = packet->completion.recv.recv_completion_tid;
646 list_add_tail(&packet->xfer_page_pkt->list_ent,
647 &net_device->recv_pkt_list);
648
649 }
650
651 /* Put the packet back */
652 list_add_tail(&packet->list_ent, &net_device->recv_pkt_list);
653 spin_unlock_irqrestore(&net_device->recv_pkt_list_lock, flags);
654
655 /* Send a receive completion for the xfer page packet */
656 if (fsend_receive_comp)
657 netvsc_send_recv_completion(device, transaction_id);
658
659}
660
661static void netvsc_receive(struct hv_device *device,
662 struct vmpacket_descriptor *packet)
663{
664 struct netvsc_device *net_device;
665 struct vmtransfer_page_packet_header *vmxferpage_packet;
666 struct nvsp_message *nvsp_packet;
667 struct hv_netvsc_packet *netvsc_packet = NULL;
668 /* struct netvsc_driver *netvscDriver; */
669 struct xferpage_packet *xferpage_packet = NULL;
670 int i;
671 int count = 0;
672 unsigned long flags;
673 struct net_device *ndev;
674
675 LIST_HEAD(listHead);
676
677 net_device = get_inbound_net_device(device);
678 if (!net_device)
679 return;
680 ndev = net_device->ndev;
681
682 /*
683 * All inbound packets other than send completion should be xfer page
684 * packet
685 */
686 if (packet->type != VM_PKT_DATA_USING_XFER_PAGES) {
687 netdev_err(ndev, "Unknown packet type received - %d\n",
688 packet->type);
689 return;
690 }
691
692 nvsp_packet = (struct nvsp_message *)((unsigned long)packet +
693 (packet->offset8 << 3));
694
695 /* Make sure this is a valid nvsp packet */
696 if (nvsp_packet->hdr.msg_type !=
697 NVSP_MSG1_TYPE_SEND_RNDIS_PKT) {
698 netdev_err(ndev, "Unknown nvsp packet type received-"
699 " %d\n", nvsp_packet->hdr.msg_type);
700 return;
701 }
702
703 vmxferpage_packet = (struct vmtransfer_page_packet_header *)packet;
704
705 if (vmxferpage_packet->xfer_pageset_id != NETVSC_RECEIVE_BUFFER_ID) {
706 netdev_err(ndev, "Invalid xfer page set id - "
707 "expecting %x got %x\n", NETVSC_RECEIVE_BUFFER_ID,
708 vmxferpage_packet->xfer_pageset_id);
709 return;
710 }
711
712 /*
713 * Grab free packets (range count + 1) to represent this xfer
714 * page packet. +1 to represent the xfer page packet itself.
715 * We grab it here so that we know exactly how many we can
716 * fulfil
717 */
718 spin_lock_irqsave(&net_device->recv_pkt_list_lock, flags);
719 while (!list_empty(&net_device->recv_pkt_list)) {
720 list_move_tail(net_device->recv_pkt_list.next, &listHead);
721 if (++count == vmxferpage_packet->range_cnt + 1)
722 break;
723 }
724 spin_unlock_irqrestore(&net_device->recv_pkt_list_lock, flags);
725
726 /*
727 * We need at least 2 netvsc pkts (1 to represent the xfer
728 * page and at least 1 for the range) i.e. we can handled
729 * some of the xfer page packet ranges...
730 */
731 if (count < 2) {
732 netdev_err(ndev, "Got only %d netvsc pkt...needed "
733 "%d pkts. Dropping this xfer page packet completely!\n",
734 count, vmxferpage_packet->range_cnt + 1);
735
736 /* Return it to the freelist */
737 spin_lock_irqsave(&net_device->recv_pkt_list_lock, flags);
738 for (i = count; i != 0; i--) {
739 list_move_tail(listHead.next,
740 &net_device->recv_pkt_list);
741 }
742 spin_unlock_irqrestore(&net_device->recv_pkt_list_lock,
743 flags);
744
745 netvsc_send_recv_completion(device,
746 vmxferpage_packet->d.trans_id);
747
748 return;
749 }
750
751 /* Remove the 1st packet to represent the xfer page packet itself */
752 xferpage_packet = (struct xferpage_packet *)listHead.next;
753 list_del(&xferpage_packet->list_ent);
754
755 /* This is how much we can satisfy */
756 xferpage_packet->count = count - 1;
757
758 if (xferpage_packet->count != vmxferpage_packet->range_cnt) {
759 netdev_err(ndev, "Needed %d netvsc pkts to satisfy "
760 "this xfer page...got %d\n",
761 vmxferpage_packet->range_cnt, xferpage_packet->count);
762 }
763
764 /* Each range represents 1 RNDIS pkt that contains 1 ethernet frame */
765 for (i = 0; i < (count - 1); i++) {
766 netvsc_packet = (struct hv_netvsc_packet *)listHead.next;
767 list_del(&netvsc_packet->list_ent);
768
769 /* Initialize the netvsc packet */
770 netvsc_packet->xfer_page_pkt = xferpage_packet;
771 netvsc_packet->completion.recv.recv_completion =
772 netvsc_receive_completion;
773 netvsc_packet->completion.recv.recv_completion_ctx =
774 netvsc_packet;
775 netvsc_packet->device = device;
776 /* Save this so that we can send it back */
777 netvsc_packet->completion.recv.recv_completion_tid =
778 vmxferpage_packet->d.trans_id;
779
780 netvsc_packet->data = (void *)((unsigned long)net_device->
