net/hyperv: Add support for promiscuous mode setting
[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         } else {
439                 netdev_err(ndev, "Unknown send completion packet type- "
440                            "%d received!!\n", nvsp_packet->hdr.msg_type);
441         }
442
443 }
444
445 int netvsc_send(struct hv_device *device,
446                         struct hv_netvsc_packet *packet)
447 {
448         struct netvsc_device *net_device;
449         int ret = 0;
450         struct nvsp_message sendMessage;
451         struct net_device *ndev;
452
453         net_device = get_outbound_net_device(device);
454         if (!net_device)
455                 return -ENODEV;
456         ndev = net_device->ndev;
457
458         sendMessage.hdr.msg_type = NVSP_MSG1_TYPE_SEND_RNDIS_PKT;
459         if (packet->is_data_pkt) {
460                 /* 0 is RMC_DATA; */
461                 sendMessage.msg.v1_msg.send_rndis_pkt.channel_type = 0;
462         } else {
463                 /* 1 is RMC_CONTROL; */
464                 sendMessage.msg.v1_msg.send_rndis_pkt.channel_type = 1;
465         }
466
467         /* Not using send buffer section */
468         sendMessage.msg.v1_msg.send_rndis_pkt.send_buf_section_index =
469                 0xFFFFFFFF;
470         sendMessage.msg.v1_msg.send_rndis_pkt.send_buf_section_size = 0;
471
472         if (packet->page_buf_cnt) {
473                 ret = vmbus_sendpacket_pagebuffer(device->channel,
474                                                   packet->page_buf,
475                                                   packet->page_buf_cnt,
476                                                   &sendMessage,
477                                                   sizeof(struct nvsp_message),
478                                                   (unsigned long)packet);
479         } else {
480                 ret = vmbus_sendpacket(device->channel, &sendMessage,
481                                 sizeof(struct nvsp_message),
482                                 (unsigned long)packet,
483                                 VM_PKT_DATA_INBAND,
484                                 VMBUS_DATA_PACKET_FLAG_COMPLETION_REQUESTED);
485
486         }
487
488         if (ret != 0)
489                 netdev_err(ndev, "Unable to send packet %p ret %d\n",
490                            packet, ret);
491         else
492                 atomic_inc(&net_device->num_outstanding_sends);
493
494         return ret;
495 }
496
497 static void netvsc_send_recv_completion(struct hv_device *device,
498                                         u64 transaction_id)
499 {
500         struct nvsp_message recvcompMessage;
501         int retries = 0;
502         int ret;
503         struct net_device *ndev;
504         struct netvsc_device *net_device = hv_get_drvdata(device);
505
506         ndev = net_device->ndev;
507
508         recvcompMessage.hdr.msg_type =
509                                 NVSP_MSG1_TYPE_SEND_RNDIS_PKT_COMPLETE;
510
511         /* FIXME: Pass in the status */
512         recvcompMessage.msg.v1_msg.send_rndis_pkt_complete.status =
513                 NVSP_STAT_SUCCESS;
514
515 retry_send_cmplt:
516         /* Send the completion */
517         ret = vmbus_sendpacket(device->channel, &recvcompMessage,
518                                sizeof(struct nvsp_message), transaction_id,
519                                VM_PKT_COMP, 0);
520         if (ret == 0) {
521                 /* success */
522                 /* no-op */
523         } else if (ret == -EAGAIN) {
524                 /* no more room...wait a bit and attempt to retry 3 times */
525                 retries++;
526                 netdev_err(ndev, "unable to send receive completion pkt"
527                         " (tid %llx)...retrying %d\n", transaction_id, retries);
528
529                 if (retries < 4) {
530                         udelay(100);
531                         goto retry_send_cmplt;
532                 } else {
533                         netdev_err(ndev, "unable to send receive "
534                                 "completion pkt (tid %llx)...give up retrying\n",
535                                 transaction_id);
536                 }
537         } else {
538                 netdev_err(ndev, "unable to send receive "
539                         "completion pkt - %llx\n", transaction_id);
540         }
541 }
542
543 /* Send a receive completion packet to RNDIS device (ie NetVsp) */
544 static void netvsc_receive_completion(void *context)
545 {
546         struct hv_netvsc_packet *packet = context;
547         struct hv_device *device = (struct hv_device *)packet->device;
548         struct netvsc_device *net_device;
549         u64 transaction_id = 0;
550         bool fsend_receive_comp = false;
551         unsigned long flags;
552         struct net_device *ndev;
553
554         /*
555          * Even though it seems logical to do a GetOutboundNetDevice() here to
556          * send out receive completion, we are using GetInboundNetDevice()
557          * since we may have disable outbound traffic already.
