b6ac152a9bd00df1a548dace4f652b466956e7ff
[linux-2.6.git] / drivers / net / hyperv / netvsc.c
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
32 #include "hyperv_net.h"
33
34
35 static 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
45         net_device->destroy = false;
46         net_device->dev = device;
47         net_device->ndev = ndev;
48
49         hv_set_drvdata(device, net_device);
50         return net_device;
51 }
52
53 static struct netvsc_device *get_outbound_net_device(struct hv_device *device)
54 {
55         struct netvsc_device *net_device;
56
57         net_device = hv_get_drvdata(device);
58         if (net_device && net_device->destroy)
59                 net_device = NULL;
60
61         return net_device;
62 }
63
64 static struct netvsc_device *get_inbound_net_device(struct hv_device *device)
65 {
66         struct netvsc_device *net_device;
67
68         net_device = hv_get_drvdata(device);
69
70         if (!net_device)
71                 goto get_in_err;
72
73         if (net_device->destroy &&
74                 atomic_read(&net_device->num_outstanding_sends) == 0)
75                 net_device = NULL;
76
77 get_in_err:
78         return net_device;
79 }
80
81
82 static int netvsc_destroy_recv_buf(struct netvsc_device *net_device)
83 {
84         struct nvsp_message *revoke_packet;
85         int ret = 0;
86         struct net_device *ndev = net_device->ndev;
87
88         /*
89          * If we got a section count, it means we received a
90          * SendReceiveBufferComplete msg (ie sent
91          * NvspMessage1TypeSendReceiveBuffer msg) therefore, we need
92          * to send a revoke msg here
93          */
94         if (net_device->recv_section_cnt) {
95                 /* Send the revoke receive buffer */
96                 revoke_packet = &net_device->revoke_packet;
97                 memset(revoke_packet, 0, sizeof(struct nvsp_message));
98
99                 revoke_packet->hdr.msg_type =
100                         NVSP_MSG1_TYPE_REVOKE_RECV_BUF;
101                 revoke_packet->msg.v1_msg.
102                 revoke_recv_buf.id = NETVSC_RECEIVE_BUFFER_ID;
103
104                 ret = vmbus_sendpacket(net_device->dev->channel,
105                                        revoke_packet,
106                                        sizeof(struct nvsp_message),
107                                        (unsigned long)revoke_packet,
108                                        VM_PKT_DATA_INBAND, 0);
109                 /*
110                  * If we failed here, we might as well return and
111                  * have a leak rather than continue and a bugchk
112                  */
113                 if (ret != 0) {
114                         netdev_err(ndev, "unable to send "
115                                 "revoke receive buffer to netvsp\n");
116                         return ret;
117                 }
118         }
119
120         /* Teardown the gpadl on the vsp end */
121         if (net_device->recv_buf_gpadl_handle) {
122                 ret = vmbus_teardown_gpadl(net_device->dev->channel,
123                            net_device->recv_buf_gpadl_handle);
124
125                 /* If we failed here, we might as well return and have a leak
126                  * rather than continue and a bugchk
127                  */
128                 if (ret != 0) {
129                         netdev_err(ndev,
130                                    "unable to teardown receive buffer's gpadl\n");
131                         return ret;
132                 }
133                 net_device->recv_buf_gpadl_handle = 0;
134         }
135
136         if (net_device->recv_buf) {
137                 /* Free up the receive buffer */
138                 free_pages((unsigned long)net_device->recv_buf,
139                         get_order(net_device->recv_buf_size));
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
152 static 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 =
166                 (void *)__get_free_pages(GFP_KERNEL|__GFP_ZERO,
167                                 get_order(net_device->recv_buf_size));
168         if (!net_device->recv_buf) {
169                 netdev_err(ndev, "unable to allocate receive "
170                         "buffer of size %d\n", net_device->recv_buf_size);
171                 ret = -ENOMEM;
172                 goto cleanup;
173         }
174
175         /*
176          * Establish the gpadl handle for this buffer on this
177          * channel.  Note: This call uses the vmbus connection rather
178          * than the channel to establish the gpadl handle.
