Loading...
1/*
2 * Copyright (c) 2009, Microsoft Corporation.
3 *
4 * This program is free software; you can redistribute it and/or modify it
5 * under the terms and conditions of the GNU General Public License,
6 * version 2, as published by the Free Software Foundation.
7 *
8 * This program is distributed in the hope it will be useful, but WITHOUT
9 * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
10 * FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License for
11 * more details.
12 *
13 * You should have received a copy of the GNU General Public License along with
14 * this program; if not, see <http://www.gnu.org/licenses/>.
15 *
16 * Authors:
17 * Haiyang Zhang <haiyangz@microsoft.com>
18 * Hank Janssen <hjanssen@microsoft.com>
19 */
20#define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
21
22#include <linux/kernel.h>
23#include <linux/sched.h>
24#include <linux/wait.h>
25#include <linux/mm.h>
26#include <linux/delay.h>
27#include <linux/io.h>
28#include <linux/slab.h>
29#include <linux/netdevice.h>
30#include <linux/if_ether.h>
31
32#include "hyperv_net.h"
33
34
35static struct netvsc_device *alloc_net_device(struct hv_device *device)
36{
37 struct netvsc_device *net_device;
38 struct net_device *ndev = hv_get_drvdata(device);
39
40 net_device = kzalloc(sizeof(struct netvsc_device), GFP_KERNEL);
41 if (!net_device)
42 return NULL;
43
44 init_waitqueue_head(&net_device->wait_drain);
45 net_device->start_remove = false;
46 net_device->destroy = false;
47 net_device->dev = device;
48 net_device->ndev = ndev;
49
50 hv_set_drvdata(device, net_device);
51 return net_device;
52}
53
54static struct netvsc_device *get_outbound_net_device(struct hv_device *device)
55{
56 struct netvsc_device *net_device;
57
58 net_device = hv_get_drvdata(device);
59 if (net_device && net_device->destroy)
60 net_device = NULL;
61
62 return net_device;
63}
64
65static struct netvsc_device *get_inbound_net_device(struct hv_device *device)
66{
67 struct netvsc_device *net_device;
68
69 net_device = hv_get_drvdata(device);
70
71 if (!net_device)
72 goto get_in_err;
73
74 if (net_device->destroy &&
75 atomic_read(&net_device->num_outstanding_sends) == 0)
76 net_device = NULL;
77
78get_in_err:
79 return net_device;
80}
81
82
83static int netvsc_destroy_recv_buf(struct netvsc_device *net_device)
84{
85 struct nvsp_message *revoke_packet;
86 int ret = 0;
87 struct net_device *ndev = net_device->ndev;
88
89 /*
90 * If we got a section count, it means we received a
91 * SendReceiveBufferComplete msg (ie sent
92 * NvspMessage1TypeSendReceiveBuffer msg) therefore, we need
93 * to send a revoke msg here
94 */
95 if (net_device->recv_section_cnt) {
96 /* Send the revoke receive buffer */
97 revoke_packet = &net_device->revoke_packet;
98 memset(revoke_packet, 0, sizeof(struct nvsp_message));
99
100 revoke_packet->hdr.msg_type =
101 NVSP_MSG1_TYPE_REVOKE_RECV_BUF;
102 revoke_packet->msg.v1_msg.
103 revoke_recv_buf.id = NETVSC_RECEIVE_BUFFER_ID;
104
105 ret = vmbus_sendpacket(net_device->dev->channel,
106 revoke_packet,
107 sizeof(struct nvsp_message),
108 (unsigned long)revoke_packet,
109 VM_PKT_DATA_INBAND, 0);
110 /*
111 * If we failed here, we might as well return and
112 * have a leak rather than continue and a bugchk
113 */
114 if (ret != 0) {
115 netdev_err(ndev, "unable to send "
116 "revoke receive buffer to netvsp\n");
117 return ret;
118 }
119 }
120
121 /* Teardown the gpadl on the vsp end */
122 if (net_device->recv_buf_gpadl_handle) {
123 ret = vmbus_teardown_gpadl(net_device->dev->channel,
124 net_device->recv_buf_gpadl_handle);
125
126 /* If we failed here, we might as well return and have a leak
127 * rather than continue and a bugchk
128 */
129 if (ret != 0) {
130 netdev_err(ndev,
131 "unable to teardown receive buffer's gpadl\n");
132 return ret;
133 }
134 net_device->recv_buf_gpadl_handle = 0;
135 }
136
137 if (net_device->recv_buf) {
138 /* Free up the receive buffer */
139 vfree(net_device->recv_buf);
140 net_device->recv_buf = NULL;
141 }
142
143 if (net_device->recv_section) {
144 net_device->recv_section_cnt = 0;
145 kfree(net_device->recv_section);
146 net_device->recv_section = NULL;
147 }
148
149 return ret;
150}
151
152static int netvsc_init_recv_buf(struct hv_device *device)
153{
154 int ret = 0;
155 int t;
156 struct netvsc_device *net_device;
157 struct nvsp_message *init_packet;
158 struct net_device *ndev;
159
160 net_device = get_outbound_net_device(device);
161 if (!net_device)
162 return -ENODEV;
163 ndev = net_device->ndev;
164
165 net_device->recv_buf = vzalloc(net_device->recv_buf_size);
166 if (!net_device->recv_buf) {
167 netdev_err(ndev, "unable to allocate receive "
168 "buffer of size %d\n", net_device->recv_buf_size);
169 ret = -ENOMEM;
170 goto cleanup;
171 }
172
173 /*
174 * Establish the gpadl handle for this buffer on this
175 * channel. Note: This call uses the vmbus connection rather
176 * than the channel to establish the gpadl handle.
177 */
178 ret = vmbus_establish_gpadl(device->channel, net_device->recv_buf,
179 net_device->recv_buf_size,
180 &net_device->recv_buf_gpadl_handle);
181 if (ret != 0) {
182 netdev_err(ndev,
183 "unable to establish receive buffer's gpadl\n");
184 goto cleanup;
185 }
186
187
188 /* Notify the NetVsp of the gpadl handle */
189 init_packet = &net_device->channel_init_pkt;
190
191 memset(init_packet, 0, sizeof(struct nvsp_message));
192
193 init_packet->hdr.msg_type = NVSP_MSG1_TYPE_SEND_RECV_BUF;
194 init_packet->msg.v1_msg.send_recv_buf.
195 gpadl_handle = net_device->recv_buf_gpadl_handle;
196 init_packet->msg.v1_msg.
197 send_recv_buf.id = NETVSC_RECEIVE_BUFFER_ID;
198
199 /* Send the gpadl notification request */
200 ret = vmbus_sendpacket(device->channel, init_packet,
201 sizeof(struct nvsp_message),
202 (unsigned long)init_packet,
203 VM_PKT_DATA_INBAND,
204 VMBUS_DATA_PACKET_FLAG_COMPLETION_REQUESTED);
205 if (ret != 0) {
206 netdev_err(ndev,
207 "unable to send receive buffer's gpadl to netvsp\n");
208 goto cleanup;
209 }
210
211 t = wait_for_completion_timeout(&net_device->channel_init_wait, 5*HZ);
212 BUG_ON(t == 0);
213
214
215 /* Check the response */
216 if (init_packet->msg.v1_msg.
217 send_recv_buf_complete.status != NVSP_STAT_SUCCESS) {
218 netdev_err(ndev, "Unable to complete receive buffer "
219 "initialization with NetVsp - status %d\n",
220 init_packet->msg.v1_msg.
221 send_recv_buf_complete.status);
222 ret = -EINVAL;
223 goto cleanup;
224 }
225
226 /* Parse the response */
227
228 net_device->recv_section_cnt = init_packet->msg.
229 v1_msg.send_recv_buf_complete.num_sections;
230
231 net_device->recv_section = kmemdup(
232 init_packet->msg.v1_msg.send_recv_buf_complete.sections,
233 net_device->recv_section_cnt *
234 sizeof(struct nvsp_1_receive_buffer_section),
235 GFP_KERNEL);
236 if (net_device->recv_section == NULL) {
237 ret = -EINVAL;
238 goto cleanup;
239 }
240
241 /*
242 * For 1st release, there should only be 1 section that represents the
243 * entire receive buffer
244 */
245 if (net_device->recv_section_cnt != 1 ||
246 net_device->recv_section->offset != 0) {
247 ret = -EINVAL;
248 goto cleanup;
249 }
250
251 goto exit;
252
253cleanup:
254 netvsc_destroy_recv_buf(net_device);
255
256exit:
257 return ret;
258}
259
260
261/* Negotiate NVSP protocol version */
262static int negotiate_nvsp_ver(struct hv_device *device,
263 struct netvsc_device *net_device,
264 struct nvsp_message *init_packet,
265 u32 nvsp_ver)
266{
267 int ret, t;
268
269 memset(init_packet, 0, sizeof(struct nvsp_message));
270 init_packet->hdr.msg_type = NVSP_MSG_TYPE_INIT;
271 init_packet->msg.init_msg.init.min_protocol_ver = nvsp_ver;
272 init_packet->msg.init_msg.init.max_protocol_ver = nvsp_ver;
273
274 /* Send the init request */
275 ret = vmbus_sendpacket(device->channel, init_packet,
276 sizeof(struct nvsp_message),
277 (unsigned long)init_packet,
278 VM_PKT_DATA_INBAND,
279 VMBUS_DATA_PACKET_FLAG_COMPLETION_REQUESTED);
280
281 if (ret != 0)
282 return ret;
283
284 t = wait_for_completion_timeout(&net_device->channel_init_wait, 5*HZ);
285
286 if (t == 0)
287 return -ETIMEDOUT;
288
289 if (init_packet->msg.init_msg.init_complete.status !=
290 NVSP_STAT_SUCCESS)
291 return -EINVAL;
292
293 if (nvsp_ver == NVSP_PROTOCOL_VERSION_1)
294 return 0;
295
296 /* NVSPv2 only: Send NDIS config */
297 memset(init_packet, 0, sizeof(struct nvsp_message));
298 init_packet->hdr.msg_type = NVSP_MSG2_TYPE_SEND_NDIS_CONFIG;
299 init_packet->msg.v2_msg.send_ndis_config.mtu = net_device->ndev->mtu;
300 init_packet->msg.v2_msg.send_ndis_config.capability.ieee8021q = 1;
301
302 ret = vmbus_sendpacket(device->channel, init_packet,
303 sizeof(struct nvsp_message),
304 (unsigned long)init_packet,
305 VM_PKT_DATA_INBAND, 0);
306
307 return ret;
308}
309
310static int netvsc_connect_vsp(struct hv_device *device)
311{
312 int ret;
313 struct netvsc_device *net_device;
314 struct nvsp_message *init_packet;
315 int ndis_version;
316 struct net_device *ndev;
317 u32 ver_list[] = { NVSP_PROTOCOL_VERSION_1, NVSP_PROTOCOL_VERSION_2,
318 NVSP_PROTOCOL_VERSION_4, NVSP_PROTOCOL_VERSION_5 };
319 int i, num_ver = 4; /* number of different NVSP versions */
320
321 net_device = get_outbound_net_device(device);
322 if (!net_device)
323 return -ENODEV;
324 ndev = net_device->ndev;
325
326 init_packet = &net_device->channel_init_pkt;
327
328 /* Negotiate the latest NVSP protocol supported */
329 for (i = num_ver - 1; i >= 0; i--)
330 if (negotiate_nvsp_ver(device, net_device, init_packet,
331 ver_list[i]) == 0) {
332 net_device->nvsp_version = ver_list[i];
333 break;
334 }
335
336 if (i < 0) {
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 if (net_device->nvsp_version <= NVSP_PROTOCOL_VERSION_4)
347 ndis_version = 0x00060001;
348 else
349 ndis_version = 0x0006001e;
350
351 init_packet->hdr.msg_type = NVSP_MSG1_TYPE_SEND_NDIS_VER;
352 init_packet->msg.v1_msg.