781 recv_buf + vmxferpage_packet->ranges[i].byte_offset);
782 netvsc_packet->total_data_buflen =
783 vmxferpage_packet->ranges[i].byte_count;
784
785 /* Pass it to the upper layer */
786 rndis_filter_receive(device, netvsc_packet);
787
788 netvsc_receive_completion(netvsc_packet->
789 completion.recv.recv_completion_ctx);
790 }
791
792}
793
794static void netvsc_channel_cb(void *context)
795{
796 int ret;
797 struct hv_device *device = context;
798 struct netvsc_device *net_device;
799 u32 bytes_recvd;
800 u64 request_id;
801 unsigned char *packet;
802 struct vmpacket_descriptor *desc;
803 unsigned char *buffer;
804 int bufferlen = NETVSC_PACKET_SIZE;
805 struct net_device *ndev;
806
807 packet = kzalloc(NETVSC_PACKET_SIZE * sizeof(unsigned char),
808 GFP_ATOMIC);
809 if (!packet)
810 return;
811 buffer = packet;
812
813 net_device = get_inbound_net_device(device);
814 if (!net_device)
815 goto out;
816 ndev = net_device->ndev;
817
818 do {
819 ret = vmbus_recvpacket_raw(device->channel, buffer, bufferlen,
820 &bytes_recvd, &request_id);
821 if (ret == 0) {
822 if (bytes_recvd > 0) {
823 desc = (struct vmpacket_descriptor *)buffer;
824 switch (desc->type) {
825 case VM_PKT_COMP:
826 netvsc_send_completion(device, desc);
827 break;
828
829 case VM_PKT_DATA_USING_XFER_PAGES:
830 netvsc_receive(device, desc);
831 break;
832
833 default:
834 netdev_err(ndev,
835 "unhandled packet type %d, "
836 "tid %llx len %d\n",
837 desc->type, request_id,
838 bytes_recvd);
839 break;
840 }
841
842 /* reset */
843 if (bufferlen > NETVSC_PACKET_SIZE) {
844 kfree(buffer);
845 buffer = packet;
846 bufferlen = NETVSC_PACKET_SIZE;
847 }
848 } else {
849 /* reset */
850 if (bufferlen > NETVSC_PACKET_SIZE) {
851 kfree(buffer);
852 buffer = packet;
853 bufferlen = NETVSC_PACKET_SIZE;
854 }
855
856 break;
857 }
858 } else if (ret == -ENOBUFS) {
859 /* Handle large packet */
860 buffer = kmalloc(bytes_recvd, GFP_ATOMIC);
861 if (buffer == NULL) {
862 /* Try again next time around */
863 netdev_err(ndev,
864 "unable to allocate buffer of size "
865 "(%d)!!\n", bytes_recvd);
866 break;
867 }
868
869 bufferlen = bytes_recvd;
870 }
871 } while (1);
872
873out:
874 kfree(buffer);
875 return;
876}
877
878/*
879 * netvsc_device_add - Callback when the device belonging to this
880 * driver is added
881 */
882int netvsc_device_add(struct hv_device *device, void *additional_info)
883{
884 int ret = 0;
885 int i;
886 int ring_size =
887 ((struct netvsc_device_info *)additional_info)->ring_size;
888 struct netvsc_device *net_device;
889 struct hv_netvsc_packet *packet, *pos;
890 struct net_device *ndev;
891
892 net_device = alloc_net_device(device);
893 if (!net_device) {
894 ret = -ENOMEM;
895 goto cleanup;
896 }
897
898 /*
899 * Coming into this function, struct net_device * is
900 * registered as the driver private data.
901 * In alloc_net_device(), we register struct netvsc_device *
902 * as the driver private data and stash away struct net_device *
903 * in struct netvsc_device *.
904 */
905 ndev = net_device->ndev;
906
907 /* Initialize the NetVSC channel extension */
908 net_device->recv_buf_size = NETVSC_RECEIVE_BUFFER_SIZE;
909 spin_lock_init(&net_device->recv_pkt_list_lock);
910
911 INIT_LIST_HEAD(&net_device->recv_pkt_list);
912
913 for (i = 0; i < NETVSC_RECEIVE_PACKETLIST_COUNT; i++) {
914 packet = kzalloc(sizeof(struct hv_netvsc_packet) +
915 (NETVSC_RECEIVE_SG_COUNT *
916 sizeof(struct hv_page_buffer)), GFP_KERNEL);
917 if (!packet)
918 break;
919
920 list_add_tail(&packet->list_ent,
921 &net_device->recv_pkt_list);
922 }
923 init_completion(&net_device->channel_init_wait);
924
925 /* Open the channel */
926 ret = vmbus_open(device->channel, ring_size * PAGE_SIZE,
927 ring_size * PAGE_SIZE, NULL, 0,
928 netvsc_channel_cb, device);
929
930 if (ret != 0) {
931 netdev_err(ndev, "unable to open channel: %d\n", ret);
932 goto cleanup;
933 }
934
935 /* Channel is opened */
936 pr_info("hv_netvsc channel opened successfully\n");
937
938 /* Connect with the NetVsp */
939 ret = netvsc_connect_vsp(device);
940 if (ret != 0) {
941 netdev_err(ndev,
942 "unable to connect to NetVSP - %d\n", ret);
943 goto close;
944 }
945
946 return ret;
947
948close:
949 /* Now, we can close the channel safely */
950 vmbus_close(device->channel);
951
952cleanup:
953
954 if (net_device) {
955 list_for_each_entry_safe(packet, pos,
956 &net_device->recv_pkt_list,
957 list_ent) {
958 list_del(&packet->list_ent);
959 kfree(packet);
960 }
961
962 kfree(net_device);
963 }
964
965 return ret;
966}