558          */
559         net_device = get_inbound_net_device(device);
560         if (!net_device)
561                 return;
562         ndev = net_device->ndev;
563
564         /* Overloading use of the lock. */
565         spin_lock_irqsave(&net_device->recv_pkt_list_lock, flags);
566
567         packet->xfer_page_pkt->count--;
568
569         /*
570          * Last one in the line that represent 1 xfer page packet.
571          * Return the xfer page packet itself to the freelist
572          */
573         if (packet->xfer_page_pkt->count == 0) {
574                 fsend_receive_comp = true;
575                 transaction_id = packet->completion.recv.recv_completion_tid;
576                 list_add_tail(&packet->xfer_page_pkt->list_ent,
577                               &net_device->recv_pkt_list);
578
579         }
580
581         /* Put the packet back */
582         list_add_tail(&packet->list_ent, &net_device->recv_pkt_list);
583         spin_unlock_irqrestore(&net_device->recv_pkt_list_lock, flags);
584
585         /* Send a receive completion for the xfer page packet */
586         if (fsend_receive_comp)
587                 netvsc_send_recv_completion(device, transaction_id);
588
589 }
590
591 static void netvsc_receive(struct hv_device *device,
592                             struct vmpacket_descriptor *packet)
593 {
594         struct netvsc_device *net_device;
595         struct vmtransfer_page_packet_header *vmxferpage_packet;
596         struct nvsp_message *nvsp_packet;
597         struct hv_netvsc_packet *netvsc_packet = NULL;
598         unsigned long start;
599         unsigned long end, end_virtual;
600         /* struct netvsc_driver *netvscDriver; */
601         struct xferpage_packet *xferpage_packet = NULL;
602         int i, j;
603         int count = 0, bytes_remain = 0;
604         unsigned long flags;
605         struct net_device *ndev;
606
607         LIST_HEAD(listHead);
608
609         net_device = get_inbound_net_device(device);
610         if (!net_device)
611                 return;
612         ndev = net_device->ndev;
613
614         /*
615          * All inbound packets other than send completion should be xfer page
616          * packet
617          */
618         if (packet->type != VM_PKT_DATA_USING_XFER_PAGES) {
619                 netdev_err(ndev, "Unknown packet type received - %d\n",
620                            packet->type);
621                 return;
622         }
623
624         nvsp_packet = (struct nvsp_message *)((unsigned long)packet +
625                         (packet->offset8 << 3));
626
627         /* Make sure this is a valid nvsp packet */
628         if (nvsp_packet->hdr.msg_type !=
629             NVSP_MSG1_TYPE_SEND_RNDIS_PKT) {
630                 netdev_err(ndev, "Unknown nvsp packet type received-"
631                         " %d\n", nvsp_packet->hdr.msg_type);
632                 return;
633         }
634
635         vmxferpage_packet = (struct vmtransfer_page_packet_header *)packet;
636
637         if (vmxferpage_packet->xfer_pageset_id != NETVSC_RECEIVE_BUFFER_ID) {
638                 netdev_err(ndev, "Invalid xfer page set id - "
639                            "expecting %x got %x\n", NETVSC_RECEIVE_BUFFER_ID,
640                            vmxferpage_packet->xfer_pageset_id);
641                 return;
642         }
643
644         /*
645          * Grab free packets (range count + 1) to represent this xfer
646          * page packet. +1 to represent the xfer page packet itself.
647          * We grab it here so that we know exactly how many we can
648          * fulfil
649          */
650         spin_lock_irqsave(&net_device->recv_pkt_list_lock, flags);
651         while (!list_empty(&net_device->recv_pkt_list)) {
652                 list_move_tail(net_device->recv_pkt_list.next, &listHead);
653                 if (++count == vmxferpage_packet->range_cnt + 1)
654                         break;
655         }
656         spin_unlock_irqrestore(&net_device->recv_pkt_list_lock, flags);
657
658         /*
659          * We need at least 2 netvsc pkts (1 to represent the xfer
660          * page and at least 1 for the range) i.e. we can handled
661          * some of the xfer page packet ranges...