179          */
180         ret = vmbus_establish_gpadl(device->channel, net_device->recv_buf,
181                                     net_device->recv_buf_size,
182                                     &net_device->recv_buf_gpadl_handle);
183         if (ret != 0) {
184                 netdev_err(ndev,
185                         "unable to establish receive buffer's gpadl\n");
186                 goto cleanup;
187         }
188
189
190         /* Notify the NetVsp of the gpadl handle */
191         init_packet = &net_device->channel_init_pkt;
192
193         memset(init_packet, 0, sizeof(struct nvsp_message));
194
195         init_packet->hdr.msg_type = NVSP_MSG1_TYPE_SEND_RECV_BUF;
196         init_packet->msg.v1_msg.send_recv_buf.
197                 gpadl_handle = net_device->recv_buf_gpadl_handle;
198         init_packet->msg.v1_msg.
199                 send_recv_buf.id = NETVSC_RECEIVE_BUFFER_ID;
200
201         /* Send the gpadl notification request */
202         ret = vmbus_sendpacket(device->channel, init_packet,
203                                sizeof(struct nvsp_message),
204                                (unsigned long)init_packet,
205                                VM_PKT_DATA_INBAND,
206                                VMBUS_DATA_PACKET_FLAG_COMPLETION_REQUESTED);
207         if (ret != 0) {
208                 netdev_err(ndev,
209                         "unable to send receive buffer's gpadl to netvsp\n");
210                 goto cleanup;
211         }
212
213         t = wait_for_completion_timeout(&net_device->channel_init_wait, 5*HZ);
214         BUG_ON(t == 0);
215
216
217         /* Check the response */
218         if (init_packet->msg.v1_msg.
219             send_recv_buf_complete.status != NVSP_STAT_SUCCESS) {
220                 netdev_err(ndev, "Unable to complete receive buffer "
221                            "initialization with NetVsp - status %d\n",
222                            init_packet->msg.v1_msg.
223                            send_recv_buf_complete.status);
224                 ret = -EINVAL;
225                 goto cleanup;
226         }
227
228         /* Parse the response */
229
230         net_device->recv_section_cnt = init_packet->msg.
231                 v1_msg.send_recv_buf_complete.num_sections;
232
233         net_device->recv_section = kmemdup(
234                 init_packet->msg.v1_msg.send_recv_buf_complete.sections,
235                 net_device->recv_section_cnt *
236                 sizeof(struct nvsp_1_receive_buffer_section),
237                 GFP_KERNEL);
238         if (net_device->recv_section == NULL) {
239                 ret = -EINVAL;
240                 goto cleanup;
241         }
242
243         /*
244          * For 1st release, there should only be 1 section that represents the
245          * entire receive buffer
246          */
247         if (net_device->recv_section_cnt != 1 ||
248             net_device->recv_section->offset != 0) {
249                 ret = -EINVAL;
250                 goto cleanup;
251         }
252
253         goto exit;
254
255 cleanup:
256         netvsc_destroy_recv_buf(net_device);
257
258 exit:
259         return ret;
260 }
261
262
263 static int netvsc_connect_vsp(struct hv_device *device)
264 {
265         int ret, t;
266         struct netvsc_device *net_device;
267         struct nvsp_message *init_packet;
268         int ndis_version;
269         struct net_device *ndev;
270
271         net_device = get_outbound_net_device(device);
272         if (!net_device)
273                 return -ENODEV;
274         ndev = net_device->ndev;
275
276         init_packet = &net_device->channel_init_pkt;
277
278         memset(init_packet, 0, sizeof(struct nvsp_message));
279         init_packet->hdr.msg_type = NVSP_MSG_TYPE_INIT;
280         init_packet->msg.init_msg.init.min_protocol_ver =
281                 NVSP_MIN_PROTOCOL_VERSION;
282         init_packet->msg.init_msg.init.max_protocol_ver =
283                 NVSP_MAX_PROTOCOL_VERSION;
284
285         /* Send the init request */
286         ret = vmbus_sendpacket(device->channel, init_packet,
287                                sizeof(struct nvsp_message),
288                                (unsigned long)init_packet,
289                                VM_PKT_DATA_INBAND,
290                                VMBUS_DATA_PACKET_FLAG_COMPLETION_REQUESTED);
291
292         if (ret != 0)
293                 goto cleanup;
294
295         t = wait_for_completion_timeout(&net_device->channel_init_wait, 5*HZ);
296
297         if (t == 0) {
298                 ret = -ETIMEDOUT;
299                 goto cleanup;
300         }
301
302         if (init_packet->msg.init_msg.init_complete.status !=
303             NVSP_STAT_SUCCESS) {
304                 ret = -EINVAL;
305                 goto cleanup;
306         }
307
308         if (init_packet->msg.init_msg.init_complete.