353 send_ndis_ver.ndis_major_ver =
354 (ndis_version & 0xFFFF0000) >> 16;
355 init_packet->msg.v1_msg.
356 send_ndis_ver.ndis_minor_ver =
357 ndis_version & 0xFFFF;
358
359 /* Send the init request */
360 ret = vmbus_sendpacket(device->channel, init_packet,
361 sizeof(struct nvsp_message),
362 (unsigned long)init_packet,
363 VM_PKT_DATA_INBAND, 0);
364 if (ret != 0)
365 goto cleanup;
366
367 /* Post the big receive buffer to NetVSP */
368 if (net_device->nvsp_version <= NVSP_PROTOCOL_VERSION_2)
369 net_device->recv_buf_size = NETVSC_RECEIVE_BUFFER_SIZE_LEGACY;
370 else
371 net_device->recv_buf_size = NETVSC_RECEIVE_BUFFER_SIZE;
372
373 ret = netvsc_init_recv_buf(device);
374
375cleanup:
376 return ret;
377}
378
379static void netvsc_disconnect_vsp(struct netvsc_device *net_device)
380{
381 netvsc_destroy_recv_buf(net_device);
382}
383
384/*
385 * netvsc_device_remove - Callback when the root bus device is removed
386 */
387int netvsc_device_remove(struct hv_device *device)
388{
389 struct netvsc_device *net_device;
390 struct hv_netvsc_packet *netvsc_packet, *pos;
391 unsigned long flags;
392
393 net_device = hv_get_drvdata(device);
394
395 netvsc_disconnect_vsp(net_device);
396
397 /*
398 * Since we have already drained, we don't need to busy wait
399 * as was done in final_release_stor_device()
400 * Note that we cannot set the ext pointer to NULL until
401 * we have drained - to drain the outgoing packets, we need to
402 * allow incoming packets.
403 */
404
405 spin_lock_irqsave(&device->channel->inbound_lock, flags);
406 hv_set_drvdata(device, NULL);
407 spin_unlock_irqrestore(&device->channel->inbound_lock, flags);
408
409 /*
410 * At this point, no one should be accessing net_device
411 * except in here
412 */
413 dev_notice(&device->device, "net device safe to remove\n");
414
415 /* Now, we can close the channel safely */
416 vmbus_close(device->channel);
417
418 /* Release all resources */
419 list_for_each_entry_safe(netvsc_packet, pos,
420 &net_device->recv_pkt_list, list_ent) {
421 list_del(&netvsc_packet->list_ent);
422 kfree(netvsc_packet);
423 }
424
425 kfree(net_device);
426 return 0;
427}
428
429
430#define RING_AVAIL_PERCENT_HIWATER 20
431#define RING_AVAIL_PERCENT_LOWATER 10
432
433/*
434 * Get the percentage of available bytes to write in the ring.
435 * The return value is in range from 0 to 100.
436 */
437static inline u32 hv_ringbuf_avail_percent(
438 struct hv_ring_buffer_info *ring_info)
439{
440 u32 avail_read, avail_write;
441
442 hv_get_ringbuffer_availbytes(ring_info, &avail_read, &avail_write);
443
444 return avail_write * 100 / ring_info->ring_datasize;
445}
446
447static void netvsc_send_completion(struct netvsc_device *net_device,
448 struct hv_device *device,
449 struct vmpacket_descriptor *packet)
450{
451 struct nvsp_message *nvsp_packet;
452 struct hv_netvsc_packet *nvsc_packet;
453 struct net_device *ndev;
454
455 ndev = net_device->ndev;
456
457 nvsp_packet = (struct nvsp_message *)((unsigned long)packet +
458 (packet->offset8 << 3));
459
460 if ((nvsp_packet->hdr.msg_type == NVSP_MSG_TYPE_INIT_COMPLETE) ||
461 (nvsp_packet->hdr.msg_type ==
462 NVSP_MSG1_TYPE_SEND_RECV_BUF_COMPLETE) ||
463 (nvsp_packet->hdr.msg_type ==
464 NVSP_MSG1_TYPE_SEND_SEND_BUF_COMPLETE)) {
465 /* Copy the response back */
466 memcpy(&net_device->channel_init_pkt, nvsp_packet,
467 sizeof(struct nvsp_message));
468 complete(&net_device->channel_init_wait);
469 } else if (nvsp_packet->hdr.msg_type ==
470 NVSP_MSG1_TYPE_SEND_RNDIS_PKT_COMPLETE) {
471 int num_outstanding_sends;
472
473 /* Get the send context */
474 nvsc_packet = (struct hv_netvsc_packet *)(unsigned long)
475 packet->trans_id;
476
477 /* Notify the layer above us */
478 if (nvsc_packet)
479 nvsc_packet->completion.send.send_completion(
480 nvsc_packet->completion.send.
481 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 u64 req_id;
509
510 net_device = get_outbound_net_device(device);
511 if (!net_device)
512 return -ENODEV;
513 ndev = net_device->ndev;
514
515 sendMessage.hdr.msg_type = NVSP_MSG1_TYPE_SEND_RNDIS_PKT;
516 if (packet->is_data_pkt) {
517 /* 0 is RMC_DATA; */
518 sendMessage.msg.v1_msg.send_rndis_pkt.channel_type = 0;
519 } else {
520 /* 1 is RMC_CONTROL; */
521 sendMessage.msg.v1_msg.send_rndis_pkt.channel_type = 1;
522 }
523
524 /* Not using send buffer section */
525 sendMessage.msg.v1_msg.send_rndis_pkt.send_buf_section_index =
526 0xFFFFFFFF;
527 sendMessage.msg.v1_msg.send_rndis_pkt.send_buf_section_size = 0;
528
529 if (packet->completion.send.send_completion)
530 req_id = (ulong)packet;
531 else
532 req_id = 0;
533
534 if (packet->page_buf_cnt) {
535 ret = vmbus_sendpacket_pagebuffer(device->channel,
536 packet->page_buf,
537 packet->page_buf_cnt,
538 &sendMessage,
539 sizeof(struct nvsp_message),
540 req_id);
541 } else {
542 ret = vmbus_sendpacket(device->channel, &sendMessage,
543 sizeof(struct nvsp_message),
544 req_id,
545 VM_PKT_DATA_INBAND,
546 VMBUS_DATA_PACKET_FLAG_COMPLETION_REQUESTED);
547 }
548
549 if (ret == 0) {
550 atomic_inc(&net_device->num_outstanding_sends);
551 if (hv_ringbuf_avail_percent(&device->channel->outbound) <
552 RING_AVAIL_PERCENT_LOWATER) {
553 netif_stop_queue(ndev);
554 if (atomic_read(&net_device->
555 num_outstanding_sends) < 1)
556 netif_wake_queue(ndev);
557 }
558 } else if (ret == -EAGAIN) {
559 netif_stop_queue(ndev);
560 if (atomic_read(&net_device->num_outstanding_sends) < 1) {
561 netif_wake_queue(ndev);
562 ret = -ENOSPC;
563 }
564 } else {
565 netdev_err(ndev, "Unable to send packet %p ret %d\n",
566 packet, ret);
567 }
568
569 return ret;
570}
571
572static void netvsc_send_recv_completion(struct hv_device *device,
573 struct netvsc_device *net_device,
574 u64 transaction_id, u32 status)
575{
576 struct nvsp_message recvcompMessage;
577 int retries = 0;
578 int ret;
579 struct net_device *ndev;
580
581 ndev = net_device->ndev;
582
583 recvcompMessage.hdr.msg_type =
584 NVSP_MSG1_TYPE_SEND_RNDIS_PKT_COMPLETE;
585
586 recvcompMessage.msg.v1_msg.send_rndis_pkt_complete.status = status;
587
588retry_send_cmplt:
589 /* Send the completion */
590 ret = vmbus_sendpacket(device->channel, &recvcompMessage,
591 sizeof(struct nvsp_message), transaction_id,
592 VM_PKT_COMP, 0);
593 if (ret == 0) {
594 /* success */
595 /* no-op */
596 } else if (ret == -EAGAIN) {
597 /* no more room...wait a bit and attempt to retry 3 times */
598 retries++;
599 netdev_err(ndev, "unable to send receive completion pkt"
600 " (tid %llx)...retrying %d\n", transaction_id, retries);
601
602 if (retries < 4) {
603 udelay(100);
604 goto retry_send_cmplt;
605 } else {
606 netdev_err(ndev, "unable to send receive "
607 "completion pkt (tid %llx)...give up retrying\n",
608 transaction_id);
609 }
610 } else {
611 netdev_err(ndev, "unable to send receive "
612 "completion pkt - %llx\n", transaction_id);
613 }
614}
615
616/* Send a receive completion packet to RNDIS device (ie NetVsp) */
617static void netvsc_receive_completion(void *context)
618{
619 struct hv_netvsc_packet *packet = context;
620 struct hv_device *device = packet->device;
621 struct netvsc_device *net_device;
622 u64 transaction_id = 0;
623 bool fsend_receive_comp = false;
624 unsigned long flags;
625 struct net_device *ndev;
626 u32 status = NVSP_STAT_NONE;
627
628 /*
629 * Even though it seems logical to do a GetOutboundNetDevice() here to
630 * send out receive completion, we are using GetInboundNetDevice()
631 * since we may have disable outbound traffic already.