662          */
663         if (count < 2) {
664                 netdev_err(ndev, "Got only %d netvsc pkt...needed "
665                         "%d pkts. Dropping this xfer page packet completely!\n",
666                         count, vmxferpage_packet->range_cnt + 1);
667
668                 /* Return it to the freelist */
669                 spin_lock_irqsave(&net_device->recv_pkt_list_lock, flags);
670                 for (i = count; i != 0; i--) {
671                         list_move_tail(listHead.next,
672                                        &net_device->recv_pkt_list);
673                 }
674                 spin_unlock_irqrestore(&net_device->recv_pkt_list_lock,
675                                        flags);
676
677                 netvsc_send_recv_completion(device,
678                                             vmxferpage_packet->d.trans_id);
679
680                 return;
681         }
682
683         /* Remove the 1st packet to represent the xfer page packet itself */
684         xferpage_packet = (struct xferpage_packet *)listHead.next;
685         list_del(&xferpage_packet->list_ent);
686
687         /* This is how much we can satisfy */
688         xferpage_packet->count = count - 1;
689
690         if (xferpage_packet->count != vmxferpage_packet->range_cnt) {
691                 netdev_err(ndev, "Needed %d netvsc pkts to satisfy "
692                         "this xfer page...got %d\n",
693                         vmxferpage_packet->range_cnt, xferpage_packet->count);
694         }
695
696         /* Each range represents 1 RNDIS pkt that contains 1 ethernet frame */
697         for (i = 0; i < (count - 1); i++) {
698                 netvsc_packet = (struct hv_netvsc_packet *)listHead.next;
699                 list_del(&netvsc_packet->list_ent);
700
701                 /* Initialize the netvsc packet */
702                 netvsc_packet->xfer_page_pkt = xferpage_packet;
703                 netvsc_packet->completion.recv.recv_completion =
704                                         netvsc_receive_completion;
705                 netvsc_packet->completion.recv.recv_completion_ctx =
706                                         netvsc_packet;
707                 netvsc_packet->device = device;
708                 /* Save this so that we can send it back */
709                 netvsc_packet->completion.recv.recv_completion_tid =
710                                         vmxferpage_packet->d.trans_id;
711
712                 netvsc_packet->total_data_buflen =
713                                         vmxferpage_packet->ranges[i].byte_count;
714                 netvsc_packet->page_buf_cnt = 1;
715
716                 netvsc_packet->page_buf[0].len =
717                                         vmxferpage_packet->ranges[i].byte_count;
718
719                 start = virt_to_phys((void *)((unsigned long)net_device->
720                 recv_buf + vmxferpage_packet->ranges[i].byte_offset));
721
722                 netvsc_packet->page_buf[0].pfn = start >> PAGE_SHIFT;
723                 end_virtual = (unsigned long)net_device->recv_buf
724                     + vmxferpage_packet->ranges[i].byte_offset
725                     + vmxferpage_packet->ranges[i].byte_count - 1;
726                 end = virt_to_phys((void *)end_virtual);
727
728                 /* Calculate the page relative offset */
729                 netvsc_packet->page_buf[0].offset =
730                         vmxferpage_packet->ranges[i].byte_offset &
731                         (PAGE_SIZE - 1);
732                 if ((end >> PAGE_SHIFT) != (start >> PAGE_SHIFT)) {
733                         /* Handle frame across multiple pages: */
734                         netvsc_packet->page_buf[0].len =
735                                 (netvsc_packet->page_buf[0].pfn <<
736                                  PAGE_SHIFT)
737                                 + PAGE_SIZE - start;
738                         bytes_remain = netvsc_packet->total_data_buflen -
739                                         netvsc_packet->page_buf[0].len;
740                         for (j = 1; j < NETVSC_PACKET_MAXPAGE; j++) {
741                                 netvsc_packet->page_buf[j].offset = 0;
742                                 if (bytes_remain <= PAGE_SIZE) {
743                                         netvsc_packet->page_buf[j].len =
744                                                 bytes_remain;
745                                         bytes_remain = 0;
746                                 } else {
747                                         netvsc_packet->page_buf[j].len =
748                                                 PAGE_SIZE;
749                                         bytes_remain -= PAGE_SIZE;
750                                 }
751                                 netvsc_packet->page_buf[j].pfn =
752                                     virt_to_phys((void *)(end_virtual -
753                                                 bytes_remain)) >> PAGE_SHIFT;
754                                 netvsc_packet->page_buf_cnt++;
755                                 if (bytes_remain == 0)
756                                         break;
757                         }
758                 }
759
760                 /* Pass it to the upper layer */
761                 rndis_filter_receive(device, netvsc_packet);
762
763                 netvsc_receive_completion(netvsc_packet->
764                                 completion.recv.recv_completion_ctx);
765         }
766
767 }
768
769 static void netvsc_channel_cb(void *context)
770 {
771         int ret;
772         struct hv_device *device = context;
773         struct netvsc_device *net_device;
774         u32 bytes_recvd;
775         u64 request_id;
776         unsigned char *packet;
777         struct vmpacket_descriptor *desc;
778         unsigned char *buffer;
779         int bufferlen = NETVSC_PACKET_SIZE;
780         struct net_device *ndev;
781
782         packet = kzalloc(NETVSC_PACKET_SIZE * sizeof(unsigned char),
783                          GFP_ATOMIC);
784         if (!packet)
785                 return;
786         buffer = packet;
787