309             negotiated_protocol_ver != NVSP_PROTOCOL_VERSION_1) {
310                 ret = -EPROTO;
311                 goto cleanup;
312         }
313         /* Send the ndis version */
314         memset(init_packet, 0, sizeof(struct nvsp_message));
315
316         ndis_version = 0x00050000;
317
318         init_packet->hdr.msg_type = NVSP_MSG1_TYPE_SEND_NDIS_VER;
319         init_packet->msg.v1_msg.
320                 send_ndis_ver.ndis_major_ver =
321                                 (ndis_version & 0xFFFF0000) >> 16;
322         init_packet->msg.v1_msg.
323                 send_ndis_ver.ndis_minor_ver =
324                                 ndis_version & 0xFFFF;
325
326         /* Send the init request */
327         ret = vmbus_sendpacket(device->channel, init_packet,
328                                 sizeof(struct nvsp_message),
329                                 (unsigned long)init_packet,
330                                 VM_PKT_DATA_INBAND, 0);
331         if (ret != 0)
332                 goto cleanup;
333
334         /* Post the big receive buffer to NetVSP */
335         ret = netvsc_init_recv_buf(device);
336
337 cleanup:
338         return ret;
339 }
340
341 static void netvsc_disconnect_vsp(struct netvsc_device *net_device)
342 {
343         netvsc_destroy_recv_buf(net_device);
344 }
345
346 /*
347  * netvsc_device_remove - Callback when the root bus device is removed
348  */
349 int netvsc_device_remove(struct hv_device *device)
350 {
351         struct netvsc_device *net_device;
352         struct hv_netvsc_packet *netvsc_packet, *pos;
353         unsigned long flags;
354
355         net_device = hv_get_drvdata(device);
356         spin_lock_irqsave(&device->channel->inbound_lock, flags);
357         net_device->destroy = true;
358         spin_unlock_irqrestore(&device->channel->inbound_lock, flags);
359
360         /* Wait for all send completions */
361         while (atomic_read(&net_device->num_outstanding_sends)) {
362                 dev_info(&device->device,
363                         "waiting for %d requests to complete...\n",
364                         atomic_read(&net_device->num_outstanding_sends));
365                 udelay(100);
366         }
367
368         netvsc_disconnect_vsp(net_device);
369
370         /*
371          * Since we have already drained, we don't need to busy wait
372          * as was done in final_release_stor_device()
373          * Note that we cannot set the ext pointer to NULL until
374          * we have drained - to drain the outgoing packets, we need to
375          * allow incoming packets.