632 */
633 net_device = get_inbound_net_device(device);
634 if (!net_device)
635 return;
636 ndev = net_device->ndev;
637
638 /* Overloading use of the lock. */
639 spin_lock_irqsave(&net_device->recv_pkt_list_lock, flags);
640
641 if (packet->status != NVSP_STAT_SUCCESS)
642 packet->xfer_page_pkt->status = NVSP_STAT_FAIL;
643
644 packet->xfer_page_pkt->count--;
645
646 /*
647 * Last one in the line that represent 1 xfer page packet.
648 * Return the xfer page packet itself to the freelist
649 */
650 if (packet->xfer_page_pkt->count == 0) {
651 fsend_receive_comp = true;
652 transaction_id = packet->completion.recv.recv_completion_tid;
653 status = packet->xfer_page_pkt->status;
654 list_add_tail(&packet->xfer_page_pkt->list_ent,
655 &net_device->recv_pkt_list);
656
657 }
658
659 /* Put the packet back */
660 list_add_tail(&packet->list_ent, &net_device->recv_pkt_list);
661 spin_unlock_irqrestore(&net_device->recv_pkt_list_lock, flags);
662
663 /* Send a receive completion for the xfer page packet */
664 if (fsend_receive_comp)
665 netvsc_send_recv_completion(device, net_device, transaction_id,
666 status);
667
668}
669
670static void netvsc_receive(struct netvsc_device *net_device,
671 struct hv_device *device,
672 struct vmpacket_descriptor *packet)
673{
674 struct vmtransfer_page_packet_header *vmxferpage_packet;
675 struct nvsp_message *nvsp_packet;
676 struct hv_netvsc_packet *netvsc_packet = NULL;
677 /* struct netvsc_driver *netvscDriver; */
678 struct xferpage_packet *xferpage_packet = NULL;
679 int i;
680 int count = 0;
681 unsigned long flags;
682 struct net_device *ndev;
683
684 LIST_HEAD(listHead);
685
686 ndev = net_device->ndev;
687
688 /*
689 * All inbound packets other than send completion should be xfer page
690 * packet
691 */
692 if (packet->type != VM_PKT_DATA_USING_XFER_PAGES) {
693 netdev_err(ndev, "Unknown packet type received - %d\n",
694 packet->type);
695 return;
696 }
697
698 nvsp_packet = (struct nvsp_message *)((unsigned long)packet +
699 (packet->offset8 << 3));
700
701 /* Make sure this is a valid nvsp packet */
702 if (nvsp_packet->hdr.msg_type !=
703 NVSP_MSG1_TYPE_SEND_RNDIS_PKT) {
704 netdev_err(ndev, "Unknown nvsp packet type received-"
705 " %d\n", nvsp_packet->hdr.msg_type);
706 return;
707 }
708
709 vmxferpage_packet = (struct vmtransfer_page_packet_header *)packet;
710
711 if (vmxferpage_packet->xfer_pageset_id != NETVSC_RECEIVE_BUFFER_ID) {
712 netdev_err(ndev, "Invalid xfer page set id - "
713 "expecting %x got %x\n", NETVSC_RECEIVE_BUFFER_ID,
714 vmxferpage_packet->xfer_pageset_id);
715 return;
716 }
717
718 /*
719 * Grab free packets (range count + 1) to represent this xfer
720 * page packet. +1 to represent the xfer page packet itself.
721 * We grab it here so that we know exactly how many we can
722 * fulfil
723 */
724 spin_lock_irqsave(&net_device->recv_pkt_list_lock, flags);
725 while (!list_empty(&net_device->recv_pkt_list)) {
726 list_move_tail(net_device->recv_pkt_list.next, &listHead);
727 if (++count == vmxferpage_packet->range_cnt + 1)
728 break;
729 }
730 spin_unlock_irqrestore(&net_device->recv_pkt_list_lock, flags);
731
732 /*
733 * We need at least 2 netvsc pkts (1 to represent the xfer
734 * page and at least 1 for the range) i.e. we can handled
735 * some of the xfer page packet ranges...
736 */
737 if (count < 2) {
738 netdev_err(ndev, "Got only %d netvsc pkt...needed "
739 "%d pkts. Dropping this xfer page packet completely!\n",
740 count, vmxferpage_packet->range_cnt + 1);
741
742 /* Return it to the freelist */
743 spin_lock_irqsave(&net_device->recv_pkt_list_lock, flags);
744 for (i = count; i != 0; i--) {
745 list_move_tail(listHead.next,
746 &net_device->recv_pkt_list);
747 }
748 spin_unlock_irqrestore(&net_device->recv_pkt_list_lock,
749 flags);
750
751 netvsc_send_recv_completion(device, net_device,
752 vmxferpage_packet->d.trans_id,
753 NVSP_STAT_FAIL);
754
755 return;
756 }
757
758 /* Remove the 1st packet to represent the xfer page packet itself */
759 xferpage_packet = (struct xferpage_packet *)listHead.next;
760 list_del(&xferpage_packet->list_ent);
761 xferpage_packet->status = NVSP_STAT_SUCCESS;
762
763 /* This is how much we can satisfy */
764 xferpage_packet->count = count - 1;
765
766 if (xferpage_packet->count != vmxferpage_packet->range_cnt) {
767 netdev_err(ndev, "Needed %d netvsc pkts to satisfy "
768 "this xfer page...got %d\n",
769 vmxferpage_packet->range_cnt, xferpage_packet->count);
770 }
771
772 /* Each range represents 1 RNDIS pkt that contains 1 ethernet frame */
773 for (i = 0; i < (count - 1); i++) {
774 netvsc_packet = (struct hv_netvsc_packet *)listHead.next;
775 list_del(&netvsc_packet->list_ent);
776
777 /* Initialize the netvsc packet */
778 netvsc_packet->status = NVSP_STAT_SUCCESS;
779 netvsc_packet->xfer_page_pkt = xferpage_packet;
780 netvsc_packet->completion.recv.recv_completion =
781 netvsc_receive_completion;
782 netvsc_packet->completion.recv.recv_completion_ctx =
783 netvsc_packet;
784 netvsc_packet->device = device;
785 /* Save this so that we can send it back */
786 netvsc_packet->completion.recv.recv_completion_tid =
787 vmxferpage_packet->d.trans_id;
788
789 netvsc_packet->data = (void *)((unsigned long)net_device->
790 recv_buf + vmxferpage_packet->ranges[i].byte_offset);
791 netvsc_packet->total_data_buflen =
792 vmxferpage_packet->ranges[i].byte_count;
793
794 /* Pass it to the upper layer */
795 rndis_filter_receive(device, netvsc_packet);
796
797 netvsc_receive_completion(netvsc_packet->
798 completion.recv.recv_completion_ctx);
799 }
800
801}
802
803static void netvsc_channel_cb(void *context)
804{
805 int ret;
806 struct hv_device *device = context;
807 struct netvsc_device *net_device;
808 u32 bytes_recvd;
809 u64 request_id;
810 struct vmpacket_descriptor *desc;
811 unsigned char *buffer;
812 int bufferlen = NETVSC_PACKET_SIZE;
813 struct net_device *ndev;
814
815 net_device = get_inbound_net_device(device);
816 if (!net_device)
817 return;
818 ndev = net_device->ndev;
819 buffer = net_device->cb_buffer;
820
821 do {
822 ret = vmbus_recvpacket_raw(device->channel, buffer, bufferlen,
823 &bytes_recvd, &request_id);
824 if (ret == 0) {
825 if (bytes_recvd > 0) {
826 desc = (struct vmpacket_descriptor *)buffer;
827 switch (desc->type) {
828 case VM_PKT_COMP:
829 netvsc_send_completion(net_device,
830 device, desc);
831 break;
832
833 case VM_PKT_DATA_USING_XFER_PAGES:
834 netvsc_receive(net_device,
835 device, desc);
836 break;
837
838 default:
839 netdev_err(ndev,
840 "unhandled packet type %d, "
841 "tid %llx len %d\n",
842 desc->type, request_id,
843 bytes_recvd);
844 break;
845 }
846
847 } else {
848 /*
849 * We are done for this pass.
850 */
851 break;
852 }
853
854 } else if (ret == -ENOBUFS) {
855 if (bufferlen > NETVSC_PACKET_SIZE)
856 kfree(buffer);
857 /* Handle large packet */
858 buffer = kmalloc(bytes_recvd, GFP_ATOMIC);
859 if (buffer == NULL) {
860 /* Try again next time around */
861 netdev_err(ndev,
862 "unable to allocate buffer of size "
863 "(%d)!!\n", bytes_recvd);
864 break;
865 }
866
867 bufferlen = bytes_recvd;
868 }
869 } while (1);
870
871 if (bufferlen > NETVSC_PACKET_SIZE)
872 kfree(buffer);
873 return;
874}
875
876/*
877 * netvsc_device_add - Callback when the device belonging to this
878 * driver is added
879 */
880int netvsc_device_add(struct hv_device *device, void *additional_info)
881{
882 int ret = 0;
883 int i;
884 int ring_size =
885 ((struct netvsc_device_info *)additional_info)->ring_size;
886 struct netvsc_device *net_device;
887 struct hv_netvsc_packet *packet, *pos;
888 struct net_device *ndev;
889
890 net_device = alloc_net_device(device);
891 if (!net_device) {
892 ret = -ENOMEM;
893 goto cleanup;
894 }
895
896 /*
897 * Coming into this function, struct net_device * is
898 * registered as the driver private data.
899 * In alloc_net_device(), we register struct netvsc_device *
900 * as the driver private data and stash away struct net_device *
901 * in struct netvsc_device *.
902 */
903 ndev = net_device->ndev;
904
905 /* Initialize the NetVSC channel extension */
906 spin_lock_init(&net_device->recv_pkt_list_lock);
907
908 INIT_LIST_HEAD(&net_device->recv_pkt_list);
909
910 for (i = 0; i < NETVSC_RECEIVE_PACKETLIST_COUNT; i++) {
911 packet = kzalloc(sizeof(struct hv_netvsc_packet), GFP_KERNEL);
912 if (!packet)
913 break;
914
915 list_add_tail(&packet->list_ent,
916 &net_device->recv_pkt_list);
917 }
918 init_completion(&net_device->channel_init_wait);
919
920 /* Open the channel */
921 ret = vmbus_open(device->channel, ring_size * PAGE_SIZE,
922 ring_size * PAGE_SIZE, NULL, 0,
923 netvsc_channel_cb, device);
924
925 if (ret != 0) {
926 netdev_err(ndev, "unable to open channel: %d\n", ret);
927 goto cleanup;
928 }
929
930 /* Channel is opened */
931 pr_info("hv_netvsc channel opened successfully\n");
932
933 /* Connect with the NetVsp */
934 ret = netvsc_connect_vsp(device);
935 if (ret != 0) {
936 netdev_err(ndev,
937 "unable to connect to NetVSP - %d\n", ret);
938 goto close;
939 }
940
941 return ret;
942
943close:
944 /* Now, we can close the channel safely */
945 vmbus_close(device->channel);
946
947cleanup:
948
949 if (net_device) {
950 list_for_each_entry_safe(packet, pos,
951 &net_device->recv_pkt_list,
952 list_ent) {
953 list_del(&packet->list_ent);
954 kfree(packet);
955 }
956
957 kfree(net_device);
958 }
959
960 return ret;
961}
1/*
2 * Copyright (c) 2009, Microsoft Corporation.