788         net_device = get_inbound_net_device(device);
789         if (!net_device)
790                 goto out;
791         ndev = net_device->ndev;
792
793         do {
794                 ret = vmbus_recvpacket_raw(device->channel, buffer, bufferlen,
795                                            &bytes_recvd, &request_id);
796                 if (ret == 0) {
797                         if (bytes_recvd > 0) {
798                                 desc = (struct vmpacket_descriptor *)buffer;
799                                 switch (desc->type) {
800                                 case VM_PKT_COMP:
801                                         netvsc_send_completion(device, desc);
802                                         break;
803
804                                 case VM_PKT_DATA_USING_XFER_PAGES:
805                                         netvsc_receive(device, desc);
806                                         break;
807
808                                 default:
809                                         netdev_err(ndev,
810                                                    "unhandled packet type %d, "
811                                                    "tid %llx len %d\n",
812                                                    desc->type, request_id,
813                                                    bytes_recvd);
814                                         break;
815                                 }
816
817                                 /* reset */
818                                 if (bufferlen > NETVSC_PACKET_SIZE) {
819                                         kfree(buffer);
820                                         buffer = packet;
821                                         bufferlen = NETVSC_PACKET_SIZE;
822                                 }
823                         } else {
824                                 /* reset */
825                                 if (bufferlen > NETVSC_PACKET_SIZE) {
826                                         kfree(buffer);
827                                         buffer = packet;
828                                         bufferlen = NETVSC_PACKET_SIZE;
829                                 }
830
831                                 break;
832                         }
833                 } else if (ret == -ENOBUFS) {
834                         /* Handle large packet */
835                         buffer = kmalloc(bytes_recvd, GFP_ATOMIC);
836                         if (buffer == NULL) {
837                                 /* Try again next time around */
838                                 netdev_err(ndev,
839                                            "unable to allocate buffer of size "
840                                            "(%d)!!\n", bytes_recvd);
841                                 break;
842                         }
843
844                         bufferlen = bytes_recvd;
845                 }
846         } while (1);
847
848 out:
849         kfree(buffer);
850         return;
851 }
852
853 /*
854  * netvsc_device_add - Callback when the device belonging to this
855  * driver is added
856  */
857 int netvsc_device_add(struct hv_device *device, void *additional_info)
858 {
859         int ret = 0;
860         int i;
861         int ring_size =
862         ((struct netvsc_device_info *)additional_info)->ring_size;
863         struct netvsc_device *net_device;
864         struct hv_netvsc_packet *packet, *pos;
865         struct net_device *ndev;
866
867         net_device = alloc_net_device(device);
868         if (!net_device) {
869                 ret = -ENOMEM;
870                 goto cleanup;
871         }
872
873         /*
874          * Coming into this function, struct net_device * is
875          * registered as the driver private data.
876          * In alloc_net_device(), we register struct netvsc_device *
877          * as the driver private data and stash away struct net_device *
878          * in struct netvsc_device *.
879          */
880         ndev = net_device->ndev;
881
882         /* Initialize the NetVSC channel extension */
883         net_device->recv_buf_size = NETVSC_RECEIVE_BUFFER_SIZE;
884         spin_lock_init(&net_device->recv_pkt_list_lock);
885
886         INIT_LIST_HEAD(&net_device->recv_pkt_list);
887
888         for (i = 0; i < NETVSC_RECEIVE_PACKETLIST_COUNT; i++) {
889                 packet = kzalloc(sizeof(struct hv_netvsc_packet) +
890                                  (NETVSC_RECEIVE_SG_COUNT *
891                                   sizeof(struct hv_page_buffer)), GFP_KERNEL);
892                 if (!packet)
893                         break;
894
895                 list_add_tail(&packet->list_ent,
896                               &net_device->recv_pkt_list);
897         }
898         init_completion(&net_device->channel_init_wait);
899
900         /* Open the channel */
901         ret = vmbus_open(device->channel, ring_size * PAGE_SIZE,
902                          ring_size * PAGE_SIZE, NULL, 0,
903                          netvsc_channel_cb, device);
904
905         if (ret != 0) {
906                 netdev_err(ndev, "unable to open channel: %d\n", ret);
907                 goto cleanup;
908         }
909
910         /* Channel is opened */
911         pr_info("hv_netvsc channel opened successfully\n");
912
913         /* Connect with the NetVsp */
914         ret = netvsc_connect_vsp(device);
915         if (ret != 0) {
916                 netdev_err(ndev,
917                         "unable to connect to NetVSP - %d\n", ret);
918                 goto close;
919         }
920
921         return ret;
922
923 close:
924         /* Now, we can close the channel safely */
925         vmbus_close(device->channel);
926
927 cleanup:
928
929         if (net_device) {
930                 list_for_each_entry_safe(packet, pos,
931                                          &net_device->recv_pkt_list,
932                                          list_ent) {
933                         list_del(&packet->list_ent);
934                         kfree(packet);
935                 }
936
937                 kfree(net_device);
938         }
939
940         return ret;
941 }