376          */
377
378         spin_lock_irqsave(&device->channel->inbound_lock, flags);
379         hv_set_drvdata(device, NULL);
380         spin_unlock_irqrestore(&device->channel->inbound_lock, flags);
381
382         /*
383          * At this point, no one should be accessing net_device
384          * except in here
385          */
386         dev_notice(&device->device, "net device safe to remove\n");
387
388         /* Now, we can close the channel safely */
389         vmbus_close(device->channel);
390
391         /* Release all resources */
392         list_for_each_entry_safe(netvsc_packet, pos,
393                                  &net_device->recv_pkt_list, list_ent) {
394                 list_del(&netvsc_packet->list_ent);
395                 kfree(netvsc_packet);
396         }
397
398         kfree(net_device);
399         return 0;
400 }
401
402 static void netvsc_send_completion(struct hv_device *device,
403                                    struct vmpacket_descriptor *packet)
404 {
405         struct netvsc_device *net_device;
406         struct nvsp_message *nvsp_packet;
407         struct hv_netvsc_packet *nvsc_packet;
408         struct net_device *ndev;
409
410         net_device = get_inbound_net_device(device);
411         if (!net_device)
412                 return;
413         ndev = net_device->ndev;
414
415         nvsp_packet = (struct nvsp_message *)((unsigned long)packet +
416                         (packet->offset8 << 3));
417
418         if ((nvsp_packet->hdr.msg_type == NVSP_MSG_TYPE_INIT_COMPLETE) ||
419             (nvsp_packet->hdr.msg_type ==
420              NVSP_MSG1_TYPE_SEND_RECV_BUF_COMPLETE) ||
421             (nvsp_packet->hdr.msg_type ==
422              NVSP_MSG1_TYPE_SEND_SEND_BUF_COMPLETE)) {
423                 /* Copy the response back */
424                 memcpy(&net_device->channel_init_pkt, nvsp_packet,
425                        sizeof(struct nvsp_message));
426                 complete(&net_device->channel_init_wait);
427         } else if (nvsp_packet->hdr.msg_type ==
428                    NVSP_MSG1_TYPE_SEND_RNDIS_PKT_COMPLETE) {
429                 /* Get the send context */
430                 nvsc_packet = (struct hv_netvsc_packet *)(unsigned long)
431                         packet->trans_id;
432
433                 /* Notify the layer above us */
434                 nvsc_packet->completion.send.send_completion(
435                         nvsc_packet->completion.send.send_completion_ctx);
436
437                 atomic_dec(&net_device->num_outstanding_sends);
438
439                 if (netif_queue_stopped(ndev))
440                         netif_wake_queue(ndev);
441         } else {
442                 netdev_err(ndev, "Unknown send completion packet type- "
443                            "%d received!!\n", nvsp_packet->hdr.msg_type);
444         }
445
446 }
447
448 int netvsc_send(struct hv_device *device,
449                         struct hv_netvsc_packet *packet)
450 {
451         struct netvsc_device *net_device;
452         int ret = 0;
453         struct nvsp_message sendMessage;
454         struct net_device *ndev;
455
456         net_device = get_outbound_net_device(device);
457         if (!net_device)
458                 return -ENODEV;
459         ndev = net_device->ndev;
460
461         sendMessage.hdr.msg_type = NVSP_MSG1_TYPE_SEND_RNDIS_PKT;
462         if (packet->is_data_pkt) {
463                 /* 0 is RMC_DATA; */
464                 sendMessage.msg.v1_msg.send_rndis_pkt.channel_type = 0;
465         } else {
466                 /* 1 is RMC_CONTROL; */
467                 sendMessage.msg.v1_msg.send_rndis_pkt.channel_type = 1;
468         }
469
470         /* Not using send buffer section */
471         sendMessage.msg.v1_msg.send_rndis_pkt.send_buf_section_index =
472                 0xFFFFFFFF;
473         sendMessage.msg.v1_msg.send_rndis_pkt.send_buf_section_size = 0;
474
475         if (packet->page_buf_cnt) {
476                 ret = vmbus_sendpacket_pagebuffer(device->channel,
477                                                   packet->page_buf,
478                                                   packet->page_buf_cnt,
479                                                   &sendMessage,
480                                                   sizeof(struct nvsp_message),
481                                                   (unsigned long)packet);
482         } else {
483                 ret = vmbus_sendpacket(device->channel, &sendMessage,
484                                 sizeof(struct nvsp_message),
485                                 (unsigned long)packet,
486                                 VM_PKT_DATA_INBAND,
487                                 VMBUS_DATA_PACKET_FLAG_COMPLETION_REQUESTED);
488
489         }
490
491         if (ret == 0) {
492                 atomic_inc(&net_device->num_outstanding_sends);