3 *
4 * This program is free software; you can redistribute it and/or modify it
5 * under the terms and conditions of the GNU General Public License,
6 * version 2, as published by the Free Software Foundation.
7 *
8 * This program is distributed in the hope it will be useful, but WITHOUT
9 * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
10 * FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License for
11 * more details.
12 *
13 * You should have received a copy of the GNU General Public License along with
14 * this program; if not, see <http://www.gnu.org/licenses/>.
15 *
16 * Authors:
17 * Haiyang Zhang <haiyangz@microsoft.com>
18 * Hank Janssen <hjanssen@microsoft.com>
19 */
20#define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
21
22#include <linux/kernel.h>
23#include <linux/sched.h>
24#include <linux/wait.h>
25#include <linux/mm.h>
26#include <linux/delay.h>
27#include <linux/io.h>
28#include <linux/slab.h>
29#include <linux/netdevice.h>
30#include <linux/if_ether.h>
31#include <linux/vmalloc.h>
32#include <asm/sync_bitops.h>
33
34#include "hyperv_net.h"
35
36/*
37 * Switch the data path from the synthetic interface to the VF
38 * interface.
39 */
40void netvsc_switch_datapath(struct net_device *ndev, bool vf)
41{
42 struct net_device_context *net_device_ctx = netdev_priv(ndev);
43 struct hv_device *dev = net_device_ctx->device_ctx;
44 struct netvsc_device *nv_dev = net_device_ctx->nvdev;
45 struct nvsp_message *init_pkt = &nv_dev->channel_init_pkt;
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
56 vmbus_sendpacket(dev->channel, init_pkt,
57 sizeof(struct nvsp_message),
58 (unsigned long)init_pkt,
59 VM_PKT_DATA_INBAND, 0);
60}
61
62static struct netvsc_device *alloc_net_device(void)
63{
64 struct netvsc_device *net_device;
65
66 net_device = kzalloc(sizeof(struct netvsc_device), GFP_KERNEL);
67 if (!net_device)
68 return NULL;
69
70 net_device->cb_buffer = kzalloc(NETVSC_PACKET_SIZE, GFP_KERNEL);
71 if (!net_device->cb_buffer) {
72 kfree(net_device);
73 return NULL;
74 }
75
76 net_device->mrc[0].buf = vzalloc(NETVSC_RECVSLOT_MAX *
77 sizeof(struct recv_comp_data));
78
79 init_waitqueue_head(&net_device->wait_drain);
80 net_device->destroy = false;
81 atomic_set(&net_device->open_cnt, 0);
82 net_device->max_pkt = RNDIS_MAX_PKT_DEFAULT;
83 net_device->pkt_align = RNDIS_PKT_ALIGN_DEFAULT;
84 init_completion(&net_device->channel_init_wait);
85
86 return net_device;
87}
88
89static void free_netvsc_device(struct netvsc_device *nvdev)
90{
91 int i;
92
93 for (i = 0; i < VRSS_CHANNEL_MAX; i++)
94 vfree(nvdev->mrc[i].buf);
95
96 kfree(nvdev->cb_buffer);
97 kfree(nvdev);
98}
99
100static struct netvsc_device *get_outbound_net_device(struct hv_device *device)
101{
102 struct netvsc_device *net_device = hv_device_to_netvsc_device(device);
103
104 if (net_device && net_device->destroy)
105 net_device = NULL;
106
107 return net_device;
108}
109
110static struct netvsc_device *get_inbound_net_device(struct hv_device *device)
111{
112 struct netvsc_device *net_device = hv_device_to_netvsc_device(device);
113
114 if (!net_device)
115 goto get_in_err;
116
117 if (net_device->destroy &&
118 atomic_read(&net_device->num_outstanding_sends) == 0 &&
119 atomic_read(&net_device->num_outstanding_recvs) == 0)
120 net_device = NULL;
121
122get_in_err:
123 return net_device;
124}
125
126static void netvsc_destroy_buf(struct hv_device *device)
127{
128 struct nvsp_message *revoke_packet;
129 struct net_device *ndev = hv_get_drvdata(device);
130 struct netvsc_device *net_device = net_device_to_netvsc_device(ndev);
131 int ret;
132
133 /*
134 * If we got a section count, it means we received a
135 * SendReceiveBufferComplete msg (ie sent
136 * NvspMessage1TypeSendReceiveBuffer msg) therefore, we need
137 * to send a revoke msg here
138 */
139 if (net_device->recv_section_cnt) {
140 /* Send the revoke receive buffer */
141 revoke_packet = &net_device->revoke_packet;
142 memset(revoke_packet, 0, sizeof(struct nvsp_message));
143
144 revoke_packet->hdr.msg_type =
145 NVSP_MSG1_TYPE_REVOKE_RECV_BUF;
146 revoke_packet->msg.v1_msg.
147 revoke_recv_buf.id = NETVSC_RECEIVE_BUFFER_ID;
148
149 ret = vmbus_sendpacket(device->channel,
150 revoke_packet,
151 sizeof(struct nvsp_message),
152 (unsigned long)revoke_packet,
153 VM_PKT_DATA_INBAND, 0);
154 /*
155 * If we failed here, we might as well return and
156 * have a leak rather than continue and a bugchk
157 */
158 if (ret != 0) {
159 netdev_err(ndev, "unable to send "
160 "revoke receive buffer to netvsp\n");
161 return;
162 }
163 }
164
165 /* Teardown the gpadl on the vsp end */
166 if (net_device->recv_buf_gpadl_handle) {
167 ret = vmbus_teardown_gpadl(device->channel,
168 net_device->recv_buf_gpadl_handle);
169
170 /* If we failed here, we might as well return and have a leak
171 * rather than continue and a bugchk
172 */
173 if (ret != 0) {
174 netdev_err(ndev,
175 "unable to teardown receive buffer's gpadl\n");
176 return;
177 }
178 net_device->recv_buf_gpadl_handle = 0;
179 }
180
181 if (net_device->recv_buf) {
182 /* Free up the receive buffer */
183 vfree(net_device->recv_buf);
184 net_device->recv_buf = NULL;
185 }
186
187 if (net_device->recv_section) {
188 net_device->recv_section_cnt = 0;
189 kfree(net_device->recv_section);
190 net_device->recv_section = NULL;
191 }
192
193 /* Deal with the send buffer we may have setup.
194 * If we got a send section size, it means we received a
195 * NVSP_MSG1_TYPE_SEND_SEND_BUF_COMPLETE msg (ie sent
196 * NVSP_MSG1_TYPE_SEND_SEND_BUF msg) therefore, we need
197 * to send a revoke msg here
198 */
199 if (net_device->send_section_size) {
200 /* Send the revoke receive buffer */
201 revoke_packet = &net_device->revoke_packet;
202 memset(revoke_packet, 0, sizeof(struct nvsp_message));
203
204 revoke_packet->hdr.msg_type =
205 NVSP_MSG1_TYPE_REVOKE_SEND_BUF;
206 revoke_packet->msg.v1_msg.revoke_send_buf.id =
207 NETVSC_SEND_BUFFER_ID;
208
209 ret = vmbus_sendpacket(device->channel,
210 revoke_packet,
211 sizeof(struct nvsp_message),
212 (unsigned long)revoke_packet,
213 VM_PKT_DATA_INBAND, 0);
214 /* If we failed here, we might as well return and
215 * have a leak rather than continue and a bugchk
216 */
217 if (ret != 0) {
218 netdev_err(ndev, "unable to send "
219 "revoke send buffer to netvsp\n");
220 return;
221 }
222 }
223 /* Teardown the gpadl on the vsp end */
224 if (net_device->send_buf_gpadl_handle) {
225 ret = vmbus_teardown_gpadl(device->channel,
226 net_device->send_buf_gpadl_handle);
227
228 /* If we failed here, we might as well return and have a leak
229 * rather than continue and a bugchk
230 */
231 if (ret != 0) {
232 netdev_err(ndev,
233 "unable to teardown send buffer's gpadl\n");
234 return;
235 }
236 net_device->send_buf_gpadl_handle = 0;
237 }
238 if (net_device->send_buf) {
239 /* Free up the send buffer */
240 vfree(net_device->send_buf);
241 net_device->send_buf = NULL;
242 }
243 kfree(net_device->send_section_map);
244}
245
246static int netvsc_init_buf(struct hv_device *device)
247{
248 int ret = 0;
249 struct netvsc_device *net_device;
250 struct nvsp_message *init_packet;
251 struct net_device *ndev;
252 int node;
253
254 net_device = get_outbound_net_device(device);
255 if (!net_device)
256 return -ENODEV;
257 ndev = hv_get_drvdata(device);
258
259 node = cpu_to_node(device->channel->target_cpu);
260 net_device->recv_buf = vzalloc_node(net_device->recv_buf_size, node);
261 if (!net_device->recv_buf)
262 net_device->recv_buf = vzalloc(net_device->recv_buf_size);
263
264 if (!net_device->recv_buf) {
265 netdev_err(ndev, "unable to allocate receive "
266 "buffer of size %d\n", net_device->recv_buf_size);
267 ret = -ENOMEM;
268 goto cleanup;
269 }
270
271 /*
272 * Establish the gpadl handle for this buffer on this
273 * channel. Note: This call uses the vmbus connection rather
274 * than the channel to establish the gpadl handle.
275 */
276 ret = vmbus_establish_gpadl(device->channel, net_device->recv_buf,
277 net_device->recv_buf_size,
278 &net_device->recv_buf_gpadl_handle);
279 if (ret != 0) {
280 netdev_err(ndev,
281 "unable to establish receive buffer's gpadl\n");
282 goto cleanup;
283 }
284
285 /* Notify the NetVsp of the gpadl handle */
286 init_packet = &net_device->channel_init_pkt;
287
288 memset(init_packet, 0, sizeof(struct nvsp_message));
289
290 init_packet->hdr.msg_type = NVSP_MSG1_TYPE_SEND_RECV_BUF;
291 init_packet->msg.v1_msg.send_recv_buf.