493         } else if (ret == -EAGAIN) {
494                 netif_stop_queue(ndev);
495                 if (atomic_read(&net_device->num_outstanding_sends) < 1)
496                         netif_wake_queue(ndev);
497         } else {
498                 netdev_err(ndev, "Unable to send packet %p ret %d\n",
499                            packet, ret);
500         }
501
502         return ret;
503 }
504
505 static void netvsc_send_recv_completion(struct hv_device *device,
506                                         u64 transaction_id)
507 {
508         struct nvsp_message recvcompMessage;
509         int retries = 0;
510         int ret;
511         struct net_device *ndev;
512         struct netvsc_device *net_device = hv_get_drvdata(device);
513
514         ndev = net_device->ndev;
515
516         recvcompMessage.hdr.msg_type =
517                                 NVSP_MSG1_TYPE_SEND_RNDIS_PKT_COMPLETE;
518
519         /* FIXME: Pass in the status */
520         recvcompMessage.msg.v1_msg.send_rndis_pkt_complete.status =
521                 NVSP_STAT_SUCCESS;
522
523 retry_send_cmplt:
524         /* Send the completion */
525         ret = vmbus_sendpacket(device->channel, &recvcompMessage,
526                                sizeof(struct nvsp_message), transaction_id,
527                                VM_PKT_COMP, 0);
528         if (ret == 0) {
529                 /* success */
530                 /* no-op */
531         } else if (ret == -EAGAIN) {
532                 /* no more room...wait a bit and attempt to retry 3 times */
533                 retries++;
534                 netdev_err(ndev, "unable to send receive completion pkt"
535                         " (tid %llx)...retrying %d\n", transaction_id, retries);
536
537                 if (retries < 4) {
538                         udelay(100);
539                         goto retry_send_cmplt;
540                 } else {
541                         netdev_err(ndev, "unable to send receive "
542                                 "completion pkt (tid %llx)...give up retrying\n",
543                                 transaction_id);
544                 }
545         } else {
546                 netdev_err(ndev, "unable to send receive "
547                         "completion pkt - %llx\n", transaction_id);
548         }
549 }
550
551 /* Send a receive completion packet to RNDIS device (ie NetVsp) */
552 static void netvsc_receive_completion(void *context)
553 {
554         struct hv_netvsc_packet *packet = context;
555         struct hv_device *device = (struct hv_device *)packet->device;
556         struct netvsc_device *net_device;
557         u64 transaction_id = 0;
558         bool fsend_receive_comp = false;
559         unsigned long flags;
560         struct net_device *ndev;
561
562         /*
563          * Even though it seems logical to do a GetOutboundNetDevice() here to
564          * send out receive completion, we are using GetInboundNetDevice()
565          * since we may have disable outbound traffic already.
566          */
567         net_device = get_inbound_net_device(device);
568         if (!net_device)
569                 return;
570         ndev = net_device->ndev;
571
572         /* Overloading use of the lock. */
573         spin_lock_irqsave(&net_device->recv_pkt_list_lock, flags);
574
575         packet->xfer_page_pkt->count--;
576
577         /*
578          * Last one in the line that represent 1 xfer page packet.
579          * Return the xfer page packet itself to the freelist
580          */
581         if (packet->xfer_page_pkt->count == 0) {
582                 fsend_receive_comp = true;
583                 transaction_id = packet->completion.recv.recv_completion_tid;
584                 list_add_tail(&packet->xfer_page_pkt->list_ent,
585                               &net_device->recv_pkt_list);
586
587         }
588
589         /* Put the packet back */
590         list_add_tail(&packet->list_ent, &net_device->recv_pkt_list);
591         spin_unlock_irqrestore(&net_device->recv_pkt_list_lock, flags);
592
593         /* Send a receive completion for the xfer page packet */
594         if (fsend_receive_comp)
595                 netvsc_send_recv_completion(device, transaction_id);
596
597 }
598
599 static void netvsc_receive(struct hv_device *device,
600                             struct vmpacket_descriptor *packet)
601 {
602         struct netvsc_device *net_device;
603         struct vmtransfer_page_packet_header *vmxferpage_packet;
604         struct nvsp_message *nvsp_packet;
605         struct hv_netvsc_packet *netvsc_packet = NULL;
606         unsigned long start;
607         unsigned long end, end_virtual;
608         /* struct netvsc_driver *netvscDriver; */
609         struct xferpage_packet *xferpage_packet = NULL;
610         int i, j;
611         int count = 0, bytes_remain = 0;
612         unsigned long flags;
613         struct net_device *ndev;