292 gpadl_handle = net_device->recv_buf_gpadl_handle;
293 init_packet->msg.v1_msg.
294 send_recv_buf.id = NETVSC_RECEIVE_BUFFER_ID;
295
296 /* Send the gpadl notification request */
297 ret = vmbus_sendpacket(device->channel, init_packet,
298 sizeof(struct nvsp_message),
299 (unsigned long)init_packet,
300 VM_PKT_DATA_INBAND,
301 VMBUS_DATA_PACKET_FLAG_COMPLETION_REQUESTED);
302 if (ret != 0) {
303 netdev_err(ndev,
304 "unable to send receive buffer's gpadl to netvsp\n");
305 goto cleanup;
306 }
307
308 wait_for_completion(&net_device->channel_init_wait);
309
310 /* Check the response */
311 if (init_packet->msg.v1_msg.
312 send_recv_buf_complete.status != NVSP_STAT_SUCCESS) {
313 netdev_err(ndev, "Unable to complete receive buffer "
314 "initialization with NetVsp - status %d\n",
315 init_packet->msg.v1_msg.
316 send_recv_buf_complete.status);
317 ret = -EINVAL;
318 goto cleanup;
319 }
320
321 /* Parse the response */
322
323 net_device->recv_section_cnt = init_packet->msg.
324 v1_msg.send_recv_buf_complete.num_sections;
325
326 net_device->recv_section = kmemdup(
327 init_packet->msg.v1_msg.send_recv_buf_complete.sections,
328 net_device->recv_section_cnt *
329 sizeof(struct nvsp_1_receive_buffer_section),
330 GFP_KERNEL);
331 if (net_device->recv_section == NULL) {
332 ret = -EINVAL;
333 goto cleanup;
334 }
335
336 /*
337 * For 1st release, there should only be 1 section that represents the
338 * entire receive buffer
339 */
340 if (net_device->recv_section_cnt != 1 ||
341 net_device->recv_section->offset != 0) {
342 ret = -EINVAL;
343 goto cleanup;
344 }
345
346 /* Now setup the send buffer.
347 */
348 net_device->send_buf = vzalloc_node(net_device->send_buf_size, node);
349 if (!net_device->send_buf)
350 net_device->send_buf = vzalloc(net_device->send_buf_size);
351 if (!net_device->send_buf) {
352 netdev_err(ndev, "unable to allocate send "
353 "buffer of size %d\n", net_device->send_buf_size);
354 ret = -ENOMEM;
355 goto cleanup;
356 }
357
358 /* Establish the gpadl handle for this buffer on this
359 * channel. Note: This call uses the vmbus connection rather
360 * than the channel to establish the gpadl handle.
361 */
362 ret = vmbus_establish_gpadl(device->channel, net_device->send_buf,
363 net_device->send_buf_size,
364 &net_device->send_buf_gpadl_handle);
365 if (ret != 0) {
366 netdev_err(ndev,
367 "unable to establish send buffer's gpadl\n");
368 goto cleanup;
369 }
370
371 /* Notify the NetVsp of the gpadl handle */
372 init_packet = &net_device->channel_init_pkt;
373 memset(init_packet, 0, sizeof(struct nvsp_message));
374 init_packet->hdr.msg_type = NVSP_MSG1_TYPE_SEND_SEND_BUF;
375 init_packet->msg.v1_msg.send_send_buf.gpadl_handle =
376 net_device->send_buf_gpadl_handle;
377 init_packet->msg.v1_msg.send_send_buf.id = NETVSC_SEND_BUFFER_ID;
378
379 /* Send the gpadl notification request */
380 ret = vmbus_sendpacket(device->channel, init_packet,
381 sizeof(struct nvsp_message),
382 (unsigned long)init_packet,
383 VM_PKT_DATA_INBAND,
384 VMBUS_DATA_PACKET_FLAG_COMPLETION_REQUESTED);
385 if (ret != 0) {
386 netdev_err(ndev,
387 "unable to send send buffer's gpadl to netvsp\n");
388 goto cleanup;
389 }
390
391 wait_for_completion(&net_device->channel_init_wait);
392
393 /* Check the response */
394 if (init_packet->msg.v1_msg.
395 send_send_buf_complete.status != NVSP_STAT_SUCCESS) {
396 netdev_err(ndev, "Unable to complete send buffer "
397 "initialization with NetVsp - status %d\n",
398 init_packet->msg.v1_msg.
399 send_send_buf_complete.status);
400 ret = -EINVAL;
401 goto cleanup;
402 }
403
404 /* Parse the response */
405 net_device->send_section_size = init_packet->msg.
406 v1_msg.send_send_buf_complete.section_size;
407
408 /* Section count is simply the size divided by the section size.
409 */
410 net_device->send_section_cnt =
411 net_device->send_buf_size / net_device->send_section_size;
412
413 netdev_dbg(ndev, "Send section size: %d, Section count:%d\n",
414 net_device->send_section_size, net_device->send_section_cnt);
415
416 /* Setup state for managing the send buffer. */
417 net_device->map_words = DIV_ROUND_UP(net_device->send_section_cnt,
418 BITS_PER_LONG);
419
420 net_device->send_section_map = kcalloc(net_device->map_words,
421 sizeof(ulong), GFP_KERNEL);
422 if (net_device->send_section_map == NULL) {
423 ret = -ENOMEM;
424 goto cleanup;
425 }
426
427 goto exit;
428
429cleanup:
430 netvsc_destroy_buf(device);
431
432exit:
433 return ret;
434}
435
436/* Negotiate NVSP protocol version */
437static int negotiate_nvsp_ver(struct hv_device *device,
438 struct netvsc_device *net_device,
439 struct nvsp_message *init_packet,
440 u32 nvsp_ver)
441{
442 struct net_device *ndev = hv_get_drvdata(device);
443 int ret;
444
445 memset(init_packet, 0, sizeof(struct nvsp_message));
446 init_packet->hdr.msg_type = NVSP_MSG_TYPE_INIT;
447 init_packet->msg.init_msg.init.min_protocol_ver = nvsp_ver;
448 init_packet->msg.init_msg.init.max_protocol_ver = nvsp_ver;
449
450 /* Send the init request */
451 ret = vmbus_sendpacket(device->channel, init_packet,
452 sizeof(struct nvsp_message),
453 (unsigned long)init_packet,
454 VM_PKT_DATA_INBAND,
455 VMBUS_DATA_PACKET_FLAG_COMPLETION_REQUESTED);
456
457 if (ret != 0)
458 return ret;
459
460 wait_for_completion(&net_device->channel_init_wait);
461
462 if (init_packet->msg.init_msg.init_complete.status !=
463 NVSP_STAT_SUCCESS)
464 return -EINVAL;
465
466 if (nvsp_ver == NVSP_PROTOCOL_VERSION_1)
467 return 0;
468
469 /* NVSPv2 or later: Send NDIS config */
470 memset(init_packet, 0, sizeof(struct nvsp_message));
471 init_packet->hdr.msg_type = NVSP_MSG2_TYPE_SEND_NDIS_CONFIG;
472 init_packet->msg.v2_msg.send_ndis_config.mtu = ndev->mtu + ETH_HLEN;
473 init_packet->msg.v2_msg.send_ndis_config.capability.ieee8021q = 1;
474
475 if (nvsp_ver >= NVSP_PROTOCOL_VERSION_5) {
476 init_packet->msg.v2_msg.send_ndis_config.capability.sriov = 1;
477
478 /* Teaming bit is needed to receive link speed updates */
479 init_packet->msg.v2_msg.send_ndis_config.capability.teaming = 1;
480 }
481
482 ret = vmbus_sendpacket(device->channel, init_packet,
483 sizeof(struct nvsp_message),
484 (unsigned long)init_packet,
485 VM_PKT_DATA_INBAND, 0);
486
487 return ret;
488}
489
490static int netvsc_connect_vsp(struct hv_device *device)
491{
492 int ret;
493 struct netvsc_device *net_device;
494 struct nvsp_message *init_packet;
495 int ndis_version;
496 const u32 ver_list[] = {
497 NVSP_PROTOCOL_VERSION_1, NVSP_PROTOCOL_VERSION_2,
498 NVSP_PROTOCOL_VERSION_4, NVSP_PROTOCOL_VERSION_5 };
499 int i;
500
501 net_device = get_outbound_net_device(device);
502 if (!net_device)
503 return -ENODEV;
504
505 init_packet = &net_device->channel_init_pkt;
506
507 /* Negotiate the latest NVSP protocol supported */
508 for (i = ARRAY_SIZE(ver_list) - 1; i >= 0; i--)
509 if (negotiate_nvsp_ver(device, net_device, init_packet,
510 ver_list[i]) == 0) {
511 net_device->nvsp_version = ver_list[i];
512 break;
513 }
514
515 if (i < 0) {
516 ret = -EPROTO;
517 goto cleanup;
518 }
519
520 pr_debug("Negotiated NVSP version:%x\n", net_device->nvsp_version);
521
522 /* Send the ndis version */
523 memset(init_packet, 0, sizeof(struct nvsp_message));
524
525 if (net_device->nvsp_version <= NVSP_PROTOCOL_VERSION_4)
526 ndis_version = 0x00060001;
527 else
528 ndis_version = 0x0006001e;
529
530 init_packet->hdr.msg_type = NVSP_MSG1_TYPE_SEND_NDIS_VER;
531 init_packet->msg.v1_msg.
532 send_ndis_ver.ndis_major_ver =
533 (ndis_version & 0xFFFF0000) >> 16;
534 init_packet->msg.v1_msg.
535 send_ndis_ver.ndis_minor_ver =
536 ndis_version & 0xFFFF;
537
538 /* Send the init request */
539 ret = vmbus_sendpacket(device->channel, init_packet,
540 sizeof(struct nvsp_message),
541 (unsigned long)init_packet,
542 VM_PKT_DATA_INBAND, 0);
543 if (ret != 0)
544 goto cleanup;
545
546 /* Post the big receive buffer to NetVSP */
547 if (net_device->nvsp_version <= NVSP_PROTOCOL_VERSION_2)
548 net_device->recv_buf_size = NETVSC_RECEIVE_BUFFER_SIZE_LEGACY;
549 else
550 net_device->recv_buf_size = NETVSC_RECEIVE_BUFFER_SIZE;
551 net_device->send_buf_size = NETVSC_SEND_BUFFER_SIZE;
552
553 ret = netvsc_init_buf(device);
554
555cleanup:
556 return ret;
557}
558
559static void netvsc_disconnect_vsp(struct hv_device *device)
560{
561 netvsc_destroy_buf(device);
562}
563
564/*
565 * netvsc_device_remove - Callback when the root bus device is removed
566 */
567void netvsc_device_remove(struct hv_device *device)
568{
569 struct net_device *ndev = hv_get_drvdata(device);
570 struct net_device_context *net_device_ctx = netdev_priv(ndev);
571 struct netvsc_device *net_device = net_device_ctx->nvdev;
572
573 netvsc_disconnect_vsp(device);
574
575 net_device_ctx->nvdev = NULL;
576
577 /*
578 * At this point, no one should be accessing net_device
579 * except in here
580 */
581 netdev_dbg(ndev, "net device safe to remove\n");
582
583 /* Now, we can close the channel safely */
584 vmbus_close(device->channel);
585
586 /* Release all resources */
587 vfree(net_device->sub_cb_buf);
588 free_netvsc_device(net_device);
589}
590
591#define RING_AVAIL_PERCENT_HIWATER 20
592#define RING_AVAIL_PERCENT_LOWATER 10
593
594/*
595 * Get the percentage of available bytes to write in the ring.