614
615         LIST_HEAD(listHead);
616
617         net_device = get_inbound_net_device(device);
618         if (!net_device)
619                 return;
620         ndev = net_device->ndev;
621
622         /*
623          * All inbound packets other than send completion should be xfer page
624          * packet
625          */
626         if (packet->type != VM_PKT_DATA_USING_XFER_PAGES) {
627                 netdev_err(ndev, "Unknown packet type received - %d\n",
628                            packet->type);
629                 return;
630         }
631
632         nvsp_packet = (struct nvsp_message *)((unsigned long)packet +
633                         (packet->offset8 << 3));
634
635         /* Make sure this is a valid nvsp packet */
636         if (nvsp_packet->hdr.msg_type !=
637             NVSP_MSG1_TYPE_SEND_RNDIS_PKT) {
638                 netdev_err(ndev, "Unknown nvsp packet type received-"
639                         " %d\n", nvsp_packet->hdr.msg_type);
640                 return;
641         }
642
643         vmxferpage_packet = (struct vmtransfer_page_packet_header *)packet;
644
645         if (vmxferpage_packet->xfer_pageset_id != NETVSC_RECEIVE_BUFFER_ID) {
646                 netdev_err(ndev, "Invalid xfer page set id - "
647                            "expecting %x got %x\n", NETVSC_RECEIVE_BUFFER_ID,
648                            vmxferpage_packet->xfer_pageset_id);
649                 return;
650         }
651
652         /*
653          * Grab free packets (range count + 1) to represent this xfer
654          * page packet. +1 to represent the xfer page packet itself.
655          * We grab it here so that we know exactly how many we can
656          * fulfil
657          */
658         spin_lock_irqsave(&net_device->recv_pkt_list_lock, flags);
659         while (!list_empty(&net_device->recv_pkt_list)) {
660                 list_move_tail(net_device->recv_pkt_list.next, &listHead);
661                 if (++count == vmxferpage_packet->range_cnt + 1)
662                         break;
663         }
664         spin_unlock_irqrestore(&net_device->recv_pkt_list_lock, flags);
665
666         /*
667          * We need at least 2 netvsc pkts (1 to represent the xfer
668          * page and at least 1 for the range) i.e. we can handled
669          * some of the xfer page packet ranges...
670          */
671         if (count < 2) {
672                 netdev_err(ndev, "Got only %d netvsc pkt...needed "
673                         "%d pkts. Dropping this xfer page packet completely!\n",
674                         count, vmxferpage_packet->range_cnt + 1);
675
676                 /* Return it to the freelist */
677                 spin_lock_irqsave(&net_device->recv_pkt_list_lock, flags);
678                 for (i = count; i != 0; i--) {
679                         list_move_tail(listHead.next,
680                                        &net_device->recv_pkt_list);
681                 }
682                 spin_unlock_irqrestore(&net_device->recv_pkt_list_lock,
683                                        flags);
684
685                 netvsc_send_recv_completion(device,
686                                             vmxferpage_packet->d.trans_id);
687
688                 return;
689         }
690
691         /* Remove the 1st packet to represent the xfer page packet itself */
692         xferpage_packet = (struct xferpage_packet *)listHead.next;
693         list_del(&xferpage_packet->list_ent);
694
695         /* This is how much we can satisfy */
696         xferpage_packet->count = count - 1;
697
698         if (xferpage_packet->count != vmxferpage_packet->range_cnt) {
699                 netdev_err(ndev, "Needed %d netvsc pkts to satisfy "
700                         "this xfer page...got %d\n",
701                         vmxferpage_packet->range_cnt, xferpage_packet->count);
702         }
703
704         /* Each range represents 1 RNDIS pkt that contains 1 ethernet frame */
705         for (i = 0; i < (count - 1); i++) {
706                 netvsc_packet = (struct hv_netvsc_packet *)listHead.next;
707                 list_del(&netvsc_packet->list_ent);
708
709                 /* Initialize the netvsc packet */
710                 netvsc_packet->xfer_page_pkt = xferpage_packet;
711                 netvsc_packet->completion.recv.recv_completion =
712                                         netvsc_receive_completion;
713                 netvsc_packet->completion.recv.recv_completion_ctx =
714                                         netvsc_packet;
715                 netvsc_packet->device = device;
716                 /* Save this so that we can send it back */
717                 netvsc_packet->completion.recv.recv_completion_tid =
718                                         vmxferpage_packet->d.trans_id;
719
720                 netvsc_packet->total_data_buflen =