596 * The return value is in range from 0 to 100.
597 */
598static inline u32 hv_ringbuf_avail_percent(
599 struct hv_ring_buffer_info *ring_info)
600{
601 u32 avail_read, avail_write;
602
603 hv_get_ringbuffer_availbytes(ring_info, &avail_read, &avail_write);
604
605 return avail_write * 100 / ring_info->ring_datasize;
606}
607
608static inline void netvsc_free_send_slot(struct netvsc_device *net_device,
609 u32 index)
610{
611 sync_change_bit(index, net_device->send_section_map);
612}
613
614static void netvsc_send_tx_complete(struct netvsc_device *net_device,
615 struct vmbus_channel *incoming_channel,
616 struct hv_device *device,
617 struct vmpacket_descriptor *packet)
618{
619 struct sk_buff *skb = (struct sk_buff *)(unsigned long)packet->trans_id;
620 struct net_device *ndev = hv_get_drvdata(device);
621 struct net_device_context *net_device_ctx = netdev_priv(ndev);
622 struct vmbus_channel *channel = device->channel;
623 int num_outstanding_sends;
624 u16 q_idx = 0;
625 int queue_sends;
626
627 /* Notify the layer above us */
628 if (likely(skb)) {
629 struct hv_netvsc_packet *nvsc_packet
630 = (struct hv_netvsc_packet *)skb->cb;
631 u32 send_index = nvsc_packet->send_buf_index;
632
633 if (send_index != NETVSC_INVALID_INDEX)
634 netvsc_free_send_slot(net_device, send_index);
635 q_idx = nvsc_packet->q_idx;
636 channel = incoming_channel;
637
638 dev_consume_skb_any(skb);
639 }
640
641 num_outstanding_sends =
642 atomic_dec_return(&net_device->num_outstanding_sends);
643 queue_sends = atomic_dec_return(&net_device->queue_sends[q_idx]);
644
645 if (net_device->destroy && num_outstanding_sends == 0)
646 wake_up(&net_device->wait_drain);
647
648 if (netif_tx_queue_stopped(netdev_get_tx_queue(ndev, q_idx)) &&
649 !net_device_ctx->start_remove &&
650 (hv_ringbuf_avail_percent(&channel->outbound) > RING_AVAIL_PERCENT_HIWATER ||
651 queue_sends < 1))
652 netif_tx_wake_queue(netdev_get_tx_queue(ndev, q_idx));
653}
654
655static void netvsc_send_completion(struct netvsc_device *net_device,
656 struct vmbus_channel *incoming_channel,
657 struct hv_device *device,
658 struct vmpacket_descriptor *packet)
659{
660 struct nvsp_message *nvsp_packet;
661 struct net_device *ndev = hv_get_drvdata(device);
662
663 nvsp_packet = (struct nvsp_message *)((unsigned long)packet +
664 (packet->offset8 << 3));
665
666 switch (nvsp_packet->hdr.msg_type) {
667 case NVSP_MSG_TYPE_INIT_COMPLETE:
668 case NVSP_MSG1_TYPE_SEND_RECV_BUF_COMPLETE:
669 case NVSP_MSG1_TYPE_SEND_SEND_BUF_COMPLETE:
670 case NVSP_MSG5_TYPE_SUBCHANNEL:
671 /* Copy the response back */
672 memcpy(&net_device->channel_init_pkt, nvsp_packet,
673 sizeof(struct nvsp_message));
674 complete(&net_device->channel_init_wait);
675 break;
676
677 case NVSP_MSG1_TYPE_SEND_RNDIS_PKT_COMPLETE:
678 netvsc_send_tx_complete(net_device, incoming_channel,
679 device, packet);
680 break;
681
682 default:
683 netdev_err(ndev,
684 "Unknown send completion type %d received!!\n",
685 nvsp_packet->hdr.msg_type);
686 }
687}
688
689static u32 netvsc_get_next_send_section(struct netvsc_device *net_device)
690{
691 unsigned long index;
692 u32 max_words = net_device->map_words;
693 unsigned long *map_addr = (unsigned long *)net_device->send_section_map;
694 u32 section_cnt = net_device->send_section_cnt;
695 int ret_val = NETVSC_INVALID_INDEX;
696 int i;
697 int prev_val;
698
699 for (i = 0; i < max_words; i++) {
700 if (!~(map_addr[i]))
701 continue;
702 index = ffz(map_addr[i]);
703 prev_val = sync_test_and_set_bit(index, &map_addr[i]);
704 if (prev_val)
705 continue;
706 if ((index + (i * BITS_PER_LONG)) >= section_cnt)
707 break;
708 ret_val = (index + (i * BITS_PER_LONG));
709 break;
710 }
711 return ret_val;
712}
713
714static u32 netvsc_copy_to_send_buf(struct netvsc_device *net_device,
715 unsigned int section_index,
716 u32 pend_size,
717 struct hv_netvsc_packet *packet,
718 struct rndis_message *rndis_msg,
719 struct hv_page_buffer **pb,
720 struct sk_buff *skb)
721{
722 char *start = net_device->send_buf;
723 char *dest = start + (section_index * net_device->send_section_size)
724 + pend_size;
725 int i;
726 bool is_data_pkt = (skb != NULL) ? true : false;
727 bool xmit_more = (skb != NULL) ? skb->xmit_more : false;
728 u32 msg_size = 0;
729 u32 padding = 0;
730 u32 remain = packet->total_data_buflen % net_device->pkt_align;
731 u32 page_count = packet->cp_partial ? packet->rmsg_pgcnt :
732 packet->page_buf_cnt;
733
734 /* Add padding */
735 if (is_data_pkt && xmit_more && remain &&
736 !packet->cp_partial) {
737 padding = net_device->pkt_align - remain;
738 rndis_msg->msg_len += padding;
739 packet->total_data_buflen += padding;
740 }
741
742 for (i = 0; i < page_count; i++) {
743 char *src = phys_to_virt((*pb)[i].pfn << PAGE_SHIFT);
744 u32 offset = (*pb)[i].offset;
745 u32 len = (*pb)[i].len;
746
747 memcpy(dest, (src + offset), len);
748 msg_size += len;
749 dest += len;
750 }
751
752 if (padding) {
753 memset(dest, 0, padding);
754 msg_size += padding;
755 }
756
757 return msg_size;
758}
759
760static inline int netvsc_send_pkt(
761 struct hv_device *device,
762 struct hv_netvsc_packet *packet,
763 struct netvsc_device *net_device,
764 struct hv_page_buffer **pb,
765 struct sk_buff *skb)
766{
767 struct nvsp_message nvmsg;
768 u16 q_idx = packet->q_idx;
769 struct vmbus_channel *out_channel = net_device->chn_table[q_idx];
770 struct net_device *ndev = hv_get_drvdata(device);
771 u64 req_id;
772 int ret;
773 struct hv_page_buffer *pgbuf;
774 u32 ring_avail = hv_ringbuf_avail_percent(&out_channel->outbound);
775 bool xmit_more = (skb != NULL) ? skb->xmit_more : false;
776
777 nvmsg.hdr.msg_type = NVSP_MSG1_TYPE_SEND_RNDIS_PKT;
778 if (skb != NULL) {
779 /* 0 is RMC_DATA; */
780 nvmsg.msg.v1_msg.send_rndis_pkt.channel_type = 0;
781 } else {
782 /* 1 is RMC_CONTROL; */
783 nvmsg.msg.v1_msg.send_rndis_pkt.channel_type = 1;
784 }
785
786 nvmsg.msg.v1_msg.send_rndis_pkt.send_buf_section_index =
787 packet->send_buf_index;
788 if (packet->send_buf_index == NETVSC_INVALID_INDEX)
789 nvmsg.msg.v1_msg.send_rndis_pkt.send_buf_section_size = 0;
790 else
791 nvmsg.msg.v1_msg.send_rndis_pkt.send_buf_section_size =
792 packet->total_data_buflen;
793
794 req_id = (ulong)skb;
795
796 if (out_channel->rescind)
797 return -ENODEV;
798
799 /*
800 * It is possible that once we successfully place this packet
801 * on the ringbuffer, we may stop the queue. In that case, we want
802 * to notify the host independent of the xmit_more flag. We don't
803 * need to be precise here; in the worst case we may signal the host
804 * unnecessarily.