721                                         vmxferpage_packet->ranges[i].byte_count;
722                 netvsc_packet->page_buf_cnt = 1;
723
724                 netvsc_packet->page_buf[0].len =
725                                         vmxferpage_packet->ranges[i].byte_count;
726
727                 start = virt_to_phys((void *)((unsigned long)net_device->
728                 recv_buf + vmxferpage_packet->ranges[i].byte_offset));
729
730                 netvsc_packet->page_buf[0].pfn = start >> PAGE_SHIFT;
731                 end_virtual = (unsigned long)net_device->recv_buf
732                     + vmxferpage_packet->ranges[i].byte_offset
733                     + vmxferpage_packet->ranges[i].byte_count - 1;
734                 end = virt_to_phys((void *)end_virtual);
735
736                 /* Calculate the page relative offset */
737                 netvsc_packet->page_buf[0].offset =
738                         vmxferpage_packet->ranges[i].byte_offset &
739                         (PAGE_SIZE - 1);
740                 if ((end >> PAGE_SHIFT) != (start >> PAGE_SHIFT)) {
741                         /* Handle frame across multiple pages: */
742                         netvsc_packet->page_buf[0].len =
743                                 (netvsc_packet->page_buf[0].pfn <<
744                                  PAGE_SHIFT)
745                                 + PAGE_SIZE - start;
746                         bytes_remain = netvsc_packet->total_data_buflen -
747                                         netvsc_packet->page_buf[0].len;
748                         for (j = 1; j < NETVSC_PACKET_MAXPAGE; j++) {
749                                 netvsc_packet->page_buf[j].offset = 0;
750                                 if (bytes_remain <= PAGE_SIZE) {
751                                         netvsc_packet->page_buf[j].len =
752                                                 bytes_remain;
753                                         bytes_remain = 0;
754                                 } else {
755                                         netvsc_packet->page_buf[j].len =
756                                                 PAGE_SIZE;
757                                         bytes_remain -= PAGE_SIZE;
758                                 }
759                                 netvsc_packet->page_buf[j].pfn =
760                                     virt_to_phys((void *)(end_virtual -
761                                                 bytes_remain)) >> PAGE_SHIFT;
762                                 netvsc_packet->page_buf_cnt++;
763                                 if (bytes_remain == 0)
764                                         break;
765                         }
766                 }
767
768                 /* Pass it to the upper layer */
769                 rndis_filter_receive(device, netvsc_packet);
770
771                 netvsc_receive_completion(netvsc_packet->
772                                 completion.recv.recv_completion_ctx);
773         }
774
775 }
776
777 static void netvsc_channel_cb(void *context)
778 {
779         int ret;
780         struct hv_device *device = context;
781         struct netvsc_device *net_device;
782         u32 bytes_recvd;
783         u64 request_id;
784         unsigned char *packet;
785         struct vmpacket_descriptor *desc;
786         unsigned char *buffer;
787         int bufferlen = NETVSC_PACKET_SIZE;
788         struct net_device *ndev;
789
790         packet = kzalloc(NETVSC_PACKET_SIZE * sizeof(unsigned char),
791                          GFP_ATOMIC);
792         if (!packet)
793                 return;
794         buffer = packet;
795
796         net_device = get_inbound_net_device(device);
797         if (!net_device)
798                 goto out;
799         ndev = net_device->ndev;
800
801         do {
802                 ret = vmbus_recvpacket_raw(device->channel, buffer, bufferlen,
803                                            &bytes_recvd, &request_id);
804                 if (ret == 0) {
805                         if (bytes_recvd > 0) {
806                                 desc = (struct vmpacket_descriptor *)buffer;
807                                 switch (desc->type) {
808                                 case VM_PKT_COMP:
809                                         netvsc_send_completion(device, desc);
810                                         break;
811
812                                 case VM_PKT_DATA_USING_XFER_PAGES:
813                                         netvsc_receive(device, desc);
814                                         break;
815
816                                 default:
817                                         netdev_err(ndev,
818                                                    "unhandled packet type %d, "
819                                                    "tid %llx len %d\n",
820                                                    desc->type, request_id,
821                                                    bytes_recvd);
822                                         break;
823                                 }
824
825                                 /* reset */
826                                 if (bufferlen > NETVSC_PACKET_SIZE) {
827                                         kfree(buffer);
828                                         buffer = packet;
829                                         bufferlen = NETVSC_PACKET_SIZE;
830                                 }
831                         } else {
832                                 /* reset */
833                                 if (bufferlen > NETVSC_PACKET_SIZE) {
834                                         kfree(buffer);
835                                         buffer = packet;
836                                         bufferlen = NETVSC_PACKET_SIZE;
837                                 }
838
839                                 break;
840                         }
841                 } else if (ret == -ENOBUFS) {
842                         /* Handle large packet */
843                         buffer = kmalloc(bytes_recvd, GFP_ATOMIC);
844                         if (buffer == NULL) {
845                                 /* Try again next time around */
846                                 netdev_err(ndev,
847                                            "unable to allocate buffer of size "
848                                            "(%d)!!\n", bytes_recvd);
849                                 break;
850                         }
851
852                         bufferlen = bytes_recvd;
853                 }
854         } while (1);
855
856 out:
857         kfree(buffer);
858         return;
859 }
860
861 /*
862  * netvsc_device_add - Callback when the device belonging to this
863  * driver is added
864  */
865 int netvsc_device_add(struct hv_device *device, void *additional_info)
866 {
867         int ret = 0;
868         int i;
869         int ring_size =
870         ((struct netvsc_device_info *)additional_info)->ring_size;
871         struct netvsc_device *net_device;
872         struct hv_netvsc_packet *packet, *pos;
873         struct net_device *ndev;
874
875         net_device = alloc_net_device(device);
876         if (!net_device) {
877                 ret = -ENOMEM;
878                 goto cleanup;
879         }
880
881         /*
882          * Coming into this function, struct net_device * is
883          * registered as the driver private data.
884          * In alloc_net_device(), we register struct netvsc_device *
885          * as the driver private data and stash away struct net_device *
886          * in struct netvsc_device *.
887          */
888         ndev = net_device->ndev;
889
890         /* Initialize the NetVSC channel extension */
891         net_device->recv_buf_size = NETVSC_RECEIVE_BUFFER_SIZE;
892         spin_lock_init(&net_device->recv_pkt_list_lock);
893
894         INIT_LIST_HEAD(&net_device->recv_pkt_list);
895
896         for (i = 0; i < NETVSC_RECEIVE_PACKETLIST_COUNT; i++) {
897                 packet = kzalloc(sizeof(struct hv_netvsc_packet) +
898                                  (NETVSC_RECEIVE_SG_COUNT *
899                                   sizeof(struct hv_page_buffer)), GFP_KERNEL);
900                 if (!packet)
901                         break;
902
903                 list_add_tail(&packet->list_ent,
904                               &net_device->recv_pkt_list);
905         }
906         init_completion(&net_device->channel_init_wait);
907
908         /* Open the channel */
909         ret = vmbus_open(device->channel, ring_size * PAGE_SIZE,
910                          ring_size * PAGE_SIZE, NULL, 0,
911                          netvsc_channel_cb, device);
912
913         if (ret != 0) {
914                 netdev_err(ndev, "unable to open channel: %d\n", ret);
915                 goto cleanup;
916         }
917
918         /* Channel is opened */
919         pr_info("hv_netvsc channel opened successfully\n");
920
921         /* Connect with the NetVsp */
922         ret = netvsc_connect_vsp(device);
923         if (ret != 0) {
924                 netdev_err(ndev,
925                         "unable to connect to NetVSP - %d\n", ret);
926                 goto close;
927         }
928
929         return ret;
930
931 close:
932         /* Now, we can close the channel safely */
933         vmbus_close(device->channel);
934
935 cleanup:
936
937         if (net_device) {
938                 list_for_each_entry_safe(packet, pos,
939                                          &net_device->recv_pkt_list,
940                                          list_ent) {
941                         list_del(&packet->list_ent);
942                         kfree(packet);
943                 }
944
945                 kfree(net_device);
946         }
947
948         return ret;
949 }