805 */
806 if (ring_avail < (RING_AVAIL_PERCENT_LOWATER + 1))
807 xmit_more = false;
808
809 if (packet->page_buf_cnt) {
810 pgbuf = packet->cp_partial ? (*pb) +
811 packet->rmsg_pgcnt : (*pb);
812 ret = vmbus_sendpacket_pagebuffer_ctl(out_channel,
813 pgbuf,
814 packet->page_buf_cnt,
815 &nvmsg,
816 sizeof(struct nvsp_message),
817 req_id,
818 VMBUS_DATA_PACKET_FLAG_COMPLETION_REQUESTED,
819 !xmit_more);
820 } else {
821 ret = vmbus_sendpacket_ctl(out_channel, &nvmsg,
822 sizeof(struct nvsp_message),
823 req_id,
824 VM_PKT_DATA_INBAND,
825 VMBUS_DATA_PACKET_FLAG_COMPLETION_REQUESTED,
826 !xmit_more);
827 }
828
829 if (ret == 0) {
830 atomic_inc(&net_device->num_outstanding_sends);
831 atomic_inc(&net_device->queue_sends[q_idx]);
832
833 if (ring_avail < RING_AVAIL_PERCENT_LOWATER) {
834 netif_tx_stop_queue(netdev_get_tx_queue(ndev, q_idx));
835
836 if (atomic_read(&net_device->
837 queue_sends[q_idx]) < 1)
838 netif_tx_wake_queue(netdev_get_tx_queue(
839 ndev, q_idx));
840 }
841 } else if (ret == -EAGAIN) {
842 netif_tx_stop_queue(netdev_get_tx_queue(
843 ndev, q_idx));
844 if (atomic_read(&net_device->queue_sends[q_idx]) < 1) {
845 netif_tx_wake_queue(netdev_get_tx_queue(
846 ndev, q_idx));
847 ret = -ENOSPC;
848 }
849 } else {
850 netdev_err(ndev, "Unable to send packet %p ret %d\n",
851 packet, ret);
852 }
853
854 return ret;
855}
856
857/* Move packet out of multi send data (msd), and clear msd */
858static inline void move_pkt_msd(struct hv_netvsc_packet **msd_send,
859 struct sk_buff **msd_skb,
860 struct multi_send_data *msdp)
861{
862 *msd_skb = msdp->skb;
863 *msd_send = msdp->pkt;
864 msdp->skb = NULL;
865 msdp->pkt = NULL;
866 msdp->count = 0;
867}
868
869int netvsc_send(struct hv_device *device,
870 struct hv_netvsc_packet *packet,
871 struct rndis_message *rndis_msg,
872 struct hv_page_buffer **pb,
873 struct sk_buff *skb)
874{
875 struct netvsc_device *net_device;
876 int ret = 0;
877 struct vmbus_channel *out_channel;
878 u16 q_idx = packet->q_idx;
879 u32 pktlen = packet->total_data_buflen, msd_len = 0;
880 unsigned int section_index = NETVSC_INVALID_INDEX;
881 struct multi_send_data *msdp;
882 struct hv_netvsc_packet *msd_send = NULL, *cur_send = NULL;
883 struct sk_buff *msd_skb = NULL;
884 bool try_batch;
885 bool xmit_more = (skb != NULL) ? skb->xmit_more : false;
886
887 net_device = get_outbound_net_device(device);
888 if (!net_device)
889 return -ENODEV;
890
891 /* We may race with netvsc_connect_vsp()/netvsc_init_buf() and get
892 * here before the negotiation with the host is finished and
893 * send_section_map may not be allocated yet.
894 */
895 if (!net_device->send_section_map)
896 return -EAGAIN;
897
898 out_channel = net_device->chn_table[q_idx];
899
900 packet->send_buf_index = NETVSC_INVALID_INDEX;
901 packet->cp_partial = false;
902
903 /* Send control message directly without accessing msd (Multi-Send
904 * Data) field which may be changed during data packet processing.
905 */
906 if (!skb) {
907 cur_send = packet;
908 goto send_now;
909 }
910
911 msdp = &net_device->msd[q_idx];
912
913 /* batch packets in send buffer if possible */
914 if (msdp->pkt)
915 msd_len = msdp->pkt->total_data_buflen;
916
917 try_batch = (skb != NULL) && msd_len > 0 && msdp->count <
918 net_device->max_pkt;
919
920 if (try_batch && msd_len + pktlen + net_device->pkt_align <
921 net_device->send_section_size) {
922 section_index = msdp->pkt->send_buf_index;
923
924 } else if (try_batch && msd_len + packet->rmsg_size <
925 net_device->send_section_size) {
926 section_index = msdp->pkt->send_buf_index;
927 packet->cp_partial = true;
928
929 } else if ((skb != NULL) && pktlen + net_device->pkt_align <
930 net_device->send_section_size) {
931 section_index = netvsc_get_next_send_section(net_device);
932 if (section_index != NETVSC_INVALID_INDEX) {
933 move_pkt_msd(&msd_send, &msd_skb, msdp);
934 msd_len = 0;
935 }
936 }
937
938 if (section_index != NETVSC_INVALID_INDEX) {
939 netvsc_copy_to_send_buf(net_device,
940 section_index, msd_len,
941 packet, rndis_msg, pb, skb);
942
943 packet->send_buf_index = section_index;
944
945 if (packet->cp_partial) {
946 packet->page_buf_cnt -= packet->rmsg_pgcnt;
947 packet->total_data_buflen = msd_len + packet->rmsg_size;
948 } else {
949 packet->page_buf_cnt = 0;
950 packet->total_data_buflen += msd_len;
951 }
952
953 if (msdp->skb)
954 dev_consume_skb_any(msdp->skb);
955
956 if (xmit_more && !packet->cp_partial) {
957 msdp->skb = skb;
958 msdp->pkt = packet;
959 msdp->count++;
960 } else {
961 cur_send = packet;
962 msdp->skb = NULL;
963 msdp->pkt = NULL;
964 msdp->count = 0;
965 }
966 } else {
967 move_pkt_msd(&msd_send, &msd_skb, msdp);
968 cur_send = packet;
969 }
970
971 if (msd_send) {
972 int m_ret = netvsc_send_pkt(device, msd_send, net_device,
973 NULL, msd_skb);
974
975 if (m_ret != 0) {
976 netvsc_free_send_slot(net_device,
977 msd_send->send_buf_index);
978 dev_kfree_skb_any(msd_skb);
979 }
980 }
981
982send_now:
983 if (cur_send)
984 ret = netvsc_send_pkt(device, cur_send, net_device, pb, skb);
985
986 if (ret != 0 && section_index != NETVSC_INVALID_INDEX)
987 netvsc_free_send_slot(net_device, section_index);
988
989 return ret;
990}
991
992static int netvsc_send_recv_completion(struct vmbus_channel *channel,
993 u64 transaction_id, u32 status)
994{
995 struct nvsp_message recvcompMessage;
996 int ret;
997
998 recvcompMessage.hdr.msg_type =
999 NVSP_MSG1_TYPE_SEND_RNDIS_PKT_COMPLETE;
1000
1001 recvcompMessage.msg.v1_msg.send_rndis_pkt_complete.status = status;
1002
1003 /* Send the completion */
1004 ret = vmbus_sendpacket(channel, &recvcompMessage,
1005 sizeof(struct nvsp_message_header) + sizeof(u32),
1006 transaction_id, VM_PKT_COMP, 0);
1007
1008 return ret;
1009}
1010
1011static inline void count_recv_comp_slot(struct netvsc_device *nvdev, u16 q_idx,
1012 u32 *filled, u32 *avail)
1013{
1014 u32 first = nvdev->mrc[q_idx].first;
1015 u32 next = nvdev->mrc[q_idx].next;
1016
1017 *filled = (first > next) ? NETVSC_RECVSLOT_MAX - first + next :
1018 next - first;
1019
1020 *avail = NETVSC_RECVSLOT_MAX - *filled - 1;
1021}
1022
1023/* Read the first filled slot, no change to index */
1024static inline struct recv_comp_data *read_recv_comp_slot(struct netvsc_device
1025 *nvdev, u16 q_idx)
1026{
1027 u32 filled, avail;
1028
1029 if (!nvdev->mrc[q_idx].buf)
1030 return NULL;
1031
1032 count_recv_comp_slot(nvdev, q_idx, &filled, &avail);
1033 if (!filled)
1034 return NULL;
1035
1036 return nvdev->mrc[q_idx].buf + nvdev->mrc[q_idx].first *
1037 sizeof(struct recv_comp_data);
1038}
1039
1040/* Put the first filled slot back to available pool */
1041static inline void put_recv_comp_slot(struct netvsc_device *nvdev, u16 q_idx)
1042{
1043 int num_recv;
1044
1045 nvdev->mrc[q_idx].first = (nvdev->mrc[q_idx].first + 1) %
1046 NETVSC_RECVSLOT_MAX;
1047
1048 num_recv = atomic_dec_return(&nvdev->num_outstanding_recvs);
1049
1050 if (nvdev->destroy && num_recv == 0)
1051 wake_up(&nvdev->wait_drain);
1052}
1053
1054/* Check and send pending recv completions */
1055static void netvsc_chk_recv_comp(struct netvsc_device *nvdev,
1056 struct vmbus_channel *channel, u16 q_idx)
1057{
1058 struct recv_comp_data *rcd;
1059 int ret;
1060
1061 while (true) {
1062 rcd = read_recv_comp_slot(nvdev, q_idx);
1063 if (!rcd)
1064 break;
1065
1066 ret = netvsc_send_recv_completion(channel, rcd->tid,
1067 rcd->status);
1068 if (ret)
1069 break;
1070
1071 put_recv_comp_slot(nvdev, q_idx);
1072 }
1073}
1074
1075#define NETVSC_RCD_WATERMARK 80
1076
1077/* Get next available slot */
1078static inline struct recv_comp_data *get_recv_comp_slot(
1079 struct netvsc_device *nvdev, struct vmbus_channel *channel, u16 q_idx)
1080{
1081 u32 filled, avail, next;
1082 struct recv_comp_data *rcd;
1083
1084 if (!nvdev->recv_section)
1085 return NULL;
1086
1087 if (!nvdev->mrc[q_idx].buf)
1088 return NULL;
1089
1090 if (atomic_read(&nvdev->num_outstanding_recvs) >
1091 nvdev->recv_section->num_sub_allocs * NETVSC_RCD_WATERMARK / 100)
1092 netvsc_chk_recv_comp(nvdev, channel, q_idx);
1093
1094 count_recv_comp_slot(nvdev, q_idx, &filled, &avail);
1095 if (!avail)
1096 return NULL;
1097
1098 next = nvdev->mrc[q_idx].next;
1099 rcd = nvdev->mrc[q_idx].buf + next * sizeof(struct recv_comp_data);
1100 nvdev->mrc[q_idx].next = (next + 1) % NETVSC_RECVSLOT_MAX;
1101
1102 atomic_inc(&nvdev->num_outstanding_recvs);
1103
1104 return rcd;
1105}
1106
1107static void netvsc_receive(struct netvsc_device *net_device,
1108 struct vmbus_channel *channel,
1109 struct hv_device *device,
1110 struct vmpacket_descriptor *packet)
1111{
1112 struct vmtransfer_page_packet_header *vmxferpage_packet;
1113 struct nvsp_message *nvsp_packet;
1114 struct hv_netvsc_packet nv_pkt;
1115 struct hv_netvsc_packet *netvsc_packet = &nv_pkt;
1116 u32 status = NVSP_STAT_SUCCESS;
1117 int i;
1118 int count = 0;
1119 struct net_device *ndev = hv_get_drvdata(device);
1120 void *data;
1121 int ret;
1122 struct recv_comp_data *rcd;
1123 u16 q_idx = channel->offermsg.offer.sub_channel_index;
1124
1125 /*
1126 * All inbound packets other than send completion should be xfer page
1127 * packet
1128 */
1129 if (packet->type != VM_PKT_DATA_USING_XFER_PAGES) {
1130 netdev_err(ndev, "Unknown packet type received - %d\n",
1131 packet->type);
1132 return;
1133 }
1134
1135 nvsp_packet = (struct nvsp_message *)((unsigned long)packet +
1136 (packet->offset8 << 3));
1137
1138 /* Make sure this is a valid nvsp packet */
1139 if (nvsp_packet->hdr.msg_type !=
1140 NVSP_MSG1_TYPE_SEND_RNDIS_PKT) {
1141 netdev_err(ndev, "Unknown nvsp packet type received-"
1142 " %d\n", nvsp_packet->hdr.msg_type);
1143 return;
1144 }
1145
1146 vmxferpage_packet = (struct vmtransfer_page_packet_header *)packet;
1147
1148 if (vmxferpage_packet->xfer_pageset_id != NETVSC_RECEIVE_BUFFER_ID) {
1149 netdev_err(ndev, "Invalid xfer page set id - "
1150 "expecting %x got %x\n", NETVSC_RECEIVE_BUFFER_ID,
1151 vmxferpage_packet->xfer_pageset_id);
1152 return;
1153 }
1154
1155 count = vmxferpage_packet->range_cnt;
1156
1157 /* Each range represents 1 RNDIS pkt that contains 1 ethernet frame */
1158 for (i = 0; i < count; i++) {
1159 /* Initialize the netvsc packet */
1160 data = (void *)((unsigned long)net_device->
1161 recv_buf + vmxferpage_packet->ranges[i].byte_offset);
1162 netvsc_packet->total_data_buflen =
1163 vmxferpage_packet->ranges[i].byte_count;
1164
1165 /* Pass it to the upper layer */
1166 status = rndis_filter_receive(device, netvsc_packet, &data,
1167 channel);
1168 }
1169
1170 if (!net_device->mrc[q_idx].buf) {
1171 ret = netvsc_send_recv_completion(channel,
1172 vmxferpage_packet->d.trans_id,
1173 status);
1174 if (ret)
1175 netdev_err(ndev, "Recv_comp q:%hd, tid:%llx, err:%d\n",
1176 q_idx, vmxferpage_packet->d.trans_id, ret);
1177 return;
1178 }
1179
1180 rcd = get_recv_comp_slot(net_device, channel, q_idx);
1181
1182 if (!rcd) {
1183 netdev_err(ndev, "Recv_comp full buf q:%hd, tid:%llx\n",
1184 q_idx, vmxferpage_packet->d.trans_id);
1185 return;
1186 }
1187
1188 rcd->tid = vmxferpage_packet->d.trans_id;
1189 rcd->status = status;
1190}
1191
1192static void netvsc_send_table(struct hv_device *hdev,
1193 struct nvsp_message *nvmsg)
1194{
1195 struct netvsc_device *nvscdev;
1196 struct net_device *ndev = hv_get_drvdata(hdev);
1197 int i;
1198 u32 count, *tab;
1199
1200 nvscdev = get_outbound_net_device(hdev);
1201 if (!nvscdev)
1202 return;
1203
1204 count = nvmsg->msg.v5_msg.send_table.count;
1205 if (count != VRSS_SEND_TAB_SIZE) {
1206 netdev_err(ndev, "Received wrong send-table size:%u\n", count);
1207 return;
1208 }
1209
1210 tab = (u32 *)((unsigned long)&nvmsg->msg.v5_msg.send_table +
1211 nvmsg->msg.v5_msg.send_table.offset);
1212
1213 for (i = 0; i < count; i++)
1214 nvscdev->send_table[i] = tab[i];
1215}
1216
1217static void netvsc_send_vf(struct net_device_context *net_device_ctx,
1218 struct nvsp_message *nvmsg)
1219{
1220 net_device_ctx->vf_alloc = nvmsg->msg.v4_msg.vf_assoc.allocated;
1221 net_device_ctx->vf_serial = nvmsg->msg.v4_msg.vf_assoc.serial;
1222}
1223
1224static inline void netvsc_receive_inband(struct hv_device *hdev,
1225 struct net_device_context *net_device_ctx,
1226 struct nvsp_message *nvmsg)
1227{
1228 switch (nvmsg->hdr.msg_type) {
1229 case NVSP_MSG5_TYPE_SEND_INDIRECTION_TABLE:
1230 netvsc_send_table(hdev, nvmsg);
1231 break;
1232
1233 case NVSP_MSG4_TYPE_SEND_VF_ASSOCIATION:
1234 netvsc_send_vf(net_device_ctx, nvmsg);
1235 break;
1236 }
1237}
1238
1239static void netvsc_process_raw_pkt(struct hv_device *device,
1240 struct vmbus_channel *channel,
1241 struct netvsc_device *net_device,
1242 struct net_device *ndev,
1243 u64 request_id,
1244 struct vmpacket_descriptor *desc)
1245{
1246 struct nvsp_message *nvmsg;
1247 struct net_device_context *net_device_ctx = netdev_priv(ndev);
1248
1249 nvmsg = (struct nvsp_message *)((unsigned long)
1250 desc + (desc->offset8 << 3));
1251
1252 switch (desc->type) {
1253 case VM_PKT_COMP:
1254 netvsc_send_completion(net_device, channel, device, desc);
1255 break;
1256
1257 case VM_PKT_DATA_USING_XFER_PAGES:
1258 netvsc_receive(net_device, channel, device, desc);
1259 break;
1260
1261 case VM_PKT_DATA_INBAND:
1262 netvsc_receive_inband(device, net_device_ctx, nvmsg);
1263 break;
1264
1265 default:
1266 netdev_err(ndev, "unhandled packet type %d, tid %llx\n",
1267 desc->type, request_id);
1268 break;
1269 }
1270}
1271
1272void netvsc_channel_cb(void *context)
1273{
1274 int ret;
1275 struct vmbus_channel *channel = (struct vmbus_channel *)context;
1276 u16 q_idx = channel->offermsg.offer.sub_channel_index;
1277 struct hv_device *device;
1278 struct netvsc_device *net_device;
1279 u32 bytes_recvd;
1280 u64 request_id;
1281 struct vmpacket_descriptor *desc;
1282 unsigned char *buffer;
1283 int bufferlen = NETVSC_PACKET_SIZE;
1284 struct net_device *ndev;
1285 bool need_to_commit = false;
1286
1287 if (channel->primary_channel != NULL)
1288 device = channel->primary_channel->device_obj;
1289 else
1290 device = channel->device_obj;
1291
1292 net_device = get_inbound_net_device(device);
1293 if (!net_device)
1294 return;
1295 ndev = hv_get_drvdata(device);
1296 buffer = get_per_channel_state(channel);
1297
1298 /* commit_rd_index() -> hv_signal_on_read() needs this. */
1299 init_cached_read_index(channel);
1300
1301 do {
1302 desc = get_next_pkt_raw(channel);
1303 if (desc != NULL) {
1304 netvsc_process_raw_pkt(device,
1305 channel,
1306 net_device,
1307 ndev,
1308 desc->trans_id,
1309 desc);
1310
1311 put_pkt_raw(channel, desc);
1312 need_to_commit = true;
1313 continue;
1314 }
1315 if (need_to_commit) {
1316 need_to_commit = false;
1317 commit_rd_index(channel);
1318 }
1319
1320 ret = vmbus_recvpacket_raw(channel, buffer, bufferlen,
1321 &bytes_recvd, &request_id);
1322 if (ret == 0) {
1323 if (bytes_recvd > 0) {
1324 desc = (struct vmpacket_descriptor *)buffer;
1325 netvsc_process_raw_pkt(device,
1326 channel,
1327 net_device,
1328 ndev,
1329 request_id,
1330 desc);
1331 } else {
1332 /*
1333 * We are done for this pass.
1334 */
1335 break;
1336 }
1337
1338 } else if (ret == -ENOBUFS) {
1339 if (bufferlen > NETVSC_PACKET_SIZE)
1340 kfree(buffer);
1341 /* Handle large packet */
1342 buffer = kmalloc(bytes_recvd, GFP_ATOMIC);
1343 if (buffer == NULL) {
1344 /* Try again next time around */
1345 netdev_err(ndev,
1346 "unable to allocate buffer of size "
1347 "(%d)!!\n", bytes_recvd);
1348 break;
1349 }
1350
1351 bufferlen = bytes_recvd;
1352 }
1353
1354 init_cached_read_index(channel);
1355
1356 } while (1);
1357
1358 if (bufferlen > NETVSC_PACKET_SIZE)
1359 kfree(buffer);
1360
1361 netvsc_chk_recv_comp(net_device, channel, q_idx);
1362}
1363
1364/*
1365 * netvsc_device_add - Callback when the device belonging to this
1366 * driver is added
1367 */
1368int netvsc_device_add(struct hv_device *device, void *additional_info)
1369{
1370 int i, ret = 0;
1371 int ring_size =
1372 ((struct netvsc_device_info *)additional_info)->ring_size;
1373 struct netvsc_device *net_device;
1374 struct net_device *ndev = hv_get_drvdata(device);
1375 struct net_device_context *net_device_ctx = netdev_priv(ndev);
1376
1377 net_device = alloc_net_device();
1378 if (!net_device)
1379 return -ENOMEM;
1380
1381 net_device->ring_size = ring_size;
1382
1383 set_per_channel_state(device->channel, net_device->cb_buffer);
1384
1385 /* Open the channel */
1386 ret = vmbus_open(device->channel, ring_size * PAGE_SIZE,
1387 ring_size * PAGE_SIZE, NULL, 0,
1388 netvsc_channel_cb, device->channel);
1389
1390 if (ret != 0) {
1391 netdev_err(ndev, "unable to open channel: %d\n", ret);
1392 goto cleanup;
1393 }
1394
1395 /* Channel is opened */
1396 netdev_dbg(ndev, "hv_netvsc channel opened successfully\n");
1397
1398 /* If we're reopening the device we may have multiple queues, fill the
1399 * chn_table with the default channel to use it before subchannels are
1400 * opened.
1401 */
1402 for (i = 0; i < VRSS_CHANNEL_MAX; i++)
1403 net_device->chn_table[i] = device->channel;
1404
1405 /* Writing nvdev pointer unlocks netvsc_send(), make sure chn_table is
1406 * populated.
1407 */
1408 wmb();
1409
1410 net_device_ctx->nvdev = net_device;
1411
1412 /* Connect with the NetVsp */
1413 ret = netvsc_connect_vsp(device);
1414 if (ret != 0) {
1415 netdev_err(ndev,
1416 "unable to connect to NetVSP - %d\n", ret);
1417 goto close;
1418 }
1419
1420 return ret;
1421
1422close:
1423 /* Now, we can close the channel safely */
1424 vmbus_close(device->channel);
1425
1426cleanup:
1427 free_netvsc_device(net_device);
1428
1429 return ret;
1430}